A Blind Module Identification Approach for Predicting Effective Connectivity Within Brain Dynamical Subnetworks.

Area of Science:

• Neuroscience
• Systems Biology
• Computational Neuroscience

Background:

• Effective brain connectivity analysis is crucial for understanding dynamic neural processes like seizures.
• Studying connectivity in real-time is challenging due to limited measurement capabilities and unobservable external inputs.
• Model fitting for such systems is a complex problem of blind module identification and model inversion.

Purpose of the Study:

• To propose a novel estimation framework for identifying nonlinear dynamic subnetworks with unknown local inputs.
• To address the challenges of model fitting in distributed dynamic phenomena using limited real-time recordings.
• To improve the estimation of subnetwork parameters and unknown inputs.

Main Methods:

• Utilized Cubature Kalman filtering for initial predictions.
• Employed residuals of local output predictions to refine local input estimates.
• Tested the algorithm on simulated EEG data and clinical EEG data from electroconvulsive therapy (ECT)-induced seizures.

Main Results:

• Significantly improved estimation accuracy for inputs and connections from noisy simulated EEG.
• Predicted increased subnetwork inputs during pre-stimulus anesthesia in clinical data.
• Identified increased frontocentral connectivity during generalized seizures, supporting hypotheses on ECT's frontal focality.

Conclusions:

• The proposed framework effectively identifies nonlinear dynamic subnetworks and their inputs.
• The method enhances the accuracy of brain effective connectivity estimation from limited EEG data.
• The framework is adaptable for various input configurations and applicable to micro and macroscale brain subnetworks.