New AI Model Significantly Improves Detection of Atrial Fibrillation
Technion Researchers have developed a new artificial intelligence model that significantly improves the detection of irregular heartbeats in heart recordings. This new tool, called RawECGNet, is better than older methods at spotting two types of heart rhythm problems: atrial fibrillation (AF) and atrial flutter (AFl). It outperforms existing methods and shows strong generalization across different patient populations and ECG lead placements. The study was led by graduate student Noam Gadot (Ben Moshe) from the Faculties of Computer Science and Biomedical Engineering and Dr. Joachim Behar of the Faculty of Biomedical Engineering, in collaboration with Lund University (Sweden) and Saitama Medical University (Japan).
RawECGNet processes raw ECG data unlike previous models that relied solely on heart rate patterns. The new model demonstrated superior performance compared to the state-of-the-art ArNet2 model, correctly identifying irregular heartbeats 91-96% of the time. RawECGNet showed improved generalization across different geographical locations, ethnicities, and ECG lead placement. The model significantly improved the detection of atrial flutter, a challenge for previous AI models, and demonstrated better performance in estimating AF burden, particularly in cases of severe AF.
Dr. Joachim Behar, the lead senior author, stated, "Our work represents a significant step forward in the automated detection of atrial fibrillation. By leveraging the raw ECG signal, we've created a more robust and generalizable model that can potentially improve early detection and monitoring of this common heart rhythm disorder."
The researchers validated their model using over 7,400 hours of manually annotated ECG data from multiple countries, demonstrating its potential for real-world application across diverse healthcare settings.
This advancement could lead to more accurate and efficient screening for atrial fibrillation, potentially reducing the risk of stroke and other AF-related complications for patients worldwide.
The study was supported among others by the Ministry of Science & Technology, the Technion-Rambam Initiative in Artificial Intelligence in Medicine (TERA) and the Hittman Family Fund.
The study titled "RawECGNet: Deep Learning Generalization for Atrial Fibrillation Detection from the Raw ECG,"was published IEEE Journal of Biomedical and Health Informatics
