Artificial intelligence allows medical devices to do better, be faster, and more. By leveraging NVIDIA Clara Holoscan, ADLINK offers high-speed data streaming, real-time AI inference, and accelerated image reconstruction and graphics to enable software-defined medical devices at the point of care.
Medical devices are mission-critical, time-sensitive applications. This is especially true for medical devices used at the point of care. For example, an intraoral scanner can help orthodontists capture and reconstruct the oral structure in colored detail, a vascular ultrasound can examine blood flow in blood vessels to identify blockage and clots, and gastrointestinal endoscopy takes tissue samples required for diagnosis and treatments.
These applications have strict requirements for low latency, which is measured in milliseconds if not shorter. Latency can occur at different stages of the data processing pipeline. These processes include acquiring sensor data, converting the data to images by running physics-based models, applying AI models to enhance image quality, running AI-assisted detection, and visualizing images. To meet the latency requirement, all aspects of the data processing pipeline must be optimized.
There are also considerations for size, weight, and power (SWaP) factors, because medical devices at the point of care are sometimes moved from one place to another. In addition, medical devices shall not obstruct surgical operation or clinic practices, so a compact size is always appreciated. Low power usage is key to medical devices that are battery-powered, and high power efficiency can yield longer uptime for medical devices that operate on an emergency backup power source during a sudden power outage.
To address latency challenges and reduce size, weight, and power consumption, ADLINK will be manufacturing medical-grade hardware based on the NVIDIA Clara Holoscan platform to bring AI software-based medical devices to production.
The medical platform powered by Clara Holoscan will integrate NVIDIA RTX GPUs with the NVIDIA Ampere architecture and power-efficient Arm Cortex-based embedded computing of the NVIDIA Jetson IGX Orin module. The platform will offer the acceleration needed to reduce latency at the different stages of the data processing pipeline while ensuring a compact footprint and high power efficiency.
- Sensor data processing: NVIDIA GPUDirect RDMA, a Magnum IO technology, through PCI Express cards allows sensor data to be streamed directly to the GPU memory, which reduces downstream processing latency.
- Accelerated image reconstruction and beamforming: Once the data has been transmitted to the GPU, the CUDA cores accelerate physics-based calculations to transform the sensor data into the image domain, reconstructing images in X-ray and CT, and beamforming in ultrasound.
- Real-time AI inference: NVIDIA Tensor Cores enable mixed-precision computing, dynamically adapting calculations to accelerate AI inference for enhanced image quality and AI-assisted detection.
- Image rendering: Tensor Cores also bring AI to graphics with capabilities like NVIDIA Deep Learning Super Sampling (DLSS) and AI denoising. NVIDIA Ampere GPUs provide significant performance gains for graphics workflows and 3D imagery through anti-aliasing techniques, high-dynamic range (HDR) color support, higher refresh rates, and up to 8K screen resolution at 60 Hz from a single cable with the DisplayPort 1.4a standard.
- SWaP optimization: The NVIDIA Ampere architecture’s CUDA cores are up to 2x more power efficient than Turing-based GPUs. In addition, Jetson IGX Orin is a power-efficient system on module, which is an ideal fit for edge applications at the point of care.
ADLINK’s medical platform will have a heterogeneous computing architecture, which enables the platform to leverage different processing cores specializing in workloads while lowering the energy footprint and thermal design power. The heterogeneous computing architecture has been proven successful in delivering an optimal balance among system performance and SWaP constraints for ADLINK’s clients in the healthcare industry.
In ultrasound applications, ADLINK’s medical platform with GPUDirect RDMA enabled shows about an 80 percent boost in data throughput and 60 percent drop in latency on the system level(i). As CPU is no longer needed to move data from a transducer to the GPU, CPU resources can be reserved for other functions.
In endoscopy applications, ADLINK’s medical platform brings down glass-to-glass latency, the amount of time it takes from a single frame of video to transfer from the camera to a display, to 50 milliseconds. ADLINK’s medical platform is also adopted in mobile C-arm X-ray machines, providing state-of-the-art computing in an extremely compact dimension.
All these medical platforms offer long product support, so that medical device companies can develop their applications with assured component availability and offer solutions with a long product life.
With ADLINK’s medical platform based on NVIDIA Clara Holoscan, developers can build their applications to offer surgical video streaming, medical image enhancement, and intraoperative guidance. Taking advantage of the platform’s modular, configurable, and heterogeneous computing architecture, developers will be able to build scalable medical device solutions and accelerate time to bring the latest technology advancement to the point of care. Learn more about ADLINK’s NVIDIA GPU-Based Medical Solutions.
(i) The software and workloads used in performance tests may have been optimized for performance only on ADLINK platforms. Performance tests are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. Contact ADLINK for more complete information about performance and benchmark results.