National Institute of Technology Calicut Department of Computer

Breast cancer is the most common form of cancer among women in many regions of India. One out of every 22 women is diagnosed with breast cancer. Mammography acts as an early detection tool for breast cancer and can detect tumors up to two years before it can be felt. Numerous studies have shown that the early detection saves lives and increases treatment options. Mammography is one of the X-ray imaging techniques to produce images (mammograms) of the internal structures of human breast. The information of these images might be used by physicians to detect abnormalities including breast cancers. Today, Computer Aided Detection (CAD) systems are widely used to interpret mammograms especially in the detection of breast cancers. Though CAD systems for mammogram analysis are available in markets, experts of this field believe that CAD systems in their present form is not effective and have demonstrated the need for the development of more appropriate algorithms for mammogram analysis. In CAD systems, one of the preprocessing steps normally carried out on mammograms is the removal or suppression of pectoral muscles as its presence within the mammograms may influence breast cancer detection. Through this research, we have attempted to obtain efficient automatic methods to identify pectoral muscle, which appears as a homogeneous region, in mediolateral oblique-view mammograms using a proposed segmentation method, the graph cut based merging. We are proposing two different approaches to identify pectoral muscle in a mammogram using the proposed graph cut based merging method. In the first approach, the proposed image segmentation method is used for identifying the pectoral muscle edge. The identified pectoral muscle edge is found to be ragged. Hence, the pectoral muscle is smoothly represented using Bezier curve which uses the control points obtained from the pectoral muscle edge. The proposed work was tested on a public dataset of medio-lateral oblique-view mammograms obtained from mammographic image analysis society database, and its performance was compared with that of the state-of-the-art methods reported in the literature. The mean false positive and false negative rates of the proposed method over randomly chosen 84 mammograms were calculated, respectively, as 0.64% and 5.58%. Also, with respect to the number of results with small error, the proposed method out performs existing methods. The second approach uses watershed transform. The watershed transformation of a mammogram shows interesting properties including the appearance of a unique watershed line corresponding to the boundary of the pectoral muscle. In addition to this, it is observed that the pectoral muscle is oversegmented due to the existence of several irrelevant catchment basins within the pectoral muscle. Hence a modified graph cut based merging algorithm is used to combine appropriate catchment basins to obtain the actual pectoral muscle boundary. The mean false positive and mean false negative of this approach is 0.85% and 4.88% respectively. With respect to the mean false negative, the performance of the proposed method is better than the existing methods. These results demonstrate that the two proposed approaches can be used effectively as a preprocessing step in the detection of breast cancers in CAD systems.