Researchers have developed a new AI model that can evaluate ECG data as well as expert cardiologists, sharing their findings in JAMA Cardiology. The team’s algorithm, which was trained to predict the presence of 38 different diagnostic classes, also outperformed a commonly used automated analysis tool.
“Building on a previously established approach, we specifically developed a convolutional neural network (CNN) to accept and train on ECG data that are readily available in most institutions: namely, XML-format ECG waveform data and cardiologist-confirmed text diagnosis labels,” wrote first author J. Weston Hughes, BA, of the department of electrical engineering and computer science at the University of California, Berkeley, and colleagues. “By training our CNN using commonly available ECG data, we aspired to demonstrate what can be achieved in many institutions and, more importantly, what could be eventually achieved by combining cross-institutional data.”
The group focused on ECG data from more than 365,000 patients treated at the University of California, San Francisco from 2003 to 2017. Data from more than 260,000 patients was used to train the AI model, and data from more than 32,000 patients was used to refine the model and then test its performance. In addition, ECG data from more than 300 patients treated in 2018 was to create an additional consensus dataset.
Overall, the CNN demonstrated an area under the ROC curve of at least 0.960 for 84.2% of 38 diagnostic classes. The model also had higher frequency weighted mean F1 scores than a group of expert cardiologists and GE Healthcare’s MUSE automated analysis tool.
“A growing body of evidence supports the aptitude of machine learning algorithms like CNNs for everyday ECG analysis,” the authors wrote. “Our results corroborate recent studies supporting strong CNN performance for common ECG diagnoses in various settings.”
Hughes et al. also explored their model’s potential for improving clinical workflow. High-performance algorithms could be integrated into ECG analysis systems, for example, or deployed on digital data before ECGs are brought into the patient’s electronic health record.
“Such integration could enable existing ECG systems to be supplemented for diagnoses where CNNs excel,” the group wrote.
Click here to read the full JAMA Cardiology study.
Michael has more than 16 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.