DESCRIPTION: State the application’s broad, long-term ...



RESEARCH PLAN

A. Specific Aims

Respiratory gated radiotherapy holds promise to reduce the incidence and severity of normal tissue complications and perhaps provide a means for increased local control by dose escalation for the management of mobile tumors in thorax and abdomen. Precise target localization in real time is particularly important for gated radiotherapy due to the reduced clinical tumor volume to planning target volume (CTV-to-PTV) margin and/or escalated dose. Our overall hypothesis is that, with precise target localization using image guidance techniques, gated radiotherapy will enable improvements in radiotherapy outcome for mobile tumors in thorax and abdomen.

Direct localization of the tumor mass in real time is often difficult, if not impossible. Various surrogates are then used to derive the tumor position during the treatment. Currently there are two forms of gated radiotherapy based on the surrogates used: internal gating and external gating. Internal gating utilizes internal tumor motion surrogates such as the implanted fiducial markers while external gating uses external respiratory surrogates such as markers placed on the surface of the patient’s abdomen. Each method has its own pros and cons. By using external markers, external gating is easy, noninvasive, and does not require radiation dose for imaging. The weakness of external gating is related to the uncertainty in correlation between external surrogates and internal target position. One can then say that external gating is “cheap” however often inaccurate. For internal gating, with the fluoroscopically tracking of implanted markers, the precision of target localization is often satisfactory, since in most cases fiducial markers implanted inside or near the tumor are good surrogates for tumor position. However, fluoroscopically tracking requires radiation dose for imaging and the marker implantation procedure is invasive. With many treatment fractions or long treatment time of a single fraction, the imaging dose can be more than what is clinically acceptable. The invasiveness of the marker implantation procedure might be clinically acceptable for abdominal tumors such as liver, but not for thoracic tumors such as lung, due to the risk of pneumothorax that could be caused by percutaneous marker insertion. Therefore, internal gating can be described as accurate but “expensive” or even impractical for some tumor sites.

The existence of various problems with the current state-of-art techniques for gated radiotherapy has prevented this new treatment modality from being widely implemented in clinical routine. These problems mainly are: 1) external gating is non-invasive but inaccurate, therefore should not be used alone; 2) abdominal tumors can be treated with internal gating by fluoroscopically tracking the implanted markers but imaging dose is a concern; and 3) for thoracic tumors internal gating does not work even if one ignores the imaging dose issue, due to the difficulty in target localization without implanted fiducial markers. These problems have to be solved before gated radiotherapy can be safely implemented in many clinics.

The project proposed here will try to solve the above mentioned problems by developing necessary tools and software infrastructure, using an in-house on-board x-ray imaging system and a commercial respiratory gating system as hardware platforms. We plan to develop two sets of tools. One will allow direct lung tumor mass localization without implanted fiducial markers. The other will allow an optimal combination of external gating with internal gating. The combined gating scheme can be called hybrid gating. Hybrid gating is the solution to both imaging dose problem for internal gating and accuracy problem for external gating. By combining external gating surrogates with internal surrogates, the imaging frequency thus the imaging dose can be greatly reduced. By frequently re-calibrating the external/internal correlation during the treatment, target localization accuracy should be greatly improved. A software infrastructure will also be developed to facilitate the use of the developed tools in a streamlined clinical gating procedure. Correspondingly, there are three specific aims of this proposal.

SA1. To develop tools for gated treatment of lung cancer without implanted fiducial markers.

Patient’s 4D CT images will be acquired using developed techniques. Target volume will be segmented either manually or automatically (using tools developed for other projects) at each breathing phase of the 4D CT scan. Tools will be developed to generate digitally reconstructed fluoroscopy (DRF) images from the 4D CT scan. Before each fraction of the treatment, two simultaneous anterior-posterior (AP) and lateral fluoroscopic images for about 15 seconds long will be acquired using the on-board x-ray imaging system. Tools will be developed to register these fluoroscopic images with corresponding DRF images 1) to identify the target contour in the fluoroscopic images and 2) to align the patient. During the treatment delivery, two simultaneous orthogonal sets of fluoroscopic images will be acquired. Tools will be developed to localize the target in every frame of these images to generate gating signals, by using the DRF images of the corresponding imaging angle.

SA2. To develop tools for combining internal surrogates with external surrogates.

The correlation between internal tumor position and external marker position will be investigated using some existing measured data. Emphasis will be given to the intra- and inter-fraction variation of the internal/external correlation. We will find out, if we use external marker position to derive the internal tumor position, how frequent we need to re-calibrate the internal/external correlation by acquiring the x-ray images. Four schemes of combining external signal with internal information will be investigated and the corresponding tools will be developed. The first approach is called external gating with internal verification. The external marker position will be used for gating, while x-ray images will be taken during the external gating window to verify the internal marker/tumor position. Tools will also be developed for the therapists to visualize the detected tumor/marker position and to monitor the treatment. If the tumor/marker position differs from the reference position by a pre-set tolerance value, the therapists will interrupt the treatment and resume the treatment after re-aligning the patient. The second approach is called double gating. The gating signal generated from the external surrogate will be used to gate the on-board imaging system. Then based on the x-ray images the target will be localized and the linac will be gated. The third and four approaches are both called hybrid gating, i.e., the target position will be derived and the gating signal will be generated by using the external and internal signals together. For the third approach, called hybrid gating with minimal imaging, the x-ray imaging takes place at a uniform rate but the rate will be minimized with the help of external signal. For the fourth approach, called hybrid gating with adaptive imaging, the x-ray images are only taken whenever necessary (so images are taken at a non-uniform rate). These four approaches are at an order of increasing technical difficulty. We will study for various clinical scenarios (tumor sites, individual patients, etc.), the optimal way of combining external gating with internal gating by looking at the clinical practicality, target localization accuracy, and the imaging dose reduction of each of the four schemes.

SA3. To develop a software infrastructure for gated radiotherapy

We propose to develop a software infrastructure to facilitate the incorporation of the proposed tools as well as existing tools into a streamlined clinical procedure for gated radiotherapy. The infrastructure should include the following functions: 1) display 4D CT data and generate DRFs, 2) display in real time, and play back fluoroscopic images, 3) detect and display marker/tumor position in fluoroscopic images, 5) register fluoroscopic images with the DRFs with/without implanted markers, 6) display reference marker/tumor position and generate warning sign when the detected marker/tumor position is outside the tolerance zone around the reference position, 7) input external surrogate signal and generate corresponding external gating signal, 8) input internal surrogate signal (detected marker/tumor position) and generate corresponding internal gating signal, 9) estimate marker/tumor position from the combined external/internal surrogate signals, 10) estimate the optimal time to image using the external surrogate signal, 11) gate the on-board imaging system, and 12) gate the linear accelerator. For each function, a corresponding software module will be developed. Proposed tools and existing tools will be tested and implemented into corresponding modules.

By developing the above mentioned tools, we believe we can treat tumors in thorax and abdomen with gated radiotherapy in a safe and clinically practical way.

B. Background and Significance

B.1. Problems with Respiratory Tumor Motion in Radiotherapy

Radiation therapy is a treatment modality directed towards local control of cancer. The primary goal is to precisely deliver a lethal dose to the tumor while minimizing the dose to surrounding healthy tissues and critical structures. Recent technological advances in radiation therapy, such as intensity-modulated radiation therapy (IMRT), provide the capability of delivering a highly conformal radiation dose distribution to a complex static target volume. However, treatment errors related to internal organ motion may greatly degrade the effectiveness of conformal radiotherapy for the management of thoracic and abdominal lesions, especially when the treatment is done in a hypo-fraction or single fraction manner [1-8]. This has become a pressing issue in the emerging era of image-guided radiation therapy (IGRT).

Intra-fraction organ motion is mainly caused by patient respiration, sometimes also by skeletal muscular, cardiac, or gastrointestinal systems. Respiration induced organ motion has been studied by directly tracking the movement of the tumor [2, 9-12], the host organ [13, 14], radio-opaque markers implanted at the tumor site [4, 15-19], radioactive tracer targeting the tumor [20, 21], and surrogate structures such as diaphragm and chest wall [22-24]. Various imaging modalities have been used for organ motion studies, including ultrasound [13, 14], CT [9, 10, 22, 25], MR [26], and fluoroscopy [2, 4, 11, 15-19, 23, 24, 27-31]. It has been shown that the motion magnitude can be clinically significant (e.g., of the order of 2 - 3 cm), depending on tumor sites and individual patients.

One category of methods to account for respiratory motion is to minimize the tumor motion, using techniques such as breath holding and forced shallow breathing (such as jet ventilation) [10, 32-39]. These techniques require patient compliance, active participation and, often, extra therapist participation. They may not be well tolerated by patients with compromised lung function which is the case for most lung cancer patients [40]. Another category of the methods accounting for respiratory motion is to allow free tumor motion while adapting the radiation beam to the tumor position by either respiratory gating or beam tracking.

Respiratory gating limits radiation exposure to the portion of the breathing cycle when the tumor is in the path of the beam [15, 23, 29, 30, 40-51]. Beam tracking technique follows the target dynamically with the radiation beam [52]. It was first implemented in a robotic radiosurgery system (CyberKnife) [53-57]. For linac-based radiotherapy, tumor motion can be compensated for using a dynamic multileaf collimator (MLC) [58-67]. Linac based beam tracking is still under development. Due to technical difficulties and quality assurance considerations, a lot of work has to be done before it can be applied to patient treatment. One the other hand, respiratory gating is technically less challenging and clinically more practical. It has been introduced in clinic practice in a limited number of cancer centers. It is believed that gated radiotherapy will be widely implemented in clinical routine for treating tumors in thorax and abdomen after some needed tools are developed. This proposal will focus on the tool development for gated radiotherapy.

B.2. Some Basic Concepts of Gated Radiotherapy

For gated radiotherapy, precise and real time tumor localization is extremely important because tighter CTV-PTV (clinical tumor volume to planning target volume) margins are often applied based on the expectation of reduced tumor motion [46]. In an idealized gated treatment, tumor position should be directly detected and the delivery of radiation is only allowed when the tumor is at the right position. However, direct detection of the tumor mass in real-time during the treatment is often difficult. Various surrogates are then used to indicate the tumor position. Based on surrogates used, we may categorize respiratory gating into internal gating and external gating. Internal gating utilizes internal tumor motion surrogates such as implanted fiducial markers while external gating relies on external respiratory surrogates such as makers placed on the patient’s abdomen.

A basic concept for the gated treatment is called gate or gating window. A gating window is a range of the surrogate signal (such as the 3D marker position in case of internal gating and the marker position in case of external gating). When the surrogate signal falls in the range (gating window), the gating signal is 1; otherwise it is 0. Therefore, the gating window converts the surrogate signal into gating signal and then the gating signal controls the linac. For internal gating, the gating window is often a small rectangular solid corresponding to the 3D position of the implanted fiducial marker. For external gating, the gating window can be either defined by two anterial-posterial (AP) positions or two phase values of the surface marker, which correspond to two types of external gating: displacement or amplitude gating, and phase gating.

Another basic concept for gated treatment is called duty cycle. Duty cycle is a measure of the treatment efficiency and defined as the ratio of beam-on time to the total treatment time. Intuitively, the larger the gating window, the higher the duty cycle. However, for the same patient, the larger the gating window, the larger the tumor residual motion. Therefore, it is always a trade-off between duty cycle and residual motion.

B.3. History and Current Status of Gated Radiotherapy

Respiratory gated radiation therapy was first developed in Japan in the late 1980s and early 1990s for linac photon beams as well as for heavy ion beams [23, 41, 68, 69]. Various external surrogates were used to monitor respiratory motion, including a combination airbag and strain gauge taped on the patient’s abdomen or back (for prone treatments) to gate a proton beam [41, 68], and position sensors placed on the patient [23, 69, 70]. A major advancement of the gated radiotherapy was the real-time tumor tracking (RTRT) system developed by Mitsubishi Electronics Co., Ltd., Tokyo, in collaboration with the Hokkaido University [29, 30, 47-51]. The RTRT system uses real-time fluoroscopic tracking of gold markers implanted in tumor.

Around the mid 1990s, Kubo and his colleagues at the University of California at Davis introduced the gated radiotherapy technique into the United States. They reported the first feasibility study of gated radiotherapy with a Varian 2100C accelerator, as well as an evaluation of different external surrogate signals to monitor respiratory motion [15]. They also reported a gated radiotherapy system which tracks inferred reflective markers on the patient abdomen using a video camera, developed jointly with Varian Medical Systems, Inc. (Palo Alto, CA) [40]. This system was later commercialized by Varian and called real-time position management (RPM) respiratory gating system. The RPM system has been implemented and investigated clinically at a number of centers [24, 42, 43, 45, 71-76].

Currently, the Mitsubishi/Hokkaido RTRT system is the only internal gating system used in clinical routine, while the Varian RPM system can be considered as the representative external gating system. Each system has its own strengths and weaknesses. The weaknesses of existing gating techniques have been the barriers for the broad implementation of gated radiotherapy. The goal of this project is to develop tools that can eliminate/mitigate the problems, and combine the strengths, of the current internal gating and external techniques.

For the RPM system, two passive inferred reflective markers pasted on a lightweight plastic block are placed on the patient's anterior abdominal surface and monitored by a charge-coupled-device (CCD) video camera which is mounted on the treatment room wall. The surrogate signal is then the abdominal surface motion. Both amplitude and phase gating are allowed by the RPM system. A periodicity filter checks the regularity of the breathing waveform and immediately disables the beam when the breathing waveform becomes irregular, such as patient movement or coughing, and re-enables the beam after establishing that breathing is again regular. The RPM can also be used for treatment simulation along with a radiotherapy simulator and/or a CT scanner to acquire the patient treatment geometry in the gating window and to setup the gating window.

The major strength of the external gating systems is the non-invasiveness and easiness of tracking external markers. The weakness is the uncertainty in the correlation between external marker position and internal target position. That is to say, tracking external is not equivalent to tracking tumor; naïvely trusting the external surrogate can cause significant errors. A solution to this problem is the frequent re-calibration of the internal/external correlation during the treatment.

The Mitsubishi/Hokkaido RTRT system as well as its application in radiotherapy has been extensively published by the Hokkaido group [19, 29, 30, 47-51, 77-88]. The system consists of four sets of diagnostic x-ray camera systems, each of the camera system consisting of an x-ray tube mounted under the floor, a 9-inch image intensifier mounted in the ceiling, and a high-voltage x-ray generator. The four x-ray tubes are placed at right caudal, right cranial, left caudal, and left cranial position with respect to the patient couch at a distance of 280 cm from the isocenter. The image intensifiers are mounted on the ceiling, opposite to the x-ray tubes, at a distance of 180 cm from the isocenter, with beam central axes intersecting at the isocenter. At a given time during patient treatment, depending on the linac gantry angle, two out of the four x-ray systems are enabled to provide a pair of unblocked orthogonal fluoroscopic images. To reduce the scatter radiation from the therapeutic beam to the imagers, the x-ray units and the linac are synchronized, i.e., the MV beam is gated off the kV x-ray units are pulsed.

Using this system, the fiducial markers implanted at the tumor site can be directly tracked fluoroscopically at a video frame rate [29]. The linear accelerator is gated to irradiate the tumor only when the marker is within the internal gating window. The size of the gating window is set at +/-1 to +/-3 mm according to the patient's characteristics and the margin used in treatment planning[30]. Techniques for the insertion of gold markers of 1.5-2.0 mm diameter into or near the tumor were developed for various tumor sites, including bronchoscopic insertion for the peripheral lung, image-guided transcutaneous insertion for the liver, cystoscopic and image-guided percutaneous insertion for the prostate, surgical implantation for spinal/paraspinal lesions[49, 51].

