Layout Image

Stefano Pedemonte

Layout Image

Stefano Pedemonte

Supervisors

Dr. Sebastien Ourselin

Prof. Simon Arridge

 

Research

Positron Emission Tomography (PET) and Single Photon Emission Computed Tomography (SPECT) are functional imaging techniques that find application in diagnosis of ischemic heart disease, mapping of local brain metabolism, detection of tumors and areas of infection, drug development.
3-D images of spatial distribution of a pharmaceutical in PET and SPECT are reconstructed indirectly by observing the radiation emitted by a radioisotope that is bound to the drug.
Iterative algorithms based on statistical inference, such as Maximum Likelihood Expectation Maximization (MLEM), have proven to be effective for tomographic reconstruction.
Correlation of drug distribution with the underlying tissue morphology provides the potential to use information from an anatomical image of the same subject to improve the reconstruction and to decrease the uncertainty associated with the estimation of nuclear activity distribution. These techniques ultimately would allow for reduction of the radiopharmaceutical dose required for given resolution. Furthermore new possibilities are emerging from combined imaging techniques: PET/CT, SPECT/CT, PET/MR and SPECT/MR.

  

SPECT/MRI MAP-EM reconstruction with three different priors, from left to right: no MRI information; Total Variation prior; prior based on Joint Entropy; prior based on Conditional Joint Entropy.
The first row shows the reconstructed activity at each time step of the iterative reconstruction algorithm: MRI information visually improves activity reconstruction. The second and third rows show the derivatives of likelihood and prior functions with respect of activity.

 

Software

Nifty-Rec provides a number of libraries for image reconstruction in Emission Tomography: 2-D and 3-D rotation-based projection and back-projection with FFT-based depth dependent Collimator Detector Response (CDR); Maximum Likelihood Expectation Maximization (MLEM), Ordered Subsets Expectation Maximization (OSEM), Maximum A Posteriori Expectation Maximization (MAPEM), Gradient Ascent (GA) Maximum Llikelihood and Maximum A Posteriori (MAP) iterative reconstruction algorithms. The code is entirely written in C and is based on niftilib data structures. Critical functions have a GPU implementation. The software is fully scriptable through Matlab and Python bindings.

OSEM reconstruction of activity concentrated in the brain gray matter. The clip is generated by the GTK based GUI of Nifty-Rec. From left to right: synthetic activity phantom; projection of reconstructed activity; estimation error (voxel-wise); sagittal view of the reconstructed activity. At each time step the reconstructed activity is a better estimate of the phantom.

 

Contact Details

University College London
Centre for Medical Image Computing
The Engineering Front Building
3rd Floor
Malet Place
London
WC1E 6BT

Office: 3.09
Tel: +44 (0)20 7679 0221 (Direct Dial)
Internal: 30216
Fax: +44 (0)20 7679 0225
Email: stefano.pedemonte.09@ucl.ac.uk

Layout Image
Layout Image Layout Image

Centre for Medical Image Computing - University College London - Gower Street - London - WC1E 6BT - Telephone: +44 (0)20 7679 0221 - Copyright © 1999-2009 UCL. Disclaimer | Accessibility | Privacy | UCL Search | UCL-CS Help