Measuring lesion–vascular context for diabetic retinopathy grading and retinal phenotyping

BACKGROUND/OBJECTIVES: Diabetic retinopathy (DR) grades summarise severity but do not quantify the vascular surroundings of individual lesions. We investigated whether measured lesion–vascular context improves lesion-based grading and characterises vascular variation beyond severity. SUBJECTS/METHODS: This cross-sectional secondary analysis of public datasets matched 2,874 APTOS photographs to lesion predictions and AutoMorph vascular measurements (1,724 training, 575 validation, 575 test). Five models trained exclusively on DDR annotations predicted four lesion families. Registered anatomy yielded 24 spatial descriptors and 1,584 lesion–biomarker products. Structured ordinal models, an anatomical graph and six frozen image encoders evaluated grading. Separate grade-adjusted analyses assessed vascular associations and prediction. RESULTS: Structured context increased test quadratic weighted kappa (QWK) from 0.764 to 0.827 (difference 0.063; 95% confidence interval [CI] 0.028–0.099). The graph achieved QWK 0.849 and macro area under the receiver operating characteristic curve 0.924 (QWK difference 0.085; 95% CI 0.050–0.119); both comparisons with lesion-only grading had false-discovery-rate-adjusted 𝑞 < 0.001. Individual contextual additions were not separately supported. Graph sensitivity for proliferative DR was 0.316, compared with 0.605 for structured context. Grade-adjusted analysis supported 103 lesion–biomarker associations, and lesion counts improved prediction of 26/66 vascular measurements, predominantly calibre/width and fractal measures. CONCLUSIONS: Combined lesion–vascular representations improved lesion-only grading and characterised vascular variation within severity categories. The measurements provide inspectable anatomical evidence, but segmentation limitations and classwise errors preclude claims of clinical readiness or prognostic validity.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11777044/v1
Primary Topic
Retinal Imaging and Analysis
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Measuring lesion–vascular context for diabetic retinopathy grading and retinal phenotyping

Shoaib Jameel, Inamullah Inamullah
3 links
Research Square
Retinal Imaging and Analysis
preprint

Measuring lesion–vascular context for diabetic retinopathy grading and retinal phenotyping

Shoaib Jameel, Inamullah Inamullah
preprint en

Abstract

BACKGROUND/OBJECTIVES: Diabetic retinopathy (DR) grades summarise severity but do not quantify the vascular surroundings of individual lesions. We investigated whether measured lesion–vascular context improves lesion-based grading and characterises vascular variation beyond severity. SUBJECTS/METHODS: This cross-sectional secondary analysis of public datasets matched 2,874 APTOS photographs to lesion predictions and AutoMorph vascular measurements (1,724 training, 575 validation, 575 test). Five models trained exclusively on DDR annotations predicted four lesion families. Registered anatomy yielded 24 spatial descriptors and 1,584 lesion–biomarker products. Structured ordinal models, an anatomical graph and six frozen image encoders evaluated grading. Separate grade-adjusted analyses assessed vascular associations and prediction. RESULTS: Structured context increased test quadratic weighted kappa (QWK) from 0.764 to 0.827 (difference 0.063; 95% confidence interval [CI] 0.028–0.099). The graph achieved QWK 0.849 and macro area under the receiver operating characteristic curve 0.924 (QWK difference 0.085; 95% CI 0.050–0.119); both comparisons with lesion-only grading had false-discovery-rate-adjusted 𝑞 < 0.001. Individual contextual additions were not separately supported. Graph sensitivity for proliferative DR was 0.316, compared with 0.605 for structured context. Grade-adjusted analysis supported 103 lesion–biomarker associations, and lesion counts improved prediction of 26/66 vascular measurements, predominantly calibre/width and fractal measures. CONCLUSIONS: Combined lesion–vascular representations improved lesion-only grading and characterised vascular variation within severity categories. The measurements provide inspectable anatomical evidence, but segmentation limitations and classwise errors preclude claims of clinical readiness or prognostic validity.

Research Square
University of Southampton (GB)
Good health and well-being
Retinal Imaging and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon