Session S84.3

Analysis of T-Wave Alternans Using the Ramanujan Transform

LT Mainardi*, R Sassi

Politecnico di Milano
Milano, Italy

T-Wave Alternans (TWA) is defined as a beat-to-beat alteration in the repolarisation morphology that repeats every other heart beat. Several methods have been proposed for TWA detection and in this study we investigate the use of the Ramanujan-Fourier Transform (RFT). The RTF decomposes a signal in a combination of Ramanujan Sums (RS), each sum being characterized by sinusoids of integer periodicity 1/q (with q={1, 2, 3, ..., N}) and its multiples p/q being p and q co-primes. This transform tends to project the original signal into patterns of finite periodicity: in the particular case of q=2, we have the 1 over 2 pattern typical of TWA. The RS have been introduced as the fundamental building blocks for the arithmetical functions in number theory. They have been recently rediscovered for signal processing and for the analysis of biological data.
We analyzed the data of the Computers in Cardiology Challenge 2008 consisting of 100 records, with 2, 3 or 12 concurrent ECG leads sampled at 500 Hz. Each record was firstly processed for QRS detection using the freely available software ECGPUWAVE. This tool also provided a first estimate of the waveforms boundaries. Such estimates guided the successive T-wave alignment. Each ECG signal was first high-pass filtered to remove major baseline wander (5th order Butterworth filter, cut-off frequency 0.5 Hz). Then an average pattern was built for the T wave and matched through cross-correlation at each T wave location. The procedure was iterated until convergence. The selection of those T-waves to be included in the TWA computation was done as follows: the RR series was screened to detect the longest RR sequences in which two successive RR intervals did not differ more than a 10% in respect to the mean global RR value. Then the T-waves belonging to this beats’ sequence were further processed. The criteria guaranteed that at least in the window under analysis no T-wave was missing (or was incorrectly add), which was fundamental for the detection of an alternating pattern. The preprocessing stage led to an array of time aligned T wave for each lead (when 12 leads were available we considered only V3, V4 and V5, where the T waves are most evident).
Finally, at each time offset n from the beginning of the T wave, the series of corresponding T wave amplitudes was transformed using the RFT. The modulus of the RFT coefficients |C(q, n)| with q=2, 3, 4 were considered for further analysis. |C(2, n)| measures the TWA, while |C(3, n)| and |C(4, n)| estimate the superimposed noise level. These coefficients were derived for each lead. The global TWA intensity was estimated as the maximum of |C(2, n)| for each offset n and every available lead. These estimates were submitted for ranking in the 2008 CinC Challenge and gained a promising preliminary score of 0.53 in the range [-1, 1]. The results show that the RFT might be a valuable tool for estimating TWA, but refinements are still necessary to improve performances.

(Abstract Control Number: 388)