Percutaneously implanting fiducial markers is an invasive procedure with potential risks of infection. Many clinicians are reluctant to use this procedure for lung cancer patients because puncturing of the chest wall may cause pneumothorax. The insertion of gold markers using bronchofiberscopy is feasible and safe only for peripheral-type lung tumors, not for central lung lesions[49, 51]. The Hokkaido group found that the markers fixed into the bronchial tree may significantly change their relationship with tumor after 2 weeks of insertion[87]. Therefore, bronchoscopic insertion of markers seems not a good solution for lung tumor treatment, especially for a large number of fractions.

Imaging dose is a concern for the RTRT treatment. The Hokkaido group has measured the air kerma rate, surface dose with backscatter, and dose distribution in depth in a solid phantom from the RTRT system [85]. It was found that the air kerma rate from one fluoroscope was about 240 mGy/h for a nominal pulse width of 2.0 ms and nominal 100 kVp of X-ray energy at the isocenter of the linear accelerator. The estimated skin surface dose from one fluoroscope in RTRT can be up to about 1200 mGy/h[85]. Therefore, approaches to reduce the amount of exposure are mandatory.

In summary, the major strength of the internal gating systems represented by the RTRT system is the precise and real-time localization of the tumor position during the treatment. The implanted internal markers are often good surrogates for tumor position. Marker migration usually is not an issue if the simulation images are acquired a few day after marker implantation[51, 78]. It is even less a concern if multiple markers are used. The two major weaknesses of internal gating are the risk of pneumothorax for implantation of markers in lung and the high imaging dose required for fluoroscopic tracking.

B.4. Clinical Gain of Gated Radiotherapy

Gated radiotherapy with precise targeting will certainly improve local control and normal tissue complication, although the magnitude of the clinical gains is still unclear. There is a lot we don’t know about the value of reacting to respiratory motion in general, and of gated radiotherapy in particular. This application is not designed to answer these questions – a process which will take many years and the engagement of many institutions. It is designed to develop tools to facilitate the large scale implementation of gated radiotherapy and thus to lead to answering these questions. This is an application without human studies designed to lead to human experimentation.

Local failure rates remain substantial despite the use of 3D conformal radiotherapy (CRT) for non-small cell lung cancer [89]. The biggest cause of local failure may not be a volume effect, but simply due to unappreciated degree of inter- and intra-fraction target movement. We may be missing parts of the target, some of the time, for many patients. We suspect the current margin sizes may not be sufficient considering large uncertainties recently discovered in conventional helical free breathing CT scans, daily setup based on lasers and/or portal imaging, and in handling of intra-fraction motion.

Gated radiotherapy with precise target localization can lead to reduced margin. The quantitative clinical gain of margin reduction depends on various factors, such as clinical site, tumor stage and size, as well as the treatment technique (e.g., fractionation scheme). For large tumors, the normal tissue volume saved by margin reduction is small relative to the target volume, but is large in absolute volume, and is large relative to the already small un-irradiated normal tissue volume. For small tumors, either primary or metastases, the margin size may not be dose limiting and may not contribute to complications for treatment with conventional fractionation scheme. However, it may be crucial for increasingly popular hypo-fractionation treatments [90].

Studies on the clinical/dosimetric gain of margin reduction are rare and preliminary. Using respiratory gating, Wagman et al were able to reduce GTV-to-PTV margin from 2 cm to 1 cm for 8 liver patients [74]. This margin reduction allowed for treatment in 2 patients who otherwise would not had been candidates for radiation therapy due to the fact that both patients had only one functioning kidney located on the ipsilateral side as the tumor. For the remaining 6 patients, the reduction in margin allowed for dose increases of 7-27% (median: 21.3%), with stable normal liver NTCPs, as calculated according to the Michigan modification of the Lyman model parameters [91]. Barnes et al found that, on average, self-gated DIBH decreased the percent of lung volume receiving >20 Gy (V20) from 12.8% to 11% without and to 8.8% with GTV-to-PTV margin reduction, which means margin reduction along can greatly reduce V20 [11]. By CT imaging of dynamically moving spheres, Keall et al found that gated radiotherapy may allow a 2-11 mm reduction in the CTV-to-PTV margin [46]. Our preliminary work using deformable registration, 4D CT data, and Monte Carlo simulation has also indicated that there is a significant dosimetric gain of using gated therapy. More details will be given in the “Preliminary Work” section.

B.5. Potential Impact of the Proposed Project

Both external and internal gating techniques have only been clinically used mainly at major academic centers with extreme caution. What prevents the gated radiotherapy from wide clinical application is mainly due to the concerns about the problems existed in the current techniques. The reluctance to implement external gating reflects the lack of confidence in deriving tumor position through external surrogates. If this confidence can be established by frequently re-calibrating the internal/external correlation, it is believed that external gating will be accepted by majority of clinicians.

One of the problems with internal gating is related to the lung tumor localization. Due to the risk of pneumothorax, percutaneous insertion of fiducial markers inside the lung will never become a popular procedure. Without fiducial markers, current technology does not allow us to track lung tumor precisely and in real time for internal gating. As a small part of a funded NIH R21 grant[92], we have been studying the feasibility of gating lung cancer treatment directly based on the fluoroscopic images[93]. The idea is to calculate the correlation score between the motion-enhanced reference template and each frame of fluoroscopic images and to generate gating signal based on the correlation score. Preliminary work shows this approach is feasible. We plan to continue our research along this direction and to develop tools to facilitate the clinical application of this approach. This work will be far beyond the R21 grant and substantial funding is needed. We propose this work as the SA1 of the current project.

The other major concern with internal gating is the high imaging dose required for fluoroscopic tracking. Imaging dose must be significantly reduced before internal gating becomes a well-accepted clinical procedure. Implanted electromagnetic transponders can be used for real time tumor localization without any ionizing radiation dose. Sub-millimeter accuracy was demonstrated in phantom experiments for such a system[94]. The weaknesses of this technology include: 1) it does not work for lung cancer patients, 2) it does not give an appreciation of patient anatomy near transponders, and 3) it requires other image guidance systems, such as a cone beam CT system, to generate 3D patient data that may be needed for adaptive therapy. Therefore, we plan to take another strategy which uses very low frequency fluoroscopy with imaging dose within the acceptable range. This can be done with an on-board x-ray imaging system. The system can also be used for lung tumor imaging without implanted markers, for cone beam CT scan during the course of the treatment to adjust the treatment plan and/or to re-calibration marker/tumor relationship, and for patient geometric information other than marker positions. Apparently this is more integrated solution not only for gated radiotherapy but also for IGRT in general.

We have investigated the reduction of imaging rate by using predictive filters to predict tumor motion[95]. Our work as well as others [96, 97] indicated that the imaging frequency may be safely reduced to 10 Hz, and thus the imaging dose can then reduced by a factor of 3 compared to the video frequency (30 Hz) that the RTRT system uses. The imaging frequency can be substantially reduced further by utilizing the external surrogate signal together with the internal marker position. This idea was preliminary demonstrated on CyberKnife systems[52, 54]. Inferred markers on the patient abdomen are constantly tracked and used to drive the tumor position. Once in a while an x-ray image is taken to re-calibrate the correlation between internal and external markers. Promising results were obtained by using this simple way of combining external and internal signals. In this project we plant to explore more sophisticated ways along this direction. This approach is actually the solution to two problems; adding internal signal from time to time improves the confidence of external gating, while using external signal reduces the imaging dose of internal gating. The SA2 of this proposal is mainly related to the development of tools for optimally combining external and internal surrogate signals.

In order to apply the tools proposed in SA1 and SA2 in the clinical routine in a streamlined fashion, a software infrastructure will be developed, based a software framework that is being developed as part of the funded NIH R21 grant for real-time tumor tracking[92]. This development work is the SA3 of this proposal.

If we successfully reach all three SA’s, we will have tools that can solves all major problems existed in current gating techniques, and a software infrastructure required for smooth clinical implementation of these tools. That is to say, this project will provide a clinically safe and practical way to implement gated radiotherapy at a large scale. Many patients with tumors in thorax, abdomen, and pelvic will benefit from this work by gaining access to gated radiotherapy of high precision that can lead to reduced margin, escalated dose, and thus improved local control.

C. Preliminary Studies

C.1. An On-board X-ray Imaging System

An on-board x-ray imaging system, as shown in figure 1, called the Integrated Radiotherapy Imaging System (IRIS), which consists of two pairs of gantry-mounted diagnostic x-ray tubes and imagers, has been developed at MGH[98]. The system was assembled in early 2004 and has been used in clinical routine for radiograph-based patient setup. The software required for real-time marker tracking is currently being under the development as part of the funded NIH R21 grant[92]. The IRIS system, along with the RPM system, will serve as the hardware platforms for the proposed tool development work.

[pic]

C.2. A Clinical Procedure for Gated Radiotherapy

An image guided respiration gated (IGRG) treatment procedure has been developed at MGH for gated liver and lung radiotherapy[76]. This procedure uses Varian RPM system as the linac gating and patient respiration monitoring/coaching tool and the IRIS system as image guidance tool. As en effort to reduce the uncertainty in tumor localization using external surrogates, 4D CT scanning, gated radiographic setup, cine EPID (Electronic Portal Imaging Device) treatment verification, as well as patient breath coaching, have been used together with the RPM system. For precise patient setup and treatment verification, implanted fiducial markers are required for liver patients and clear anatomic features near the target are required lung cancer patients. The procedure includes seven major steps:

Step 1: Patient selection and preparation. In addition to medical considerations judged by the attending clinician, we determine the suitability of the patient for receiving IGRG treatment according to three factors. First, since breath coaching technique is used throughout the whole treatment course, the patient should be able to follow the breath coaching instruction. Second, the patient should have radio-opaque makers implanted inside or near the liver tumor or clear anatomic features near the lung tumor. Third, the patient has a large intra-fraction tumor motion in order to have a significant gain from the gated treatment.

Step 2: Breath training and motion assessment. Patient’s breathing pattern can vary inter- and intra-fractionally. To have a reasonable duty cycle for the gated treatment, and more importantly, to keep good reproducibility throughout the whole treatment course, breath coaching is required for our IGRG technique. A breath training session of one hour is scheduled on the simulator, where fluoroscopic images are taken and initial gating window is determined.

Step 3: 4D CT simulation. After the breath training session, a 4D CT simulation is scheduled before the treatment. At first, a free breathing helical CT scan is taken. Then, a 4DCT scan with breathing coaching is acquired. Usually 10 sets of CT data are reconstructed corresponding to 10 different breathing phases. Special attention is paid on the CT data corresponding to breathing phases within the gating window. At this point, the gating window is fine tuned and finalized, considering the balance between residual motion and duty cycle.

Step 4: Treatment planning. The GTV and/or CTV are contoured in each of the 4DCT data sets in the gating window and then combined to define a composite target volume that includes the residual motion. The composite target volume is fused to the 4D CT data set at end-of-exhale (EOE) phase which is used as the planning CT. The critical structures are contoured on the EOE CT data set. A margin is added to CTV to obtained PTV. A 3D CRT or IMRT treatment plan is developed. A backup plan with larger margin for non-gated treatment is also developed using the free breath CT scan.

Step 5: Imaged guided patient setup. The patient is initially setup using laser alignment to skin tattoos as in a conventional treatment. The RPM system is applied to the patient to monitor and coach the patient’s breathing. After the patient’s breathing is properly coached, a pair of gated AP and lateral IRIS radiographs are taken at the EOE phase. Using an in-house software called DIPS [99], the gated radiographs are then matched with DRRs to detect patient shifts.

Step 6: Gated treatment delivery. After the patient alignment, the gated treatment starts with the patient under breath coaching with the RPM system. If it is a 3DCRT treatment, EPID images are taken in cine mode for treatment verification during the delivery of each field.

Step 7: Treatment verification and assessment. The recorded EPID images are analyzed retrospectively to verify the gated treatment. The residual marker motion in the gating window is measured. If the residual motion is significantly larger than what was estimated during the simulation session, the treatment will be modified by simply reducing the gating window. This technique has been published[100] and will be described in section C.5.

C. 3. A 4D CT Scanning Protocol

For gated radiotherapy, patient/tumor geometry within the gating window should be used for treatment planning and patient setup. To acquire such information, gated CT or 4D CT scan should be an integral part of the gated radiotherapy treatment. At MGH, in collaboration with GE Medical Systems (Waukesha, Wisconsin), we have developed a scanning protocol for generating 4D CT image data sets for patients with respiratory tumor motion [101]. This protocol has been implemented in our clinic since 2004 and by now a few hundred of lung and liver patients have been scanned [102-104].

C.4. A Patient Breath Coaching Technique

The RPM gating system has audio instruction and visual feedback functions for patient breath coaching. The audio instruction function tells the patient when to breath in and out, while the visual feedback function guides the patient to have constant end-of-exhale (EOE) position and constant end-of-inhale (EOI) position by letting the patient to look his/her own breathing waveform in real time. Sometimes the audio instruction technique is used along [72, 75]. A commonly used coaching protocol is based both audio instruction and visual feedback [72]. This audio/video protocol has been tested at MGH, on five healthy volunteers, observed during 6 sessions, and 33 lung cancer patients, observed during one session when undergoing 4D CT scans, with free breathing as a control [105]. For all 5 volunteers, breath coaching was well tolerated and the intra- and inter-session reproducibility of the breathing pattern was greatly improved. However, about half of the patients could not follow both audio and video instructions simultaneously, suggesting that the audio/video coaching protocol is too complicated for patients and needs to be simplified.

For amplitude gating, the variation of breathing period has no effect on the treatment. Therefore, audio prompting may not be necessary here. If the gating window is set at EOE, there is no need to have a stable EOI position of breathing waveform. Based on these considerations, we developed a new breath coaching protocol. Two straight lines that contain EOE positions are used to define the amplitude gating window. By looking at his/her own breathing waveform on a pair of video goggles, the patient is asked to put the EOE position in between two lines while breathing out. When breathing in, there is no constraint on the EOI position. An example of this coaching technique is given in figure 2, along with the free breathing trace for the same patient for comparison.

[pic]

C.5. A Treatment Verification Technique for Gated 3DCRT

[pic]

For external gating, the verification of tumor position when beam turns on is important. For gated 3DCRT, we have developed a new technique for treatment verification using EPID in cine mode [100]. Implanted radiopaque fiducial markers inside or near the target are required for this technique. During the treatment, a sequence of EPID images can be acquired without disrupting the treatment. Implanted markers are visualized in the images and their positions in the beam’s eye view are calculated off-line and compared to the reference position by matching the field apertures in corresponding EPID and DRR images, as shown in figure 3. The precision of the patient setup, the placement of the beam-gating window, as well as the residual tumor motion can be assessed for each treatment fraction. Since this technique uses the exit image of a treatment beam, it might be difficult to be applied to gated IMRT treatment. Also, the current technique is off-line. In the proposed project, we plan to improve this technique, by developing software infrastructure to allow on-line treatment verification/monitoring based on real-time IRIS images. This is part of SA3.

C. 6. Studies on the Dosimetric Gain of Gated Radiotherapy

In an early treatment planning study, we investigated the dosimetric gain of margin reduction for beam gating and beam tracking for one lung patient and one liver patient [106]. Tissue motion and deformation were estimated across breathing phases of 4D CT data using deformable registration. The CTV (in the reference phase) volumes were 118 cc for the lung case and 162 cc for the liver case. It was found that the integral dose to healthy tissue outside the CTV was reduced by 20% (lung) and 30%(liver) by using respiratory gating.

To have a more realistic, quantitative, and systematic estimation of the gains of margin reduction in dose and then in TCP/NTCP, we have been trying to compare the gated treatment with non-gated treatment by calculating dose distributions using Monte Carlo simulation on deformably-registered multiple-phase 4D CT scans (cite AAPM). For patients with large tumor motion, gated radiotherapy can significant improve the target coverage. For example, for a patient with 2 cm peak-to-peak respiratory motion, the equivalent uniform dose (EUD) of CTV based on the treatment plan should be 60 Gy. However, the actual delivered EUD is only 30 Gy without gating. With gating, 60 Gy can be delivered.

