Automated Cerebral Edema Detection using Electroencephalography in Post-Cardiac Arrest Patients.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Diffuse cerebral edema is a critical complication in post-cardiac arrest patients.
• Early detection of cerebral edema is crucial for timely intervention and improved outcomes.
• Current diagnostic methods like neuroimaging may not detect edema in its earliest stages.

Purpose of the Study:

• To develop and evaluate an electroencephalography (EEG)-based machine learning model for identifying diffuse cerebral edema in post-cardiac arrest patients.
• To assess the model's ability to predict cerebral edema prior to radiographic detection.
• To compare the performance of Transformer and Long Short-Term Memory (LSTM) architectures.

Main Methods:

• Retrospective, single-center cohort study of adult cardiac arrest survivors (2016-2024) with neuroimaging and EEG monitoring.
• Machine learning models (Transformer, LSTM) trained on 4- and 8-hour EEG segments to detect diffuse cerebral edema.
• Prediction model evaluated using EEG segments preceding radiographic edema detection.

Main Results:

• The best detection model (Transformer, 4-hour EEG) achieved 80.0% AUC, 80.0% accuracy, 80.0% sensitivity, and 90.0% specificity.
• The top prediction model (Transformer, 8-hour EEG) demonstrated 92.2% AUC, 90.6% accuracy, 100% sensitivity, and 88.9% specificity.
• Transformer models outperformed LSTM in both detection and prediction tasks.

Conclusions:

• Machine learning applied to routine EEG data can effectively identify diffuse cerebral edema in cardiac arrest survivors.
• Transformer-based models show superior performance for early detection and prediction of cerebral edema.
• This strategy holds potential for earlier recognition of evolving edema during EEG monitoring, enabling timely interventions to reduce secondary brain injury.