Nvidia launched Clara earlier in 2018 as a platform to give more facilities access to the GPU horsepower needed to run computationally intensive applications such as artificial intelligence algorithms. Clara enables sites to send data to a centrally located GPU server for processing, rather than have GPU-enabled computers located at every point of data acquisition.
At RSNA 2018, Nvidia is taking the next step in Clara's evolution with the launch of an SDK that will enable third parties to develop apps to run on the platform. The SDK will include accelerated libraries that will enable developers to get acceleration out of the GPU and foster a "develop once, run everywhere" functionality, according to Abdul Hamid Halabi, global business development lead for healthcare and AI at Nvidia.
A major problem in AI and healthcare is the fact that algorithms written on a dataset at one location often don't work as well at another. To address this issue, Nvidia is launching at RSNA 2018 a transfer learning toolkit for medical imaging.
Users of the toolkit can take existing AI algorithms and combine them with a subset of private data from a local hospital or imaging center to "tune" the algorithm to run at that location. The toolkit should be available in early 2019.
Another new launch at RSNA 2018 is a patient annotation SDK that addresses the issue of labeling data in scans, currently a time-consuming effort that can take up to four hours for an image set. The SDK will run as a plug-in to any image viewer, and it will enable users to direct the AI algorithm to a region of interest in the image with just a few clicks.
Finally, Nvidia is announcing partnerships with Ohio State University and the U.S. National Institutes of Health (NIH) on artificial intelligence and medical imaging. Nvidia is working with Ohio State on a marketplace for AI algorithms, and it is working with the NIH to develop algorithms for liver and brain cancer that are designed to measure changes in tumors over time.