Correction of Motion Artifacts in CT images Using Motion ...
SENIOR DESIGN PROJECT 2017University of Central Oklahoma08FallCorrection of Motion Artifacts in CT images Using Motion Modeling ____________________________________________________________KushalThapa, AbdulRahman Alqahtani, YahyaHadadi, Mohammed Alhaseem and Michael MartinezFaulty advisor- Dr. NesreenAlsbou from the Department of Engineering and PhysicsOther advisors- Drs. Imad Ali and Salahuddin Ahmad from the Department of Radiation Oncology at the StephensonOklahoma Cancer Center, University of Oklahoma Health Sciences Center 5/2/2017IntroductionComputed Tomography imaging (CT) is one of the most common medical imaging modality in the modern world of medicine. Despite being implemented nearly for five decades, this imaging modality, due to its enormous scope and a wide range of advantages, is the household tool of hospitals and medical centers around the world. Since its development in the early 1970’s, its applications are continuously evolving as well. CT imaging has proved to be a vital tool for diagnostic imaging of different diseases particularly cancer. CT imaging plays an increasing role in the screening and staging of different cancers. The prognosis of tumor is determined by stage of the disease, the size of the primary tumor and histologic grade. In radiotherapy, CT images are used to outline the tumors and critical structures required for treatment planning. Even over this long time period, new technical advancement and development are trying to improve image quality and speed the process of CT imaging. There are several drawbacks that affect the quality of CT-images including geometric distortions, noise and metal and motion artifacts. Motion artifacts are produced in CT images due to the motion of the patients during the scanning process. These kinds of artifacts usually have a significant effect especially if those CT are being used for radiation therapy. Patient motion may cause substantial image artifact in CT imagesparticularly in lung, abdominal and pelvictumors. The degradation of image quality by motion leads to variations in the position, shape and size of tumors. CT images are obtained from different methods such as axial or helical CT imaging where motionartifactsvary depending on the imaging techniques while some techniques can reduce these artifacts, they cannot eliminate the artifacts completely. The goal of our project is to intervene and evaluate quantitatively the performance of different deformable image registration (DIR) algorithms on CT images obtained from imaging a mobile phantom with controlled motion patterns. We are using research software called DIRART (Deformable Image Registration and Adaptive Radiotherapy Treatment) developed at the University of Washington in Saint Louis. This software has several algorithms to match up two or more CT image sets so that the artifacts will, theoretically, differentiate away leaving the image without any errors. Our goal is to study these algorithms, understand the process of registration, evaluate quantitatively the performance of several selected algorithms and make changes to the image registration process so as to optimize error reduction in the final CT image. CT (Computed Tomography)HistoryThe prototype of the first medical CT scanner was constructed by Godfrey Newbold Hounsfield in 1972. For his accomplishments in an invention of CT and medical imaging, he was awarded Nobel Prize in 1979 for Physiology and Medicine. He is also considered the father of computed tomography and unit of attenuation coefficient (CT number) is named after him (Hounsfield unit). The scope and applications of CT imaging systems was quickly recognized beyond medical community, and in 1980’s, industries began adopting this technology for non-destructive product testing and quality check. PrincipleA basic CT consists of an X-ray source, which emits X-ray onto an object set still or in a rotary table and an X-ray detector behind the object to obtain projection images. Then the data collected in the X-ray detector goes to a data processing unit for computation, visualization and data analysis of the results. Individual CT images are similar to 2D X-ray images; however, CT merges multiple cross section images through multiple planes of an object from angle positions performing one revolution, which creates a 3D image. As the X-rays pass through the object, some of them are absorbed or scattered while some are transmitted. The X-ray intensity in detector screen is reduced because of scattered or absorbed X-rays. This process is called attenuation. The transmitted photons are collected on the detector screen and visualized by computer, creating a complete reconstruction of the scanned object.Fig 1: Basic components and working of CT(Source: Cantator, A and Muller, P, Introduction to Computed Tomography, March 2011, Technical University of Denmark)Types of CT scannersCT scanners can be broadly categorized into two systems: 1) 2D-CT 2)3D-CT. 2D-CT has a fan beam source and a line detector which enable the acquisition of a slice of a 3D object by coupling a translation and rotation movement of the object. The drawback of this system is long scanning times, however, it is useful is some industrial systems. The 3D-CT systems consist of a flat area detector and a cone beam source which enables the acquisition of a slice of the object just with one revolution of the rotary table. The drawback of this system is that the scanning quality deteriorates from the center to the borders of the detectors. The main difference between the medical and industrial CT scanners is that the object rotates on a rotary table and the X-ray source is steady in the last one. While in medical CT, the X-ray source rotates and the object (usually patient) is steady. Medical CT’s are further categorized into five generations depending on the scanner geometry. They are illustrated in the figure below.Fig 2. Basic illustration of the principle of five generations of medical CT scanners.