Prediction of spontaneous termination of atrial fibrillation in surface ECG by frequency analysis

We developed an automatic algorithm for predicting the spontaneous termination of atrial fibrillation (AFJ using the 2004 PhysioNet / Computers in Cardiology Challenge database. The algorithm consisted of QRST subtraction to cancel ventricular and correlated activiry followed by peak frequency analysis on the remninder signal. A learning set was used to determine a 5.9 Hi peak frequency threshold to separate immediately terminating (T) from non-terminating (N) segments. To separate T from nan-immediately terminating (S) segments, the frequency analysis was preceded by QRST morphology matching to identify pairs of S and T segments from each individual patient. Higher frequency segments were labeled as N or S respectively. This algorithm correctly separated 19/20 T vs. N segments from 20 different patients and 38/20 segments in IO patients in the learning set and correctly identified 25/30 segments from Test Set A nnd 16/20 segments from Test Set 3.