{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T17:07:27Z","timestamp":1790269647389,"version":"4.1.0"},"reference-count":53,"publisher":"MDPI AG","issue":"19","license":[{"start":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T00:00:00Z","timestamp":1790208000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Diagnostics"],"abstract":"<jats:p>Background\/Objectives: Retinal fundus photography is the most widely available means of screening for sight-threatening disease, but it is difficult to read: several conditions leave overlapping signs on the same photograph, the features that separate them are often subtle and confined to small regions of the retina, and a single image carries no systemic context, so an automated reader must resolve categories on which experienced clinicians themselves agree only moderately. Methods: We compared five deep learning architectures\u2014EfficientNet-B3, DenseNet-121, ViT-Base\/16, Inception-V3, and RETFound ViT-Large\u2014for simultaneous seven-class retinal disease classification on a multi-source fundus image benchmark, in a retrospective computational study using stratified five-fold cross-validation. We pooled 11,223 patient-eye fundus images from three publicly available, de-identified datasets (ODIR-5K, RFMiD, and APTOS 2019) spanning age-related macular degeneration (AMD), cataract, diabetic retinopathy (DR), glaucoma, hypertensive retinopathy, pathological myopia, and normal fundus; the cohort was markedly class-imbalanced (DR, 62.5%; normal, 25.6%; each remaining class &lt; 3%), and no new patient data were collected. Images were deduplicated by MD5 hash and split at the patient level to eliminate data leakage, with a fixed 15% held-out test set and five stratified cross-validation folds; training images received Ben Graham illumination normalization and class-balancing augmentation, while validation and test images were unaugmented. Three ImageNet-pretrained CNNs (EfficientNet-B3, DenseNet-121, Inception-V3), ViT-Base\/16, and RETFound ViT-Large (pretrained on 1.6 million retinal photographs via masked autoencoding) were trained; the primary outcome was the five-fold mean weighted-averaged F1 score, pairwise comparisons used paired t-tests on fold-level F1 (df= 4), and Grad-CAM was applied for explainability. Results: EfficientNet-B3 achieved the highest five-fold mean accuracy (95.53%; weighted-F1, 0.9507) and significantly outperformed RETFound ViT-Large (\u0394F1=+0.0255; t=9.70; p&lt;0.001), ViT-Base\/16 (\u0394F1=+0.0075; t=9.20; p&lt;0.001), and Inception-V3 (\u0394F1=+0.0103; t=8.38; p&lt;0.001), while performing comparably to DenseNet-121 (t=2.24; p=0.089); Grad-CAM showed clinically plausible activation localization for the CNN architectures. Conclusions: A compact, resource-efficient CNN thus matched or exceeded a substantially larger retinal foundation model and Vision Transformers, indicating that larger domain-specific foundation models are not inherently superior for heterogeneous retinal screening; because such compact models run on commodity hardware, they may offer more practical screening solutions for community, rural, and resource-limited settings where ophthalmologist coverage is scarcest. External validity across imaging sources and populations remains unproven, underscoring that prospective, multicenter validation must precede clinical deployment. A provenance audit of this pooled cohort subsequently found it confounded. The audit identified five failure modes, three of them decisive: image width alone reproduced the diabetic retinopathy label at 99.98% accuracy using no retinal pixels, source and label were almost perfectly dependent with a Cram\u00e9r\u2019s V of 0.9996, and diabetic retinopathy recall was exactly 1.0000 for all five architectures. The cohort was rebuilt from a single source with per-eye labels, and nine architectures were retrained under one protocol with external validation. The corrected headline is a macro-AUC of 0.930 internally and 0.978 externally, with a macro-F1 of 0.686. No architecture is declared superior, and the pooled-cohort conclusion does not stand.