| Author: | Kong, Luoyi |
| Title: | Deep neural networks for computed tomography image super-resolution and tumor segmentation |
| Degree: | M.Sc. |
| Year: | 2023 |
| Subject: | Diagnostic imaging -- Digital techniques Image segmentation Tomography Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Electrical and Electronic Engineering |
| Pages: | iv, 46 pages : color illustrations |
| Language: | English |
| Abstract: | Cancer is one of the diseases with the highest incidence and mortality rates globally. Specifically, in Hong Kong, liver cancer consistently ranks among the top five in terms of both incidence and fatality rates. Timely cancer diagnosis and treatment play a pivotal role in improving patient outcomes and reducing mortality. In clinical practice, pathological examination is commonly regarded as the gold standard for cancer diagnosis. However, invasive procedures, such as biopsies, can lead to severe complications. Conversely, computer-aided diagnosis, particularly in the field of medical image analysis, holds significant value as a non-invasive method to facilitate early cancer detection. Of utmost importance within computer-aided diagnostic systems, medical image segmentation serves as a critical component. Enhancing the accuracy of medical image segmentation provides healthcare professionals with vital clues, expediting cancer diagnosis and optimizing treatment timelines. To address these issues, this study focuses on the low accuracy of medical image segmentation, which can be attributed to the use of low-resolution image acquisition methods to accommodate patient comfort. This often results in poor-quality medical image data, leading to incorrect judgments, not only by artificial intelligence algorithms, but also by experienced doctors. Considering these factors, this study emphasizes the enhancement of medical images to improve the quality of medical image segmentation. To achieve this goal, two innovative approaches are proposed in this dissertation. Firstly, a tailored image enhancement scheme is designed, taking into account the unique pixel characteristics of medical images. Experimental results demonstrate that this scheme produces high-contrast and visually smooth images. Secondly, a novel two-path super-resolution reconstruction algorithm with multi-scale convolutional kernels is developed to simultaneously enlarge images while restoring fine details and textures. This algorithm yields high-resolution and high-quality images, serving as valuable inputs for subsequent analysis. Experimental results show that the reconstructed images obtained by the proposed super-resolution reconstruction algorithm achieves superior results in terms of visual effects and quantitative scores (4.581 higher PSNR score and 0.024 higher SSIM score than the state-of-the-art HAT algorithm). Furthermore, this study reproduces the classic medical image segmentation algorithm Res-UNet and conducts multidimensional comparisons with different image enhancement schemes and image super-resolution reconstruction algorithms, demonstrating the advancement of the proposed medical image enhancement scheme and two-path medical image super-resolution reconstruction algorithm. Finally, the two proposed approaches can effectively improve the quality of medical images, benefiting not only the accuracy of medical image segmentation but also further pathological analysis and radiogenomics analysis. It is believed that medical image enhancement can greatly contribute to the field of medicine. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 8267.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 1.61 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13816