{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,9]],"date-time":"2026-10-09T16:25:54Z","timestamp":1791563154464,"version":"4.3.4"},"reference-count":25,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,3,31]],"date-time":"2026-03-31T00:00:00Z","timestamp":1774915200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Med."],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Automated analysis of color fundus photographs can support scalable screening for diabetic retinopathy (DR), age-related macular degeneration (AMD), and glaucoma, but single-run reporting and accuracy-only summaries can mask clinically relevant instabilities and failure modes.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>\n                      Using the public FIVES dataset, we benchmarked six deep learning configurations for four-class fundus screening (AMD, DR, glaucoma, normal): three DeepLabv3\u2013backbone hybrids (ResNet50, DenseNet121, EfficientNet-B0) and three backbone-only classifiers. All experiments were evaluated using five independent stratified splits generated with different random seeds (\n                      <jats:italic>n<\/jats:italic>\n                      = 5 runs), each defining a distinct 20% held-out test set. Models were trained on the remaining 80% (training\/validation), and all reported metrics are computed on the 20% test set of each run and summarized as mean \u00b1 SD across runs. Performance was summarized with accuracy, sensitivity, specificity, and one-vs.-rest AUC; we further characterized clinical behavior via row-normalized confusion matrices, per-class precision\/recall\/F1, and a screening-style binary triage setting (referable = AMD \u222a DR \u222a glaucoma vs. normal).\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Hybrid models consistently achieved higher discrimination than simple classifiers (AUC 0.969\u20130.979 vs. 0.908\u20130.920), despite similar accuracies (0.924\u20130.941). The selected model, DeepLabv3\u2013DenseNet121, reached the highest AUC (0.979 \u00b1 0.009). Class-wise analysis revealed strong performance for Normal (F1 0.970 \u00b1 0.014) and Glaucoma (F1 0.896 \u00b1 0.048), while DR was the main bottleneck (sensitivity 0.738 \u00b1 0.117), with most DR errors redistributed to AMD (13.6%) and Glaucoma (12.0%) and minimal confusion with Normal (0.5%). In binary triage, the model achieved sensitivity 0.993 \u00b1 0.011 and specificity 0.963 \u00b1 0.034, with PPV 0.987 \u00b1 0.013 and NPV 0.980 \u00b1 0.032, and a stable referral rate (\u223c0.73\u20130.77) across runs.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusion<\/jats:title>\n                    <jats:p>DeepLabv3-based hybrids provide a robust advantage in AUC for multiclass fundus screening on FIVES. The residual risk concentrates in the DR\u2013AMD\u2013Glaucoma decision boundary, suggesting that deployment-oriented policies should prioritize conservative handling of DR-adjacent cases while leveraging the stability of Normal predictions for screening workflows.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fmed.2026.1771083","type":"journal-article","created":{"date-parts":[[2026,3,31]],"date-time":"2026-03-31T05:32:30Z","timestamp":1774935150000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":2,"title":["Robust benchmarking of DeepLabv3 hybrid models for multiclass fundus screening and referable-disease triage on the FIVES dataset"],"prefix":"10.3389","volume":"13","author":[{"given":"Igor","family":"Garcia-Atutxa","sequence":"first","affiliation":[{"name":"Escuela Polit\u00e9cnica Superior, Universidad Cat\u00f3lica de Murcia (UCAM)","place":["Murcia, Spain"]},{"name":"Universidad Tecmilenio","place":["Azcapotzalco, Mexico"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Laura","family":"Martinez-Baliu","sequence":"additional","affiliation":[{"name":"Universitat Oberta de Catalunya","place":["Barcelona, Spain"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Francisca","family":"Villanueva-Flores","sequence":"additional","affiliation":[{"name":"Centro