A Multimodal Deep Learning Approach to Intraoperative Nociception Monitoring: Integrating Electroencephalogram,...

Area of Science:

• Anesthesiology
• Biomedical Engineering
• Neuroscience

Background:

• Monitoring nociception (pain perception) under general anesthesia is difficult.
• Existing single-parameter methods have limitations in accurately assessing pain pathways.
• There is a need for advanced, reliable methods for real-time nociception assessment.

Purpose of the Study:

• To develop and evaluate a multimodal approach for predicting nociception during general anesthesia.
• To compare the performance of deep learning models (MLP and LSTM) in nociception monitoring.
• To identify optimal data normalization strategies for clinical application.

Main Methods:

• Collected electroencephalogram (EEG), photoplethysmography (PPG), and electrocardiogram (ECG) data from patients under general anesthesia.
• Developed and trained Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM) deep learning models on expert-assessed nociception data.
• Evaluated various data normalization techniques, including Min-Max normalization, for offline and online use.

Main Results:

• The Multilayer Perceptron (MLP) model demonstrated high accuracy in capturing nociceptive changes in response to surgical stimuli.
• The Long Short-Term Memory (LSTM) model provided smoother nociception predictions but showed lower sensitivity to rapid changes.
• Min-Max normalization was identified as the most effective strategy for the analyzed dataset.

Conclusions:

• Multimodal data integration with deep learning offers a promising solution for real-time nociception monitoring.
• The MLP model shows potential for accurate and sensitive pain assessment in clinical settings.
• These findings can advance the development of improved patient monitoring during general anesthesia.