We develop and implement various image reconstruction techniques, especially to reconstruct the data obtained from our tomographic systems (YAP-(S)PET+, MicroCT), as well as the data from our systems with stationary planar detectors (PEM, SPEM, and DoPET).

Iterative Methods

Most of our iterative algorithms are based on the maximum likelihood - expectation maximisation algorithm (ML-EM) or its' accelerated version ordered subset expectation maximisation (OSEM). In this context our research is mainly focused on the following subjects:

• Efficient ways of setting up accurate system matrices. In particular, our aim is to compare various methods such as fully Monte Carlo based methods, analytical integration (multi-ray) methods and methods based on the acquistion of point sources (i.e. by measuring the response of the detectors directly).
• Evaluation of attenuation corrections for PET reconstructions based on attenuation maps obtained from CT scans.
• Estimation of normalisation factors for crystal efficiency correction.
• Optimisation of the algorithms: We concentrate on platform dependent optimisation for sequential execution and on OpenMP/MPI implementations for parallel versions on shared/distributed memory architectures.

In Figure 1 we show two results of images which were reconstruced with one of our iterative algorithms.

Figure 1: Reconstructions of PET data acquired with the YAP-(S)PET+ scanner. On the left side a transaxial slice of a reconstructed Derenzo phantom with 3.0, 2.5, 2.0, 1.5, 1.2 mm rods and on the right side a maximum intensity projection of reconstructed mouse. In both cases OSEM along with a multi-ray generated system matrix was used.

Analytical Methods

Additionally to our iterative methods, we employ Filtered Backprojection (FBP) as analytical method to reconstruct data from our tomographs. Since FBP is computationally less expensive, it remains the method of choice, especially for transmission tomography, where the reconstructed volumes are considerably larger than those used in emission tomography. Most of our research in this field is related to high resolution MicroCT imaging, more specifically to:

• Accurate and automatic measurement of misalignment parameters of cone-beam scanners, using redundancies in the scan data of a generic objects.
• Efficient reconstruction of large volumes, using symmetries of the imaging system, cache memory optimization, and parallel computing (OpenMP).
• Pre- and post-reconstruction corrections of image artifacts (beam hardening, ring artifacts).
• Generation of (coregistered) attenuation maps for attenuation correction in PET.

In Figure 2 we show the results of reconstructed data obtained from our MicroCT and the YAP-(S)PET+.

Figure 2: CT, PET and PET/CT images of a mouse, injected with 18F-fluoride (bone tracer). The CT image was reconstructed with a modified cone-beam FBP (Feldkamp algorithm) with perspective correction. The PET image was reconstructed with the ML-EM algorithm.

Literature

• S. Moehrs, M. Defrise, N. Belcari, A. Del Guerra, A. Bartoli, S. Fabbri and G. Zanetti, "Multi-Ray based system matrix generation for 3D PET reconstruction", Physics in Medicine and Biology, 53(23):6925-6945, October 2008.
  
• D. Panetta, N. Belcari, A. Del Guerra and S. Moehrs, "An optimization-based geometrical calibration method in cone-beam CT without dedicated phantoms", Physics in Medicine and Biology, 53(14):3841-3861, July 2008.
  
• A. Motta, A.Del Guerra , N.Belcari, S.Moehrs, D.Panetta, S.Righi, D.Valentini, "Fast, 3D-like Expectation Maximization Reconstruction using Planograms for the Stationary Planar Positron Emission Mammography camera", Computerized Medical Imaging and Graphics, 29(8), 587-596, 2005.
  
• A.Motta, C.Damiani, A.Del Guerra, G.Di Domenico, G.Zavattini, "Use of a fast deconvolution EM algorithm for 3D Image Reconstruction with the YAP-PET tomograph", Computerized Medical Imaging and Graphics, 26(5), 293-302, 2002.