Development Of A Deep Learning Model To Classify X-Ray Of Covid-19, Normal And Pneumonia-Affected Patients

Abstract

Pneumonia is commonly seen in several diseases, including Covid-19 that has put countries under lockdown today [1]. Other than antigen rapid test kit (RTK) and reverse transcription-polymerase chain reaction (RT-PCR), an alternative method to detect COVID-19 is through the examination of patients’ chest radiography (CXR). However, the results of manual inspections may be false and the misdiagnosis could lead to fatal consequences such as delayed treatment and death. The manual inspection can be inconsistent, inaccurate and may differ from different individuals due to different perspectives. Often, Covid-19 Xrays are misinterpreted as bacterial pneumonia. With the advancement of technology, this issue can be overcome by developing a Convolutional Neural Network (CNN) model to categorize X-ray of normal, pneumonia-affected and COVID-19 patients via deep learning. In this work, various CNN models (ResNet-50, ResNet-101, Vgg-16, Vgg-19 and SqueezeNet) were trained with the public databases that contain a combination of 1345 viral pneumonia, 1200 COVID-19 in addition to 1341 regular CXR images. The transfer learning method was employed, aided by image augmentation for training and validation of ResNet-50, ResNet-101, Vgg-16 and Vgg-19 architectures. Meanwhile, SqueezeNet was trained from scratch to investigate the importance of transfer learning to the model. The highest training accuracy achieved in this study was 97.38% by the VGG-16 model using a learning rate of 0.01 whereas the highest weighted average accuracy achieved was 94% by the VGG-16 model using a learning rate of 0.01 and the VGG-19 model using a learning rate of 0.001. The reliability and high accuracy of the CNN model would open a new avenue for the diagnosis of Covid-19.