(Source: Cantator, A and Muller, P, Introduction to Computed Tomography, March 2011, Technical University of Denmark)ApplicationsBesides its broad use in Medicine, CT scanners, as mentioned above, are widely used in industries for non-destructive analysis of faults and the material composition or density inside the volume of the scanned parts. They are helpful in both quantitative and qualitative analysis of materials. CT scanners are also used in military, naval and aerospace technologies. Their widespread use in security (like airport check in) is apparent to all of us. CT systems are also very popular in archeology because of its ability to scan fragile objects inside and out without doing any harm to the object itself. Advantages and disadvantages The main advantage of CT over other scanning machines in medical field is its scanning time. Modern CT scanners are very fast and they are able to image bone, soft tissue and blood vessels all at the same time. It is cheaper than some other imaging modalities like MRI and PET. However, it uses X-rays radiation, which can cause medical problems if used for a long period. CT images also have less soft tissue differentiation than some other modalities. The reconstruction of the image after the scan can also have some problems. Since the object being scanned is a mobile system (humans), the CT images can be blurred somewhat which are called motion artifacts. There are some techniques to remove these motion artifacts; however, none of them are perfect. The goal of our project is to zoom in to this problem of motion artifacts, evaluate the existing techniques that deal with these problems and hopefully come up with a new and better way to reduce these artifacts in the CT images.Deformable Image Registration (DIR)One of the techniques to deal with the image artifacts caused by motion is deformable image registration. Image registration is the procedure of transforming diverse image sets into one coordinate system where the objects or organs are matched anatomically. When acquiring CT images, a sequence of images of a slice is taken from different angles and each image differs from the others. The difference in the geometric positioning is normal since the angles are different and these differences in the images can be adjusted using simple geometric transformation. However, human body is a mobile system and the interior movement, like breathing and heartbeat, can cause the tissue being scanned to shift, expand or shrink which cannot be resolved by spatial transformation techniques. Thus non-rigid or deformable image registration is required. There are various algorithms, which can perform deformable or non-rigid transformations and thus are useful in deformable image registration. Some of the main ones are described below:1) Demon’s algorithm:Demon’s algorithm considers the object boundary in one image as semi-permeable membrane and lets the other image diffuse through these interfaces (Thirion, 1998), by the action of effectors (or demons) within the membranes. It uses optical flow method, which considers the intensity of moving object to be constant over time. If ‘s’ is the intensity of the static image and ‘m’ is the intensity of moving the object and ‘v’ is the velocity vector of the moving image, then, demon’s algorithm implies: v=m-s?s?s2+m-s22) Horn-Schunck method:The Horn-Schunck is a method that solves the variations in the intensity of a sequence of images usingthe optical flowequation.The displacements of the voxels are assumed to be small and intensity of the image is nearly constant. The optical flow equation is solved using a global smoothness term for entire optical flow over the whole image. 3) Lucas-Kanade method:The Lucas-Kanade algorithm is a technique that can offer estimation for the displacement movement in other image by using the pixel. Therefore, we can assume by using the equation in solving an optical flow in other pixel proximity. So, we can collect the information and combine for each nearby pixel and look for the pixel intensity changing. The Lucas-Kanade work best for slow moving object because the small time increasing between the images. In addition, The Lucas-Kanade flows in which the direction the object moves, so the changing in intensity can be justified.Ixq1Vx+Iyq1Vy=-Itq1Ixq2Vx+Iyq2Vy=-Itq2IxqnVx+IyqnVy=-Itqn∴ Where q1,q2,…,qn are the pixels and Ixqi,Iyqi,Itqi are the partial derivation of I with respect to position (x , y) , and time t.4) Fast Demon’s method:Fast Demon’s or accelerated demon’s is extends the demon’s algorithm with an additional ‘active force’ using the intensity gradient of the moving image. The Demon’s algorithm only