Machine Learning Algorithms for Breast Cancer CADx System in the Mammography

A computer aided diagnosis and detection system (CADx) enhances the early detection accuracy for breast cancer which in turn save women life all over the world. The existing early detection CADx systems are costly and demands highly complex algorithms with complicated computation thus unsuitable for developing countries. The segmentation process is a key issue in designing CADx system for breast cancer and the most computationally costing. Hence, the contribution of this paper is proposing two CADx systems for breast cancer in mammographic datasets that compromise between efficiency, speed, and cost. The segmentation step in the first system is performed by a Particle Swarm Optimization (PSO) Algorithm while it is performed by a Gaussian Mixture Model (GMM) in the second proposed system. Textural, statistical and shape features are extracted from segmented ROIs and input for the classifier to distinguish the mammograms if normal, benign or malignant. The proposed method utilizes the nonlinear support vector machines for classification. Experiments are conducted on mini MIAS database. The resulted accuracy rates based on 10-folds cross validation for classifying normal from abnormal are 85.4% and 82.4% for PSO and GMM respectively. Besides, the ratios become 89.5%, 87.5% for classifying malignant from benign for PSO and GMM respectively. On the other hand, the experimental results show the overall average classification accuracy of 81% and 85% when classifying normal, malignant or benign. Confusion matrix is additionally evaluation metric that was also used to check and assess the process comprehensive balanced performances for both PSO and GMM based methods.