April 26, 2022 – Deaths from COVID-19 may have caught more attention lately, but heart disease remains the leading cause of death in the U.S.
More than 300,000 Americans will die this year of sudden cardiac arrest (also called sudden cardiac death, or SCD), when the heart abruptly stops working.
These events happen suddenly and often without warning, making them nearly impossible to predict. But that may be changing, thanks to 3D imaging and artificial intelligence (AI) technology under study at Johns Hopkins University.
There, researchers are working to create more accurate and personalized models of the heart – and not just any heart, your heart, if you have heart disease.
“Right now, a clinician can only say whether a patient is at risk or not at risk for sudden death,” says Dan Popescu, PhD, a Johns Hopkins research scientist and first author of a new study on AI’s ability to predict sudden cardiac arrest. “With this new technology, you can have much more nuanced predictions of probability of an event over time.”
Put another way: With AI, clinicians may be able not only to predict if someone is at risk for sudden cardiac arrest, but also when it is most likely to happen. They can do this using a much clearer and more personalized look at the electrical “wiring” of your heart.
Your Heart, the Conductor
Your heart isn’t just a metronome responsible for keeping a steady stream of blood pumping to tissues with every beat. It’s also a conductor through which vital energy flows.
To make the heart beat, electrical impulses flow from the top to the bottom of the organ. Healthy heart cells relay this electricity seamlessly. But in a heart damaged by inflammation or a past heart attack, scar tissue will block the energy flow.
When an electrical impulse encounters a scarred area, the signal can become erratic, disrupting the set top-to-bottom path and causing irregular heartbeats (arrhythmias), which increase someone’s danger of sudden cardiac death.
Seeing the Heart in 3D
Today’s tests offer some insights into the heart’s makeup. For example, MRI scans can reveal damaged areas. PET scans can show inflammation. And EKGs can record the heart’s electrical signals from beat to beat.
But all these technologies offer only a snapshot, showing heart health at a moment in time. They can’t predict the future. That’s why scientists at Johns Hopkins are going further to develop 3D digital replicas of a person’s heart, known as computational heart models.
Computational models are computer-simulated replicas that combine mathematics, physics, and computer science. These models have been around for a long time and are used in many fields, ranging from manufacturing to economics.
In heart medicine, these models are populated with digital “cells,” which imitate living cells and can be programmed with different electrical properties, depending on whether they are healthy or diseased.
“Currently available imaging and testing (MRIs, PETs, EKGs) give some representation of the scarring, but you cannot translate that to what is going to happen over time,” says Natalia Trayanova, PhD, of the Johns Hopkins Department of Biomedical Engineering.
“With computational heart models, we create a dynamic digital image of the heart. We can then give the digital image an electrical stimulus and assess how the heart is able to respond. Then you can better predict what is going to happen.”
The computerized 3D models also mean better, more accurate treatment for heart conditions.
For example, a common treatment for a type of arrhythmia known as atrial fibrillation is ablation, or burning some heart tissue. Ablation stops the erratic electrical impulses causing the arrhythmia, but it can also damage otherwise healthy heart cells.
A personalized computational heart model could allow doctors to see more accurately what areas should and shouldn’t be treated for a specific patient.
Using Deep Learning AI to Predict Health Outcomes
Trayanova’s colleague Popescu is applying deep learning and AI to do more with computerized heart models to predict the future.
In a recent paper in Nature Cardiovascular Research, the research team showed their algorithm assessed the health of 269 patients and was able to predict the chance of sudden cardiac arrest up to 10 years in advance.
“This is really the first time ever, as far as we know, where deep learning technology has been proven to analyze scarring of the heart in a successful way,” Popescu says.
Popescu and Trayanova say the AI algorithm gathers information from the 3D computational heart models with patient data like MRIs, ethnicity, age, lifestyle, and other clinical information. Analyzing all this data can produce accurate and consistent estimates about how long patients might live if they are at risk for sudden death.
“You can’t afford to be wrong. If you are wrong, you can actually impact a patient’s quality of life dramatically,” Popescu says. “Having clinicians use this technology in the decision-making process will provide confidence in a better diagnosis and prognosis.”
While the current study was specifically about patients with a particular type of heart disease, Popescu says his algorithm can also be trained to assess other health conditions.
So when might you see this being used outside of a research study? Trayanova predicts 3D imaging of heart models could be available in 2 years, but first the technique must be tested in more clinical trials – some of which are happening right now.
Adding AI to the heart models will require more studies and FDA approval, so the timeline is less clear. But perhaps the biggest hurdle is that after approval, the technologies would need to be adopted and used by clinicians and caregivers.
“The much harder question to answer is, ‘When will doctors be perfectly comfortable with AI tools?’ And I don’t know the answer,” Popescu says. “How to use AI as an aid in the decision-making process is something that’s not currently taught.”