Abstract

Breast cancer is one of the most common diseases affecting women worldwide, and early diagnosis plays an important role in improving treatment outcomes and reducing mortality. Among the available imaging techniques, breast ultrasound is widely used because it is safe, non-invasive, and particularly effective for examining dense breast tissue. However, interpreting ultrasound images remains difficult due to image noise, low contrast, and the visual similarity between benign and malignant lesions. This study presents an explainable deep learning framework for breast ultrasound image classification by combining EfficientNet-B0 with the Convolutional Block Attention Module (CBAM). In addition, Gradient-weighted Class Activation Mapping (Grad-CAM) was incorporated to provide visual explanations for the model's predictions and improve their interpretability. The proposed framework was evaluated using the publicly available BUSI dataset, which contains 780 breast ultrasound images divided into three classes: normal, benign, and malignant. The dataset was split into 70% for training, 15% for validation, and 15% for testing. To evaluate the effectiveness of the proposed method, two baseline models, ResNet50 and EfficientNet-B0, were implemented and compared with the proposed EfficientNet-B0 + CBAM model. ResNet50 was trained for 2 epochs, while EfficientNet-B0 and EfficientNet-B0 + CBAM were trained for 5 epochs using the Adam optimizer with a learning rate of 0.0001 and a batch size of 16. The experimental results showed that the proposed model achieved a test accuracy of 81.20%, compared with 79.49% obtained by the baseline EfficientNet-B0 model. In addition, the recall for the malignant class increased from 0.6452 to 0.6774, demonstrating an improved ability to identify malignant breast lesions. The generated Grad-CAM visualizations also showed that the model focused on clinically relevant regions of the ultrasound images, providing useful visual explanations for its predictions. Overall, the results indicate that integrating CBAM with EfficientNet-B0 can improve classification performance while Grad-CAM enhances the transparency of the decision-making process. These findings suggest that the proposed framework has the potential to support radiologists by providing accurate predictions together with interpretable visual explanations.