Classification of spatially unaligned fMRI scans.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biostatistics

Background:

• Functional MRI (fMRI) data analysis is complex due to high dimensionality and low signal-to-noise ratio.
• Existing methods often require spatial alignment across subjects, limiting applicability.
• Understanding brain network interactions is crucial for diagnosing neurological and psychiatric conditions.

Purpose of the Study:

• To develop a novel classification method for fMRI data.
• To overcome limitations of spatial alignment in fMRI analysis.
• To accurately discriminate between different subject groups using fMRI data.

Main Methods:

• Modeled fMRI scans as distance matrices representing temporal signal divergence.
• Utilized single-subject independent components analysis (ICA) to extract spatial networks and time courses.
• Classified subjects based on the temporal activity of independent components (ICs) without spatial normalization.

Main Results:

• Achieved up to 90% classification accuracy on diverse datasets (schizophrenia/normal, Alzheimer's/age groups).
• Demonstrated effective classification without requiring spatial alignment of fMRI scans.
• Showcased the method's ability to perform multivariate classification and identify network interactions.

Conclusions:

• The proposed method offers a robust and unique approach to fMRI data analysis and classification.
• Independent components (ICs) may represent fundamental imaging basis functions reflecting network-driven neural activity.
• This technique holds promise for advancing diagnostic capabilities in neurological and psychiatric disorders.