Adaptive tracking of human ECoG network dynamics.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Modeling multi-site electrocorticogram (ECoG) network activity is crucial for understanding brain function and developing neurotechnologies.
• Existing dynamic latent state models often overlook non-stationarity in ECoG data, which can arise from changing brain states or recording issues.
• The potential benefits of adaptive tracking for dimensionality reduction and model parsimony remain largely unexplored.

Purpose of the Study:

• To investigate whether adaptive tracking of ECoG network dynamics can lead to improved dimensionality reduction and more precise modeling.
• To compare the performance of adaptive versus non-adaptive models in predicting ECoG network dynamics.

Main Methods:

• Employed an adaptive linear state-space model for ECoG network activity.
• Utilized ECoG power feature time-series data from 10 human subjects with epilepsy, spanning tens of hours.
• Compared adaptive modeling against non-adaptive models that do not account for non-stationarity.

Main Results:

• Adaptive modeling significantly enhanced the prediction of ECoG network dynamics compared to non-adaptive approaches, particularly at lower latent state dimensions.
• Adaptive modeling enabled further dimensionality reduction without compromising prediction accuracy.
• The findings indicate that ECoG network dynamics exhibit non-stationarity over the recording periods, which can be effectively captured by adaptive modeling.

Conclusions:

• Adaptive modeling offers a more parsimonious and precise approach to analyzing ECoG network dynamics by accounting for non-stationarity.
• These results support the development of adaptive neurotechnologies for enhanced decoding and modulation of brain states in neurological disorders.