A comparison of the options for brain partial volume correction using PET/MRI

Partial volume effects affect the quantitative accuracy of PET images. Many approaches to partial volume correction (PVC) have been proposed, however most rely on additional, patient-specific anatomical information from structural imaging modalities such as MRI. In order to utilize anatomical data, image registration is required. With the recent advent of simultaneous PET/MRI scanners comes the ability to acquire accurately registered data. In this study, applied eight different PVC techniques to Monte Carlo simulated data, derived from a clinical brain FDG PET/MRI study. Reconstruction-based and post-reconstruction PVC methods were evaluated. Their performance was investigated in terms of bias vs. noise, lesion contrast and when faced with registration errors. Excellent quantification, with reduced noise, can be achieved by applying PVC when accurately aligned data are available. Reconstruction-based methods produced images with low bias and reduced noise. Post-reconstruction techniques appeared to be more sensitive to registration and segmentation errors. All PVC techniques improved recovery compared to the uncorrected data.