Related Experiment Video
Updated: Jun 15, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.6K
Sequential sensitivity analysis of multimodal large language models for rare orbital disease detection
Chaoyu Lei1,2,3, Kaiyuan Ji4,5, Chen Zhao3
1Hainan Research Institute, Shanghai Jiao Tong University, Sanya, Hainan, China.
Communications Medicine
|February 20, 2026
Summary
Multimodal large language models (MLLMs) show promise in diagnosing rare orbital diseases. Integrating diverse clinical data significantly improves MLLM diagnostic accuracy, aiding clinical decisions.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Diagnostics
Background:
- Delayed diagnosis of rare orbital diseases is common due to limited clinical awareness.
- Prior studies show multimodal large language models (MLLMs) can detect common ocular conditions.
Purpose of the Study:
- To evaluate if integrating multimodal clinical data enhances MLLM diagnostic accuracy for rare orbital diseases.
- To assess the performance of MLLMs compared to traditional and next-generation models.
Main Methods:
- A multinational, retrospective study analyzed two datasets.
- A contrastive language-image pre-training (CLIP) model was fine-tuned for preliminary classification.
- A MLLM (GPT-4o-Latest) was evaluated with sequential sensitivity analysis of multimodal inputs, including an AI agent combining CLIP and GPT-4o-Latest.
Main Results:
- The CLIP model achieved 90.21% preliminary detection accuracy, outperforming baseline models.
- MLLM accuracy significantly improved with multimodal inputs; the combined agent reached 85.29% top-5 accuracy.
- Generated medical reports and recommendations were accurate, readable, complete, and had low potential for harm.
Conclusions:
- MLLMs demonstrate significant potential for improving diagnostic accuracy in rare orbital diseases.
- This technology can support clinical decision-making and enhance patient care for rare orbital conditions.
