Neural Network Detection of Atrial Fibrillation by Lorenz Plot Images of Interbeat Interval Variation

We developed artificial intelligence (AI) atrial fibrillation (AF) detection system using the big data of Allostatic State Mapping by Ambulatory electrocardiogram (ECG) Repository (ALLSTAR). Detection of atrial fibrillation (AF) is important for preventing acute cerebral infarction, but some waveforms that clinicians can easily diagnose existed that cannot be successfully detected by conventional programs. We let AI learn the detection of AF by convolutional neural network (CNN) using Lorenz plot of heart rate variability of 24-h ECG in subjects with sinus rhythm (SR) and AF whose diagnosis was confirmed as teacher data. Among 10000 datasets for SR and AF each, 80% of data was used as teacher data. With remaining 20% validation data, the CNN developed by AI detected AF with 100% sensitivity and 100% specificity.