Outcome Prediction in Postanoxic Coma With Deep Learning.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• Visual electroencephalogram (EEG) assessment by experts predicts outcomes in about half of comatose cardiac arrest patients.
• Deep neural networks (DNNs) offer a potential for more objective and consistent outcome prediction.

Purpose of the Study:

• To develop and validate a deep learning model for predicting neurological outcomes in comatose patients post-cardiac arrest.
• To compare the performance of the DNN model against traditional visual EEG assessment.

Main Methods:

• A prospective cohort study involving 895 comatose patients after cardiac arrest across five Dutch teaching hospitals.
• A convolutional neural network (VGG architecture) was trained using continuous EEG data from the first 3 days to predict 6-month functional outcomes.
• Model performance was evaluated using internal validation (n=661) and external validation (n=234).

Main Results:

• The DNN model predicted poor neurological outcome at 12 hours with 58% sensitivity and 0% false positive rate in the external validation set.
• Good outcome prediction at 12 hours achieved 48% sensitivity with a 5% false positive rate in the external validation set.
• The deep learning approach demonstrated superior performance compared to previous outcome predictors.

Conclusions:

• Deep learning analysis of EEG signals surpasses traditional visual assessment for predicting neurological outcomes in comatose cardiac arrest survivors.
• This AI-driven method provides objective, real-time prognostic insights at the bedside.
• The findings highlight the potential of DNNs to enhance clinical decision-making in critical care settings.