COVID-19 AND PNEUMONIA DETECTION SYSTEM USING DEEP LEARNING WITH CHEST X-RAY IMAGES

Abstract

By now, the world has been suffering from the COVID-19 pandemic for almost 1.5 years. While the knowledge about the virus variants, the symptoms, the tests, and the treatment have improved, and various vaccines have been developed, the pandemic is still going on. Severe SARS-CoV-2 infection could cause respiratory failure and damage to the internal organs, especially the lungs. Thus, diagnosis needs to be done using radiological examination (X-ray, CT scan or MRI). The challenge is that the COVID-19 markers are difficult to distinguish from some other diseases such as Pneumonia, and not every health center has a radiologist to make the diagnosis. This is particularly true in many ASEAN and other developing countries. This research proposes the use deep learning technique for an automatic detection of COVID 19 with chest X-ray (CXR) images, which is the most commonly available and affordable medical images. The focus is to accurately distinguish between COVID-19, Pneumonia and healthy CXR. The VGG-16 Convolutional Neural Network (CNN) algorithms and its variants are exploited and compared with other transfer learning models e.g. InceptionV3, ResNet50 and Xception. The accuracy graph, confusion matrix and performance metrics are produced for evaluation and analysis in searching for better approaches for the final model. While there are many recent studies in the same topic, most of them only consider the software parts of the solution, that often still rely on expensive computer system to run the computational algorithm. This research offers one step further solution, proposing the Graphical User Interface (GUI) design and the embedded system implementation, allowing to build an affordable but reliable portable detection system to assist radiologist to diagnose COVID-19 and Pneumonia through patients CXR more accurately in shorter time, and to help  other medical practitioners to make emergency diagnosis of the diseases, in the absent of a  radiologist.