• Timothy A Heintz, Anusha Badathala, Avery Wooten, Cassandra W Cu, Alfred Wallace, Benjamin Pham, Arthur W Wallace, and Julien Cobert.
• Department of Anesthesiology, Perioperative and Pain Medicine, Brigham and Women's Hospital, Boston, MA, USA.
• Anesthesiology. 2025 Jan 13.

BackgroundEffective pain recognition and treatment in perioperative environments reduce length of stay and decrease risk of delirium and chronic pain. We sought to develop and validate preliminary computer vision-based approaches for nociception detection in hospitalized patients.MethodsProspective observational cohort study using red-green-blue camera detection of perioperative patients. Adults (≥18 years) admitted for surgical procedures to the San Francisco Veterans Affairs Medical Center (SFVAMC) were included across 2 study phases: (1) algorithm development phase and (2) internal validation phase. Continuous recordings occurred perioperatively across any postoperative setting. We inputted facial images into convolutional neural networks using a pretrained backbone, to detect (1) critical care pain observation tool (CPOT) and (2) numerical rating scale (NRS). Outcomes were binary pain/no-pain. We performed external validation for CPOT and NRS classification on data from University of Northern British Columbia-McMaster University (UNBC) and Delaware Pain Database. Perturbation models were used for explainability.ResultsWe included 130 patients for development, 77 patients for validation cohort and 25 patients from UNBC and 229 patients from Delaware datasets for external validation. Model area under the curve of the receiver operating characteristic for CPOT models were 0.71 (95% confidence interval [CI] 0.70, 0.74) on the development cohort, 0.91 (95% CI 0.90, 0.92) on the SFVAMC validation cohort, 0.91 (0.89, 0.93) on UNBC and 0.80 (95% CI 0.75, 0.85) on Delaware. NRS model had lower performance (AUC 0.58 [95% CI 0.55, 0.61]). Brier scores improved following calibration across multiple different techniques. Perturbation models for CPOT models revealed eyebrows, nose, lips, and foreheads were most important for model prediction.ConclusionsAutomated nociception detection using computer vision alone is feasible but requires additional testing and validation given small datasets used. Future multicenter observational studies are required to better understand the potential for automated continuous assessments for nociception detection in hospitalized patients.Copyright © 2025 American Society of Anesthesiologists. All Rights Reserved.

13 keywords...

hide…