The accuracy of PET for measuring regional radiotracer concentrations in the human brain is limited by the finite resolution capability of the scanner and the resulting partial volume effects (PVEs). We designed a new algorithm to correct for PVEs by characterizing the geometric interaction between the PET system and the brain activity distribution.
Methods: The partial volume correction (PVC) algorithm uses high-resolution volumetric MR images correlated with the PET volume. We used a PET simulator to calculate recovery and cross-contamination factors of identified tissue components in the brain model. These geometry-dependent transfer coefficients form a matrix representing the fraction of true activity from each distinct brain region observed in any given set of regions of interest. This matrix can be inverted to correct for PVEs, independent of the tracer concentrations in each tissue component. A sphere phantom was used to validate the simulated point-spread function of the PET scanner. Accuracy and precision of the PVC method were assessed using a human basal ganglia phantom. A constant contrast experiment was performed to explore the recovery capability and statistic error propagation of PVC in various noise conditions. In addition, a dual-isotope experiment was used to evaluate the ability of the PVC algorithm to recover activity concentrations in small structures surrounded by background activity with a different radioactive half-life. This models the time-variable contrast between regions that is often seen in neuroreceptor studies.
Results: Data from the three-dimensional brain phantom demonstrated a full recovery capability of PVC with less than 10% root mean-square error in terms of absolute values, which decreased to less than 2% when results from four PET slices were averaged. Inaccuracy in the estimation of 18F tracer half-life in the presence of 11C background activity was in the range of 25%-50% before PVC and 0%-6% after PVC, for resolution varying from 6 to 14 mm FWHM. In terms of noise propagation, the degradation of the coefficient of variation after PVC was found to be easily predictable and typically on the order of 25%.
Conclusion: The PVC algorithm allows the correction for PVEs simultaneously in all identified brain regions, independent of tracer levels.