Only a few months after it emerged from stealth and burst onto the medtech scene, Cleerly is already living up to its name.
The newly published results of a study show that the New York City-based startup’s artificial intelligence-powered software does, indeed, allow healthcare providers to (ahem) clearly spot signs of heart disease without requiring the invasive evaluations that make up the current gold standard for coronary artery disease diagnosis.
In the latter method, the coronary angiogram, a catheter is threaded through the arm or leg to reach the heart, where it injects dye into its arteries so X-ray images can capture how blood flows through the vessels. Though serious complications from the procedure are rare, they potentially include injury to the catheterized artery, an allergic reaction or kidney damage from the dye, as well as heart attack, stroke and more.
RELATED: Cleerly launches with $43M to tackle heart attacks with AI-powered imaging
Cleerly’s AI, meanwhile, analyzes noninvasive coronary CT scans to measure the amount of plaque build-up in the coronary arteries. Its aim is to detect heart disease before symptoms arise—especially since the first symptom for more than half of coronary artery disease patients is a heart attack, according to the company.
The FDA-cleared technology comprises more than two dozen AI and machine learning algorithms developed on research conducted at Weill Cornell Medicine and New York-Presbyterian Hospital. The algorithms were trained to recognize signs of heart disease using data from large-scale coronary imaging studies, including millions of annotated lab images.
RELATED: Canon Medical thinks Cleerly to combine AI imaging analysis with cardiac CT
In the study, led by researchers from the George Washington University School of Medicine, the AI’s analysis went head-to-head with angiography—or QCA, for quantitative coronary angiography—in detecting and grading the severity of a narrowed coronary artery.
The AI produced analyses of the blockages with 94% sensitivity and between 68% and 82% specificity, depending on the severity of stenosis. In total, the AI’s assessments resulted in accuracy levels of at least 84% and showed a strong correlation with the QCA’s on a per-vessel and per-patient basis.
Additionally, when the AI and angiogram significantly disagreed, the cases were settled using invasive fractional flow reserve, or FFR, where a sensor-tipped catheter is inserted into the arteries to poll blood flow.
In more than two-thirds of those cases, the FFR analyses agreed with the AI, rather than the angiogram, suggesting that “Cleerly analysis may have performed better than the QCA gold standard used,” according to the company, and indicating that the AI could be used to help cut down on the number of unnecessary catheterizations that are performed.