Unsupervised machine-learning classification of electrophysiologically active electrodes during human cognitive task...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Identifying active electrodes for task-relevant neurophysiological activity is crucial for brain function research and clinical applications.
• Existing methods may lack objectivity and efficiency in classifying electrophysiological signals.

Purpose of the Study:

• To develop an unsupervised, fully automated approach for classifying active electrodes based on event-related intracranial EEG (iEEG) responses.
• To introduce interpretable metrics for quantifying spectral characteristics of iEEG signals.

Main Methods:

• Developed an unsupervised machine learning framework utilizing novel interpretable metrics (power-in-band, synchrony) to analyze normalized iEEG signals.
• Applied unsupervised clustering to classify active electrodes from 11,869 electrodes across 115 patients performing a verbal memory task.

Main Results:

• The automated method achieved high performance: 97% sensitivity and 92.9% specificity with the most efficient metric.
• Unsupervised clustering identified distinct sets of active electrodes across subjects.
• Active electrodes were significantly localized to brain regions supporting verbal memory processing.

Conclusions:

• The proposed machine learning framework offers an objective and efficient method for classifying and interpreting electrophysiological signals.
• This approach facilitates the study of brain activities related to memory and cognition.
• The findings support the use of automated analysis for identifying task-relevant neural activity.