2021/02/22 by Thomas Grandits, Grandits, Thomas, Simone Pezzuto +11 · 1 citation
Medicine · #Artificial Intelligence (cs.AI) #Atrial Fibrillation Management and Outcomes #Cardiac electrophysiology and arrhythmias #ECG Monitoring and Analysis #FOS: Computer and information sciences #Machine Learning (cs.LG)
paper · pdf · doi:10.48550/arxiv.2102.10863
openalex publication_date 2021/02/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Electroanatomical maps are a key tool in the diagnosis and treatment of\natrial fibrillation. Current approaches focus on the activation times recorded.\nHowever, more information can be extracted from the available data. The fibers\nin cardiac tissue conduct the electrical wave faster, and their direction could\nbe inferred from activation times. In this work, we employ a recently developed\napproach, called physics informed neural networks, to learn the fiber\norientations from electroanatomical maps, taking into account the physics of\nthe electrical wave propagation. In particular, we train the neural network to\nweakly satisfy the anisotropic eikonal equation and to predict the measured\nactivation times. We use a local basis for the anisotropic conductivity tensor,\nwhich encodes the fiber orientation. The methodology is tested both in a\nsynthetic example and for patient data. Our approach shows good agreement in\nboth cases, with an RMSE of 2.2ms on the in-silico data and outperforming a\nstate of the art method on the patient data. The results show a first step\ntowards learning the fiber orientations from electroanatomical maps with\nphysics-informed neural networks.\n