Classification of Breast Micro-calcifications As Benign or Malignant Using Subtraction of Temporally Sequential Digital Mammograms and Machine Learning

Cancer ranks as the second leading cause of mortality worldwide with breast cancer accounting for approximately 20% of all new cancer cases reported globally. Mammography is the most effective screening tool for the early diagnosis of breast cancer. However, the current practice of evaluating mammograms by two radiologists, and a third in case of disagreement, highlights the challenges faced even by experts in identifying potential abnormalities. To address these challenges, Computer-Aided Diagnosis (CAD) systems are being developed to assist radiologists in breast cancer diagnosis. This study proposes a classification approach for biopsy-confirmed benign and malignant Micro-Calcifications (MCs), using subtraction of temporally sequential digital mammograms combined with feature-based machine learning. The algorithm's performance was evaluated on a dataset retrospectively collected for this work, including 128 images from 32 patients, with precisely annotated MC locations and biopsy confirmations. Several features were extracted and a combination of feature selection algorithms was employed to identify the most critical subset of features. Ten classifiers were evaluated using leave-one-patient-out and k-fold cross-validation (k = 4 and 8). An Artificial Neural Network (ANN) achieved the highest performance, with 90.63% sensitivity, and 85.39% accuracy. These findings demonstrate the potential of the proposed algorithm to be translated into clinical practice as a second-reading tool for the classification of breast MCs as biopsy-confirmed benign or malignant.