Classifying Unconscious, Psychedelic, and Neuropsychiatric Brain States with Functional Connectivity, Graph Theory,...

Area of Science:

• Neuroscience
• Machine Learning
• Computational Psychiatry

Background:

• Accurate brain state classification is crucial for neuroscience research and clinical applications.
• Traditional methods rely on functional connectivity and graph theory, but may miss global organizational patterns.
• Cortical gradients offer a complementary view of brain organization.

Purpose of the Study:

• To develop and validate a machine learning model that integrates functional connectivity, graph-theoretic metrics, and cortical gradient features.
• To discriminate between baseline and atypical brain states across diverse conditions using a unified approach.
• To assess the generalizability and transferability of the integrated model across different brain states.

Main Methods:

• Extracted features from brain states including unconsciousness, psychedelic states, and neuropsychiatric disorders.
• Employed a support vector machine with nested cross-validation for model construction.
• Utilized a soft voting ensemble model to combine predictions from base models.

Main Results:

• The ensemble model achieved an average balanced accuracy of 79%, outperforming individual models (70-76%).
• Performance varied across conditions, indicating the need for tailored approaches.
• Feature importance analysis revealed significant differences in neural mechanisms across meta-conditions.

Conclusions:

• Integrating multiple feature sets enhances the robustness of brain state classification.
• The developed model shows promise for broader brain state discrimination, though further validation is needed.
• Tailored approaches are necessary for accurate classification of specific brain states due to varying neural underpinnings.