A Deep Neural Network-Based Pain Classifier Using a Photoplethysmography Signal

Area of Science:

• Biomedical Engineering
• Anesthesiology
• Machine Learning

Background:

• Accurate intraoperative pain assessment is vital to prevent analgesic side effects.
• Current methods for pain evaluation can be subjective and challenging during surgery.

Purpose of the Study:

• To develop and evaluate a novel pain classifier utilizing deep belief networks (DBN) and photoplethysmography (PPG) signals.
• To assess the efficacy of a bagging ensemble model in enhancing pain classification performance.

Main Methods:

• Feature extraction from photoplethysmography (PPG) signals.
• Development of a deep belief network (DBN) model to learn nonlinear relationships between PPG features and pain status (Numeric Rating Scale - NRS).
• Implementation of a selective bagging ensemble method to improve classification accuracy.

Main Results:

• The DBN-based pain classifier demonstrated superior performance compared to traditional Multilayer Perceptron Neural Network (MLPNN) and Support Vector Machine (SVM) models.
• The selective bagging ensemble model further enhanced the classification accuracy of the DBN classifier.
• The model successfully learned complex, nonlinear associations between PPG features and patient-reported pain levels.

Conclusions:

• A DBN utilizing PPG signals offers a promising, objective approach for intraoperative pain assessment.
• Ensemble methods, particularly selective bagging, can significantly improve the robustness and accuracy of PPG-based pain classification.
• This developed pain classifier has the potential to aid in the creation of advanced, automated pain management systems.