Abstract:
Breast cancer is the major cause of death among women across the world. In Thailand, the number of patients with breast cancer is 16.3% of patients with all types of cancer, which is one of the leading causes of death. This, of course, encourages cancer detection and treatment before it escalates to other parts of the body. For a better treatment of cancer, there are developments of diagnostic approach, laboratory as well as medical instruments. There developments along with increasing knowledge of pathology have tremendously altered the current trend of diagnosis and treatment of breast cancer. This is evidenced by appropriate advice of an accurate selection of appliances for detection and treatment of breast cancer based on academic principles and in a most cost effective manner. In this connection, computer-aided detection/diagnosis (CAD) is developed as an alternative for radiologists breast cancer diagnosis through initial detection and diagnosis of mammograms before undergoing the final diagnosis by radiologists. The research on CAD consists of 2 characteristics. The first one is detection and diagnosis of calcification while the second one is detection and diagnosis of mass. This research focuses on developing an algorithm for classifying masses based on their shape features. In the research, an experiment is conducted consisting of 3 processes of (1) preprocessing, (2) mass boundary segmentation and (3) mass classification. Preprocessing is a process of preparation of mammographic masses to be used in the experiment. Mass boundary segmentation is a process of segmenting only the boundary of masses using an algorithm called Active Contour Models (snake). Finally, mass classification is a process of classifying masses into groups defined using Fishers linear discriminant analysis and based on 5 features of area, perimeter, compactness, radius and Fourier descriptor (FD). In the experiment, 40 mammographic masses derived from MIAS were used. It composed of 21 circumscribed and 19 speculated masses. However, when severity was considered, it was composed of 28 benign and 12 malignant masses. We have divided our experiment into 2 sub-experiments based on 5 shape features of area, perimeter, compactness, radius and FD. The first one performed classification based on class. It yielded 90% accuracy rate compared to a comparative method of Rangayyans method, using 4 features of FD, compactness, moment and chord length, which yielded 82.5% accuracy rate. The second one performed classification based on severity, which produced 72.5% accuracy rate compared to Rangayyans method, which provided 52.5% accuracy rate.