Examination of Mammogram Image Classification using Fuzzy Co-occurrence Matrix

Abstract

Breast cancer is one of the leading causes of mortality in women. Detection in earlier stage can help reduce the mortality rate. We develop a breast abnormalities detection system to help radiologists. The abnormalities considered are calcification (CALC), well-defined/circumscribed masses (CIRC), spiculated masses (SPIC), and architectural distortion (AD). The fuzzy co-occurrence matrix is utilized to generate 14 features in our system. We also utilized multi-class support vector machine with one-versus-all strategy as a classifier. The feature set generated from the gray level co-occurrence matrix is also used for the purpose of comparison. We found out that the features extracted from our fuzzy co-occurrence matrix have a better performance than those from the regular gray level co-occurrence matrix. The best blind test data set results for AD, SPIC, CALC, and CIRC detection from the feature set generated from our fuzzy co-occurrence matrix are 100% with 9.46 false positives per image (FPI), 90% with 13.72 FPI, 100% with 3.39 FPI, and 81.25% with 18 FPI, respectively. While those for AD, SPIC, CALC, and CIRC detection from the feature set extracted from the gray level co-occurrence matrix are 100% with 9.46 FPI, 70% with 4.45 FPI, 89.47% with 10.81 FPI, and 68.75% with 6.78 FPI, respectively. Our system performs better than other existing methods in AD and CALC detection. The result from our system is comparable with those methods in SPIC and CIRC detection. However, there is no pre-processing or ROI selection in our system at all.