A DEEP LEARNING APPROACH FOR DIAGNOSIS OF COVID-19 INFECTION AND ITS RELATED FACTORS: A POPULATION-BASED STUDY
Today, there is a high demand for artificial intelligence (AI) applications in distinct areas of research. AI can be used in the medical context to help in clinical decision-making and limited resource allocation. The present study proposes the best model for the detection of COVID-19, the prediction of disease in new cases, and also determines the top significant features related to COVID-19, using DL algorithms as a subset of AI techniques.
In this retrospective population-based study, 10862 individuals suspicious of COVID-19 participated. The information was collected from 35 different hospitals across Khorasan-Razavi province, Northeast of Iran, from 20 February 2020 to 21 June 2021. We employed artificial neural networks (ANN), random forests (RF), decision tree (DT), support vector machines (SVM), boosted trees (BT), and logistic regression (LR) DL algorithms. Our findings indicated that the RF model had higher performance than all other algorithms. The RF algorithm had a sensitivity of 66%, specificity of 95%, precision of 88%, accuracy of 85%, and AUC of 74%. Our study found that the common top predictors for detecting COVID-19 were: age, SpO2, reception season, CT result, contact history, sex, and fever. RF model can aid in clinical decision-making and limited resource allocation. This model needs to be externally validated in larger populations, more features, and multicenter settings.
artificial intelligence, deep learning, random forest, COVID-19, diagnosis.