Different BI-RADS breast cancer diagnosis using MobileNetV1 and vision transformer based on explainable artificial intelligence (XAI).
Authors
Affiliations (2)
Affiliations (2)
- Physics Department, Faculty of Science, Mansoura University, Mansoura, 35516, Egypt.
- Information Technology Department, Faculty of Computers and Information, Mansoura University, Mansoura, 35516, Egypt. [email protected].
Abstract
Breast cancer (BC) remains one of the leading causes of death among women in the world, depending on the requirement for precise, effective, and interpretable computer-aided diagnosis systems (CADs). In this work, a hybrid deep learning (DL) framework is presented for multi-class BI-RADS BC classification using mammographic images. This framework fuses MobileNetV1, a lightweight convolutional neural network (CNN), to capture fine-grained local features and combines it with a Vision Transformer (ViT) to model global contextual connections, thereby enabling corresponding representation learning through a dual-stream structure. The evaluation was performed on the publicly available King Abdulaziz University BC Mammogram Dataset (KAUBC), which includes multi-view mammograms (craniocaudal (CC) and mediolateral oblique (MLO)) arranged according to the BI-RADS classification scheme and characterized by class imbalance. Feature-level fusion is performed, followed by a bagging-based logistic regression (LR) classifier to enhance robustness and decrease prediction conflict. The proposed approach was extensively analyzed using 5-fold cross-validation and compared with multiple state-of-the-art CNN and transformer models, each fused with various machine learning (ML) classifiers. The experimental results demonstrate higher and stable performance across all BI-RADS categories, with accuracy (ACC), sensitivity (SEN), and specificity (SPE) exceeding 99%. In addition, explainable artificial intelligence (XAI) techniques, including Grad-CAM and Grad-CAM++, were applied to provide clinically interpretable visual explanations by highlighting diagnostically relevant regions in mammograms. These results indicate that the proposed MobileNetV1-ViT-Bagging framework recommends an effective, computationally structured, and explainable solution for multi-class BI-RADS BC diagnosis, with strong potential for clinical decision-support applications.