C.7. Marker Tracking Techniques

The implanted markers can be in either spherical (e.g., 1.5-2.0 mm diameter) or cylindrical in shape (e.g., 0.8 mm diameter and 3 mm length). Although tracking high-density markers may seem easy, there are still some technical challenges for precise and reliable real-time tracking. These challenges include the change of the marker shape from frame to frame (for cylindrical and wire markers), the occlusion by and the confusion with bony structure, air bubbles, etc, the correspondence issue when multiple markers are present, and the poor image quality due to such as MV beam interference. We have developed a parameterized template matching method that addresses some of the issues, which has been used for analysis of the tumor motion due to respiration[107, 108]. As part of the funded NIH R21 grant[92], we are also developing a technique called multiple hypothesis tracking (MHT) that is able to track multiple markers simultaneously without mixing them up and is robust enough to continue tracking even when the marker is moving behind bony anatomy[109, 110]. The developed marker tracking algorithm will be integrated into SA3 to facilitate gated radiotherapy based on internal marker position.

C.8. Tracking Failure Detection Techniques

Tracking algorithms may fail when the signal quality is poor, or when background clutter mimics the target. Therefore, any tracking system should have the ability to detect and correct the tracking failures. When a tracking failure happens, the treatment beam should be held off until the recovery of the correct tracking, or the therapist should interrupt the treatment and re-set the tracking system. We have developed an algorithm that utilizes four signals to detect tracking errors that include distance between rays, pattern recognition score, instantaneous velocity and acceleration[111]. Those signals will be displayed in real time for the therapist to monitor the treatment. In case of lung tumor tracking without implanted markers, the correlation score will be displayed. This is part of SA3.

C.9. Gating Based on Lung Tumor Mass

As part of the fund NIH R21 grant, we have investigated the feasibility of gating the treatment using fluoroscopic information without the implantation of radiopaque fiducial markers. Some promising preliminary results have been obtained [112]. We found that the average image intensity in a region of interest in lung fluoroscopy fluctuates following the patient breathing pattern. This is due to the fact that, as the lungs fill/empty, the radiological pathlength through them shortens/lengthens, giving brighter/darker fluoroscopic intensities. The motion, for which we would like to compensate, is the key to marker-less gating. Our results show that a moving tumor can be isolated from the rest of the anatomy by using a motion enhancement algorithm. The correlation scores calculated between a motion-enhanced reference template and motion-enhanced fluoroscopic images are then used as surrogate signal to generate gating signal. The resulting beam-on pattern is similar to one produced by the RPM external gating system. We plan to investigate along this direction to develop accurate, clinically robust, and computational efficient methods for gated treatment of lung cancer without implanted fiducial markers. This will be SA1 of this proposal.

C.10. Residual Tumor Motion in External Gating

External gating assumes that the correlation between the external surface and the internal tumor position remains constant during the treatment. This assumption has yet to be validated. As the first step, we have measured the residual tumor motion within an external gating window[113]. If the correlation is stable at least at EOE, the residual motion should be similar to what was measured during treatment simulation and used for treatment planning. We used synchronized internal/external data from our collaborators in Japan. Eight lung patients with implanted fiducial markers were studied at the NTT Hospital in Sapporo, Japan. Synchronized internal marker positions and external abdominal surface positions were measured during the entire course of treatment. The RTRT system was used to find the internal markers in four dimensions. We then used the data retrospectively to assess the residual tumor motion within the external gating window. We found that the residual motion (95th percentile) was between 0.7-5.8 mm, 0.8-6.0 mm, and 0.9-6.2 mm, for 20%, 30%, and 40% duty cycle windows, respectively. Large fluctuations (>300%) were seen in the residual motion between some beams, which indicates that for some patients the internal/external correlation may vary significantly even during the same treatment fraction, and frequent re-calibration of the correlation and on-line monitoring are necessary during the treatment.

C.11. Modeling Tumor Trajectory

Modeling tumor trajectory not only helps understanding tumor motion characteristics, but also facilitates the development of tumor motion prediction filters. Tumor motion is often very complicated. A simple cosine model certainly cannot handle such complexity[19, 114]. We have tested two methods for tumor trajectory modeling. One is based on a new concept that we introduced, called Average Tumor Trajectory (ATT), which scales and averages several cycles of tumor trajectory measured during treatment simulation session and then is used to estimate tumor trajectory during the treatment[59]. This works only when breathing coach is applied to enforce a stable and regular breathing pattern. The other method is more general. It utilizes a finite state model that breaks a normal breathing cycle into three states: inhale (IN), exhale (EX), and end of exhale (EOE)[115]. A forth state (IRR) is also introduced to handle irregular breathing. The transition from one state to another is automatically guided by the finite state automaton (FSA) as illustrated in figure 4. The model has been successfully applied to analyze the tumor motion data of 23 lung cancer patients and is being tested for tumor motion prediction (cite AAPM).

[pic]

C.12 Tumor Motion Prediction

Prediction filters are needed in a tumor tracking system to handle the following issues: 1). the system latency (the time delay between imaging and treatment beam reaction), 2) the detection of irregular tumor motion (such as unexpected patient cough or movement), 3) the reduction of imaging frequency for lower imaging dose. We have evaluated various predictive models for reducing tumor localization errors when a real-time tumor-tracking system targets a moving tumor at a slow imaging rate and with large system latencies[95]. We compared linear prediction, neural network prediction and Kalman filtering, against a system which uses no prediction. We also developed a model-based probabilistic solution for mining and prediction of tumor respiratory motion (cite AAPM). By analyzing the historical motion data based on a finite state model, two probability distributions are proposed for knowledge discovery of tumor moving status. These probabilities can be used to determine the current motion state and capture transitions from one state to another. They are dynamically built and used in real-time motion prediction. Results from both studies indicate that using prediction filters, reasonable marker localization accuracy can be achieved for gated treatment for systems that have latencies as large as 200 ms and for systems that have imaging rates as low as 10 Hz. To further reduce the imaging rate, external signal, which can always be at high frequency such as 30 Hz, should be included into the prediction process. This is actually part of SA2.

C.13. A Dynamic Phantom

As part of the PI’s Whitaker grant[116], a computer-controlled motor-driven motion phantom has been developed to simulate external marker motion and tumor motion projected in beam’s eye view (BEV), using measured external and internal signals as input. The design of the phantom composed of three main parts: three sets of step motors and linear stages, an electrical interface and a computer system. The RPM marker box, attached to one stage, can move vertically to simulate external gating surrogate. A mechanical platform, fixed on top of two stages, can move in 2D to simulate tumor motion in BEV. For dosimetric measurement, the platform can hold solid water. All three motors can replicate input external and internal motion signals precisely spatially and temporally. The accuracy of the phantom has been tested using a calibrated camera system. The phantom’s ability to recreate variable patterns of movement makes it useful for testing tools developed in SA2.

C.14. A Radiograph-based Patient Setup Software

A radiograph-based patient setup software, called Digital Image Positioning System (DIPS), has been developed and used at MGH proton therapy center since 2001. To date, more than 50,000 images have been collected and used for patient set-up. Recently, we have developed a semi-automatic feature extracting technique for DIPS and integrated it into the IRIS system for patient setup in our photon therapy clinic [117]. The procedure uses the manually selected features from the first treatment fraction to automatically locate the same features on the second and subsequent fractions. Some software modules of DIPS can be used for the development of the infrastructure for this proposal.

C.15. Software Infrastructure

A software framework for tracking implanted markers is currently under active development at MGH as part of the NIH funded research collaboration between MGH and Varian Medical Systems[92]. This framework, called the IRIS Software Environment (ISE), is responsible for real-time communication with the dual fluoroscopic imagers hardware, real-time dual display, and real-time tracking of implanted markers in 2D and 3D. Completion of ISE framework is scheduled for October 2006, with development of advanced marker tracking and prediction algorithms continuing through October 2007. ISE will serve as the basis for the proposed software infrastructure.

D. Research Design and Methods

D.1. A General Description of the Proposed Clinical Procedure for Gated Radiotherapy

In this section, we propose a new clinical procedure for gated radiotherapy using IRIS and RPM as hardware platforms, and based on the tools and software infrastructure to be developed. As described in the section of “Preliminary Work”, the clinical procedure developed at MGH for gated radiotherapy treatment includes the following steps: 1) patient selection and preparation, 2) breath training and motion assessment, 3) 4D CT simulation, 4) treatment planning, 5) image-guided patient setup, 6) gated treatment delivery, and 7) treatment verification and assessment. Various techniques have been developed and clinically implemented; however, there still exist major technical challenges in the last three steps that prevent the clinical use of gated radiotherapy in an effective and safe way. The proposed tools and infrastructure in this project will handle those challenges, covering steps 5 to 7, which are all related to the treatment delivery part of gated radiotherapy and happen on the linac.

The proposed new clinical procedure for gated radiotherapy is the same as the existing one for the first four steps. With the tools to be developed, the last three steps will be different (improved). For step 5, i.e., image-guided patient setup, before each fraction of treatment, there will be a fluoro setup session, in which about 15 seconds of AP and lateral fluoroscopic images will be taken using the on-board imaging system (IRIS) while the patient’s breathing is coached and stabilized. The reason to choose 15 seconds is that for this time period we can have 3-5 breathing cycles and that is considered necessary for fluoro-based patient setup. These images will be processed off-line to calculate the necessary patient shift in order to align the target properly. For patients with implanted fiducial markers, such as liver patients, markers will be automatically tracked. The mean position of all markers within the gating window will be calculated and compared to the reference position derived from the 4D CT scan. Then patient shift can be calculated.

In general, tracking implanted fiducial markers seems an easy problem however there are still some technical challenges in clinical practice. These challenges include: 1) for cylindrical or wire markers, the marker shape in projection images can change constantly with motion; 2) markers may be occluded by and confused with bony structure, air bubbles, or other features in the fluoroscopic images; 3) from some imaging angles, multiple markers may located closely in the images and the correspondence issue is not trivial; and 4) during the treatment the quality of the fluoroscopic images is often degraded due to the MV beam interference. We have developed a technique for tracking a single marker of cylindrical shape (cite Greg’s abstract). In a funded NIH grant, we are currently developing tools for tracking multiple markers simultaneously (cite). Those marker tracking tools will be used in the proposed project.

For patients with tumors in thorax, we may not have implanted fiducial markers. The detection of tumor location for patient setup is a major technical challenge which will be addressed in the next section of this proposal (SA1). The basic idea is to match the setup fluoroscopic images in the gating window to the corresponding DRFs generated from the patient’s 4D CT data. Therefore, we need to develop tools to generate DRFs and to detect lung tumor position by registering the fluoroscopic images with DRFs.

For gated treatment delivery (step 6), we propose to use both external and internal surrogate signals for gating to combine the merits of external gating and internal gating, i.e., reduced imaging dose and still sufficient treatment accuracy. We plan to test four schemes of combining external and internal signals for gated treatment delivery:

G1. External gating with internal verification

Gating signal will be generated using the external surrogate signal (Varian RPM system) to control the linac. Within the external gating window, x-ray fluoroscopic images will be taken at a pre-set rate. Marker/tumor position will be detected and displayed on a computer monitor along with the reference position and tolerance zone. Warning signal will be given when the detected marker/tumor position is outside of the tolerance zone. When this happens persistently, the therapist should interrupt the treatment and re-align the patient. Research issues here include: 1) to determine the required minimum imaging rate (SA2), 2) to detect the lung tumor position without implanted markers (SA1), and 3) to develop software for displaying measured and reference target positions as well as the tolerance zone, and for generating warning signal (SA3).

G2. Double gating

The basic idea is that, RPM gates IRIS, and then IRIS gates linac. Gating signal generated from the external surrogate signal will be used to gate the IRIS system. Therefore, fluoroscopic images will only be taken within the external gating signal. Marker/tumor position will be detected in the gated fluoroscopic images and then used to gate the linac. Research issues here include: 1) to gate the IRIS system using external gating signal (SA3) and 2) to detect the lung tumor position without implanted markers (SA1), and 3) to gate the linac using internal gating signal (SA3).

G3. Hybrid gating with minimized imaging

External surrogate signal will be used to derive the target position. X-ray images will be taken at a constant rate to re-establish the internal/external correlation. Research issues here include: 1) to model the internal/external correlation (SA2), 2) to determine the minimum imaging rate needed to maintain a good internal/external correlation (SA2), and 3) to detect the lung tumor position without implanted markers (SA1).

G4. Hybrid gating with adaptive imaging

External surrogate signal is constantly acquired and the target position as well as a confidence value will be estimated using the external signal. The internal surrogate signal is only acquired (i.e., an x-ray image is taken) whenever the confidence value falls below a pre-set threshold value. The estimated marker/tumor position will be used to gate the linac. Research issues here include: 1) to model the internal/external correlation (SA2), 2) to develop algorithms to derive the target position with a confidence value from the external signal (SA2), and 3) to detect the lung tumor position without implanted markers (SA1).

The technical difficulty increases from scheme 1 to 4. Each scheme has its pros and cons. All four schemes will be developed, tested, and compared with each other as well as pure external gating and internal gating, to evaluate their practicality, imaging dose reduction, marker/tumor localization precision, etc. Recommendations will be then given for various clinical scenarios.

In this new clinical procedure of gated radiotherapy, treatment verification (step 7) will be integrated into treatment delivery (step 6) and becomes more like treatment monitoring. No matter which treatment delivery scheme is used, we will always display the detected marker/tumor positions within the external gating window on a computer monitor along with the reference position and tolerance zone. Therapists will always be asked to monitor the detected marker/tumor position and interrupt the treatment when any abnormity happens. Statistics of marker/tumor positions within the external gating window will be computed and stored for off-line assessment. Research issues here include: 1) to develop software for displaying measured and reference target positions and the tolerance zone, and for generating warning signal (SA3), and 2) to develop software for calculating and displaying statistics of the target positions in the external gating window (SA3).

D.2. SA1: To develop tools for gated treatment of lung cancer without implanted fiducial markers

D.2.1. Patient data

4D CT scan has become a routine clinical procedure at MGH since 2004 [101]. More than 100 lung patients have been scanned with our 4D CT protocol [102-104]. Currently, for our lung cancer patients, 4D CT data is always acquired for purposes of motion assessment and treatment planning. Most of those patients are treated with non-gated 3DCRT or IMRT if motion is less than 1.5 cm peak-to-peak, using the concept of internal target volume (ITV) to define PTV. For some lung cancer patients, if tumor motion is greater than 1.5 cm, and tumor mass has high contrast in fluoroscopy, or tumor is attached or close to an easily detectable anatomic feature such as diaphragm, we treat them with gated 3D CRT or IMRT using the protocol described in section C.2.. For those patients, GTV will be contoured on each phase of the 4D CT scan. CTV contours will be generated for phases within the gating window and will be combined to generate a PTV with proper margin to take into account the setup error and estimated organ motion that is additional to what is shown during the 4D CT scan. Critical structures will be contoured only on the phase at the center of the gating window, such as the end of exhale (EOE) phase. Currently, we perform manual contouring of GTV for all the phases which is a time consuming process. Automatic segmentation tools have been under the development through a collaborative effort between MGH and Rensselaer Polytechnic Institute (RPI) [118], which is beyond the scope of this grant proposal. For those lung patients under gated treatment, we also take occasional (mostly weekly) IRIS fluoroscopic images to assess, in an off-line manner, the tumor motion and motion pattern change during the treatment course. The 4D CT data and fluoroscopic data acquired for routine patient treatment will be used retrospectively for the development and test tasks proposed fro SA1. An IRB will be applied later.

