Mass detection and false positive reduction in mammographic images
- Publication Type:
- Issue Date:
Breast cancer is the most common type of cancer for women in America. Currently the most effective method for early detection of breast cancer is mammography. Mammography is the only widely accepted imaging method used for routine breast cancer screening. Masses are one of the important signs of breast cancer. However it is difficult to detect masses because masses have different size and shape and their features can be obscured or similar to the normal breast parenchyma. Reading mammograms is a demanding job for radiologists. A computer aided detection (CAD) system can provide a consistent second opinion to a radiologist and greatly improve the mass detection accuracy. In this thesis, a computer aided detection system is developed which can segment the breast region from the background in the whole mammographic image, detect the suspicious regions from the breast region and then classify the suspicious regions to mass or normal breast tissue. The suspicious regions in the full mammographic image can be found by contrast limited adaptive histogram equalization and thresholding. These suspicious regions can be masses or normal breast tissue (false positives). To reduce the number of false positives in mass detection, a feature selection and classification approach using particle swarm optimization (PSO) and support vector machine (SVM) is proposed. Firstly, texture features are derived from the gray level co-occurrence matrix (GLCM) of each suspicious region. A PSO and SVM based feature selection is proposed to determine the significant features. The significant features found by PSO-SVM based feature selection are used by the SVM classifier to classify the suspicious region to mass or normal breast tissue. One advantage of the proposed mass detection system is that it can detect different types of masses, including spiculated, circumscribed and ill-defined masses from the whole mammographic image. The number of false positives in mass detection can be reduced by the PSO and SVM based feature selection and mass classification method proposed in this thesis. Experimental results show that the proposed PSO-SVM based feature selection technique can find the significant features that can improve the classification accuracy of SVM and perform better than other widely used feature selection methods. The proposed mass classification approach using PSO and SVM has better or comparable performance when compared to other state-of-the-art mass classification techniques, using sensitivity and specificity as the evaluation criteria. In order to perform accurate image segmentation of the mass from the suspicious region, a mass segmentation method by PSO based image clustering is proposed. Two new fitness functions are proposed which can improve the performance of image clustering by generating more compact clusters and larger inter-cluster distance. The proposed PSO based image clustering, with the new fitness function, can improve the segmentation of the mass from mammographic image. It has been shown experimentally that PSO based image clustering can have better mass segmentation performance when compared to K-means, a widely used clustering technique.
Please use this identifier to cite or link to this item: