Enforcing Multilabel Consistency for Automatic Spatio-Temporal Assessment of Shoulder Pain Intensity

Area of Science:

• Biomedical Engineering
• Computer Vision
• Pain Medicine

Background:

• Current pain assessment relies on subjective self-reports or clinician observations, which have limitations.
• Objective, reliable pain measurement is crucial, especially when self-reporting is not feasible.
• Automatic facial expression analysis presents a promising avenue for objective pain assessment.

Purpose of the Study:

• To develop and evaluate a video-based approach for automatic measurement of self-reported pain and observer pain intensity.
• To explore the added value of integrating multiple pain scales (Visual Analog Scale, Sensory Scale, Affective Motivational Scale) and Observer Pain Intensity ratings.
• To improve the state-of-the-art in automatic pain estimation by enhancing prediction quality and consistency.

Main Methods:

• Utilized a spatio-temporal Convolutional Neural Network - Recurrent Neural Network (CNN-RNN) architecture.
• Jointly minimized mean absolute error for pain score estimation across different scales.
• Maximized consistency between pain scores derived from various self-reported scales and observer ratings.

Main Results:

• The proposed method demonstrated enhanced prediction quality by enforcing consistency between different pain scales.
• Achieved state-of-the-art results in automatic self-reported pain estimation.
• Validated reliability on the UNBC-McMaster Pain Archive benchmark dataset.

Conclusions:

• Automatic assessment of self-reported pain intensity from videos is feasible and reliable.
• Integrating multiple pain scales improves the accuracy of automated pain measurement.
• This technology can serve as a complementary tool to support caregivers, particularly for vulnerable populations requiring continuous monitoring.