Artificial intelligence making its way into hospitals

University of Saskatchewan researchers are developing artificial intelligence to predict heart disease.

Dr. Scott Adam, a cardiothoracic radiologist at Royal University Hospital in Saskatoon and an assistant professor at the university, spoke with Gormley about AI’s great potential for clinicians.

“There are some AI models and deep-learning models which have become extraordinarily capable of analyzing medical images that we acquire in daily practice,” he said.

“There are so many features that can be extracted from medical images that are really beyond physicians’ perception.”

With the AI, physicians are able to extract and classify those features.

“We’re able to use those features to develop clinical insight that can be used to stratify a patient’s cardiovascular risk and subsequently inform treatment planning,” said Adam.

There’s a number of ways to look at a patient’s cardiovascular risk factor. Currently, doctors use high-risk models that include demographic information and biochemical markers such as cholesterol.

With AI, it can look at biochemical markers that humans are unable to detect. It will be able to look at the entire 3D data set from CT images to get a better idea of patients’ cardiovascular risk.

These multi-modal AI models are able to bring together multiple layers of data into one singular model.

Adam said the overlaying of these different identifiers like chemical markers, age factors and imaging and putting genetics into that mix would make it a very strong diagnostic tool.

The AI is also being used to measure biological age rather than chronological age.

Adam said this AI is not ready to be used just yet, but its certainly something they will look at in the future.

“We’re taking steps to improve our models in each of those domains — improving AI models for genetic analysis (and) improving models for imaging analysis, for example,” he said. “Over the next few years, I really do think that multi-modal models will become the norm and, with careful testing and validation, will be translated into the clinic.”

As for a timeline, researchers are still in the research phase and need more clinical trials, not just in Saskatchewan or Canada but throughout the world with international trials.

According to Adam, many AI models already have Health Canada and FDA approval. But they’re still quite some time away from creating an accurate multi-modal AI algorithm.

Research and trials will be done on the general population as a whole, not just those at higher risk.

“If you just begin with those at highest risk, there’s a potential problem of getting biased data sets,” said Adam.

He said there have been a number of population-based studies done that researchers can draw upon and use those publicly available data sets for AI algorithm development.

But he does acknowledge that those algorithms will be used more often on those higher-risk populations.

“Ensuring an accurate prediction model for those populations is absolutely critical,” said Adam.