Multimodal Pain Recognition in Postoperative Patients: Machine Learning Approach.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Pain Management

Background:

• Effective acute pain management is crucial for postoperative patients, especially those unable to self-report pain.
• Current pain assessment methods (subjective reports, behavioral tools) lack objectivity and consistency.
• Multimodal pain assessment using physiological and behavioral data offers a path to more accurate pain measurement, but real-world clinical data is limited.

Purpose of the Study:

• To develop and evaluate a multimodal machine learning framework for objective pain assessment in postoperative patients.
• To utilize biosignals like electrocardiogram, electromyogram, electrodermal activity, and respiration rate (RR) in real clinical settings.
• To address challenges in clinical pain monitoring, including motion artifacts and imbalanced data.

Main Methods:

• The iHurt study involved 25 postoperative patients, collecting multimodal biosignals and Numerical Rating Scale pain scores.
• Data preprocessing included noise filtering and feature extraction, combining handcrafted and autoencoder-derived features.
• Machine learning classifiers were trained using weak supervision and oversampling to manage sparse and imbalanced pain data.

Main Results:

• Multimodal pain recognition models achieved an average balanced accuracy exceeding 80% across pain levels.
• Respiration rate (RR) models showed strong performance, especially for lower pain intensities; electromyogram was effective for higher intensities.
• While single modalities like RR performed well, the multimodal framework demonstrated improved overall accuracy compared to previous studies.

Conclusions:

• A novel multimodal machine learning framework for objective postoperative pain recognition was developed.
• Integrating multiple biosignal modalities shows significant potential for enhancing pain assessment accuracy.
• The framework offers valuable applications for real-world clinical pain monitoring.