Enhancing Clinical Trust in Diabetic Retinopathy Screening: A Multi-Strategy Rule Extraction Framework from CNN
2026 IEEE International Conference on Digital Health (ICDH) · Sydney, Australia · 2026 · pp. 135-144
DOI: 10.1109/ICDH72779.2026.00024
• Published at IEEE ICDH 2026, an international conference on digital health. • Developed an XAI framework combining EfficientNet-B5, Grad-CAM++, and rule extraction with Decision Tree, RIPPER, and LIME; the post-hoc strategy preserves the CNN's 82.95% accuracy and provides symbolic explanations for 88.2% of predictions.