D.2.2. Registration of the fluoroscopic images to DRFs for patient setup

For each patient, AP and lateral DRFs will be generated from the 4D CT data. The frequency of the DRFs is about 2-5 Hz depending on the period of the patient breathing cycle. The outer contour of the GTV will be projected in each frame of the DRFs. The mean position of the lung tumor in the gating window will be calculated and used later as the reference position for patient setup. We will also acquire about 15 seconds of simultaneous AP and lateral fluoroscopic images during the patient setup session prior to each fraction of treatment. The fluoroscopic images will be processed off-line and registered to corresponding DRFs to calculate the necessary patient shift. An example of 4D CT data, DRFs, and fluoroscopic images is shown in figure x.

We plan to explore two approaches for the registration of fluoroscopic images to DRFs. The first approach is automatic registration. Both fluoroscopic images and DRFs will be motion-enhanced and averaged over all the frames in the gating window. The motion-enhanced method has been shown to work well in our preliminary work, for removing irrelevant static structures [112]. Various existing image registration algorithms, such as cross-correlation or mutual information, will be tested to match the motion-enhanced average fluoroscopic image to the motion-enhanced average DRF image. The registration can be done for both AP and lateral fluoroscopic images either simultaneously or sequentially. When it is done sequentially, the shifts along the common direction, i.e., the super-inferior direction, will be averaged. Both simultaneous and sequential registration methods will be tested and compared.

The automatic registration procedure proposed here has not been tested and it may not work for some patients. As an alternative and backup, we will also develop a manual registration procedure. Various techniques will be explored to facilitate the manual registration. One technique is to play back the setup fluoroscopic images and DRFs with GTV contours side by side, and the clinician will use computer mouse to drag the GTV contour in fluoroscopic images to the right position. Another technique is to overlay DRFs on the fluoroscopic images with different colors or transparency. Motion-enhancement and image subtraction will also be tested. Software tools and user interface to allow manual registration will be developed as part of SA3.

D.2.3. Generation of gating signal from the fluoroscopic images during treatment

During the treatment delivery, two simultaneous orthogonal sets of fluoroscopic images will be acquired. The central axis of each set of images is 45 degrees from the therapy beam central axis. Depending on the technique of combining external and internal signals, the time and frequency of taking fluoroscopic images might be different. For both G1 and G2, fluoroscopic images will only be taken within the external gating window at a pre-set imaging rate. For G3, fluoroscopic images will be taken at a constant imaging rate to re-calibrate the internal/external correlation. For G4, fluoroscopic images will only be taken when required by the algorithm. The required minimum imaging rate might be different for G1, G2, and G3, and will be investigated as part of SA2. Also, how to determine the time to take images for G4 will be studied in SA2. No matter which technique is used for combining external/internal signals, we need to locate the tumor position in each fluoroscopic image, or to identify if the tumor is at/near the reference position as defined by the template using the correlation score.

We propose two strategies for gated treatment for lung cancer patients without implanted fiducial markers: score-based gating and location-based gating. Score-based gating uses the gating signal generated based on the correlation score between a reference template and a region of interest (ROI) in each fluoroscopic image acquired during the treatment, while location-based gating uses the gating signal generated directly from the tracked tumor location. Note that in both strategies, we pre-process all the image frames by motion-enhancement [112].

D.2.3.1. Score-based gating

Our preliminary work belongs to this strategy [112], which will be extended in this proposal. We plan to examine three scoring schemes: (1) [pic], (2) [pic], and (3) [pic], in increasing generality. Here, [pic] is the ROI in a measured fluoroscopic image, [pic] is the i-th reference template (such as one of EOE templates in the EOE gating window), the sign [pic] represents correlation, [pic] is the score used for generating the gating signal, i.e., [pic], where [pic] is the threshold score, [pic] is the Heaviside step function. [pic] for [pic] and [pic]for [pic]. The gating signal [pic] means beam on while [pic]means beam off.

Scheme 1: [pic]. Apparently, our preliminary work is a special case of this scheme, with [pic]. This work can be improved by applying automatic thresholding to the score to generate the gating signal. As shown in figure x, an ROI containing the tumor for all breathing phases is selected. This ROI is determined based on the registration of DRF to the fluoroscopic image during the patient setup session. Assume we plan to gate the treatment at EOE with 35% duty cycle. Using the fluoroscopic images acquired in the setup session, we can determine the intensity threshold to have 35% duty cycle, as shown in figure x. All images with average ROI intensity value less than this threshold are then motion-enhanced and averaged to generate the reference template. During the treatment, the correlation score between the reference template and each motion-enhanced frame of the fluoroscopic images is calculated. High correlation score indicates that the ROI contains an image that is similar to the reference template, i.e., the tumor is in the gating window. In the setup session, the intensity threshold can be translated into the correlation score threshold. This threshold for correlation score is then applied through out the whole treatment fraction. This procedure is illustrated in figure x.

[pic]

[pic]

This method works well for some patients. However, we observed that with this simple implementation of score-based method, gating could be erratic, as shown in figure x. We plan to investigate other forms of combining templates into one reference template to improve the accuracy and robustness of the score-based gating.

[pic]

Scheme 2: [pic]. In stead of combining all EOE templates into one reference template, we can compute the correlation score of each template with the ROI of the fluoroscopic images, i.e., [pic], then combine the scores. There are various ways to combine the correlation scores to generate a more robust score. We could simply find the total correlation overall [pic] to serve as our scoring function. In a more elegant way, we plan to apply a voting scheme to generate the final score. Voting several scores in a sense can provide smoother scores, and hence, can give us smoother gating boundaries. The idea of combining the scores through voting is inspired by bagging classifiers [119]. Bagging is an ensemble technique for combining multiple classifiers. The idea is to generate different expert classifiers by diversification through random sampling with replacement of the training data and then take the majority vote as the final classification. Breiman has proved that bagging improves the performance of the base classifier [119]. In our case, we diversify by random sampling different EOE templates. We will also investigate the performance, if we pick representative templates at fixed phase points within the EOE gating window. If we have images more than one cycle in setup session, we can average all templates at the same breathing phase point to generate [pic].

When we only use templates in the gating window for correlation calculation, what we know is how close the tumor is to the reference position. If we also use other parts of the phase cycle to generate multiple templates, then we may have a better idea of the tumor motion during the whole breathing cycle and then generate gating signal in a more robust way. If we know that the tumor image is in the inhale state, then it cannot be in the EOE state. We will develop techniques that combine correlation scores from templates at various breathing phases (e.g., templates from inhale, exhale, and intermediate phases). As a simple example, we can the correlation scores as: [pic], where[pic] and [pic] are reference templates developed at EOE and inhale phases, respectively, and [pic] means multiply the correlation signal by [pic] or invert the signal. We plan to try even more templates and test various ways to combine their scores to generate robust gating signal.

Scheme 3: [pic]. Schemes 1 and 2 both use correlation scores. In general, we can take advantage of other functions of [pic] and [pic]. In Schemes 1 and 2, which are actually special cases of Scheme 3, we apply a threshold to the final score to generate gating signal. Apparently, the ultimate goal is to determine the gating signal (when to turn the radiation beam on or off). We either turn the beam on when we are in the gating window or turn it off otherwise. This exclusivity condition provides us with a clue that the gating problem can actually be recast as a classification problem. This opens a vast resource of classification algorithms from the machine learning literature that we can explore, such as support vector machines (SVM) [120], decision trees [121], Bayes classifier [122]. An SVM projects instances into high dimensional space via kernels and then learns a linear separator that maximizes the margin between the two classes. Kernels allow SVM to perform dot products in high dimensional space by working on kernel functions in the low dimensional space, thus avoiding the computational cost in high dimensions. SVMs have good theoretical properties; it learns an optimal bound on the expected error, and finds an optimal solution as opposed to many learning algorithms that provides local optima (such as, neural networks [123]). SVM has shown success and demonstrated good generalization performance in several classification tasks, examples are gene, text, and character recognition. We will also try decision trees, which build axes parallel decision boundaries of the feature space. A Bayes classifier provides the optimal decision rule if we know the true probability distributions of each class. Since we do not know these distributions, we will estimate them by assuming that each class comes from a mixture of multivariate Gaussians. Note that in all cases, training (or estimating the parameters and decision surfaces) is done offline. During on-line treatment, we simply look at where the test instance, [pic], is and then make our decision. Each of these classifiers provides the output: turn on or turn off. They also have versions that provide the decision confidence. For example, for SVMs, [pic], where [pic] is the class for template [pic] in the training set, [pic] is the Lagrange multiplier, [pic]is the kernel function, and [pic] is the offset. For decision trees, score can be computed as the proportion of class [pic] in the leaf node that [pic] belongs to. For Bayes, score is equal to the probability of class [pic] given [pic], [pic].

For all score-based methods, the final score used to generate gating signal will also be used for treatment monitoring. The score, along with the pre-set threshold value, will be graphically displayed on the screen of the control computer. If the score cannot reach the threshold, i.e., beam cannot be turned on, for several breathing cycles, the therapist will interrupt the treatment and re-set the patient. This is part of SA3.

D.2.3.2. Location-based gating:

[pic]

The second strategy is to perform lung tumor tracking so as to identify the tumor location and then generate gating signal. The tumor location results here serve as internal signal inputs for lung tumor to the methods developed in SA2. A simple tracking approach is to simply perform an exhaustive search and to find the highest correlation between a tumor template and an image region. The image region with the highest correlation to the template is the tracked result. Preliminary experiments show that this leads to erratic locations. When we adaptively update the tumor template at time [pic], [pic], to be equal to the tracked region at time [pic], [pic], the tracking results could be improved. The reason is that the tumor's appearance changes with time. However, we do not want to use the previous estimate as the template to search for the tumor in the next frame, because this is not robust to errors made in previous frames and the tracking may drift.

[pic]

Tumor shape projected in the images may vary throughout the breathing cycle. A common problem in object recognition is to be able to detect the object at various poses (due to changes in rotation, scale, illumination, and other factors). One way to deal with this problem is by building templates for these different poses [124]. We propose to apply multiple templates to represent every fix duration intervals within one breathing cycle. In a preliminary study, we sample twelve templates, [pic] at equal time intervals, based on the intensity signal from the setup session, as shown in figure x. We compute the cross-correlation between each template, [pic], and each measured fluoroscopic image by allowing the template to shift ([pic]) along the x axis and y axis. The [pic] and [pic] coordinates of the tumor centroid for template [pic], ([pic]), are determined during patient setup by registering DRFs to setup fluoroscopic images. The tumor location at this time point is then given as [pic], where ([pic]) is the shift required for template [pic] to produce the best match to the image, and among all the 12 templates, [pic] generates the highest correlation score. For each frame of fluoroscopic images, we perform this operation to identify tumor position, so the tumor position can be tracked continuously.

Figure x is a plot of the best correlation scores for each template with respect to each frame of fluoroscopic images. The y axis is the template ID, the x axis is fluoro frame ID (time), and the grayscale value signifies the correlation score. The brighter the pixel is, the higher is the score. Note that the correlation score in general cycles through the different templates as one would expect. At each frame the tumor region is highly correlated with several templates (bright intensity values along a vertical line in the figure). This is due to the fact that tumor shapes in neighboring phases are similar. Therefore, instead of calculating [pic] using only one template [pic], a more robust way might be to weight the location estimates from all templates that generate high correlation scores (above a pre-set threshold):

[pic]

where [pic] is the weighting factor for template [pic]. We will try various ways of computing the weighting factor, such as the normalized correlation score, the truncated Gaussian function of correlation score, etc. A simple way is averaging, i.e., setting [pic] for the templates with scores above the threshold and [pic] for the templates with scores below the threshold. Figure x shows the estimated x and y coordinates of the tumor centroid versus time, by averaging templates with correlation scores above 95% of the maximum score.

D.2.3.3.Several directions for improving the basic approaches mentioned above

1. Methods to generate templates.

We plan to find out the optimal and practical way of generating templates for each imaging angle. We will investigate three options. The first option is to use DRFs to generate templates. This method is handy. However, the low spatial and temporal resolution, as well as the different image quality, of DRFs, will pose technical challenges when using them to match fluoroscopic images to generate scores or locations for gating. The second option is to acquire some fluoroscopic images for each imaging angle during the treatment simulation or before the first treatment fraction. The third option is to update the templates adaptively during the treatment. For example, we can generalize [pic] and include the current EOE templates for averaging and tapering off the influence of EOE templates that occurred earlier. All three methods will be investigated.

2. Improve the speed for finding the tumor region and dimensionality reduction.

Performing an exhaustive search of the entire fluoroscopic image to find the best match for each twelve template is time consuming. We can speed up the search process by searching only within a surrounding area of [pic] pixels from the previous location. Another strategy for speeding up is by multi-scale search, such as in [125, 126].

A typical template is around 100 pixels x 100 pixels in dimensionality. Another means of reducing the computational complexity (which translates to increase in speed) is by applying dimensionality reduction techniques (e.g., principal component analysis (PCA), local linear embedding (LLE), kernel-PCA). PCA finds a linear transformation, [pic], that projects the original high-dimensional data [pic] with [pic] dimensions to [pic] with [pic] dimensions where [pic], such that the mean squared error between [pic] and [pic] is as small as possible. [pic], here is [pic], where [pic] is the number of data points, and [pic] is a [pic] matrix. The solution is the transformation matrix [pic] whose columns correspond to the [pic] eigenvectors with the [pic] largest eigenvalues of the data covariance. PCA finds a global linear transformation. LLE finds an optimal nonlinear transformation to a low-dimensional space or embedding that preserves the local neighborhood structures. LLE exploits local symmetries by minimizing the reconstruction mean squared error between a data point [pic] and weighted combinations of its neighbors. Kernel-PCA is a generalized version of PCA that allows nonlinear transformations but avoids the combinatorial explosion of time. After any of these transformations, matching will then be computed with Euclidean distance rather than cross-correlation. We will investigate each of these methods. The idea for applying PCA to images or eigen images was introduced by Turk and Pentland for face recognition and has shown to be successful in other object recognition tasks [127]. Not only do the dimensionality reduction methods improve speed, they can also improve the tracking performance because they only keep the most informative dimensions (irrelevant pixels can lead to poor matches). For similar reasons, we will also apply these dimensionality reduction techniques before the classification techniques under score-based Scheme 3.

3. Intelligent means for selecting multiple templates.

The basic algorithm described earlier is to generate multiple templates by sampling at a uniform rate. We can intelligently pick representative templates by clustering the possible set of templates. Clustering is the process of grouping similar objects together. The idea is to group similar templates and use the group's mean to represent that cluster. We will apply k-means clustering [128] or a finite Gaussian mixture model [129, 130] to perform this task. We can perform the task of dimensionality reduction and clustering together through a finite mixture of probabilistic PCA [131, 132]. When clustering, we will apply techniques that also automatically finds the number of clusters using penalty methods, such as Bayesian information criterion [133]. A penalty term is needed because the maximum likelihood estimate increases as more clusters are used. Without the penalty, the likelihood is at a maximum when each data point is considered as an individual cluster (a case of overfitting).

4. Probabilistic model for the breathing cycle.

Another way to perform tumor tracking is to build and take advantage of a model for the breathing cycle. In our preliminary work, we showed that the breathing cycle could be segmented into different states: exhale, end-of-exhale, inhale and irregular states, as shown in figure x. Observe that given that the current state is exhale, the next state can only be either exhale again, end-of-exhale, or irregular state. Also, we do not need to know what the other states are given the previous state to predict the next state. We say, that this random process observes a first-order Markov property, [pic], i.e., the current state is only dependent on the previous state, where [pic] is a random variable that represents the state at time [pic]. We can take advantage of this property to predict the next state. The transitions from one state to another can be modeled by a finite state machine as shown in figure x. However, these states are not directly observable (``hidden''); we only observe the image of the tumor region. We will investigate hidden Markov models (HMMs) [134] for modeling the sequence of observed image regions. Let [pic] be the state transition probability matrix, where [pic], for [pic]. Based on the models shown in figure x, [pic] only on transitions represented by the arrows. We will model the probability of the observed image region given each state, [pic], as a finite mixture of multivariate Gaussians. In this case, we need to first extract features from the image. One simple way is to apply PCA presented earlier. We will automatically find the number of mixture components through the Bayesian information criterion described in (2) above. The parameters will be estimated through the expectation-maximization algorithm [135], and will be computed during training only.