de Investigaci\u00f3n en Ciencia Aplicada y Tecnolog\u00eda Avanzada, Unidad Morelos del Instituto Polit\u00e9cnico Nacional","place":["Xochitepec, Mexico"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,3,31]]},"reference":[{"key":"B1","volume-title":"World Report on Vision.","year":"2019"},{"key":"B2","doi-asserted-by":"publisher","first-page":"e106","DOI":"10.1016\/S2214-109X(13)70145-1","article-title":"Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: a systematic review and meta-analysis.","volume":"2","author":"Wong","year":"2014","journal-title":"Lancet Glob Health."},{"key":"B3","doi-asserted-by":"publisher","first-page":"e144","DOI":"10.1016\/S2214-109X(20)30489-7","article-title":"Causes of blindness and vision impairment in 2020 and trends over 30 years, and prevalence of avoidable blindness in relation to VISION 2020: the Right to Sight: an analysis for the global burden of disease study.","volume":"9","author":"Steinmetz","year":"2021","journal-title":"Lancet Glob Health."},{"key":"B4","doi-asserted-by":"publisher","first-page":"e053805","DOI":"10.1136\/bmjopen-2021-053805","article-title":"Time trends, associations and prevalence of blindness and vision loss due to glaucoma: an analysis of observational data from the global burden of disease study 2017.","volume":"12","author":"Sun","year":"2022","journal-title":"BMJ Open."},{"key":"B5","doi-asserted-by":"publisher","first-page":"1580","DOI":"10.1016\/j.ophtha.2021.04.027","article-title":"Global prevalence of diabetic retinopathy and projection of burden through 2045: systematic review and meta-analysis.","volume":"128","author":"Teo","year":"2021","journal-title":"Ophthalmology."},{"key":"B6","doi-asserted-by":"publisher","first-page":"112346","DOI":"10.1016\/j.diabres.2025.112346","article-title":"IDF diabetes Atlas: a worldwide review of studies utilizing retinal photography to screen for diabetic retinopathy from 2017 to 2024 inclusive.","volume":"226","author":"Owens","year":"2025","journal-title":"Diabetes Res Clin Pract."},{"key":"B7","doi-asserted-by":"publisher","first-page":"2200","DOI":"10.4103\/jfmpc.jfmpc_111_20","article-title":"Glaucoma blindness-A rapidly emerging non-communicable ocular disease in India: addressing the issue with advocacy.","volume":"9","author":"Senjam","year":"2020","journal-title":"J Family Med Prim Care."},{"key":"B8","doi-asserted-by":"publisher","first-page":"712","DOI":"10.1055\/a-2617-1575","article-title":"Understanding glaucoma: why it remains a leading cause of blindness worldwide.","volume":"242","author":"Qi","year":"2025","journal-title":"Klin Monbl Augenheilkd."},{"key":"B9","doi-asserted-by":"publisher","first-page":"10750","DOI":"10.1038\/s41598-019-47181-w","article-title":"Deep learning fundus image analysis for diabetic retinopathy and macular edema grading.","volume":"9","author":"Sahlsten","year":"2019","journal-title":"Sci Rep."},{"key":"B10","doi-asserted-by":"publisher","first-page":"100203","DOI":"10.1016\/j.cosrev.2019.100203","article-title":"Application of deep learning for retinal image analysis: a review.","volume":"35","author":"Badar","year":"2020","journal-title":"Comput Sci Rev."},{"key":"B11","doi-asserted-by":"publisher","first-page":"20406223231209895","DOI":"10.1177\/20406223231209895","article-title":"Deep learning-based fundus image analysis for cardiovascular disease: a review.","volume":"14","author":"Chikumba","year":"2023","journal-title":"Ther Adv Chronic Dis."},{"key":"B12","doi-asserted-by":"publisher","first-page":"1608994","DOI":"10.3389\/fcell.2025.1608994","article-title":"Research progress in deep learning-based fundus image analysis for the diagnosis and prediction of hypertension-related diseases.","volume":"13","author":"Li","year":"2025","journal-title":"Front Cell Dev Biol."},{"key":"B13","doi-asserted-by":"publisher","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":"B14","doi-asserted-by":"publisher","first-page":"3603","DOI":"10.3390\/electronics13183603","article-title":"Fundus image generation and classification of diabetic retinopathy based on convolutional neural network.","volume":"13","author":"Zhang","year":"2024","journal-title":"Electronics."