Optimal control-based bayesian detection of clinical and behavioral state transitions

Area of Science:

• Neuroscience
• Computational Medicine
• Biomedical Engineering

Background:

• Accurate detection of hidden clinical/behavioral states from sequential data is crucial for applications like neural prosthetics and drug delivery.
• Early seizure detection from electroencephalography (EEG) can prevent impairment and overtreatment.
• Existing detection algorithms have limitations in accuracy and adaptability.

Purpose of the Study:

• To develop a Bayesian paradigm for state transition detection combining optimal control and Markov processes.
• To create a detection policy minimizing false positives and transition time lag.
• To introduce an adaptive, time-varying threshold policy for improved detection.

Main Methods:

• Defined a hidden Markov model for state evolution.
• Developed a detection policy optimizing a loss function for false positives and accuracy.
• Applied the paradigm to Parkinson's disease movement onset detection (subthalamic recordings) and rodent seizure detection (intracranial EEG).

Main Results:

• The Bayesian paradigm significantly outperformed chance in both applications.
• The method demonstrated superior performance compared to widely used detection algorithms.
• The adaptive policy effectively adjusted to new measurements and model parameters.

Conclusions:

• The developed Bayesian paradigm offers a robust and adaptive method for detecting hidden clinical and behavioral states.
• This approach has significant implications for advancing neural prosthetics, brain-computer interfaces, and personalized medicine.
• The paradigm's effectiveness in real-world neuroscience and medical applications was validated.