During the treatment session, we apply the HMM model and the previous observation sequence to predict the next state, [pic]. Then, we can compute for the probability of observing a candidate image region given the estimated next state, [pic]. The image region with the highest probability becomes the tracked result. Predicting the next state, narrows down the possible ``templates'' to match. Unlike the previous models, HMM takes advantage of previous states. Computing [pic] for a finite mixture model is analogous to a weighted correlation score for each template within a state, [pic]. If we model the HMM such that each state generates image regions from a single Gaussian distribution, then computing [pic] is analogous to a correlation score for the single most likely template. At the expense of speed, but with the gain in accuracy, we will also examine computing [pic] for each candidate region [pic] and state [pic]. And return the location for the highest probability score among all the states [pic]. This is the probabilistic formulation for finding the best matching region for each possible next state and is analogous to finding the best match among different templates.

D.2.3.4. Validation for robustness

Like in most object tracking approaches, there is a chance of losing track of the tumor. We will, thus, validate the tracked results to make sure that we are not drifting. We can use the correlation score for validation. If the best correlation score is low, we warn the system, and ask the operator to re-calibrate. We can also apply classification methods pointed out in score-based gating (such as, support vector machines, decision trees, Bayes classifier) to check whether the selected region contains the tumor or not, and again even provide confidence scores to warn the operator when to re-calibrate. This validation mechanism will be applied both in the score-based and location-based gating.

D.2.4. Evaluation

We will build a training set with ``ground truth'' location values by asking attending/resident radiation oncologists to retrospectively mark the tumor contours/centroids in measured fluoroscopic images. Then, we can apply mean-squared-error to measure performance. Labeling many frames of fluoroscopic data can be very time consuming. To minimize the number of image frames needed for labeling, we will apply active learning techniques [122, 136, 137]. Active learning deals with the problem of selecting data points for labeling given a large set of unlabeled data. Typical strategies are to pick representative points and points that are difficult to label. Representative points can be selected by applying clustering and choose the points near the cluster means to be labeled. Points that are difficult to label are points near the boundary. These too will be chosen for labeling. We will create an interactive environment to facilitate this labeling process, and at the same time take advantage of the few labeled examples currently stored to query new samples to be manually labeled through active and semi-supervised learning methods [138, 139]. This environment will be part of SA3. We will explore and develop various strategies for active learning in the context of location tracking.

D.3. SA2: To develop tools for combining internal gating with external gating

D.3.1. Patient data

For this SA, we need patient data with measured internal and external marker positions for the development and test of the proposed tools. For lung tumor, we have measured internal/external data from our collaborators in Japan[113]. For abdominal and pelvic tumors, we will collect data from serial weekly 4D CT scans for patients undergoing conventional radiation therapy with the abdominal and pelvic malignancies including cancer of the pancreas, liver, rectum, uterine cervix, and endometrium. This is an IRB-approved clinical study led by Dr. Theodore S. Hong, who is also a co-investigator of this proposed project. The study includes 10 patients of abdominal tumors and 10 patients of pelvic tumors. For each patient, the 4D CT scan is performed weekly for a maximum of 6 additional scans. For patients with pancreatic tumors, oral contrast will be administered thirty minutes prior to CT scan to allow visualization of the duodenum. RPM signal will be acquired during to the 4D CT scan. The RPM signals and contoured tumor locations in 4D CT data will be used to study the inter-fraction variation of internal/external correlation. Some of these patients will be treated with gated radiotherapy. During the treatment, synchronized RPM data and fluoroscopic data will be acquired weekly as part of treatment to assess tumor motion and motion pattern change during the treatment course. These data will also be used retrospectively for this study.

D.3.2 A general framework for combining internal and external signals

The correlation between the locations of an external marker and an internal tumor can be studied and modeled in a number of ways. By far, a linear model is the simplest: an internal tumor position ([pic],[pic],or [pic]) is modeled as a linear function of the external marker position. Such linear functions can account for differences in shift and scale, but they cannot account for differences in the phase between the external and internal position signals, and it is well known that such phase differences often exist[71]. Modeling the internal tumor and external marker positions by sinusoidal functions of time allow one to account for shift, scale, and phase differences, and such “cosine” models (and their generalizations) are often used to model breath patterns and hence internal tumor (and external marker) positions[19, 114]. A generalized cosine model is given below

[pic]

where [pic], [pic], [pic], and [pic] correspond to angular velocity, scale, shift, and phase parameters, respectively, and [pic] is a parameter dictating the order of the cosine model. One could employ such a model to predict future internal tumor positions by first finding the model parameters which fit a set of historical points as well as possible, and then using these model parameters and the above equations to predict future points. Constant or periodic re-calibration would be necessary in order to adapt to changing breath patterns. Assuming that external marker positions are correlated with internal tumor positions to some degree, one might expect some improvement in the predictability of internal position by including the external signal in the above formulation.

We propose to explore extending standard tumor motion models to include an external signal in order to improve future internal position predictability and/or reduce the internal signal sampling rate. However, standard tumor motion models exhibit one general shortcoming: the model itself dictates the form of the predicted tumor motion (linear, sinusoidal, etc.). For regular breath patterns (such as those obtained through the result of breath coaching), such models may well approximate actual breath patterns, given an appropriate breathing model. However, for irregular or varying breath patterns, such models would likely yield poor results or require constant local re-calibration.

With the above in mind, we propose to explore an alternative approach as well. Instead of dictating a fixed (parametric) motion model and finding those parameters which fit the data as well as possible, we propose to explore general motion models which are expressive and encompass many natural motion behaviors and to let the data itself dictate the exact motion model and associated parameters. A promising recent approach of this type was devised by Tao, et al[140]. The motion of an object is treated as a sequence of points in time (e.g., [pic]), and the general motion model is a linear recurrence over retrospective points. For example, in one dimension [pic] using [pic] retrospective points, we would have

[pic]

Such recurrences can model quite complex motion behavior. For example, given appropriate constants [pic], [pic], and [pic], the above recurrence can perfectly model any quadratic motion of [pic] as a function of [pic]; in general, any polynomial motion of degree [pic] can be perfectly modeled by a linear recurrence using [pic] retrospective points. By examining the Taylor series expansion of any candidate motion function, one can find an appropriate polynomial (and hence linear recursive) approximation to any desired degree of accuracy, given a sufficient number of retrospective points. Simultaneous linear recurrences in multiple dimensions can capture dependent motion in the associated space; for example, the following two-dimensional recurrence using two retrospective points models circular behavior in the [pic] plane

[pic]where the recurrence is now expressed in matrix form. The [pic] matrix on the right-hand side of this equation is referred to as the motion matrix [pic], where [pic] now refers to the dimensionality of the space. By appropriately modifying the above motion matrix [pic], one can perfectly model arbitrary elliptical motion in the [pic] plane, and thus a two-dimensional recurrence with two retrospective points encompasses the standard cosine motion models. By using an appropriate number of retrospective points, the above model can encompass (or approximate) arbitrarily complex motion in two-dimensions, and the model can be extended to three-dimensional space by adding the appropriate recurrences for [pic].

Finally, the form of the motion (and the exact model parameters) can be inferred from the data by solving for the [pic] matrix using historical data. For example, given [pic] historical points, one could solve for the motion matrix [pic] which “best fits” the following matrix equation:

[pic]If the quality of fit is measured by mean squared prediction error, then the best fit motion matrix [pic] can be determined by simple matrix operations on the above system.

Such an approach has many potential advantages for our problem. (1) The form of the motion is inferred from the data itself; it is not dictated by a fixed motion model. Given a sufficient number of retrospective points, a very general class of motion models is subsumed by the above formulation, and training on historical data infers the “best” motion model. One can also limit the form or class of motion models considered by placing constraints on the motion matrix and solving for [pic] using constrained optimization techniques (cite). (2) Future tumor positions are predicted from positions in the recent past, not from a globally derived and fixed motion equation. Thus, the model described can more easily adapt to changing local behavior. (3) The information provided by one or more external signals can be incorporated easily and naturally by simply adding one or more dimensions (in addition to [pic],[pic],or [pic]) to the above formulation. (4) The formulation given above can be adapted to predict positions at time points arbitrarily far in the future (with concomitant increasing prediction error, of course) and from partial internal signals and mixed (partial internal, full external) signals. We propose to investigate this model in addition to standard motion models in order to accomplish our specific aims: the reduction of imaging dose through the use of an external signal. A number of preliminary results obtained through the use of this model are described in what follows.

[pic]

A block diagram of our proposed system is shown in figure x. The historical (past) external and internal signal is stored in a memory module. The current external and internal signal (if present) is passed to an operator monitor and graphically displayed. This module serves the purpose of treatment monitoring which is part of SA3. Based on this current information, the operator may choose to halt treatment and re-calibrate if current (external and/or internal) signal is outside the tolerance range. The historical external signal is passed directly to a prediction module when using scheme G1 (external gating with internal verification) wherein gating is determined by the external signal alone. The historical external and internal signals are passed to a model learning module which infers a motion model and its associated parameters. A future internal (and external) position prediction is made using the inferred motion model (and possibly current and past position information). Depending on the type of gating employed, these predictions can dictate treatment and/or dictate whether a new internal image is taken.

With this system diagram in mind, we next describe possible implementations of each of the four gating strategies described earlier: (1) external gating with internal verification [G1], (2) double gating [G2], (3) hybrid gating with minimized imaging [G3], and (4) hybrid gating with adaptive imaging [G4]. We propose to study these implementations and others with the end goal of accurate treatment using a minimal imaging dose.

D.3.3. G1: External gating with internal verification

The gating scheme G1 is similar to the IGRG technique developed at MGH. The major difference is the internal verification part. Images will be taken within the external gating window. The minimal imaging rate will be studied by simulating the treatment using measured data. For treatment monitoring and verification, the display of the detected marker/tumor location or correlation score, along with the reference position or score, on the control computer’s monitor is part of SA1.

D.3.4. G2: Double gating

[pic]

For double gating, the major development work is the interface between the software infrastructure and the x-ray generators and linac, for gating the generators and linac. This is part of SA3. We also need to determine the minimal imaging rate within the external gating window, using methods similar to G1. Another research issue is to determine the optimal size of external gating window. If external gating window is too small, then the treatment duty cycle is low. On the other side, if external gating window is too large, then the imaging duty cycle is large which means too much imaging is given. The treatment duty cycle is also limited by the internal gating window. That means, after a threshold point, the treatment duty cycle saturates and will not increase with the increase of external gating window size. Therefore, the optimal external gating window size should be the turning point on the curve shown in figure x. More work will be done on the issue.

D.3.5. G3: Hybrid gating with minimized imaging

[pic]

In this gating methodology, our goal is to use the full external signal sampled at a fairly high rate (e.g., 30Hz) while making use of an internal signal sampled at a much lower, though constant, rate (e.g., 3Hz). Lowering the rate at which the internal signal is sampled will reduce the imaging dose, but at a cost of increased tumor position prediction error possibly leading to mistreatment. If the internal signal is highly correlated with the external signal, then the full or partial loss of the internal signal need not increase prediction error greatly; if the internal and external signals are not highly correlated, then the presence of a full external signal cannot replace the loss of much or all of the internal signal. Thus, one of our research goals is to establish the correlation between the internal and external signals for various tumor and external marker locations. Our preliminary experiments have shown that the internal and external signals are correlated to some degree and that the external signal can be effectively used to compensate for a loss of internal signal. In figure X, we show the results of predicting the internal y-axis tumor motion 330ms in the future from an internal signal captured at a rate of 3Hz. We used the recursive motion model described earlier by inferring the motion matrix from points within the past 660ms (history) and predicting from points within the past 66ms (retrospect). (The motion model provides predictions for internal [pic],[pic],or [pic]. For the data used, the [pic] motion (S-I direction) was most prominent, so we describe those results.) Note that the sharp peaks in the bottom-half of the plot correspond to the transition between inhale and exhale, while the softer peaks in the top-half of the plot correspond to end-of-exhale (EOE). It can be seen that the prediction of tumor position is quite good except in this EOE range. Unfortunately, this is the range where gated treatment usually occurs; thus, it is most important to make accurate predictions in this range. Conversely, consider the results given in figure x. Here the training and prediction is identical except that an external signal sampled at a rate of 30Hz is added to the model. Note that the predictions made in the critical EOE range are vastly improved. However, the external signal alone is insufficient to predict the internal signal accurately (See figure Z.) Here a motion matrix is trained in the manner described above; however, future prediction are made from the actual external signal and the inferred past internal signal. Thus, an internal signal is only used for a short time (during training), and future predictions are effectively made from the external signal alone. Note the eventual degradation in prediction accuracy due to the lack of a regular internal signal acquisition effectively used for re-calibration.

While our preliminary results are encouraging, many question yet remain. For a given prediction error target (e.g., RMS error threshold or 95 confidence interval), what is the minimal required internal sampling rate? In the context of the recursive motion model described above, what combination of retrospective points (used for prediction) and historical points (used for training) are required to achieve the minimal sampling rate? More generally, how can other motion models be extended to predict future tumor locations from a mixed external and partial internal signal? We propose to study these questions and others.

D.3.6. G4: Hybrid gating with adaptive imaging

In this gating methodology, our goal is to use the full external signal sampled at a fairly high rate (e.g., 30Hz) while making use of the internal signal only “when necessary” to re-calibrate the motion model. How can one determine that re-calibration is likely necessary without sampling the internal signal? One approach using the recursive motion model described above is as follows. Given the external and (partial) internal signal captured in the past, the recursive motion model predicts future internal and external positions: the external signal is treated as another “axis”, and the motion model effectively predicts movement in the resulting four (or higher) dimensional space. One can compare the predicted external position with the actual external position to infer the local accuracy of the motion model. By correlating the internal and external prediction errors on historical data, one could conceivably determine an external error threshold beyond which a given internal error is likely to have occurred. At such times, one could then capture internal signal in order to re-calibrate the motion model (by solving for a new [pic] matrix). One complication arises in the potential use of recursive motion models as described. Recursive motion models most often assume that the input signals are sampled at uniform rates. If the internal signal were sampled at a non-uniform rate, there would be data “missing” for prediction and training. One possible solution is to use the predicted internal positions as a surrogate for the missing actual internal signal; thus, future positions would be predicted from past external, internal, and predicted internal data. We propose to study such approaches and the use of other motion models as well with adaptive internal imaging.

D.4. SA3: To develop a software infrastructure for gated radiotherapy

A software infrastructure for gated radiotherapy will be built to integrate the developed tools into a streamlined gated radiotherapy procedure, to facilitate collaboration between project members, and to allow easy dissemination of tools. Our infrastructure will be based on the IRIS Software Environment (ISE), a software framework currently under development as part of a funded NIH R21 research collaborative between MGH and Varian Medical systems. ISE provides a set of key services, including real-time dual fluoroscopic image acquisition, real-time display, and real-time fiducial marker tracking on the IRIS physical hardware. However, additional software services are needed to support gated radiotherapy. These include a virtual IRIS machine, communication with new physical devices, DRF generation, registration of fluoroscopy with DRFs, an architecture for safety-critical software, treatment monitoring, and statistics and logging.

D.4.1. Hardware abstraction layer and virtual IRIS machine

Because physical access to the IRIS hardware platform is limited for collaborator at Northeastern University (NEU), we propose the implementation of a hardware abstraction layer and virtual IRIS machine. Software will be used to emulate a subset of the physical hardware, including the imaging device, external sensor, x-ray generator, and linear accelerator. The use of a hardware abstraction layer also allows the ISE infrastructure to support a variety of physical devices. For example, the abstraction of the external sensor will support optical systems and spirometry. Thus the developed tools and infrastructure can be readily useful for other investigators who may have hardware platforms different from Varian RPM and linacs, and MGH IRIS system.

