A Clinician-Friendly Platform for Ophthalmic Image Analysis Without Technical Barriers

TL;DR

GlobeReady presents a training-free ophthalmic AI platform enabling immediate ocular disease diagnosis across CPF and OCT without local model retraining. By combining large-scale synthetic data generation, self-supervised pretraining (DINOv2) and vision-language contrastive learning (CLIP) with a Bayesian confidence framework and local feature retrieval, it achieves high diagnostic accuracy and robust generalization across centers and diseases. The system also includes open-set anomaly detection, image- and case-retrieval capabilities, and clinician usability validation, collectively reducing practical deployment barriers. These features collectively suggest GlobeReady can enhance global ophthalmic care by delivering reliable, interpretable predictions and facilitating rapid case matching and data annotation without specialized AI expertise.

Abstract

Artificial intelligence (AI) shows remarkable potential in medical imaging diagnostics, yet most current models require retraining when applied across different clinical settings, limiting their scalability. We introduce GlobeReady, a clinician-friendly AI platform that enables fundus disease diagnosis that operates without retraining, fine-tuning, or the needs for technical expertise. GlobeReady demonstrates high accuracy across imaging modalities: 93.9-98.5% for 11 fundus diseases using color fundus photographs (CPFs) and 87.2-92.7% for 15 fundus diseases using optic coherence tomography (OCT) scans. By leveraging training-free local feature augmentation, GlobeReady platform effectively mitigates domain shifts across centers and populations, achieving accuracies of 88.9-97.4% across five centers on average in China, 86.3-96.9% in Vietnam, and 73.4-91.0% in Singapore, and 90.2-98.9% in the UK. Incorporating a bulit-in confidence-quantifiable diagnostic mechanism further enhances the platform's accuracy to 94.9-99.4% with CFPs and 88.2-96.2% with OCT, while enabling identification of out-of-distribution cases with 86.3% accuracy across 49 common and rare fundus diseases using CFPs, and 90.6% accuracy across 13 diseases using OCT. Clinicians from countries rated GlobeReady highly for usability and clinical relevance (average score 4.6/5). These findings demonstrate GlobeReady's robustness, generalizability and potential to support global ophthalmic care without technical barriers.

Figures (6)

• Figure 1: Overview of the GlobeReady. a, Data collection and model pre-training. b, GlobeReady disease diagnosis; c, Adaptive ocular disease diagnosis with local retrieval augmentation; d, Confidence-quantifiable ocular disease diagnosis; e, Feature-based case retrieval.
• Figure 2: Performance of GlobeReady for identifying ocular diseases on different datasets. JLHW11 is a CFP dataset from the Joint Shantou International Eye Center (JSIEC), Longchuan People's Hospital, Haifeng Pengpai Memorial Hospital, and Wuhan Aier Eye Hospital (WAEH), covering 11 categories. JSIEC-OCT15 is an OCT dataset created by Joint Shantou International Eye Center with 15 categories.
• Figure 3: Results of local retrieval augmented ocular disease diagnosis and confidence-quantifiable ocular disease diagnosis. a, The Guangxi Jingliang Eye Hospital (GJEH) dataset includes the same disease categories as JLHW11; b, QCH dataset, from Qingdao Central Hospital, comprises images labeled as normal, AMD, DR, and glaucoma. c, PNH dataset, from Puning People's Hospital, also shares the same categories as JLHW11. d, The Singapore Epidemiology of Eye Diseases (SEED) dataset, collected in Singapore, includes annotations for diagnosing AMD, DR, and glaucoma. e, BDEH dataset originates from Binh Dinh Eye Hospital in Vietnam. f, UKDR dataset, was collected from a DR screening program in the University of Liverpool, the United Kingdom. g, WAEH-OCT13 dataset is a local OCT dataset from the Wuhan Aier Eye Hospital. h, The Zeiss-OCT11 dataset is a local OCT dataset acquired by the device of Zeiss.
• Figure 4: Confidence-quantifiable ocular disease diagnosis. JLHW11 and JSIEC-OCT15 represent the GlobeReady’s performance without introducing confidence-quantification; JLHW11$\_$Co and JSIEC-OCT15$\_$Co indicate the GlobeReady’s performance with confidence-quantification; JLHW11$\_$CoT and JSIEC-OCT15$\_$CoT are the performance of GlobeReady after implementing a confidence threshold recommendation. In this last scenario, samples exhibiting low confidence are flagged for re-evaluation by a clinician.
• Figure 5: GlobeReady's performance in detecting out-of-distribution (OOD) fundus diseases in an open-set scenario. a, Data distribution and corresponding OOD detection rates for the 49 disease categories in CFPOOD49 dataset. b, Data distribution and corresponding OOD detection rates for the 13 disease categories in OCT-OOD13 dataset.