Development of mammogram segmentation algorithm to detect microcalcification and mass a hyper elastic model based approach

Abstract

Breast cancer is the most commonly occurring cancer in women and the second most common cancer overall. In 2018, over 2 million new cases were diagnosed. Breast cancer is the second most death causing cancer in women next to Lung cancer. Early detection of breast cancer is very essential for increasing the survival rate. This can be screened by digital mammography. Even though mammography is an effective tool for early diagnosis of breast anomalies, the detection of tumors in mammogram film is the most challengeable clinical task. This is due to the greater number of non-pathological structures of the breast .This leads the radiologists towards false diagnosis in many cases. To reduce false cases, physicians use computer aided detection as a second reading option. Segmentation of mass and Mcs can be achieved by CAD based algorithms with high impacts. A successful CAD system initiates a good computer aided-diagnosis and this greatly saves the work of radiologists and profits the victims. This new enhancement technique is focused on mammograms. This research work involves enhancement and segmentation of mammogram. Enhancement technique involves a new hyper elastic model, which mimics the breast and preserves the topology of mass and microcalcification in mammogram. It is very much essential in medical images for better decision making. The proposed approach has been experimented with MIAS data set. The technique has earned a higher effective measure with an average of 7.57 against the traditional methods. Most significantly, the visual inspection has suggested that the proposal protects topological information well without destroying the important features. On close observation of mammogram segmentation, most of the algorithms found in literature are exclusively built to deal with identifying the mass or microcalcification. Reason might be that the existing algorithms do not consider the property of the subject involved. Only few model based on segmentation or detection works in literature. This proposal defines s