sMRI Analysis: Reproducibility Considerations
OverviewTeaching: 20 min
Exercises: 10 minQuestions
How sensitive are the findings to your MR pipeline parameters?Objectives
Understand impact of software and atlas choices
You Are Here!
MR image processing pipeline selection choices
- Compute environment
- OS / system math libararies
- Programming libraries Python / R versions
- Software (algorithms and their versions)
- Image clean-up
- Image normalization
- Image quantification
- Quality control
- Manual protocol specifics
- Automatic outlier criteria
- Biological priors
- Atlases/parcellations: ROI definitions
Example software analysis
Task: compare cortical thickness between FreeSurfer and CIVET
Note: See this for details on brain plots
Correlation between thickness measurements:
Software: FreeSurfer (v5.1, v5.3, v6.0), CIVET (2.1) and ANTs
Possible choices for FreeSurfer parcellations
Different parcellation different results?
Inference pertaining to neuroantomical differences and/or prediction models based on individual neuroanatomical feature sets can be sensitive to parcellation choice.
Note: Image adopted from Madan 2021
It is crucial to assess biological vs methodological variation in your findings to avoid reproducibility crisis.