Artificial intelligence to evaluate postoperative pain based on facial expression recognition

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Pain management

Background:

• Self-reported pain intensity is subjective and challenging in non-communicating patients, potentially leading to inadequate pain management.
• Automated facial expression analysis using artificial intelligence (AI) for pain assessment in clinical settings remains largely unevaluated.

Purpose of the Study:

• To develop and validate a deep-learning system for objective pain intensity evaluation using facial expressions.
• To compare the AI system's pain assessment performance against that of experienced nurses.

Main Methods:

• A ResNet-18 convolutional neural network was trained and validated on 2810 facial expressions from 1189 patients, correlated with self-reported pain intensity (Numeric Rating Scale, NRS).
• AI performance was assessed by accuracy, sensitivity, and specificity for detecting pain thresholds (≥4/10 and ≥7/10).
• AI performance was benchmarked against pain intensity evaluations by 33 nurses using the same facial expressions.

Main Results:

• The AI system achieved 53% exact prediction of pain intensity (0-10 NRS) with a mean error of 2.4 points on an external test set (120 images).
• AI demonstrated high sensitivity in detecting significant pain (89.7% for ≥4/10, 77.5% for ≥7/10).
• Nurses' performance was significantly lower, with 14.9% mean accuracy and lower sensitivities for detecting pain (44.9% for ≥4/10, 17.0% for ≥7/10).

Conclusions:

• AI-based facial expression analysis shows potential as an objective tool to assist in pain assessment, particularly for patients unable to self-report.
• Further AI training may enhance performance, offering a rapid, standardized method for pain detection and management.
• This technology represents a significant advancement towards automated, objective pain measurement.