The implementation of the virtual IRIS machine requires the production of appropriate images and external sensor input. We will provide two sources for this: synthetic and recorded. The recorded input source will replay images and data from previously recorded sessions. It is more accurate and realistic than the synthetic source, and will be used to exercise SA1, SA2 and SA3. The synthetic input source will generate a tumor trajectory from mathematical models of tumor motion. Synthetic images will depict one or more implanted markers moving against a realistic, but static ribcage. It will be used to exercise SA2 and SA3.

D.4.2. High-speed device communication

The software infrastructure will support communication and control of all input and output hardware devices required for gated treatment. In addition to the imaging panels, communication with the external sensor, x-ray generators, and linear accelerator will be developed. A Varian RPM camera system will be used for tracking the external surrogate, and will be controlled through the RPM serial interface provided in RPM version 1.7. Two CPI Indico 100 x-ray generators will be gated on and off for internal imaging, and will be controlled using custom hardware and a digital I/O card located in the IRIS computer. A Varian Clinac 21EX linear accelerator will be gated on and off for treatment, and will be controlled through an Ontrak ADU208 relay box and other stock hardware. In addition to basic communication, the software infrastructure will provide configuration management, status checking, user interface, and display feedback for each devices.

D.4.3.DRF generation

Digitally reconstructed fluoroscopy (DRF) will be used to match the tumor isocenter in the treatment plan with fluoroscopy in SA1. This project requires routines for generating DRF from 4DCT. The basic procedure is identical to generating digitally reconstructed radiographs (DRR) from a CT scan [141]. However, to support automatic registration, the conversion table from Hounsfeld unit to radiographic attenuation (HUA) will be calibrated to match the IRIS hardware [142]. Lookup tables for the HUA will be generated for a range of energy values, from 40-120 kVp, using a CT density phantom. Another technical challenge is the choosing of the fluoroscopy kV setting for both DRF generation and fluoroscopic image acquisition, which is dependent on many factors such as imaging angle, tumor site, patient size, etc.. During the patient setup session of the first treatment fraction, kV settings can be tuned iteratively to achieve the optimal image quality (such as the highest tumor contrast in the image). This process can be time consuming and also gives the patient unnecessary x-ray dose. Therefore, instead of tuning the kV settings on IRIS using an initial guess based on previous experience, during the treatment planning stage, we could determine an initial setting by examining the tumor contrast in the motion-enhanced DRF templates while interactively changing the kV setting for DRF generation. The initial setting can then fine tuning during the patient setup session. Software tools to facilitate this function will be developed.

D.4.4.Interactive setup with DRFs

In addition to the automatic setup procedure described in SA1, we will implement an interactive registration procedure for matching the planning target volume with the tumor mass visible in fluoroscopy. This will be done by dragging the treatment planning contours, or its DRF, over the fluoro using a mouse. In addition to its use during initial setup, these visualization capabilities will be available throughout the treatment, for easy monitoring and adjustment by the therapists.

D.4.5. Safety critical system architecture

Automatic control of radiation delivery requires careful attention to safety. The software infrastructure will implement a comprehensive strategy for safety-critical system software, including a dual-channel architecture, watchdog processes, and watchdog timers [143]. For gating the linac, our architecture will include dual redundant relay hardware and software control. For imaging and sensing, we will implement analytic redundancy to confirm proper operation. If any sensor or actuator failure is detected, then the radiation is disabled and the system is placed in a safe state. We will guard against unexpected software design errors using a watchdog process. The watchdog process monitors the state of the primary process, and throws a hardware interlock if the primary process is not responding or responds unexpectedly. In addition, we will implement watchdog timers in the hardware interface to the linear accelerator. The primary software process is required to send a heartbeat signal to the external hardware at regular intervals. If the software fails to send a message within the time limit, the hardware will throw an interlock.

D.4.6. Treatment monitoring

D.4.7. Statistics and logging

Decision processes used for automatic control of the treatment and imaging systems must be recorded for proper record keeping, and for post-mortem analysis of system performance. Our software infrastructure will provide a comprehensive real-time statistics and logging subsystem to perform this task.

D.4.8 Software engineering

The software infrastructure for gated radiotherapy be developed and maintained using methods that conform to industry standard practice [144]. This includes design review, code review, unit testing, integration testing, regression testing, source code control, and defect tracking.

Design review: Documents describing the application programming interface (API) and functional capabilities of the software framework and all software modules shall be documented and reviewed. A formal design review meeting will be held to identify and correct design errors.

Code review: Software source code that implements the framework and all software modules will be reviewed. A formal code review meeting will be held to identify and correct errors in implementation.

Unit tests: A unit test is a test of a procedure or module of code. Unit tests exercise the implementation and ensure that design specifications are met. Because a detailed knowleG2e of the code is required for unit test, the same developer who implements a design provides this testing.

Integration tests: Combining several units or modules together is the function of integration testing. This phase tests the combined functionality of a group of units, checking interface compatibility across all the anticipated contexts and intended uses of the module.

Regression tests: Regression tests ensure that new versions of the software do not break previously working functionality. A set of tests for standard operations is run, and the test results are verified against previously expected results. Only after a software version passes regression testing should it be considered stable and suitable for release.

Source code control: For both code management and release of a sequence of versions, we will use a source code control system. This will allow project members to develop on multiple branches independently and then to merge changes into a common main trunk for major releases. All old code remains in the system so that recovery is possible, as is fixing bugs on a previously released version and re-releasing it without dealing with changes made since the previous release.

Defect tracking: A centralized bug tracking system will be used to organize and prioritize software defects and feature requests. Using a defect tracking system improves software quality by allowing developers to see which problems are still outstanding, and prioritize problem resolution.

D.4.9. Software dissemination

To make this software available to widest audience, we will release the software developed for this proposal under an open source license. Major releases will be available for download in source code form on the world wide web, together with all required configuration and make files necessary to create executable programs on a remote system. In addition, each release of the software will include an updated set of documentation, including release notes, a user guide, installation instructions, and a manual for developers who wish to extend the capabilities of the system. For documentation to be both accurate and contemporary, it must be an integral part of the design, implementation, and release processes. The design description is the first element of documentation and from this comes the more detailed explanation of the purpose, algorithms, and use of that piece of software. Documentation will be written in an open format, such as XML, capable of producing output formats appropriate for both printing and online browsing.

D.5. Project Milestones

e. Human Subjects

There will be no human subjects used in this work. All patient data utilized in our system will be fully anonymized, will conform to HIPPA standards and will also be subject to the standards in place by the submitting biomedical research laboratory at MGH.

f. Vertebrate Animals

N/A

g. Literature Cited

h. Consortium/Contractual Arrangements

1. Jacobs, I., J. Vanregemorter, and P. Scalliet, Influence of respiration on calculation and delivery of the prescribed dose in external radiotherapy. Radiother Oncol, 1996. 39(2): p. 123-8.

2. Engelsman, M., E.M. Damen, K. De Jaeger, K.M. van Ingen, and B.J. Mijnheer, The effect of breathing and set-up errors on the cumulative dose to a lung tumor. Radiother Oncol, 2001. 60(1): p. 95-105.

3. Langen, K.M. and D.T. Jones, Organ motion and its management. Int J Radiat Oncol Biol Phys, 2001. 50(1): p. 265-78.

4. Ozhasoglu, C. and M.J. Murphy, Issues in respiratory motion compensation during external-beam radiotherapy. Int J Radiat Oncol Biol Phys, 2002. 52(5): p. 1389-99.

5. Bortfeld, T., K. Jokivarsi, M. Goitein, J. Kung, and S.B. Jiang, Effects of intra-fraction motion on IMRT dose delivery: statistical analysis and simulation. Phys Med Biol, 2002. 47(13): p. 2203-20.

6. Jiang, S.B., C. Pope, K.M. Al Jarrah, J.H. Kung, T. Bortfeld, and G.T. Chen, An experimental investigation on intra-fractional organ motion effects in lung IMRT treatments. Phys Med Biol, 2003. 48(12): p. 1773-84.

7. Chui, C.S., E. Yorke, and L. Hong, The effects of intra-fraction organ motion on the delivery of intensity-modulated field with a multileaf collimator. Med Phys, 2003. 30(7): p. 1736-46.

8. Kung, J.H., P. Zygmanski, N. Choi, and G.T. Chen, A method of calculating a lung clinical target volume DVH for IMRT with intrafractional motion. Med Phys, 2003. 30(6): p. 1103-9.

9. Ross, C.S., D.H. Hussey, E.C. Pennington, W. Stanford, and J.F. Doornbos, Analysis of movement of intrathoracic neoplasms using ultrafast computerized tomography. International Journal of Radiation Oncology, Biology, and Physics, 1990. 18: p. 671-677.

10. Hanley, J., M.M. Debois, D. Mah, G.S. Mageras, A. Raben, K. Rosenzweig, B. Mychalczak, L.H. Schwartz, P.J. Gloeggler, W. Lutz, C.C. Ling, S.A. Leibel, Z. Fuks, and G.J. Kutcher, Deep inspiration breath-hold technique for lung tumors: the potential value of target immobilization and reduced lung density in dose escalation. Int J Radiat Oncol Biol Phys, 1999. 45(3): p. 603-11.

11. Barnes, E.A., B.R. Murray, D.M. Robinson, L.J. Underwood, J. Hanson, and W.H. Roa, Dosimetric evaluation of lung tumor immobilization using breath hold at deep inspiration. Int J Radiat Oncol Biol Phys, 2001. 50(4): p. 1091-8.

12. Stevens, C.W., R.F. Munden, K.M. Forster, J.F. Kelly, Z. Liao, G. Starkschall, S. Tucker, and R. Komaki, Respiratory-driven lung tumor motion is independent of tumor size, tumor location, and pulmonary function. Int J Radiat Oncol Biol Phys, 2001. 51(1): p. 62-8.

13. Davies, S.C., A.L. Hill, R.B. Holmes, M. Halliwell, and P.C. Jacson, Ultrasound quantitation of respiratory organ motion in the upper abdomen. British Journal of Radiology, 1994. 67: p. 1096-1102.

14. Bryan, P.J., S. Custar, J.R. Haaga, and V. Balsara, Respiratory movement of the pancreas: an ultrasonic study. J Ultrasound Med, 1984. 3(7): p. 317-20.

15. Kubo, H.D. and B.C. Hill, Respiration gated radiotherapy treatment: a technical study. Phys Med Biol, 1996. 41(1): p. 83-91.

16. Malone, S., J.M. Crook, W.S. Kendal, and J. Szanto, Respiratory-induced prostate motion: quantification and characterization. Int J Radiat Oncol Biol Phys, 2000. 48(1): p. 105-9.

17. Murphy, M.J., J.R. Adler, Jr., M. Bodduluri, J. Dooley, K. Forster, J. Hai, Q. Le, G. Luxton, D. Martin, and J. Poen, Image-guided radiosurgery for the spine and pancreas. Comput Aided Surg, 2000. 5(4): p. 278-88.

18. Chen, Q.S., M.S. Weinhous, F.C. Deibel, J.P. Ciezki, and R.M. Macklis, Fluoroscopic study of tumor motion due to breathing: facilitating precise radiation therapy for lung cancer patients. Med Phys, 2001. 28(9): p. 1850-6.

19. Seppenwoolde, Y., H. Shirato, K. Kitamura, S. Shimizu, M. van Herk, J.V. Lebesque, and K. Miyasaka, Precise and real-time measurement of 3D tumor motion in lung due to breathing and heartbeat, measured during radiotherapy. Int J Radiat Oncol Biol Phys, 2002. 53(4): p. 822-34.

20. Weiss, P.H., J.M. Baker, and E.J. Potchen, Assessment of hepatic respiratory excursion. J Nucl Med, 1972. 13(10): p. 758-9.

21. Harauz, G. and M.J. Bronskill, Comparison of the liver's respiratory motion in the supine and upright positions: concise communication. J Nucl Med, 1979. 20(7): p. 733-5.

22. Giraud, P., Y. De Rycke, B. Dubray, S. Helfre, D. Voican, L. Guo, J.C. Rosenwald, K. Keraudy, M. Housset, E. Touboul, and J.M. Cosset, Conformal radiotherapy (CRT) planning for lung cancer: analysis of intrathoracic organ motion during extreme phases of breathing. Int J Radiat Oncol Biol Phys, 2001. 51(4): p. 1081-92.

23. Minohara, S., T. Kanai, M. Endo, K. Noda, and M. Kanazawa, Respiratory gated irradiation system for heavy-ion radiotherapy. Int J Radiat Oncol Biol Phys, 2000. 47(4): p. 1097-103.

24. Ford, E.C., G.S. Mageras, E. Yorke, K.E. Rosenzweig, R. Wagman, and C.C. Ling, Evaluation of respiratory movement during gated radiotherapy using film and electronic portal imaging. Int J Radiat Oncol Biol Phys, 2002. 52(2): p. 522-31.

25. Shimizu, S., H. Shirato, K. Kagei, T. Nishioka, X. Bo, H. Dosaka-Akita, S. Hashimoto, H. Aoyama, K. Tsuchiya, and K. Miyasaka, Impact of respiratory movement on the computed tomographic images of small lung tumors in three-dimensional (3D) radiotherapy. Int J Radiat Oncol Biol Phys, 2000. 46(5): p. 1127-33.

26. Korin, H.W., R.L. Ehman, S.J. Riederer, J.P. Felmlee, and R.C. Grimm, Respiratory kinematics of the upper abdominal organs: a quantitative study. Magn Reson Med, 1992. 23(1): p. 172-8.

27. Wade, O.L., Movements of the thoracic cage and diaphragm in respiration. J Physiol, 1954. 124(2): p. 193-212.

28. Ekberg, L., O. Holmberg, L. Wittgren, G. Bjelkengren, and T. Landberg, What margins should be added to the clinical target volume in radiotherapy treatment planning for lung cancer? Radiother Oncol, 1998. 48(1): p. 71-7.

29. Shirato, H., S. Shimizu, T. Kunieda, K. Kitamura, M. van Herk, K. Kagei, T. Nishioka, S. Hashimoto, K. Fujita, H. Aoyama, K. Tsuchiya, K. Kudo, and K. Miyasaka, Physical aspects of a real-time tumor-tracking system for gated radiotherapy. Int J Radiat Oncol Biol Phys, 2000. 48(4): p. 1187-95.

30. Shimizu, S., H. Shirato, S. Ogura, H. Akita-Dosaka, K. Kitamura, T. Nishioka, K. Kagei, M. Nishimura, and K. Miyasaka, Detection of lung tumor movement in real-time tumor-tracking radiotherapy. Int J Radiat Oncol Biol Phys, 2001. 51(2): p. 304-10.

31. Murphy, M.J., D. Martin, R. Whyte, J. Hai, C. Ozhasoglu, and Q.T. Le, The effectiveness of breath-holding to stabilize lung and pancreas tumors during radiosurgery. Int J Radiat Oncol Biol Phys, 2002. 53(2): p. 475-82.

32. Mah, D., J. Hanley, K.E. Rosenzweig, E. Yorke, L. Braban, C.C. Ling, S.A. Leibel, and G. Mageras, Technical aspects of the deep inspiration breath-hold technique in the treatment of thoracic cancer [In Process Citation]. Int J Radiat Oncol Biol Phys, 2000. 48(4): p. 1175-85.

33. Rosenzweig, K.E., J. Hanley, D. Mah, G. Mageras, M. Hunt, S. Toner, C. Burman, C.C. Ling, B. Mychalczak, Z. Fuks, and S.A. Leibel, The deep inspiration breath-hold technique in the treatment of inoperable non-small-cell lung cancer. Int J Radiat Oncol Biol Phys, 2000. 48(1): p. 81-7.

