NVIDIA and Siemens Healthineers release NV-Raw2Insights-US, an AI model that learns directly from raw ultrasound sensor data to estimate sound speed and improve image focusing in real time.

NVIDIA and researchers from Siemens Healthineers developed NV-Raw2Insights-US, a reconstruction model that learns from raw ultrasound sensor signals instead of finished images, enabling the system to generate a personalized map of sound speed for each patient.

The model runs accelerated inference on a Blackwell-class GPU via NVIDIA Holoscan (an edge AI sensor processing platform) and uses Data over DisplayPort technology to stream raw ultrasound channel data from an ACUSON Sequoia scanner to NVIDIA IGX for processing, allowing real-time correction of ultrasound images.

By learning directly from raw ultrasound channel data rather than reconstructed images, the system reduces errors introduced by traditional assumptions (such as constant speed of sound) and adapts imaging for each patient, while establishing a modular foundation for integrating new AI models.