Radiomics is the branch of image analysis concerned with the extraction of statistics (features) concerning not just uptake and shape, but the heterogeneity of the ROI, using spatially-aware textural features to describe the distribution of voxel values. These texture features have shown potential for providing diagnostic and prognostic information. This work examines the robustness of these textural features in order to advise on their future usage in PET studies. The signal-noise-ratio (SNR) of PET data can be approximated by the noise-equivalent count rate (NECR), a scanner- and geometry-dependent performance metric. An investigation was carried out to determine whether textural image features were correlated to the NECR, achieved by acquiring data from phantoms filled with a high activity of 18-F on a Siemens Biograph mCT TrueV over a 12 hour period. Four phantoms were utilised; a cylinder, the NEMA IQ (Image Quality) phantom and two variants using custom-printed tumour-like inserts for the NEMA IQ phantom. The data was recorded in successive 5 and 25 minute frames and images were reconstructed using clinically-appropriate parameters. Radiomics features were extracted using an IBSI (Image Biomarker Standardisation Initiative) compliant method. Strong correlations (|PMCC|>0.9) were found with NECR for 32 out of 75 textural features for large-volume, long time frame images, enabling their characterisation. Multiplication factors were calculated enabling correction of texture features from the value obtained at clinically-expected activity levels to their expected value at peak NECR. Such correlations diminish for short time frame images and when considering smaller ROIs such as the NEMA spheres and phantom inserts. This thesis discusses these results, suggesting methods for calculating a `tumour-specific' noise equivalent count rate to address the diminished textural feature correlations. In addition, work undertaken towards building a Monte Carlo simulation framework to improve this study is discussed.
Date of Award | 31 Dec 2022 |
---|
Original language | English |
---|
Awarding Institution | - The University of Manchester
|
---|
Supervisor | David Cullen (Supervisor) & Peter Julyan (Supervisor) |
---|
- harmonisation
- 3D printing
- NECR
- robustness
- radiomics
- PET
Assessing the robustness of radiomics features in oncology PET
Needham, G. (Author). 31 Dec 2022
Student thesis: Phd