Related Experiment Videos
Commentary and opinion: I. Principal component analysis, variance partitioning, and "functional connectivity"
S C Strother1, I Kanno, D A Rottenberg
1PET Imaging Service, VA Medical Center, Minneapolis, Minnesota 55417, USA.
Summary
Functional positron emission tomography (PET) studies show intrasubject variance is small compared to intersubject variance. Careful analysis is needed to understand functional connectivity in PET data.
Area of Science:
- Neuroimaging
- Biostatistics
- Medical Data Analysis
Background:
- Functional positron emission tomography (PET) data analysis requires robust methods for variance partitioning and optimal model selection.
- Heterogeneous spatial covariance structures are a key consideration in PET data analysis.
Purpose of the Study:
- To review the importance of variance partitioning and optimal model selection in functional PET data analysis.
- To investigate the proportion of intrasubject signal variance in functional PET studies.
- To evaluate the impact of different analytical approaches on variance components.
Main Methods:
- Review of variance partitioning and optimal model selection techniques.
- Application of principal component analysis (PCA) to an [15O]water PET dataset.
- Comparison of analysis of covariance with scaled subprofile model processing before PCA.
Main Results:
- The intrasubject signal component of interest in baseline activation studies is a very small fraction of the intersubject variance in processed PET data.
- Analysis of covariance subtly but significantly alters this small intrasubject variance component compared to scaled subprofile model processing prior to PCA.
Conclusions:
- The interpretation of functional connectivity in PET imaging should be broad until inter- and intrasubject variability are better understood in both healthy and diseased states.
- Further research is needed to elucidate the roles of different variability sources in PET datasets.