Computer aided diagnosis of breast cancer based on level set segmentation of masses and classification using ensemble classifiers

Abstract

Breast cancer has been ranked number one cancer in Indian females with rates occurrence of 25.8 per 1,00,000 females and death rate 12.7 among 1,00,000. Whereas in USA, the estimated new detected cases of breast cancer are 2,52,710 with 15% of all new cancer cases with estimated deaths of 40,610 women i.e. 6.8% of all cancer deaths in 2017. The mammograms can help an early detection of lesions by radiologists before it becomes incurable. But the degree of variations among the different radiologists is very high resulting in false positives and false negatives. So a great amount of research is focused on the design of computer aided diagnosis (CAD) for early detection of breast cancer from the mammograms. In quest of high accuracy, this paper aims at developing an automated computer aided diagnostic (CAD) system that detects the malignant neoplasms from the mammograms. A novel technique is used to remove the pectoral muscle in pre-processing stage in order to make the segmentation of suspicious masses from breasts easier. The active contour based level set method is used to segment the mammograms. The texture features being most implemented in mammographic analysis, the standard gray-level co-occurrence matrix (GLCM) texture descriptors by Harallick are extracted from the segmented images. Finally, ensemble classifier is used to classify the mammograms into normal and abnormal, and then abnormal ones into malignant and benign. The respective accuracies obtained are 97.46% and 82.05% respectively.