aiqbench analysis
This study investigates the cross-domain generalization limits of a dual-resolution deep learning framework designed for diabetic retinopathy grading. By employing an attention-gated fusion mechanism and ordinal regression, the authors quantify significant performance drops when models encounter unseen imaging domains, highlighting that while relative severity ordering remains robust, absolute classification accuracy degrades substantially under domain shift.
Dual-Resolution Attention-Gated Deep Learning with Ordinal Regression for Diabetic Retinopathy Grading: A Quantified Assessment of Cross-Domain Generalization
arXiv:2604.17341v2 Announce Type: replace-cross Abstract: Diabetic retinopathy (DR) is a leading cause of preventable blindness, and automated grading could extend screening capacity. However, most reported DR models are validated only on the dataset they were trained on, leaving their behaviour under real screening variability unmeasured. This study presents a dual-resolution grading framework and quantifies how far performance falls when the imaging domain shifts. Two EfficientNet backbones process complementary views of each fundus image: B0 receives Ben Graham-normalised input at 224x224, emphasising vascular structure, while B3 receives CLAHE-enhanced input at 300x300, emphasising focal lesions. A learnable attention gate fuses the branches per image, and an ordinal binary-decomposition head models severity as an ordered scale rather than as unordered categories. Training used a combined set of 4,149 images (APTOS 2019, n = 2,929; Messidor-2 training portion, n = 1,220); evaluation used a held-out APTOS split (n = 733) and a Messidor-2 test set (n = 524) excluded from training and from all model selection. Quadratic weighted kappa was 0.882 (95% CI 0.853-0.906) on APTOS and 0.679 (95% CI 0.613-0.735) on Messidor-2 for this run, a significant gap of 0.202 (95% CI 0.142-0.273); across three random seeds the held-out kappa was 0.689 +/- 0.021. Critically, accuracy fell 19.3 points while 93.7% of predictions stayed within one grade of reference: ordering survives domain shift, threshold placement does not. Referable-DR sensitivity fell from 0.879 to 0.620.
Abstract:Diabetic retinopathy (DR) is a leading cause of preventable blindness, and automated grading could extend screening capacity. However, most reported DR models are validated only on the dataset they were trained on, leaving their behaviour under real screening variability unmeasured. This study presents a dual-resolution grading framework and quantifies how far performance falls when the imaging domain shifts. Two EfficientNet backbones process complementary views of each fundus image: B0 receives Ben Graham-normalised input at 224x224, emphasising vascular structure, while B3 receives CLAHE-enhanced input at 300x300, emphasising focal lesions. A learnable attention gate fuses the branches per image, and an ordinal binary-decomposition head models severity as an ordered scale rather than as unordered categories. Training used a combined set of 4,149 images (APTOS 2019, n = 2,929; Messidor-2 training portion, n = 1,220); evaluation used a held-out APTOS split (n = 733) and a Messidor-2 test set (n = 524) excluded from training and from all model selection. Quadratic weighted kappa was 0.882 (95% CI 0.853-0.906) on APTOS and 0.679 (95% CI 0.613-0.735) on Messidor-2 for this run, a significant gap of 0.202 (95% CI 0.142-0.273); across three random seeds the held-out kappa was 0.689 +/- 0.021. Critically, accuracy fell 19.3 points while 93.7% of predictions stayed within one grade of reference: ordering survives domain shift, threshold placement does not. Referable-DR sensitivity fell from 0.879 to 0.620.
v2: added multi-seed ablation; corrected component-contribution claims; expanded evaluation with figures; code and data available (Zenodo DOI: https://doi.org/10.5281/zenodo.21739226)
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2604.17341 [cs.CV]
(or arXiv:2604.17341v2 [cs.CV] for this version)
From: Afshan Hashmi Dr. [view email] [v1] Sun, 19 Apr 2026 09:27:37 UTC (2,297 KB) [v2] Mon, 3 Aug 2026 11:32:38 UTC (19,292 KB)
Full text under the publisher's open licence · 285 words
Related coverage
Multi-Scale Feature Attention Network for Polymer Classification using THz Dual-Comb Spectroscopy
arXiv · Jun 8, 2026
Game Design Creativity evals
LessWrong · Oct 12, 2026
News
Quantum Computing Report · Jun 7, 2026
Multi-Scale Feature Attention Network for Polymer Classification Using Terahertz Spectroscopy
arXiv · Jun 9, 2026