A proposed two-stage algorithm using nonlinear HRV features and nearest-neighbor classifiers demonstrated robustness in predicting the onset of paroxysmal atrial fibrillation.
We propose a two-stage solution algorithm to predict the onset of paroxysmal atrial fibrillation (PAF) based on half-hour heart rate variability (HRV) signals. Nonlinear feature based on vectors calculated from return map and difference map constructed by HRV signal were developed. The extracted features were fed into their corresponding nearest-neighbor classifiers for parameter adjustment and classification. According to the official scoring results, our algorithm scored 34 points in the screening stage and 40 points in the prediction stage. In addition, the developed algorithm appears to he very robust against measuring noises. For example, with different QRS defectors, the classification results only change slightly (within 5%).
Lynn et al. (Wed,) studied this question.