EyeVQA: Benchmarking Ophthalmic Vision-Language Models from Recognition to Spatial Grounding

Gujie Shao; Zixun Xie; Xuechun Xing; Ruixiang Wang; Ziyun Lan; Yanlin Qi; Gangyi Zhang; Yuxin Yang

cs.CVSep 26, 2026

EyeVQA: Benchmarking Ophthalmic Vision-Language Models from Recognition to Spatial Grounding

Authors: Gujie Shao, Zixun Xie, Xuechun Xing, Ruixiang Wang, Ziyun Lan, Yanlin Qi, Gangyi Zhang, Yuxin Yang, +2 more

Abstract

Vision-language models (VLMs) have shown increasing potential for medical image understanding, yet their capabilities in ophthalmic imaging remain insufficiently characterized. Existing ophthalmic datasets are typically designed for individual diseases or specialized tasks, making it difficult to systematically evaluate whether VLMs can move beyond disease recognition toward comparative reasoning and fine-grained spatial grounding. We introduce EyeVQA, a unified visual question answering benchmark for comprehensive evaluation of ophthalmic VLMs. EyeVQA is constructed from 21 available ophthalmic datasets and contains 20,000 question-answer pairs spanning six disease groups and seven question types: Single-Choice, Multi-Select, Variable-Select, True-False, Ranking, Point Location, and Bounding Box. Gold answers are deterministically derived from source-provided diagnoses, severity grades, clinical findings, segmentation masks, bounding boxes, and anatomical landmarks, enabling reproducible evaluation without relying on model-generated annotations. Notably, 44.5% of the questions require reasoning across multiple images, extending evaluation beyond conventional single-image medical VQA. We benchmark fourteen representative general-purpose, scientific, and medically specialized VLMs under a unified zero-shot protocol. The best-performing model only achieves an overall score of 62.8, while substantial gaps remain in spatial grounding and cross-task generalization. These results highlight the limitations of current VLMs in comprehensive ophthalmic visual understanding and establish EyeVQA as a diagnostic benchmark for developing more reliable and spatially grounded ophthalmic multimodal models. The project page is available at https://github.com/PKUTHM/EyeVQA.

Figures & tables

Figure 1: Overview of the EyeVQA benchmark.

Figure 2: Dataset overview of EyeVQA: (a) distribution of question types; (b) question length distribution by type; (c) question-type composition within each disease group; (d) clinical vocabulary frequency over the 20,000 questions.

Model	Total	DR	GL	PM	MD	MU	OT	SC	MS	VS	TF	RK	PL	BB
GLM-4V-Flash	39.3	37.5	45.0	41.6	41.5	35.7	25.4	37.1	48.9	31.8	42.1	53.2	51.3	4.0
GLM-4.1V-Flash	49.5	49.1	46.3	46.9	55.5	46.6	55.7	43.1	51.2	38.5	60.3	57.3	50.6	34.5
GLM-4.6V-Flash	45.6	44.8	44.0	44.6	45.2	38.5	57.3	37.6	54.0	36.4	51.6	57.4	27.9	51.8
Intern-S1-Pro	54.1	58.7	51.9	51.9	58.6	54.0	44.4	49.6	59.0	44.5	70.2	60.0	50.7	17.0
Intern-S2-35B	59.3	60.0	57.1	55.6	60.8	60.4	68.1	60.1	63.3	50.5	70.5	59.1	45.2	49.4
Intern-S2-397B	62.8	63.2	57.9	57.9	73.4	62.5	67.7	60.2	63.0	51.6	76.2	55.8	70.7	45.5

Table 1: Performance comparison of the evaluated VLMs. Bold indicates the best result, and Underline indicates the second-best result.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Dataset	Images	Source (institution/country)	Disease	License
ACRIMA [ 9 ] (2019)	705	FISABIO Oftalmología Médica, Valencia, Spain	Glaucoma	CC BY 4.0
ADAM [ 10 ] (2018)	1,200	Zhongshan Ophthalmic Center, Sun Yat-sen University, China	Macula Disease	CC BY-NC-ND 4.0
AIROGS [ 8 ] (2021)	101,442	500 screening centers across the United States of America	Glaucoma	CC BY-NC-ND 4.0
BRSET [ 35 ] (2024)	16,266	Department of Ophthalmology, São Paulo Federal University, Brazil	Diabetic Retinopathy, Macula Disease, etc.	PhysioNet Credentialed Health Data License 1.5.0
CSDI [ 47 ] (2026)	187	Peking University Third Hospital, China	Cataract	CC BY 4.0
Retina Dataset [ 6 ] (2016)	601	8 open access databases (specific sources undisclosed)	Glaucoma, Cataract, etc.	Undisclosed

Appendix

Table S1: Overview of the collected ophthalmic datasets.

