Classification of fMRI independent components using IC-fingerprints and support vector machine classifiers.

Area of Science:

• Neuroimaging
• Machine Learning
• Cognitive Neuroscience

Background:

• Functional MRI (fMRI) generates complex data requiring robust methods for analyzing independent components (ICs).
• Accurate classification of fMRI-ICs is crucial for understanding brain function and identifying task-related activity.

Purpose of the Study:

• To present a general, machine learning-based method for classifying fMRI-ICs.
• To validate the method's accuracy and reproducibility across subjects and datasets.
• To uncover neurophysiological insights, particularly regarding sensorimotor cortex involvement in visual processing.

Main Methods:

• Developed an IC-fingerprint representation for each fMRI-IC based on global parameter estimates.
• Employed a machine learning algorithm trained on expert-labeled components for automatic classification.
• Applied the method to fMRI data from a visual structure-from-motion study involving faces and control stimuli.

Main Results:

• IC-fingerprints proved effective for inspecting, characterizing, and selecting fMRI-ICs.
• Automatic classification showed high correspondence with expert visual inspection.
• Identified a reproducible task-related activation in the primary sensorimotor cortex's 'face' region, suggesting its role in visual perception.

Conclusions:

• The proposed method offers a reliable and generalizable approach for fMRI-IC classification.
• The findings suggest a role for sensorimotor cortex in processing moving visual stimuli, potentially beyond the mirror system.
• The classification algorithm demonstrates robustness across diverse fMRI acquisition parameters and experimental designs.