The National Institutes of Health (NIH) is gearing up to launch artificial intelligence programs to integrate the technology into medical imaging and other use cases, a top agency official said on Sept. 3.
NIH will soon launch a program using AI to conduct precision medicine using imaging scans across departments, Chris Kinsinger, assistant director for catalytic data resources at the NIH Common Fund, shared while speaking at the AFCEA Health IT Summit 2025 in Washington.
“It’s called PRIMED-AI, which is precision medicine with AI with a focus on imaging,” said Kinsinger. “So, we know that there’s a lot of machine learning classifiers for imaging out there – that’s going great, I think industry is doing a fantastic job there.”
“We’re really looking at working across imaging departments within the hospital, we’ve got radiology, pathology, cardiology, they do their own stuff,” continued Kinsinger, saying that NIH is trying to “bring those images together and as well as reports, other types of data, and really do precision medicine with AI models that come out of that.”
Kinsinger said that those models could be either small or large language models or “some kind of picture block management generation.”
Other AI uses that the NIH Common Fund has been looking into include applying the technology to biomedical data analysis which could make data process more efficient.
“I think, like, 85% of time is spent on wrangling the data,” said Kinsinger. “And so if you can just automate that or outsource that to the AI and then really have our PhDs spending time on doing the higher-level thinking, I think that’s an incredible use case we’re looking for.”
Kinsinger said that while “we’re not there yet,” he thinks “that’s coming in the near future.”
In the meantime, he pointed to the need for AI-centric data, saying that data must be trustworthy at its foundation and create clean datasets to train the AI.
“NIH has a lot of great data, but it’s really not ready for AI,” Kinsinger explained, adding that the agency had funded four projects to make flagship data sets that AI models could be trained on.
One of those datasets that NIH is looking at creating takes information from intensive care units which AI could then use to predict adverse events that are deadly in the ICU, such as cardiac arrest, sepsis, and stroke.
“If we got just an early warning from an AI model that has been trained on hundreds of thousands of patients from the ICU, maybe we can save some lives there,” said Kinsinger.
So far, NIH has begun creating a data set based on 100,000 patients across the country and from 14 ICUs to train a generative AI model.