Adaptive Ultrasound Imaging with Physics-Informed NV-Raw2Insights-US AI
Learns directly from raw sensor data, not processed images, to adapt sound physics.
NVIDIA, in collaboration with Siemens Healthineers, introduced NV-Raw2Insights-US, a physics-informed AI model that transforms ultrasound imaging by learning directly from raw sensor data rather than traditional reconstructed images. Traditional ultrasound pipelines compress rich raw measurements, discarding valuable information about how sound interacts with tissue. This new model leverages raw channel data—the closest representation of sound-tissue interaction—to estimate a personalized speed-of-sound map for each patient. This map is then used to adaptively correct image focusing in real time, replacing complex, time-consuming computations with a single AI inference pass on a Blackwell-class GPU. The approach, called Raw2Insights, aims to shift ultrasound from fixed algorithms to end-to-end AI that actively understands and adapts to individual patient physics.
Deployment is enabled by NVIDIA's Holoscan Sensor Bridge (HSB), an open-source FPGA IP that streams high-bandwidth raw ultrasound data from scanners like the ACUSON Sequoia via DisplayPort outputs over Ethernet to an NVIDIA IGX system. This Data over DisplayPort approach allows software-only integration with existing clinical hardware, facilitating continuous AI model updates without hardware changes. The system runs on NVIDIA Holoscan, an edge AI platform, and supports modular expansion for new AI models. By moving beyond hand-engineered pipelines, this technology promises to unlock new diagnostic capabilities, improve image quality, and enable real-time adaptive imaging across diverse patient anatomies.
- NV-Raw2Insights-US learns from raw ultrasound channel data, not processed images, to create a personalized sound-speed map per patient.
- It corrects image focus in real time using a single AI pass on a Blackwell-class GPU, replacing complex traditional computations.
- Deployed via NVIDIA Holoscan Sensor Bridge and Data over DisplayPort, enabling software-only upgrades on existing scanners like ACUSON Sequoia.
Why It Matters
Real-time, patient-specific ultrasound correction could improve diagnostic accuracy and enable continuous AI-driven improvements in medical imaging.