considers the gradient driven by the static image while fast demon’s considers the diffusion process is bidirectional and thus moving image has effect on the gradient too. Thus a reaction force is necessary in addition to the passive demon force, which is given by: v=s-m?m?m2+s-m2Then, the total force will be:f=fs+fm=(m-s)×(?s?s2+s-m2+?m?m2+s-m2)5) Symmetric Force Demons Method:In this algorithm, image gradient points to the densest region. The gradient indicates color intensity difference. The image gradient information is used to extract a force field. In the Demons algorithm, image gradient information from 1 image computes force field at each iteration, whilethe Symmetric Forces Demons compare moving image gradient instead of the reference image.6) Iterative Optical Flow Method:The Optical Flow algorithm estimation approximates the motion direction (flow vector) for each pixel between 2 frames at t and t+1. Iterative optical flow method defines frame 1 at t and frame 2 at t+1. Iterative algorithm- SWIFT-Flow computes optical flow for image frame. Flow vector for each pixel in image is predicted in SWIFT-flow method. 7) Other iterative methods, Free form deformation method Lu:This one of the fasts algorithms in optical flow methods because it takes only 3 minute to do the registration. This method is used to measure the smoothness and similarity between two images.The vectors u (x, t) are the displacement vectors and displacement field is u.8) Demons methods, Double force Demons:This algorithm allow the length displacement vector to be fix in every iteration by ∝. It makes the factor normalized to the original demons algorithm. Is(0) and Im(0)areknown for the moving image. The demons force algorithm dr(n+1)=(Imn-Is0)?Is0(Imn-Is0)2+|?Is0|2The double demons algorithm dr(n+1)=(Imn-Is0)?Is0∝2+(Imn-Is0)2+|?Is0|2+Imn-Is0?Imn∝2(Imn-Is0)2+|?Imn 9) Original level set motion: The level set method introduced in 1987 by Oshar and Sethin for catching motion. It is a good numerical device in image science linking curves and surfaces on grids. Generally, the image science here discusses problems in image processing,computergraphics,andcomputervision. Also, this method is easy in flowing shape changes.10) Affine approximation of level set motion:The affine approximation of level set motion uses basically the general linear affine approximation function. It is applied on the level set motion algorithm approximating the result to solve the motion equation using first orders methods.DIRART (Deformable Image Registration and Adaptive Radio Therapy)In 2006, Dr. Deshan Yang and his colleagues from University of Washington in St. Louis developed the DIRTART software, which has more than 20 different image registration algorithms.. The scope of this software goes beyond just doing DIR to perform adaptive radiation therapy (ART); however, our project here is concerned with this specific functionality.DIRART is collection of DIR algorithms andit has different visualization and validation features with ART toolkit to perform dose and structure remapping, dose accumulation and analysis using the DIR results. It is a complimentary package to CERR (A Computational Environment for Radiotherapy Research) to provide additional DIR and ART functions. This software interacts with different treatment planning systems using DICOM‐RT, which is a standard for the viewing and distributing medical image files, (via CERR) to compute deformation between scans, apply deformation for planning adaptation purposes, daily dose deformation, accumulation and comparison as well as visualizes and analyzes the results. Fig 3. Model of interaction from CERR to DIRARTFig 4. A summary of working of DIRARTProjectProject SetupA mobile thorax phantom with well-known targets with different sizes that is made from water-equivalent material and inserted in foam was used to simulate lung lesions. The thorax phantom was imaged with helical, axial and cone-beam CT. The phantom was moved with a cyclic motion with different motion amplitudes and frequencies along the superior-inferior direction. Different deformable image registration algorithms including the demons method, fast demons method, Horn-Schunck and other iterative optical flow algorithms from the DIRART software was used to deform CT images for the phantom with different motion patterns. The CT images of the mobile phantom were deformed to the corresponding CT images of the stationary phantom. DeliverablesThe project will deliver the following: The values of the displacement vector fields will be calculated for different deformable image registration algorithms. A new approach will be developed to extract the motion parameters of mobile targets from helical, axial and cone-beam CT images for mobile phantoms.The performance of different deformable image registration algorithms will be evaluated. Testing new deformable image registration algorithms using motion modeling.Measure of successThe project measures of success include the following: Establishing correlation between the displacement vector fields and motion parameters.Investigation of the performance of the different deformable image registration algorithms and identifying the one with the best performance for the different targets and CT imaging techniques.Extraction of motion parameters form helical, axial and cone-beam CT images of