Evaluation of MR-driven versus MR-free PET/CT spatial normalization approaches for quantification of regional uptake in Dementia

Abstract

Background:

Spatial normalization is a critical preprocessing step in PET imaging. While MR-driven normalization represents the gold standard, MR-free approaches offer an alternative that may enhance accessibility.

Purpose:

This study aims to evaluate the performance of MR-free normalization compared to MR-driven normalization across three PET modalities (18F-FDG, early-phase amyloid, and late amyloid PET), examinating regional agreement.

Methods:

56 patients with dementia performing MR and the three PET modalities were retrospectively enrolled in the study. Three different spatial normalization methods were performed: MR-driven and two MR-free (${}^{15}$O-H${}_2$O template and site-specific PET template). The agreement among different methods was pair-wise compared using intraclass correlation coefficients (ICCs).

Results:

MR-free normalization showed strong agreement with MR-driven methods for FDG and early amyloid PET, with ICCs $>$ 0.99 in cortical regions. The spatial normalization with early-phase amyloid site-specific template had a median ICC of 0.927, confirming its agreement with MR-driven approach. The performance was slightly lower with late amyloid-derived templates (ICC = 0.822) compared to the ${}^{15}$O-H${}_2$O template. The worst reproducibility was observed in cerebellar and thalamic sub-regions, though aggregating sub-ROIs in parent regions brought to strong ICCs ($>$ 0.920).

Conclusion:

MR-free normalization is a viable alternative to MR-driven approaches, particularly for FDG and early amyloid PET. In addition, the results confirm the similarity of early amyloid and FDG tracers distributions.