OI and fMRI signal separation using both temporal and spatial autocorrelations

Area of Science:

• Neuroimaging
• Signal Processing
• Biomedical Engineering

Background:

• Blind source separation (BSS) is crucial for analyzing brain imaging data.
• Canonical correlation analysis (CCA) is a common BSS technique for optical imaging (OI) and functional magnetic resonance imaging (fMRI).
• Standard CCA faces limitations in separating temporal signal sources due to dimensionality reduction and ignoring spatial autocorrelation.

Purpose of the Study:

• To introduce a novel method, low-dimensional canonical correlation analysis (LD-CCA), for improved brain imaging signal separation.
• To address the limitations of traditional CCA in handling temporal signal sources.
• To leverage both spatial and temporal autocorrelations for more effective signal recovery.

Main Methods:

• Developed LD-CCA by incorporating a "generalized timecourse" technique.
• Reorganized data artificially to define combined spatial and temporal autocorrelations.
• Maximized combined temporal and spatial autocorrelations to identify genuine signal sources.
• LD-CCA inherently operates in a low-dimensional space, avoiding the need for explicit dimension reduction.

Main Results:

• LD-CCA demonstrated superior effectiveness in recovering signal sources compared to temporal CCA and temporal independent component analysis (tICA) in simulated data.
• Validation using real intrinsic OI and fMRI data confirmed the efficacy of LD-CCA.
• The generalized timecourse approach in LD-CCA preserves potentially useful information by eliminating the need for aggressive dimension reduction.

Conclusions:

• LD-CCA offers a more robust and effective approach for separating brain imaging signals, particularly temporal sources.
• The method's ability to integrate spatial and temporal information enhances the accuracy of signal source recovery.
• LD-CCA represents a significant advancement in BSS techniques for neuroimaging analysis.