mobile phantoms that simulate patient motion. Development of deformable image registration algorithms based on patient motion modeling.Results of Algorithm Analysis in DIRART 1) Time Analysis of the DIRART algorithms registration timeFigure 4: Time analysis of the algorithms: This graph shows the time required by each algorithm in order to perform DIR for CT images with different motion amplitudes. The figure shows that the Affine “Approximation of Level Set Motion” algorithm took the most amount of time while “Free form Deformation” algorithm took the least amount of time. The time needed by each algorithm to perform deformation in itself isn’t the best measure of efficiency of algorithms but it importantof the speed for the different algorithms.2) Coronal View (registration only):Fig 5&6: Coronal view of the mobile targets from the cone-beam CT images. The first row represents the unedited, original CT images of the moving phantom. The motion amplitudes are indicated on the top and ascend left to right. The rows following this, show the CT-images after DIR with algorithms mentioned on the left side. This figure shows the performance of the different algorithms.These views show qualitative evaluation of the performance of the different DIR algorithms.A trend of more blurring in images is seen from left to right in all algorithms because of the increasing motion amplitudes. 3) Profiles (registration only):5mmFig 7: CT number vs. Pixel graph of big target (two figures on the left), and medium and small target (two figures on the right). The black profilerepresent the CT-number distribution for the static image while the other profiles are the CT-number distributionfrom the CT images of phantom that moved sinusoidally with amplitude of 2.5mm. The top two figures show the CT profiles of five algorithms (mentioned in the index) and the bottom two figures represent another set of five DIR algorithms. We didn’t do all ten algorithms in one graph because of the resultant clarity of the graph and limitations the code we used to do profiling.Therefore, these profiles show quantitatively the variations in the CT-number values from the different DIR algorithms. The figure shows how well each algorithm managed to trace the blurry edge of eachmobile target in the images and tried to map it back to the artifact-free -static image. Most of the algorithms were able to reproduce the shape and CT-number level of the large and medium targets. However, the CT-number level was not reproduced for the small target these algorithms.The CT-number profiles shows that the Demon’s algorithm clearly has some problems because the image is visibly shifted; Original Horn-Schunck algorithm appears to be the best for all size targets.10mmFig 8: The graphs above are the profiles for the medium and small target (left two figures) and large target (right two figures) for the 10mm amplitude image. Like above, the black line is the static profile of CT numbers and the other lines are obtained from the indicated DIR algorithms. The profiles are more spread out for this motion amplitude than the smaller (2mm) amplitude because of strong image artifacts. That is expected because of larger motion. It is also apparent from these graphs and the ones above, which tracing back the small target is the hardest because of the high gradient in the CT-number distributions. Most of the algorithms were not able to reproduce the shape and CT-number level of the small, medium and large targets. Other inferences from these graphs are: Demons (red on the top two) still shows significant shift, there is no useable algorithm to reconstruct the small target, Free Form Deformation is best for the medium target and Iterative Optical Flow is best for large target. 20mmFig 9: CT-number profiles of medium and small target (left two figures) and large target (right two figures) for the 20mm motion amplitude. It is apparent from the left two graphs that the deformation of the images of the mobile targets with motion artifacts was unsuccessful in the medium and small targets by any algorithms. For the large target, original Horn Schunck came the closet to reproduce the shape of the larger target in the static image. Software Component: MATLAB Deblurring Code:MATLAB Code description:The description for the code used for deblurring process has been described below. The description order follows the order of the code and hence the execution of the code. Prior to running this code we already have a 3D 512x512x81 (suppose axbxc) vector loaded, which is named ‘img1’. Illustration of the general makeup and the specific views of a stationary 3D image have been included too for better understanding of the steps in the code.80010073660Fig 10: Representation of the planes of view of CT image Fig 11: Coronal view (81x512 or cxb)Fig 12: Sagittal view (81x512 or cxa)Fig 13: Transverse view (512x512 or axb)J=zeros(81,512); A 2D 81x512 vector with zero as its all elements is formed and is assigned name ‘J’. This is done to initalize a vector ‘J’ so that it can be manipulated in the for-loop below. The exact use of this vector can be seen.fori=1:1:512 Main for loop is started. The loop goes from i=0 to i=512 with increment of 1.I=squeeze(img1(i,:,:)); Initializing a new variable ‘I’. In this step, ‘I’ is a vector corresponding to the squeeze of ‘img1’. Squeeze function slices