PLoS ONE (Jan 2022)

Clinical evaluation of AI-assisted screening for diabetic retinopathy in rural areas of midwest China

  • Shaofeng Hao,
  • Changyan Liu,
  • Na Li,
  • Yanrong Wu,
  • Dongdong Li,
  • Qingyue Gao,
  • Ziyou Yuan,
  • Guanyan Li,
  • Huilin Li,
  • Jianzhou Yang,
  • Shengfu Fan

Journal volume & issue
Vol. 17, no. 10

Abstract

Read online

Background Although numerous studies have described the application of artificial intelligence (AI) in diabetic retinopathy (DR) screening among diabetic populations, studies among populations in rural areas are rare. The purpose of this study was to evaluate the application value of an AI-based diagnostic system for DR screening in rural areas of midwest China. Methods In this diagnostic accuracy study, diabetes mellitus (DM) patients in the National Basic Public Health Information Systems of Licheng County and Lucheng County of Changzhi city from July to December 2020 were selected as the target population. A total of 7824 eyes of 3933 DM patients were enrolled in this screening; the patients included 1395 males and 2401 females, with an average age of 19–87 years (63±8.735 years). All fundus photographs were collected by a professional ophthalmologist under natural pupil conditions in a darkroom using the Zhiyuan Huitu fundus image AI analysis software EyeWisdom. The AI-based diagnostic system and ophthalmologists were tasked with diagnosing the photos independently, and the consistency rate, sensitivity and specificity of the two methods in diagnosing DR were calculated and compared. Results The prevalence rates of DR according to the ophthalmologist and AI diagnoses were 22.7% and 22.5%, respectively; the consistency rate was 81.6%. The sensitivity and specificity of the AI system relative to the ophthalmologists’ grades were 81.2% (95% confidence interval [CI]: 80.3% 82.1%) and 94.3% (95% CI: 93.7% 94.8%), respectively. There was no significant difference in diagnostic outcomes between the methods (χ2 = 0.329, P = 0.566, P>0.05), and the AI-based diagnostic system had high consistency with the ophthalmologists’ diagnostic results (κ = 0.752). Conclusion Our research demonstrated that DR patients in rural area hospitals can be screened feasibly. Compared with that of the ophthalmologists, however, the accuracy of the AI system must be improved. The results of this study might lend support to the large-scale application of AI in DR screening among different populations.