Unsupervised Classification of Atrial Fibrillation Triggers Using Heart Rate Variability Features Extracted from Implantable Cardiac Monitor Data

Catheter ablation is a common treatment of atrial fibrillation (AF), but its success rate is around 60%. It is believed that the success rate can be improved if the procedure were to be guided by the specific AF triggers found in the "Flashback", i.e. the trend of around 500 ventricular beats preceding the AF onset stored in an implantable cardiac monitor (ICM). The need to automatically classify these different triggers: atrial tachycardia (AT), atrial flutter, premature atrial contractions (PAC) or Spontaneous AF has motivated the design in this paper of an unsupervised classification method evaluating statistical and geometrical Heart Rate Variability (HRV) features extracted from the Flashback. From a cohort of 132 patients (57± 12 years, male 67%), 528 Flashbacks were extracted and classified into 5 different clusters after the Principal Component Analysis (PCA) was computed on the HRV features. 2 principal components explained more than 95% of the variance and were a combination of the mean R-R interval, Square root of the mean squared differences of successive R-R intervals (RMSSD), Standard deviation of the R-R intervals (SDNN) and Poincare descriptors, SD1 and SD2. RMSSD and SD1 were significantly different among all clusters (p-value < 0.05, with Holm's correction) showing that distinct patterns can be found using this method.Clinical Relevance-Preliminary step towards ablation strategy guidance using the AF trigger patterns to improve catheter ablation success rates.