Time-frequency analysis compensating missing data for Atrial Fibrillation ECG assessment

We propose a novel algorithm for temporal tracking of the fibrillatory frequency for Atrial Fibrillation (AF) ECG. Both Atrial activity extraction and fibrillatory frequency tracking are combined into a single algorithm and analytical expressions are derived for final time-frequency distributions, thus the proposed method is computationally efficient. It can also serve the general purpose of time-frequency analysis of a non stationary signal with missing or corrupted data segments. The method uses short-time expansion of an orthogonal basis set, and regularized least squares solution used to compute a stable coefficients for the basis. A simplified and computationally efficient time-frequency distribution is derived based only on the coefficient vector. The algorithm is successfully applied to simulated as well as real ECG with AF, and its ad vantages over conventional average beat subtraction method are discussed.

1. INTRODUCTION

Atrial Fibrillation (AF) is the most common sustained cardiac arrhythmia, increasing in prevalence with age, accounting for approximately one third of hospitalizations for cardiac rhythm disturbances [1]. AF is characterized by the replacement of consistent P-waves on the ECG by rapid oscillations (fibrillatory waves) that vary in amplitude, frequency, and shape, associated with an irregular ventricular response. AF affects approximately 10% of the population over age of 75 and is associated with an increased risk of stroke [2], [1]. Noninvasive analysis of AF and other supraventricular tach-yarrhythmia conditions require cancelation of the ventricular activity (QRS complex and T-wave) from the ECG [3], [4]. Due to the overlapping spectral contents of atrial and ventricular activities, linear filtering is not possible, and also the higher dominance of ventricular activity in the ECG makes the extraction of atrial activity non-trivial [5], [6].