{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"medRxiv"}],"indexed":{"date-parts":[[2026,8,1]],"date-time":"2026-08-01T09:59:09Z","timestamp":1785578349456,"version":"3.56.0"},"posted":{"date-parts":[[2026,7,29]]},"group-title":"Ophthalmology","reference-count":29,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,7,29]],"date-time":"2026-07-29T00:00:00Z","timestamp":1785283200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"funder":[{"DOI":"10.13039\/100000002","name":"National Institutes of Health","doi-asserted-by":"crossref","award":["R01 EY036222 and P30 EY003790"],"award-info":[{"award-number":["R01 EY036222 and P30 EY003790"]}],"id":[{"id":"10.13039\/100000002","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,7,29]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:p>\n                  Glaucoma is a leading cause of irreversible blindness worldwide. Ophthalmologists diagnose glaucoma through a structured reasoning process by sequentially evaluating optic nerve head characteristics before reaching a final diagnosis, whereas existing AI systems typically perform direct image classification without providing clinically meaningful reasoning. We present the first clinically annotated fundus reasoning dataset, comprising 1,077 fundus photographs paired with expert-authored six-step diagnostic reports. Building on this dataset, we develop a reasoning-driven vision-language framework that explicitly models the ophthalmologist\u2019s diagnostic workflow by generating structured clinical reasoning prior to diagnosis. The generated reports are clinically validated, achieving the best performance across all evaluated clinical findings, including a cup-to-disc ratio mean absolute error of 0.070, an ISNT Kendall distance of 1.73, and the highest semantic agreement with expert reports (BERTScore-F1 = 0.874). The resulting framework also improves glaucoma diagnosis, achieving a balanced accuracy of 94.7% and precision of 94.8%, demonstrating that explicitly modeling expert clinical reasoning simultaneously improves interpretability and diagnostic performance. Code and data are available at\n                  <jats:ext-link xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" ext-link-type=\"uri\" xlink:href=\"https:\/\/glaucoma-cot.github.io\/\">https:\/\/glaucoma-cot.github.io\/<\/jats:ext-link>\n                  .\n                <\/jats:p>","DOI":"10.64898\/2026.07.28.26359057","type":"posted-content","created":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T01:10:28Z","timestamp":1785373828000},"source":"Crossref","is-referenced-by-count":0,"title":["Learning Ophthalmologist Clinical Reasoning for Glaucoma Diagnosis from Fundus Images"],"prefix":"10.64898","author":[{"given":"Kaichen","family":"Zhou","sequence":"first","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"},{"name":"Media Lab, Massachusetts Institute of Technology, Cambridge, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yuzhen","family":"Chen","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Elif","family":"Yildiz","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"},{"name":"International School of Medicine, Istanbul Medipol University, Istanbul, Turkey"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Min","family":"Shi","sequence":"additional","affiliation":[{"name":"School of Computing and Informatics, University of Louisiana at Lafayette, Lafayette, LA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"David","family":"Dai","sequence":"additional","affiliation":[{"name":"Media Lab, Massachusetts Institute of Technology, Cambridge, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Grace","family":"Chen","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jiale","family":"Zheng","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"He","family":"Wang","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fangneng","family":"Zhan","sequence":"additional","affiliation":[{"name":"Media