<< Click to Display Table of Contents >> Navigation:  Partial-Volume Correction (PVC) > PVC of Brain Images based on MRI Segmentation > Validation and Recommendations

Validation

The performance of the PVC (Brain MR based) method has been studied by Olivier Barret, PhD, Institute for Neurodegenerative Disorders, New Haven, CT, USA, and the results presented at the Symposium on Functional Neuroreceptor Mapping of the Living Brain 2010 in Glasgow. The abstract is reproduced below.

Performance Evaluation of PMOD integrated Partial Volume Correction Method for Brain PET data

Partial volume effects in emission tomography (PET) come from the limited resolution of the cameras and are due to two factors: the underlying heterogeneity of the tissues (grey and white matter in brain tissue) and the cross contamination of adjacent regions by the point spread function (PSF) of the tomographs. The result of these effects is an erroneous estimation of the true local tissue radioactivity concentration and in studies where quantitative estimates are required, it is often essential to perform partial volume corrections (PVC), in particular to take into account pathological or structural changes for instance as these can have a significant impact. Several correction methods, often requiring an additional structural imaging, have been proposed to attempt recovery of the true signal, and among these, one of the correction the most commonly used is an MR-based solution.

We report here of a fully integrated implementation of such an MR-based correction in the PMOD software, where the MR is automatically segmented into grey matter and white matter maps and subsequently registered with the PET image. Activity concentration in white matter is automatically determined by linear regression and the functional image is corrected by the maps smoothed by the PSF of the system, approximated by a 3-dimensional Gaussian distribution with a full-width half-maximum (FWHM) adjustable by the user. At the end of the process, the user is presented with a grey matter partial volume corrected functional image. The correction has been validated using simulated subjects from a Monte Carlo database built using the PET-SORTEO simulator which implements a realistic model of the ECAT EXACT HR+ tomograph. Ten subjects simulated with a FDG functional model were used to estimate the recovery coefficient as a function of the set FWHM and to evaluate the robustness of the correction with regard to the white matter activity concentration estimate.

Results for the simulated subjects show a very good recovery of the cortical regions with corrected values within 3% of the simulated activity concentration for an optimal FWHM around 6-7mm, in agreement with an effective resolution of the space-variant published resolution of the simulated scanner. The recovery coefficient also showed good stability against the user defined FWHM, with variation of 4-5% for a FWHM set between 5mm and 9mm. The method was shown to be robust against the white matter activity concentration estimates, with variation of only 2-3% of the final grey matter corrected values for large errors of 10% in the WM estimates.

An MR-based PVC method fully automated and integrated with all the steps necessary to perform the correction has been implemented in the PMOD software and validated against a simulated data set. The pertinence of the correction is being evaluated for different tracers (FDOPA from the PET-SORTEO database) and also for subject data from the ADNI database, in particular for the beta amyloid PIB tracer where this correction might prove to be critical because of the high white matter uptake.

Recommendations

The PVC (Brain MR based) method is a fully automatic procedure which has been tested in detail. However, the user should be aware that in practical situations there are several sources of potential errors which may affect the PVC outcome, particularly

▪distortions of the grey and white matter segments, and

▪inaccurate matching of the PET to the MRI.

Therefore the user is advised to inspect the additional PVC results. He should fuse the segment images with the anatomical MRI, and the matched PET image with the MRI to ensure that the anatomical information, which is used for the PVC, is correct and aligned.

Furthermore, VOI statistics should be calculated both with the original as well as with the PVC corrected images in order to detect consistencies in the data.