Smart Digital Mammographic Screening System for Bulk Image Processing

Abstract

Treating breast cancer is easier at early stages. However, proper diagnosis is essential for this purpose. Mammography helps in early detection of cancer cells. Existence of masses, calcification and mammogram are the evidences that help radiologists in early cancer identification. This paper proposes a smart digital mammographic screening system for processing images in large volumes irrespective of the nature of images. Watershed segmentation is performed based on appropriate selection of internal and external markers using multiple threshold extended maxima transformations in this technique. Distinguishing between healthy breast tissue and masses can be performed efficiently using a two-stage classifier. Extreme Learning Machine based single layer feed forward network along with Bayesian classifier is used for reducing false positive areas. Feature vector with features like texture and contrast are calculated using these approaches. Digital Mammography Screening database (DMS) is created with 100 mammographic images for the purpose of evaluation. Further, online databases like Breast Cancer Database (BCDB) and BreakHis are also used for analysis. Overall sensitivity of the datasets using the Bayesian classifier and Extreme Learning Machine is found to be 85% and 90% respectively.