1880 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 56, NO. 7, JULY 2009 Multilead Analysis of T-Wave Alternans in the ECG Using Principal Component Analysis Violeta Monasterio ∗ , Pablo Laguna, Senior Member, IEEE, and Juan Pablo Mart´ ınez Abstract—T-wave alternans (TWA) is a cardiac phenomenon associated with the mechanisms leading to sudden cardiac death. Several methods exist to automatically detect and estimate TWA in the ECG on a single-lead basis, and their main drawback is their poor sensitivity to low-amplitude TWA. In this paper, we propose a multilead analysis scheme to improve the detection and estimation of TWA. It combines principal component analysis with a single-lead method based on the generalized likelihood ratio test. The proposed scheme is evaluated and compared to a single-lead scheme by means of a simulation study, in which different types of simulated and physiological noise are considered under realistic conditions. Simulation results show that the multilead scheme can detect TWA with an SNR 30 dB lower and allows the estimation of TWA with an SNR 25 dB lower than the single-lead scheme. The two analysis schemes are also applied to stress test ECG records. Results show that the multilead scheme provides a higher detection power and that TWA detections obtained with this scheme are significantly different in healthy volunteers and ischemic patients, whereas they are not with the single-lead scheme. Index Terms—ECG, multilead analysis, principal component analysis (PCA), T-wave alternans (TWA). I. INTRODUCTION T -WAVE alternans (TWA) isa cardiac phenomenon exten- sively studied as an index of high risk of malignant ar- rhythmias and sudden cardiac death (SCD) [1], [2]. This paper presents a multilead analysis scheme that improves the detection and estimation of TWA in the ECG. ECG signals are measured by placing electrodes on the body surface and recording the electrical activity of the heart. The simultaneous recording of the ECG on different chest loca- tions (channels or leads) provides a spatial perception of cardiac events. The standard 12-lead system is the most widely used sys- tem in clinical practice, and consists of eight independent leads, named V1–V6, I, and II, and four additional leads that can be derived from the independent ones. The ECG usually presents Manuscript received October 3, 2008; revised January 15, 2009. First published March 4, 2009; current version published June 12, 2009. This work was supported by the Centro de Investigaci´ on Biom´ edica en Red (CIBER) de Bioingenier´ ıa, Biomateriales y Nanomedicina through Instituto de Salud Carlos III (ISCIII), by the Comisi´ on Interministerial de Ciencia y Tecnolog´ ıa (CICYT) under Project TEC-2007-68076-C02-02, and by the Grupo Consoli- dado T30 (Spain). Asterisk indicates corresponding author. ∗ V. Monasterio is with the Centro de Investigaci´ on Biom´ edica en Red de Bioingenier´ ıa, Biomateriales y Nanomedicina (CIBER-BBN), Communications Technology Group, Arag´ on Institute of Engineering Research, University of Zaragoza, Zaragoza 50018, Spain (e-mail: violeta.monasterio@unizar.es). P. Laguna and J. P. Mart´ ınez are with the Centro de Investigaci´ on Biom´ edica en Red de Bioingenier´ ıa, Biomateriales y Nanomedicina (CIBER-BBN), Com- munications Technology Group, Arag´ on Institute of Engineering Research, University of Zaragoza, Zaragoza 50018, Spain (e-mail: laguna@unizar.es; jpmart@unizar.es). Digital Object Identifier 10.1109/TBME.2009.2015935 Fig. 1. (a) ECG signal with visible TWA. (b) Superposition of two consecutive beats. (c) Alternans waveform: difference between odd and even beats. three characteristic waves on each beat: P-wave, QRS complex, and T-wave [Fig. 1(a)]. The interval between the end of the QRS complex and the end of the T-wave is known as ST-T complex, and reflects the repolarization activity of the ventricles. TWA is defined as a consistent fluctuation in the repolariza- tion morphology on an every-other-beat basis [Fig. 1(b) and (c)]. TWA amplitude is in the range of microvolts and can be even be- low the noise level, making its detection a difficult task. Several signal processing methods exist to detect and estimate TWA. A comprehensive review can be found in [3]. The most widely used techniques are the spectral method (SM) [1], [4] and the modified moving average method [5]. Alternative techniques are the complex demodulation method [6] and the recently pro- posed Laplacian likelihood ratio method (LLR) [7], [8]. The main drawback of existing techniques is either their sensitivity to the presence of nonalternant components with high amplitude or their poor sensitivity to low-level TWA [2], [3]. Furthermore, some of these techniques measure TWA amplitude, but do not estimate the TWA waveform. An accurate waveform estimation is desirable because, in addition to the presence and magnitude of TWA, the distribution of TWA within the ST-T complex has been shown to indicate arrhythmic risk [9]. To date, TWA analysis techniques have been mostly applied to each lead individually. In commercial TWA analysis systems, only basic multilead strategies are performed, such as analyzing the vector-magnitude lead (CH2000 and Heartwave systems, Cambridge Heart, Inc., Bedford, MA). However, ECG signals present a high spatial redundancy that can be better exploited with techniques based on the eigenanalysis of input data, such as 0018-9294/$25.00 © 2009 IEEE Authorized licensed use limited to: Universidad de Zaragoza. Downloaded on July 10, 2009 at 04:32 from IEEE Xplore. Restrictions apply.