Explore similar work

May 21, 2026cs.CV

Towards Clinically Interpretable Ophthalmic VQA via Spatially-Grounded Lesion Evidence

Visual Question Answering (VQA) holds great promise for clinical support, particularly in ophthalmology, where retinal fundus photography is essential for diagnosis. However, ophthalmic VQA benchmarks primarily emphasize answer accuracy, neglecting the explicit visual evidence necessary for clinical interpretability. In this work, we introduce FundusGround, a new benchmark for clinically interpretable ophthalmic VQA with spatially-grounded lesion evidence. Specifically, we propose a three-stage pipeline that collects 10,719 fundus images with 15,595 image-level meticulously annotated lesions. To ensure anatomical consistency and clinical validity, all lesions are spatially localized using the Early Treatment Diabetic Retinopathy Study (ETDRS) grid, enabling standardized mapping to nine clinically meaningful retinal regions. Built upon this structured lesion evidence, 72,706 questions are then generated spanning four formats: open-ended, closed-ended, single-choice, and multiple-choice. We further benchmark multiple general- and medical- large vision-language models using dual metrics for answer accuracy and lesion-level reasoning. The experiments demonstrate that incorporating lesion-level visual evidence consistently improves model performance and transparency, highlighting the necessity of explicit spatial grounding for reliable and explainable ophthalmic VQA.

Xingyue Wang, Bo Liu, Meng Wang +4

Department of Computer Science and Engineering, Southern University of Science and Technology, Shenzhen 518055, China · The Hong Kong Polytechnic University, Hong Kong S.A.R., China. · National University of Singapore, Singapore. +2

Medical Visual Question Answering Medical Vision-Language Models Semantic Region

Ask PDF

Jul 5, 2026cs.CV

IRIS: An Intelligent Vision-Language System for Ocular Surface Diseases via Topic Tree and Scene-Driven VQA Generation

While Large Vision-Language Models (VLMs) demonstrate remarkable generic capabilities, their clinical reasoning in specialized domains like ocular surface diseases (OSDs) is severely hindered by a paucity of high-fidelity, multimodal instruction-tuning data. To dismantle this data bottleneck, we introduce IRIS, an Intelligent Recognition and Interaction System tailored for fine-grained OSD understanding via external eye photography. First, we curate IRIS-120K, the largest and most comprehensive OSD visual question-answering (VQA) dataset to date. Crucially, to overcome the semantic shallowness of conventional image-caption pairs, we propose a synergistic data generation paradigm to explicitly inject clinical priors. Our data engine operates via a dual-branch framework: 1) a Topic Finding Tree (TFT) that hierarchically anchors visual features to precise anatomical and pathological concepts, enforcing rigorous medical deduction logic; and 2) a Scene-driven strategy that synthesizes role-adaptive clinical dialogues to ensure pragmatic generalization. By explicitly aligning a compact 4B-parameter VLM on this structurally enriched corpus, IRIS achieves state-of-the-art performance, comprehensively outperforming both generalist and specialized medical VLMs with up to 34B parameters. Our findings underscore that structured knowledge injection profoundly prevails over sheer parameter scaling, unlocking the potential for resource-efficient, expert-level AI deployment on mobile edge devices for scalable OSD screening. Code, datasets, and model weights will be publicly released by this repo.

Hao Wei, Wenjin Qi, Dasen Dai +2

The Chinese University of Hong Kong, Shatin, Hong Kong, China · Westlake University, Xihu District, Hangzhou, Zhejiang Province, China

Recent Vision-Language Models Melanoma

Ask PDF

Aug 13, 2026cs.AI

Polish Medical Visual Question Answering: Vision-Language Models Underutilize Visual Evidence

We introduce a Polish-language medical visual question answering (VQA) benchmark, built from Polish Board Certification Examination questions for licensed physicians and dentists pursuing specialist certification. The benchmark comprises image-containing questions spanning diverse medical specialties and visual domains, together with a text-only question answering (QA) control set. We evaluate Polish-oriented, general-purpose open-weight, and commercial vision-language models. The task remains challenging: the best model achieves 79.0% accuracy on the full VQA set, and only GPT-5.6 surpasses the approximate human reference on the subset with available candidate responses; all other evaluated models perform worse than humans. To assess visual grounding, we compare complete inputs with configurations omitting the image, the question, or both, and categorize questions by image importance. Models derive more useful information from the question text than from the image and perform worse on image-dominant questions. Across both QA and VQA, they nevertheless achieve above-chance accuracy from the answer choices alone, showing that non-trivial performance can persist even when key task components are missing.

Jakub Pokrywka, Łukasz Grzybowski, Antoni Lasik +3

Adam Mickiewicz University · ARAAI Poland · NASK National Research Institute +2

Medical Visual Question Answering Multiple-Choice Visual Question Answering Vision-Language Foundation Models Visual Evidence Diagnostic Benchmark Assessment

Ask PDF