<\/jats:p>","DOI":"10.3390\/diagnostics16193109","type":"journal-article","created":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T16:29:53Z","timestamp":1790267393000},"page":"3109","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Comparative Evaluation of EfficientNet-B3, Vision Transformers, and RETFound for Seven-Category Retinopathy Fundus Image Classification"],"prefix":"10.3390","volume":"16","author":[{"given":"Majid","family":"Alutaibi","sequence":"first","affiliation":[{"name":"College of Computer Science and Engineering, University of Ha\u2019il, Ha\u2019il 81411, Saudi Arabia"}],"role":[{"vocabulary":"crossref","role":"author"},{"role":"formal-analysis","vocabulary":"credit"},{"role":"validation","vocabulary":"credit"},{"role":"software","vocabulary":"credit"},{"role":"methodology","vocabulary":"credit"},{"role":"visualization","vocabulary":"credit"},{"role":"data-curation","vocabulary":"credit"},{"role":"writing-original-draft","vocabulary":"credit"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9074-1029","authenticated-orcid":false,"given":"Meshari","family":"Alazmi","sequence":"additional","affiliation":[{"name":"College of Computer Science and Engineering, University of Ha\u2019il, Ha\u2019il 81411, Saudi Arabia"}],"role":[{"vocabulary":"crossref","role":"author"},{"role":"investigation","vocabulary":"credit"},{"role":"conceptualization","vocabulary":"credit"},{"role":"resources","vocabulary":"credit"},{"role":"supervision","vocabulary":"credit"},{"role":"project-administration","vocabulary":"credit"},{"role":"formal-analysis","vocabulary":"credit"},{"role":"writing-review-editing","vocabulary":"credit"}]}],"member":"1968","published-online":{"date-parts":[[2026,9,24]]},"reference":[{"key":"ref_1","unstructured":"World Health Organization (2019). World Report on Vision, World Health Organization."},{"key":"ref_2","doi-asserted-by":"crossref","first-page":"2402","DOI":"10.1001\/jama.2016.17216","article-title":"Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs","volume":"316","author":"Gulshan","year":"2016","journal-title":"JAMA"},{"key":"ref_3","doi-asserted-by":"crossref","first-page":"39","DOI":"10.1038\/s41746-018-0040-6","article-title":"Pivotal Trial of an Autonomous AI-Based Diagnostic System for Detection of Diabetic Retinopathy in Primary Care Offices","volume":"1","author":"Abramoff","year":"2018","journal-title":"npj Digit. Med."},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"2211","DOI":"10.1001\/jama.2017.18152","article-title":"Development and Validation of a Deep Learning System for Diabetic Retinopathy and Related Eye Diseases Using Retinal Images From Multiethnic Populations With Diabetes","volume":"318","author":"Ting","year":"2017","journal-title":"JAMA"},{"key":"ref_5","doi-asserted-by":"crossref","first-page":"1342","DOI":"10.1038\/s41591-018-0107-6","article-title":"Clinically Applicable Deep Learning for Diagnosis and Referral in Retinal Disease","volume":"24","author":"Ledsam","year":"2018","journal-title":"Nat. Med."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"1410","DOI":"10.1016\/j.ophtha.2018.02.037","article-title":"A Deep Learning Algorithm for Prediction of Age-Related Eye Disease Study Severity Scale for Age-Related Macular Degeneration From Color Fundus Photography","volume":"125","author":"Grassmann","year":"2018","journal-title":"Ophthalmology"},{"key":"ref_7","doi-asserted-by":"crossref","first-page":"1199","DOI":"10.1016\/j.ophtha.2018.01.023","article-title":"Efficacy of a Deep Learning System for Detecting Glaucomatous Optic Neuropathy Based on Color Fundus Photographs","volume":"125","author":"Li","year":"2018","journal-title":"Ophthalmology"},{"key":"ref_8","doi-asserted-by":"crossref","first-page":"167","DOI":"10.1136\/bjophthalmol-2018-313173","article-title":"Artificial intelligence and deep learning in ophthalmology","volume":"103","author":"Ting","year":"2019","journal-title":"Br. J. Ophthalmol."},{"key":"ref_9","doi-asserted-by":"crossref","first-page":"101971","DOI":"10.1016\/j.media.2021.101971","article-title":"Applications of deep learning in fundus images: A review","volume":"69","author":"Li","year":"2021","journal-title":"Med. Image Anal."