Personalized Deep Bi-LSTM RNN Based Model for Pain Intensity Classification Using EDA Signal

Area of Science:

• Physiology
• Machine Learning
• Signal Processing

Background:

• Automatic pain assessment from physiological signals is an emerging field.
• Traditional methods rely on domain-specific features, limiting automation.
• Deep learning offers potential for automated feature extraction from raw signals.

Purpose of the Study:

• To investigate deep learning models for personalized pain intensity classification.
• To compare Bidirectional Long short-term memory Recurrent Neural Networks (BiLSTM RNN) and an ensemble of BiLSTM RNN with Extreme Gradient Boosting Decision Trees (XGB).
• To evaluate the performance of deep learning-generated features versus knowledge-based features.

Main Methods:

• Recorded Electrodermal Activity (EDA) signals during a cold pressor test from 29 subjects.
• Decomposed EDA signals into tonic and phasic components and augmented them.
• Developed and evaluated personalized BiLSTM RNN and BiLSTM-XGB models for four-category pain classification.

Main Results:

• The BiLSTM-XGB model achieved an average F1-score of 0.81 and an Area Under the ROC curve (AUROC) of 0.93.
• The ensemble model outperformed the standalone BiLSTM model.
• Combining deep learning features with knowledge-based features further improved XGB model performance.

Conclusions:

• Deep learning models, particularly the BiLSTM-XGB ensemble, show significant promise for automated pain intensity assessment.
• This approach can surpass limitations of traditional feature engineering, leveraging raw physiological signal data.
• Personalized deep learning models offer a path towards more objective and real-time pain monitoring.