Wavelet Statistical Feature Selection Using Genetic Algorithm with Fuzzy Classifier for Breast Cancer Diagnosis

Breast cancer diagnosis at its early stage is achieved through mammogram analysis. This paper presents a genetic fuzzy system (GFS) for feature selection and mammogram classification. Mammogram image is decomposed into sub-bands using wavelet transform. Wavelet statistical features are obtained from 100 biggest wavelet coefficients from each sub-band. From each level of decomposition, 20 WSFs are extracted. Therefore, total 80 WSFs are extracted from four levels of decomposition. At first level, 20 WSFs are given to GFS, which selects five features with classification accuracy of 60.94%. For second level, 18 features are selected from 40 features and classification accuracy of 80.66% is obtained. Further, at third level, 18 features are selected from 60 features with classification accuracy of 85.25%. At last, for fourth level, 21 features are selected from 80 features and classification accuracy improved to 93.77%.