New AI Tool Detects Heart Disease
A recently developed AI tool is making waves in the field of cardiology by detecting structural heart disease (SHD) before it escalates into a more severe condition. SHD includes various defects within the heart’s valves, walls, or chambers, which can either be congenital or develop over time, leading to complications in blood circulation.
This type of heart disease often remains “hidden.” That is, it can progress without noticeable symptoms until a serious event like a heart attack or stroke occurs.
Researchers at Columbia University and New York Presbyterian are at the forefront of this innovation, unveiling a screening tool that identifies individuals who might need a crucial ultrasound to diagnose structural heart issues. Dr. Pierre Elias, a cardiologist and AI specialist, emphasizes the growing role of AI in medical diagnostics. “We’ve seen a surge in AI models capable of detecting diseases and facilitating opportunistic screenings,” he said.
He further explains, “While many models currently exist for diseases like coronary artery disease through CT scans or mammograms, Echonext stands out as the first model designed to detect structural heart diseases using any form of ECG.” An electrocardiogram, or ECG, is a quick and non-invasive test that measures the electrical activity of the heart.
ECGs are among the most commonly performed cardiac tests and are typically ordered for patients showing symptoms such as shortness of breath or chest pain. However, while valuable, ECGs alone aren’t reliable for identifying SHD.
This is where Echonext comes into play. Developed over the past four years, this tool analyzes ECG data to ascertain whether further echocardiograms are necessary. An echocardiogram is another form of imaging used to diagnose various heart conditions, including congenital defects and valve disorders.
“Echonext leverages inexpensive tests to flag who requires more costly ultrasounds,” Elias adds. “We believe that integrating ECG with AI could revolutionize the screening process.” The tool was trained using a substantial dataset of over 1.2 million ECG-echocardiogram pairs from around 230,000 patients.
In evaluations, Echonext outperformed a group of thirteen cardiologists, who successfully detected 77% of structural heart issues in a set of 3,200 ECGs with a recorded accuracy of 64%.
Through this research, Echonext flagged over 7,500 individuals from nearly 85,000 study participants as having a higher likelihood of previously undiagnosed SHD. The researchers monitored these patients for a year without informing their doctors about the predictions.
About 55% of these flagged individuals had their first echocardiogram, and nearly 75% were diagnosed with SHD, significantly higher than typical diagnosis rates. The study’s findings were outlined in a recent issue of the Journal Nature.
“Our goal is straightforward: to connect the right patients with the right specialists sooner rather than later,” Elias explained. “The unfortunate reality is that many patients who should see a cardiologist go unnoticed, and Echonext aims to change that by guiding them to needed treatments.”
Columbia is already looking ahead, having submitted a patent application for the Echonext ECG algorithm, and clinical trials are continuing in multiple emergency departments.
