Skip to content
Conference Open access

Attention-Enhanced ResNet-50 for Explainable Breast Cancer Ultrasound Classification

Sep 2026 · 2026 International Conference on Data Science and Its Applications (ICoDSA) · pp. 90-95 · 0 citations · 29 references

Abstract

The high mortality rate from breast cancer drives the need for objective and precise diagnostic methods to increase the chance of patient recovery. Ultrasound (US) is a commonly used screening modality due to its accessibility, but it is susceptible to speckle noise and high visual variability, which can complicate lesion interpretation and lead to diagnostic inconsistencies. Although deep learning-based Computer-Aided Diagnosis (CAD) systems have shown potential for medical image classification, conventional Convolutional Neural Network (CNN) models remain limited in capturing inter-channel feature relationships and lack interpretability in decision-making. Furthermore, studies integrating attention mechanisms into residual networks for ultrasound-based multiclass breast cancer classification, using a combination of public datasets and Grad-CAM validation, remain limited. Therefore, this study developed the SE-ResNet-50 architecture by integrating the Squeeze-and-Excitation (SE) Block into ResNet-50 to improve feature representation and classification performance. The model was evaluated using a combination of BUSI and Mendeley datasets comprising 1,028 images, with 5-Fold Stratified Cross-Validation and an independent test set. Experimental results show that SE-ResNet-50 achieves 94.17% accuracy, 96.08% precision, 94.75% recall, and 95.27% F1-Score, with improvements of 4.85%, 7.48%, 3.08%, and 5.30%, respectively, compared to ResNet-50 as the baseline. In addition, Gradient-weighted Class Activation Mapping (Grad-CAM) visualization reveals more focused activation areas over lesions, thereby improving model interpretability and supporting the application of SE-ResNet-50 as a reliable CAD system for US-based breast cancer diagnosis.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.