off ‘img1’ which is a axbxc vector into ith of ‘a’ bxcπvector. So every time the loop is started from i=1, this step takes out 512x81 vector corresponding to a=1, a=2…….a=512. Every slice is then a coronal image.I=permute(I,[2 1]);Permute function rearranges the vector in the order defined. In this case, this function changes I which was a 512x81 vector from above step in to 81x512 vector and reassigns this vector into variable ‘I’. This is necessary to align the coronal image in right orientation because the deblurring steps below works in only one direction.T=4;This is the time period of the one human breath cycle. Normal breathing rate ranges from 12-20 breaths per minute. In this case, we took the lower range of that i.e. 12/min which gives us one cycle of breath (i.e. inspiration and expiration) in 4secs. This is the rate that corresponded to the phantom movement in our CT imaging experiments. However, this can be varied and can be made patient specific. f=1/T; w=2*pi*f; In this step, we define the frequency of motion, which is given by the inverse of time period defined above and we also define angular frequency ‘w’ which is given by w=2πf.A=[2.5 5 7.5 10 12.5 15 17.5 20]; Here, we define the motion amplitudes. We used the above set of motion amplitudes in our experiment so we included all of those in this vector.for n=1:8 Nested for loop is started where we define n which goes from 1 to x. x is the number of elements in the vector ‘A’ defined above.yy=1:1:A(n)-1;Defining ‘yy’ for nested for loop.p=(2/(T*w)).*((1-((yy(:)).^2)./(A(n))^2).^-0.5); Defining probability function. For more information on this function, refer to # in refrences.N=sum(p(:)); This gives the normalization factor. For more information on this function, refer to # in refrences.P{n}=p./N; This gives the normalized probability function. For more information on this function, refer to # in refrences.endThis ends the nested for-loop. The main point of this nested for-loop is to define a probability density function. Probability density function determines the charactersitics of motion of the object in the blurred image. We now use this information to deblurr the image with the motion artifact.I = edgetaper(I,P{8}); This function tapers the edges of the image using above defined probability density function. In this case P{8} is used because the image we were trying to deblur had motion amplitude of 20mm. Corresponding n in P{n} should be used for other motion amplitudes. The new image is reassigned as ‘I’.Edgetaper function blurs the edges of image I using the probability densityfunctionP{}. The output image is the weighted sum of the originalimageI and its blurred version. The weighting array, determined by the autocorrelation function of P{}, makes outputimage equal to input image in its central region, and equal to the blurred version of I near the edges. The edgetaper function reduces the ringing effect in image deblurring methods that use the discrete Fourier transform, such as DECONWNR (Wiener Deconvolution) which we have used below.estimated_nsr = 0.05; NSR is noise to signal ratio in the blurred image. There are various ways to figure out this ratio if the parameters of the imaging are known. In this case, we used experimentally determined value of nsr(=0.05). This value gave the best output with least noises after deconvoultion.I= deconvwnr(I, P{8}, estimated_nsr); This the actual step of deblurring. We used Weiner’s deconvolution filter to deblur because this function uses the parameters that we knew or could be estimated, and gave the best results. In MATLAB, deconvwnr(I,PDF,nsr) does the weinerdeconvolution in Image ‘I’ with probability density funciton ‘PDF’ and noise to signal ration of ‘nsr’.J=cat(3,I,J); Until this point, we have an vector image ‘I’ that is 81x512. Now as we go through the loop we want to save each of the deblurredslice and then stack them on top of each other to get a 3D image. We use function ‘cat’ to do this. Cat is short for concatenate. ‘cat(x,I,J)’ concatenate or stacks images I and J in the x dimension. In our case, it stacks I and J which are both 81x512 vectors in third dimension giving 81x512x(i+1). Thus at the end of the loop we will get a vector J that is 81x512x513. endNow we end the main loop. The end-product of this loop will be a 81x512x513 (cxbxa)vector that is the deblurred version of img1.Mo=permute(J,[3 2 1]); The product of the for-loop above gives us a 3D vector but its not in the original configuration of axbxc. Thus, we use function permute again to change ‘J’ into 513x512x81 vector.Note: Even though the end-vector ‘Mo’ has one more stack of elements than the original one (i.e. 513 in ‘a’ as opposed to 512), we don’t need to revert back to 512 in ‘a’ because we can do the same process without deblurring for stationary image and use two 513x512x81 images in DIRART to do registration.Results after Deblurring:1) Profile Diagrams:a. 5mmFig. 14: Profile Diagrams for large target (left) and medium and small target (right) for Demons (red), fast demons (bright green), Iterative Optical Flow (dotted red) and Original Level Set Motion (Neon Blue) after deblurring process as compared with stationary (black). The profiles are shifted but show great improvement on the preservation of CT number (amplitude) which means they the size of the targets were well recorded. b. 10mmFig 15: Profiles Diagram with similar specifications as above for 10mm amplitude motion. Still, the amplitude preservation from the stationary one was commendable.c. 15mmFig. 16: Profile Diagram for 15mm amplitude motion CT images after deblurring and registration with four algorithms with same specifications. There is no apparent discrepancy in amplitude (CT numbers) of the registered images as compared to the stationary one.d. 20mmFig 17: Profile Diagram for the 20mm amplitude CT images after being deblurred and registered with different algorithms. No loss in amplitude shows that deblurring works for all motion amplitudes unlike registration only, which becomes ineffective in larger amplitudes.2. Coronal Diagrams:Fig 18: Coronal Views of the CT images of lung phantom of different motion amplitudes after deblurring and registering with labeled algorithms. As can be seen above, demons and fast demons algorithm images had shift in the targets; however, the intensity of targets (corresponding to the density) remainedpreserved from the stationary ones. There was also no blurriness even in the high amplitude original motion images. Affine and Original Level Set produced almost perfect replica of the stationary ones. Comparison between registration only and deblurred+registration images:1) Profile Analysis:a. Comparing fast demons before and after deblurringFig 19: Profile Diagram for original 20mm amplitude CT image after registering with fast demons only (red) and, after deblurring and registering with fast demons (bright green). The first two peaks correspond to the middle and small target while the peak on the right correspond to large target. After deblurring, the image profile followed the stationary profile better than when not deblurred as can be seen in the picture. This proves the effectiveness and importance of deblurring before registration.b. Comparing 20mm motion image before and after deblurring using four algorithmsFig 20: Profiles for small and medium targets, left: by registered with four labeled algorithms and right: by deblurring and then registering with same four algorithms. In the left image, the profiles for different algorithms come nowhere close to mimicking the stationary profile (black) indicating the DIR wasn’t successful. On the right the profiles have same shape and amplitude but are shifted from one another. So the DIR worked at least partially after deblurring.Fig 21: Similar profile and comparison as above except for large target. This proves deblurring works in all size of targets.2) Coronal Comparison:a. For Fast DemonsFig 22: Top left: Coronal CT image of lung phantom moved with 20mm motion during scanning. Top right: Same image after registering with fast demons algorithm in DIRART. Bottom left: Coronal image of the original 20mm image after deblurring with our code. Bottom right: Coronal image after registering the deblurred image using fast demons algorithm. This shows the effectiveness of deblurring specially when combined with DIRART registration afterwards. b. Between working of demons and fast demons before and after registration:Fig 23: Coronal comparison of the CT images with different degree of motion errors when registered through demons and fast demons (row 2 and 3) and when these images were deblurred first and then registered through demons and fast demons (row 4 and 5). This goes to show how much deblurring as image pre-processing helped the algorithms in DIRART to do successful DIR.Hardware Component: 3D Respiratory Gating PlatformMechanical Parts: The Platform is designed to move in 3 directions x, y and z-axis. The whole platform was made of polyethylene (C2H4) as it mainly contains water to avoid any overlap or interference with the x-rays. The biggest challenge in the designing was to avoid any metals in the phantom board and in the same time can hold up to 34 lbs. The design of the base has four motors in the four corners, each motor can hold up to 40 lbs. The base was a high dense 0.5-inch polyethylene as to be rigid enough to hold the weight. On the top of the base, there are two boards for the x and y axis motion. Every board is controlled with two motors. The first board is moving in the x-axis and the main board that will hold the phantom is moving in the y-axis. The phantom board dimensions are (26" x 13" x 11”).Fig 24: View from the top Fig 25: DesignComponent:1-L16-R Miniature Linear Servos for RC &Arduino:These linear actuators use the 3 wires connector, ground power and control. Our L16 linear actuator is constructed using an anodized aluminum shaft, metal gearbox, and steel ball bearings. The L16-R has a mass of between 56g-84g.?It is light weight makes the L16-R ideal for applications where we need a higher force.The L16-R line includes three unique models featuring:Stroke lengths of 50mm3 gearing options for maximum forces between 200N (45lbs) and maximum speeds 8mm/s.Available in 6V-12V to ensure compatibility with RC receiversFig 26: MotorFig 27: Nuts and BoltsGearing Option150:1Peak Power Point175N @4mm/sPeak Efficiency Point75N @7mm/sMax Speed (no load)8mm/sMax Force (lifted)200NStroke Option50mmMass56gRepeatability0.3mmInput voltage0-15 VDC. Rated at 12VDCStall Current650mA @ 12VTechnical Spaces:Fig 28: Mechanical and Electrical specification of Motor2- polyethylene(C2H4) Board:HDPE (High-Density Polyethylene) is an