Continuous, Contactless, and Multimodal Pain Assessment During Surgical Interventions.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Pain Medicine

Background:

• Objective pain assessment remains a challenge in clinical settings.
• Multimodal approaches integrating physiological and behavioral data show promise for pain detection.

Purpose of the Study:

• To introduce a continuous, multimodal pain classification technique using camera-based data.
• To validate the integration of facial Action Units (AUs) and video-derived vital signs for pain assessment.

Main Methods:

• Facial AUs and heart rate variability (HRV) from video-derived photoplethysmography were extracted.
• A contactless, multimodal algorithm for binary pain classification was developed using logistic regression.
• The approach was validated on the BioVid Heat Pain Database and surgical patient data.

Main Results:

• An F1-score of 53% was achieved on the BioVid Heat Pain Database.
• An F1-score of 48% was achieved on surgical patient data using ultra-short-term processing.
• Video-based HRV measurement was validated against electrocardiogram.

Conclusions:

• The developed approach offers a robust foundation for contactless, multimodal pain classification.
• Camera-based vital signs and facial mimic parameters from short recordings can be utilized for pain assessment.
• This technique has potential applications in clinical pain monitoring.