Session S74.4

Estimation of T-Wave Alternans from Multi-Lead ECG Signals Using a Modified Moving Average Method

GM Nijm*, S Swiryn, AC Larson, AV Sahakian

Northwestern University
Evanston, IL, USA

T-wave alternans (TWA) is characterized in the surface ECG by alternating T-wave amplitudes, usually at heart rates in the range of 90 to 120 beats per minute. TWA magnitude is typically in the microvolt range, so it is not evident at normal ECG display scales. There is a predictive relationship between the degree of TWA and risk of sudden cardiac death. A wide variety of signal processing methods have been presented for detection and estimation of TWA, but further investigation is needed to determine which method provides the most consistent performance.
The objective of the Physionet/Computers in Cardiology Challenge 2008 is to estimate the peak magnitude of TWA for 100 multi-lead ECGs using a fully automated method. Each entry is compared with a reference ranking and assigned a Kendall rank correlation coefficient (1 = perfect agreement, -1 = perfect disagreement).
This paper presents a detailed algorithm for estimation of TWA using a modified moving average method. After baseline wander removal and QRS detection, even and odd beats are separated into different streams of beats. Leads that are too noisy or corrupted by artifact are eliminated from additional analysis. The modified moving average is computed for each beat stream, and the T-wave is isolated for both the even and odd averaged beats. Aberrant beats are removed from analysis by template matching. The absolute value of the maximum amplitude difference between the T-waves of the even and odd averaged beats is calculated as the peak magnitude of the TWA. The largest of the TWA magnitude estimates for all the leads is used to produce the final estimate of TWA peak magnitude.
A preliminary score (Kendall rank correlation coefficient) of 0.41 was obtained. Further work to improve the algorithm will include enhanced methods of combining data from multiple ECG leads. In addition, refinement of QRS detection may also help to improve accuracy of TWA estimation, since correct QRS detection is critical for the success of the modified moving average method.

(Abstract Control Number: 366)