Development of a Machine Learning Model for Supporting Brain Death Using EEG Suppression Ratio (P11-1.012)

Area of Science:

• Neurology
• Medical Technology
• Artificial Intelligence

Background:

• Electroencephalography (EEG) is an ancillary test for brain death diagnosis.
• Diagnosing brain death can be challenging, potentially delaying organ transplantation.
• Electrocerebral inactivity (ECI) is mandatory for brain death declaration in Korea.

Purpose of the Study:

• To develop a machine learning (ML) model to aid in brain death diagnosis using EEG suppression ratio (SR).
• To improve the speed and accuracy of brain death diagnosis.
• To potentially reduce delays in organ transplantation.

Main Methods:

• Analyzed EEG data from 180 patients (81 brain death, 99 unresponsive wakefulness syndrome).
• Calculated suppression ratio (SR) using Persyst® v13 software, defining suppression as <3μv amplitude lasting ≥0.5 seconds.
• Applied supervised ML models, including Quadratic Discriminant Analysis (QDA), to predict brain death using SR, with a 70/30 training/test split.

Main Results:

• The QDA model achieved the highest performance in predicting brain death, with an Area Under the Curve (AUC) of 0.9806 ± 0.02.
• A cutoff SR value of >74.76% was identified as the maximum SR index for brain death.
• QDA demonstrated superior accuracy (0.9120 ± 0.05), precision (0.9304 ± 0.04), recall (0.9120 ± 0.05), and F1-score (0.9101 ± 0.05).

Conclusions:

• A machine learning model, specifically QDA, can effectively support brain death diagnosis using EEG SR.
• This ML-driven approach may prevent diagnostic delays, thereby expediting organ transplantation.
• The developed model shows high potential for clinical application in brain death determination.