News Categories

NVIDIA advances healthcare research with new edge AI platform and large language models

By Liu Hongzuo - on 21 Sep 2022, 3:58am

NVIDIA advances healthcare research with new edge AI platform and large language models

Large Language Model for Biology by NVIDIA BioNeMo.

At NVIDIA’s GTC keynote, the company introduced a host of new AI-related advances for the healthcare industry, with the final parts of the 90-minute presentation highlighting its healthcare-related tools and technologies.


NVIDIA IGX edge AI platform

Medical devices developed using Clara Holoscan with IGX.

Newly introduced is the NVIDIA IGX platform for medical edge AI use cases. IGX is a combination of both hardware and software components. It delivers AI inference to medical devices, granting critical instant insights across various medical requirements (devices, sensors, and even robotic-assisted surgery or patient monitoring).

What’s included on an NVIDIA IGX is an Orin Module (250 TOPS), the ConnectX-7 with 400GbE streaming input/output, an RTX Ampere GPU (up to 600 TOPS), and a functional safety island and safety microcontroller unit.

The IGX platform is compatible with NVIDIA Clara Holoscan. The latter is an application framework that offers real-time AI computing platform for medical devices, processing streamed data and visualisation at low latencies. 

Three medical device startups already use NVIDIA Clara Holoscan on the new IGX platform for their robotic surgical systems. 

Activ Surgical, for example, uses this combination for surgical guidance – surgeons can view physiological structures and functions like blood flow, which the naked eye cannot see.

Moon Surgical used NVIDIA Clara Holoscan and the IGX platform to design Maestro, an accessible, adaptive surgical-assistant robotics system that works with existing medical equipment and workflows. 

Finally, Proximie uses Clara Holoscan to provide local video processing in the operation room (for remote surgery). This implementation helps to cloud computing costs, while maintaining data privacy.

You can learn more about NVIDIA IGX here.


NVIDIA BioNeMo framework for training LLMs

NVIDIA BioNeMo is an AI-powered drug discovery cloud service and framework for training and deploying biomolecular transformer AI models at a supercomputing scale. It’s an extension of the NVIDIA NeMo Megatron framework.

NVIDIA BioNeMo is both a framework and a service. As a framework, BioNeMo can help researchers develop pre-trained large language models (LLMs) at any scale and with any biological sequence (chemistry, protein, DNA, and RNA). As an LLM cloud service, it grants researchers access to pre-trained chemistry and biology language models. 

The output from BioNeMo lets researchers generate new proteins and chemicals, or allow them to predict structure, function, and reaction properties.

NVIDIA BioNeMo will be available in early October 2022.


Broad Institute partnership

NVIDIA has partnered with the Broad Institute to avail NVIDIA’s AI and acceleration tools on Broad’s Terra Cloud platform.

This collaboration connects NVIDIA’s AI expertise to Broad’s researchers and scientists, allowing them to analyse massive amounts of healthcare data quickly.

One example would be NVIDIA Clara Parabricks, now available in six new Terra workflows. Parabricks is a GPU-accelerated software suite for secondary analysis of sequencing data. Users can now analyse a whole genome sequence in a little more than one hour, much faster than the original 24 hours in a CPU-based environment, while reducing compute costs by more than half.

The researchers will also be able to use NVIDIA BioNeMo to develop foundational models for DNA and RNA.

Finally, NVIDIA is also contributing a new deep learning model directly to Broad Institute’s Genome Analysis Toolkit (GATK), an industry-standard toolkit for identifying genetic variants associated with diseases (which subsequently leads to drug discovery and new therapies).

Source: NVIDIA (blog)

Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.