NVIDIA supercomputer keeps medical imaging systems up to date

NVIDIA’s Project Clara, a medical imaging supercomputer, is renewing the capabilities of some 3 million ageing medical imaging instruments, in place.

Unveiled at the NVIDIA GPU Technology Conference, Project Clara makes use of NVIDIA GPUs to help reduce radiation exposure, improve medical image quality and produce images in real time.

Computational game-changers like computer tomography (CT) iterative reconstruction and magnetic resonance (MR) compressed sensing are reducing radiation exposure by up to 90% and shortening the time it takes for an MR image to be captured. The next level up for medical advanced image analysis and quantification involves artificial intelligence (AI) and deep learning.

A recent algorithm called V-Net uses a neural network to compute 3D volumetric segmentation and can automatically measure the volume of blood flowing through the heart, for example. Fifteen years ago, this algorithm would have needed a computer that cost US$10 million and consumed 500 kW of power per hour. Today, the same code can run on a handful of NVIDIA Tesla V100 GPUs.

Building on 10 years in medical imaging and work with development partners, NVIDIA is redefining medical imaging. Through NVIDIA virtual GPU (vGPU) technology, which shares the power of NVIDIA GPUs across virtual machines (VMs) and virtual applications, Clara can act as the brain of existing computational instruments, serving multiple users simultaneously. Clara is also universal, handling the computation for any instrument, whether it is hardware for CT, MR, ultrasound, X-ray or mammography. It may be 3 million today, but many more tomorrow - Clara uses Kubernetes on GPUs to efficiently scale compute with demand.

Dozens of healthcare companies, startups and research hospitals are already working with Project Clara, creating AI applications like V-Net, NVIDIA said. "Modern medical imaging applications demand new levels of computing, scale and accessibility. Clara is our computing platform to revolutionise medical imaging," the company said in a blog post.

Source: NVIDIA blog. This 3D ultrasound depicts the left ventricle of the heart segmented by V-Net, a fully convolutional 3D neural network, running on a Tesla V100 GPU.