Automated Assessment of Children's Postoperative Pain Using Computer Vision.

摘要: BACKGROUND: Current pain assessment methods in youth are suboptimal and vulnerable to bias underrecognition of clinical pain. Facial expressions a sensitive, specific biomarker the presence severity pain, computer vision (CV) machine-learning (ML) techniques enable reliable, valid measurement pain-related facial from video. We developed evaluated CVML approach measure for automated youth. METHODS: A CVML-based model pediatric postoperative was videos 50 neurotypical 5 18 years old both endogenous/ongoing exogenous/transient conditions after laparoscopic appendectomy. Model accuracy assessed self-reported ratings children time since surgery, compared with by-proxy parent nurse estimates observed RESULTS: detection versus no-pain demonstrated good-to-excellent (Area under receiver operating characteristic curve 0.84–0.94) ongoing transient conditions. moderate-to-strong correlations (r = 0.65–0.86 within; r 0.47–0.61 across subjects) The performed equivalently nurses but not as well parents detecting conditions, estimating severity. Nurses were more likely than underestimate ratings. Demographic factors did affect performance. CONCLUSIONS: models derived automatic expression measurements binary classifications, strong patient ratings, parent-equivalent estimation children’s levels over typical trajectories