Edge-to-Cloud Architecture for Real-Time Surgical Intelligence with AWS and NVIDIA

This article describes an edge-to-cloud architecture for surgical AI that enables real-time clinical decision support in operating rooms while maintaining privacy and compliance.

• NVIDIA IGX and Holoscan platform process surgical video at the edge with sub-20ms latency for de-identification, phase recognition, and instrument detection
• AWS SageMaker trains models on anonymized surgical data using YOLOv8 for instrument detection and CNN+LSTM for phase recognition
• AWS IoT Greengrass manages deployment, versioning, and orchestration of AI models across hospital devices at scale
• De-identification occurs at the edge before data leaves the operating room, ensuring privacy compliance across jurisdictions
• Feedback loop enables continuous improvement: procedures generate anonymized data that retrains models deployed back to edge devices
• Flexible connectivity supports three deployment models: real-time streaming, fully disconnected, or telemetry-only modes
• CloudWatch and SageMaker Monitor provide fleet-wide observability for model performance and device health

The architecture transforms surgical video into a learning system by combining edge inference for real-time support with cloud-based model training and fleet management.