Support Vector Machine for Analyzing Contributions of Brain Regions During Task-State fMRI

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Machine Learning

Background:

• Task-based functional Magnetic Resonance Imaging (fMRI) typically identifies active brain regions using generalized linear models.
• Machine learning (ML) methods are emerging as powerful, data-driven approaches for analyzing complex fMRI datasets.

Purpose of the Study:

• To compare ML-based classification of cognitive tasks with traditional statistical analysis of fMRI data.
• To investigate the contribution of both activated and non-activated brain regions to task differentiation using ML.

Main Methods:

• Utilized language task fMRI data (math vs. story) from the Human Connectome Project (HCP).
• Employed a linear Support Vector Machine (SVM) classifier to differentiate between math and story tasks.
• Compared SVM classification results with brain regions identified as active by Statistical Parametric Mapping (SPM).

Main Results:

• The SVM classifier successfully distinguished between math and story tasks.
• Out of 25 regions used by the SVM, 13 were statistically identified as active, while 12 were not.
• Notably, non-activated regions, specifically the right Paracentral Lobule and right Rolandic Operculum, were major contributors to the classification accuracy.

Conclusions:

• Machine learning approaches can identify distinct patterns of brain activity beyond traditionally defined 'active' regions.
• The findings suggest that non-activated brain regions play a crucial role in cognitive task differentiation.
• This study provides a novel perspective on the physiological mechanisms underlying different cognitive tasks by integrating ML with fMRI analysis.