Fusing functional signals by sparse canonical correlation analysis improves network reproducibility.

Area of Science:

• Neuroimaging
• Functional MRI
• Biomedical Engineering

Background:

• Arterial spin labeling (ASL) MRI provides insights into brain function.
• Standard methods for analyzing functional brain networks often average signals within regions of interest (ROIs).
• Existing approaches may not fully exploit the rich information within ASL data or fuse multiple signal types effectively.

Purpose of the Study:

• To develop a novel multivariate strategy for computing brain functional network structure using ASL MRI.
• To enhance the reproducibility and interpretability of functional network analysis.
• To investigate brain connectivity differences in traumatic brain injury (TBI) using the proposed method.

Main Methods:

• Employed sparse canonical correlation analysis (SCCA), an interpretable dimensionality reduction technique.
• Fused and correlated multiple functional signals, including ASL-BOLD and ASL-based cerebral blood flow (CBF) time series.
• Compared the SCCA approach with traditional ROI region-averaging methods in test-retest and TBI studies.

Main Results:

• SCCA extracts biologically plausible and stable functional network structures from ASL data.
• The SCCA approach demonstrated significantly improved reproducibility compared to region-averaging.
• Joint BOLD-CBF network analysis using SCCA revealed connectivity differences in TBI not observed with region averaging.

Conclusions:

• The proposed SCCA strategy offers a more reproducible and sensitive method for analyzing brain functional networks from ASL MRI.
• This multivariate approach effectively utilizes the full information within ROIs and fuses multiple ASL-derived signals.
• The findings suggest SCCA is a valuable tool for neuroimaging research, especially in clinical applications like TBI assessment.