34. Yorke, E.D., L. Wang, K.E. Rosenzweig, D. Mah, J.B. Paoli, and C.S. Chui, Evaluation of deep inspiration breath-hold lung treatment plans with Monte Carlo dose calculation. Int J Radiat Oncol Biol Phys, 2002. 53(4): p. 1058-70.

35. Wong, J.W., M.B. Sharpe, D.A. Jaffray, V.R. Kini, J.M. Robertson, J.S. Stromberg, and A.A. Martinez, The use of active breathing control (ABC) to reduce margin for breathing motion. Int J Radiat Oncol Biol Phys, 1999. 44(4): p. 911-9.

36. Stromberg, J.S., M.B. Sharpe, L.H. Kim, V.R. Kini, D.A. Jaffray, A.A. Martinez, and J.W. Wong, Active breathing control (ABC) for Hodgkin's disease: reduction in normal tissue irradiation with deep inspiration and implications for treatment. Int J Radiat Oncol Biol Phys, 2000. 48(3): p. 797-806.

37. Remouchamps, V.M., N. Letts, F.A. Vicini, M.B. Sharpe, L.L. Kestin, P.Y. Chen, A.A. Martinez, and J.W. Wong, Initial clinical experience with moderate deep-inspiration breath hold using an active breathing control device in the treatment of patients with left-sided breast cancer using external beam radiation therapy. Int J Radiat Oncol Biol Phys, 2003. 56(3): p. 704-15.

38. Remouchamps, V.M., N. Letts, D. Yan, F.A. Vicini, M. Moreau, J.A. Zielinski, J. Liang, L.L. Kestin, A.A. Martinez, and J.W. Wong, Three-dimensional evaluation of intra- and interfraction immobilization of lung and chest wall using active breathing control: A reproducibility study with breast cancer patients. Int J Radiat Oncol Biol Phys, 2003. 57(4): p. 968-78.

39. Remouchamps, V.M., F.A. Vicini, M.B. Sharpe, L.L. Kestin, A.A. Martinez, and J.W. Wong, Significant reductions in heart and lung doses using deep inspiration breath hold with active breathing control and intensity-modulated radiation therapy for patients treated with locoregional breast irradiation. Int J Radiat Oncol Biol Phys, 2003. 55(2): p. 392-406.

40. Kubo, H.D., P.M. Len, S. Minohara, and H. Mostafavi, Breathing-synchronized radiotherapy program at the University of California Davis Cancer Center. Med Phys, 2000. 27(2): p. 346-53.

41. Ohara, K., T. Okumura, M. Akisada, T. Inada, T. Mori, H. Yokota, and M.J. Calaguas, Irradiation synchronized with respiration gate. Int J Radiat Oncol Biol Phys, 1989. 17(4): p. 853-7.

42. Ramsey, C.R., I.L. Cordrey, and A.L. Oliver, A comparison of beam characteristics for gated and nongated clinical x-ray beams. Med Phys, 1999. 26(10): p. 2086-91.

43. Ramsey, C.R., D. Scaperoth, D. Arwood, and A.L. Oliver, Clinical efficacy of respiratory gated conformal radiation therapy. Med Dosim, 1999. 24(2): p. 115-9.

44. Kubo, H.D. and L. Wang, Compatibility of Varian 2100C gated operations with enhanced dynamic wedge and IMRT dose delivery. Med Phys, 2000. 27(8): p. 1732-8.

45. Vedam, S.S., P.J. Keall, V.R. Kini, and R. Mohan, Determining parameters for respiration-gated radiotherapy. Med Phys, 2001. 28(10): p. 2139-46.

46. Keall, P.J., V.R. Kini, S.S. Vedam, and R. Mohan, Potential radiotherapy improvements with respiratory gating. Australas Phys Eng Sci Med, 2002. 25(1): p. 1-6.

47. Shirato, H., S. Shimizu, T. Shimizu, T. Nishioka, and K. Miyasaka, Real-time tumour-tracking radiotherapy. Lancet, 1999. 353(9161): p. 1331-2.

48. Shimizu, S., H. Shirato, K. Kitamura, N. Shinohara, T. Harabayashi, T. Tsukamoto, T. Koyanagi, and K. Miyasaka, Use of an implanted marker and real-time tracking of the marker for the positioning of prostate and bladder cancers. Int J Radiat Oncol Biol Phys, 2000. 48(5): p. 1591-7.

49. Shirato, H., T. Harada, T. Harabayashi, K. Hida, H. Endo, K. Kitamura, R. Onimaru, K. Yamazaki, N. Kurauchi, T. Shimizu, N. Shinohara, M. Matsushita, H. Dosaka-Akita, and K. Miyasaka, Feasibility of insertion/implantation of 2.0-mm-diameter gold internal fiducial markers for precise setup and real-time tumor tracking in radiotherapy. Int J Radiat Oncol Biol Phys, 2003. 56(1): p. 240-7.

50. Shirato, H., S. Shimizu, K. Kitamura, T. Nishioka, K. Kagei, S. Hashimoto, H. Aoyama, T. Kunieda, N. Shinohara, H. Dosaka-Akita, and K. Miyasaka, Four-dimensional treatment planning and fluoroscopic real-time tumor tracking radiotherapy for moving tumor. Int J Radiat Oncol Biol Phys, 2000. 48(2): p. 435-42.

51. Harada, T., H. Shirato, S. Ogura, S. Oizumi, K. Yamazaki, S. Shimizu, R. Onimaru, K. Miyasaka, M. Nishimura, and H. Dosaka-Akita, Real-time tumor-tracking radiation therapy for lung carcinoma by the aid of insertion of a gold marker using bronchofiberscopy. Cancer, 2002. 95(8): p. 1720-7.

52. Murphy, M.J., Tracking moving organs in real time. Semin Radiat Oncol, 2004. 14(1): p. 91-100.

53. Adler, J.R., Jr., M.J. Murphy, S.D. Chang, and S.L. Hancock, Image-guided robotic radiosurgery. Neurosurgery, 1999. 44(6): p. 1299-306; discussion 1306-7.

54. Schweikard, A., G. Glosser, M. Bodduluri, M.J. Murphy, and J.R. Adler, Robotic motion compensation for respiratory movement during radiosurgery. Comput Aided Surg, 2000. 5(4): p. 263-77.

55. Ozhasoglu, C., M.J. Murphy, G. Glosser, M. Bodduluri, A. Schweikard, K.M. Forster, D.P. Martin, and J.R. Adler. Real-time tracking of the tumor volume in precision radiotherapy and body radiosurgery - A novel approach to compensate for respiratory motion. in Proc. 14th Int. Conf. on Computer Assisted Radiology and Surgery (CARS 2000). 2000. San Francisco, CA, USA.

56. Murphy, M.J., S.D. Chang, I.C. Gibbs, Q.T. Le, J. Hai, D. Kim, D.P. Martin, and J.R. Adler, Jr., Patterns of patient movement during frameless image-guided radiosurgery. Int J Radiat Oncol Biol Phys, 2003. 55(5): p. 1400-8.

57. Murphy, M.J., Fiducial-based targeting accuracy for external-beam radiotherapy. Med Phys, 2002. 29(3): p. 334-44.

58. Keall, P.J., V.R. Kini, S.S. Vedam, and R. Mohan, Motion adaptive x-ray therapy: a feasibility study. Phys Med Biol, 2001. 46(1): p. 1-10.

59. Neicu, T., H. Shirato, Y. Seppenwoolde, and S.B. Jiang, Synchronized moving aperture radiation therapy (SMART): average tumour trajectory for lung patients. Phys Med Biol, 2003. 48(5): p. 587-98.

60. Suh, Y., B. Yi, S. Ahn, J. Kim, S. Lee, S. Shin, and E. Choi, Aperture maneuver with compelled breath (AMC) for moving tumors: a feasibility study with a moving phantom. Med Phys, 2004. 31(4): p. 760-6.

61. Papiez, L., The leaf sweep algorithm for an immobile and moving target as an optimal control problem in radiotherapy delivery. Mathematical and Computer Modelling, 2003. 37(7-8): p. 735-745.

62. Rangaraj, D. and L. Papiez, Synchronized delivery of DMLC intensity modulated radiation therapy for stationary and moving targets. Med Phys, 2005. 32(6): p. 1802-17.

63. Papiez, L. and D. Rangaraj, DMLC leaf-pair optimal control for mobile, deforming target. Med Phys, 2005. 32(1): p. 275-85.

64. Keall, P.J., S. Joshi, S.S. Vedam, J.V. Siebers, V.R. Kini, and R. Mohan, Four-dimensional radiotherapy planning for DMLC-based respiratory motion tracking. Med Phys, 2005. 32(4): p. 942-51.

65. Wijesooriya, K., C. Bartee, J.V. Siebers, S.S. Vedam, and P.J. Keall, Determination of maximum leaf velocity and acceleration of a dynamic multileaf collimator: implications for 4D radiotherapy. Med Phys, 2005. 32(4): p. 932-41.

66. Webb, S., The effect on IMRT conformality of elastic tissue movement and a practical suggestion for movement compensation via the modified dynamic multileaf collimator (dMLC) technique. Phys Med Biol, 2005. 50(6): p. 1163-90.

67. Webb, S., Limitations of a simple technique for movement compensation via movement-modified fluence profiles. Phys. Med. Biol., 2005. 50(14): p. N155-N161.

68. Okumara, T., H. Tsuji, and Y. Hayakawa. Respiration-gated irradiation system for proton radiotherapy. in Proceedings of the 11th international conference on the use of computers in radiation therapy. 1994. Manchester: North Western Medical Physics Dept. Christie Hospital.

69. Tada, T., K. Minakuchi, T. Fujioka, M. Sakurai, M. Koda, I. Kawase, T. Nakajima, M. Nishioka, T. Tonai, and T. Kozuka, Lung cancer: intermittent irradiation synchronized with respiratory motion--results of a pilot study. Radiology, 1998. 207(3): p. 779-83.

70. Hara, R., J. Itami, and T. Kondo, Stereotactic single high dose irradiation of lung tumors under respiratory gating. Radiother Oncol, 2002. 63: p. 159-163.

71. Vedam, S.S., V.R. Kini, P.J. Keall, V. Ramakrishnan, H. Mostafavi, and R. Mohan, Quantifying the predictability of diaphragm motion during respiration with a noninvasive external marker. Med Phys, 2003. 30(4): p. 505-13.

72. Kini, V.R., S.S. Vedam, P.J. Keall, S. Patil, C. Chen, and R. Mohan, Patient training in respiratory-gated radiotherapy. Med Dosim, 2003. 28(1): p. 7-11.

73. Mageras, G.S., E. Yorke, K. Rosenzweig, L. Braban, E. Keatley, E. Ford, S.A. Leibel, and C.C. Ling, Fluoroscopic evaluation of diaphragmatic motion reduction with a respiratory gated radiotherapy system. J Appl Clin Med Phys, 2001. 2(4): p. 191-200.

74. Wagman, R., E. Yorke, E. Ford, P. Giraud, G. Mageras, B. Minsky, and K. Rosenzweig, Respiratory gating for liver tumors: use in dose escalation. Int J Radiat Oncol Biol Phys, 2003. 55(3): p. 659-68.

75. Mageras, G.S. and E. Yorke, Deep inspiration breath hold and respiratory gating strategies for reducing organ motion in radiation treatment. Semin Radiat Oncol, 2004. 14(1): p. 65-75.

76. Jiang, S.B., R.I. Berbeco, J. Wolfgang, G.C. Sharp, K. Doppke, T. Neicu, Y. Chen, P. Busse, and G.T.Y. Chen, Image-Guided Respiration-Gated Treatment. Medical Physics, 2005.

77. Onimaru, R., H. Shirato, H. Aoyama, K. Kitakura, T. Seki, K. Hida, K. Fujita, K. Kagei, T. Nishioka, T. Kunieda, Y. Iwasaki, and K. Miyasaka, Calculation of rotational setup error using the real-time tracking radiation therapy (RTRT) system and its application to the treatment of spinal schwannoma. Int J Radiat Oncol Biol Phys, 2002. 54(3): p. 939-47.

78. Kitamura, K., H. Shirato, S. Shimizu, N. Shinohara, T. Harabayashi, T. Shimizu, Y. Kodama, H. Endo, R. Onimaru, S. Nishioka, H. Aoyama, K. Tsuchiya, and K. Miyasaka, Registration accuracy and possible migration of internal fiducial gold marker implanted in prostate and liver treated with real-time tumor-tracking radiation therapy (RTRT). Radiother Oncol, 2002. 62(3): p. 275-81.

79. Kitamura, K., H. Shirato, Y. Seppenwoolde, R. Onimaru, M. Oda, K. Fujita, S. Shimizu, N. Shinohara, T. Harabayashi, and K. Miyasaka, Three-dimensional intrafractional movement of prostate measured during real-time tumor-tracking radiotherapy in supine and prone treatment positions. Int J Radiat Oncol Biol Phys, 2002. 53(5): p. 1117-23.

80. Kitamura, K., H. Shirato, N. Shinohara, T. Harabayashi, R. Onimaru, K. Fujita, S. Shimizu, K. Nonomura, T. Koyanagi, and K. Miyasaka, Reduction in acute morbidity using hypofractionated intensity-modulated radiation therapy assisted with a fluoroscopic real-time tumor-tracking system for prostate cancer: preliminary results of a phase I/II study. Cancer J, 2003. 9(4): p. 268-76.

81. Shirato, H., Y. Seppenwoolde, K. Kitamura, R. Onimura, and S. Shimizu, Intrafractional tumor motion: Lung and liver. Semin Radiat Oncol, 2004. 14(1): p. 10-8.

82. Yamamoto, R., A. Yonesaka, S. Nishioka, H. Watari, T. Hashimoto, D. Uchida, H. Taguchi, T. Nishioka, K. Miyasaka, N. Sakuragi, and H. Shirato, High dose three-dimensional conformal boost (3DCB) using an orthogonal diagnostic X-ray set-up for patients with gynecological malignancy: a new application of real-time tumor-tracking system. Radiother Oncol, 2004. 73(2): p. 219-22.

83. Shirato, H., M. Oita, K. Fujita, S. Shimizu, R. Onimaru, S. Uegaki, Y. Watanabe, N. Kato, and K. Miyasaka, Three-dimensional conformal setup (3D-CSU) of patients using the coordinate system provided by three internal fiducial markers and two orthogonal diagnostic X-ray systems in the treatment room. Int J Radiat Oncol Biol Phys, 2004. 60(2): p. 607-12.

84. Ahn, Y.C., S. Shimizu, H. Shirato, T. Hashimoto, Y. Osaka, X.Q. Zhang, T. Abe, M. Hosokawa, and K. Miyasaka, Application of real-time tumor-tracking and gated radiotherapy system for unresectable pancreatic cancer. Yonsei Med J, 2004. 45(4): p. 584-90.

85. Shirato, H., M. Oita, K. Fujita, Y. Watanabe, and K. Miyasaka, Feasibility of synchronization of real-time tumor-tracking radiotherapy and intensity-modulated radiotherapy from viewpoint of excessive dose from fluoroscopy. Int J Radiat Oncol Biol Phys, 2004. 60(1): p. 335-41.

86. Onimaru, R., H. Shirato, M. Fujino, K. Suzuki, K. Yamazaki, M. Nishimura, H. Dosaka-Akita, and K. Miyasaka, The effect of tumor location and respiratory function on tumor movement estimated by real-time tracking radiotherapy (RTRT) system. Int J Radiat Oncol Biol Phys, 2005. 63(1): p. 164-9.

87. Imura, M., K. Yamazaki, H. Shirato, R. Onimaru, M. Fujino, S. Shimizu, T. Harada, S. Ogura, H. Dosaka-Akita, K. Miyasaka, and M. Nishimura, Insertion and fixation of fiducial markers for setup and tracking of lung tumors in radiotherapy. Int J Radiat Oncol Biol Phys, 2005.