Lab, Massachusetts Institute of Technology, Cambridge, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chhavi","family":"Saini","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lucy Q.","family":"Shen","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yike","family":"Guo","sequence":"additional","affiliation":[{"name":"Dept of CSE, The Hong Kong University of Science and Technology, Hongkong, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Paul Pu","family":"Liang","sequence":"additional","affiliation":[{"name":"Media Lab, Massachusetts Institute of Technology, Cambridge, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mengyu","family":"Wang","sequence":"additional","affiliation":[{"name":"Harvard Ophthalmology AI Lab, Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026080102150526000_2026.07.28.26359057v1.1","doi-asserted-by":"publisher","DOI":"10.1016\/j.ophtha.2014.05.013"},{"issue":"10414","key":"2026080102150526000_2026.07.28.26359057v1.2","doi-asserted-by":"crossref","first-page":"1788","DOI":"10.1016\/S0140-6736(23)01289-8","article-title":"Glaucoma: now and beyond","volume":"402","year":"2023","journal-title":"The Lancet"},{"key":"2026080102150526000_2026.07.28.26359057v1.3","unstructured":"D. Kri\u017eaj , What is glaucoma?, Webvision: The organization of the retina and visual system [Internet] (2019)."},{"issue":"1","key":"2026080102150526000_2026.07.28.26359057v1.4","doi-asserted-by":"crossref","first-page":"13762","DOI":"10.1038\/s41598-021-92971-w","article-title":"Prevalence of primary open angle glaucoma in the last 20 years: a meta-analysis and systematic review","volume":"11","year":"2021","journal-title":"Scientific reports"},{"issue":"1","key":"2026080102150526000_2026.07.28.26359057v1.5","doi-asserted-by":"crossref","first-page":"235","DOI":"10.1146\/annurev-vision-111815-114417","article-title":"Understanding glaucomatous optic neuropathy: the synergy between clinical observation and investigation","volume":"2","year":"2016","journal-title":"Annual review of vision science"},{"key":"2026080102150526000_2026.07.28.26359057v1.6","doi-asserted-by":"publisher","DOI":"10.1001\/archopht.120.10.1268"},{"issue":"2","key":"2026080102150526000_2026.07.28.26359057v1.7","doi-asserted-by":"crossref","first-page":"169","DOI":"10.1111\/ceo.13138","article-title":"Evolution of optic nerve photography for glaucoma screening: a review","volume":"46","year":"2018","journal-title":"Clinical & experimental ophthalmology"},{"key":"2026080102150526000_2026.07.28.26359057v1.8","doi-asserted-by":"publisher","DOI":"10.1016\/j.survophthal.2022.08.005"},{"key":"2026080102150526000_2026.07.28.26359057v1.9","doi-asserted-by":"crossref","first-page":"1324109","DOI":"10.3389\/frai.2024.1324109","article-title":"Transforming glaucoma diagnosis: transformers at the fore-front","volume":"7","year":"2024","journal-title":"Frontiers in Artificial Intelligence"},{"issue":"1","key":"2026080102150526000_2026.07.28.26359057v1.10","doi-asserted-by":"crossref","first-page":"90","DOI":"10.1097\/ICU.0000000000001091","article-title":"Foundation models in ophthalmology: opportunities and challenges","volume":"36","year":"2025","journal-title":"Current opinion in ophthalmology"},{"key":"2026080102150526000_2026.07.28.26359057v1.11","doi-asserted-by":"publisher","DOI":"10.1136\/bjophthalmol-2020-317112"},{"key":"2026080102150526000_2026.07.28.26359057v1.12","doi-asserted-by":"crossref","unstructured":"S. L. Groth , K. M. Joos , Primary open-angle glaucoma, in: Albert and Jakobiec\u2019s Principles and Practice of Ophthalmology, Springer, 2020, pp. 1\u201315.","DOI":"10.1007\/978-3-319-90495-5_169-1"},{"issue":"7","key":"2026080102150526000_2026.07.28.26359057v1.13","first-page":"1764","article-title":"Ranking of optic disc variables for detection of glaucomatous optic nerve damage","volume":"41","year":"2000","journal-title":"Investigative