SVM Based Detection and Classification of Breast Abnormalities in Mammography Images
No Thumbnail Available
Date
2024-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Breast cancer is the second leading cause of death among females. Digital mammography is the most effective screening technique for early detection of breast cancer and other abnormalities. However, interpreting digital mammograms can be challenging due to the small differences in attenuation among various soft tissue structures in the female breast. This results in low-quality X-ray images, high background noise, and artifacts, making diagnosis difficult. Therefore, it is essential to develop an effective computer-aided diagnosis system to improve the detection and classification of abnormalities in mammogram images. The proposed Support Vector Machine (SVM)-based mammography image detection and classification method involves four stages: preprocessing, segmentation, feature extraction, and classification. During preprocessing, the images undergo denoising with median and winner filters, pectoral muscle removal using the seeded region grow algorithm, background removal through morphological operations, and image enhancement with range contrast adjustment and Contrast Limited Adaptive Histogram Equalization (CLAHE). Segmentation is performed using global thresholding. In the feature extraction stage, 24 features were calculated from locally computed Gray Co-occurrence Matrices (GLCM). Finally, SVM was employed to classify the mammogram images. The study utilized 274 mammograms, of which 102 were abnormal-samples (malignant or benign) and 172 were normal images taken from control subjects. Of these, 80% were used for training and 20% for testing the SVM classification model. Based on the performance metrics, the system has an 84.82% probability of detecting abnormalities in patients breast tumor with positive predictive value (PPV) of 94.06%. It correctly classifies mammogram images as normal in 96.3% of patients without breast tumor (specificity = 96.3%), with an overall accuracy of 91.6% with negative predictive value (NPV) of 90.17%. Among the various feature combinations, dissimilarity and inverse difference normalized features provided the highest AUC value of 0.92.
Description
Keywords
Mammogram, Mammography, Support Vector Machine, Pectoral Muscle, Classification.