Type 2 Fuzzy Logic for mammogram breast tissue classification

BIRADS, Breast Imaging, Reporting and Data System, is a standard for preparing mammogram reports and it reduces confusion during mammogram image evaluation. In some cases, classifying BIRADS is a quite challenging task to the radiologist due to inter or intra personal variability. Therefore, this paper is aimed to develop a BIRADS classification model based on Type-2 Fuzzy Logic using mammogram data reports. The study begins with data aggregation comprising more than 100 images and their reports from Radiology Department of The National University of Malaysia Medical Center. Then, we select only complete data instances comprising calcification, masses and distortion as inputs and its expert decision as its target output. Next, Type-2 Fuzzy Logic based on Mamdani model produces membership linguistic variables automatically using those three inputs and an output. In advance, we also infer expert rules according to their experience for defuzzification phase. Lastly, the model was tested on three membership functions namely Gaussian, Trapezoidal and Triangular. The study shows that Triangular membership function based on expert driven rules outperform Gaussian and Trapezoid functions.