ScholarDeck

StudyACM transactions on computing for healthcare · 2026

LMOD+: A Comprehensive Multimodal Dataset and Benchmark for Developing and Evaluating Multimodal Large Language Models in Ophthalmology.

Zhenyue Qin, Yang Liu, Y U Yin … Qingyu Chen

OphthalmologyStudy

Qwen-7B achieved 58.26% accuracy in zero-shot ophthalmic disease screening.

LMOD+: A Comprehensive Multimodal Dataset and Benchmark for Developing and Evaluating Multimodal Large Language Models in Ophthalmology.

Zhenyue Qin et al. · ACM transactions on computing for healthcare · 2026

Background

The rising prevalence of vision-threatening eye diseases poses a major global health and economic burden, yet timely diagnosis remains limited by workforce shortages, diagnostic delays, and restricted access to specialized care.

Methods

Third, we systematically evaluated 24 state-of-the-art MLLMs, including recent models from the InternVL, Qwen, and DeepSeek families.

58.26%

Results

Qwen-7B correctly found eye diseases in over half of cases without specific training

More results

Artificial intelligence (AI) offers potential solutions.

However, advancing MLLMs for ophthalmology is hindered by the lack of unified, comprehensive benchmark datasets for development and evaluation.

More results

In this work, we present LMOD+, a large-scale multimodal ophthalmology benchmark dataset comprising 32,633 instances with multi-granular annotations across 12 common ophthalmic conditions and 5 imaging modalities.

First, we expanded the dataset by nearly 50% (from 21,933 to 32,633 instances).

“

Conclusion

The dataset website, benchmark leaderboard, and download link are available at https://kfzyqin.github.io/lmod_plus.

Post

LMOD+: A Comprehensive Multimodal Dataset and Benchmark for Developing and Evaluating Multimodal Large Language Models in Ophthalmology.

Qwen-7B achieved 58.26% accuracy in zero-shot ophthalmic disease screening.

Related papers