Hybrid RNN-ANN Based Deep Physiological Network for Pain Recognition.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Pain Management

Background:

• Quantitative pain assessment is crucial for effective patient treatment and distress relief.
• Current self-report methods lack objectivity and accuracy.
• Existing physiological signal-based methods rely on handcrafted features.

Purpose of the Study:

• To develop an objective pain classification system using physiological signals.
• To enhance pain assessment accuracy by combining deep learning with handcrafted features.
• To introduce a hybrid classifier integrating auto-extracted and expert-defined features.

Main Methods:

• A deep Recurrent Neural Network (RNN), specifically a bidirectional Long Short-Term Memory (biLSTM) network, was employed to automatically extract temporal features from physiological signals.
• Handcrafted features, based on domain expertise, were extracted and fused with the RNN-generated features.
• An Artificial Neural Network (ANN) was trained on the combined feature set for pain intensity classification.

Main Results:

• The hybrid classifier achieved a classification accuracy of 83.3%.
• The integration of handcrafted features significantly enhanced the performance of the RNN classifier.
• Comparative analysis on an open dataset demonstrated superior performance over existing methods.

Conclusions:

• The proposed hybrid feature approach offers a significant advancement in objective pain assessment.
• Combining deep learning's feature extraction with handcrafted features improves classification accuracy.
• This research validates the effectiveness of hybrid features for physiological signal-based pain evaluation.