A new set of AI models released by the U.S. National Institutes of Health (NIH) in collaboration with Nvidia is designed to help clinicians detect and examine novel coronavirus infections using chest CT scan data.
The AI models are part of a package of tools and partnerships announced this week by the GPU leader as part of its expanding Clara healthcare platform. Nvidia also rolled out a genomic pipeline running its new Ampere A100 GPU along with new automated patient monitoring and other telemedicine tools.
Data quality remains an issue as public health researchers scramble to understand mechanisms and the spread of COVID-19. Nvidia said it worked with NIH clinicians and data scientists to generate the AI models via a cooperative R&D agreement.
Built in less than three weeks, the AI models use infection data from China, Italy, Japan and the U.S. They were built using the Clara application framework for medical imaging that includes training and deployment workflow tools.
Nvidia announced an expansion of its Clara AI platform last December, billing it as a “federated” learning tool for radiologists. The goal is to promote collaboration while protecting patient privacy. Clara federal learning runs on Nvidia’s EGX edge platform.
Building robust AI models has been hampered by limits on sharing patient data across borders. Nvidia said it is working with Mass General Brigham (a partnership formed last year to combine the former Massachusetts General Hospital and Brigham and Women’s Hospital) to expand COVID-19 AI models to x-ray imaging.
The AI models are available now in the latest release of Clara imaging on Nvidia’s software hub.
Public health officials struggling to stay keep up with the pandemic stress the importance of speed in areas like COVID-19 testing and contact tracing. With that in mind, Nvidia also announced this week it has achieved record DNA sequencing speed. The accelerated genomics analysis promises to generate high-quality data sets that could be used for applications such as diagnosing rare childhood diseases.
Among those pediatric diseases is an emerging complication from COVID-19 that resembles Kawasaki disease.
The accelerated sequencing was accomplished using a combination of Nvidia’s Parabricks computational genomics software and its new A100 GPUs. DNA sequencing analysis for an entire genome was reduced to under 20 minutes, the chipmaker said Thursday (May 14). The GPU-accelerated, RNA-sequencing pipeline is touted as returning results in less than two hours, giving researchers quick insights into patient susceptibility, disease progression and response to treatment.
Nvidia said it is partnering with the DNAnexus genomics cloud platform to deliver DNA sequencing results. The cloud provider is offering license-free access to test and validate Clara Parabricks pipelines.
As COVID-19 exposes doctors, nurses and other healthcare workers to growing risks of infection, telemedicine and other remote diagnostic tools have grown in importance. In response, Nvidia also this week unveiled a suite of video analytics and automated speech recognition technologies. They are designed to help overworked hospital staffers limit their exposure while performing tasks like vital sign monitoring.
The Clara Guardian suite is deployed on Nvidia’s EGX edge platform with the goal of enabling providers to deploy and remotely manage fleets of medical devices. Guardian’s software stack incorporates Nvidia’s DeepStream speech recognition and natural language processing tools designed for video analytics apps and services. It runs on the company’s Jarvis application framework for conversational AI services.
“The COVID-19 pandemic has supercharged the collaboration of technology, research and the healthcare industry to develop new computing solutions that accelerate the understanding of the spread, scale and severity of the disease,” said Kimberly Powell, Nvidia’s vice president of healthcare.