Recently, artificial intelligence algorithm has made great progress in the prediction of diabetic retinopathy based on fundus images，showing very high sensitivity and specificity. However，the real-world diagnosis effectiveness of deep learning model is still unclear. This study is designed to evaluate the clinical efficacy of such an algorithm in detecting referable diabetic retinopathy.
This prospective clinical study is designed to evaluate the real-world diagnostic performance of an AI model in detecting referable diabetic retinopathy (RDR, defined as more than mild NPDR), by evaluating its sensitivity and specificity compared to the clinical reference standard- seven-field stereoscopic photography. The subjects enrolled in this study are patients with T1DM or T2DM. Qualified color fundus images and seven-field stereoscopic photography images of each eyes of the subject are taken. The fundus images are graded for RDR by the algorithm under test, and seven-field stereoscopic photography images of the same eye are graded by ophthalmologist, which serving as the gold standard to compare the algorithm performance against. The trial plans to enroll 1000 subjects. With a 95% confidence interval, the sensitivity is expected to be at least 87% whereas the specificity at 89% or above. The quality of fundus images are assessed according to the National DR Screening Imaging and Grading Guideline published by Chinese Ophthalmological Society and Chinese Medical Doctor Association in 2017. The grading of RDR is based on the National DR Clinical Diagnosis and Treatment Guideline published by Chinese Ophthalmological Society in 2014. A brief overview of the clinical protocol is as follows: Candidate recruiting phase: recruiting qualified participants; Clinical phase: imaging and diagnosing by AI and ophthalmologist ; Statistical analysis phase: comparing two outputs; Closing phase: final report and archiving
Study Type
OBSERVATIONAL
Enrollment
1,000
Patients with diabetes enrolled will undergo nonmydriatic fundus imaging and seven-field stereoscopic photography. The images will be run on an artificial intelligence (AI) algorithm. The diagnosis of the AI algorithm will be compared to the diagnosis of seven-field stereoscopic photography by ophthalmologist. Sensitivity and specificity will be calculated to evaluate the performance of AI algorithm.
Shenzhen second peoples's hospital
Shenzhen, Guangdong, China
RECRUITINGSensitivity and specificity
To evaluate the sensitivity and specificity of the model in detecting referable DR (more than mild NPDR)
Time frame: No more than 1 day for each subject
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.