28 Mar 2018 Silicon Valley - There are about 3 million medical imaging instruments installed around the world. With only a couple hundred thousand new ones sold each year, it would take decades to update this install base. NVIDIA's Project Clara, a medical imaging supercomputer, renews the capabilities of these machines in place. Unveiled at the GPU Technology Conference in Silicon Valley, Project Clara takes advantage of incredible advancements in computation.
Medical imaging instruments have been vital to early detection and improvement of patient outcomes for more than four decades. Innovation in the field has come from improvements in detector technology and, more recently, parallel computing.
A decade ago, researchers realized NVIDIA GPUs provide the most efficient architecture for medical imaging applications and could help reduce radiation exposure, improve image quality and produce images in real time. More recently, deep learning is dominating, with more than half of new research in medical imaging applications involving AI.
Computational game-changers like CT iterative reconstruction and MR compressed sensing are reducing radiation exposure up to 90 percent and shortening the time it takes for an MRI image to be captured.
Deep learning and AI are generating exciting opportunities for advanced image analysis and quantification. A recent algorithm called V-Net uses 3D volumetric segmentation and can automatically measure the volume of blood flowing through the heart. Fifteen years ago, this algorithm would have needed a computer that cost $10 million and consumed 500 kW of power. Today, it can run on a few Tesla V100 GPUs.
Building on 10 years in medical imaging and work with NVIDIA's development partners, the company sees the potential to re-imagine how computing can improve medical imaging.
Clara is virtual: it can run many computational instruments simultaneously. Clara is remote: it leverages NVIDIA vGPUs to enable multi-user access. Clara is universal: it can perform the computation for any instrument, whether CT, MR, ultrasound, X-ray or Mammography. And Clara is scalable: it uses Kubernetes on GPUs to efficiently scale compute with demand.
Working with NVIDIA are dozens of health care companies, start-ups and research hospitals. Their AI applications like AutoMap and V-Net bring intangible value to radiology.
AutoMap, from the MGH Martinos centre, can shorten acquisition time of MRI and boosts image quality. V-Net can automatically measure anatomy and assess functionality. Cinematic rendering pioneered by Elliot Fishman at Johns Hopkins University brings a new level of quality, ultimately saving time for radiologists and improving patient outcomes.
Subtle Medical, which is working on dozens of applications in medical imaging, recently won over more than a quarter of a million dollars in the healthcare category in NVIDIA's Inception programme awards.
"New technologies are transforming health care", stated Dr. Greg Zaharchuk, founder of Subtle Medical, radiologist and associate professor in Radiology at Stanford. "NVIDIA's vision for a virtualized imaging supercomputer is an exciting new chapter that will revolutionize our ability to deliver AI-powered health care."
Modern medical imaging applications demand new levels of computing, scale and accessibility. Clara is NVIDIA's computing platform to revolutionize medical imaging.