The impact of FreeSurfer variability on structural brain analyses in Parkinson’s disease
Poster No:
264
Submission Type:
Abstract Submission
Authors:
Andrzej Sokolowski1, Nikhil Bhagwat2, Dimitrios Kirbizakis1, Yohan Chatelain1, Mathieu Dugré1, Jean-Baptiste Poline2, Madeleine Sharp2, Tristan Glatard1
Institutions:
1Concordia University, Montreal, Quebec, 2McGill University, Montreal, Quebec
First Author:
Co-Author(s):
Introduction:
Software variability impacts the reproducibility of neuroimaging studies. Image processing software impacts quantification of brain measures and may impact clinical research. The goal of this study was to investigate the impact of variability between major FreeSurfer releases on the estimation of structural MRI-derived measures in patients with Parkinson's disease (PD). Clinical research questions were derived from previous studies on MRI-derived biomarkers of PD (Hanganu et al., 2014; Mak et al., 2015; Mitchell et al., 2021). We hypothesized that the software version would impact the magnitude of the group differences between healthy controls (HC) and PD patients in subcortical volume and cortical thickness (at baseline and longitudinally). We also hypothesized that software variability would impact the strength of the relationship between disease severity and subcortical volume as well as cortical thickness in patients with PD (at baseline and longitudinally).
Methods:
Two hundred and nine PD patients (Mage = 62.5; SD = 9.6) and 106 HC (Mage = 60.5; SD = 10.2) were selected from Parkinson's Progression Markers Initiative database. 125 PD patients (Mage = 61.1; SD = 9.3) had two scans that were used in the longitudinal analyses. T1-weighted brain images were processed using FreeSurfer. We measured the differences in the estimation of volume, surface area, and cortical thickness between three major FreeSurfer releases (i.e., 5.3, 6.0.1, and 7.3.2). Longitudinal preprocessing stream was used to calculate the change in cortical thinning and subcortical volumes between the two study visits (Reuter et al., 2012). Unified Parkinson's disease rating scale was used to measure disease severity. We compared clinical results obtained from different software versions.
Results:
The code and results are available at https://github.com/LivingPark-MRI/freesurfer-variability. We report high software variability in the estimation of all three structural measures. Estimations significantly differed between software versions in 62% to 86% regions depending on metric and FreeSurfer pair. Some regions display higher between-version than between-subject variability (Fig. 1). The variability did not differ between patients and healthy controls (ps < .05). Importantly, software variability impacted the clinical outcomes. Group differences between patients and healthy controls in subcortical volumes depended on software version; results differed between versions for the left hippocampus, right pallidum, right amygdala, and right nucleus accumbens (ps < .05) Vertex-wise analyses of group differences in cortical thickness and its correlation with disease severity showed distinct results depending on the software version (Fig. 2). More significant clusters were reported in FreeSurfer 5 than in more recent versions.
·Fig. 1
·Fig. 2
Conclusions:
We report that software variability is not only associated with the estimation of structural measures but it also impacts the interpretation of the correlations between estimates of brain structure and clinical outcomes that are commonly used in research and in the clinic. Such variability limits the utility of MRI-derived measures of brain structure as outcomes in clinical research and poses significant challenges to their eventual integration into clinical practice. We recommend users to implement the latest available release of FreeSurfer. The most recent software versions usually have improved algorithms and fixed issues discovered in previous releases. Toolbox version should not be changed throughout the same study. Developers could track differences between versions by analyzing the same dataset with current and future releases to provide information about the degree of software variability. Providing software long-term support would be beneficial. Our study provides insight into the reproducibility of neuroimaging studies in neurodegenerative disorders.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Segmentation and Parcellation 2
Keywords:
Aging
Computational Neuroscience
Degenerative Disease
Segmentation
1|2Indicates the priority used for review
Provide references using author date format
Mak, E., Su, L., Williams, G. B., Firbank, M. J., Lawson, R. A., Yarnall, A. J., ... & O’Brien, J. T. (2015). Baseline and longitudinal grey matter changes in newly diagnosed Parkinson’s disease: ICICLE-PD study. Brain, 138(10), 2974-2986.
Mitchell, T., Lehéricy, S., Chiu, S. Y., Strafella, A. P.,
Stoessl, A. J., & Vaillancourt, D. E. (2021). Emerging neuroimaging
biomarkers across disease stage in Parkinson disease: a review. JAMA neurology,
78(10), 1262-1272.
Reuter, M., Schmansky, N. J., Rosas, H. D., & Fischl, B. (2012). Within-subject template estimation for unbiased longitudinal image analysis. Neuroimage, 61(4), 1402-1418.