Medical Engineering & Physics 20 (1998) 773–779 Ventricular late potential analysis with musical and harmonic wavelets Arnaldo Batista a,* , Michael English b a Grupo de Biofı ´sica e Engenharia Biome ´dica, Departamento de Fı ´sica, FCT/UNL, Monte de Caparica, Portugal b Graduate Division of Biomedical Engineering, University of Sussex, Falmer, Brighton, UK Received 18 March 1997; received in revised form 9 November 1998; accepted 9 November 1998 Abstract Harmonic and musical wavelets were introduced by DE Newland in 1994, and have their spectrum tightly defined, therefore greatly reducing spectral leakage that may disturb signal frequency analysis. We have explored the ability of these wavelets to perform detection and quantification of ventricular late potentials (VLP) through our multiresolution time-scale method of energy comparison between the ST and TP segments of the ECG. Since reduction of spectral leakage improves the method’s reliability, Newland wavelets provided better results than Daubechies wavelets in our study cases. The only drawback is the comparatively reduced time resolution of Newland wavelets. This required us to concatenate a number of ST segments to form a longer data set that is more representative of the high-resolution ECG (HR-ECG) of a patient than one individual beat. This approach may also be considered for other applications in the HR-ECG field. The spectral properties of the Newland wavelets play a major role in the improvement in our results.  1999 IPEM. Published by Elsevier Science Ltd. All rights reserved. Keywords: Ventricular late potentials; High-resolution electrocardiography; Wavelets 1. Introduction Ventricular late potentials (VLP) are cardiac microvolt signals arising from the delayed inhomo- geneous propagation of the depolarisation wave through ventricular tissue. They have diagnostic value as predic- tors of malignant arrhythmias and cardiac arrest. Tra- ditional time analysis methods excessively depend on the noise level and the location of the QRS limits [1,2]. Time–frequency methods are a promising modern alter- native still under scrutiny, for which the spectrogram [3] and the Wigner–Ville methods have mainly been used [4]. The spectrogram’s limitations are the low resolution and dependence on the window type, while the Wigner– Ville Transform exhibits cross terms. Wavelets are one of the most recent tools adopted for high-resolution ECG (HR-ECG) research. Reports have pointed out the superior ability to discriminate low energy transients * Corresponding author. Departamento de Fı ´sica da FCT/UNL, Quinta da Torre, 2825 Monte de Caparica, Portugal. Tel.: + 35-11- 294-8576; fax: + 35-11-294-8549; e-mail: agb@mail.fct.unl.pt 1350-4533/99/$19.00  1999 IPEM. Published by Elsevier Science Ltd. All rights reserved. PII:S1350-4533(98)00087-3 such as VLPs [4,5]. We present a quantitative method based on harmonic and musical wavelets [6–8]. Our HR- ECG acquisition set-up had the following parameters: ADC effective resolution, 12 bits; group delay error, 0.77%; noise level below 1 V; final stage sampling fre- quency, 2.2 kHz; and bandwidth, 550 Hz. 2. Discrete Wavelet Transform (DWT) with harmonic and musical wavelets Newland wavelets are given by Newland [8]: w m,n x - k n - m  = (1) exp  in2  x - k n - m  - exp  im2  x - k n - m  i2(n - m)  x - k n - m  For m = 2 j and n = 2 j+ 1 the harmonic wavelet family