},{"key":"ref_10","doi-asserted-by":"crossref","first-page":"962","DOI":"10.1016\/j.ophtha.2017.02.008","article-title":"Automated Identification of Diabetic Retinopathy Using Deep Learning","volume":"124","author":"Gargeya","year":"2017","journal-title":"Ophthalmology"},{"key":"ref_11","doi-asserted-by":"crossref","first-page":"1170","DOI":"10.1001\/jamaophthalmol.2017.3782","article-title":"Automated Grading of Age-Related Macular Degeneration From Color Fundus Images Using Deep Convolutional Neural Networks","volume":"135","author":"Burlina","year":"2017","journal-title":"JAMA Ophthalmol."},{"key":"ref_12","doi-asserted-by":"crossref","first-page":"565","DOI":"10.1016\/j.ophtha.2018.11.015","article-title":"DeepSeeNet: A Deep Learning Model for Automated Classification of Patient-based Age-related Macular Degeneration Severity from Color Fundus Photographs","volume":"126","author":"Peng","year":"2019","journal-title":"Ophthalmology"},{"key":"ref_13","doi-asserted-by":"crossref","first-page":"4828","DOI":"10.1038\/s41467-021-25138-w","article-title":"Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks","volume":"12","author":"Cen","year":"2021","journal-title":"Nat. Commun."},{"key":"ref_14","doi-asserted-by":"crossref","first-page":"158","DOI":"10.1038\/s41551-018-0195-0","article-title":"Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning","volume":"2","author":"Poplin","year":"2018","journal-title":"Nat. Biomed. Eng."},{"key":"ref_15","doi-asserted-by":"crossref","first-page":"2310","DOI":"10.1056\/NEJMra032865","article-title":"Hypertensive Retinopathy","volume":"351","author":"Wong","year":"2004","journal-title":"N. Engl. J. Med."},{"key":"ref_16","doi-asserted-by":"crossref","first-page":"1264","DOI":"10.1016\/j.ophtha.2018.01.034","article-title":"Grader Variability and the Importance of Reference Standards for Evaluating Machine Learning Models for Diabetic Retinopathy","volume":"125","author":"Krause","year":"2018","journal-title":"Ophthalmology"},{"key":"ref_17","unstructured":"Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., and Gelly, S. (2021, January 3\u20137). An Image Is Worth 16 \u00d7 16 Words: Transformers for Image Recognition at Scale. Proceedings of the International Conference on Learning Representations (ICLR), Virtual."},{"key":"ref_18","unstructured":"Bommasani, R., Hudson, D.A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M.S., Bohg, J., Bosselut, A., and Brunskill, E. (2021). On the Opportunities and Risks of Foundation Models. arXiv."},{"key":"ref_19","doi-asserted-by":"crossref","first-page":"156","DOI":"10.1038\/s41586-023-06555-x","article-title":"A Foundation Model for Generalizable Disease Detection From Retinal Images","volume":"622","author":"Zhou","year":"2023","journal-title":"Nature"},{"key":"ref_20","doi-asserted-by":"crossref","first-page":"e271","DOI":"10.1016\/S2589-7500(19)30123-2","article-title":"A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: A systematic review and meta-analysis","volume":"1","author":"Liu","year":"2019","journal-title":"Lancet Digit. Health"},{"key":"ref_21","doi-asserted-by":"crossref","first-page":"m689","DOI":"10.1136\/bmj.m689","article-title":"Artificial intelligence versus clinicians: Systematic review of design, reporting standards, and claims of deep learning studies","volume":"368","author":"Nagendran","year":"2020","journal-title":"BMJ"},{"key":"ref_22","doi-asserted-by":"crossref","first-page":"665","DOI":"10.1038\/s42256-020-00257-z","article-title":"Shortcut learning in deep neural networks","volume":"2","author":"Geirhos","year":"2020","journal-title":"Nat. Mach. Intell."},{"key":"ref_23","doi-asserted-by":"crossref","unstructured":"Zech, J.R., Badgeley, M.A., Liu, M., Costa, A.B., Titano, J.J., and Oermann, E.K. (2018). Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: A cross-sectional study. PLoS Med., 15.","DOI":"10.1371\/journal.pmed.1002683"},{"key":"ref_24","doi-asserted-by":"crossref","first-page":"610","DOI":"10.1038\/s42256-021-00338-7","article-title":"AI for radiographic COVID-19 detection selects shortcuts over signal","volume":"3","author":"DeGrave","year":"2021","journal-title":"Nat. Mach. Intell."