Exchange Programme

The purpose of the exchange programme was to process and reconstruct list-mode data acquired from the Siemens PET-MR scanner as processing data from a PET-MR scanner is not straightforward, generally due to the incompatibility between the format of the scanner and the user software. In addition, Reconstruction taking into account attenuation, normalisation, random and scatter effects is an essential step. During the visit to UCL the main achievements were: that all necessary datasets are now readable; a script now exists for different reconstruction algorithms; and that reconstructions were successfully performed creating a range of images. We plan to make this script available to other users of the Siemens PET-MR scanner.

The purpose of the exchange was to become familiar with the data acquisition process with the GE PET-MR system and with the use of the respiratory tracking device currently utilized in the centre (that is a pressure belt), and to acquire patient data with MR sequences that could allow the detection of an MR navigator (as representative of the internal respiratory motion of the organs). Moreover, data acquired with the GE PET-MR scanner were meant to be used as test datasets for the utilization of a feature of the STIR code, that allows the user to unlist the listmode files into sinograms (after the recent addition of the scanner to the library).

The main intention of the visit was to explore the use of a spatially variant penalty strength in penalized image reconstruction using anatomical information. As Parallel Level Sets (PLS) has shown promising results in literature, it was chosen to be incorporated into the previously proposed preconditioned algorithm (L-BFGS-B-PC) for achieving both good image quality and fast convergence rate. Since the penalty function has been well-studied by the host group, the other purpose of this exchange was to learn from their knowledge and experience on parameter optimization for PLS. The software developed (on top of STIR) during the visit can be used as a starting-point for integration into the CCP PETMR software which would allow users to utilize PLS with anatomical information derived from MR images during PET reconstructions. During this visit I interacted with several members of the group of Prof. J. Nuyts, most intensively with Dr Anna Turco and Dr Georg Schramm. In addition to helping me build up essential knowledge for the proposed joint project, they also provided plenty information regarding live in the center and the city so that I could get used to the new environment quickly and have a productive visit.

This exchange programme was proposed to contribute towards the current devel-opments in Synergistic Image Reconstruction Framework (SIRF) to read and reconstruct positron emission tomography (PET) raw data from GE SIGNA PET/magnetic resonance (MR) scanner. There were three main challenges that were faced during PET data reconstruction with SIRF for real data produced by GE SIGNA PET/MR scanner prior to this exchange programme. These challenges included:
1. Random data correction: Randoms correction could not be carried out for the data acquired by GE SIGNA scanner.
2. Rotated images: There was a view angle o set between GE reconstructed image and SIRF for reconstructed images.
3. Time of Flight (ToF) reconstructions: Reconstructions with recent ToF code (to be included in SIRF) for real phantom data from the scanner was not validated. This ToF code will allow SIRF users to be able to reconstruct real ToF-PET data from GE SIGNA PET/MR (a scanner with 390 ps timing resolution) with all data corrections.
This visit was aimed to resolve all the above challenges. Additionally, PET/MR datasets were acquired using VQC calibration phantom provided by GE with GE SIGNA scanner located at Imanova, UK with the help of Dr. Gaspar Delso. This VQC phantom data was used to validate the rotation angle between GE and STIR. The dataset was also used to validate the random correction after the code was included in STIR. This dataset will prove to be bene cial for PET and MR data alignment that will be included within SIRF and will be made open source via Dementia Platform United Kingdom (DPUK). Another dataset with 2 GE-68 spheres as discussed in section 2.2, was collected at Imanova and this was used to validate the ToF code as discussed later in this report. Finally, all the code-based developments and outcomes of the visit are aimed to be available within SIRF as a part of this exchange programme.

