Whybra, Philip
2021.
Standardisation and optimisation of radiomic techniques for the identification of robust imaging biomarkers in oncology.
PhD Thesis,
Cardiff University.
Item availability restricted. |
PDF
- Accepted Post-Print Version
Download (56MB) |
|
PDF (Cardiff University Electronic Theses and Dissertations Publication Form)
- Supplemental Material
Restricted to Repository staff only Download (369kB) |
Abstract
Radiomics is a rapidly evolving field within oncology. It explores the extraction of quantitative features from medical scans to aid in diagnosis, prognosis and monitoring of disease. In effect, these features may act as imaging biomarkers. Radiomics is a potential piece of a multifaceted data puzzle, powering precision medicine approaches, where treatment strategies could be tailored more to the individual than relying on a one-size-fits-all strategy. However, there are crucial challenges within the field regarding reproducibility and reliability of many common radiomic features. This thesis explored the hypothesis that varying but valid approaches to engineered feature extraction can cause discrepancy that harms identification and validation of potential radiomic biomarkers. As a research community, we require guidelines, standards and references to see forward progression and to avoid a replication crisis. Through development of radiomics software, results from this work significantly contributed to a large collaborative consensus benchmarking effort to address this standardisation need. This work investigated the effect of implementation choices on compliance to this new standard. Alongside this, the role of interpolation in radiomics became a key focus through the lens of robustness. Optimal feature extraction should avoid redundancy and utilise robust features. Finally, benchmarking methodology and tools were developed in an effort to standardise the application of filters in image processing steps prior to feature extraction. Discrepancies between radiomics software were identified and evaluated using these tools. The uncertainties in developing optimal and robust radiomic imaging biomarkers that result in clinically useful models are discussed.
Item Type: | Thesis (PhD) |
---|---|
Date Type: | Completion |
Status: | Unpublished |
Schools: | Engineering |
Uncontrolled Keywords: | Imaging Biomarkers, Texture Analysis, Standardisation, Cancer, Robustness, Radiomics |
Date of First Compliant Deposit: | 24 September 2021 |
Last Modified: | 24 Sep 2021 15:37 |
URI: | https://orca.cardiff.ac.uk/id/eprint/144441 |
Actions (repository staff only)
Edit Item |