Ophthalmology & Visual Science"},{"key":"2026080102150526000_2026.07.28.26359057v1.14","doi-asserted-by":"publisher","DOI":"10.1001\/archopht.124.11.1579"},{"key":"2026080102150526000_2026.07.28.26359057v1.15","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2019.2927226"},{"issue":"1","key":"2026080102150526000_2026.07.28.26359057v1.16","doi-asserted-by":"crossref","first-page":"291","DOI":"10.1038\/s41597-022-01388-1","article-title":"Papila: Dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessment","volume":"9","year":"2022","journal-title":"Scientific Data"},{"key":"2026080102150526000_2026.07.28.26359057v1.17","first-page":"24824","article-title":"Chain-of-thought prompting elicits reasoning in large language models","volume":"35","year":"2022","journal-title":"Advances in neural information processing systems"},{"key":"2026080102150526000_2026.07.28.26359057v1.18","unstructured":"S. Bai , Y. Cai , R. Chen , K. Chen , X. Chen , Z. Cheng , L. Deng , W. Ding , C. Gao , C. Ge , et al., Qwen3-vl technical report, arXiv preprint arXiv:2511.21631 (2025)."},{"key":"2026080102150526000_2026.07.28.26359057v1.19","unstructured":"A. Yang , A. Li , B. Yang , B. Zhang , B. Hui , B. Zheng , B. Yu , C. Gao , C. Huang , C. Lv , et al., Qwen3 technical report, arXiv preprint arXiv:2505.09388 (2025)."},{"key":"2026080102150526000_2026.07.28.26359057v1.20","unstructured":"M. Chen , J. Tworek , H. Jun , Q. Yuan , H. P. D. O. Pinto , J. Kaplan , H. Edwards , Y. Burda , N. Joseph , G. Brockman , et al., Evaluating large language models trained on code, arXiv preprint arXiv:2107.03374 (2021)."},{"issue":"1","key":"2026080102150526000_2026.07.28.26359057v1.21","doi-asserted-by":"crossref","first-page":"5528","DOI":"10.1038\/s41467-025-60577-9","article-title":"Enhancing diagnostic accuracy in rare and common fundus diseases with a knowledge-rich vision-language model","volume":"16","year":"2025","journal-title":"Nature communications"},{"key":"2026080102150526000_2026.07.28.26359057v1.22","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06555-x"},{"key":"2026080102150526000_2026.07.28.26359057v1.23","unstructured":"M. Oquab , T. Darcet , T. Moutakanni , H. Vo , M. Szafraniec , V. Khalidov , P. Fernandez , D. Haziza , F. Massa , A. El-Nouby , et al., Dinov2: Learning robust visual features without supervision, arXiv preprint arXiv:2304.07193 (2023)."},{"key":"2026080102150526000_2026.07.28.26359057v1.24","doi-asserted-by":"publisher","DOI":"10.7326\/M14-0698"},{"key":"2026080102150526000_2026.07.28.26359057v1.25","unstructured":"R. J. Hyndman , G. Athanasopoulos , Forecasting: principles and practice, OTexts, 2018."},{"key":"2026080102150526000_2026.07.28.26359057v1.26","doi-asserted-by":"publisher","DOI":"10.1093\/biomet\/30.1-2.81"},{"issue":"4","key":"2026080102150526000_2026.07.28.26359057v1.27","first-page":"35","article-title":"Modern information retrieval: A brief overview","volume":"24","year":"2001","journal-title":"IEEE Data Eng. Bull"},{"key":"2026080102150526000_2026.07.28.26359057v1.28","unstructured":"T. Zhang , V. Kishore , F. Wu , K. Q. Weinberger , Y. Artzi , Bertscore: Evaluating text generation with bert, arXiv preprint arXiv:1904.09675 (2019)."},{"key":"2026080102150526000_2026.07.28.26359057v1.29","unstructured":"C.-Y. Lin , Rouge: A package for automatic evaluation of summaries, in: Text summarization branches out, 2004, pp. 74\u201381."}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.07.28.26359057","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,1]],"date-time":"2026-08-01T09:15:17Z","timestamp":1785575717000},"score":1,"resource":{"primary":{"URL":"http:\/\/medrxiv.org\/lookup\/doi\/10.64898\/2026.07.28.26359057"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7,29]]},"references-count":29,"URL":"https:\/\/doi.org\/10.64898\/2026.07.28.26359057","relation":{},"subject":[],"published":{"date-parts":[[2026,7,29]]},"subtype":"preprint"}}