Segmentation of Digital Breast Tomograms using clustering techniques

In current scenario Breast cancer is the most common cause of cancer death in women. Recently, mammography is used as the most convenient examination technique for the detection of early signs of breast cancer. Clustered Microcalcification (MC) in mammograms plays a vital role in indication for early detection of breast cancer. The conventional 2D mammography has a severe limitation of decrease in contrast of structures due to the superimposition of overlying tissue and this leads to false positives and false negatives. Digital Breast Tomosynthesis (DBT), a 3D mammographic technique, is a new modality for breast imaging that overcomes the limitation of planar mammography. As breast is composed of soft tissues, segmentation of a cancerous and a non cancerous tissue is considered as the most important and significant step in image processing of breast tomograms. In this work, we consider segmentation of tomograms obtained from Filtered back projection (FBP) reconstruction of raw images. Fully automated, robust and efficient clustering methods, Fuzzy C Means and Expectation Maximization, are used in our work to segment the microcalcifications from the other non cancerous anatomical structures of the breast such as connective tissue, glandular tissue, contractile tissue, skin and chest muscle. Our work resulted in greater sensitivity value for EM than FCM. The algorithm is extended to sequence of tomograms for segmentation of the microcalcification content in each of the slices. The segmentation done will also lead to effective and accurate feature extraction and classification of malignant and benign microcalcifications.