Enhancing Breast Mass Diagnosis: Leveraging Temporal Subtraction for Accurate Classification of Benign and Malignant Masses.

Despite recent advancements, breast cancer continues to be a significant cause of mortality among women worldwide. While mammography has notably reduced mortality rates, accurate classification of breast masses in mammograms remains a challenge. This study proposes a novel approach for the detection of important Regions of Interest (ROIs) and their classification as normal (0), benign (1), or malignant (2) masses. This approach exploits the subtraction of temporally sequential digital mammograms, combined with machine learning. The algorithm was evaluated on a new dataset consisting of 352 images from 88 patients with precisely annotated mass locations. A comprehensive feature extraction process yielded 98 features, which, were subsequently, ranked using eight different feature selection algorithms to identify the most discriminative characteristics. Ten classifiers were evaluated with leave-one-patient-out and k-fold cross-validation. An Artificial Neural Network (ANN) emerged as the most effective classifier, achieving 99.4% overall accuracy, 0.97 AUC for class 0, 0.91 AUC for class 1, and 0.93 AUC for class 2. These results are a significant improvement compared to existing state-of-the-art methods. They also underscore the efficacy of utilizing temporally consecutive mammograms in combination, with advanced machine learning algorithms, for the precise classification of the detected ROIs as normal, benign, or malignant. When clinically applied, the outcome of this study could significantly improve the precision of breast cancer diagnosis, potentially leading to better patient outcomes and more personalized treatment approaches.