Inter-Method Discrepancies in Brain Volume Estimation May Drive Inconsistent Findings in Autism
Gajendra J Katuwal1, Stefi A Baum2, Nathan D Cahill3
1Autism and Developmental Medicine Institute, Geisinger Health SystemDanville, PA, USA; Chester F. Carlson Center for Imaging Science, Rochester Institute of TechnologyRochester, NY, USA.
Frontiers in Neuroscience
|October 18, 2016
Summary
Inconsistent autism brain volume findings stem from differing analysis methods. SPM, FSL, and FreeSurfer yield varied results, highlighting the need for cross-validation in neuroimaging research.
Area of Science:
- Neuroimaging
- Developmental Neuroscience
- Biomedical Engineering
Background:
- Previous studies on autism spectrum disorder (ASD) using structural magnetic resonance imaging (sMRI) report conflicting neuroanatomical abnormalities.
- These inconsistencies may arise from variations in automatic preprocessing methods used for data analysis.
Purpose of the Study:
- To investigate inter-method differences in brain volume estimation as a cause for inconsistent neuroimaging findings in ASD.
- To compare the performance of three popular preprocessing methods (SPM, FSL, FreeSurfer) for estimating gray matter, white matter, cerebrospinal fluid, and total intracranial volume.
Main Methods:
- T1-weighted sMRIs from 417 ASD subjects and 459 typically developing controls (TDC) from the ABIDE dataset were processed using SPM, FSL, and FreeSurfer.
- Brain volumes were estimated, inter-method differences analyzed, and group differences (ASD vs. TDC) assessed for each method.
- Manual segmentation was performed on a subset of subjects to validate automated total intracranial volume (TIV) estimates.
Main Results:
- Significant inter-method differences were found in brain volume estimations, except for TIV between SPM and FreeSurfer.
- ASD vs. TDC group differences were method-dependent; SPM indicated larger TIV, gray matter, and CSF in ASD, while FSL and FreeSurfer showed no significant differences or opposite trends.
- Manual validation showed SPM estimates were closest to manual segmentation, followed by FreeSurfer, with FSL estimates being significantly lower.
Conclusions:
- Inter-method discrepancies in brain volume estimation significantly impact the detection of neuroanatomical differences between ASD and TDC groups.
- Differential biases exist across methods, with some biases exceeding the observed group differences.
- Cross-validation across methods and the development of more robust neuroimaging analysis techniques are crucial for reliable findings in ASD research.
