A multiple kernel learning approach to perform classification of groups from complex-valued fMRI data analysis:...

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Functional magnetic resonance imaging (fMRI) data are complex-valued spatiotemporal images.
• fMRI phase images contain valuable information but are often discarded due to noise.
• Existing methods struggle with between-group inference or classification using fMRI magnitude and phase data.

Purpose of the Study:

• To develop a novel multiple kernel learning (MKL) algorithm (ν-MKL) for improved schizophrenia classification.
• To leverage both magnitude and phase fMRI data for enhanced diagnostic accuracy.
• To identify brain regions with discriminative activation patterns between healthy controls and schizophrenia patients.

Main Methods:

• Developed a new ν-MKL algorithm for tunable sparse selection of brain region patterns.
• Applied the ν-MKL algorithm to fMRI data from healthy controls and schizophrenia patients.
• Validated the method's ability to detect discriminative information in simulated fMRI data.

Main Results:

• The ν-MKL methodology improved classification accuracy by 5% when including fMRI phase data.
• Achieved classification accuracy equivalent to state-of-the-art methods on the schizophrenia dataset.
• Demonstrated superior identification of brain regions showing discriminative activation between groups.

Conclusions:

• The proposed MKL-based methodology effectively utilizes both magnitude and phase fMRI data for improved schizophrenia characterization.
• The ν-MKL algorithm accurately identifies brain regions critical for distinguishing between schizophrenia patients and controls.
• This approach offers a promising tool for neuroimaging-based psychiatric research and diagnosis.