Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques

Lung diseases impose a financial burden on society. Early detection of lung diseases may result in lifesaving treatments. In view of the need for an efficient treatment, scientists contend that deep learning has a great potential for diverse applications in aiding the diagnosis of lung diseases in m...

Full description

Saved in:

Bibliographic Details
Main Author:	Kieu, Stefanus Tao Hwa
Format:	Thesis
Language:	English English
Published:	2023
Subjects:	RC705-779 Diseases of the respiratory system
Online Access:	https://eprints.ums.edu.my/id/eprint/41411/1/24%20PAGES.pdf https://eprints.ums.edu.my/id/eprint/41411/2/FULLTEXT.pdf
Tags:	Add Tag No Tags, Be the first to tag this record!

id	my-ums-ep.41411
record_format	uketd_dc
institution	Universiti Malaysia Sabah
collection	UMS Institutional Repository
language	English English
topic	RC705-779 Diseases of the respiratory system
spellingShingle	RC705-779 Diseases of the respiratory system Kieu, Stefanus Tao Hwa Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques
description	Lung diseases impose a financial burden on society. Early detection of lung diseases may result in lifesaving treatments. In view of the need for an efficient treatment, scientists contend that deep learning has a great potential for diverse applications in aiding the diagnosis of lung diseases in medical imaging. In previous research, it was shown that deep learning has been utilized to classify lung diseases in a variety of publications. However, the majority of researchers employed features extracted automatically using convolutional neural networks (CNN) in their published studies. To the best of our knowledge, the number of ensemble-based works is likewise restricted. Thus, this research aims to produce a lung diseases classification framework by ensembling classifiers trained from features extracted from x-ray images and edge images. This research employs a modified edge detection technique to produce a new type of feature, uses image augmentation to increase the number of training images, and uses a modified weighted averaging ensemble to increase classification accuracy. The methods applied in this research is suitable to tackle the various problems in the field of computer vision, including limited available dataset, data imbalance and the lack of diverse features during ensemble. This research is significant because the production of a deep learning aided lung disease classification system can assist medical officers to detect lung diseases. There are three reasons to develop a computer-aided lung disease classification system. Reasons to develop this system also include reducing human workload, overcoming human exhaustion, and help health services in areas with a lack of medical expertise. In this research, classifiers were developed to classify chest x-rays into four conditions: COVID-19, pneumonia, tuberculosis, and normal (healthy). In this respect, the deep learning methods employed in this work include CNN, transfer learning, data augmentation, and ensemble. VGG16 and InceptionV3 were the CNN architectures used to extract features in this research. This is due to the fact that these two CNNs had been applied in other works of literature and have produced high accuracy classification models. Also, an enhanced Canny edge detection technique was introduced. This enhanced approach addresses many shortcomings of the conventional Canny technique and has been shown to be more accurate. This enhanced Canny approach was then used to generate an alternative edge image training dataset. With this alternative dataset available, a novel ensemble approach called accuracy-based weighted averaging was presented to combine classification result from classifiers trained from different features. This ensemble approach was utilized to increase the classification accuracy, sensitivity, and specificity of the individual classifiers by combining their probability scores. Accordingly, a closer analysis of the results reveals that the best performing ensemble combination achieved an accuracy of 92 %, a sensitivity of 98%, 86.9%, 95.6%, 87.5% for COVID-19, normal, pneumonia, and tuberculosis, respectively, and a specificity of 97.4%, 96.17%, 98.61%, 96.61% for COVID-19, normal, pneumonia, and tuberculosis. Moreover, the findings provide consistent accuracies ranging from 82 % to 96 %, indicating that this ensemble method has better classification results than single classifiers. We believe that this paradigm may be applicable to various diseases and image types, such as computed tomography scans or sputum smear microscopy images.
format	Thesis
qualification_name	Doctor of Philosophy (PhD.)
qualification_level	Doctorate
author	Kieu, Stefanus Tao Hwa
author_facet	Kieu, Stefanus Tao Hwa
author_sort	Kieu, Stefanus Tao Hwa
