Uncovering relationships in multi-channel EEG data using principal Hessian directions and Ricci flow

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Science

Background:

• High-dimensional electroencephalography (EEG) data presents challenges for classification.
• Understanding interchannel connectivity is crucial for improving EEG signal recovery and feature selection.

Purpose of the Study:

• To develop a systematic framework for extracting meaningful interchannel network structure from EEG data.
• To enhance downstream EEG signal classification through improved feature selection.

Main Methods:

• Implemented a supervised algorithm using principal Hessian directions (pHd) for classification-relevant community structure.
• Utilized a discrete Ricci flow-based unsupervised algorithm for community detection.
• Applied both methods to high-dimensional EEG datasets for classifying imagined digits and detecting emotional valence.

Main Results:

• The combined pHd and Ricci flow methods successfully detected interchannel relationships applicable to unseen data.
• Extracted interchannel structure demonstrably improved downstream EEG signal classification performance.
• The framework provides a systematic approach to EEG data preprocessing.

Conclusions:

• The pHd and Ricci flow framework offers a novel method for analyzing EEG network structure.
• This approach enhances feature selection, leading to better EEG signal recovery and classification.
• The study expands the EEG preprocessing toolkit for real-world applications.