Compact vision models match domain-specific foundation models for several retinal imaging classification tasks: A systematic benchmark
Open Access
- 18 August 2026
- journal article
- research article
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 21 (8) , e0356202
Abstract
Large domain-specific foundation models have been widely adopted for retinal image analysis, yet systematic evidence for their advantage over compact general-purpose architectures remains scarce. We benchmarked nine model configurations spanning 22.8M to 303M parameters (vision transformers, hierarchical Swin Transformers, ConvNeXt, and the domain-specific RETFound models) across four tasks: OCT 8-class disease classification, and three fundus photography tasks (DME severity, glaucoma detection, and DR severity grading). All models were evaluated under identical training conditions, with both pretrained (on natural-domain image datasets) and from-scratch initializations compared using Mann-Whitney U tests. Pretraining improved accuracy by 5.18–18.41 percentage points across all tasks (p < 0.05 throughout), with larger benefits for CFP modalities and harder tasks. Compact hierarchical models (27–29M parameters) matched or exceeded larger architectures on three of four tasks. For instance, the SwinV2-tiny architecture ranked first on OCT, DME, and GL classification. The domain-specific RETFound model (303M) achieved the highest accuracy only on the most challenging task (DR severity grading, where the most severe class is underrepresented at 8% of images), where it outperformed the best compact model by 1.54 percentage points. These results indicate that compact general-purpose models may be sufficient for most retinal classification benchmarks, and that domain-specific foundation models may add higher value mainly for severity grading tasks with skewed class distributions.Keywords
Funding Information
- Werner H. Spross-Stiftung (n/a)
This publication has 20 references indexed in Scilit:
- Oculomics: Current concepts and evidenceProgress in Retinal and Eye Research, 2025
- Oculomics: A Crusade Against the Four Horsemen of Chronic DiseaseOphthalmology and Therapy, 2024
- Deep Learning and Machine Learning Algorithms for Retinal Image Analysis in Neurodegenerative Disease: Systematic Review of Datasets and ModelsTranslational Vision Science & Technology, 2024
- A foundation model for generalizable disease detection from retinal imagesNature, 2023
- An interpretable transformer network for the retinal disease classification using optical coherence tomographyScientific Reports, 2023
- PAPILA: Dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessmentScientific Data, 2022
- IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045Diabetes Research and Clinical Practice, 2022
- Insights into Systemic Disease through Retinal Imaging-Based OculomicsTranslational Vision Science & Technology, 2020
- The meaning and use of the area under a receiver operating characteristic (ROC) curve.Radiology, 1982
- A Coefficient of Agreement for Nominal ScalesEducational and Psychological Measurement, 1960