extremely versatile product with outstanding properties and good chemical resistance for a wide variety of applications. HDPE has a low coefficient of friction.Density of HDPE can range from 0.93 to 0.97 g/cm3 or 970 kg/m3.Moisture/chemical resistant, impact resistant, superior tensile strength and FDA approved/meets NSF standards.Fig 28: Polyethylene boards3D design:Fig 29: 3D design of motor’s boxFig 30: 3D design of Plastic bracket Electrical Parts:The platform is functioning with 8 actuator motors, four at the base and two on each linear board. Each motor is powered by 650 mA @ 12V with a static force of 250N. Motors limits were controlled by built in PCB with amplitude range of ± 25 mm with a cycle time of 1- ∞ that can vary based on the amplitude. A mega Arduino is used to control the system, 8 signal wires are connected to the Arduino to adjust the signal input in a range of 3.3 - 5v volts.A switching Power Supply with an output of (5V/12V/-5V 5A/2.5A/1A 46.5W) is used to power the Arduino and the 8 Motors. The Power supply input is connected to a Power Socket with a fuse of 10A-110V to save the system from any Prospects of high voltage. The speed of the motors is controlled with the PWM through the Arduino pins by giving signal pulses from 1.3 volts to 4.2 volts, which controlled the incoming voltage to the motors that varies from 7 volts to 12 volts through a TIP122 transistor. The system is controlled wirelessly though a Bluetooth module HC-05 as the system can be controlled from any device that has a Bluetooth transmitter. Fig 31: Electrical DiagramCode Required: The code mainly was controlling the motors;the motors limits are controlled with the signal wires connected to the MC that moves in a range of 45-180 converted from analog values of 8 bits (0-1023). The motions of the 8 motors are controlled with a function of 1-2cos4 (t) with the increase of the amplitude by +1.Fig 32: Sinusoidal WaveBy summing the 8 signals of the motors and starting from the same offset positions the motors start to synchronize with each other by moving with the same amplitude and time signal.Fig 33: Synchronized Sin waveThe motors were functioned to move in the limits of ± 20 mm and ± 40 mm. The motors were controlled by a Bluetooth module and by adjusting the Serial Software library to control two imaginary pins to act as the RX and TX. By adding the Bluetooth, serial in the code we were able to control the Platform through any device has a Bluetooth Receiver. We are using an Android phone to control the motion by sending variables declared as “Char”.Budgeting:Fig 33: Budget digramParts:ItemDescriptionQuantity1Platform -Cutting Boards HDPE (Cut-to-Size) - White 1/2 (.500)" Thick, 13" Wide, 26" Long all board as bundle has the same shipping as one cart shipping ($37.93)12Platform base-Cutting Boards HDPE (Cut-to-Size) - White 1/2 (.500)" Thick, 19" Wide, 35"Long13Polyethylene Sheets - HDPE (Cut-to-Size) - Natural1/4 (.236)" Thick, 2-11/16" Wide, 35" Long24Polyethylene Sheets - HDPE (Cut-to-Size) - Natural1/4 (.236)" Thick, 2-11/16" Wide, 19" Long25Nain Base-Cutting Boards HDPE (Cut-to-Size) - White1/2 (.500)" Thick, 26" Wide, 44-1/2" Long (Oversize charge.)16Polyethylene Sheets - HDPE (Cut-to-Size) - Natural 1/4 (.236)" Thick, 2-11/16" Wide, 39" Long27Polyethylene Sheets - HDPE (Cut-to-Size) - Natural1/4 (.236)" Thick, 2-11/16" Wide, 23" Long28Polyethylene Sheets - HDPE (Cut-to-Size) - Natural1/4 (.236)" Thick, 2" Wide, 39" Long29Polyethylene Sheets - HDPE (Cut-to-Size) - Natural1/4 (.236)" Thick, 2" Wide, 23" Long210Polyethylene Sheets - HDPE (Cut-to-Size) - Natural1/4 (.236)" Thick, 3" Wide, 3" Long10119V battery holder with switch & 5.5mm/2.1mm plug +shipping $9.121127 Pcs 7Pin 6mm Dia D Shaft 20 Detents Points 360 Degree Rotary Encoder Push Button Switch+ shipping is free113Arduino Compatible Mega 2560 Atmega2560 Mega2560 R3 Board + USB Cable+shipping $8.32 for all downward114Smraza 120pcs Multicolored Jumper Wire 40pin Male to Female, 40pin Male to Male, 40pin Female to Female Breadboard Jumper 115SMAKN? TB6600 Upgraded Version 32 Segments 4A 40V 57/86 Stepper Motor Driver116L16-R Miniature Linear Servos for RC & Arduinok,Stroke4:: 50 mm , Ratio: 150:1 Max Voltage: 6 Volt+ shipping $15.05 for all downward 817DSD TECH HC-05 Bluetooth Serial Pass-through Module Wireless Serial Communication with Button for Arduino118Cables Unlimted 6-feet Mickey Mouse Power Cord119Switching Power Supply1 Engineering Standards(a) In this project, digital imaging and communication standard in medicine (DICOM) will be used which is employed internationally to store, exchange and transmit medical imaging data. The measurement, Reporting, and Management of Radiation Dose in CT (AAPM NO.96)(b) The phantoms used in the data acquisition include standard (NEMA XR21) water-equivalent materials that represent human tissues, which in turn are composed mostly from water. The phantom has low-density foam material is used to represent lung tissues and it is compatible for imaging with different imaging modalities such as CT and MRI. (c) Determination of Slice Thickness In Diagnostic Magnetic Resonance Imaging (NEMA MS 5).Economic IssuesOur project requires taking CT images. CT machines are multi-million dollar machines, which pose financial issue to someone without CT machines trying to replicate our project or do similar work. CT