OBUSight: Clinically Aligned Generative AI for Ophthalmic Ultrasound Interpretation and Diagnosis.
Authors
Affiliations (10)
Affiliations (10)
- State Key Laboratory of Transvascular Implantation Devices of the Second Affiliated Hospital, Zhejiang University School of Medicine, and Transvascular Implantation Devices Research Institute, Hangzhou, China.
- Zhejiang University, Eye Center of Second Affiliated Hospital, School of Medicine, China, Zhejiang Provincial Key Laboratory of Ophthalmology, Zhejiang Provincial Clinical Research Center for Eye Diseases, Zhejiang Provincial Engineering Institute on Eye Diseases, Hangzhou, China.
- Zhongshan Ophthalmic Center, Sun Yat-sen University, WHO Collaborating Centre for Eye Care and Vision, State Key Laboratory of Ophthalmology, Guangdong Provincial Key Laboratory of Ophthalmology and Visual Science, Guangdong Provincial Clinical Research Center for Ocular Diseases, Guangzhou, China.
- Department of Ophthalmology, National University of Singapore, Singapore, Singapore.
- Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore.
- Centre for Innovation and Precision Eye Health, National University of Singapore, Singapore, Singapore.
- Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts, USA.
- Department of Genetics and Biomedical Informatics, Zhongshan School of Medicine, Institute for Frontier Interdisciplinary Research in Health Sciences and Technology, Sun Yat-sen University, Guangzhou, China.
- Hainan Eye Hospital and Key Laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou, China.
- Liangzhu Laboratory and WeDoctor Cloud, Zhejiang Key Laboratory of Medical Imaging Artificial Intelligence, Hangzhou, China.
Abstract
Ocular B-scan ultrasonography (OBU), widely used for diagnosing posterior segment ocular disorders, poses unique challenges for ophthalmologists in image interpretation. In this study, a clinically aligned generative artificial intelligence (AI) model, OBUSight, was proposed to jointly generate reports and diagnose diseases for comprehensive OBU image interpretation. OBUSight was trained and validated on a large multi-center OBU dataset consisting of 39 654 images and 17 586 corresponding reports from 11 381 patients. By evaluating the quality of generated reports using natural language generation (NLG) metrics and clinical efficacy (CE) metrics, OBUSight outperformed eight state-of-the-art models and demonstrated robust performance across multi-center and multimorbidity validation datasets. The expert rating further indicated that OBUSight can provide clinically aligned reports without major corrections. The ancillary role of OBUSight in enhancing diagnostic efficiency was evaluated by providing ophthalmologists, residents, and ophthalmology students with its generated reports and predicted diagnoses during the diagnostic process. In both retrospective and prospective evaluations, OBUSight significantly outperformed residents and ophthalmology students (all p < 0.05), achieved diagnostic performance comparable to ophthalmologists, and reduced diagnostic time. In conclusion, OBUSight represents a promising AI tool for enhancing diagnostic efficiency in ophthalmic ultrasound practice, especially for less experienced clinicians.