Restricted canonical correlation analysis in functional MRI-validation and a novel thresholding technique.

Area of Science:

• Neuroimaging
• Functional Magnetic Resonance Imaging (fMRI)
• Biostatistics

Background:

• Functional Magnetic Resonance Imaging (fMRI) is a key neuroimaging technique.
• Analysis of fMRI data is crucial for understanding brain activity.
• Existing methods like the General Linear Model (GLM) have limitations.

Purpose of the Study:

• To validate the performance of restricted canonical correlation analysis (rCCA) for fMRI data.
• To enhance the usability of rCCA by introducing a novel significance estimation technique.
• To compare the efficacy of rCCA against the GLM in fMRI analysis.

Main Methods:

• fMRI data from a language task and resting-state were collected from eight volunteers.
• Data were analyzed using both rCCA and GLM.
• A modified Receiver Operating Characteristic (ROC) method evaluated analysis performance, with the area under the ROC curve as the primary metric.

Main Results:

• The rCCA method demonstrated significantly higher performance (area under the ROC curve) than the GLM.
• The novel significance estimation technique for rCCA maps showed good agreement with selected thresholds.
• rCCA effectively identified activated voxels in fMRI datasets.

Conclusions:

• The rCCA method is a highly effective tool for fMRI data analysis.
• The developed significance estimation method improves the practical usability of rCCA for fMRI.
• rCCA offers a more performant alternative to GLM for analyzing fMRI data.