Endoshare: A Source-Available Solution to De-Identify and Manage Surgical Videos

Research Group CAMMA / University of Strasbourg / IHU Strasbourg

http://camma.u-strasbg.fr/

Overview

Endoshare is a cross-platform, source-available desktop application that merges, standardizes, and de-identifies surgical videos for research, education, and clinical quality improvement.
It was developed at the University of Strasbourg – ICube Laboratory and IHU Strasbourg to address the lack of accessible, privacy-preserving tools for managing endoscopic recordings.

The application is distributed under the PolyForm Noncommercial License 1.0.0, ensuring open access for research and educational purposes while restricting commercial use.

---

Architecture

Endoshare follows a three-layer architecture:

1. Presentation Layer –
   A PyQt5-based GUI allowing clinicians to perform all operations without coding.
   Functions include:

   • Case selection
   • Video merging
   • Out-of-body detection and removal
   • Filename pseudonymization
   • Batch processing with progress indicators

2. Application Layer –
   Implements the processing pipeline in Python, integrating:

   • FFmpeg and VidGear for format standardization and merging
   • OpenCV for frame manipulation
   • TensorFlow (MobileNetV2 + LSTM) for automatic out-of-body (OoB) detection
   • Dual processing modes:
     • Fast Mode – minimal re-encoding for speed
     • Advanced Mode – full re-encoding and parameter control

3. Data Layer –
   Manages I/O and ensures data integrity and confidentiality.

   • All metadata is removed using FFmpeg.
   • Filenames are replaced by pseudonyms (UUIDs).
   • A local CSV log securely maps patient IDs to pseudonyms for traceability.

---

Core Features

• 🧠 AI-based Out-of-Body Scene Detection
  Identifies and blurs non-endoscopic frames using a MobileNetV2-LSTM network.

• 🔒 Privacy-by-Design De-Identification
  Automatic removal of metadata and identifiers; local storage of pseudonym mappings.

• 💻 Cross-Platform Availability
  Runs on macOS, Windows, and Linux — from laptops to high-performance servers.

• ⚡ Dual Processing Modes
  Fast vs. Advanced, balancing efficiency and full control.

• 🔁 Multithreaded Batch Execution
  Enables unattended overnight processing with progress monitoring.

• 🧾 Transparent and Reproducible
  Full source code and documentation released under a noncommercial license.

---

Installation

From Binaries

Pre-compiled builds are available for:

• macOS (.dmg)
• Windows (.exe)
• Linux (.AppImage or .tar.gz)

Simply download the appropriate version from the official release page, extract, and launch Endoshare.

From Source

# 1. Clone the repository
git clone https://github.com/CAMMA-public/Endoshare_code.git
cd Endoshare_code

# 2. Install Python dependencies
pip install -r requirements.txt

# 3. Download model weights (OOBNet)
# Create the checkpoint directory and download weights into it
cd endoshare/resources
mkdir -p ckpt
cd ckpt
wget https://s3.unistra.fr/camma_public/github/oobnet_detection/ckpt/oobnet_weights.h5
cd ../../../

# 4. Install FFmpeg binaries
# The FFmpeg executable must be placed inside: endoshare/resources/Externals/ffmpeg/

# ── macOS / Linux ────────────────────────────────
# Create the folder and download the static build
mkdir -p endoshare/resources/Externals/ffmpeg
cd endoshare/resources/Externals/ffmpeg
wget https://johnvansickle.com/ffmpeg/releases/ffmpeg-release-amd64-static.tar.xz
tar -xf ffmpeg-release-amd64-static.tar.xz
mv ffmpeg-*-static/ffmpeg .
cd ../../../..

# ── Windows ─────────────────────────────────────
# 1. Go to https://www.gyan.dev/ffmpeg/builds/
# 2. Download "ffmpeg-git-full.7z"
# 3. Extract it and copy the ffmpeg.exe binary into:
#      endoshare\\resources\\Externals\\ffmpeg\\ffmpeg.exe

# 5. Launch Endoshare
python main.py

Requirements:

• Python ≥ 3.11
• FFmpeg, OpenCV, TensorFlow, PyQt5, loguru, tqdm, psutil

---

How It Works

1. Load Cases – Select a folder containing endoscopic video segments.
2. Merge Videos – Automatically concatenates clips from the same case.
3. De-Identify – Detects and removes out-of-body scenes; strips metadata.
4. Export – Produces anonymized videos with consistent format and naming.
5. Review – Preview results directly in the GUI.

---

Validation and Usability

• External survey of 10 surgeons
  → Perceived usefulness 5.07 ± 1.75 / 7
  → Ease of use 5.15 ± 1.71 / 7
  → Recommendation likelihood 9.2 ± 0.79 / 10
• Processing time was significantly affected by:
  • Mode (Fast vs. Advanced)
  • Video duration (p ≤ 0.001)
  • Machine computational power (p = 0.041)

---

Privacy and Compliance

Endoshare is research and educational software. It is not a medical device and is not certified for clinical or diagnostic use; manual review is recommended for privacy assurance.

---

Planned Extensions

• Integration with surgical data lakes and PACS systems
• Multi-user network mode
• Compliance audit logging
• Optional cloud-based anonymization queue

---

🤝 Contributors

Lorenzo Arboit, Dennis N. Schneider, Britty Baby, Vinkle Srivastav, Pietro Mascagni, Nicolas Padoy
University of Strasbourg, ICube, INSERM, IHU Strasbourg, and Fondazione Policlinico Universitario Agostino Gemelli IRCCS.

---

📜 License

This software is distributed under the PolyForm Noncommercial License 1.0.0.
You may use, copy, and modify this software for noncommercial purposes only.
See the full text in LICENSE.

---