},{"key":"B15","doi-asserted-by":"publisher","first-page":"155","DOI":"10.29194\/NJES.27020155","article-title":"Comprehensive survey of the state-of-the-art deep learning models for diabetic retinopathy detection and grading using retinal fundus photography.","volume":"27","author":"Sadek","year":"2024","journal-title":"Nahrain J Eng Sci."},{"key":"B16","doi-asserted-by":"publisher","first-page":"104599","DOI":"10.1016\/j.compbiomed.2021.104599","article-title":"Deep learning for diabetic retinopathy detection and classification based on fundus images: a review.","volume":"135","author":"Tsiknakis","year":"2021","journal-title":"Comput Biol Med."},{"key":"B17","doi-asserted-by":"publisher","first-page":"233","DOI":"10.1007\/s10791-025-09761-5","article-title":"A review of deep learning for fundus image enhancement.","volume":"28","author":"Lin","year":"2025","journal-title":"Discov Comput."},{"key":"B18","doi-asserted-by":"publisher","first-page":"1679310","DOI":"10.3389\/frai.2025.1679310","article-title":"Early-fusion hybrid CNN-transformer models for multiclass ovarian tumor ultrasound classification.","volume":"8","author":"Garcia-Atutxa","year":"2025","journal-title":"Front Artif Intell."},{"key":"B19","doi-asserted-by":"publisher","first-page":"1649746","DOI":"10.3389\/frai.2025.1649746","article-title":"Artificial intelligence for ovarian cancer diagnosis via ultrasound: a systematic review and quantitative assessment of model performance.","volume":"8","author":"Garcia-Atutxa","year":"2025","journal-title":"Front Artif Intell."},{"key":"B20","doi-asserted-by":"publisher","first-page":"29","DOI":"10.3390\/data8020029","article-title":"Retinal fundus multi-disease image dataset (RFMiD) 2.0: a dataset of frequently and rarely identified diseases.","volume":"8","author":"Panchal","year":"2023","journal-title":"Data."},{"key":"B21","doi-asserted-by":"publisher","first-page":"840","DOI":"10.3390\/bioengineering12080840","article-title":"Deep learning for comprehensive analysis of retinal fundus images: detection of systemic and ocular conditions.","volume":"12","author":"Aghabeigi Alooghareh","year":"2025","journal-title":"Bioengineering."},{"key":"B22","doi-asserted-by":"publisher","first-page":"103365","DOI":"10.1016\/j.media.2024.103365","article-title":"RFMiD: retinal image analysis for multi-disease detection challenge.","volume":"99","author":"Pachade","year":"2025","journal-title":"Med Image Anal."},{"key":"B23","doi-asserted-by":"publisher","first-page":"130","DOI":"10.1038\/s41746-025-01473-w","article-title":"A hybrid multi model artificial intelligence approach for glaucoma screening using fundus images.","volume":"8","author":"Sharma","year":"2025","journal-title":"NPJ Digit Med."},{"key":"B24","doi-asserted-by":"publisher","DOI":"10.48550\/arxiv.1706.05587","article-title":"Rethinking atrous convolution for semantic image segmentation.","author":"Chen","year":"2017","journal-title":"arXiv [Preprint]."},{"key":"B25","doi-asserted-by":"publisher","first-page":"475","DOI":"10.1038\/s41597-022-01564-3","article-title":"FIVES: a fundus image dataset for artificial intelligence based vessel segmentation.","volume":"9","author":"Jin","year":"2022","journal-title":"Sci Data."}],"container-title":["Frontiers in Medicine"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fmed.2026.1771083\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,31]],"date-time":"2026-03-31T05:32:31Z","timestamp":1774935151000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fmed.2026.1771083\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,3,31]]},"references-count":25,"alternative-id":["10.3389\/fmed.2026.1771083"],"URL":"https:\/\/doi.org\/10.3389\/fmed.2026.1771083","relation":{},"ISSN":["2296-858X"],"issn-type":[{"value":"2296-858X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,3,31]]},"article-number":"1771083"}}