How SDSC Uses AI to Transform Surgical Training and Practice - Ep. 241

What It Covers

Margo Mason Forsyth, Director of Machine Learning at Surgical Data Science Collective, explains how the nonprofit analyzes 40 terabytes of surgical video using computer vision to standardize techniques, improve education, and reduce surgical mortality worldwide.

Key Questions Answered

• •Data Collection Challenge: Surgeons must manually press record, export video from devices, transfer via USB, and upload to cloud—a multi-step process that prevents most of the thousands of terabytes of surgical footage from being captured and analyzed.
• •Surgical Mortality Scale: Five billion people lack access to safe surgery globally, and 4.2 million die within 30 days of surgery annually. If treated as a disease, surgery would rank as the leading cause of death worldwide, making standardization critical.
• •Technical Architecture Approach: Start with simple models on small datasets for proof of concept, then iterate and scale. Combine CNN architectures with temporal models and vision transformers to handle long surgical videos with temporal dependencies and environmental challenges like obstructions.
• •Standardization Gap: Every surgeon in every hospital performs procedures differently with no agreed ABCD protocol for specific operations. Analyzing shared surgical videos creates the first mechanism for global surgeon collaboration and identification of best practices through outlier detection and pattern analysis.