COVID-Net Architecture Modification for Covid-19 Detection on Chest X-ray Images

The coronavirus-based illness known as Covid-19 is easily spread. Because Covid-19 spread quickly and easily, many people were exposed to it. As a result, finding a quick and accurate way to diagnose Covid-19 is necessary, and using chest X-ray pictures is one such option. The best way to describe the health of the lungs in Covid-19 patients is through X-ray scans. The Kaggle website provided the dataset for this study. The 4345 photos in this collection are divided into three classes: 1500 Covid-19 images, 1500 normal images, and 1345 pneumonia images. The Convolutional Neural Network (CNN), which employs the modified COVID-Net architecture, is the technique suggested in the Covid-19 detection procedure. Therefore, an architecture known as COVID-Net was created specifically for diagnosing Covid-19. The COVID-Net architecture uses epoch 10, batch size 32, and a learning rate of 0.0001 with kernel sizes ranging from 7x7 to 1x1. As a result, Covid-19 could be identified from chest X-ray images by COVID-Net. Based on the findings of tests performed using 5-fold Cross Validation, an average F1-score accuracy value of 93% and an AUC value of 98.78% were obtained. Based on the F1 score and AUC values, it can be said that the modified COVID-Net architecture is a useful model.