CSCC - WSEAS



Investigating the race factor in Mammography

MAJDI AL-QDAH[pic], ROZI MAHMUD[pic], NANI ADILAH BINTI SUHAIMI [pic], ABD. RAHMAN RAMLI [pic], and RAHMITA WIRZA[pic]

[pic] Faculty of Information Technology, Multimedia University

Jalan Multimedia, 63100 Cyberjaya Selangor, Malaysia

[pic] Faculty of Radiology, University Putra Malaysia

43400 UPM Serdang, Selangor D.E. Malaysia

[pic] Institute of Advanced Technology, University Putra Malaysia

43400 UPM Serdang 43400 UPM, Selangor D.E. Malaysia

[pic] Faculty of Information Technology, University Putra Malaysia

43400 UPM Serdang, Selangor D.E. Malaysia

Abstract: - In this paper the race factor is investigated in two independent mammography studies. First, the glandular breast optical density (OD) is measured manually using a densitometer device for 45 mammograms from each of the three main races in Malaysia divided into three different age groups. Second, different wavelets were compared and eventually the db4 wavelet filter was used to detect micro calcifications in a set of seventy-five mammograms from each Malaysian race. The glandular OD measured results showed that there are no significant differences of the OD for the breast glandular area among the three races of the same age group but there are little variations among different age groups’ OD for the glandular area. Similarly in the detection study the findings only confirmed the conclusion of the first study in that mammograms from all races equally have variations in their contrast which results in no one race mammogram is said to be different for radiologists’ calcifications detection over any other race mammograms for all age groups.

Key-Words: - mammogram, optical density, glandular, classifications, cancer, wavelets.

1. Introduction

The breast of women can be classified into three board categories, depending on the relative amounts of fibro- glandular tissue and fatty tissue present. There are fibro – glandular breast, fibro- fatty breast and fatty breast [1]. Meson [1] classified breast into three categories, which are fatty breast, mixed density breast and dense breast. Recently, the Breast Imaging Report and Data system (BI- RADS) adopted four categories of breast tissue densities: almost entirely fat, scattered fibroglandular tissue, heterogeneously dense, or extremely dense. The common age grouping for the fibro- glandular category is postpuberty to about age thirty. Pregnant or lactating females of any age also tend to have fibro-glandular type breast. Women ages 30 to 50 breasts’ is not usually as dense as that of younger women’s’ breasts that usually have fibro- fatty breast types. Differently, women over 50 years of age commonly have fatty breast type. The breast of children and most male contain mostly fat in small proportion therefore fall into the fatty breast types.

Several factors have been reported to be associated with variations in the female breast such as age, body mass index (BMI), parity, obesity, hormone replacement therapy (HRT), breast size, smoking, family history of breast cancer and dietary [2], several factors which are associated with the accumulation of fat in breast tissue such as obesity, oestrogens in post – menopausal hormone replacement regimes, parity and dietry factor [3]. The density of the breast can be determined from a mammogram film and can be measured manually using a densitometer device, which measures the optical density of the mammogram. In this study the breast tissue of a sample that came from each one of the three main races in Malaysia was classified according to the BI-RADS classification categories then a comparison is performed.

Several studies have attempted to relate ethnicity to the occurrence of breast cancer. For example, white women have been found to have higher rates of breast cancer more than the incidence of breast cancer in black women [4]. In Malaysia a similar study was done by University Malay Medical Center in Kuala Lumpur (HUKL) in the year 2000 has shown that 60% of 952 cancer patients admitted to the UHKL in the years 1993 to 2000 were Chinese patients. It was concluded that the incidence of breast cancer in Chinese appears to be higher than the other two races, namely Malays and Indians. In diagnosing cancer, radiologists try to distinguish normal from cancerous tissue by looking at shape and density of an up normal mass in a tissue area. Usually the malignant area is characterized by indistinct border-shapes. For early detection radiologists try to identify micro calcifications and usually the clustered ones of them as indicators of cancer (malignant), which usually appear in clusters with very sharp edges, irregular shapes, and very small in size. The micro calcifications are normally known to be of size range 0.1 mm to 1 mm isolated or clustered pixels that constitute the first warning signs for breast cancer [5].

