Deep learning-guided postoperative pain assessment in children.

Area of Science:

• Medical informatics
• Computer vision
• Pediatric medicine

Background:

• Current automated pain assessment tools are limited to infants and youth, posing challenges for diverse pediatric populations.
• Postoperative pain management in children requires objective and efficient assessment methods across a wide age range (0-14 years).

Purpose of the Study:

• To introduce a large-scale dataset, the Clinical Pain Expression of Children (CPEC), for pediatric postoperative pain assessment.
• To develop and validate a deep learning framework, the Children Pain Assessment Neural Network (CPANN), for automated pain evaluation using facial expressions.

Main Methods:

• Collected 4104 preoperative and 4865 postoperative videos of 4104 children (0-14 years) from January to December 2020.
• Developed the Children Pain Assessment Neural Network (CPANN), a deep learning model analyzing facial expressions for pain detection.
• Trained and evaluated the CPANN using the CPEC dataset, measuring performance with accuracy and macro-F1 scores.

Main Results:

• The CPANN achieved 82.1% accuracy and 73.9% macro-F1 score on the CPEC testing set.
• Demonstrated superior speed, convenience, and objectivity compared to traditional pain scales.
• Validated the effectiveness of deep learning for automated pediatric pain assessment.

Conclusions:

• The CPANN offers a promising, objective, and efficient solution for assessing postoperative pain in children.
• The CPEC dataset provides a valuable resource for advancing research in automated pediatric pain assessment.
• Deep learning-based facial expression analysis represents a significant advancement in pediatric pain management.