TY - JOUR T1 - Impact of different spatial normalization procedures on single subject testing in FDG brain PET: Comparison between Hermes BRASS and Statistical Parametric Mapping JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1724 LP - 1724 VL - 59 IS - supplement 1 AU - Hans-Georg Buchholz AU - Torsten Manthey AU - Stefanie Meckbach AU - Mathias Schreckenberger Y1 - 2018/05/01 UR - http://jnm.snmjournals.org/content/59/supplement_1/1724.abstract N2 - 1724Aim: Impact of different spatial normalization procedures on single subject testing in FDG brain PET: Comparison between Hermes BRASS and Statistical Parametric Mapping Objectives: Single subject testing based on FDG brain PET is frequently used for accurate classification of patients with neurodegenerative disorders. Typically, the patient’s individual FDG brain PET needs to be spatially normalized to a template in standard space in order to fit to a reference database consisting of FDG brain PET of healthy normal controls. The deviation of regional glucose consumption from reference is normally expressed in Z-scores calculated as difference between the reference mean and the patient’s value related to the standard deviation of the reference. In this study we used Hermes BRASS software and Statistical Parametric Mapping (SPM) to test the regional performance of reference databases using two different settings for spatial normalization. Methods: Spatial normalization was performed with two conditions in Hermes BRASS and in SPM12: affine registration only (“-A”) and affine and non-linear registration (“-W”). 37 datasets of normal healthy controls were used to build four reference databases (BRASS-A, BRASS-W, SPM-A and SPM-W). Brain VOI-template defined in standard space from Hermes BRASS FDG tool was used for calculations of the regional means and coefficient of variations (CV) of all reference databases. FDG brain PET of 30 patients with neurodegenerative disorders were selected from clinical routine and equally preprocessed as the normal controls. Single subject analyses were conducted using the VOI-template and each of the four reference databases to calculate regional Z-scores. Results: Comparing the four reference databases we found significant differences between BRASS and SPM in CV of each VOI. With warping CV in the cortical VOIs (BRASS-W: 5.5 %, SPM-W: 4.2 %) were lowest compared to BRASS-A (6.7 %) and SPM-A (5.5 %). However with warping, mismatches of subcortical VOIs (especially the Nucl. Caudate and Thalamus) in some datasets of the 37 normal controls were detected in BRASS, leading to a lower mean and a higher CV of these structures in the reference database BRASS-W. This warping effect in BRASS was very different compared to SPM. While BRASS showed a slightly better registration of striatal VOIs when warping was omitted (mean CV: BRASS-A 11.4 % vs. BRASS-W 13.2 %, p=0.0004), SPM performed better when warping was applied (SPM-W: 8.0 %, SPM-A: 11.1 %, p=0.0003). When testing 20 patients with Alzheimer’s disease (AD) warping led to highest Z-scores in cortical VOIs known to be involved in AD. Results in BRASS-W and SPM-W were comparable. However, when analyzing 10 patients with Parkinson syndrome (PS) with the focus mainly on striatal structures the results were discordant: Using warping in BRASS, subcortical mismatches occurred in some PS patients’ datasets. Therefore, results of single subject testing of these patients in BRASS and SPM depended strongly on the normalization methods (i.e. with or without warping) and the corresponding reference databases. Conclusions: Constitution of reference databases and their spatial normalization procedures played a crucial role in single subject testing. Performance in testing AD patients did not differ between BRASS and SPM and was better when warping was used. Due to the higher CV of subcortical VOIs in BRASS reference databases compared to those of SPM, testing of PS patients with BRASS and SPM showed discordant results. ER -