Computerized classification of malignant and normal microcalcifications on mammograms: Using soft set theory

Abstract

This paper introduces an automatic classification of mammogram images by categorizing malignant or normal after segmenting the suspected region. Fuzzy and fuzzy soft set approaches have been used successfully to deal with diverse uncertainties, imprecision and vagueness in data. We have advocated a method of fuzzy soft set using fuzzy soft aggregation operator for solving the problem. The proposed method includes five important stages namely pre-processing, segmentation, feature extraction feature selection and classification. Preprocessing concentrates on the algorithmic development of automated noise removal, contrast enhancement and pectoral muscle removal. A hierarchical spatial fuzzy c- means clustering algorithm is employed for segmentation and the micro-calcification clusters have been identified with adaptive h-dome transformation. Among the various features present in the image, an initial set of 14 textural and statistical features are calculated from the micro-calcification clusters using Gray Level Co-occurrence Matrix (GLCM) for 00 angle and 3 pixel distances. Out of 14 extracted features from the clusters, we have selected seven parameters for fuzzy soft set implementation. Using fuzzy aggregation operator, the classification process has been performed for malignant or normal cases. The experiment is performed with 322 images and resulted in 94.09% accuracy. Our method also benefited in dimensionality reduction with high accuracy and therefore the overall performance is improved.