Due to the distance travelled for this exchange, it was decided that the exchange should take place between multiple host facilities and over a longer period of time, in this case two months, than typical CCP PETMR Exchanges. Additionally, due to the extra cost, the exchange was funded by CSIRO, CCP PETMR and University of Queensland. The project was originally divided between two projects, which became four projects during the exchange. These were: implementing HKEM into STIR and SIRF; developing a motion modelling technique to capture discontinuous lung motion; developing PET data-driven head motion estimation techniques; and implementing geometrical awareness and consistency into STIR and SIRF. Due to the increased scope of work and limited time, all projects require some additional work for completion. However, this was a worthwhile trade-off, as the benefit from the short time on exchange was maximised.

The purpose of the exchange was the integration of a numerical simulation framework for dynamic PET-MR data into the Synergistic Image Reconstruction Framework (SIRF). The exchange was part of a larger effort to implement a simulation framework able to produce MR and PET rawdata in a format which makes simulation results compatible with standard image reconstruction packages such as SIRF. In addition, the simulation framework should also provide ground truth motion information on the displacement of organs due to respiratory and cardiac motion to validate image registration and motion correction methods. After implementing the MR part of the framework at PTB, Berlin, the simulation was able to simulate the MR acquisition process. The goal of the exchange was to integrate the PET rawdata simulation and to design and implement the software structures to simulate dynamic processes (e.g. respiration, heart beat, contrast uptake) for both modalities.

This exchange was co-funded by CCP PET-MR and CSIRO. In this exchange, I aimed to achieve four primary goals: to make progress on (1) the development of an automated pipeline for PET reconstruction from raw data extracted from the scanner through to reconstructed equivalent to manufacturer DICOM output; to (2) apply the recently integrated HKEM technique in STIR/SIRF to a larger dataset in order to determine its effectiveness in a 18F-Florbetaben study; to (3) learn from UCL’s experience in kinetic modelling of dynamic PET for use at CSIRO; and finally, to (4) develop the work to be presented at MIC on PCA motion regression to be expanded into a journal publication. Given the broad scope of work to be achieved, most of the visit was spent developing plans and preliminary implementations of the intended products, and work will be continued by myself in Australia.

The purpose of the research exchange visit to UCL was to familiarise myself with the algorithms developed within the CCP PET-MR for PET reconstruction with MR sideinformation, and to investigate PET-MR synergistic reconstruction using the SIRF software. Another purpose of this visit was to see if these algorithms could be extended to SPECT reconstruction using PET side-information as a potential method for Partial Volume Correction (PVC) in theranostic PET/SPECT studies, for improved activity estimates in small lesions in the reconstructed SPECT image.
After attending the SIRF conference/symposium in Chester I headed down to London to begin my research exchange visit to UCL. The first few days of my visit was spent setting up SIRF and STIR natively on my Macbook so that I could begin testing and debugging the implementation of anatomical priors using the algorithms developed within the CCP for PETMR. Once I had STIR and SIRF installed on my laptop I started investigating PET and SPECT reconstruction using Daniel Deidda’s Hybrid Kernelised Expectation Maximisation (HKEM) algorithm developed for PET reconstruction using MR side-information as an anatomical prior. A few bugs and potential improvements were discovered throughout my investigations; this led to several Github issues raised as well as suggestions for future additions and alterations to the HKEM algorithm implemented in STIR.
Since my visit to UCL I have been using the HKEM algorithm in STIR to investigate the potential of PET-SPECT synergistic reconstruction for the theranostic pairing Copper-64 (Cu-64) PET and Copper-67 (Cu-67) SPECT as a method for PVC; using clinical data from a meningioma cohort. The initial results have been very promising and has led to an abstract submission to the SNMMI 2020 conference: “SPECT/CT-based Dosimetry in PRRT: Using Theranostics to Minimise the Impact of the Partial Volume Effect”.

Due to comparatively long measurement times in PET/MR imaging, patient movements during the measurement are likely. These lead to artefacts which have a negative impact on the visual assessment and quantitative validity of the image data and, in the worst case, lead to misinterpretations. Simultaneous PET/MR systems allow the registration of movements and enable correcting the PET data. In order to assess the effectiveness of the motion correction methods, it is necessary to carry out measurements on phantoms that are reproducible moved. For this purpose an experimental setup was developed for specific movements during my master thesis in Germany. Due to a cooperation between Leipzig (UKL) and London the data I measured during my master thesis could be used. An MR-compatible robotic system (Innomotion by INNOMEDIC GmbH) was used to generate rigid movements of a brain-like phantom acquired on a Siemens mMR. Different motion estimates were compared with the robot-induced motion.

