Mammogram image analysis - a soft computing approach

Abstract

In this thesis CAD system is designed to diagnose breast cancer through mammograms, using image processing techniques along with swarm intelligence tools and clustering method, such as Particle Swarm Optimization Algorithm (PSO), Artificial Bee Colony Optimization (ABC), Fuzzy Spatial Dependency C-Means (FSDCM) Clustering Algorithm and Artificial Neural Network. The detection of micro calcifications is performed in two phases: preprocessing and segmentation in the first phase and feature extraction, selection and classification in the second phase. 161 pairs of digitized mammograms obtained from the Mammography Image Analysis Society (MIAS) database is used to design the proposed diagnosing system. Initially, the film artifacts and labels are eliminated from the mammogram images and median filter is applied to eliminate the high frequency components from the image. Then the mammogram images are normalized. The suspicious region or micro calcifications is segmented using Fuzzy Spatial Dependency CMeans (FSDCM) Clustering Algorithm (FSDCM ), Particle Swarm Optimization Algorithm (PSO) and Markov Random Field (MRF) hybrid with Artificial Bee Colony Optimization (ABC) algorithm for mammogram images. The MRF and ABC based image segmentation method is a process seeking the optimal labeling of the pixels. The textural features can be extracted from the segmented mammogram image to classify the micro calcifications into benign, malignant or normal. The Surrounding Region Dependency Matrix textural analysis method is used to extract the fourteen Haralick features from the segmented image. The features are selected from the extracted set of features using Genetic Algorithm (GA), PSO and ABC.