Auditory stimulation and deep learning predict awakening from coma after cardiac arrest.

Area of Science:

• Neuroscience and Artificial Intelligence
• Critical Care Medicine
• Signal Processing

Background:

• Prognosticating outcomes for patients in coma after cardiac arrest is challenging due to subjective clinical scoring.
• Current methods for assessing neural function in coma, like electroencephalogram (EEG) analysis, are often complex and not standardized for routine clinical use.
• A significant number of patients fall into a 'grey zone' with uncertain prognoses, highlighting the need for more objective assessment tools.

Purpose of the Study:

• To investigate the potential of convolutional neural networks (CNNs) to extract interpretable EEG patterns predictive of coma outcome.
• To assess if CNNs can accurately predict awakening and 3-month survival in patients following cardiac arrest using auditory stimuli.
• To evaluate the performance of CNNs in patients with uncertain prognoses (clinical 'grey zone').

Main Methods:

• Utilized convolutional neural networks (CNNs) to model single-trial EEG responses to auditory stimuli within the first 24 hours of coma.
• Analyzed data from a multicentre, multiprotocol patient cohort managed under standardized sedation and targeted temperature management.
• Validated CNN predictions against actual patient outcomes at 3 months, including awakening and survival.

Main Results:

• CNNs achieved a positive predictive power for awakening of approximately 0.83 (therapeutic hypothermia) and 0.81 (normothermia).
• The area under the curve for predicting overall outcome was approximately 0.70 for both patient groups.
• Predictive accuracy was maintained in the subset of patients within the clinical 'grey zone', and network confidence correlated with neural synchrony, complexity, and clinical markers like EEG reactivity.

Conclusions:

• Interpretable deep learning algorithms, specifically CNNs, show significant promise in improving the accuracy of coma outcome prognostication.
• Combining CNNs with auditory stimulation provides an objective and potentially more reliable method for assessing neural function in comatose patients.
• This approach can aid clinicians in managing uncertainty and making more informed decisions for patients recovering from cardiac arrest.