According to radiologists, breast density plays an important factor in breast cancer detection. Radio graphically visible density includes ducts, lobular elements, and fibrous connective tissue. The fibrous connective tissue can be of two types, intralobular or extralobular tissue, and this latter tissue type seen as the major component of gross density variation in mammograms [6]. Breast density is an important factor in the interpretation of a mammogram. In a breast that is considerably dense, the sensitivity of mammography for the early detection of malignancy and large cancers is reduced because of the difficulty in locating ill-defined cancers within an opaque uniform background [6]. In mammography the contrast between the whole soft tissues of the breast is minimal and small change in the breast tissue structure can mean malignant breast tumor [7]. Therefore, tools for detection should be available for image enhancement and feature extraction.

Wavelets are good tools for medical image enhancement that perform enhancement by amplification or some modification to wavelet coefficients prior to reconstruction. The wavelet transform is a decomposition of a signal with respect to a real orthonormal basis of functions,[pic] obtained by translations and dilations of a single mother wavelet [pic]. Using the wavelet transform, the mammogram is decomposed into high frequency (details) and low frequency components (approximations) and micro calcifications are believed to exist in the high frequency components of the decomposed mammogram image at various levels of decomposition [8].

2. Materials and Methods

2.1 Classifications of mammograms

Some sample mammograms were obtained from two major Malaysian hospitals: Hospital Selayang and Hospital Tunku Ampuan Rahimah. All studies from Hospital Selayang were performed on Mammography unit Mammo Diagnost 300, Philips. KODAK, Min – R 200 film and KODAK MN – R2 casstte ( size 18 x 24 ). All studies from Hospital Tunku Ampuan Rahimah performed on PHILIPS MAMMO DIAGNOSTIC unit, KODAK, MIN – R 200 film and KODAK MIN – R2 with C – IN WINDOW KODAK cassette ( 18 x 24 cm ). A densitometer 07 – 443, Nuclear Associates, Victoreen and white board marker. This densitometer can be used to read the Optical Density (OD) of a mammogram. White board marker is used to draw a region on the mammogram. The study was done inclusively of breast augmentation which needs 24 x 30 cm cassette size, history of breast cancer or breast biopsy, taking HRT or any hormone, breast screening are abnormal from calculus, fibroadenama or micro calcification, breast implant, and family history of breast cancer.

135 women mammograms: Malays, Chinese and Indians aged 40 to 54 years old who had screening mammogram at the above mentioned hospitals and followed the exclusion criteria were included in this study. They were 45 Malay women, 45 Chinese women, and 45 Indian women. The age was divided into three age groups: 40 to 44, 45 to 49 and 50 to 54 years of age. For each group the breast tissue is supposed to vary with age. In each mammogram, the mean OD of the glandular area was used to classify the mammograms. Firstly, the mammograms had to be put on an illuminator and the nipple had to be recognized. A radial line was drawn from the nipple to the chest wall and the glandular area is located 2 cm from the nipple to 5 cm from the nipple.

The optical density (OD) value for each region in each mammogram was measured by the densitometer device. The glandular area was split into three same size compartments and the OD value was measured for every compartment; then the mean of the three is recorded. The mean OD of the glandular area from LMLO was added to the mean OD from RMLO and the resulting OD value is divided by two. The mean OD of the glandular area was used in classifying the breast types since these densities are located in the middle OD for glandular and fatty tissue. The breast densities were classified according to the American Collage of Radiology (ACR), breast imaging and reporting Data System categories (BIRAD): Almost entirely fatty, Scattered fibroglandular tissue, Heterogeneously dense, and Extremely dense. Table 1 summarizes the classifications results and their corresponding OD value range used for this study and as used by other researchers such as Kimme Smith [9].

Table 1 shows the classification of breast types with their corresponding OD range.

|OD range |Breast types |

|0 – 0.76 |Extremely dense |

|0.8 – 1.2 |Heterogeneously dense |

|1.26 – 1.33 |Scattered fibro glandular tissue |

|1.4 – 1.8 |Almost entirely fatty |

After recording the OD values, the SPSS software was used to analyze the data. A ANOVA Mean Test at significance value of p < 0.05 was used to compare the races’ breast types and the Khai square test was used to find out the relationship of the breast types among all races and age groups.

