Machine Learning in Mammography Classification: A Radiomics-Based Evaluation and Developments

Breast cancer remains the second leading cause of cancer-related mortality among women worldwide. Early detection and accurate classification of breast lesions, including masses and microcalcifications, are therefore essential to improving diagnostic precision and patient outcomes. This study investigates the performance of three supervised machine learning (ML) classifiers—Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and K-Nearest Neighbours (KNN)—in the context of a radiomics-based approach to mammographic image classification. The analysis was conducted using the matRadiomics toolbox on 1,219 mammograms obtained from the publicly available CBIS-DDSM dataset. All three classifiers—LDA, SVM, and KNN—were tested on the same mammographic subsets of masses and microcalcifications. LDA achieved the highest overall performance in terms of AUC (69.7% for microcalcifications; 62.9% for masses) and F1-score (0.741 and 0.638, respectively). While KNN showed slightly higher accuracy on masses (60%), its F1-score (0.576) remained lower than LDA. For microcalcifications, both SVM and KNN reached comparable F1-scores (0.741 and 0.742), though LDA retained the best AUC and overall balance. These findings highlight both the potential and the current limitations of traditional ML-based radiomics techniques in breast cancer classification. They also underscore the importance of exploring more advanced artificial intelligence approaches, such as deep learning and ensemble methods, to enhance diagnostic accuracy and clinical applicability in mammographic screening.