Preserving privacy in surgical video analysis using a deep learning classifier to identify out-of-body scenes in endoscopic videos
Problem
Endoscopic surgical videos can record privacy-sensitive out-of-body scenes, exposing patients and operating room staff when videos are used for education and research.
Why it matters
Identifiable footage in surgical videos threatens patient and staff privacy while blocking valuable educational and research use.
What the AI did
A deep learning classifier was trained on an internal dataset of 356,267 images from 12 laparoscopic and robotic surgery types to identify out-of-body scenes, and externally validated on two independent multicentric test datasets.
Results
Performance measured as ROC AUC against human ground-truth annotations; numeric value not stated in abstract.
Impact
The model was externally validated on two independent multicentric datasets, but the abstract does not state the performance value.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Joël Lukas Lavanchy, Armine Vardazaryan, Pietro Mascagni, Giovanni Guglielmo Laracca, Ludovica Guerriero, Andrea Spota, Claudio Fiorillo, Giuseppe Quero et al. |
| Citations | 25 (OpenAlex, at import time) |
| References | 30 |
| Model or method | deep learning classifier (specific architecture not stated in source) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41598-023-36453-1 |
| Open access | gold |
| Keywords | surgical video, privacy, deep learning, endoscopy, out-of-body detection |
Importance rating
real but narrower advance
Privacy protection in surgical videos enables education/research use; real but narrow advance. Validated on 356,267 internal images across 12 surgery types, but abstract gives no population-scale impact evidence.