MedPath

Real-time Artificial Intelligence System for Detecting Multiple Ocular Fundus Lesions by Ultra-widefield Fundus Imaging

Conditions
Diagnostic Screening Programs
Artificial Intelligence
Diagnostic Imaging
Abnormality of the Fundus
Interventions
Device: Taking an ultra-widefield fundus image
Registration Number
NCT04859634
Lead Sponsor
Sun Yat-sen University
Brief Summary

This prospective multicenter study will evaluate the efficacy of a real-time artificial intelligence system for detecting multiple ocular fundus lesions by ultra-widefield fundus imaging in real-world settings.

Detailed Description

The ocular fundus can show signs of both ocular diseases (e.g., lattice degeneration, retinal detachment and glaucoma) and systemic diseases (e.g., hypertension, diabetes and leukemia). The routine fundus examination is conducive for early detection of these diseases. However, manual conducting fundus examination needs an experienced retina ophthalmologist, and is time-consuming and labor-intensive, which is difficult for its routine implementation on large scale.

This study will develop an artificial intelligence system integrating with ultra-widefield fundus imaging to automatically screen for multiple ocular fundus lesions in real time and evaluate its performance in different real-world settings. The efficacy of the system will compare to the final diagnoses of each participant made by experienced ophthalmologists.

Recruitment & Eligibility

Status
UNKNOWN
Sex
All
Target Recruitment
2000
Inclusion Criteria

All the participants who agree to take ultra-widefield fundus images.

Exclusion Criteria
  1. Patients who cannot cooperate with a photographer such as some paralytics, the patients with dementia and severe psychopaths.
  2. Patients who do not agree to sign informed consent.

Study & Design

Study Type
OBSERVATIONAL
Study Design
Not specified
Arm && Interventions
GroupInterventionDescription
IKang Physical Examination CenterTaking an ultra-widefield fundus imageThe participant only needs to take an ultra-widefield fundus image as usual.
Beijin Tongren HospitalTaking an ultra-widefield fundus imageThe participant only needs to take an ultra-widefield fundus image as usual.
Shenzhen Ophthalmic CenterTaking an ultra-widefield fundus imageThe participant only needs to take an ultra-widefield fundus image as usual.
Xudong Ophthalmic CenterTaking an ultra-widefield fundus imageThe participant only needs to take an ultra-widefield fundus image as usual.
Zhongshan Ophthalmic CenterTaking an ultra-widefield fundus imageThe participant only needs to take an ultra-widefield fundus image as usual.
Yangxi General Hospital People's HospitalTaking an ultra-widefield fundus imageThe participant only needs to take an ultra-widefield fundus image as usual.
Guangdong Provincial People's HospitalTaking an ultra-widefield fundus imageThe participant only needs to take an ultra-widefield fundus image as usual.
Primary Outcome Measures
NameTimeMethod
Accuracy8 months

Performance of artificial intelligence system for detecting multiple ocular fundus lesions based on ultra-widefield fundus imaging.

Secondary Outcome Measures
NameTimeMethod
False-positive rate8 months

Features of Misclassification

Sensitivity8 months

Performance of artificial intelligence system for detecting multiple ocular fundus lesions based on ultra-widefield fundus imaging.

Cohen's kappa coefficient8 months

The comparison between the performacne of AI system and ophthalmologists of three degrees of expertise.

Specificity8 months

Performance of artificial intelligence system for detecting multiple ocular fundus lesions based on ultra-widefield fundus imaging.

False-negative rate8 months

Features of Misclassification

Data processing time of AI system8 months

Data processing time of AI system.

Trial Locations

Locations (1)

Zhongshan Ophthalmic Center, Sun Yat-sen University

🇨🇳

Guangzhou, Guangdong, China

© Copyright 2025. All Rights Reserved by MedPath