Privacy-proof Live Surgery Streaming: Development and Validation of a Low-cost, Real-time Robotic Surgery...

Area of Science:

• Medical imaging and computer vision in surgery.
• Development of AI-driven surgical tools.

Background:

• Surgical video data is crucial for research and training.
• Anonymization of endoscopic surgery videos is necessary for privacy.
• Existing solutions lack real-time, platform-independent anonymization.

Purpose of the Study:

• To develop a novel surgical anonymization algorithm for real-time, accurate removal of out-of-body images.
• To validate the algorithm across diverse robotic surgical platforms.
• To create a procedure-independent anonymization solution.

Main Methods:

• A deep learning algorithm was developed using 63 surgical videos from 6 procedures on 4 robotic systems.
• The algorithm automatically detected out-of-body frames from 496,828 images.
• Performance was benchmarked against existing methods, with a postprocessing step and real-time streaming setup tested.

Main Results:

• The algorithm achieved a 99.46% ROC AUC on unseen procedures, improving to 99.89% after postprocessing.
• The Robotic Anonymization Network outperformed state-of-the-art algorithms on unseen procedures.
• The solution demonstrated effectiveness regardless of the robotic platform or procedure type.

Conclusions:

• The Robotic Anonymization Network provides reliable, accurate, and safe real-time surgical anonymization.
• The deep learning model is effective across various robotic platforms and complex procedures.
• The tool is suitable for live surgical streaming and is openly accessible.