Real-Time Non-Invasive Imaging and Detection of Spreading Depolarizations through EEG: An Ultra-Light Explainable...

Area of Science:

• Neurocritical care
• Neuroscience
• Medical technology

Background:

• Secondary brain injury is a critical concern in neurocritical care.
• Spreading depolarizations (SDs) are a major cause of secondary brain injury.
• Current SD detection methods rely on invasive electrocorticography, limiting clinical application.

Purpose of the Study:

• To develop a non-invasive method for detecting SDs using scalp electroencephalogram (EEG).
• To improve the accuracy and efficiency of SD detection compared to conventional methods.
• To enable flexible EEG electrode configurations for SD detection.

Main Methods:

• A novel ultra-light-weight multi-modal deep-learning network was developed.
• The model fuses EEG spectrogram imaging with temporal power vectors.
• The approach transforms SD identification from a 1-D time-series detection to a 2-D image-based task.

Main Results:

• The proposed model achieves ultra-fast processing speeds (<0.3 seconds), a significant improvement over conventional methods (2 hours).
• The model enhances SD identification accuracy by incorporating frequency information from spectrograms.
• This method allows for SD detection using ultra-low-density EEG with variable electrode positioning.

Conclusions:

• The novel deep-learning approach offers a significant advancement in early SD detection for neurocritical care.
• Incorporating frequency-domain information via spectrograms improves SD detection accuracy.
• This non-invasive technique facilitates instant brain injury prognosis and supports flexible clinical application.