A machine learning approach for real-time cortical state estimation.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Cortical function is dynamically regulated by internal variables known as 'cortical state'.
• Current methods for estimating cortical state are often imprecise and not suitable for real-time applications.
• Accurate, real-time decoding of cortical states is crucial for understanding brain function and developing advanced neurotechnologies.

Purpose of the Study:

• To develop and implement robust, data-driven algorithms for fast, online cortical state estimation.
• To model the temporal dynamics of cortical state transitions for improved inference.
• To provide a real-time software tool for continuous decoding of cortical states.

Main Methods:

• Utilized unsupervised Gaussian mixture models to identify emergent clusters in local field potential (LFP) signals.
• Extended the approach using a temporally-informed hidden semi-Markov model (HSMM) with Gaussian observations.
• Implemented HSMM algorithms in a real-time system and evaluated performance through emulation experiments.

Main Results:

• Unsupervised clustering revealed emergent state-like structures in electrophysiological data, correlating with arousal states.
• HSMMs enabled real-time cortical state inference by modeling state-switching dynamics.
• HSMM-based state estimates demonstrated robustness against noisy, sequential electrophysiological data.

Conclusions:

• This work presents the first real-time software for continuous cortical state decoding with high temporal resolution (40 ms).
• The developed algorithms and software facilitate understanding of how cortical states modulate neural function dynamically.
• This tool provides a foundation for state-aware brain-machine interfaces in both health and disease contexts.