A new AI model called Multimodal AI for ventricular Arrhythmia Risk Stratification (MAARS) is much better than doctors at identifying patients likely to experience cardiac arrest.
The linchpin is the system's ability to analyze long-underused heart imaging, alongside a full spectrum of medical records, to reveal previously hidden information about a patient's heart health.
The federally funded work, led by Johns Hopkins University researchers, could save many lives and also spare many people unnecessary medical interventions, including the implantation of unneeded defibrillators.
"Currently we have patients dying in the prime of their life because they aren't protected and others who are putting up with defibrillators for the rest of their lives with no benefit," said Senior Author Natalia Trayanova, a researcher focused on using artificial intelligence in cardiology. "We have the ability to predict with very high accuracy whether a patient is at very high risk for sudden cardiac death or not."
The findings are published today in Nature Cardiovascular Research.
Here is an exclusive Tech Briefs interview, edited for length and clarity, with Trayanova.
Tech Briefs: What was the biggest technical challenge you faced while developing MAARS?
Trayanova: It is a multimodal AI, and one of the most important features of the multimodal AI predictor is that in one of the channels it takes raw signal intensity images. Typically, when an arrhythmia assessment or other predictive application has been done in cardiology , the data is taken from an MRI via a radiologist: What is the chamber volume? How much scar is there? Or something like that.
Here we have the raw signal intensity of the image, which has been fed into one of the data inputs for the predictor. That is, in itself, pretty complex.
This is a very new and very novel development, so it took us some time to develop it. But it is very important because, mechanistically, we know that arrhythmias occur when you have scarring in the heart. And because these patients have a lot of scarring, it's reflected in the distribution of scar, or fibrosis, in the heart, which you see on the MRI. So, if you just say, “Well, this patient has 10 percent and that patient has 12 percent,” it can be completely different. The one who has 12 may not have an arrhythmia and the one who has 10 might have it. That could be because the distributions and the patterns of the scarring can be very different.
Tech Briefs: The article I read says, “The team plans to further test the new model on more patients and expand the new algorithm to use with other types of heart disease, including sarcoidosis.” Do you have any plans for further research?
Trayanova: We are working on that. While this was being published, we did research on that. So, with this disease it is in general hard to predict which patient is at risk with severe cardiac death. What we are trying to do is predict risk for patients with hypertrophic cardiomyopathy, so they're better protected. And we’re trying to do the same thing with sarcoidosis, a disease for which it is very difficult to predict who is at risk. What we are trying to do, is work with a population in which it is exceptionally difficult to stratify patients fromlow to high risk so they can have some protective measures. Sarcoidosis is one of those diseases, however, it is more of an inflammatory disease rather than a genetic disease.
Tech Briefs: Do you have any updates you could share?
Trayanova: All I can say is that the paper is submitted.
Tech Briefs: Do you have any plans for even further research? Where you go from here?
Trayanova: What we would like to do is to expand external validation. It is a little bit complicated because these diseases are generally only seen in specialized centers. I would like to pull more data together, which requires sharing data from different healthcare centers, which is a really complex task. For example, it involves, data use agreements and so forth. It is a challenge, but we would like to have more data.
The other route is we would like really to have some sort of a deployment in our hospital; we are working toward that. If it starts being used in our clinic, I think that more centers will follow. On the other hand, we made the software available open source, so if anyone is interested, they can just download it and apply it in their healthcare center rather than sharing data with us.
Tech Briefs: Doc, those are all the questions I have. Is there anything else you'd like to add that I didn't touch upon?
Trayanova: AI is making inroads in healthcare. It’s just not as fast as we would like, but there is a reason for that. So, we are doing what we can step-by-step.
