REACH: Remote Evaluation and Analysis with Certificate-free signcryption for community-level Health in Diabetic Retinopathy Screening

Abstract

Diabetic retinopathy (DR) is a common complication of diabetes that often does not show symptoms in the early stages, so timely screening and diagnosis are crucial. This study proposes a remote DR screening system called REACH. This system integrates a portable fundus imaging device, an artificial intelligence (AI) inference module, and an information management system to make the DR screening process more efficient. The REACH system uses a deep learning model to analyze and classify fundus images. The experimental results demonstrate that the system achieved excellent classification performance on the DDR, APTOS, and EyePACS datasets, with accuracy rates of 89.48%, 85.31%, and 85.85%, respectively, and Kappa coefficients of 0.9297, 0.9207, and 0.9043. Additionally, the REACH system achieved an initial screening accuracy of 88.89% in clinical trials conducted at Zhejiang Provincial People’s Hospital and three affiliated primary care hospitals. These findings indicate that the REACH system has the potential to significantly enhance the coverage and efficiency of DR screening in regions with limited medical resources. In the data collection process, an unsigned encryption algorithm is integrated to unify the signing and encryption functions, making it suitable for resource-constrained remote areas. This approach ensures secure data transmission and protects patient privacy.