New study shows AI performs as well as expert echocardiographers at recognizing abnormal heart wall movement
Millions of patients rely on cardiac ultrasound imaging for an initial assessment of a suspected heart condition before being referred for further examination. However, assessing a sequence of ultrasound images to detect abnormal movement of the heart walls – known as regional wall motion abnormalities (RWMA) – requires substantial experience. This process can also be time-consuming and often subjective in nature, potentially leading to significant user-to-user variability, especially when attempting to make quantitative assessments.
Internationally renowned cardiologist Dr. Roberto M. Lang and his team at the University of Chicago Medical Center (Chicago, US) investigated whether artificial intelligence (AI) could consistently identify RWMA and perform as well as expert echocardiographers when reading ultrasound images of the beating heart. To test their ideas, they developed an AI model to identify RWMA using divisions of the myocardial wall that are consistent with current ASE guidelines. The AI model was trained on an impressive dataset consisting of 15,746 clinical transthoracic echocardiography studies, with ground truth established by the corresponding clinical reports. The group then performed a comprehensive evaluation where the AI model went head-to-head against a group of clinical readers to evaluate RWMA in a test dataset unseen by either the AI model or the human readers, with ground truth for this test established by a separate panel of experts.
The results were astounding. Not only did the AI model perform the RWMA assessment in seconds, but its performance showed no statistical difference (defined as p < 0.05) as compared to the expert readers in the group. Moreover, the AI model demonstrated statistical superiority over a majority of novice readers to whom it was compared.
“What’s even more remarkable is that the model achieved expert-level performance when presented with images of the heart from a single angle (the apical view), whereas the study’s expert readers had the benefit of additional views from different angles. When pitted against novice echocardiographers – fellows with three or more months dedicated training in echocardiography but not yet passed their echocardiography board examinations – it outperformed the majority, potentially making it a valuable teaching tool,” Dr. Lang added.
The study demonstrates how AI has the potential to accelerate clinical workflows and to help clinicians make informed decisions in situations where accurate and consistent diagnosis is critical to determining patient treatment and outcomes.
“AI has the potential to remove much of the subjectivity in echocardiography so that expert and less highly trained echocardiographers alike can acquire high-quality images and make accurate quantitative measurements within seconds,” said Dr. Lang. “Due to the ubiquity and inherent safety of transthoracic echocardiography, we believe this study could further cement ultrasound imaging as a first-line diagnostic tool that benefits many more heart patients worldwide.”
Dr. Lang will present the results of this study, first published in early 2024 [1], during a presentation on Saturday, June 15 at 4:00 PM PST in Theater #2 at the 2024 American Society of Echocardiography Scientific Sessions (ASE 2024) Portland, Oregon, US.
Dr. Noah Liel-Cohen, co-founder of DiA Imaging Analysis, acquired by Philips in 2023, and Associate Professor of Cardiology at the Ben-Gurion University of the Negev (Israel), will co-present a related paper at ASE 2024 titled ‘Automated Calculation of Wall Motion Score Index by Machine Learning Based Algorithm Utilizing Artificial Intelligence Based Platform.’