Differentiating loss of consciousness causes through artificial intelligence-enabled decoding of functional
Young-Tak Kim1, Hayom Kim2, Mingyeong So2
1Department of Biomedical Sciences, Korea University College of Medicine, 73, Goryeodae-ro, Seongbuk-gu, Seoul, 02841, Republic of Korea.
Neuroimage
|July 21, 2024
Summary
Artificial intelligence (AI) models can now differentiate causes of acute loss of consciousness (LOC) using electroencephalography (EEG) functional connectivity. A convolutional neural network (CNN) achieved high accuracy, aiding diagnosis and treatment selection.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Diagnostics
Background:
- Acute loss of consciousness (LOC) presents diagnostic challenges due to overlapping symptoms across various etiologies.
- Altered functional connectivity (FC) is implicated in LOC pathophysiology, yet specific patterns for differential diagnosis remain underexplored.
- Developing AI models to identify distinct FC patterns for different LOC causes is essential for targeted therapies.
Purpose of the Study:
- To develop and evaluate artificial intelligence (AI) models for the differential diagnosis of acute loss of consciousness (LOC) based on electroencephalography (EEG) functional connectivity (FC).
- To compare different feature extraction methods and machine learning algorithms, including deep learning with convolutional neural networks (CNNs), for classifying LOC etiologies.
- To determine the optimal EEG data parameters, such as epoch size and brain wave bands, for accurate LOC differentiation.
Main Methods:
- Extracted features using three-dimensional FC adjacency matrices, vectorized FC values, and graph theoretical measurements.
- Implemented deep learning (CNN) and various machine learning algorithms using electroencephalography (EEG) data.
- Optimized classification accuracy by varying epoch sizes and analyzing brain wave band contributions (delta and theta).
Main Results:
- The CNN model utilizing FC adjacency matrices achieved the highest classification accuracy, with an Area Under the Curve (AUC) of 0.905.
- Optimal performance was observed using 20-second EEG epochs.
- Key distinguishing features were identified within the delta and theta brain wave frequency bands.
- High accuracy was validated in a prospective cohort study.
Conclusions:
- AI models, particularly CNNs leveraging FC patterns from EEG, can accurately differentiate causes of acute LOC.
- The use of 20-second EEG epochs and analysis of delta/theta bands offers a clinically practical approach for diagnosis.
- This research enhances understanding of LOC mechanisms and holds promise for improving diagnostic accuracy and treatment selection.
