Enhanced Functional Connectivity for EEG Classification with a Modified Maximum Entropy Model: a Case Study of
Summary
This study introduces a penalized Maximum Entropy Model (pMEM) to improve brain functional connectivity analysis. The new method enhances classification accuracy for Alzheimer's disease detection and offers better network interpretability.
Area of Science:
- Neuroscience
- Computational Biology
- Medical Imaging
Background:
- Functional connectivity (FC) analysis is vital for understanding brain disorders.
- Traditional correlation-based methods may miss global connectivity patterns.
- Pairwise Maximum Entropy Models (pMEM) offer global insights but face instability and divergence.
Purpose of the Study:
- To develop a penalized pMEM framework integrating correlation-based FC as a prior.
- To enhance the stability and interpretability of pMEM for neural signal analysis.
- To improve the classification of brain disorders using hybrid FC modeling.
Main Methods:
- Proposed a penalized pMEM framework with correlation-based FC as a prior constraint.
- Regulated pMEM parameters to maintain adherence to traditional FC.
- Evaluated the framework on an EEG dataset of Alzheimer's disease patients and healthy controls.
Main Results:
- The penalized pMEM achieved 83.13% classification accuracy, surpassing standalone pMEM and correlation-based methods.
- The hybrid model demonstrated improved stability and preserved key linear relationships.
- Revealed more pronounced group differences in network small-worldness, enhancing interpretability.
Conclusions:
- The penalized pMEM offers a stable and interpretable hybrid approach to functional connectivity analysis.
- This method enhances diagnostic classification for brain disorders like Alzheimer's disease.
- The framework provides deeper insights into neural network properties and group differences.