For the 45 pair mammogram from the Malay race used for this study the ANOVA test based on some age group showed little significant difference (p > 0.05) OD glandular area for Malay women among age group 40 - 44, 45 – 49 and 50 – 54 and the Khai square test showed little significant relationship between the breast type and age group among Malay women. Table 2 shows the OD values for Malay women for three age groups. Table 5 shows the Mean OD glandular area for Malay women for the three age groups. Similarly, for the 45 pair mammogram from the Chinese race used for this study the ANOVA test based on some age group showed little significant difference (p > 0.05) OD glandular area for Chinese women among age group 40 - 44, 45 – 49 and 50 – 54 and the Khai square test showed little significant relationship between the breast type and age group among Chinese women. Table 3 shows the OD values for Chinese women for three age groups. Table 6 shows the Mean OD glandular area for Chinese women for the three age groups. Likewise, for the 45 pair mammogram from the Indian race used for this study the ANOVA test based on some age group showed little significant difference (p > 0.05) OD glandular area for Indian women among age group 40 - 44, 45 – 49 and 50 – 54 and the Khai square test showed little significant relationship between the breast type and age group among Indian women. Table 4 shows the OD values for Indian women for three age groups. Table 7 shows the Mean OD glandular area for Indian women for the three age groups. Regarding race, all races can be observed to have comparable values and classifications among the four different BI-RADS breast type classifications.

Table 2 shows the Mean OD glandular area for Malay women for the three age groups studied.

|Age Group |Min OD |Test |p |

|40 - 44 year |0.963 ± 0.477 |ANOVA |0.068 |

| | | F= 2.871 | |

|45 - 49 year |1.111 ± 0.379 | | |

|50 - 54 year |1.139 ± 0.458 | | |

Table 3: Mean OD glandular area for Chinese women for the three age groups studied

|Age Group |Min OD |Test |p |

|40 - 44 year |1.050 ± 0.324 |ANOVA  |0.136 |

| | |F = 2.092 | |

|45 - 49 year |1.034 ± 0.422 | | |

|50 - 54 year |1.27 ± 0.298 | | |

Table 4: Mean OD glandular area for Indian women for the three age groups studied

|Age Group |Min OD |Test |p |

|40 - 44 year |1.033 ± 0.353 |ANOVA |0.478 |

| | |F = 0.751 | |

|45 - 49 year |1.121 ± 0.448 | | |

|50 - 54 year |1.203 ± 0.326 | | |

Table 5 shows the classification of the 45 Malay mammograms among 4 BIRADS breast types.

|Age Group |Extremely |Scattered(SF) |Heterogeneously |Almost entirely |Total |

| |dense (ED) |Fibroglandular |dense (HD) |fatty (AEF) | |

| | |tissue | | | |

|  40 - 44 year  |6 ( 40.0 % ) |4 ( 26.7 % ) |0  |5 ( 33.3 % ) |15 |

| 45 - 49 year |3 ( 20.0 % ) |7 ( 46.7 % ) |2 ( 13.3 % ) |3 ( 20.0 % ) | 15 |

| 50 - 54 year  |1 ( 6.7 % ) | 3 (20.0 % ) |3 ( 20.0 % )  |8 ( 53.3 % ) |15 |

|Total |10 |14 |5 |16 |45 |

Table 6 shows the classification of the 45 Chinese mammograms among 4 BIRADS breast types.

|Age Group |Extremely |Scattered(SF) |Heterogeneously |Almost entirely |Total |

| |dense (ED) |Fibroglandular |dense (HD) |fatty (AEF) | |

| | |tissue | | | |

|40 - 44 year |4 ( 26.7 % ) |8 ( 53.3 % ) |1 ( 6.7 % ) |2 ( 13.3 % ) |15 |

|45 - 49 year |4 ( 26.7 % ) |7 ( 46.7 % ) |1 ( 6.7 % ) |3 ( 20.0 % ) |15 |

|50 - 54 year |1 ( 6.7 % ) |6 ( 40.0 % ) |2 (13.3 % ) |6 ( 40.0 % ) |15 |

|Total |9 |21 |4 |11 |45 |

Table 7 shows the classification of the 45 Indian mammograms among 4 BIRADS breast type

