- Title
- Investigation of methods for synthetic CT generation, evaluation and implementation for male and female pelvis MRI-only radiotherapy
- Creator
- Choi, Jae Hyuk
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2023
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- MRI-only radiotherapy treatment planning has gained increasing attention in recent years because it reduces systematic errors between CT and MRI and improves treatment planning workflows. It involves generating a CT-like scan since MRI does not provide electron densities of different tissue classes necessary for radiation dose calculations. Studies have investigated different algorithms to generate sCT from the MRI, and demonstrated the accuracy of the sCT radiation dose calculation using retrospective patient data. However, there are still many issues that need to be addressed for the safe and accurate implementation of MRI-only planning into clinical care: 1) The sCT scan is a computer-generated image that can contain errors propagated from the artefacts of the MRI. Methods to ensure the quality of the generated scans are required. 2) Commonly used metrics for measuring the image representation similarity of the sCT may not correlate with dose calculation agreement to the reference CT. Also, the dosimetric verification of the sCT is a time-consuming multi-step process and cannot be easily integrated into the sCT generation and evaluation pipeline. 3) The sCT generation methods currently implemented for clinical application require high computational time to generate new sCT scans. The image quality is highly susceptible to the registration accuracy between CT and MRI and the registration to the atlas MRI scans. 4) Studies have reported that deep learning-generated sCT scans have the highest accuracy in image similarity and absorbed dose calculation compared to the corresponding reference CT. However, the registration error present in the training and test MRI-CT pairs could affect the learning process of the models and result in sub-optimal models and inaccurate output sCT. 5) Many studies have performed the comparison analysis between the sCT and the reference CT calculations for prostate cancer, but few studies have extented these comparisons to the female pelvis or other pelvic treatment sites such as the rectum, anal canal, cervix and endometrium. This thesis addresses the outlined issues. Firstly, an effective and robust dosimetric QA method was developed to validate sCT on a patient-specific basis in a clinical MRI-only radiation therapy workflow. The method provided accurate dose agreement to the reference CT (rCT) calculation and the result are within clinically acceptable limits. Secondly, the developed sCT QA method was applied as a dosimetric verification tool in a large scale multi-centre MRI-only prostate radiotherapy clinical trial. In addition, the method was used to further optimize the dosimetric agreement of the output sCT of a commercially available sCT generation product to the reference CT calculation. Thirdly, a simple and fast dosimetric evaluation metric of newly generated sCT scans was developed to facilitate the process of sCT generating model development. The metric calculation and the dose difference between the sCT and rCT were observed to be highly correlated with a linear fit. The metric was then used to develop a dosimetrically robust deep learning-based sCT generation method for male and female pelvic cancer MRI-only treatment planning. During the development process, the importance of registration accuracy in the training and test datasets was investigated. Finally, the output sCT of the developed sCT generation deep learning model was then compared to that of the other common synthetic CT generation methods for male and female rectum, anal canal, cervix and endometrium treatments.
- Subject
- CT; MRI; pelvis; radiotheraphy
- Identifier
- http://hdl.handle.net/1959.13/1500761
- Identifier
- uon:55010
- Rights
- Copyright 2022 Jae Hyuk Choi
- Language
- eng
- Full Text
- Hits: 234
- Visitors: 252
- Downloads: 28
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details Download | ATTACHMENT01 | Thesis | 3 MB | Adobe Acrobat PDF | View Details Download | ||
| View Details Download | ATTACHMENT02 | Abstract | 295 KB | Adobe Acrobat PDF | View Details Download |