Machine learning models for image-based diagnosis and prognosis of COVID-19: A systematic review.

Researchers

Journal

Modalities

Models

Abstract

In order to provide the best possible care for COVID-19 patients and reduce the burden on the health care system, accurate and timely diagnosis and effective prognosis of this disease is important. Machine learning methods can play vital roles in diagnosing COVID-19 by processing chest x-ray images.
Our aim of this study is to summarize information on the use of intelligent models for diagnosing and prognosing the COVID-19 to help early and timely diagnosis of the disease to help with health.
A systematic search of the PubMed, Web of Science, IEEE, ProQuest, Scopus, bioRxiv, and medRxiv databases up to 24 May 2020 is performed. To conduct this study, PRISMA guidelines were followed. All original articles applying image processing for predicting and diagnosing the COVID-19 disease were considered. Two reviewers independently assessed original papers to determine eligibility for inclusion. Risk of bias was evaluated by using Prediction Model Risk of Bias Assessment Tool (PROBAST).
Of the 629 articles retrieved, 45studies were included. The review identified 4 prognosis models for calculating prediction of disease severity and estimation of confinement time, for individual patients, and 41 diagnosis models for detecting COVID-19 from normal or other pneumonias. Most articles used deep learning methods based on CNN networks which have been used widely as a classification algorithm The most frequently reported predictors of prognosis in patients with COVID-19 included age, CT data, gender, comorbidities, symptoms and laboratory findings. Deep CNN obtained better results compared with non-Neural Network-based methods. Moreover, all of the models are in high risk of bias due to the lack of information about study population, intended groups and inappropriate reporting.
Machine learning models for diagnosis and prognosis of COVID-19 showed excellent discriminative performance approximately. However, these models were at high risk of bias, because of various reasons like low information about participants, randomizing process and lack of external validation. Therefore, it leads to optimistic report in their models. In hence this review doesn’t recommend any of the current models to be used in practice.