|Age Group |Extremely |Scattered(SF) |Heterogeneously |Almost entirely |Total |

| |dense (ED) |fibroglandular |dense (HD) |fatty (AEF) | |

| | |tissue | | | |

|40 - 44 year |3 (20.0 %) |8 ( 53.3 % ) |1 ( 6.7 % ) |3 ( 20.0 % ) |15 |

|45 - 49 year |3 (20.0 %) |5 ( 33.3 % ) |1 ( 6.7 % ) |6 ( 40.0 % ) |15 |

|50 - 54 year  |2 ( 13.3 % ) |7 ( 46.7 % ) |1 ( 6.7 % ) |5 ( 33.3 % ) |15 |

|Total |8 |20 |3 |14 |45 |

2.2 Mammogram calcification detection

Special hardware and software tools such as wavelet filters have been used to detect micro calcifications and other details in medical images. The detection performance of a wavelet filter depends on the length and amplitude of the wavelet filter used to decompose and reconstruct the mammogram images. For example, wavelet filters db4 and coif2 characteristics make them more sensitive to details and thus they both show a lot of positive existence of calcifications in the detection process which maybe corrupted with some noise with db4 filter detecting less noise. The sym4 filter detects little micro calcifications. The db20 is smoother filter than db4 filter and therefore is less sensitive to details and will detect less calcifications and less noise. This is true since shorter wavelet filters are more sensitive to existing micro calcifications but they tend to produce more false calcification detection. The detection of the db1 filter detects most calcifications and most noise. Experimentally in this study it is found that the detection ability of the db4 filter was best for most of the mammograms from the three races so it is finally chosen for analysis and comparisons among the samples from each race. The db4 wavelet filter is applied to a set of seventy-five cutouts and whole breast image mammograms from each main Malaysian race.

The choice of the db4 wavelet filter is based on that this wavelet filter is biorthogonal and compactly supported filter that makes it appropriate for detail detection. All of the mammogram images have been visually identified to contain small clusters and isolated pixels areas of micro calcifications. The images are obtained from the national cancer center in Kuala Lumpur and verified by radiologists at the center. Using the forward wavelet transform the images are decomposed into few levels of wavelet decomposition using the db4 wavelet filter. The micro calcification detected pixels at the detail levels of each decomposition level are combined and reconstructed to constitute one detail image. Since calcification corresponds to higher contrasts than surrounding texture of breast tissue, thresholding is applied based on a statistical measure in a Region of Interest (ROI) of the mammogram; namely, the histogram of the cutout. In order to extract the calcifications, the tissue background is set to zero so the final resulting image is bright white spots corresponding to micro calcification present in the original mammogram image on a black background corresponding to normal tissue. The wavelet filter detection procedure is repeated on the mammogram sample images from the three races. Detection is counted if the filter correctly identified the radiologist’s identified area of micro calcifications in the mammogram. After experimenting with a ROI in the mammogram images from all of the three races for different age groups, the detection percentage of the wavelet filter applied on the set of mammogram samples from each race is recorded to obtain the detection comparison chart. Fig. 1 shows a sample cutout mammogram image. Fig. 2 shows the micro calcification detected pixels from applying the db4 wavelet filter. Fig. 3 shows the detection comparison among five different wavelet filters. Fig. 4 shows the detection comparison chart among the three main races in Malaysia: Chinese, Malays, and Indians.

[pic]

Figure1: Micro calcifications visible in a mammogram

[pic]

Figure 2: shows the detected micro calcifications from applying db4 filter

[pic]

Figure 3: shows the detection comparison among five different wavelet filters

[pic]

Figure 4: shows the detection comparison chart among the three races

4. Conclusion

In classifying the mammograms no significant difference was observed of the OD fibro glandular areas among all the races studied. Malays, Chinese, and Indians can have any type and comparable breast types since the Khai square test showed that there is no significant relationship between breast types and race for the three races; but there is a little significant difference of the OD glandular areas among different age groups. The younger age group had lower OD value while older age groups had higher OD value.

Comparably, a technique of breast cancer detection confirmed the findings of the classification technique in that the race factor does not play a major part if any in the easiness or difficulty of detecting cancer since all mammograms from all races compared almost equally in detecting calcification for all age groups studied. Using wavelet tools to detect early signs of cancer, calcifications, it is concluded from studying the race factor in breast cancer detection that the characteristics of the tissue for women vary from one woman to another for the three main races in Malaysia: Malays, Chinese and Indians. All women vary in their density of breast tissue from the low Fatty tissue type to the extremely dense type of tissue that can come from all races with no particular race of any specific tissue type.

Therefore, it should recommend to radiologists to spend time in studying mammograms of all races equally with no one race in particular easier for breast cancer diagnosis and detection contrarily to the initial hypothesis that some race maybe easier or harder for detection. Of course, since both classification and detection studies used a low number sample definite conclusion of the race factor in detecting cancer can be done only with a larger sample over a long period of time in real hospitals by real radiologists.

