cs.CVNov 14, 2026

Lesion-Aware Explainable AI for Diabetic Retinopathy Grading from Fundus Images

Authors: Truong Thinh Nguyen, Hai Phong Nguyen, Truong Thanh Hung Nguyen, Giang Son Tran, Hung Cao

Organizations: University of Science and Technology of Hanoi, Hanoi, Vietnam · University of New Brunswick, Fredericton, New Brunswick, Canada

Abstract

Post-hoc explanations of deep-learning models for diabetic retinopathy (DR) grading are difficult to trust because different attribution methods highlight different retinal regions for the same image, and a visually plausible heatmap does not show that the highlighted pixels drive the prediction. We propose the Consensus-Calibrated Explanation Map (CCEM), which fuses Grad-CAM++, Ada-SISE, and SmoothGrad Integrated Gradients (SG-IG) using image-specific weights derived from annotation-free reliability cues: deletion sensitivity, inter-method agreement, compactness, retinal containment, and peak structure. CCEM explains a frozen ordinal grader, an EfficientNet-B7 with CBAM and a scalar regression head, pre-trained on EyePACS 2015 in Stage 1 and fine-tuned on APTOS 2019 in Stage 2 with quadratic weighted kappa (QWK)-optimized thresholds; it reaches a QWK of 0.920 on the hidden Kaggle test set and 0.559 on the external MAPLES-DR cohort. Explanations are evaluated on MAPLES-DR, 198 MESSIDOR images with expert pixel-level lesion annotations that are never used for training, threshold fitting, or fusion. On the 193 images with annotated lesions, CCEM obtains the highest Soft Pointing Game score (69.61%; 95% CI 63.7-75.3), defined as the ratio of the strongest lesion-supported saliency to the global map maximum rather than a lesion detection rate, together with the lowest deletion AUC (0.570) and the best deletion-insertion balance (0.115); each gain is significant against the strongest single explainer on paired tests. Annotation-free, per-image reliability weighting therefore turns three disagreeing explainers into one map whose peak evidence is more lesion-supported and more faithful to the model than that of any single explainer.

Explore similar work

Apr 25, 2026

From Pixels to Explanations: Interpretable Diabetic Retinopathy Grading with CNN-Transformer Ensembles, Visual Explainability and Vision-Language Models

The quality of diabetic retinopathy (DR) screening relies on the ability to correctly grade severity; however, many deep-learning (DL) classifiers cannot be easily interpreted in the clinical context. This study presents a methodology that combines strong discriminative models with multimodal explanations, converting retinal pixels into clinically interpretable outputs. Using the APTOS 2019 benchmark, we evaluated six representative CNN- and transformer-based backbones under a controlled protocol with stratified five-fold cross-validation. We then compared ensembling strategies (hard voting, weighted soft voting, stacking) and investigated a hybrid class-level fusion variant to exploit grade-specific advantages. For interpretability, we produced Grad-CAM++ visual attribution maps and short textual rationales using vision-language models (VLMs) conditioned on the fundus image and classifier outputs under conservative prompting constraints. Modern CNN backbones (ResNet-50 and ConvNeXt-Tiny) provided the strongest single-model baselines, with cross-validated QWK up to 0.919 and 0.914, respectively. Ensembling improved ordinal agreement, and weighted soft voting was the most consistent across folds (QWK 0.934 +/- 0.017). Hybrid class-level fusion was competitive but did not yield a statistically reliable improvement over standard fusion in paired fold comparisons (Holm-adjusted p >= 1.000). For explanation quality, Grad-CAM++ offered plausible but coarse localization, and VLM rationales were generally grade-consistent. Quantitatively, VLM variants showed a trade-off between clinical completeness and template-level semantic similarity (coverage 0.700 vs. BERTScore 0.072), while image-text alignment was comparable (CLIPScore approximately 0.34).
Oct 28, 2026

CalibCause: A Quality-Gated, Interpretable and Calibrated Deep Learning Framework for Diabetic Retinopathy Grading

Most deep learning systems for diabetic retinopathy (DR) grading are evaluated on a single data source, with limited evidence of cross-site generalisation, uncertainty calibration, and interpretability. We propose CalibCause, a quality-gated mixture-of-experts framework that combines per-expert visual explanations, label-free image-quality conditioning, knowledge distillation, and calibrated confidence with deferral. We evaluated the framework using 51,310 fundus images from EyePACS, DDR, and APTOS 2019, with IDRiD and DeepDRiD for external testing. Evaluation included pooled testing, leave-one-environment-out experiments, paired bootstrap confidence intervals, and local grade-threshold adaptation. On the pooled test split, the label-free distilled student achieved quadratic weighted kappa (QWK) of 0.863 and accuracy of 0.865 with 27.8 million parameters, 52% fewer than its teacher. Across held-out environments, the label-free model achieved mean QWK of 0.741 compared with 0.743 for a plain ConvNeXt-Tiny baseline, while adversarial invariance did not improve cross-site grading accuracy. Refitting grade thresholds using 200 labelled images increased held-out EyePACS QWK from 0.533 to 0.621 for the label-free model. After vector scaling, expected calibration error ranged from 0.025 to 0.036 across CalibCause variants, and deferring the least confident 20% of images increased student accuracy from 0.865 to 0.923 on the remaining images. These findings demonstrate that quality-aware grading, interpretable expert routing, model compression, and calibrated deferral can be integrated without sacrificing cross-site performance relative to a plain baseline, although reliable deployment still requires local validation and adaptation.
Aug 28, 2026

Explainable Diabetic Retinopathy Classification Using Vision Foundation Models

Diabetic retinopathy (DR) is a leading cause of preventable blindness, and automated screening must be both accurate and trustworthy. We study an explainable DR classification framework built on vision foundation models and three transfer learning strategies. Three backbones, DINOv2, CLIP, and Vision Transformer (ViT), were adapted by full fine-tuning, linear probing, and Low-Rank Adaptation (LoRA). Models were trained and internally evaluated on the ODIR dataset and externally evaluated on APTOS to assess generalization. DINOv2-LoRA achieved the highest internal AUROC (0.758), whereas DINOv2 and ViT full fine-tuning achieved the highest external AUROC (0.920). Probability calibration was assessed with reliability analysis after isotonic regression. For explainability, Grad-CAM and HiResCAM maps were compared with expert-annotated lesion masks from the IDRiD dataset using Dice, Intersection over Union (IoU), and Pointing Game metrics. The results show that foundation models, particularly DINOv2, provide strong predictive performance, and that LoRA is a parameter-efficient alternative to full fine-tuning. Spatial overlap between explanation maps and annotated lesions was limited (mean Dice of about 0.13 for the combined lesion mask), which highlights the value of quantitatively testing whether model attention corresponds to clinically relevant retinal lesions.