Regional quantitation of functional Positron Emission Tomography (PET) brain images suffers from the subjective determination of the regions of interest. An accurate and reproducible method of mapping the functional information to specific anatomical structures is by correlating PET and Magnetic Resonance (MR) images. Correlation of PET and MR images of the human brain has been implemented by matching of anatomical or external markers or by stereotactic methods, sometimes employing normal anatomy atlases. All of these methods have well documented drawbacks. The 'surface fitting' algorithm developed by Pellizzari and Chen elegantly overcomes those drawbacks, but its accuracy is critically dependent on the spatial distortions peculiar to the PET and MR images. This paper describes the design and application of two phantoms to the characterization and correction of data distortion (as implemented by Cutler and al. at UCLA), and the following correlation of corrected PET and MR brain images by means of the 'surface fitting' technique. MR images of the two phantoms were acquired on a Fonar B3000 resistive whole body scanner, and PET images on a CTI/Siemens 831/08 neuropet scanner. Filled holes appear in the images ad small dots whose centroids may be used to determine the spatial mispositioning in different regions of the image volume. The values for planar and axial distortion in PET images were 0.1 ± 0.3 mm and 0.1 ± 0.5 mm respectively, or substantially lower than the voxel size for PET data (3.1 mm3), indicating that geometrically corrected PET images are spatially accurate. The magnetic field gradient causes a slight barrel or pin-cushion distortion in MR image planes. In the axial dimension the effects are a slgiht tilt of individual image planes and a rather severe skewing (up to 6 mm between edge planes) of the image volume when all image planes are stacked. Interleaved or multiple acquisitions create additional alignment and spacing problems. Correction for spatial distortions in MR images was implemented through pixel by pixel repositioning and shifting of the images, based on the phantom data. Patient MR images were acquired on a FONAR B3000 whole body scanner using a T1-weighted inversion recovery sequence (TE=30 msec, TI=300 msec, TR=1276 msec) to maximize the anatomic boundary between grey matter and white matter. Patient transmission and emission PET images were acquired on a CTI/Siemens 831/08 scanner. Registration of distortion-corrected PET and MR brain image volumes was performed according to the 'surface fitting' technique. This correlation method can be easily extended to image volumes generated from other modalities, such as CT or SPECT. In any case, spatial distortions of the images must be characterized and corrected before correlation is attempted.
|Translated title of the contribution||Correlation of PET imaging and MR, corrections for distorsions, for anatomical mapping and cerebral function|
|Number of pages||6|
|Journal||Italian Current Radiology|
|Publication status||Published - 1991|
ASJC Scopus subject areas
- Radiology Nuclear Medicine and imaging