Artificial Intelligence/Machine Learning Screening for COVID-19 using a US-Patent- Pending Technology known as iDetect COVID-19 Testing Application

Abstract

The COVID-19 pandemic necessitated the development of accurate diagnostics in order to control and minimize viral propagation; however, accurate and remote means of COVID-19 diagnosis have yet to be developed. The objective of this research was to explore the potential of external eye imaging and machine learning (ML) algorithms for early COVID-19 diagnosis via a patent-pending protocol known as iDetect. iDetect employs deep learning techniques to analyze external eye images captured using mobile eye imaging devices. Images were collected through rigorous data collection procedures, including polymerase chain reaction (PCR) testing and image acquisition from participants in multiple countries. Local interpretable model-agnostic explanations (LIME) analysis was utilized for ML model interpretability. The developed convolutional neural network (CNN) and support vector machine (SVM) models demonstrated high diagnostic performance, with the CNN models achieving an area-under-receiver operating characteristic (AUROC) of 92% and exhibiting sensitivity and specificity values exceeding 90% in most eye positions. The SVM models also showed promising performance, with an AUROC of 89%. The findings of this study present a practical and scalable solution for accurate diagnosis of COVID-19, contributing to ongoing efforts to explore innovative diagnostic methods and enhance early detection and management of infectious diseases, such as COVID-19.