88. Hashimoto, T., H. Shirato, M. Kato, K. Yamazaki, N. Kurauchi, T. Morikawa, S. Shimizu, Y.C. Ahn, Y. Akine, and K. Miyasaka, Real-time monitoring of a digestive tract marker to reduce adverse effects of moving organs at risk (OAR) in radiotherapy for thoracic and abdominal tumors. Int J Radiat Oncol Biol Phys, 2005. 61(5): p. 1559-64.

89. Lagerwaard, F.J., S. Senan, J.P. van Meerbeeck, and W.J. Graveland, Has 3-D conformal radiotherapy (3D CRT) improved the local tumour control for stage I non-small cell lung cancer? Radiother Oncol, 2002. 63(2): p. 151-7.

90. Onimaru, R., H. Shirato, S. Shimizu, K. Kitamura, B. Xu, S. Fukumoto, T.C. Chang, K. Fujita, M. Oita, K. Miyasaka, M. Nishimura, and H. Dosaka-Akita, Tolerance of organs at risk in small-volume, hypofractionated, image-guided radiotherapy for primary and metastatic lung cancers. Int J Radiat Oncol Biol Phys, 2003. 56(1): p. 126-35.

91. Lawrence, T.S., R.K. Ten Haken, M.L. Kessler, J.M. Robertson, J.T. Lyman, M.L. Lavigne, M.B. Brown, D.J. DuRoss, J.C. Andrews, and W.D. Ensminger, The use of 3-D dose volume analysis to predict radiation hepatitis. Int J Radiat Oncol Biol Phys, 1992. 23(4): p. 781-8.

92. Jiang, S.B., R.I. Berbeco, G.T.Y. Chen, A. Jeung, H. Mostafavi, and G.C. Sharp, A Tumor Tracking System for Image Guided Radiotherapy. NIH/NCI R21 grant, 2005-2007.

93. BERBECO, r., h. MOSTAFAVI, g. SHARP, and s. JIANG. Tumor Tracking in the Absence of Radiopaque Markers. in The 14th International Conference on the Use of Computers in Radiation Therapy. 2004. Seoul, Korea.

94. Balter, J.M., J.N. Wright, L.J. Newell, B. Friemel, S. Dimmer, Y. Cheng, J. Wong, E. Vertatschitsch, and T.P. Mate, Accuracy of a wireless localization system for radiotherapy. Int J Radiat Oncol Biol Phys, 2005. 61(3): p. 933-7.

95. Sharp, G.C., S.B. Jiang, S. Shimizu, and H. Shirato, Prediction of respiratory tumour motion for real-time image-guided radiotherapy. Phys Med Biol, 2004. 49(3): p. 425-40.

96. Murphy, M.J., J. Jalden, and M. Isaksson. Adaptive filtering to predict lung tumor breathing motion during image-guided radiation therapy. in Proc. 16th Int. Conf. on Computer Assisted Radiology (CARS 2002). 2002.

97. Vedam, S.S., P.J. Keall, A. Docef, D.A. Todor, V.R. Kini, and R. Mohan, Predicting respiratory motion for four-dimensional radiotherapy. Med Phys, 2004. 31(8): p. 2274-83.

98. Berbeco, R.I., S.B. Jiang, G.C. Sharp, G.T. Chen, H. Mostafavi, and H. Shirato, Integrated radiotherapy imaging system (IRIS): design considerations of tumour tracking with linac gantry-mounted diagnostic x-ray systems with flat-panel detectors. Phys Med Biol, 2004. 49(2): p. 243-55.

99. Sharp, G.C., S. Kollipara, T. Madden, S.B. Jiang, and S. Rosenthal, Anatomic feature-based registration for patient setup in head and neck cancer radiotherapy. Phys Med Biol, 2005. to be published.

100. Berbeco, R.I., T. Neicu, E. Rietzel, G.T. Chen, and S.B. Jiang, A technique for respiratory-gated radiotherapy treatment verification with an EPID in cine mode. Phys Med Biol, 2005. 50(16): p. 3669-79.

101. Pan, T., T.Y. Lee, E. Rietzel, and G.T. Chen, 4D-CT imaging of a volume influenced by respiratory motion on multi-slice CT. Med Phys, 2004. 31(2): p. 333-40.

102. Rietzel, E., S.J. Rosenthal, D.P. Gierga, C.G. Willet, and G.T. Chen, Moving targets: detection and tracking of internal organ motion for treatment planning and patient set-up. Radiother Oncol, 2004. 73 Suppl 2: p. S68-72.

103. Rietzel, E., G.T. Chen, N.C. Choi, and C.G. Willet, Four-dimensional image-based treatment planning: Target volume segmentation and dose calculation in the presence of respiratory motion. Int J Radiat Oncol Biol Phys, 2005. 61(5): p. 1535-50.

104. Rietzel, E., T. Pan, and G.T. Chen, Four-dimensional computed tomography: image formation and clinical protocol. Med Phys, 2005. 32(4): p. 874-89.

105. Neicu, T., R.I. Berbeco, J. Wolfgang, and S.B. Jiang, Synchronized Moving Aperture Radiation Therapy (SMART): Investigation of the breathing pattern reproducibility under respiratory coaching. Phys Med Biol, 2005. submitted.

106. Trofimov, A., E. Rietzel, H.M. LU, B. Martin, S.B. Jiang, C.T.Y. Chen, and T. Bortfeld, Temporo-spatial IMRT optimization: Cocepts and first results. Phys Med Biol, 2004. submitted.

107. Gierga, D.P., G.T. Chen, J.H. Kung, M. Betke, J. Lombardi, and C.G. Willett, Quantification of respiration-induced abdominal tumor motion and its impact on IMRT dose distributions. Int J Radiat Oncol Biol Phys, 2004. 58(5): p. 1584-95.

108. Betke, M., J. Ruel, G.C. Sharp, S.B. Jiang, D.P. Gierga, and C.T.Y. Chen, Tracking and prediction of tumor movment in the abdomen. International Journal of Radiation Oncology, Biology, and Physics, 2005(submitted).

109. Reid, D.B., An algorithm for tracking multiple targets. IEEE Transactions on Automatic Control, 1979. AC-24(6)(6): p. 843-854.

110. Bar-Shalom, Y. and T.E. Fortmann, Tracking and Data Association. 1988: Academic Press.

111. Sharp, G.C., S.B. Jiang, S. Shimizu, and H. Shirato, Tracking errors in a prototype real-time tumour tracking system. Phys Med Biol, 2004. 49(23): p. 5347-56.

112. Berbeco, R.I., H. Mostafavi, G.C. Sharp, and S.B. Jiang, Towards fluoroscopic respiratory gating for lung tumours without radiopaque markers. Phys Med Biol, 2005. 50(19): p. 4481-90.

113. Berbeco, R.I., S. Nishioka, H. Shirato, G.T. Chen, and S.B. Jiang, Residual motion of lung tumours in gated radiotherapy with external respiratory surrogates. Phys Med Biol, 2005. 50(16): p. 3655-67.

114. Lujan, A.E., E.W. Larsen, J.M. Balter, and R.K. Ten Haken, A method for incorporating organ motion due to breathing into 3D dose calculations. Med Phys, 1999. 26(5): p. 715-20.

115. Wu, H., G.C. Sharp, B. Salzberg, D. Kaeli, H. Shirato, and S.B. Jiang, A finite state model for respiratory motion analysis in image guided radiation therapy. Phys Med Biol, 2004. 49(23): p. 5357-72.

116. Jiang, S.B., Compensating for Tumor Motion Using A Computer-Controlled Multi-leaf Collimator in Radiation Therapy. Whitaker Foundation Research Grant, 2001-2004.

117. Sharp, G.C., S. Kollipara, T. Madden, S.B. Jiang, and S.J. Rosenthal, Anatomic feature-based registration for patient set-up in head and neck cancer radiotherapy. Phys Med Biol, 2005. 50(19): p. 4667-79.

118. Freedman, D., R.J. Radke, T. Zhang, Y. Jeong, D.M. Lovelock, and G.T. Chen, Model-based segmentation of medical imagery by matching distributions. IEEE Trans Med Imaging, 2005. 24(3): p. 281-92.

119. Breiman, L., Bagging Predictors. Machine Learning, 1996. 24(2): p. 123-140.

120. Vapnik, V., Statistical Learning Theory. 1998, New York: Wiley.

121. Quinlan, J.R., Induction of decision trees. Machine Learning, 1986. 1(1): p. 81-106.

122. Duda, R.O., P.E. Hart, and D.G. Stork, Pattern Classification (Second Edition). 2001, NY: Wiley & Sons.

123. Bishop, C.M., Neural Networks for Pattern Recognition. 1995: Oxford University Press.

124. Murase, H. and S. Nayar, Visual Learning and recognition of 3-D objects from appearance. International Journal of Computer Vision, 1995. 14: p. 5-24.

125. Black, M.J. and A.D. Jepson, EigenTracking: Robust Matching and Tracking of Articulated Objects Using a View-Based Representation. International Journal of Computer Vision, 1998. 26(1): p. 63-84.

126. Sista, S., C.A. A. Bouman, and J.P. P. Allebach, Fast image search using a multiscale stochastic model, in IEEE International Conference on Image Processing. 1995: Washington, DC. p. 22-25.

127. Turk, M. and A. Pentland, Face recognition using eigenfaces, in Proceedings in Computer Vision and Pattern Recognition. 1991: Maui. p. 586-591.

128. Jolliffe, I.T., Principal Component Analysis. 2002: Springer.

129. McLachlan, G.J. and K.E. Basford, Mixture Models, Inference and Applications to Clustering. 1988, New York: Marcel Dekker.

130. Dy, J.G. and C.E. Brodley, Feature selection for unsupervised learning. Journal of Machine Learning Research, 2004. 5: p. 845-889.

131. Tipping, M.E. and C.M. Bishop, Mixtures of probabilistic principal component analysers. Neural Computation, 1999. 11(2): p. 443-482.

132. Su, T. and J.G. Dy, Automated Hierarchical Mixtures of Probabilistic Principal Component Analyzers, in Proceedings of the Twenty-First International Conference on Machine Learning. 2004: Banff, Alberta, Canada. p. 775-782.

133. Schwarz, G., Estimating the Dimension of a Model. The Annals of Statistics, 1978. 6(2): p. 461-464.

134. Rabiner, L.R., A tutorial on Hidden Markov Models and selected applications in speech recognition. Proceedings of the IEEE, 1989. 77(2).

135. Dempster, A.P., N.M. Laird, and D.B. Rubin, Maximum likelihood from incomplete data via the EM algorithm. Journal Royal Statistical Society, Series B, 1977. 39(1): p. 1-38.

136. Cohn, D., L. Atlas, and R. Ladner, Improving generalization with active learning. Machine Learning (Historical Archive), 1994. 15(2): p. 201-221.

137. Cohn, D.A., Z. Ghahramani, and M.I. Jordan, Active Learning with Statistical Models. JOURNAL OF ARTIFICIAL INTELLIGENCE RESEARCH, 1996. 4: p. 129.

138. Zhu, X., Semi-Supervised Learning Literature Survey, in Computer Sciences TR 1530. 2005, University of Wisconsin-Madison.

139. Zhu, X., J. Lafferty, and Z. Ghahramani, Combining Active Learning and Semi-Supervised Learning Using Gaussian Fields and Harmonic Functions, in International Conference on Machine Learning 2003 workshop on The Continuum from Labeled to Unlabeled Data in Machine Learning and Data Mining. 2003.

140. Tao, Y., C. Faloutsos, D. Papadias, and B. Liu, Prediction and indexing of moving objects with unknown motion patterns, in Proceedings of the 2004 ACM SIGMOD international conference on Management of data. 2004, ACM Press: Paris, France. p. 611-622.

141. Siddon, R.L., Fast calculation of the exact radiological path for a three-dimensional CT array. Medical Physics, 1985. 12: p. 252-255.

142. Killoran, J.H., E.H. Baldini, C.J. Beard, and L. Chin, A technique for optimization of digitally reconstructed radiographs of the chest in virtual simulation. Int J Radiat Oncol Biol Phys, 2001. 49(1): p. 231-9.

143. Siewiorek, D.P. and R.P. Swarz, Reliable Computer Systems Design and Evaluation. 3rd ed. 1998, Natick, MA: AK Peters.

144. Pressman, R.S., Software Engineering, A Practitioner's Approach. 2nd ed. 2004: McGraw-Hill.

-----------------------

|[pic] |

|[pic] |

|[pic] |

|Figure x. Estimated internal tumor position using: (top) internal signal |

|(3Hz) alone, (middle) internal signal (3Hz) and external signal (30Hz), and|

|(bottom) external signal (30Hz) alone. |

|[pic] |Figure x. The internal |

| |(treatment) duty cycle as a |

| |function of external (imaging)|

| |duty cycle for double gating, |

| |computed using measured |

| |internal and external marker |

| |data. |

|[pic] |

| |

|Figure 2. The software infrastructure provides services for gated radiotherapy through a hardware abstraction layer, which can be |

|used on the IRIS physical hardware (left) or a virtual IRIS machine (right). |

|[pic] |

|Figure x. A block diagram showing a general scheme of combining external |

|signal with internal signal for gated radiotherapy. |

|[pic] |[pic] |

|(a) |(b) |

|[pic] |

|(c) |

|Figure 4. The finite state model for tumor respiratory motion: (a) three |

|states of a regular breathing cycle, (b) finite state automation for |

|respiratory motion, (c) raw and modeled respiratory motion with irregular |

|breathing. |

|[pic] |[pic] |

|Figure 3. Cine EPID image (right) is compared with the DRR image (left) for |

|the same beam to determine the residual marker motion within the gating |

|window. |

|[pic] |

|[pic] |

|Figure 2. The breath waveforms of a lung cancer patient, treated with |

|gated radiotherapy, with and without breath coaching using a protocol |

|developed at MGH. The two dashed lines define a gating window. |

|[pic] |[pic] |

|Figure x. Left: the correlation score (in gray scale) as functions of |

|template ID (y-axis) and the measured image frame ID (x-axis). Right: the |

|estimated tumor position as a function of time. |

|[pic] |

|[pic] |

|Figure x. Twelve motion-enhanced tumor templates built by averaging the |

|images in ROI (as shown in figure x) falling in the same bin. |

|[pic] |

|Figure x. The average intensity in the ROI shown in figure x as a function of|

|time (top row). The first 12 seconds (as indicated by the vertical line) is |

|used to simulated setup session while the rest images are used to simulated |

|treatment. The image intensity threshold is determined to have 35% gating |

|duty cycle in the setup session. The threshold for correlation score (middle |

|row) is automatically translated from the intensity threshold and then used |

|for generating gating signal (bottom row) for the treatment session. |

|[pic] |[pic] |

|Figure x. A frame of setup fluoroscopic images. Left: original image. Right:|

|motion-enhanced image. The selected ROI is shown in both images as a |

|rectangular region containing tumor positions of all phases. |

| |4D CT |DRF |Fluoro |

|0% |[pic] |[pic] |[pic] |

|30%|[pic] |[pic] |[pic] |

|50%|[pic] |[pic] |[pic] |

|Figure x. Coronal slices of 4D CT data, AP DRFs, and AP fluoroscopic |

|images at various breathing phases: 0% (inhale), 30%, and 50%(exhale). The|

|GTV contour of EOE is also shown in the CT slices. |

|[pic] |

|Figure x. Same as figure x, except for a different patient. |

|[pic] |Figure 1. A dual-head on-board |

| |x-ray imaging system, called |

| |Integrated Radiotherapy Imaging |

| |System (IRIS), developed at MGH. |

| |The system consists of two x-ray |

| |tubes, two flat panels, and two |

| |x-ray generators. |

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download