Evaluation of partial volume correction and segmentation methods for dynamic dopaminergic PET study based on simulated PET

Objectives Partial volume effect is a considerable issue in PET striatum study. In this work we evaluate several sophisticated partial volume correction (PVC) algorithms implemented in the software PVELab. SPM5 was employed to perform T1-MR segmentation.

Methods One dynamic PET with [18F]FDOPA characteristic was generated by SORTEO PET simulator and Zubal phantom. Striatum region on grey matter segment was enhanced by: a) manual striatum mask; b) atlas formed by FIRST (FMRIB’s integrated Registration and Segmentation Tool). These segmentations combined with a static PET and with dynamic PET were passed to PVELab, where each combination underwent following PVC, respectively: Müller-Gärtner(MGCS), Rousset (Rou) and modified MG (mMG). Activity values of nucleus caudatus (NC) and putamen (PU) after PVC were extracted by mask (SPM5 segmentation with and without striatum adjustment) or by FIRST atlas.

Results With adjustment, misclassification of PU voxel on gray matter segment decreases from 50% to 25% (mask) and 16% (FIRST). DICE coefficient between mask and atlas are: 0.85±0.05, 0.86±0.05 for left/right NC and 0.85±0.05, 0.89±0.02 for left/right PU. In static study, sufficient recovery was achieved after MGCS (recovery coefficient NC=0.87±0.01, PU=0.84±0.01), mMG (0.88±0.01, 0.85±0.01) and Rou (1.08±0.02, 1.02±0.07), all with adjustment of segmentation. Dynamic study reveals very good correlations between time-activity curves (TACs) after MGCS (R2>0.96) and mMG(R2>0.96) compare to the reference. Slightly poor regression was observed for the Rou method (R2>0.92).

Conclusions Adjustment of striatum region on SPM5 segments is necessary. Atlas formed by FIRST is comparable to manually identified mask. MGCS and mMG are capable for distortion of shape and amplitude of TACs in striatum, thus they could be considered in future FDOPA modelling study