Electroencephalography-driven brain-network models for personalized interpretation and prediction of neural...

Develop an encephalography (EEG)-driven method that integrates interpretability, predictiveness, and personalization to assess the dynamics of the brain network, with a focus on pathological conditions such as pharmacoresistant epilepsy.

We propose a method to identify dominant coherent oscillations from EEG recordings. It relies on the Koopman operator theory to achieve individualized EEG prediction and electrophysiological interpretability. We extend it with concepts from adiabatic theory to address the nonstationary and noisy EEG signals.

By simultaneously capturing the local spectral and connectivity aspects of patient-specific oscillatory dynamics, we are able to clarify the underlying dynamical mechanism. We use it to construct the corresponding generative models of the brain network. We demonstrate the proposed approach on recordings of patients in status epilepticus.

The proposed EEG-driven method opens new perspectives on integrating interpretability, predictiveness, and personalization within a unified framework. It provides a quantitative approach for assessing EEG recordings, crucial for understanding and modulating pathological brain activity.

This work bridges theoretical neuroscience and clinical practice, offering a novel framework for understanding and predicting brain network dynamics. The resulting approach paves the way for data-driven insights into brain network mechanisms and the design of personalized neuromodulation therapies.