Nvidia is building a supercomputer specifically designed to help with the increasingly computationally complex area that is medical imaging.
Unveiled at Nvidia’s annual GPU Technology Conference in San Jose, Clara—named after the Californian city where the company is headquartered, Santa Clara—was labelled by Nvidia founder and CEO Jensen Huang as “a data center-virtualized, multimodality, multi-user, medical computational medical instrument”.
With around 3 million instruments installed in hospitals throughout the world, and 100,000 new ones installed each year, Huang predicts it will take about 30 years before the installed base is replaced with upgraded equipment.
According to Huang, supercomputing is now a fundamental pillar of science.
“We need larger computers; even then we need larger computers; the world needs larger computers, because there’s serious work to be done, there’s serious ground-breaking work to be done,” he said.
Speaking with ZDNet about the future a machine like Clara promises, Nvidia VP of healthcare and AI business development Kimberly Powell said they have been focusing on healthcare for over 10 years, alongside the medical imaging industry, which has been driven by computing for at least as long.
“If you think about what medical imaging does, it does computation, it does processing, digitalization, and artificial intelligence,” she said.
As Huang detailed, Nvidia started by performing image reconstruction inside instruments and serving up the volume rendering and image processing required for the interpretation of such images.
For the past five years, Nvidia has also worked with the research and startup communities exploring the future of AI from a healthcare lens.
While Clara is still being built by Nvidia, what the supercomputer will be capable of is no pipedream.
At its most basic, Clara is a computing platform that allows the virtualization of workloads—a virtualized data center for all the computation that happens in medical imaging.
According to Powell, the dream for Nvidia is to provide Clara to regions where radiologists or specialists, for example, aren’t as common, or where the high-level training that is required to become a radiologist isn’t as accessible as it is in many Western regions.
Pointing to AlphaGo, Powell said that upon reflection, in the first three games the computer beat the human—but in the fourth game, the human beat the computer because he was taught by the computer something he didn’t always know how to do.
Discussing some potential use cases, Powell said she came across a case recently that saw 23,000 unread X-rays in the UK; similarly, she said in Japan there’s a very “disorienting number” of instruments and medical exams that are performed when placed in a ratio against the number of practicing radiologists.
“This, I think, is an opportunity to balance some of that out,” she said.
Powell said being able to sequence genomes requires a lot of data and computational power to process such data, and that the data is only growing thanks to the increasing cost efficiency of genomics.
Powell expects more of Clara to be unveiled at the Radiological Society of North America conference in November.