The effects of image reconstruction algorithms on topographic characteristics, diagnostic performance and clinical correlation of metabolic brain networks in Parkinson's disease.

Purpose: The purpose of this study was to evaluate the effects of different image reconstruction algorithms on topographic characteristics and diagnostic performance of the Parkinson's disease related pattern (PDRP).
Methods: FDG-PET brain scans of 20 Parkinson's disease (PD) patients and 20 normal controls (NC) were reconstructed with six different algorithms in order to derive six versions of PDRP. Additional scans of 20 PD, 25 atypical parkinsonism (AP) patients and 20 NC subjects were used for validation. PDRP versions were compared by assessing differences in topographies, individual subject scores and correlations with patient's clinical ratings. Discrimination of PD from NC and AP subjects was evaluated across cohorts.
Results: The region weights of the six PDRPs highly correlated (R ≥ 0.991; p < 0.0001). All PDRPs' expressions were significantly elevated in PD relative to NC and AP subjects (p < 0.0001) and correlated with clinical ratings (R ≥ 0.47; p < 0.05). Subject scores of the six PDRPs highly correlated within each of individual healthy and parkinsonian groups (R ≥ 0.972, p < 0.0001) and were consistent across the algorithms when using the same reconstruction methods in PDRP derivation and validation. However, when derivation and validation reconstruction algorithms differed, subject scores were notably lower compared to the reference PDRP, in all subject groups.
Conclusion: PDRP proves to be highly reproducible across FDG-PET image reconstruction algorithms in topography, ability to differentiate PD from NC and AP subjects and clinical correlation. When calculating PDRP scores in scans that have different reconstruction algorithms and imaging systems from those used for PDRP derivation, a calibration with NC subjects is advisable.
(Copyright © 2018. Published by Elsevier Ltd.)