HMC-net: a ResNet fused hierarchical multi-scale cross-attention architecture for mammographic breast malignancy recognition incorporating explainable AI
Abstract
Accurate and understandable interpretation of mammograms is fundamental to dependable identification of breast cancer, which facilitates clinical trust and usefulness. The framework proposed in this paper is known as the ResNet50HierarchicalMultiScaleCross-Attention (HMC) Network, building upon ResNet50 by embedding Hierarchical Self Attention and Multi-Scale Cross Attention modules for enriched feature representation toward mammogram-based detection of breast cancer. Intra-layer self-attention together with inter-layer cross-attention may enable the model to learn local as well as global patterns and hence improve performance for classification tasks on the MIAS dataset. For explainability issues, Grad-CAMs, Grad-CAM++, and Score-CAM are included in the architecture. Such methods yield heatmaps whose more clinically relevant regions can be made explicit for automated diagnostics to become transparent and trusted. With a 5-fold cross-validation run, it attained a mean accuracy of 0.9972 (± 0.05), a mean precision value of 0.9851 (± 0.13), a recall of 0.9899 (± 0.19), F1-score that amounted to 0.9864 (± 0.07). The values for AUC-ROC and specificity were found to be quite high at 0.99(± 0.01) and about 0.9978(± 0.09), respectively, basically beating most baseline models like ResNet50, VGG19, VGG16, and ViT among others in performance metrics variance as indicated by the Friedman test (p-value=0.002post-hoc test, HMC-Attention clearly outperformed standard ResNet50; learning…
Read the paper
Related studies
- In silico toxicology protocols
- QM7-X, a comprehensive dataset of quantum-mechanical properties spanning the chemical space of small organic molecules
- Quantum Informed Machine-Learning Potentials for Molecular Dynamics Simulations of CO2's Chemisorption and Diffusion in Mg-MOF-74
- In-silico strategies in nano-drug design: Bridging nanomaterials and pharmacological applications
- mRNA codon optimization with quantum computers
- The computational power of the human brain
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.