A Generalizable CNN Enhancement for Robust Mammogram Classification

Hana Mechria, Lamjed BEN SAID

Keywords: Convolutional Neural Networks, Mammogram images classification, Imbalanced data, multiclass classification

Abstract: Deep convolutional neural networks (CNNs) have demonstrated outstanding capabilities in analyzing mammogram images. However, their overall performance is frequently hampered by class imbalance and the inherent complexity of mammogram images. In this work, we propose a novel generalizable enhancement that can be seamlessly integrated into any CNN architecture aimed at improving classification outcomes for mammographic images. In order to rigorously evaluate its effectiveness, the proposed approach was applied to six commonly used CNN models and assessed on an imbalanced multiclass mammogram images dataset. The experimental results showed consistent and significant improvements in all key performance metrics, such as accuracy, precision, recall, F1-score, Precision-Recall Curve (PRC), and Area Under the Curve (AUC), which highlights the robustness and adaptability of our method. This enhancement provides a generalizable strategy for strengthening CNN-based mammogram classification systems, thereby promoting more reliable computer-aided diagnosis in breast cancer screening.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 17634