Modelling subject variability in the spatial and temporal characteristics of functional modes

Area of Science:

• Neuroscience
• Cognitive Neuroscience
• Data Science

Background:

• Subject variability in functional MRI (fMRI) data is significant and can confound cross-subject analyses.
• Accurate representations of functional brain data are crucial for reliable downstream analysis.

Purpose of the Study:

• To extend the Probabilistic Functional Modes (PFMs) framework to capture cross-subject variability in spatial maps, functional coupling, and mode amplitudes.
• To develop a new implementation for analyzing large-scale fMRI datasets.

Main Methods:

• Extended the Probabilistic Functional Modes (PFMs) framework to model cross-subject variability.
• Developed a new inference implementation for large-scale fMRI data analysis (PROFUMO).
• Validated the approach using simulated data, Human Connectome Project resting-state data, and task-state fMRI data.

Main Results:

• PFMs effectively capture spatio-temporal structure variations in resting-state fMRI activity across subjects within a single model.
• Compared to independent component analysis with dual regression (ICA-DR), PFMs attribute more variability to spatial maps.
• Functional coupling between modes, after accounting for spatial variability, reflects cognitive state.

Conclusions:

• PFMs provide a richer description of cross-subject variability in fMRI data compared to traditional methods.
• The findings have significant implications for interpreting cross-sectional functional connectivity studies.
• The PROFUMO package facilitates the analysis of large-scale fMRI datasets, accounting for individual differences.