Prediction error connectivity: A new method for EEG state analysis.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Traditional methods for analyzing brain connectivity often rely on frequency-domain analysis, like spectral coherence.
• Understanding brain interregional communication is crucial for diagnosing and treating neurological disorders.

Purpose of the Study:

• To introduce and validate a novel time-domain method, prediction error connectivity (PEC), for analyzing brain dynamics and detecting abrupt changes.
• To compare the efficacy of PEC against spectral coherence in identifying transitions in brain networks and predicting neurological events like seizures.

Main Methods:

• Developed a variable-order Markov model algorithm for data compression to estimate network transitions via error matrices (EMs).
• Analyzed virtual EEG signals from neural mass models and real EEG/stereo-EEG data from epilepsy patients.
• Utilized support vector machines for classifying EMs to predict seizure onset and identify sleep stages.

Main Results:

• PEC demonstrated superior performance over spectral coherence in detecting network transitions in both simulated and real EEG data.
• PEC-based models achieved high accuracy in predicting seizures (up to 98% ROC AUC) and classifying seizure-free periods (96% vs. 83%).
• PEC successfully classified and forecasted up to 88% of REM sleep phases from both deep and scalp EEG data.

Conclusions:

• Prediction error connectivity (PEC) is a novel and effective time-domain method for analyzing brain connectivity and dynamics.
• PEC offers significant advantages over frequency-domain methods for detecting transitions and predicting neurological events, showing high sensitivity and specificity.
• The PEC method holds promise for real-world applications in brain connectivity research, seizure detection, and prediction.