Deep learning-based diagnostic system for malignant liver detection
dc.contributor.advisor | Wu, FangXiang | |
dc.contributor.advisor | Luo, Yigang | |
dc.contributor.committeeMember | Guo, Huiqing | |
dc.contributor.committeeMember | Bui, Francis | |
dc.contributor.committeeMember | Johnston, James | |
dc.contributor.committeeMember | Hu, Ting | |
dc.creator | Khan, Rayyan Azam | |
dc.creator.orcid | 0000-0001-5816-1251 | |
dc.date.accessioned | 2022-12-19T20:49:16Z | |
dc.date.available | 2022-12-19T20:49:16Z | |
dc.date.copyright | 2022 | |
dc.date.created | 2022-11 | |
dc.date.issued | 2022-12-19 | |
dc.date.submitted | November 2022 | |
dc.date.updated | 2022-12-19T20:49:17Z | |
dc.description.abstract | Cancer is the second most common cause of death of human beings, whereas liver cancer is the fifth most common cause of mortality. The prevention of deadly diseases in living beings requires timely, independent, accurate, and robust detection of ailment by a computer-aided diagnostic (CAD) system. Executing such intelligent CAD requires some preliminary steps, including preprocessing, attribute analysis, and identification. In recent studies, conventional techniques have been used to develop computer-aided diagnosis algorithms. However, such traditional methods could immensely affect the structural properties of processed images with inconsistent performance due to variable shape and size of region-of-interest. Moreover, the unavailability of sufficient datasets makes the performance of the proposed methods doubtful for commercial use. To address these limitations, I propose novel methodologies in this dissertation. First, I modified a generative adversarial network to perform deblurring and contrast adjustment on computed tomography (CT) scans. Second, I designed a deep neural network with a novel loss function for fully automatic precise segmentation of liver and lesions from CT scans. Third, I developed a multi-modal deep neural network to integrate pathological data with imaging data to perform computer-aided diagnosis for malignant liver detection. The dissertation starts with background information that discusses the proposed study objectives and the workflow. Afterward, Chapter 2 reviews a general schematic for developing a computer-aided algorithm, including image acquisition techniques, preprocessing steps, feature extraction approaches, and machine learning-based prediction methods. The first study proposed in Chapter 3 discusses blurred images and their possible effects on classification. A novel multi-scale GAN network with residual image learning is proposed to deblur images. The second method in Chapter 4 addresses the issue of low-contrast CT scan images. A multi-level GAN is utilized to enhance images with well-contrast regions. Thus, the enhanced images improve the cancer diagnosis performance. Chapter 5 proposes a deep neural network for the segmentation of liver and lesions from abdominal CT scan images. A modified Unet with a novel loss function can precisely segment minute lesions. Similarly, Chapter 6 introduces a multi-modal approach for liver cancer variants diagnosis. The pathological data are integrated with CT scan images to diagnose liver cancer variants. In summary, this dissertation presents novel algorithms for preprocessing and disease detection. Furthermore, the comparative analysis validates the effectiveness of proposed methods in computer-aided diagnosis. | |
dc.format.mimetype | application/pdf | |
dc.identifier.uri | https://hdl.handle.net/10388/14385 | |
dc.language.iso | en | |
dc.subject | Computer-aided diagnosis | |
dc.subject | Liver malignancy | |
dc.subject | Preprocessing | |
dc.subject | Medical imaging | |
dc.subject | Hepatocellular carcinoma | |
dc.subject | Metastasis | |
dc.subject | Segmentation | |
dc.subject | Deblurring | |
dc.subject | Enhancement | |
dc.title | Deep learning-based diagnostic system for malignant liver detection | |
dc.type | Thesis | |
dc.type.material | text | |
thesis.degree.department | Mechanical Engineering | |
thesis.degree.discipline | Mechanical Engineering | |
thesis.degree.grantor | University of Saskatchewan | |
thesis.degree.level | Doctoral | |
thesis.degree.name | Doctor of Philosophy (Ph.D.) |