Robust detection of atrial fibrillation for a long term telemonitoring system

We present a method to automatically detect atrial fibrillation (AF) for ambulatory monitoring with arbitrary lead placements. We aim for robust AF detection in the face of significant muscle artifact and potentially changing morphology. Our approach is based on the variance of R-R intervals since the QRS spike is the most prominent feature of an ambulatory ECG and the least confounded by muscle noise. Specifically, we use the morphology-independent QRS detector wqrs to compute R-R intervals and variance and then smooth the resulting classifications for further robustness. Experiments on the MIT-BIH AF database (AFDB) show our AF detection algorithm has sensitivity 0.96 with specificity 0.89 which is sufficient for AF screening