Deep learning of early brain imaging to predict post-arrest electroencephalography

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Current guidelines recommend both CT and EEG for post-cardiac arrest prognostication.
• The potential for strong associations between CT and EEG findings may allow for streamlined diagnostic pathways.

Purpose of the Study:

• To quantify the associations between CT imaging and EEG findings in comatose patients after cardiac arrest using deep learning.
• To evaluate the predictive performance of machine learning models based on clinical data and CT imaging for EEG classification.

Main Methods:

• A retrospective study of 500 comatose cardiac arrest patients.
• Deep learning models were developed to predict EEG classifications (generalized suppression, highly pathological, benign) from clinical data and processed CT scans.
• Model performance was evaluated using AUC in test sets.

Main Results:

• Clinical machine learning models demonstrated moderate discrimination (AUC 0.73-0.80).
• Image-based deep learning models showed lower performance (AUC 0.51-0.69), particularly in distinguishing benign from pathological EEG.
• Integrating CT-based deep learning with clinical models improved prediction of generalized suppression by identifying cerebral edema.

Conclusions:

• CT and EEG provide complementary information crucial for assessing post-arrest brain injury.
• Selective acquisition of only one modality is not supported, except in cases of severe injury.