Diagnosis of atrial fibrillation from surface electrocardiograms based on computer-detected atrial activity

A computerized method to detect atrial fibrillatory activity on the surface electrocardiogram is presented. After ventricular activity was canceled by creating a remainder electrocardiogram, significant differences were found in the percent power of the remainder electrocardiograms for a group of rhythms with atrial fibrillation (mean +/- SD; lead V1, 47.4 +/- 29.7%; lead II, 39.4 +/- 26.8%) and a control group (irregular rhythms or rhythms without readily detectable P waves; lead V1, 17.6 +/- 14.6%; lead II, 19.2 +/- 13.9%) for both leads (p less than 0.0001). A discrimination algorithm that classified a rhythm as atrial fibrillation if the percent power was greater than 32% and if noncoupled P waves were not present had a specificity of 90.0% and a sensitivity of 69.7% for the training set and a specificity of 87.8% and a sensitivity of 68.3% for the test set. In addition, the algorithm correctly detected all 66 of the 66 sinus rhythms tested. The algorithm produced good results that may be incorporated into arrhythmia interpretation systems to improve their specificity.