Título / Title

LONGITUDINAL COMPARATIVE ANALYSIS OF THE EFFECTIVENESS OF ARTIFICIAL INTELLIGENCE IN THE SCREENING/DIAGNOSIS OF RETINOPATHIES AND SELF-AWARENESS OF THE DISEASE.

Introdução / Purpose

The study aimed to evaluate the sensitivity and specificity of portable retinal cameras with artificial intelligence (AI) in screening for diabetic patients at HUCAM/UFES in 2022 and 2023, as well as comparing and observing the technological evolution of these devices over the course of a year, and also to understand the level of self-knowledge of diabetic patients about the disease and its progression during the period studied.

Material e Método / Methods

This cross-sectional observational study assessed the performance of an algorithm using retinal images obtained from 190 patients during diabetes screening campaigns in 2022 and 2023 at HUCAM/UFES. Retinal images were captured using handheld devices and interpreted by both human specialists and an AI system. The AI system utilized a modified Xception CNN and GradCam technique for automated Retinopaties detection. Statistical analysis compared device outputs to human readings for sensitivity, specificity, accuracy, PPV, and NPV.

Resultados / Results

In 2022, the device demonstrated a sensitivity/specificity of 95.7% and 86.5% for retinopathy, respectively, with an accuracy of 90%, a positive predictive value (PPV) of 81.5%, and a negative predictive value (NPV) of 97%. In 2023, the device demonstrated a sensitivity/specificity of 95.7% and 94%, respectively, with an accuracy of 94.6%, a PPV of 90%, and an NPV of 97.5%. In the first year, 36 (54%) had never heard of diabetic retinopathy, 76% were unaware of related ophthalmologic exams, and 41 (61%) did not have regular follow-ups with an ophthalmologist. In the second year, these numbers were 54 (56%), 71 (74%), and 65 (68%), respectively.

Discussão e Conclusões / Conclusion

These results demonstrate a significant increase in the sensitivity, specificity, and accuracy of AI in detecting retinopathies, indicating an improvement in the effectiveness of the examination over the course of a year. However, the level of self-awareness among diabetic patients about the disease still requires improvement.

Palavras Chave

Diabetes, Diabetic retinopathy, Handheld camera, Single image, Artificial intelligence

Area

CLINICAL RETINA

Institutions

Federal University of Espirito Santo - Vitória - Espírito Santo - Brasil

Authors

THIAGO CABRAL, BIANCA CABRAL, LEONARDO ZAMPROGNO, JOÃO SAMPAIO