Search papers, labs, and topics across Lattice.
This paper introduces a dual-edge spatial-Jacobian image graph for the interpretable grading of diabetic retinopathy (DR) from color fundus photographs, integrating multiple evidence streams to enhance clinical interpretability. By representing each image as a graph node with aligned data on vessel information, lesion evidence, contrastive embeddings, and morphometric biomarkers, the model effectively encodes the geometry of lesions in relation to vascular structures. The proposed framework achieves high accuracy metrics, including 0.9055 for referable DR, positioning it as a valuable tool for generating hypotheses about lesion-biomarker relationships rather than a direct clinical classifier.
The dual-edge spatial-Jacobian image graph reveals intricate relationships between retinal lesions and vascular biomarkers, achieving up to 0.9055 accuracy in referable diabetic retinopathy grading.
Automated diabetic retinopathy (DR) grading from colour fundus photographs can achieve strong predictive performance, but clinical interpretation requires more than an image-level label. It requires understanding how lesion evidence is distributed around retinal vessels and how this evidence relates to quantitative vascular biomarkers. We present a dual-edge spatial-Jacobian image graph for interpretable DR grading. Each fundus image is represented as a graph node with four aligned evidence streams: AutoMorph vessel information ($X_1$), DR-XAI-style lesion evidence maps ($X_2$), a 128-dimensional lesion-based contrastive image embedding ($X_3$), and AutoMorph morphometric biomarkers ($X_4$). The spatial edge branch ($X_{12}$) encodes vessel-lesion geometry, while the Jacobian branch ($X_{34}$) models embedding-biomarker sensitivity. Lightweight two-token attention fuses both edge families into a final image graph. On 2,910 matched non-augmented APTOS images, the full graph achieves 0.8076 accuracy, 0.8312 quadratic weighted kappa, 0.5915 macro-F1, and 0.9330 adjacent-grade accuracy; referable DR reaches 0.9055 accuracy and 0.9711 AUROC. The framework is positioned as an explainable representation-learning tool for lesion-biomarker hypothesis generation, rather than as a deployment-ready clinical classifier. The code is available at https://github.com/Inamullah-Colab/dual-edge-dr-graph-xai.
