Convolutional Neural Network (CNN) Based Covid-19 Detection Using Chest X-Ray Image

 




 

Chon, Chee Hao (2022) Convolutional Neural Network (CNN) Based Covid-19 Detection Using Chest X-Ray Image. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

[img] Text
CHON CHEE HAO_Thesis.pdf
Restricted to Registered users only

Download (6MB)

Abstract

The Covid-19 caused a severe pandemic to the world since the year 2019. It is at the urge of need to introduce a Computer-Aided Diagnosis (CAD) system for Covid-19, which can assist medical industries in the world to detect the enormous amount of suspected Covid-19 cases and simultaneously reduce the burden of medical workers. The increase in confirmed cases of Covid- 19 has never stopped since the spread of the virus begins. A Covid-19 detection system utilizing machine learning technology can provide a better classification of Covid-19 positive and negative cases with a minimal labor force. The efficiency of a machine learning-based detection system can detect the positive Covid-19 case at an early stage, which can contribute to slowing down the spread of the virus. This project aims to construct an optimum Convolutional Neural Network (CNN) based Covid-19 detection system using chest X-ray image. Moreover, this project explores possible techniques that can improve the accuracy and reliability of the CNN-based model by applying two different techniques on building the architecture of CNN, which are transfer learning and Faster R-CNN. An experimental study had been conducted to evaluate the architecture building techniques for optimum design of CNN-based Covid-19 detection using chest X-ray images. In this project, a total of 3 different CNN based models are built, trained and evaluated. The first CNN model is a basic CNN based model, which acts as a reference model is constructed for comparison purpose. This followed by Transfer Learning CNN model and Faster R-CNN model which both models are deisgned accourding to ResNet50 architecture. All CNN models are trained and tested with same dataset to ensure the final results obtained are unbiased. The training dataset consists a total of 985 chest X-ray images, where 500 images are negative Covid-19 cases and 485 images are positive Covid-19 cases. While the testing dataset consists of 624 chest X-ray images, 234 images are negative Covid-19 cases and 390 images are positive Covid-19 cases. After all CNN models are trained and tested, the basic CNN model achieved an overall accuracy of 87.3%, while the ResNet50 with transfer learning obtained a higher value which is 93.4%. Faster R-CNN that is also based on ResNet50 attained two sets of results, because the model was trained with a mini batch size of one and four.The overall accuracy accomplished is 83.0% for mini batch size of one and 84.3% for mini batch size of four.

Item Type: Final Year Project
Subjects: Technology > Mechanical engineering and machinery
Technology > Electrical engineering. Electronics engineering
Faculties: Faculty of Engineering and Technology > Bachelor of Mechatronics Engineering with Honours
Depositing User: Library Staff
Date Deposited: 04 Mar 2022 07:26
Last Modified: 04 Mar 2022 07:26
URI: https://eprints.tarc.edu.my/id/eprint/20392