Perioperative medicine & outcomes
Preliminary Development and Validation of Automated Nociception Recognition Using Computer Vision in Perioperative Patients
Authors' conclusion
Automated nociception detection using computer vision alone is feasible but requires additional testing and validation given the small datasets used.
Study context
- Population
- Adults (≥18 yr) admitted for surgical procedures, San Francisco Veterans Affairs Medical Center; external data from University of Northern British Columbia-McMaster University and Delaware Pain Database130 development, 77 validation, 25 + 229 external
- Exposure
- Computer vision CNN models classifying Care Pain Observation Tool (CPOT) and NRS pain/no pain
- Outcome
- Binary pain/no pain classification (AUC)
Methodological detailsdiagnostic model development and validation, internal and external validation
- Diagnostic model development and validation
- Internal and external validation
Main finding
As reported in the abstract
Binary pain/no pain classification (AUC of ROC)
| Group | AUC, development cohort | AUC, San Francisco VA validation cohort | AUC, University of Northern British Columbia-McMaster University | AUC, Delaware | AUC |
|---|---|---|---|---|---|
| CPOT | 0.71 (95% CI, 0.70 to 0.74) | 0.91 (95% CI, 0.90 to 0.92) | 0.91 (95% CI, 0.89 to 0.93) | 0.80 (95% CI, 0.75 to 0.85) | — not stated in the abstract |
| Numeric rating scale | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | 0.58 [95% CI, 0.55 to 0.61] |
AUC, development cohort
CPOT0.71 (95% CI, 0.70 to 0.74)
Numeric rating scale— not stated in the abstract
AUC, San Francisco VA validation cohort
CPOT0.91 (95% CI, 0.90 to 0.92)
Numeric rating scale— not stated in the abstract
AUC, University of Northern British Columbia-McMaster University
CPOT0.91 (95% CI, 0.89 to 0.93)
Numeric rating scale— not stated in the abstract
AUC, Delaware
CPOT0.80 (95% CI, 0.75 to 0.85)
Numeric rating scale— not stated in the abstract
AUC
CPOT— not stated in the abstract
Numeric rating scale0.58 [95% CI, 0.55 to 0.61]
— not stated in the abstract
Observational data: associations, not effects.
Further findings
Calibration (Brier scores)
- Brier scores improved after calibration across multiple techniques
Explainability (perturbation models, CPOT)
- Eyebrows, nose, lips and forehead most important for model prediction
Original publication
Preliminary Development and Validation of Automated Nociception Recognition Using Computer Vision in Perioperative Patients. Anesthesiology. Published 13 January 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →