Transformer encoder with multiscale deep learning for pain classification using physiological signals.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Artificial Intelligence in Healthcare

Background:

• Pain is a widespread health issue with challenging assessment due to self-report limitations.
• Conventional pain scales suffer from inconsistency and bias, necessitating advanced assessment methods.

Purpose of the Study:

• To introduce PainAttnNet, a novel deep-learning model for precise pain intensity classification.
• To evaluate PainAttnNet's ability to outperform existing models in capturing temporal dependencies in physiological signals.

Main Methods:

• Developed PainAttnNet, integrating multiscale convolutional networks, squeeze-and-excitation residual networks, and a transformer encoder block.
• Utilized a deep learning approach to extract robust features and analyze temporal dependencies from physiological signals.
• Evaluated the model on the BioVid heat pain dataset.

Main Results:

• PainAttnNet achieved superior performance compared to existing models in pain intensity classification.
• The model effectively captured temporal dependencies within physiological data.
• Demonstrated the model's capability in extracting robust and relevant features.

Conclusions:

• PainAttnNet represents a significant advancement in automated pain assessment.
• The model shows potential for more accurate and individualized pain management strategies.
• Highlights the utility of deep learning in analyzing physiological signals for pain evaluation.