XClose

Institute of Nuclear Medicine

Home
Menu

Image reconstruction in emission tomography

Our group focuses on the reconstruction of images from single photon emission tomography (SPECT) and positron emission tomography (PET) raw data using iterative algorithms such as maximum likelihood expectation maximisation (MLEM) and Maximum A Posteriori (MAP) type algorithms.

Our current research is targeted towards accurate quantification, including aspects of system modelling, patient motion, and numerical convergence. Our research also includes reconstruction of multi-modality data such as PET/CT and PET/MRI.

We work closely with researchers from the UCL Centre for Medical Imaging Computing and in particular the groups of Profs. Sébastien Ourselin and Simon Arridge.

Since April 2015, we lead the EPSRC-funded Collaborative Computational Platform on Synergistic PET-MR Reconstruction.

Current research projects related to image reconstruction in our group:

Acceleration of MAP Image Reconstruction Algorithms

Traditional Maximum Likelihood (ML)-type image reconstruction algorithms suffer from problems with either reduced quantification or increased noise due to the ill-posedness of the ML inverse problem. Maximum a Posteriori (MAP) regularises the solution, but can need a higher number of iterations to obtain a clinically useful image. To circumvent this, we have proposed a fast algorithm, LBFGS-B-PC [1]. The algorithm combines LBFGS-B with diagonal preconditioning. Preliminary results show good performance which is relatively independent of the choice of prior etc, which might make this algorithm a good candidate for use in a clinical context.

Joint activity/attenuation reconstruction in SPECT/PET

Image reconstruction in emission tomography needs an estimate of the attenuation of the photons. This is nowadays often obtained from CT, but there are situations where this is not possible or desirable (e.g. due to dose considerations or standalone-SPECT or PET-MR).
In SPECT, our team has investigated the utilisation of dual energy window data (i.e. photopeak and scatter windows) to jointly reconstruct the activity distribution image and the attenuation map, thus removing the need for CT measurement [2]. We are currently working on improving this approach with the utilisation maximum-likelihood methods and GPU-accelerated implementation using the CUDA library (NVidia), as well as priors. Another work in progress is the joint activity/attenuation reconstruction in PET using dynamic data.

Joint image reconstruction/motion estimation in respiratory gated PET/CT

Gating can be used in PET/CT to reduce the effect of respiratory motion. Obtaining gated CT images however considerably increases the dose delivered to the patient. To avoid the need for gated CT, we developed an algorithm that jointly estimates motion and activity distribution directly from the raw data in PET/CT using a single CT image [2]. The CT image is warped alongside the activity for attenuation correction. Figure [1] shows the motion corrected PET image using this algorithm. More recently, this approach was extended to TOF-PET/CT [4].

Image 1 -non-motion corrected reconstructed PET image

Image 2 - motion-corrected reconstructed PET image using the algorithm proposed in [3]

Joint-Reconstruction in PET-MR

We have developed methods for joint reconstruction of multi-modality data [5] which will allow us to investigate the potential advantages of using common structural information in for instance PET/MR applications that need fast imaging. We have recently shown that the joint-priors can be used for more "traditional" image reconstruction of PET data with anatomical information from MR [6].

In addition to the above projects, we are also working on image reconstruction for the INSERT SPECT/MR system [LINK TO our INSERT page]. Other PET/MR related projects including head motion correction and GPU-enabled image reconstruction with our UCL collaborators are described on the TiG page.

To be able to develop and evaluate new image reconstruction algorithms, a considerable amount of time needs to be invested in the software implementation. We therefore lead various Open Source Software projects.

References


[1] Tsai Y, Bousse A, Ehrhardt MJ, Hutton BF, Arridge SR and Thielemans K, "Performance Evaluation of MAP Algorithms with Different Penalties, Object Geometries and Noise Levels", IEEE MIC Conf. Proc., 2015.

[2] Cade CS, Arridge S, Evans MJ, Hutton BF, "Use of measured scatter data for the attenuation correction of single photon emission tomography without transmission scanning," Med. Phys., vol. 40, no. 8, 2013. http://dx.doi.org/10.1118/1.4812686

[3]  Bousse A, Bertolli O, Atkinson D, Arridge S, Ourselin S, Hutton BF, and Thielemans K, "Maximum-likelihood joint image reconstruction/motion estimation in attenuation-corrected respiratory gated PET/CT using a single attenuation map," IEEE Trans. Med. Imag., vol. 35, no. 1, pp. 217-228, 2016. http://dx.doi.org/10.1109/tmi.2015.2464156

[4]  Bousse A, Bertolli O, Atkinson D, Arridge S, Ourselin S, Hutton BF, and Thielemans K,
"Maximum-likelihood joint image reconstruction and motion estimation with misaligned
attenuation in TOF-PET/CT," Phys. Med. Biol., vol. 61, pp. L11-19, 2016. http://dx.doi.org/10.1088/0031-9155/61/3/L11

[5]  Ehrhardt MJ, Thielemans L, Pizarro L, Atkinson D, Ourselin S, Hutton BF, and Arridge SR, "Joint reconstruction of PET-MRI by exploiting structural similarity," Inverse Problems, vol. 31, 2015. http://dx.doi.org/10.1088/0266-5611/31/1/015001

[6]  Ehrhardt MJ, Markiewicz P, Liljeroth M, Barnes A, Kolehmainen V, Duncan J, Pizarro L, Atkinson D, Hutton B, Ourselin S, Thielemans K, Arridge SR, "PET Reconstruction with an Anatomical MRI Prior using Parallel Level Sets.", IEEE Trans. Med. Imag., 2016 (in press) http://dx.doi.org/10.1109/tmi.2016.2549601