title	Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques
title_short	Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques
title_full	Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques
title_fullStr	Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques
title_full_unstemmed	Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques
title_sort	framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques
granting_institution	Universiti Malaysia Sabah
granting_department	Faculty Of Science And Natural Resources
publishDate	2023
url	https://eprints.ums.edu.my/id/eprint/41411/1/24%20PAGES.pdf https://eprints.ums.edu.my/id/eprint/41411/2/FULLTEXT.pdf
_version_	1818611402228629504
spelling	my-ums-ep.414112024-11-07T07:41:11Z Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques 2023 Kieu, Stefanus Tao Hwa RC705-779 Diseases of the respiratory system Lung diseases impose a financial burden on society. Early detection of lung diseases may result in lifesaving treatments. In view of the need for an efficient treatment, scientists contend that deep learning has a great potential for diverse applications in aiding the diagnosis of lung diseases in medical imaging. In previous research, it was shown that deep learning has been utilized to classify lung diseases in a variety of publications. However, the majority of researchers employed features extracted automatically using convolutional neural networks (CNN) in their published studies. To the best of our knowledge, the number of ensemble-based works is likewise restricted. Thus, this research aims to produce a lung diseases classification framework by ensembling classifiers trained from features extracted from x-ray images and edge images. This research employs a modified edge detection technique to produce a new type of feature, uses image augmentation to increase the number of training images, and uses a modified weighted averaging ensemble to increase classification accuracy. The methods applied in this research is suitable to tackle the various problems in the field of computer vision, including limited available dataset, data imbalance and the lack of diverse features during ensemble. This research is significant because the production of a deep learning aided lung disease classification system can assist medical officers to detect lung diseases. There are three reasons to develop a computer-aided lung disease classification system. Reasons to develop this system also include reducing human workload, overcoming human exhaustion, and help health services in areas with a lack of medical expertise. In this research, classifiers were developed to classify chest x-rays into four conditions: COVID-19, pneumonia, tuberculosis, and normal (healthy). In this respect, the deep learning methods employed in this work include CNN, transfer learning, data augmentation, and ensemble. VGG16 and InceptionV3 were the CNN architectures used to extract features in this research. This is due to the fact that these two CNNs had been applied in other works of literature and have produced high accuracy classification models. Also, an enhanced Canny edge detection technique was introduced. This enhanced approach addresses many shortcomings of the conventional Canny technique and has been shown to be more accurate. This enhanced Canny approach was then used to generate an alternative edge image training dataset. With this alternative dataset available, a novel ensemble approach called accuracy-based weighted averaging was presented to combine classification result from classifiers trained from different features. This ensemble approach was utilized to increase the classification accuracy, sensitivity, and specificity of the individual classifiers by combining their probability scores. Accordingly, a closer analysis of the results reveals that the best performing ensemble combination achieved an accuracy of 92 %, a sensitivity of 98%, 86.9%, 95.6%, 87.5% for COVID-19, normal, pneumonia, and tuberculosis, respectively, and a specificity of 97.4%, 96.17%, 98.61%, 96.61% for COVID-19, normal, pneumonia, and tuberculosis. Moreover, the findings provide consistent accuracies ranging from 82 % to 96 %, indicating that this ensemble method has better classification results than single classifiers. We believe that this paradigm may be applicable to various diseases and image types, such as computed tomography scans or sputum smear microscopy images. 2023 Thesis https://eprints.ums.edu.my/id/eprint/41411/ https://eprints.ums.edu.my/id/eprint/41411/1/24%20PAGES.pdf text en public https://eprints.ums.edu.my/id/eprint/41411/2/FULLTEXT.pdf text en validuser dphil doctoral Universiti Malaysia Sabah Faculty Of Science And Natural Resources

Framework for lung images classification based on weighted averaging ensemble and enhanced edge detection techniques

Similar Items