NVIDIA is continuing its expansion into the healthcare sector, this time partnering with startup Abridge, which builds AI-powered note-taking applications for physicians, to jointly train a specialized healthcare artificial intelligence model.
The two companies stated that this model is specifically designed for clinical communication scenarios and will be exclusively applied on Abridge's platform to optimize tasks such as clinical decision support and medical documentation. NVIDIA is also an investor in Abridge.
Kimberly Powell, Vice President of Healthcare at NVIDIA, explained that this medical model will be trained based on NVIDIA's open-source Nemotron series of models. Open-source models typically allow users to download and modify them for free; fully open-source models also grant users access to all training data and code.
"We now have the opportunity to integrate clinical expertise early in the model development process to specifically optimize existing models," Powell said. Dr. Shiv Rao, co-founder and CEO of the startup, stated that during the collaboration, Abridge will use de-identified clinical data to conduct further training and customized optimization of the Nemotron series models.
Powell noted that this partnership with Abridge is a prime example of NVIDIA's open-source models being deployed in the healthcare and life sciences fields. The chip giant's open-source models can also be customized for various applications including drug discovery, medical devices, and digital health.
She added, "NVIDIA is committed to collaborating with all types of AI companies across the entire industry."
Dr. Shiv Rao, the startup's co-founder and CEO, reiterated that during the partnership, Abridge will leverage de-identified clinical data for additional training and tailored refinement of the Nemotron models.
"General-purpose models are powerful, but clinical expertise requires specialized training, targeted refinement, and evaluation within real-world application contexts. This is the primary focus of our team's current experiments and work with the Nemotron team," Rao said.
Davis Liang, Head of Applied Science at the company, indicated that this new AI model is expected to be officially deployed later this year and will become one of several models supporting the platform's operations. He explained that previously, because third-party models could not adapt to hospital environments and professional medical terminology, Abridge had already trained its own speech recognition model.
Liang candidly stated that cost considerations were also a significant factor in choosing to collaborate based on NVIDIA's open-source models. Lightweight, specialized open-source models are less expensive than proprietary commercial models and can be deployed on Abridge's own hardware infrastructure.
Comments