A Lightweight Unidimensional Deep Learning Model for Atrial Fibrillation Detection

Quenaz Bezerra Soares and Marco Gutierrez
Heart Institute University of Sao Paulo


Abstract

Continuous rhythm monitoring using wearables devices is a potential tool for early identification of atrial fibrillation (AF), the most frequent cardiac arrhythmia (with 0,51% worldwide prevalence, increasing with time), and is also a tool for remote monitoring patients after cardiac surgery. However, AF detection directly through wearable devices is limited by the computational complexity of the classifier model. In this work we propose a lightweight AF classifier model based on the VGG-11 architecture (LiteVGG-11), focusing on reducing the number of parameters and numerical operations. Using a low number of filters, depthwise separable convolution, and global pooling, this model has only 20,454 parameters and needs 6.9 MFLOP to make an inference for an input of 10 seconds of the ECG leads I and II, sampled at 200 Hz (simulating an ECG signal acquired by a wearable device). To test its effectiveness for AF detection we used the dataset ensemble of the PhysioNet/CinC challenge 2021, stratifying the classes into sinus rhythm, AF, and other rhythms. After 10 Monte Carlo cross-validation splits, with 24,260 unbalanced samples for training and 1,536 balanced samples for validation and testing, the observed metrics (mean[standard deviation]) were: Sensitivity (Se) 81.3[0.6]%; Specificity (Sp) 90.6[0.3]%; F1-score 80.4[0.7]%; and area under the receiver operating characteristic curve (AUC) 91.5[0.8]%. Considering only AF performance, the metrics were: Se 94.1[0.1]%; Sp 91.9[0.8]%; F1-Score 89.5[0.7]%; and AUC 96.1[0.6]%. The best model submitted to the Physionet challenge, considering only AF detection over two leads, has a higher AUC (98.0%) but a lower F1-score (88.3%) than our model, showing that the LiteVGG-11 has a competitive performance. Furthermore, due to its computational simplicity, it is light enough to be embedded into smartphones or even directly into wearable devices, enabling continuous ECG rhythm monitoring.