References:

[1] Meeson, S., Young, K. C., Ramsdale, M. L., Wallis, M. G. & Cooke, J. Analysis of optical density and contras in mammograms. The British Journal of Radiology 72: 670- 677. 1999.

[2] Lucassen, A., Watson, E. & Eccles, D. 2. Advice about mammography for a young woman with a family history of breast cancer. Clinical review. BMJ ; 2001. 322 : 1040 – 1042.

[3] Salminen, T. M., Saarenmaa, I. E., Heikkila, M. M. & Hakama, M. Unfavourable change inmammographic patterns and the breast cancer risk factors. Breast Cancer Res Treat 57( 2 ): 165–173. 1999.

[4] Malaysian Study on the Internet



[5] M. Unser and A. Aldroubi, "A review of wavelets in biomedical applications," Proc. IEEE Int. Conf. Image Processing, Santa Barbara, CA, Oct. 26-29, 1997.

[6] Pankow, J. S., Vachon, C. M., Kuni, C.C., et al. Genetic analysis of mammographic breast density in adult women : evidence of a gene effect. J Natl Cancer Inst ; 89 : 1997. 549 – 556.

[7] A. F. Laine, S. Schuder, J. Fan, and W. Huda, "Mammographic feature enhancement by multiscale analysis," IEEE Trans. Med. Image., vol 8, no 5, pp.491 - 504, 1997.

[8] S. G. Mallat, "A theory for multi resolution signal decomposition: The wavelet representation," IEEE Trans. Pattern Anal. Machine Intell., vol.11 July 1989.

[9] L. Sheng , Charles F. and Edward J. "Multiresolution Detection of Spiculated Lesions in Digital Mammograms," IEEE Trans. Med. Image., vol 6, No. 6 June 2001.

[10] Bontrager, K. L. Textbook of Radiographic Positioning and Related Anatomy. Ed ke–3. St Louis: Mosby year Book. 1993.

[11] Chang, Y. H., Wang, X. H., Hardesty, L. A., Chang, T. S., Poller, W. R., Good, F. D. & David Gur. Computerized Assessment of Tissue Composition on Digitized Mammograms. Academic Radiology 9 ( 8 ) 2002.

[12] Warwick, J., Pinney, E., Warren, R. M. L., Duffy, S. W., Howell, A., Wilson, M. & Cuzick, J. Breast density and breast - cancer risk factors in a high-risk population. The Breast 12: 10-16. 2003.

[13] El – Bastawissi, A. Y., White, E., Mandelson, M. & Taplin, S. Reproductive and hormonal factors associated with mammographic breast density by age. Cancer Causes and Control 11: 955-963. 2000.

[14] Stomper, P. C., D’ Souza, D.J., DiNitto, P.A, Arredondo M. A. Analysis of parenchymal density on mammograms in 1353 women 25- 79 years old. AJR Am J Roentgenol; 167 : 1261 – 1265. 1996.

[15] S.K. Aditya, C.H. Chu, and H.H. Szu. Application of adaptive subband coding for noisy bandlimited ecg signal processing. SPIE, 2762, 1996.

[16] American College of Radiology. (1998)Breast Imaging Reporting And Data System. Third Edition. Reston[VA]: American College of Radiology.

[17] J.N. Bradely, C.M. Brislawn, and T. Hopper. The fbi wavelet/scalar quantization fingerprint image compression standard.

[18]Technical report, Los Almos National Laboratory, Los Almos, NM 87545, 1994.

[19] Boyd, N. F., Byng J. W., Jong, R. A., Fishell, E. K., Little, L. E., Miler, A. B., Lockwood, G. A., Tritchler, D. L. & Yaffe, M. J. Quantitative classification of mammographic densities and breast cancer risk : result from the Canadian National Breast Screening Study. J Natl Cancer Inst 87( 9 ): 670 – 675. 1995.

[20] Kerlikowske, K., Grady, D., Rubin SM., Sandrock, C. & Ernster, VL. 1995. Efficacy of screening mammography. A meta – analysis. JAMA ; 273: 149 – 154.

[21] Vachon, C. M., Kuni, C. C., Anderson, K., Anderson. V.E. & Seller T. A. 2000. Association of mammographically defined percent breast density with epidemiologis risk factors for breast cancer ( United States ). Cancer Causes Control ; 11 : 653 – 662.

[22] Bontrager, K. L. 1993. Textbook of Radiographic Positioning and Related Anatomy. Ed ke–3. St Louis: Mosby year Book.

 

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

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

Google Online Preview   Download