NVIDIA and Siemens Healthineers Release NV-Raw2Insights-US for Adaptive Ultrasound Imaging

NVIDIA and Siemens Healthineers have released NV-Raw2Insights-US, an AI model that learns directly from raw ultrasound sensor data to estimate patient-specific speed of sound for adaptive image focusing. Traditional ultrasound imaging compresses raw sensor measurements into images using assumptions like constant sound speed, discarding information. NV-Raw2Insights-US processes raw channel data via the Holoscan Sensor Bridge, an open-source FPGA IP, streamed from an ACUSON Sequoia scanner's DisplayPort outputs over Ethernet to an NVIDIA IGX for inference on a Blackwell-class GPU. The model produces a personalized sound-speed map that corrects the live image in real time. This software-defined approach enables continuous improvement and modular expansion. The technology is under investigational development and not yet cleared for sale. The project involved collaboration with Ismayil Guracar and Rickard Loftman of Siemens Healthineers.