A new deep neural network, dubbed CHAIS, could soon become the new gold standard for monitoring heart health without the use of invasive procedures like catheterization.
The ancient Greek philosopher and polymath Aristotle, for example, once deducted that the human heart is tri-chambered, and that it was the one and only most important organ in the entire body, governing motion, sensation and thought.
Today we know the human heart has four chambers and that the brain controls movement, sensitivity and thought. But Aristotle was right that the heart is a vital organ that pumps the blood to the rest of the body so that it can reach other vital organs. As a life-threatening condition such as heart failure takes hold, the heart gradually becomes unable to pump blood and nutrients to other organs, depriving them of the sustenance that allows them to do their jobs.
The new tool, developed by researchers from MIT and Harvard Medical School and described in an open-access paper they published recently in Nature Communications Medicine, is a noninvasive deep learning model that analyzes electrocardiogram (ECG) signals in order to predict, with accuracy, a patient’s risk of heart failure. In a clinical trial, the model achieved accuracy on par with gold-standard but more-invasive procedures, offering hope to those at risk of heart failure. The condition has recently experienced an alarming increase in mortality, especially among young adults, that is likely attributable to the growing epidemic of obesity and diabetes.
“Parts of this paper are a distillation of things I’ve said in various other places over the course of years,” says paper senior author Collin Stultz, director of Harvard-MIT Program in Health Sciences and Technology and an affiliate of the MIT Abdul Latif Jameel Clinic for Machine Learning in Health (Jameel Clinic). “The purpose of this work is to find people who are starting to get sick even before they have symptoms so you can intervene early enough to prevent hospitalization.”
The four chambers of the heart consist of two atria and two ventricles — one of each type on the right side of the heart and one of each on the left side. In a healthy human heart, the chambers work in rhythmic synchrony: oxygen-poor blood enters the heart through the right atrium. The right atrium contracts creating the pressure to push the blood into the right ventricle and from here the blood is pumped into the lungs for oxygenation. Oxygen-filled blood from the lungs then flows into the left atrium, which contracts to push the blood into the left ventricle. There is another contraction, and blood is pumped through the left ventricle and out of the heart via the aorta into the veins, which branch throughout the rest of the body.
“The left atrial pressures may become elevated, and when that happens the blood drain from the lungs into the left atrium gets impeded because it’s a higher-pressure system,” says Stultz. Stultz, a professor of electrical engineering and computer science and a practicing cardiologist at Mass General Hospital (MGH), designed the wearable version of the platform. “The higher that pressure in the left atrium, the more pulmonary symptoms you develop — dyspnea, shortness of breath, things like that. “The right side of the heart sends blood through the pulmonary vasculature to the lungs, so the higher pressures in the left atrium also translate into higher pressures in the pulmonary vasculature.”
Currently, right heart catheterization (RHC), an invasive procedure requiring insertion of a thin reader tube (the catheter) connected to a pressure transmitter into the right heart and pulmonary arteries, is the gold-standard method for assessing left atrial pressure. Instead, commenting in writing, physicians often prefer noninvasive assessment of risk to RHC, through examination of weight, blood pressure, and heart rate.
But as Stultz sees it, these measures are coarse, given that one-in-four heart failure patients winds up re-hospitalized within 30 days. “What we’re looking for is something that provides you information the same way an invasive device would, though, short of a simple weight scale,” Stultz says.
To obtain a fuller picture of a patient’s heart problem, physicians usually perform a 12-lead ECG, in which 10 adhesive patches are applied to the patient and connected to a machine that issues data from 12 different angles about the heart. But 12-lead ECG machines are only available in clinical environments and are also not commonly used to evaluate the risk of heart failure.
What Stultz and other researchers suggest instead is a Cardiac Hemodynamic AI monitoring System or CHAIS, a deep neural network that can learn to read ECG data from a single lead — or in other words, they need to have just a single adhesive, commercially-available patch on their chest that they can wear outside of the hospital, untethered to a machine.
In order to compare CHAIS to the current gold standard, RHC, the researchers enrolled patients with imminent catheterization and asked them to wear the patch for 24 to 48 hours prior to the procedure (but to remove previous to the catheterization). “When you get to an hour-and-a-half [before the procedure], it’s 0.875, so it’s very, very good,” says Stultz. “So a measure from the device is just as good and tells you the same thing as your being cathed the next hour-and-a-half.”
“Every cardiologist knows the importance of measuring left atrial pressure for defining cardiac function and for guiding treatment strategies in heart failure,” says Aaron Aguirre SM ’03, PhD ’08, who works as a cardiologist and critical care physician at MGH. “This work has significance as it provides a noninvasive means of estimating this critical clinical parameter utilizing widely available cardiac monitoring.”
Aguirre, who earned a PhD in medical engineering and medical physics at MIT, anticipates that with additional clinical validation CHAIS will be beneficial to two main areas: first, it will help identify patients who will benefit the most from the more invasive cardiac test via RHC; and second, as CHAIS technology could enable serial monitoring and tracking of left atrial pressure in patients with heart disease. “A noninvasive and quantitative method may aid in optimizing treatment strategies both in patients at home or in hospital,” Aguirre says. “I look forward to seeing where this goes from here with the MIT team.”
But it is not just patients who reap the benefits — for patients with difficult-to-manage heart failure, it becomes a challenge to prevent them from being readmitted to the hospital without a permanent implant, occupying more space and more time of an already beleaguered and understaffed medical workforce.
The researchers also have another active clinical study using CHAIS with MGH and Boston Medical Center that they want to wrap up soon so they can start analyzing the data.
“In my opinion, the real promise of AI in health care is to provide equitable, world-class care to everyone, regardless of socioeconomic status, background, and where they live,” Stultz says. “This work is one step towards achieving this goal.”
