Título / Title

RETINA SPECIALIST VERSUS DEEP LEARNING SYSTEM FOR DIABETIC RETINOPATHY SCREENING USING TELEMEDICINE

Introdução / Purpose

Diabetic Retinopathy (DR) affects approximately 95% of patients with type 1 Diabetes Melitus (DM) and 60% of patients with type 2 DM, making it the primary cause of legal blindness in adults. Screening for DR, coupled with appropriate referral and treatment, is a universally employed strategy for preventing blindness. In the field of ophthalmology, artificial intelligence (AI) is enhancing diagnostic imaging capabilities. This study aim to assess the accuracy of an AI algorithm in identifying fundus changes, and the necessity of referral to a specialist, in diabetic patients.

Material e Método / Methods

A prospective study was conducted, which involved collecting color retinographies from diabetic patients, both with and without DR. Subsequently, the retinography images were evaluated using the deep learning algorithm created and validated in the Singapore Eye Research Institute. The same images were also analyzed by two Retina specialists, whose impressions were considered the gold-standard. Assessment of agreement between them and the screening accuracy performed by artificial intelligence was then analyzed.

Resultados / Results

A total of 396 eyes from 198 patients were evaluated, with quality images captured in 371 eyes using color retinography. In terms of the presence of DR or macular edema (referrable) or absence of DR (non-referrable), the Kappa correlation index between observers was 0,859. The image analysis algorithm for DR screening demonstrated a sensitivity of 97.3%, specificity of 44.3%, false negative of 1.6%, and false positive rate of 55.7%.

Discussão e Conclusões / Conclusion

AI represents a significant advancement in the field of retinology. The AI algorithm tested in this study encompasses the characteristics of a good screening tool, as it exhibits high sensitivity and a low number of false negatives. This feature ensures that the vast majority of patients with any signs of DR are referred for further evaluation. Therefore, using AI programs alongside retinography can effectively aid in the referral of patients to specialists.

Palavras Chave

Artificial Intelligence; Diabetes; Diabetic Retinopathy; Screening

Area

CLINICAL RETINA

Institutions

Hospital das Clínicas de Ribeirão Preto - USP - Ribeirão Preto - São Paulo - Brasil

Authors

JOÃO PEDRO ROMERO BRAGA, MOISÉS MOURA DE LUCENA, MATEUS CALORI, LEANDRO CHAVEZ, THIAGO CALIL, DANIEL FERRAZ, RODRIGO JORGE