Machine learning-based classification of Covid-19 using chest radiography images / Chan Yu T'ng

• Main Author: Chan, Yu T'ng
• Format: Thesis
• Published: 2021
• Subjects: R Medicine (General); T Technology (General)
• Online Access: http://studentsrepo.um.edu.my/13412/1/Chan_Yu_T'ng.jpg; http://studentsrepo.um.edu.my/13412/8/chan.pdf; http://studentsrepo.um.edu.my/13412/

The novel coronavirus disease, also known as COVID- 19, was first reported in Wuhan, China and has since spread around the world. Up to July 2021, it has infected over 197 million people and caused over 4 million associated death worldwide. As the number of reported cases escalates, most countries are running out of resources. The scarcity of testing kits, lengthy testing time, and the growing number of daily cases urged researchers around the world to devise alternative methods such as medical imaging to be used in conjunction with computer aided diagnosis systems, to assist radiologists and physicians in detecting COVID- 19 cases more quickly and reliably. The aim of this project is to implement a machine learning- based binary classifier that can detect COVID- 19 positive cases from COVID- 19 negative cases using chest CT images. Transfer learning technique in conjunction with VGG16 and ResNet50 architecture has been adopted in developing the binary classifier. To achieve an optimal performance of the baseline models, many performance improvement strategies such as data augmentation, re- training of weights, fine- tuning of hyperparameters, and 5- fold cross validation have been implemented and incorporated. Thorough experimentation demonstrates that the proposed classification models are computationally less expensive while yielding astoundingly good results where Model 1 based on VGG19 architecture achieved an accuracy of 95.19% and Model 2 based on ResNet50 architecture achieved an accuracy of 98.29%.