AI Model Helps Clinicians Detect Severe Coronary Artery Disease in Stress Echocardiograms
Posted on 20 Dec 2021
A breakthrough study has shown that artificial intelligence (AI) is capable of independently reading stress echocardiograms and predicting coronary artery disease (CAD), significantly increasing detection compared to traditional methods.
The new study found that Ultromics Limited’s (Oxford, UK) AI-powered EchoGo platform significantly improved the accuracy and confidence in CAD detection from a stress echocardiogram. The study revealed that doctors are more confident in CAD diagnoses with the support of EchoGo. Further, the study revealed that the EchoGo AI improved the sensitivity of detection of the disease by over 10%, which brings the effectiveness of stress echocardiography - as a modality - on par with much more expensive and invasive diagnostic procedures.
Ultromics has developed an AI algorithm that automatically analyzes stress echocardiograms and tells the clinician whether that patient is at risk of coronary artery disease. The study assessed Ultromics’ deep learning algorithms trained on hundreds of thousands of data to predict CAD risk, and was able to distinguish the CAD risk in patients with 10% more sensitivity than manual analysis. During the first portion of the study, Ultromics AI was trained and tested to identify patients with severe coronary artery disease. Patients without CAD were correctly identified by AI 92.7% of the time and patients with CAD were correctly identified 84.4% of the time. The result showed a 10% increase in accuracy compared to the human eye.
The second part of the study sought to explore how an AI classification might impact physicians’ and echocardiographers’ clinical interpretation of stress echocardiograms. Four physicians/echocardiographers tested Ultromics’ AI and were asked to review heart ultrasounds to determine whether patients had CAD. Each physician / echocardiographer was shown each heart ultrasound two times, one month apart – once with the AI’s decision and once without – and were asked to determine whether they were confident in their clinical interpretation. When provided with the AI-generated classifications, the level of agreement between all four physicians/echocardiographers increased 15% or more - from between 68% and 79% accuracy, to between 83% and 97% accuracy.
“Using AI to make CAD predictions is a real breakthrough for science, it will help spot earlier signs of a problem, prevent unnecessary tests and procedures, and help save more lives”. said Dr. Ross Upton, a study author and the CEO and co-founder of Ultromics. “Traditionally, CAD analysis is done through manual observations. However, the complexity of identifying subtle features within an image means that it is not always easy to correctly identify CAD on first visit, and accurate diagnoses can be subject to the operator’s expertise. This means heart disease, the number one global killer, gets missed one in five times. With the support of our AI service, clinicians can make confident diagnoses and spot features which would have been impossible to detect using traditional methods.”
“Echocardiography is such a critical modality for diagnosing cardiovascular disease”, added Professor Paul Leeson, Professor of Cardiovascular Medicine at the University of Oxford and the co-founder of Ultromics. “It is affordable, accessible, portable, comfortable, non-invasive, and doesn’t emit radiation. The fact that we can now apply artificial intelligence to echocardiography to perform at this level is truly revolutionary for patient care.”
Related Links:
Ultromics Limited
