IntroductionPositron emission tomography (PET) is a molecular imaging technique; three-dimensional images of functional processes within the body are typically produced using iterative reconstruction methods. Image optimisation is a significant challenge, as the optimal combination of acquisition and reconstruction parameters is dependent upon image context and the clinical task, and there are thousands of possible combinations of selectable parameters. Furthermore, as PET technology continues to evolve, advances need to be optimised for different clinical indications. Manufacturers of PET imaging systems often suggest generic reconstruction strategies for tumour imaging of standard patients; however, imaging departments should validate and optimise reconstruction strategies for clinical applications of interest. This is particularly true when the use of such advances prove to be controversial, as remains the case for Point Spread Function (PSF) modelling.Whilst many publications have assessed the effects of various reconstruction parameters, there is no established methodology for the assessment and optimisation of clinical reconstruction parameters. The primary aim of this thesis is therefore to develop a generic methodology to assess and optimise PET image reconstruction that can be applied to any clinical application. The ability to detect small, low intensity lesions within the liver is critical for effective patient management; however, such lesions are challenging to identify in Fluorine-18 Fluorodeoxyglucose (18F-FDG) PET imaging due to the relatively high liver background activity. Despite its clinical importance, there is no established optimal method for PET liver reconstruction. The secondary aim of this thesis is therefore to optimise image reconstruction for small liver lesion detection in 18F-FDG-PET imaging for a specific PET system: the General Electric Medical Systems (GEMS) Discovery 690 PET-CT system.MethodsPhantom studies were undertaken to assess the effects of varying acquisition and reconstruction parameters upon image noise, spatial resolution and lesion detection. The effects of slice overlap upon image quality were assessed to determine if the GEMS recommended setting was appropriate. The effects of Time of Flight (TOF), PSF, effective iterations, post-reconstruction filtering and voxel size on image quality were then assessed. A human observer study using patient data was also undertaken to determine if recommendations based on phantom data were applicable to clinical liver imaging. Different phantom acquisition and analysis techniques were compared and used to develop a generic methodology that can be used to optimise PET acquisition and reconstruction for different clinical tasks. This methodology includes instructions for a patient observer study. Phantom image acquisitions were largely designed to reflect liver imaging but can be adapted to other clinical scenarios. Phantom and patient work therefore led to the recommendation of a specific reconstruction strategy to optimise liver lesion detection. ResultsThe use of a 23% slice overlap between image acquisition frames, as recommended by GEMS, was shown to produce acceptable image quality under routine clinical practice. An amendment to European guidelines was proposed to better reflect the relationship between image noise and slice overlap when calculating minimum patient injection activities. A clinically relevant methodology for spatial resolution measurement was developed using activity concentrations and voxel values that reflect clinical imaging. Full width half maximum (FWHM) measurements were shown to be reliable and the use of a simplistic background activity correction method was assessed and justified. The effects of reconstruction parameters upon image noise and clinical spatial resolution were assessed using phantom data. PSF was shown to degrade spatial resolution at low iterations (<180) when applied without TOF. Furthermore, noise analyses revealed the GEMS Gaussian filter implementation did not function as intended at certain filter widths; larger voxels demonstrated greater noise levels than smaller voxels. This was unknown to GEMS engineers and was a novel finding.Qualitative phantom assessments concluded applying TOF and PSF together, with no filtering and approximately 108 iterations, was optimal for small lesion detection. Joint analysis of Hot Contrast Recovery Coefficient (HCRC) and noise were shown to be a better predictor of observer lesion detection preferences than signal-to-noise ratios (SNR). Furthermore, a novel SNR calculation (based on region-to-region noise) was shown to be a better predictor of human observer preference than traditional SNR calculations (based on voxel-to-voxel noise). Reassuringly, reconstruction parameters suggested by GEMS (54 iterations, 4mm filter, 3.65mm voxels) were found to be close to optimal. The patient observer study (n = 30) did not demonstrate any statistically significant differences in lesion detection between the GEMS suggested reconstruction and three progressively sharper reconstructions based on the phantom results. However, results did suggest the use of a sharper reconstruction (54 iterations, no filtering, 2.73mm voxels) may improve lesion detection.ConclusionsA four-step generic methodology for optimising PET acquisition and reconstruction is proposed by this thesis. This methodology includes instructions for selecting an appropriate slice overlap for image acquisition. Phantom acquisition and analysis techniques are included for assessing spatial resolution, image noise and lesion detection in a clinically relevant manner. Recommendations are also made for conducting a patient observer study.This thesis further concludes that all 18F-FDG oncology patient images acquired using the GEMS Discovery 690 PET-CT system should be reported using two reconstructions in tandem. The GEMS suggested reconstruction should be used for general image interpretation. The sharper reconstruction should be used for liver lesion detection.
【 预 览 】
附件列表
Files
Size
Format
View
Optimisation and assessment techniques for clinical PET reconstruction
PDF
download
878987797
028-85220240
