Optimising reconstruction to accommodate complex system models for SPECT.

Lead Research Organisation: University College London
Department Name: Nuclear Medicine

Abstract

There are a number of techniques used to image the human body to provide information on anatomy (structure) or function, to aid in diagnosing a range of diseases. One such technique is Single Photon Emission Computed Tomography (SPECT) which permits physicians to visualise the internal 3D distribution of administered radioactive compounds that typically reflect functional differences between normal and diseased tissues. SPECT is a widely available clinical tool. There are several fundamental difficulties in obtaining accurate and useful images using SPECT. Images tend to be quite blurred but also can have grainy appearance that detracts from the image interpretation. The main reason for this is the need to use a collimator to determine the origin of detected photons. The problem is that to improve image quality requires more counts which either results in additional study time and patient inconvenience or increased radiation dose, which is clearly undesirable. Optimising the collimator offers potential for both reduced study time and reduced radiation dose. A further practical problem is the time taken to compute the final images, which can be quite lengthy as the complexity increases. We propose to adapt computer cards that are currently used in domestic game systems that provide very fast computation. Given that we can improve the efficiency for processing of images, this also opens the possibility to explore even more complex approaches to processing that utilise additional information which is currently acquired but ignored. This offers further potential not only to reduce study time but to also improve clinical image quality. The project has a focus on delivering a practical solution that can be implemented in routine clinical practice.The work will be undertaken under the supervision of three investigators each with complementary skills; physics applied to nuclear medicine (Hutton), reconstruction algorithms (Arridge) and acceleration using graphics computer boards (Ourselin). The work will proceed initially with simulation studies in order to optimise design of collimators prior to these being manufactured and development of the computer programs that will be used for image reconstruction. The developed approach will be implemented on fast hardware and this implementation will be directly compared with conventional approaches to implementation in terms of both speed and image quality. Finally the complete system including collimators and fast computation will be evaluated in human subjects (ethics approval will be sought at the appropriate time prior to commencing human studies). The human studies will involve patients having additional images acquired but will not involve any additional radiation dose. The translation of our research findings directly into clinical practice is an important goal in the research.

Publications

10 25 50
publication icon
Hutton BF (2011) Recent advances in iterative reconstruction for clinical SPECT/PET and CT. in Acta oncologica (Stockholm, Sweden)

publication icon
Thomas BA (2011) The importance of appropriate partial volume correction for PET quantification in Alzheimer's disease. in European journal of nuclear medicine and molecular imaging

publication icon
Pedemonte S (2012) Steady-state model of the radio-pharmaceutical uptake for MR-PET. in Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention

publication icon
Pedemonte S (2012) Steady-state model of the radio-pharmaceutical uptake for MR-PET. in Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention

publication icon
Pedemonte S (2011) 4-D generative model for PET/MRI reconstruction. in Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention

publication icon
Thomas B (2014) Framework for the construction of a Monte Carlo simulated brain PET-MR image database in Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment

publication icon
Hutton B (2013) What approach to brain partial volume correction is best for PET/MRI? in Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment

 
Description Development of robust methods for evaluation of system design

The design of advanced medical imaging instrumentation requires decisions on the design of components and the acquisition protocols that should be used. It is difficult to envisage how changes in design will affect the final acquired images and the clinical usefulness of the system. The original objectives in the project included the optimized design of collimators but this was extended to include more general design considerations. A novel extension of methodology for evaluating system design was developed and has been applied to a unique system that permits adaptive acquisition where data are preferentially acquired from a selected sub-volume. The results provide insight to the necessary acquisition protocols and have demonstrated previously unrecognised noise properties when data are truncated.

Development of novel multi-modal algorithms and a framework for GPU implementation

Throughout the project novel algorithms were explored that not only compensate for variable resolution but also attempt to adapt the final image reconstruction based on complementary information from a second modality (CT or MRI). A challenge is to optimize use of the additional information while maintaining the contrast in regions that are not identified by the second modality. The resulting algorithms can be quite complex and computationally demanding. Use of GPU for acceleration of the developed algorithms has therefore been a critical step resulting in a range of high-speed algorithms; these are provided to fellow researchers as open-source software.

