Automatic Estimation of Self-Reported Pain by Interpretable Representations of Motion Dynamics

Area of Science:

• Medical imaging
• Computer vision
• Machine learning

Background:

• Objective pain assessment remains challenging.
• Facial dynamics offer potential indicators of pain intensity.
• Automated analysis of subtle facial changes is needed.

Purpose of the Study:

• To develop an automated system for pain intensity measurement from video data.
• To utilize facial movement dynamics for quantitative pain assessment.
• To compare the proposed method against state-of-the-art techniques.

Main Methods:

• Utilized 66 facial points to capture facial movement dynamics.
• Employed Gram matrices on Riemannian manifolds for trajectory representation.
• Applied curve fitting and temporal alignment for trajectory smoothing.
• Trained a Support Vector Regression model for pain level encoding.

Main Results:

• The system successfully encoded facial trajectories into ten pain intensity levels.
• Evaluated on the UNBC McMaster Shoulder Pain Archive, achieving competitive performance.
• Cross-validation results demonstrated robustness and comparable accuracy to existing methods.

Conclusions:

• The proposed automated method provides a viable approach for objective pain intensity measurement from video.
• Facial movement analysis, when combined with advanced computational techniques, can effectively quantify pain.
• This method holds potential for clinical applications in pain assessment.