Performance evaluation of CNN architectures for COVID-19 detection from X-ray images

Early detection of the COVID-19 infection is the key to avoiding fatalities. Chest radiography has proven to be an effective and low-cost solution for detecting the virus. It is important to evaluate the potential of deep learning models for COVID-19 detection from the x-ray images for quick and early detection of COVID-19 with high accuracy. We conducted a study that evaluates the potential and performance of various Convolutional Neural Networks (CNN) architectures for detecting the COVID-19 on a dataset consisting of 5902 chest X-ray images having 2276 instances of X-ray images of COVID-19 patients and 3626 images of healthy and non-COVID-19 pneumonia X-rays. The performance of the models is assessed using metrics like accuracy, specificity, sensitivity, F1 Score, ROC curve, etc. The results suggest that the DenseNet-121 model proved to be the better choice among evaluated architectures for COVID-19 detection from X-ray images in terms of overall performance with an accuracy of 98.2%, sensitivity of 97.6%, and specificity of 98.4%. We conclude that there is a need for further evaluation of the CNN architectures on large, real-world, and diverse datasets for obtaining generalizable results for a reliable diagnosis.

Automated summary

Early detection of COVID-19 infection is crucial to prevent fatalities. Chest radiography has been proven effective and affordable for detecting the virus. This study evaluates the performance of various CNN architectures on a dataset of 5902 chest X-ray images, and finds that the DenseNet-121 model performs best in terms of accuracy, sensitivity, and specificity.