Nature Medicine：孙晓东/黄天荫/盛斌合作开发AI系统，精准诊断遗传性视网膜疾病
Merged summary
TL;DR - A Nature Medicine study presents Retina4IRD, an AI decision-support system that predicts inherited retinal disease genotypes from multimodal retinal images. In a randomized trial, it substantially improved clinicians’ diagnostic accuracy and downstream management decisions.
- Retina4IRD uses a RETFound-pretrained Vision Transformer with fundus photographs and optical coherence tomography images to predict 17 genotype categories.
- Training and validation included 1,843 genetically confirmed patients from China, South Korea, and Poland; top-5 accuracy was 90.4% internally and 85.6% externally.
- Among 295 trial participants, AI-assisted clinicians achieved 88.5% top-5 gene accuracy versus 67.3% for clinicians alone.
- The system could help prioritize genetic testing, accelerate diagnosis, and extend specialist-level support to resource-limited clinics.
Sources (1)
Nature Medicine：孙晓东/黄天荫/盛斌合作开发AI系统，精准诊断遗传性视网膜疾病
TL;DR - A Nature Medicine study presents Retina4IRD, an AI decision-support system that predicts inherited retinal disease genotypes from multimodal retinal images. In a randomized trial, it substantially improved clinicians’ diagnostic accuracy and downstream management decisions.
- Retina4IRD uses a RETFound-pretrained Vision Transformer with fundus photographs and optical coherence tomography images to predict 17 genotype categories.
- Training and validation included 1,843 genetically confirmed patients from China, South Korea, and Poland; top-5 accuracy was 90.4% internally and 85.6% externally.
- Among 295 trial participants, AI-assisted clinicians achieved 88.5% top-5 gene accuracy versus 67.3% for clinicians alone.
- The system could help prioritize genetic testing, accelerate diagnosis, and extend specialist-level support to resource-limited clinics.