Session P73.2

Screening Patients with Paroxysmal Atrial Fibrillation (PAF) from Non-PAF Heart Rhythm Using HRV Data Analysis

YV Chesnokov*

Kuban State University
Krasnodar, Russia

Paroxysmal atrial fibrillation (PAF) is the most common abnormal heart rhythm encountered in clinical practice, and has serious associated morbidity and mortality as a sudden stroke. As PAF occurrence usually hard to catch using conventional ECG recording during short visit to clinic, screening if a patient is prone to PAF from non-PAF heart rhythm would facilitate diagnosis. To achieve this goal we studied non-PAF heart rhythms from PAF documented patients and patients without that disease.
The data for analysis was taken from Physionet databases. We used atrial fibrillation prediction database (AFPDB), consisted of 30 minute non-PAF ECG segments from PAF patients, healthy controls and diseased non-PAF patients, MIT-BIH AF database (AFDB), consisted of 10 hour recordings from PAF suffering patients with PAF and non-PAF rhythms and normal sinus rhythm database (NSRDB), consisted of 24 hour recordings. We annotated each ECG record using our developed algorithm [Y.V. Chesnokov et al, Individually Adaptable Automatic QT Detector. In Proceedings of Computers Cardiol: 17-20 September 2006, Valencia.] and extracted HRV data. We used entire 30 minute segment for analysis from AFPDB database. Long-term records from AFDB and NSRDB databases were divided to 30 minute segments. Obtained HRV segments were interpolated to 2Hz and processed with Fourier analysis (FFT) estimation in the 0.01 – 0.5Hz frequency range. In order to obtain automatic classification we applied artificial neural networks (ANN). First we trained ANN on the non-PAF HRV data from AFPDB database to distinguish between patients prone to PAF and non-PAF patients. Then we applied that ANN model to AFDB (on the non-PAF HRV data) and NSRDB database for testing.
FFT spectral analysis of the 30 minute HRV records, on the heart rhythm without PAF, from AFPDB database for patients with PAF history provided statistically significant (p<0.001, T-test) increase in the 0.1 – 0.5Hz energy compared to non-PAF patients. FFT spectrum for AFDB and NSRDB non-PAF rhythms has the same distribution compared to AFPDB database. Results on per-segment basis for ANN classification on non-PAF 30 minute segments for 16 patients with PAF history provided Sensitivity of 94.5%. For 16 healthy patients from NSRDB database we achieved Specificity of 96.5%.

(Abstract Control Number: 44)