Critical biomarkers for responsive deep brain stimulation and responsive focal cortex stimulation in epilepsy field

To derive critical signal features from intracranial electroencephalograms of epileptic patients in order to design instructions for feedback-type electrical stimulation systems. The Detrended Fluctuation Analysis (DFA) exponent is chosen as the classification exponent, and the disparities between indicators representing distinct seizure states and the classification efficacy of rudimentary machine learning models are computed. The DFA exponent exhibited a statistically significant variation among the pre-ictal, ictal period, and post-ictal stages. The Linear Discriminant Analysis model demonstrates the highest accuracy among the three basic machine learning models, whereas the Naive Bayesian model necessitates the least amount of computational and storage space. The set of DFA exponents is employed as an intermediary variable in the machine learning process. The resultant model possesses the capability to function as a feedback trigger program for electrical stimulation systems of the feedback variety, specifically within the domain of neural modulation in epilepsy. We propose a complete method from raw data to intermediate variables and then to machine learning, which is beneficial for automated triggering of feedback type electric stimulators. Through this method, we can greatly reduce the computational and storage requirements of the feedback electrical stimulation system, while ensuring minimal decrease in discrimination accuracy. [Display omitted]