An improvement of texture-based classification of microcalcification clusters in mammography using PSO-SVM approach

Texture-based analysis of microcalcification (MC) clusters provides a robust tool for the development of a computer-aided diagnosis (CADx) in mammography. Unlike shape-based schemes, a texture approach does not require a microcalcification segmentation stage. This paper presents a new texture-based CADx that accomplishes feature selection and classification stages using a PSO-SVM framework. The proposed CADx mainly consists of texture feature extraction and heuristic parameter selection stages. The first stage characterizes MC clusters using 28 texture features from graylevel co-occurrence matrices (GLCMs). The second stage involves a heuristic feature selection and performance optimization of a kernel-based support vector machine (SVM) classifier using a PSO-SVM approach. This step uses a particle swarm optimization (PSO) algorithm to heuristically search for the most discriminative texture features and to find the optimal SVM learning model that comprises the regularization and kernel parameters. Testing the proposed parameter selection approach using MC clusters from the mini-MIAS dataset produced perfect classification accuracy and demonstrated a promising performance of parameter selection using PSO-SVM method.