A novel seizure detection algorithm informed by hidden Markov model event states.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• FDA-approved responsive, closed-loop intracranial devices for epilepsy treatment require rapid and accurate seizure detection.
• Current devices' high sensitivity leads to frequent false positive stimulations, decreasing battery life.
• A more robust seizure detection model is needed to improve device efficacy and longevity.

Purpose of the Study:

• To develop and validate a novel, robust seizure detection model for intracranial EEG (iEEG) data.
• To reduce false positive detections and improve the accuracy of seizure onset identification in real-time.

Main Methods:

• Utilized a Bayesian nonparametric Markov switching process to model iEEG data into distinct dynamic event states.
• Modeled each event state as a multidimensional Gaussian distribution for predictive state assignment.
• Translated the algorithm into a real-time application and validated it using iEEG data from dogs with epilepsy, comparing it to a control detector.

Main Results:

• The novel method achieved a false negative rate of 0/55 seizures missed, compared to 2/55 for the control.
• Demonstrated a significantly reduced false positive rate (0.0012 h⁻¹ vs. 0.058 h⁻¹).
• Detected all seizures an average of 12.1 ± 6.9 seconds before unequivocal epileptic onset (UEO).

Conclusions:

• The proposed seizure detection algorithm is computationally inexpensive, individualized, and suitable for real-time application in implantable antiepileptic devices.
• This method significantly reduces false positive rates compared to current industry standards.
• Offers potential for improved battery life and more effective epilepsy treatment with closed-loop devices.