},{"key":"ref_25","doi-asserted-by":"crossref","first-page":"31","DOI":"10.1038\/s41746-019-0105-1","article-title":"Deep learning predicts hip fracture using confounding patient and healthcare variables","volume":"2","author":"Badgeley","year":"2019","journal-title":"npj Digit. Med."},{"key":"ref_26","doi-asserted-by":"crossref","first-page":"100804","DOI":"10.1016\/j.patter.2023.100804","article-title":"Leakage and the reproducibility crisis in machine-learning-based science","volume":"4","author":"Kapoor","year":"2023","journal-title":"Patterns"},{"key":"ref_27","doi-asserted-by":"crossref","first-page":"22544","DOI":"10.1038\/s41598-021-01681-w","article-title":"Effect of data leakage in brain MRI classification using 2D convolutional neural networks","volume":"11","author":"Yagis","year":"2021","journal-title":"Sci. Rep."},{"key":"ref_28","doi-asserted-by":"crossref","first-page":"580","DOI":"10.1038\/s41597-022-01618-6","article-title":"Inflation of test accuracy due to data leakage in deep learning-based classification of OCT images","volume":"9","author":"Tampu","year":"2022","journal-title":"Sci. Data"},{"key":"ref_29","doi-asserted-by":"crossref","first-page":"48","DOI":"10.1038\/s41746-022-00592-y","article-title":"Machine learning for medical imaging: Methodological failures and recommendations for the future","volume":"5","author":"Varoquaux","year":"2022","journal-title":"npj Digit. Med."},{"key":"ref_30","doi-asserted-by":"crossref","first-page":"199","DOI":"10.1038\/s42256-021-00307-0","article-title":"Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans","volume":"3","author":"Roberts","year":"2021","journal-title":"Nat. Mach. Intell."},{"key":"ref_31","doi-asserted-by":"crossref","first-page":"e200029","DOI":"10.1148\/ryai.2020200029","article-title":"Checklist for Artificial Intelligence in Medical Imaging (CLAIM): A Guide for Authors and Reviewers","volume":"2","author":"Mongan","year":"2020","journal-title":"Radiol. Artif. Intell."},{"key":"ref_32","doi-asserted-by":"crossref","first-page":"e078378","DOI":"10.1136\/bmj-2023-078378","article-title":"TRIPOD+AI statement: Updated guidance for reporting clinical prediction models that use regression or machine learning methods","volume":"385","author":"Collins","year":"2024","journal-title":"BMJ"},{"key":"ref_33","unstructured":"(2025, November 07). International Competition on Ocular Disease Intelligent Recognition. Available online: https:\/\/odir2019.grand-challenge.org\/."},{"key":"ref_34","doi-asserted-by":"crossref","unstructured":"Pachade, S., Porwal, P., Thulkar, D., Kokare, M., Deshmukh, G., Sahasrabuddhe, V., Giancardo, L., Quellec, G., and M\u00e9riaudeau, F. (2021). Retinal Fundus Multi-Disease Image Dataset (RFMiD): A Dataset for Multi-Disease Detection Research. Data, 6.","DOI":"10.3390\/data6020014"},{"key":"ref_35","unstructured":"Kaggle (2026, April 20). APTOS 2019 Blindness Detection. Available online: https:\/\/www.kaggle.com\/c\/aptos2019-blindness-detection."},{"key":"ref_36","unstructured":"Graham, B. (2015). Kaggle Diabetic Retinopathy Detection Competition Report, University of Warwick. Technical Report."},{"key":"ref_37","unstructured":"Tan, M., and Le, Q.V. (2019, January 9\u201315). EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks. Proceedings of the International Conference on Machine Learning (ICML), Long Beach, CA, USA."},{"key":"ref_38","doi-asserted-by":"crossref","unstructured":"Huang, G., Liu, Z., van der Maaten, L., and Weinberger, K.Q. (2017). Densely Connected Convolutional Networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE.","DOI":"10.1109\/CVPR.2017.243"},{"key":"ref_39","doi-asserted-by":"crossref","unstructured":"Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., and Wojna, Z. (2016). Rethinking the Inception Architecture for Computer Vision. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE.","DOI":"10.1109\/CVPR.2016.308"},{"key":"ref_40","doi-asserted-by":"crossref","first-page":"1895","DOI":"10.1162\/089976698300017197","article-title":"Approximate Statistical Tests for Comparing Supervised Classification Learning Algorithms","volume":"10","author":"Dietterich","year":"1998","journal-title":"Neural Comput."},{"key":"ref_41","first-page":"65","article-title":"A simple sequentially rejective multiple test procedure","volume":"6","author":"Holm","year":"1979","journal-title":"Scand. J. Stat."