DETECTION OF COVID-19 FROM CHEST X-RAY USING DEEP LEARNING

Abstract

As the Coronavirus Disease 2019 (COVID-19) have left devastating consequences over the world, an effective screening test procedure is crucial to strengthen public health and put the disease to a halt. This study introduces a refined pipeline to train deep learning predictive models for the detection of COVID-19 from chest X-rays, as well as the resulting models themselves. The pipeline involves multiple techniques to combat overfitting and optimize predictive results, such as data augmentation, Bayesian optimization for hyperparameter tuning, selecting the appropriate performance metric, and early stopping during model training. On the COVID-XRay-5K v3 dataset, the three models, ResNet50, NASNet-A-Mobile, and Xception, achieved the areas under the precision-recall curves of 0.9773, 0.9633, and 0.9003; and the areas under the receiver operating characteristic curves of 0.9940, 0.9964, and 0.9812, respectively. At 98% recall (sensitivity), they sustained high specificity of 97.53%, 97.60%, and 86.00%. With such performance, these deep learning models are promising tools to aid in the combat against the pandemic.