🔩 Hacking Real-time AI workflows for Surgery 🔧

CMIC seminar

Zhehua Mao and Miguel Xochicale

ARC/WEISS at UCL

15th of November 2023, 13:00 GMT

🔧 Challenges

🚀 Real-time AI Workflow for Surgery

♻️📊 Model training & comparison

🔁🏥 Model deployment (IEC 62304)

♻️ Model deployment

🌐 Automatic deployment of gh-pages

📢 Takeaways

• Employed annotation open-source tools (e.g., Labelme and foundation model Segment Anything Model (SAM)).
• Prototyped Software problem resolution (Clause 9) following IEC 62304 standard.
• Participants trained models, investigating hyperparameter effects and data enhancement parameters.
• Testing and evaluation of trained models, and optimizing of real-time performance.
• Collaborated with NVIDIA engineer for exploring and deploying the optimized models on NVIDIA's cutting-edge device, Clara AGX.
• Produced an automatic deployment of GitHub pages as way to improve accessibility of our work.
• Documented the entire learning process on GitHub.

🏆 Best team name

🙌 Acknowledgements

• Snigdha Sen and Florence Townend

  • UCL Centre for Medical Image Computing (CMIC)

• Mikael Brudfors

  • NVIDIA Healthcare AI

• Steve Thompson

  • Advanced Research Computing Centre (ARC)

• Sophia Bano and Matt Clarkson

  • Wellcome / EPSRC Centre for Interventional and Surgical Sciences (WEISS)