},{"key":"ref_42","doi-asserted-by":"crossref","unstructured":"Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., and Batra, D. (2017). Grad-CAM: Visual Explanations From Deep Networks via Gradient-Based Localization. Proceedings of the IEEE International Conference on Computer Vision (ICCV), IEEE.","DOI":"10.1109\/ICCV.2017.74"},{"key":"ref_43","doi-asserted-by":"crossref","first-page":"239","DOI":"10.1023\/A:1024068626366","article-title":"Inference for the generalization error","volume":"52","author":"Nadeau","year":"2003","journal-title":"Mach. Learn."},{"key":"ref_44","doi-asserted-by":"crossref","first-page":"153","DOI":"10.1007\/BF02295996","article-title":"Note on the sampling error of the difference between correlated proportions or percentages","volume":"12","author":"McNemar","year":"1947","journal-title":"Psychometrika"},{"key":"ref_45","doi-asserted-by":"crossref","first-page":"171","DOI":"10.1080\/01621459.1987.10478410","article-title":"Better Bootstrap Confidence Intervals","volume":"82","author":"Efron","year":"1987","journal-title":"J. Am. Stat. Assoc."},{"key":"ref_46","doi-asserted-by":"crossref","first-page":"837","DOI":"10.2307\/2531595","article-title":"Comparing the areas under two or more correlated receiver operating characteristic curves: A nonparametric approach","volume":"44","author":"DeLong","year":"1988","journal-title":"Biometrics"},{"key":"ref_47","doi-asserted-by":"crossref","first-page":"404","DOI":"10.1093\/biomet\/26.4.404","article-title":"The use of confidence or fiducial limits illustrated in the case of the binomial","volume":"26","author":"Clopper","year":"1934","journal-title":"Biometrika"},{"key":"ref_48","unstructured":"Raghu, M., Zhang, C., Kleinberg, J., and Bengio, S. (2019, January 8\u201314). Transfusion: Understanding Transfer Learning for Medical Imaging. Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), Vancouver, BC, Canada."},{"key":"ref_49","doi-asserted-by":"crossref","unstructured":"Ke, A., Ellsworth, W., Banerjee, O., Ng, A.Y., and Rajpurkar, P. (2021, January 8\u20139). CheXtransfer: Performance and Parameter Efficiency of ImageNet Models for Chest X-ray Interpretation. Proceedings of the Conference on Health, Inference, and Learning (CHIL), Virtual.","DOI":"10.1145\/3450439.3451867"},{"key":"ref_50","doi-asserted-by":"crossref","first-page":"215","DOI":"10.1016\/S0161-6420(92)31990-6","article-title":"Expert Agreement in Evaluating the Optic Disc for Glaucoma","volume":"99","author":"Varma","year":"1992","journal-title":"Ophthalmology"},{"key":"ref_51","doi-asserted-by":"crossref","first-page":"1662","DOI":"10.1016\/S0161-6420(94)31118-3","article-title":"Agreement among Optometrists, Ophthalmologists, and Residents in Evaluating the Optic Disc for Glaucoma","volume":"101","author":"Abrams","year":"1994","journal-title":"Ophthalmology"},{"key":"ref_52","doi-asserted-by":"crossref","first-page":"39","DOI":"10.1016\/j.ajo.2008.07.023","article-title":"Agreement Among Glaucoma Specialists in Assessing Progressive Disc Changes From Photographs in Open-Angle Glaucoma Patients","volume":"147","author":"Jampel","year":"2009","journal-title":"Am. J. Ophthalmol."},{"key":"ref_53","doi-asserted-by":"crossref","first-page":"1061","DOI":"10.1007\/s00371-024-03384-5","article-title":"HRDC challenge: A public benchmark for hypertension and hypertensive retinopathy classification from fundus images","volume":"41","author":"Qian","year":"2025","journal-title":"Vis. Comput."}],"container-title":["Diagnostics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/2075-4418\/16\/19\/3109\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T16:42:52Z","timestamp":1790268172000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/2075-4418\/16\/19\/3109"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,24]]},"references-count":53,"journal-issue":{"issue":"19","published-online":{"date-parts":[[2026,10]]}},"alternative-id":["diagnostics16193109"],"URL":"https:\/\/doi.org\/10.3390\/diagnostics16193109","relation":{},"ISSN":["2075-4418"],"issn-type":[{"value":"2075-4418","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,24]]}}}