Key points are not available for this paper at this time.
BACKGROUND Diabetic retinopathy (DR) is the leading cause of visual impairment and blindness among individuals with diabetes mellitus. Early detection and timely intervention are crucial for preventing irreversible vision loss. However, traditional methods of DR screening are labor-intensive and reliant on the availability of skilled personnel, posing challenges in resource-constrained settings. Objective This study aims to evaluate the effectiveness of artificial intelligence (AI) in detecting the severity of DR compared to conventional ophthalmological assessments. METHODS A hospital-based observational study was conducted at the ophthalmology outpatient department of Tertiary care hospital, India over a six-month period. A total of 300 diabetic patients were included, and fundus photographs were obtained using a fundus camera. The images were then analyzed using an AI model trained on a diabetic retinopathy dataset. The severity of DR was graded according to established criteria, and the accuracy of the AI model was compared to that of ophthalmologist grading. RESULTS The AI model demonstrated an accuracy rate of 95.25% in grading the severity of DR. Comparison between AI and ophthalmologist grading showed close sensitivity and specificity rates across different DR grades, with the AI model slightly outperforming in certain categories. CONCLUSIONS Artificial intelligence shows promise as an effective and efficient tool for the screening and diagnosis of diabetic retinopathy. Its integration into healthcare systems could enhance early detection and treatment of DR, particularly in underserved regions with limited access to ophthalmological services. Further research and validation are warranted to optimize the use of AI in diabetic eye care and ensure its equitable distribution and ethical use.
Sheetal et al. (Tue,) studied this question.