• 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • benign masses The abnormalities on the images cover br


    benign masses. The abnormalities on the images cover 2–6 209
    152 presented by Zheng et al. [24] detected cue suspected lesions categories of BIRADS mammographic assessment. At least one 210
    154 The matching areas of the suspicious lesions are identified considered in this GW-311616 study. The size of abnormal lesions on 212
    156 increasing the robustness of the scheme. Yuan et al. [25] images in DICOM (Digital Imaging and Communications in 214
    157 presented a two stage computerized framework to match Medicine) format captured on 'Hologic Selenia System' and on 215
    158 suspicious lesion on CC view to its correlative counterpart 'GE Medical Senograph System' contain 4096 3328 pixels 216
    160 0.03. However, the metrics such as sensitivity, specificity, images are marked under the supervision of expert radiolo- 218
    161 accuracy, false positives are not assessed to show diagnostic gists participating in this study. Another set consists of 148 219
    162 abilities of their CAD system. Wei et al. [6] analyzed two mammograms selected from publicly available DDSM dataset 220
    163 views of mammograms to improve case-based sensitivities [30]. This set comprises 38 pairs benign masses and 36 pairs 221
    165 mass detection. Nipple detection is manual and not fully lesions on all these images are delineated by the radiologists in 223
    166 automatic. Padayachee [26] explored GLCM textural fea- the DDSM dataset.
    167 tures, distance similarity, and mutual information for
    168 matching suspicious lesion on ipsilateral views for a
    4. Methods
    170 The scheme reported two shortcomings, one related to
    171 ground truth and the other regarding smaller dataset. Zheng Methods for detection of breast abnormalities require that 226 172 et al. [27] investigated three methods for deciding width of suspicious anatomical lesions in mammograms be segmented 227 173 search area used to match the suspicious lesion from and characterized. The locations specific information of 228 174 ipsilateral views and showed that straight strip search lesions are then used to restrict processing to relevant areas, 229 175 method works better with case-based sensitivity of 86% at to perform location-dependent processing, or to correlate 230 176 0.29 FPs/I and hence is suitable for CAD systems. Samulski findings in different views. This section describes the proposed 231 177 and Karssemeijer [28] proposed a multiview CAD system for method covering four stages such as pre-processing, single 232 178 mass detection with improved case-based sensitivity by view diagnosis, ipsilateral view diagnosis, fusion of single with 233 179 4.7% at 0.5 FPs/I using information collected from ipsilateral two view diagnosis. Pre-processing of medical images 234 180 views of mammograms. If the chest wall is not parallel to increases the efficiency and accuracy of the CAD algorithms 235 181 the border of CC view or nipple on larger breast is not in identifying the abnormalities easily [31,32]. Pre-processing 236 182 detected, the performance of the system may degrade. stage includes detection of air–skin interface, nipple in the 237 183 Tanner et al. [29] exploited region-based 71 features profile and pectoral muscle on MLO views, etc. It is covered in 238 184 combined with modified search area techniques for corre- brief in the subsection as below. 239
    185 lating mammographic masses on ipsilateral views with
    187 0.915. In this approach, distance to manually delineated
    188 location of nipple seems to degrade the performance. On an average, the actual size of the FFDM image is almost 241 189 The literature surveyed has helped us to identify the double the size of the breast parenchyma as shown in Fig. 1(a). 242 190 research gap and we have attempted to bridge this gap by The proposed work starts with minimizing the size of the input 243 191 devising computer algorithms which can improve perfor- mammogram to reduce the computational burden during 244 192 mance of CAD systems using two views mammogram segmentation phase. The air–skin interface and nipple of the 245 193 analysis. breast are detected using fuzzy c-means clustering technique. 246
    Please cite this article in press as: Sapate S, et al. Breast cancer diagnosis using abnormalities on ipsilateral views of digital mammograms.
    The experiment is performed by selecting three clusters of the given mammographic image. The three cluster centres separate the mammogram into three partitions background, breast profile and a dense portion in the profile.