Initially written for support of PET and MR data, the Synergistic Image Reconstruction Framework (SIRF) was extended in v3.3.0 to handle SPECT data with parallel-hole collimators using the ‘SPECTUB’ projector class from the Software for Tomographic Image Reconstruction (STIR). A 3D SPECT system matrix modelling library specific for pinhole collimators was recently developed for STIR to enable corrections for the spatially variant collimator-detector response and attenuation by incorporating their effects into the system matrix. The inclusion of the pinhole-SPECT library in SIRF could greatly benefit the research community given the recent advancements in imaging technology, namely in the preclinical setting. This opens up exciting potential for synergistic imaging with pinhole-SPECT systems. Therefore, the objectives of the exchange were as follows:
1. Finalize the integration of the pinhole-SPECT library into STIR,
2. Expand SIRF by adding a new pinhole-SPECT acquisition model, and
3. Investigate the use of SIRF in multi-radionuclide pinhole-SPECT acquisitions.
To elaborate on Objective 3, multi-radionuclide SPECT data reconstructed with conventional energy windows can contain cross-talk which refers to undesired γ-rays detected in an energy window. Cross-talk can be caused by down-scattered photons from higher energies or direct overlap of photopeaks in an energy window. My previous work explored the triple energy window cross-talk correction method to subtract undesired events from the primary window. While the subtraction of presumed cross-talk events improved contrast between simultaneously acquired radionuclides, it also increased noise and reconstruction uncertainty due to compromised count statistics. Thus, a multi-radionuclide reconstruction method was desired to optimize cross-talk correction without compromising count statistics.

Traditional X-ray radiotherapy is increasingly being sidelined by high-energy charged particle irradiation techniques, for the treatment of various types of tumours. These new techniques result in a precise high-dose delivery to the tumour area, reducing the risk to surrounding healthy tissues. Neutrons allow a highly selective therapeutic modality when used in neutron capture therapy (NCT), providing additional advantages due to their higher relative biological effectiveness (RBE). The working principle of NCT is to exploit the high thermal neutron cross-section of specific nuclides, loaded in the target to damage the tumour volume. Boron Neutron Capture Therapy (BNCT) arises from these considerations. A radiopharmaceutical containing 10B is conveyed to the tumour area and irradiated with thermal (or epithermal) neutrons. The capture reaction of 10B with thermal neutrons produces high Linear Energy Transfer (LET) particles, which deposit the total therapeutic dose at cellular level, and γ-rays emitted at particular energies that exit the patient’s body. Since these γ-rays are emitted in situ, and their emission rate is proportional to the neutron capture reaction rate. Their detection allows real-time monitoring and quantifying of the delivered radiation dose and its distribution during the therapy. For this purpose, dedicated tomographic imaging systems based on SPECT (Single Photon Emission Computed Tomography) have been proposed, but never made. In recent years, Boron Neutron Capture Therapy (BNCT) has experienced a significant surge in popularity, largely attributed to remarkable innovations within the particle accelerator field. This evolution has made BNCT more accessible and practical for clinical applications. Unlike the past, when neutrons were primarily available only within nuclear reactors, the contemporary breakthroughs in accelerator technology have enabled hospitals to integrate neutron sources directly within their facilities.

The primary aim of this project was to successfully read and utilise listmode data from the Siemens Biograph Vision and Siemens Biograph Vision Quadra scanners. Using phantom or patient data from the University Medical Centre Groningen, the goal was to generate data necessary for developing motion detection and motion tracking algorithms. One potential candidate for motion
detection explored was Bottom-Up Segmentation (BUS), which could be used with listmode data to achieve high temporal and spatial resolution in motion detection.