Abstract
Commercial positron emission tomography (PET) systems are continually evolving as manufacturers strive to improve image quality, whilst reducing radiation dose and acquisition times. The impact of these developments on clinically relevant measures must be evaluated to ensure PET imaging is used appropriately. Clinical trials designed specifically to evaluate new developments in PET technology or methodology are expensive and time consuming, so alternative evaluation approaches are required. One potential way to evaluate new PET developments is to use simulated PET data along with forward modelling in a ‘virtual imaging trial’ framework. The purpose of this project was to develop and validate the methodology for generating PET datasets created by insertion of simulated lesions into real clinical PET datasets. This approach required an accurate and reliable method for combining simulated lesions where the ground truth was known, with the physiological uptake in real patient data with a realistic range of weight, body mass indices and image noise levels. The validated methodology was then used for the evaluation of new PET technologies, in particular use of time-of-flight (TOF), point spread function (PSF) modelling and a Bayesian penalised likelihood (BPL) reconstruction algorithm.The first stage of the work in this thesis involved designing a model of the General Electric (GE) Healthcare Discovery 710 PET scanner to perform Monte Carlo simulations of realistic lesions and to develop the methodology for inserting the lesions into real PET datasets in projection space. To validate the scanner model, measurements of spatial resolution and sensitivity were performed according to the National Electrical Manufacturers Association (NEMA) NU-2 standard using simulated phantoms and results were compared to those measured for the real scanner. Additionally, accuracy of corrections was assessed using a simulated acquisition of a uniform cylinder.
The validity of the insertion technique was tested by comparing recovery coefficients derived from simulated spheres inserted into real PET datasets of the background compartment of the NEMA image quality phantom to those from real acquisitions of the phantom containing identical sized spheres. Anonymised PET datasets from 10 patients who had a measurable pulmonary lesion were used to assess the ability of the technique to generate realistic lesions. Characteristics from the real lesions were used to generate simulated lesions that were then inserted into the contralateral lung of the same patient. To demonstrate the simulated lesions were indistinguishable from real lesions, a two-alternative forced choice task was performed by an experienced PET physician whereby they were asked to review the reconstructed PET images for each patient and choose which of the two lesions they thought was simulated and rate their confidence in identifying it.
To ensure that PET images created using the insertion technique could be used in place of real PET data and a real PET imaging system for performing clinically relevant tasks, PET images were generated for a cohort of 97 patients consisting of simulated lesions with characteristics matching those from a population of real patients with known solitary pulmonary nodules (SPNs). Quantitative measures of 18Fluorine-labelled fluorodeoxyglucose (18F-FDG) uptake and diagnostic accuracy for the assessment of malignancy risk were then compared between the cohorts of patients with real and simulated lesions.
Finally, the impact of new PET technologies on image quality and task-based measures was evaluated using phantom and patient images generated using the validated methodology. The influence of TOF, PSF modelling and BPL reconstruction on image quality was investigated using technical image-based measures for the NEMA image quality phantom. For the task-based assessment, 194 lesions were simulated with characteristics representative of real benign and malignant SPNs and inserted into the anonymised raw PET datasets from 194 patients. The PET data were reconstructed using parameters in use across clinical PET Centres in the UK that incorporated TOF, PSF modelling and BPL. The resultant datasets were used to determine the influence of incorporating new technologies on measurements of 18F-FDG uptake using standardised uptake values (SUVs) and the subsequent impact on the diagnostic accuracy for categorising the lesions as malignant or benign. The diagnostic accuracy for each reconstruction was assessed using lesion SUVmax, lesion SUVmean and the Herder score which combines the 18F-FDG uptake with the patient clinical and radiological characteristics to determine risk of malignancy.
| Date of Award | 1 Jul 2023 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Sally Barrington (Supervisor) & Paul Marsden (Supervisor) |