A Dual Edge Spatial Jacobian Image Graph for Interpretable Diabetic Retinopathy Grading

2026-06-24 · Source: stat.ML updates on arXiv.org · Field: Health & Wellbeing — Artificial Intelligence & Machine Learning, Health & Medical Research, Medical Devices & Health Technology · Depth: Expert, medium

Summary

A dual-edge spatial–Jacobian image graph is introduced for interpretable diabetic retinopathy (DR) grading from color fundus photographs. This framework represents each image as a graph node, integrating 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 system constructs two distinct edge branches: a spatial branch ($X_{12}$) encoding vessel–lesion geometry and a Jacobian branch ($X_{34}$) modeling embedding–biomarker sensitivity. These branches are fused using a lightweight two-token attention module. Evaluated on 2,910 non-augmented APTOS images, the full graph achieved 0.8076 accuracy, 0.8312 quadratic weighted kappa, 0.5915 macro-F1, and 0.9330 adjacent-grade accuracy. For referable DR, it reached 0.9055 accuracy and 0.9711 AUROC. The framework is designed as an explainable representation-learning tool for lesion–biomarker hypothesis generation, rather than a deployment-ready clinical classifier.

Key takeaway

For AI Scientists developing diagnostic tools for ophthalmology, you should consider graph-based approaches that explicitly model multi-modal relationships. This framework demonstrates how integrating spatial vessel-lesion geometry and embedding-biomarker sensitivity can yield more interpretable DR grading, moving beyond simple image-level labels. Your focus should shift towards building explainable representations that facilitate hypothesis generation, rather than solely optimizing for raw predictive performance in clinical deployment.

Key insights

Interpretable DR grading benefits from graph representations linking spatial lesion-vessel geometry and embedding-biomarker sensitivity.

Principles

• Clinical interpretability requires understanding lesion distribution and vascular biomarkers.
• Graph edges can encode interpretable relationships between different data streams.
• Fusing diverse evidence streams enhances both prediction and explanation.

Method

Fundus images are represented as graph nodes with four evidence streams. Spatial and Jacobian edge branches are constructed and fused via two-token attention for DR grading.

In practice

• Combine vessel maps ($X_{1}$) and lesion evidence ($X_{2}$) for spatial interaction channels.
• Use lesion-based embeddings ($X_{3}$) and morphometric biomarkers ($X_{4}$) for Jacobian sensitivity.
• Employ attention mechanisms to fuse multi-modal graph edges for interpretation.

Topics

• Diabetic Retinopathy Grading
• Image Graph
• Explainable AI
• Multi-modal Fusion
• Oculomics
• Retinal Biomarkers

Code references

• Inamullah-Colab/dual-edge-dr-graph-xai

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• Geometry-Aware Superpixel Graph Transformer with Metadata for Skin Lesion Classification — Explicitly modeling skin lesions as geometry-aware superpixel graphs with integrated metadata improves classification accuracy.
• Improving clinical interpretability of linear neuroimaging models through feature whitening — A novel whitening method improves neuroimaging model interpretability by decorrelating anatomically informed brain regions.
• HSQ-VLM: A Novel Spatially-Constrained Quadrant Segmentation VLM Model for Explainability in Diabetic Retinopathy — HSQ-VLM provides explainable DR diagnostics by segmenting fundus images with fovea-centered anatomical precision.
• Leptomeningeal Collateral Detection on DSA via Vessel-Graph Neural Networks — A hybrid graph-pixel neural network enables objective, individualized detection of leptomeningeal collaterals on high-resolution DSA images.
• Systematically decoding pathological morphologies and molecular profiles with unified multimodal embedding — Multi-Embed unifies morphological and molecular data for superior disease understanding and prediction.
• Decision processes in 3D structural MRI schizophrenia classification evaluated with saliency maps — Explainable AI methods are crucial for validating deep learning models in medical diagnostics, revealing decision plausibility beyond performance metrics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.