High-throughput Large-model-based AI-assisted Diagnosis Using OCT

N/ANot Yet RecruitingNCT07249307

Peking Union Medical College Hospital2,000 enrolled

Overview

This observational study aims to establish key technologies for high-throughput, large-model-based AI-assisted diagnosis using optical coherence tomography (OCT) and OCT angiography (OCTA). The study will collect real-world OCT/OCTA images and corresponding clinical information from patients with common blinding retinal and optic nerve diseases at Peking Union Medical College Hospital. A high-throughput diagnostic framework based on large-scale artificial intelligence models will be developed and evaluated. The primary objective is to determine the diagnostic performance of the AI system, including its ability to identify diabetic retinopathy, branch retinal vein occlusion, central retinal vein occlusion, age-related macular degeneration, pathologic myopic choroidal neovascularization, and glaucoma-related optic nerve damage. The results of this study are expected to support the development of standardized, efficient, and scalable AI-assisted diagnostic pathways for OCT imaging in clinical practice.

This study investigates key technologies for high-throughput, large-model-based AI-assisted diagnosis using optical coherence tomography (OCT) and OCT angiography (OCTA). OCT/OCTA imaging has become an essential non-invasive tool for detecting and monitoring retinal and optic nerve diseases, yet manual interpretation remains time-consuming, experience-dependent, and limited by inter-observer variability. Recent advances in large artificial intelligence models provide an opportunity to develop scalable, generalizable diagnostic tools that can process large multimodal datasets and support clinical decision-making. This observational study will enroll patients who undergo routine OCT and/or OCTA examinations at Peking Union Medical College Hospital and who are diagnosed with one or more of the following conditions: diabetic retinopathy, branch retinal vein occlusion, central retinal vein occlusion, age-related macular degeneration, pathologic myopic choroidal neovascularization, or glaucoma with optic nerve damage. The study will include both retrospectively collected and prospectively acquired imaging and clinical data, following standardized quality control and data-management procedures. The high-throughput diagnostic framework will be trained and validated using large-scale image and clinical datasets. Primary outcomes include diagnostic performance metrics such as the area under the receiver operating characteristic curve (AUC). Secondary outcomes include sensitivity, specificity, and lesion-level or structural feature assessment when applicable. No experimental intervention will be introduced, and all imaging and clinical evaluations will follow standard clinical care. The study aims to produce a robust, clinically relevant benchmark for large-model-based AI systems in OCT/OCTA interpretation and provide technical support for future integration of AI-assisted diagnostic tools into routine ophthalmic practice.

Eligibility

Sex: ALL

Medical Language ↔ Plain English

Inclusion Criteria: * 1\. Patients of any age or sex who undergo OCT and/or OCT angiography (OCTA) examinations as part of routine clinical care at Peking Union Medical College Hospital. 2\. Clinical diagnosis of at least one of the following conditions: Diabetic retinopathy, Branch retinal vein occlusion, Central retinal vein occlusion, Age-related macular degeneration, Pathologic myopia with choroidal neovascularization and Glaucoma with optic nerve damage. 3\. Imaging quality sufficient for analysis based on predefined OCT/OCTA quality control criteria. 4\. Ability to provide informed consent (for prospective participants), or availability of medical records that meet institutional ethical requirements (for retrospective data). Exclusion Criteria: \- 1. Poor-quality OCT/OCTA images that do not meet analysis standards (e.g., severe motion artifacts, media opacity, incomplete scans). 2\. Patients unable to cooperate with standard ophthalmic imaging procedures. 3. Any condition judged by investigators to preclude accurate imaging evaluation or reliable diagnostic interpretation.

Outcomes

Primary Outcomes

Diagnostic performance of the AI-assisted OCT/OCTA model (AUC for multi-disease classification)

Area under the receiver operating characteristic curve (AUC) of the high-throughput large-model-based OCT/OCTA diagnostic system for identifying major retinal and optic nerve diseases, including diabetic retinopathy, branch retinal vein occlusion, central retinal vein occlusion, age-related macular degeneration, pathologic myopic choroidal neovascularization, and glaucoma.

Time frame: Baseline imaging visit (time of image acquisition and model inference).

Secondary Outcomes

Sensitivity and specificity of the AI-assisted OCT/OCTA model

Sensitivity and specificity of the AI system for detecting each target disease (diabetic retinopathy, branch retinal vein occlusion, central retinal vein occlusion, age-related macular degeneration, pathologic myopic choroidal neovascularization, and glaucoma) compared with masked clinician consensus.

Time frame: At the time of image acquisition and model inference (baseline imaging visit).

Agreement between AI-assisted diagnosis and clinician diagnosis

Level of agreement between the AI system's classification and the final clinical diagnosis by retina and glaucoma specialists, quantified using Cohen's kappa statistics or similar agreement measures.