Classification and Extraction of Resting State Networks Using Healthy and Epilepsy fMRI Data.

Area of Science:

• Neuroimaging
• Machine Learning
• Clinical Neuroscience

Background:

• Functional magnetic resonance imaging (fMRI) advances brain activity understanding in healthy and patient populations.
• Resting-state fMRI (rs-fMRI) examines neural activity without task confounds, providing valuable functional mapping.
• Current clinical workflows can benefit from automated organization and labeling of rs-fMRI data.

Purpose of the Study:

• To develop an automatic resting state network (RSN) labeling method for clinical rs-fMRI mapping.
• To classify spatial maps into specific functional networks (auditory, visual, default-mode, sensorimotor, executive control).
• To validate the method in epilepsy patients and healthy subjects.

Main Methods:

• Applied independent component analysis (ICA) to rs-fMRI data from 23 epilepsy patients and 30 healthy subjects.
• Utilized machine learning algorithms (naïve Bayes and perceptron) for network classification.
• Extracted and classified spatial maps into predefined RSNs.

Main Results:

• ICA identified distinct and consistent functional network components across both patient and healthy groups.
• Achieved high classification accuracy: 88% for epilepsy patients and 90% for healthy subjects.
• Successfully labeled auditory, visual, default-mode, sensorimotor, and executive control RSNs.

Conclusions:

• The developed automatic RSN labeling method is effective for clinical rs-fMRI analysis.
• This automated approach provides valuable, complementary functional information for clinicians.
• The tool enhances the organization and interpretation of RSN spatial maps in clinical settings.