Novel methods of quantifying tumour characteristics on PET/CT and their application in personalised dosimetry and personalised therapy for patients with Neuroendocrine tumours

This study investigated a range of essential tasks required to establish a high quality Theranostic service, including an estimation of the potential occupation exposure, an investigation into local shielding requirements, and the calibration and optimisation of imaging and dosimetric equipment for use with 68Ga and 177Lu. In particular, a thorough investigation of PET quantitative metrics was performed, including the optimisation of 68Ga radiomic feature extraction and selection, an assessment of radiomic feature stability, and a pilot study to investigate the clinical application of radiomic features in NET patient classification. Monte Carlo simulations were employed to determine occupational exposure rates in terms of Hp(0.07), Hp(3) and Hp(10) for a range of clinically relevant positron emitting radionuclides and 177Lu. A range of commercially available local shields were assessed and the optimal shields for use with 177Lu and 68Ga were identified. A multi-centre study was performed to investigate the measured skin dose of staff working with 68Ga labelled radiopharmaceuticals across 8 centres. 68Ga PET SUV metrics were found to be accurate, repeatable and stable over a range of clinically relevant activities. This study confirmed that 68Ga PET imaging has poorer spatial resolution than 18F PET imaging, resulting in reduced recovery coefficients (RCs) for 68Ga. RCs of 68Ga were found to be adversely affected by low activity levels and low contrast, particularly for volumes < 22 mm in diameter. PSF reconstructions resulted in increased SUV metrics and improved contrast recovery, although a smoothing filter was required to reduce edge enhancement. The optimal reconstruction settings for standard and advanced OSEM reconstructions for PET quantification were identified. Optimisation of 68Ga specific radiomic feature extraction and selection methods were performed in this study, using a range of uniform and heterogeneous phantoms, including a novel 3D printed textured phantom. The ability of a reduced set of radiomic features to generate a prediction model for NET patient tumour grade and Ki-67 proliferation rate, using machine learning algorithms, was assessed for various feature set reduction methods. Five robust 68Ga radiomic features were identified that could accurately predict the patient’s NET grade. An investigation into the application of machine learning algorithms to the optimisation of administered activity of 68Ga-DOTATOC was performed, whereby a model that predicted the qualitative perception of IQ, based on the quantitative metrics of 68Ga-DOTATOC patient scans, was generated. The prediction model was found to be a very useful tool in ranking the quantitative metrics of PET by their ability to predict the qualitative perception of IQ, and in identifying patient scans of sub-optimal IQ. The most important quantitative metrics for predicting qualitative IQ were identified, and the most appropriate method of calculating CNR for IQ assessment was confirmed. 177Lu activities assayed using the default 177Lu calibration factor resulted in inaccuracies that exceed the NPL and AAPM tolerance of 5%, for two out of three dose calibrators in our centre, confirming that the optimal calibration factor should be determined independently on each individual dose calibrator. A geometry dependent calibration factor was not required for 177Lu, a single calibration factor was sufficient for all clinically relevant containers and volumes. SPECT sensitivity for 177Lu was found to be stable across a range of activities. The optimal reconstruction parameters for qSPECT were identified on a new generation CZT detector system and a NaI-based imaging system. While comparable RCs were obtained on both systems, the CZT detector system was found to have higher sensitivity, better energy resolution, lower noise properties and to be less affected by the Gibbs artefact.