Incorporation of scatter models to derive data-driven attenuation correction

Correction for attenuation is important both for accurate quantification and for reduction of image artefacts. However direct measurement of attenuation is not always possible, especially with some recent novel designs of SPECT system. We have developed an algorithm which permits reconstruction of the attenuation map directly from the acquired emission data, making use of multi-spectral data that arises from Compton scattering of the gamma photons in the body, whose distribution is directly influenced by the attenuation. A feasibility study has been completed but extension to full 3D modelling of scatter, with appropriate corrections for multiple order scatter is work-in-progress. The technique will have widespread clinical application provided the computation can be sufficiently accelerated.

Extension of algorithms to accommodate large-scale problems

As a result of improving computational efficiency through use of GPU, algorithms that incorporate higher dimensions can be accommodated that were previously impractical. For example we have investigated the direct reconstruction of kinetic parameters from a time series of tomographic studies treating the data as 4-dimensional (3D over time) rather than as a series of independent 3D acquisitions. Alternative approaches can be taken to incorporate motion through joint registration and reconstruction, development that is highly relevant to emission tomography, where the study duration is relatively long and some degree of patient motion unavoidable.

Evaluation of algorithm performance for clinical application

In the latter stages of the project the quantitative accuracy of the developed reconstruction algorithms that incorporate prior anatomical information has been directly compared with alternative post-reconstruction approaches that aim to reduce the bias that results from limited resolution. This comparison requires knowledge of truth (a gold standard) unavailable from clinical data. However our goal was to simulate clinical data as closely as possible, including possible uncertainties introduced as a result of realistic biological variation in tracer distribution. The resulting datasets provide a useful basis for further comparison of quantitative reconstruction and post-reconstruction processing algorithms and will be made available to a wider research community subsequent to the publication of current results.

Summary

The defined objectives of this challenging project were largely met. As with most projects of this scale there was need to redefine realistic end-points and to adapt the project so as to minimize risk. Early work that aimed to assist design of novel collimators was put on hold, pending the development of suitable evaluation methodology; the demonstration of a robust method for system evaluation means this can now be revisited. Overall we exceeded our expectations in developing a number of novel algorithms for multi-modal reconstruction, simultaneous emission/transmission reconstruction, partial volume correction and 4D reconstruction with further publication based on this work in progress.
Exploitation Route Open-source software has been made available which may assist other researchers in evaluating the new approaches to reconstruction with ultimate use in improving quantification of imaging studies in healthcare. A patent was lodged on one of the developed computer algorithms. However UCL decided not to pursue this for economic reasons.
Sectors Healthcare

 
Description Reconstruction software developed under the project has been made available as open source for evaluation of its potential use in healthcare. Optimised methods of data acquisition have been demonstrated for a specific SPECT system (D-SPECT) which can be implemented in clinical practice. Developed methods are being extended for use in the current grant on exploiting PET/MRI technology.
First Year Of Impact 2012
Sector Healthcare
Impact Types Societal

 
Description EC FP7
Amount € 4,600,000 (EUR)
Organisation European Commission 
Sector Public
Country European Union (EU)
Start 03/2013 
End 02/2017
 
Title PV correction evaluation 
Description Synthetic data based on real clinical studies provides a ground truth for evaluating partial volume correction algorithms. 
Type Of Material Database/Collection of data 
Year Produced 2014 
Provided To Others? Yes  
Impact Work is in progress to evaluate different PV correction techniques with a view to identifiying the most promising methods for clinical implementation. 
 
Title Niftyrec 
Description Provides routines for tomographic reconstruction of PET/SPECT data. 
Type Of Technology Software 
Year Produced 2012 
Open Source License? Yes  
Impact The software has been downloaded a large number of times. 
URL http://cmictig.cs.ucl.ac.uk/research/software/24-niftyrec
 
Title PV correction software 
Description Software for performing partial volume correction of emission tomography data using the RBV / iterative Yang correction methods. URL currently being tested internally. 
Type Of Technology Software 
Year Produced 2014 
Open Source License? Yes  
Impact In addition to this software an open-source library of test cases is being prepared for validation purposes (as part of an EC COST project). 
 
Description UCH open days 
Form Of Engagement Activity Participation in an open day or visit at my research institution
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Public/other audiences
Results and Impact UCH organise an annual open day at which INM promotes activities including research projects.
Year(s) Of Engagement Activity 2014,2015
 
Description UCL open days 
Form Of Engagement Activity Participation in an open day or visit at my research institution
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Postgraduate students
Results and Impact UCL based open days organised for postgraduate students and for Centre for Medical Image Computing.
Year(s) Of Engagement Activity 2013,2014,2015