A Full-Stack Application for Detecting Seizures and Reducing Data During Continuous Electroencephalogram Monitoring

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Technology

Background:

• Continuous electroencephalogram (EEG) monitoring in critical care settings is linked to reduced mortality but is underutilized due to the labor-intensive manual data interpretation.
• Manual review of prolonged EEG data streams presents a significant bottleneck in clinical practice, limiting its widespread adoption.

Purpose of the Study:

• To develop and validate a novel, real-time machine learning-based system for alerting and monitoring epilepsy and seizures using continuous EEG data.
• To significantly reduce the manual electroencephalogram review required for critically ill patients.

Main Methods:

• A custom data reduction algorithm utilizing a random forest was developed and deployed on a cloud-based platform.
• The system streams EEG data and interacts with caregivers via a web interface, displaying real-time algorithm results.
• The machine learning system was trained on continuous EEG recordings from 77 intensive care unit (ICU) patients and tested on an additional 20 patients.

Main Results:

• The system achieved a mean seizure sensitivity of 84% (cross-validation) and 85% (testing).
• Mean specificity was 83% (cross-validation) and 86% (testing), indicating substantial data reduction.
• A public, high-quality annotated dataset of 97 ICU continuous EEG recordings was released to facilitate further research.

Conclusions:

• The study validates a platform for machine learning-assisted continuous EEG analysis.
• This AI-driven approach represents a significant advancement in improving the utility and reducing the cost of continuous EEG monitoring.
• The developed system shows promise for more efficient and effective seizure detection in critical care environments.