Deep-learning based Solution to Detect COVID-19 Infection Using Chest X-rays Images

Ong, Joe Huei (2021) Deep-learning based Solution to Detect COVID-19 Infection Using Chest X-rays Images. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Ong Joe Huei.pdf Restricted to Registered users only Download (4MB)

Abstract

The COVID-19 outbreak has a significant impact on global health. However, the dangerous infections need to prevent from emerging by dampening the hazardous viruses' effects on society. The critical action that can be taken to fight COVID-19 is to possess robust monitoring of the site by forming suspected or infected patients. In the on-going process, the most preferred way to detect the COVID-19 infection through radiological imaging, where a chest X-ray was being used in diagnosis. Early studies stated that the COVID-19 is present in the abnormality in chest X-rays of the infected patients. This may help to express the motivation to research some convincing way to detect the infection of COVID-19, which is a deep-learning-based solution to detect COVID-19 disease using chest X-ray images based on several studies. In this study, the proposed project aims to develop a diagnostic system on COVID-19 detection using deep-leaning and the SqueezeNet model will be introduced and used in the process of recognition. The SqueenzeNet model obtains a high detection rate compared to other methods in the diagnosis of COVID-19 with the use of chest X-ray images. By evaluating this model, a dataset called COVIDx will be used to train the model. The purpose of using this dataset because it is the largest benchmark dataset and open to public. Since the symptoms of COVID-19 may be similar to viral pneumonia so that Explainable Artificial Intelligent was implemented (XAI) into this project. XAI was used to gain better insights into how the COVID-19 diagnostic system makes decisions and prove whether the detection decisions made by XAI are based on relevant information instead of erroneous information that biases the decisions made based on irrelevant visual indicators. Besides, a web application is created by deploying the trained model. It is an independent platform run by everyone who is willing to do the COVID-19 test by accessing the internet. The accuracy obtained by this project is 73.10%.