OLAYIWOLA, JOY OLUWABUKOLA and Covenant University, Theses (2022) CLASSIFICATION OF CHEST X-RAY IMAGES OF LUNG DISEASES USING DEEP CONVOLUTIONAL NEURAL NETWORK. Masters thesis, Covenant University.
![[img]](http://eprints.covenantuniversity.edu.ng/style/images/fileicons/application_pdf.png) |
PDF
Download (227kB) |
Abstract
The accurate diagnosis of lung disease in infected patients is a crucial step in coping with and combating such diseases. Lung opacity, tuberculosis, COVID-19, bacterial pneumonia,and viral pneumonia are examples of infectious diseases that similarly affect the lungs. Classifying accurately which of these diseases (lung opacity, tuberculosis, COVID-19, bacterial pneumonia, and viral pneumonia or normal) an image of chest Xray is being infected with despite the similarities in the images is crucial. Therefore, this research aimed at developing a convolutional neural network, CNN-based model to classify the lung diseases. In this research work, four convolutional neural network models, MobileNetV2, Resnet-50, ResNet-101, and AlexNet were empirically analysed in order to classify lung diseases from images of chest X-rays. The models were utilised in three classification modes: 6-subclass (lung opacity, tuberculosis, COVID-19, bacterial pneumonia, viral pneumonia, and normal), 5-subclass (lung opacity, viral pneumonia, COVID-19, tuberculosis, and normal), and 4-subclass (lung opacity, viral pneumonia, COVID-19, and normal); to investigate the effect of high interclass similarity. The retrained ResNet-50 architecture provided the best classification accuracy with 97.22%, 92.14%, and 96.08% for 6-subclass, 5-subclass, and 4-subclass respectively. Nevertheless, ResNet-101 has the lowest classification accuracy with 78.12% and 79.49% for 6- subclass and 5-subclass respectively while MobileNetV2 has the lowest classification accuracy of 88.89% for 4-subclass. The findings suggest that the ResNet-50 model can be applied to accurately diagnose lung diseases from chest images of X-rays even with high interclass similarity. Also, this corroborates the success of adopting computer-aided detection (CAD) systems designed for decision support in theclassification of lung diseases.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Lung disease, Deep Learning, Diagnosis, ResNet-50, MobileNetV2, Transfer Learning. |
| Subjects: | T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | AKINWUMI |
| Date Deposited: | 20 Jan 2023 15:12 |
| Last Modified: | 20 Jan 2023 15:12 |
| URI: | http://eprints.covenantuniversity.edu.ng/id/eprint/16541 |
Actions (login required)
 |
View Item |
