Development of Clinical Workflow

To allow our methodology to be used within the clinic, or for a larger scale validation study, a number of steps would need to be undertaken.

9.2.2.1

Protocol

The extra one minute motion model sequence utilised in the last few chapters could be acquired immediately after the routine clinical protocol (as per the methodology in this work), or could be incorporated within the clinical protocol. The duration of PET to be motion corrected can range from 3 minutes per bed position for whole-body scans, to up to 60-90 minutes for research studies with one bed position. For a whole-body scan with four bed positions, two of the positions adversely affected by motion (thorax and abdomen) may be extended from three minutes to four minutes each, with no change to the protocol for the initial three minute bed position (see Figure 9.1). In some clinical

9.2. Future Work 170 PET MR Model Clinical 4 min 3 min

Figure 9.1: Proposed whole-body clinical workflow with extended PET bed position durations for motion model acquisitions.

protocols, the duration of MR acquisition may be shorter than the PET duration, in which case the motion-capturing MR sequence can be acquired with no time penalty. Acquiring the extra MR close in time to the clinical data reduces the effects of breathing style changes and tracer washout.

9.2.2.2

Computational Pipeline

Throughout this thesis, data processing has been carried out on a number of operating systems (Mac OS, Ubuntu, Windows) with various scripts in Matlab, STIR and bash. In order to produce results efficiently, a pipeline is required to automate processes, from raw data exported from the scanner to motion-compensated PET reconstruction. This will allow the method to be tested with a larger patient cohort or potentially to be used for specific research projects or clinical use.

9.2.2.3

Image Visualisation

For all motion correction work in this thesis, the resulting PET image has been warped to an ’exhale’ state, and reference images for image registrations are chosen at exhale. For clinical use, it may be useful for the respiratory state of the motion-corrected PET image to be flexible, to allow for it to match the respiratory state of any of the diagnostic MR images from the same PET bed. For example, if a patient cannot successfully hold their breath at exhale for a breath-hold MR acquisition and does so at inhale (as seen for MRAC scans throughout the thesis), a motion-corrected PET image at inhale would be better for image visualisation, especially for fusion of PET and MR images.

An example is given in Figure 9.2, where the PET-derived signals P(t) and P’(t) are plotted for a four minute bed position, as part of a whole-body scan. Four diagnostic MR sequences are acquired throughout at various respiratory states, either at inhale

9.2. Future Work 171 −10 0 10 20 30 P(t) P’(t) MR ! BREATH-HOLD (INHALE)! p = 10! p’ = 0! MR ! BREATH-HOLD (EXHALE)! p = -8! p’ = 0! MR ! FREE BREATHING! (TRIGGERED – EXHALE)! p = -8! p’ = 0!

Figure 9.2: Typical 4 min PET/MR scan for a single bed position. Four different MR sequences acquired at either inhale breath-hold, exhale breath-hold or free breathing. Values of PET-derived signals can be used to warp motion-corrected PET images to spatially align with each MR acquisition.

or exhale breah-hold, or free-breathing, triggered to only acquire data at the exhale position. The values of signals during each MR acquisition can therefore be used as input to the motion model to estimate deformation fields, and a motion-corrected PET image can be reconstructed to spatially align with each MR acquisition. PET images could also be reconstructed at a respiratory state dependant on usage, e.g. at mid-cycle for certain types of radiation planning.

9.2.2.4

Metric for Motion Correction Performance

In general, clinicians have adapted to read PET images with artefacts known to be caused by respiration, and may not trust motion correction techniques due to a lack of robustness, clinical validation and clinical practicality [McClelland et al., 2013]. These issues have been addressed in this thesis, but examples have been shown where motion correction makes no significant change to images, or even adds artefacts in extreme cases. For clinical use, our proposal is that motion corrected PET is to be used alongside non motion-corrected AC and NAC PET images, rather than instead of. A metric to allow the clinician to understand how well the motion model performed, and therefore how much to ’trust’ the motion corrected PET image would be useful.

One option would be to use a visual tool, showing a movie of MR slices at each slice location warped by model-estimated deformation fields. The more stationary the images appear, the better the model is performing. The drawback to this method is

9.2. Future Work 172 that only MR images that were used to form the model can be tested, so data from the clinical section of the scan is untested. A movie of reconstructed PET gates could also be used as a visual tool, but a simplified version of the model would need to be used with a low number of gates to ensure high enough contrast in each.

Other image-based quantitative measures used throughout the thesis such as Euclidean Distance and Mutual Information on MR data, and Tenengrad Variance Sharpness on PET data could be utilised, but absolute values may be difficult to interpret.