A region-of-interest embedded graph neural architecture for gallbladder cancer detection

Saiful Islam, Md. Injamul Haque, Mushrat Jahan, Md. Zahid Hasan, Md. Awlad Hossen Rony, Kaniz Fatema, Taslima Ferdaus Shuva, Muhammad Ali Abdullah Almoyad, Abdullah Al-Mamun Bulbul, Md. Tanvir Rahman, Md Whaiduzzaman, Touhid Bhuiyan, Mohammad Ali Moni

Abstract: Highlights•The RGBNet with ROI embedding outperformed baseline models for the classification of complex medical images.•Statistically, the improvements made by the model are significant, with p-values less than 0.05.•The model performs well across various cancer detection datasets, showing its suitability for different medical scenarios.•Grad-CAM and Guided Grad-CAM visualizations improve interpretability, aiding clinical decision-making.•ROI embedding enhances feature extraction, enabling accurate identification of important regions in complex cases.

External IDs:doi:10.1016/j.rineng.2025.104624