imaging itself is an expensive procedure. The phantoms that we have been using and the platforms for the phantoms cost thousands of dollars too. These were all accessible to us by the courtesy of Stephenson Cancer Center. Beside these however, the software in which we have been working most of our project on, DIRART software, is free software accessible to anyone online. The goal of this project is to derive a new, more efficient way of doing deformable registration. If successful, they will cut the cost not only from the patients, since they won’t have to take CT images more often, and consequently health insurance companies, but also from the hospital or research centers that own the CT machines.Health and safety issuesSince our project revolves around medical equipment and procedures, our product will have significant impact on health and safety of its beneficiaries. We are trying to come up with better way of locating and characterizing the tumors on patients. If we are successful, we will enable medical personnel to provide treatment to the patients in more efficient way, leading to a healthier lifestyle for the patients. Then, we can decrease the radiation-imaging dose. However, if there is something wrong with our product, then its application might mean wrong treatment plan for tumors and thus the health and safety of the patients is compromised.Potential impact of the projectDeformable image registration has a large potential with increasing clinical applications that include diagnostic imaging as well as radiation therapy in recent years. This involves correction of variations in patient anatomy in CT images due to motion, changes in the position or fillings of certain organs such as stomach, bladder or rectum. The goal of this project is to investigate correlation of displacement vector fields (DVF) calculated by deformable image registration algorithms with motion parameters in helical, axial and cone-beam CT images with motion artifacts. If such correlation can be found, then motion artifacts in CT images can be significantly removed after post-scan image reconstruction work. This would mean better and more accurate images and thus better dose and therapeutic planning and better treatment of cancer patients. Another phase, is developing software andhardware components that can be used with the DIR algorithms investigated in this project.AcknowledgementsWe would like to thank Dr. Imad Ali and Dr. Salahuddin Ahmad from the University of Oklahoma Stephenson Cancer Center for their collaboration with us. We also thank Dr. Nesreen Alsbou, our project advisor and Dr. Deshan Yang, one of the creators of the DIRART software. We would also like to acknowledge Amjad and Mohamad for their help in the project.References:1) Thirion, J. (1998). Image matching as a diffusion process: An analogy with Maxwell's demons. Medical Image Analysis,2(3), 243-260.2) Cantator, A and Muller, P. March 2011.Introduction to Computed Tomography, Technical University of Denmark.3) Wang, H., Dong, L., O'Daniel, J., Mohan, R., Garden, A., Ang, K., Cheung, R. (2005). Validation of an accelerated demons algorithm for deformable image registration in radiation therapy. Physics in Medicine and Biology,50(12), 2887-2905.4) Weiguo Lu. (2004). Fast free-form deformable registration via calculus of variations.Physics in Medicine and Biology,49(14), 3067-3087.5) Gu, Xuejun, Pan, Hubert, Liang, Yun, Castillo, Richard, Yang, Deshan, Choi, Dongju, Jiang, Steve B. (2010). Implementation and evaluation of various demons deformable image registration algorithms on a GPU. Physics in Medicine and Biology,55(1), 207-219.6) Computational Applied Mathematics Publications. (n.d.). Retrieved December 04, 2016, from ) Horn KPaSB. Determining Optical Flow. Massachusetts Institute of Technology Artificial Intelligence Laboratory 1980.8) Barron JL, Fleet DJ, Beauchemin SS, Burkitt TA.Performance of optical flow techniques.? Computer Vision and Pattern Recognition, 1992 Proceedings CVPR '92, 1992 IEEE Computer Society Conference on; 1992 15-18 Jun 1992; 1992. p. 236-42.9) Chen GTY, Kung JH, Beaudette KP. Artifacts in computed tomography scanning of moving objects.Seminars in Radiation Oncology; 14: 19-26.10) Yang D. BS, El Naqa I., Aditya A., Wu Y., Goddu M., Mutic S., Deasy J., Low D. Technical Note: DIRART-A software suite for deformable image registration and adaptive radiotherapy research. Med Physics 2010; 38: 67-77 ................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.
Related download
- summary details of the ats test tracks http
- vanderbilt university
- mica spec flextherm thermal insulation spec guide for
- flexible conduit cable carrying capacity wire fill
- colonoscopy surveillance of colorectal polyp recurrence in
- section 10674
- comparisons and combined application of two dimensional
- university of arizona
- shelter repair project crs
- correction of motion artifacts in ct images using motion
Related searches
- most unexplained artifacts in world
- images of skin rashes in adults
- newton s laws of motion pdf
- laws of motion worksheet pdf
- newton s laws of motion printables
- newton s second law of motion example
- newton s second law of motion state
- newton s law of motion examples
- 3 laws of motion examples
- newton s first law of motion formula
- newtons laws of motion facts
- rate of motion formula