A COVID-19 screening method based on eyes photographs and artificial intelligence. (20th October 2021)

• Record Type: Journal Article
• Authors: Li, F; Xue, XY; Boned Fustel, P; Rong, SS; Sun, Q; Tang, HC; Wang, WX; Fu, Y; Boned-Ombuena, A; Gu, MW
• Is Part Of: European journal of public health. Volume 31(2021)Supplement 3
• Publication Date: 2021-10-20
• Subjects: Epidemiology -- Europe -- Periodicals; Public health -- Europe -- Periodicals; 362.109405
• DOI: 10.1093/eurpub/ckab164.503
• Languages: English
• ISSNs: 1101-1262

Abstract

Abstract: Background: SARS-CoV-2 can spread both from symptomatic and asymptomatic individuals. Ocular manifestations due to SARS-CoV-2 have been described, being conjunctival inflammation the most common affectation. Evidence shows that conjunctivitis could be the first and/or only manifestation of COVID-19. This study aimed to develop and validate a COVID-19 screening method based on eyes photographs and artificial intelligence. Methods: In this multicentre study, 1, 200 participants were enrolled from Shanghai Public Health Clinical Center (SPHCC) Fudan University, AIMOMICS LAB and La Fe University and Polytechnic Hospital (LFUPH) of Valencia (Spain). Pictures of participants' ocular surface were taken in four different positions with mobile phone cameras, and a Deep Learning System (DLS) was developed through machine learning to identify characteristic conjunctival inflammation patterns. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committees of SPHCC and LFUPH. Results: The area under the receiver-operating-characteristic curve (AUC), sensitivity, specificity, and accuracy were calculated according to the results of our binary classification network. Bootstrapping with 1, 000 replicates was used to estimate 95% confidence intervals of the performance metrics, with photography as the resampling unit. On the subject-level classification, the network achieved the AUC of 0.976 (95% CI 0.965-0.988) among Asian population and 0.892 (95% CI 0-763-1.000) among Caucasian population. Conclusions: Preliminary results show that this DLS performed well in identifying probable asymptomatic COVID-19 cases through the analysis of participants' eyes pictures. This method could be an innocuous, accessible, low cost and quick COVID-19 screening method. Eventually, it could potentially contribute to pandemic control. Key messages: In the context of the COVID-19 pandemic it would be useful to have a screening method to easily and quickly detect asymptomatic individuals, in addition to using temperature control. Preliminary results show that this Deep Learning System (DLS) based on eyes pictures taken with mobile phone cameras could be an innocuous, accessible, low cost and quick COVID-19 screening method.