Automated EEG Classification to Track Levels of Consciousness.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Technology

Background:

• Accurate prognostication in acute brain injury is hampered by a lack of bedside consciousness biomarkers.
• The ABCD framework classifies electroencephalography (EEG) to assess thalamocortical network function, aiding diagnosis and prognosis in severe brain injuries.
• Current ABCD classification relies on labor-intensive visual inspection of EEG power spectra, requiring specialized expertise.

Purpose of the Study:

• To develop and validate an automated, accurate, and well-calibrated convolutional neural network (CNN)-based classifier for EEG into ABCD categories.
• To demonstrate the clinical utility of the automated classifier for continuous EEG analysis in intensive care unit (ICU) patients.

Main Methods:

• Developed a CNN-based automated classifier using a dataset of 4,611 manually classified EEG power spectra.
• Compared the performance of the automated classifier against the gold standard visual inspection and an alternative automated spectral analysis method.
• Applied the classifier to a continuous EEG recording from an ICU patient with acute severe traumatic brain injury.

Main Results:

• The automated CNN classifier achieved high accuracy and calibration, performing comparably to the gold standard visual inspection.
• The automated classifier outperformed an alternative automated spectral analysis method.
• Continuous ABCD classifications from the patient's EEG demonstrated the classifier's ability to capture state fluctuations with high temporal and spatial resolution.

Conclusions:

• The automated ABCD classifier enables efficient analysis of continuous EEG, facilitating bedside application of the ABCD framework for acute severe brain injury patients.
• This technology enhances the potential for analyzing large EEG datasets, potentially yielding new insights into consciousness electrophysiology.
• The automated classifier promises to improve prognostication and clinical decision-making for brain-injured patients in the ICU.