Chest X-Ray Image Analysis of Convolutional Neural Network Models with Transfer Learning for Prediction of COVID Patients

The COVID-19 pandemic grounds a major outbreak around the world, having a severe impact on the health and life of many people globally. The critical step in fighting COVID-19 is the capability to identify the infected patients during initial stages and situate them under extraordinary care. Detecting COVID-19 disease from radiography and radiology images is one of the effective way to detect the infected patients. Inspired by the chest radiograms of patients infected with COVID-19, we study the application of machine learning and convolutional neural network models to detect COVID-19 patients from their chest radiography images. We first prepare a dataset of chest X-rays from the publicly available ieee8023/covid-chestxray-dataset. This paper aims to provide the following contributions. Firstly, the dataset is preprocessed and segregated as healthy, COVID-19, bacterial pneumonia and viral pneumonia. Secondly, the dataset is processed to form the initial trained layers of base model and is fitted with several convolutional neural network models like VGG, ResNet, Xception and DenseNet to extract the high-level general features. Thirdly, the base model of several convolutional neural network models are added to the custom layers developed with transfer learning deep learning approach to analyze the performance of prediction of COVID-19 patients. Fourth, the performance of the convolutional neural network models along with transfer learning is analyzed with the metrics like model loss, precision, accuracy, recall and F-Score. The project is implemented with Python in Spyder under anaconda navigator. Experimental results show that Xception CNN model is found to experience 95% accuracy with the available chest X-ray images.