Overview:
Philips Healthcare, a global leader in medical ultrasound systems (with ~300,000 installations and ~1.33 billion procedures annually) tackled a dual challenge: lengthy R&D innovation cycles and clinical limitations of ultrasound imaging (patient-dependence, variable acoustic windows, lower 3D volume rates).
The solution: migrating from custom FPGA-based beamforming hardware to a software-defined architecture powered by NVIDIA RTX GPUs and the CUDA framework. This enables massively parallel processing of ultrasound signal chains (beamforming, filtering, envelope detection, 3D volume rendering) and adaptive imaging in real time.
Within the organization, this operates in the R&D / product engineering domain of Philips’ ultrasound business, and in clinical imaging workflows across cardiology, general imaging, obstetrics/gynecology and point-of-care units. The data and processes include raw ultrasound signals, beam-forming and image reconstruction pipelines, 3D volume data, and adaptive imaging workflows. The core technologies include GPUs for accelerated compute, software-defined beamforming, and image reconstruction/visualization algorithms.
Key Features:
Migrates beam-forming pipeline from custom FPGA hardware to a software-defined GPU architecture (using RTX GPUs)
Accelerates core ultrasound processing (beamforming, filtering, envelope detection, 3D volume reconstruction and rendering) via the CUDA parallel-compute framework
Enables adaptive, patient-specific imaging that adjusts to each acoustic window and anatomical context in real time
Improves imaging throughput (higher 3D volume frame-rates) and image quality (sharper, photorealistic 3D visualisation)
Modularizes hardware-software architecture so future algorithm and GPU upgrades can extend system life and reduce total cost of ownership
Results & Impact:
Achieved approximately a 70% improvement in 3D color volume rates (one of the most computationally intensive ultrasound modes) compared to the prior FPGA-based system.
Reduced algorithm R&D cycle time from “years” to “months”, enabling faster time to market for new imaging capabilities.
Enabled real-time adaptive 3D cardiac imaging at near 2D frame rates, allowing clinicians to visualise complex valve anatomy in interventional procedures, thereby improving diagnostic confidence and expanding minimally invasive treatment options.
Transition to software-defined architecture contributed to sustainability goals: less custom hardware, longer lifecycle via software upgrades, and more energy-efficient GPU processing.
AI Technology:
AI Model Types: LLMs not explicitly referenced; primary compute acceleration via GPUs and parallel algorithms. The architecture supports image-reconstruction and adaptive processing (so include: Generative/Adaptive imaging algorithms, image-processing AI).
AI Purpose: Accelerate (compute-intensive imaging), Adapt (patient-specific imaging), Improve quality (image reconstruction & volume rendering)
Application Type: Medical Imaging / Clinical Diagnostics / R&D Engineering
Target Users:
R&D Engineers (Ultrasound algorithm and hardware teams)
Ultrasound Systems Product Managers
Sonographers and Cardiac Imaging Specialists
Cardiologists and Interventional Radiologists
Hospital Imaging IT and equipment procurement teams
Sources:
https://www.nvidia.com/en-us/customer-stories/philips-healthcare