Deep learning enables diagnosis of atrial cardiomyopathy from routine 12-lead electrocardiogram

BackgroundAtrial cardiomyopathy (AtCM) is both a cause and a consequence of atrial fibrillation and flutter (AF) and can lead to ischemic stroke. Imaging derived left atrial (LA) structure and function are used to diagnose AtCM. Considering the tight coupling of heart structure and rhythm generation, this information might also be derived from 12-lead electrocardiogram (ECG), which is low-cost and readily available. MethodsFirst, we finetuned a deep learning ECG foundational model (ECG-FM) pretrained on over 1 million ECG samples to predict LA imaging indices based on 26134 ECGs from the UK Biobank cohort. We then investigated if the ECG-predicted imaging features improved detection of patients with previous diagnosis of AF as well as prediction of incident AF beyond the CHARGE-AF Score on a test set from the UK Biobank. We externally validated our model on a Brazilian cohort of primary care ECGs (n = 64851) as well as a cohort of ischemic stroke patients from the University Hospital Zurich (n = 312) ResultsOur deep learning model successfully predicted LA imaging indices from 12 lead ECG with Pearson correlation of predictions and ground truths ranging from 0.41 - 0.52 (p < 0.001). In the UK Biobank test set, the ECG-predicted imaging features significantly improved detection of participants with previous AF diagnosis and five-year risk prediction of incident AF beyond the CHARGE-AF score. ECG-predicted imaging markers showed superior test performance compared to established ECG markers of AtCM and an alternative deep learning approach trained to detect patients with previous diagnosis of AF directly. This also held on external validation sets. We further successfully validated our model on Holter-ECG with reduced number of leads. DiscussionWe established a novel deep learning approach for the diagnosis of AtCM from 12 lead ECG. Due to the wide availability of ECG, our approach has the potential to improve screening and diagnosis of AtCM. The code for the analysis is available under: https://github.com/jul-des/DL-AtCM.git Clinical PerspectiveWhat is new? O_LILeft atrial imaging indices derived from cardiac magnetic resonance imaging can be predicted from 12 lead sinus rhythm ECG using deep learning. C_LIO_LIThe predicted imaging indices allow improved diagnosis of patients with previous episodes of atrial fibrillation, as well as prediction of incident atrial fibrillation. C_LIO_LIFurther, they allow improved detection atrial fibrillation as cause of stroke in a cohort of ischemic stroke patients. C_LI What are the clinical implications? O_LIWe developed and validated a novel approach for diagnosing atrial cardiomyopathy from 12 lead ECG. C_LIO_LIIt has the potential to improve diagnosis of atrial cardiomyopathy, which is key for preventing atrial fibrillation and ischemic stroke. C_LI