A Blinded, Self-control Trial to Evaluate an Artificial Intelligence Based CAD System for Diabetic Retinography

A Blinded, Self-control Trial to Evaluate an Artificial Intelligence Based CAD System for Diabetic Retinography

A Blinded, Self-control Trial to Evaluate an Artificial Intelligence Based CAD System for Diabetic Retinography

To evaluate the safety and performance of an innovative artificial intelligence based Computer-Aided Diagnosis(CAD) system for diabetic retinography, Retinal images of patients with diabetes mellitus or diabetic retinopathy(DR) were collected retrospectively. All images were graded by a retinal specialists expert panel and the CAD device using the International Clinical Diabetic Retinopathy severity scale criteria. Investigator responsible for DR grading by CAD system is blinded to the DR grading results from the expert panel. Finally, DR grading results of the CAD system and experts were compared using sensitivity and specificity.

Retinal images were collected retrospectively according to the following inclusion/exclusion criterion: Inclusion Criterion: Clinical history of diabetes mellitus or diabetic retinopathy; Fully Gradable Images; around 45° field which covers optic disc and macula; complete patient identification information; Exclusion Criterion: incomplete patient identification information; DR grading by expert panel At first, retinal images are graded by three experts independently, then they met for a consensus meeting to discuss cases without initial agreement. If they can't achieve consensus, a final decision is made by the principal investigator. Experts give a grading of both DR and Diabetic Macular Edema (DME) for each image according to the International Clinical Diabetic Retinopathy severity scale criteria and hard exudates around optic disc. Blinding and DR grading by CAD system Before DR grading by CAD system, a randomized identification(ID) is assigned to each retinal image, which ensures that investigator responsible for CAD system operation is masked to the expert panel grading result. Both DR and DME grading is generated by the CAD system and the results are exported. Unblinding Finally, all data are unblinded and results of the CAD system are compared to the results of human grading, which is considered the gold standard, using measures as sensitivity and specificity;

This trial aims to evaluate the diagnostic performance of a CAD system for retinal images； And DR grading by clinicians is used as the golden standard.

1.investigation target 3: Negative: DR grading of 0 or 1; Positive: DR grading of 2 or higher; After completion of DR grading by expert panel and CAD system, results of the CAD system were compared to the results of human grading, which is considered the gold standard, using measures as sensitivity(Se) and specificity(Sp).

investigation target 1: Negative : DR grading of 0 ; Positive : DR grading of 1 or higher; investigation target 2: Negative : DR grading of 0,1 or 2 ; Positive : DR grading of 3 or higher; investigation target 4: Negative : no clinically significant DME; Positive : clinically significant DME; investigation target 5: Negative : no clinically significant DME and DR grading of 0, 1 or 2 ; Positive : clinically significant DME or DR grading of 3 or higher; After completion of DR grading by expert panel and CAD system, results of the CAD system were compared to the results of human grading, which is considered the gold standard, using measures as sensitivity(Se) and specificity(Sp).

Inclusion Criteria: Clinical history of diabetes mellitus or diabetic retinopathy; Fully Gradable Images; around 45° field which covers optic disc and macula; complete patient identification information; Exclusion Criteria: incomplete patient identification information

Xu Y, Wang L, He J, Bi Y, Li M, Wang T, Wang L, Jiang Y, Dai M, Lu J, Xu M, Li Y, Hu N, Li J, Mi S, Chen CS, Li G, Mu Y, Zhao J, Kong L, Chen J, Lai S, Wang W, Zhao W, Ning G; 2010 China Noncommunicable Disease Surveillance Group. Prevalence and control of diabetes in Chinese adults. JAMA. 2013 Sep 4;310(9):948-59. doi: 10.1001/jama.2013.168118.

Williams GA, Scott IU, Haller JA, Maguire AM, Marcus D, McDonald HR. Single-field fundus photography for diabetic retinopathy screening: a report by the American Academy of Ophthalmology. Ophthalmology. 2004 May;111(5):1055-62. doi: 10.1016/j.ophtha.2004.02.004.

Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan S, Widner K, Madams T, Cuadros J, Kim R, Raman R, Nelson PC, Mega JL, Webster DR. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. JAMA. 2016 Dec 13;316(22):2402-2410. doi: 10.1001/jama.2016.17216.