DCAF-Net: An Interpretable Dual-Pathway Deep Learning Framework for Accurate Gastrointestinal Endoscopic Image Classification

Gastrointestinal (GI) diseases such as colorectal cancer and inflammatory bowel disease, contribute substantially to global morbidity with colorectal cancer accounting for nearly 10% of all cancer diagnoses worldwide. This study aimed to develop and validate an interpretable, multi-model deep learning framework for accurate and generalizable classification of endoscopic images across eight clinically relevant GI categories. Deep features were fused using a Dual Context Attentive Feature Network (DCAF-Net) and optimized via the Aquila Optimizer without retraining. Multiple machine learning classifiers were evaluated. Performance was assessed using accuracy, precision, recall, F1-score, ROC analysis and statistical testing. Interpretability was examined using Grad-CAM++ and SHAP. The proposed framework offers a reliable, interpretable, and clinically applicable solution for AI-assisted GI disease classification, supporting real-world endoscopic deployment.