derive non-rigid deformation fields of myocardial wall motion on a physical beating cardiac phantom, and the motion fields used in the PET reconstruction with the MCIR approach.
Some work has demonstrated the use of deformation information found from both PET and MR images. One method has been proposed with an MR motion model, using results from registration of gated PET images to choose which MR-derived deformation fields to use [Balfour et al., 2015]. The method was tested on MR data from four human volunteers, with simulated PET data. Results were similar to using only MR data to form deformation fields, but the proposed method only required a short MR acquisition to form the model, rather than acquiring throughout the whole PET scan. In this methodology, the gated PET data acts as the surrogate to the motion model, though in practice with real data, a respiratory signal would need to be acquired in order to gate the PET data. Another approach used PET and MR data together to form the deformation fields, by utilising both sets of data in a registration cost function [Fieseler et al., 2014]. This work shows some local improvements over using MR-derived deformations only, though it was tested with simulation data only.
3.10
Summary
PET imaging of the thorax and/or abdomen can be adversely affected by patient respiration, potentially causing quantification errors, blurring of tracer-avid regions, and image artefacts. This Chapter has explained the mechanics that cause this motion through breathing, and some of the tools currently used to try to correct for this motion.
The various methods used to track breathing with a respiratory signal were evaluated, from imaging methods that directly measure diaphragmatic displacement, to external hardware devices and data-driven approaches. Different methods to measure physical displacements with PET and MR imaging were also covered, along with tools such as image registration and PET gating. Once deformations throughout a PET scan have been estimated, and PET data has been grouped into different motion states, a motion-corrected PET reconstruction can then be carried out with one of two methods, either in image space after gated reconstructions (RTA), or during
3.10. Summary 64 reconstruction (MCIR).
In the last few years, many methods to utilise the simultaneity of MR with PET in the simultaneous PET/MR scanner have been proposed. Although work has shown promising results, the main limitation of other proposed methods is the issue of applicability for clinical use. One issue is scan time, and the need to keep this low for high patient throughput in hospitals. For ’snapshot’ imaging or use of bespoke MR sequences such as tagging, MR images are acquired throughout the PET scan to track motion, meaning useful diagnostic MR images cannot be acquired at the same time. This means scan time must be extended to accommodate diagnostic imaging, and extended further if the method is used at multiple bed positions.
Use of motion models partly overcome this problem by collecting motion information in a short space of time, then using this information to infer deformations throughout the PET scan without the use of motion-capturing MR. However, to use motion models, a surrogate signal is required. An MR pencil-beam navigator can provide the signal, but this requires sequence development by insertion into each clinical MR sequence, can’t be continuously acquired, may introduce artefacts in the MR images, and extends scan time. External hardware may also be used, but at a monetary and time cost, with methods requiring careful set up prior (and often during) each patient scan. Data-driven methods may overcome this issue, with PET-derived respiratory signals possible to acquire from the PET data with no need to extend scan time or change imaging protocol. Anatomical location is also an issue, as some motion-capturing MR techniques are designed for motion of a specific anatomy, whereas in clinical practice there is a need to motion-correct all areas as often the diseased area is unknown prior to the patient scan.
Another limitation of recent PET/MR motion correction literature is the lack of clinical validation on patient data. Many studies have been conducted on simulated and physical moving phantoms, which is useful for comparing computational methods such as types of reconstruction, but they do not account for the complex motion of real breathing. For example, with simulated PET data, the deformation fields used to apply respiratory motion are often taken from registration results of 3D low resolution MR images, but these are often unrealistic as these are only an estimation of complex non-rigid diffeomorphic motions. This type of motion will be more easily recovered
3.10. Summary 65 by generic registration methods than real respiratory motions. Some studies have been conducted with primates and human patients, but often subject number is low, and there is a still a need to test methods on a large patient cohort.
In this thesis many of these limitations of previous work are addressed. We demonstrate the capability of a joint PET-MR motion model to predict respiratory motion, and use this to motion-correct any length of PET acquisition with minimal extra scan time and no external hardware used, and finally, test the method on a large patient cohort.
Chapter 4
Validation of a PET-derived
Respiratory Signal
4.1
Motivation
In this chapter, we introduce a technique devised by Thielemans et al. to extract a respiratory signal from raw PET data alone [Thielemans et al., 2012]. In this chapter, the signal was used for PET gating only, and compared with RPM data. We perform a validation of the PET-derived respiratory signal, by comparison with an absolute measure of diaphragmatic displacement using an MR pencil-beam navigator. The aim of this study is to find if the PET-derived signal can provide an accurate measure of diaphragmatic displacement, and therefore be used as a respiratory signal for motion monitoring and correction. We also explore the use of a respiratory cushion for a measure of chest wall displacement, for comparison with the PET-derived signal.