Ensuring privacy protection in the era of big laparoscopic video data: development and validation of an inside outside discrimination algorithm (IODA)

• DOI: 10.1007/s00464-023-10078-x
• Publication Date: 2023-05-05T15:01:53Z
• Authors: A. Schulze; D. Tran; M. T. J. Daum; A. Kisilenko; L. Maier-Hein; S. Speidel; M. Distler; J. Weitz; B. P. Müller-Stich; S. Bodenstedt; M. Wagner
• Keywords: Frame rate; Censoring (clinical trials); Data set

Abstract Background Laparoscopic videos are increasingly being used for surgical artificial intelligence (AI) and big data analysis. The purpose of this study was to ensure privacy in video recordings laparoscopic surgery by censoring extraabdominal parts. An inside-outside-discrimination algorithm (IODA) developed protection while maximizing the remaining data. Methods IODAs neural network architecture based on a pretrained AlexNet augmented with long-short-term-memory. set training testing contained total 100 23 different operations length 207 h (124 min ± per video) resulting 18,507,217 frames (185,965 149,718 video). Each frame tagged either as abdominal cavity, trocar, operation site, outside cleaning, or translucent trocar. For testing, stratified fivefold cross-validation used. Results distribution annotated classes were cavity 81.39%, trocar 1.39%, site 16.07%, cleaning 1.08%, 0.07%. Algorithm binary all five showed similar excellent results classifying mean F1-score 0.96 0.01 0.97 0.01, sensitivity 0.02 0.0.97 false positive rate 0.99 respectively. Conclusion IODA is able discriminate between inside high certainty. In particular, only few misclassified therefore at risk breach. anonymized can be multi-centric development AI, quality management educational purposes. contrast expensive commercial solutions, made open source improved scientific community.