WO2010084211A1

WO2010084211A1 - Device and method for detecting alternation of ventricular repolarisation using windowing

Info

Publication number: WO2010084211A1
Application number: PCT/ES2009/000296
Authority: WO
Inventors: Eduardo MORENO MARTÍNEZ; Manuel Blanco Velasco; Pedro AMO LÓPEZ; Fernando CRUZ ROLDÁN; Concepción MORO SERRANO; Antonio HERNÁNDEZ MADRID
Original assignee: Universidad De Alcalá
Priority date: 2009-01-20
Filing date: 2009-05-28
Publication date: 2010-07-29
Also published as: ES2343054A1; ES2343054B2

Abstract

Device and method for detecting alternation of cardiac ventricular repolarisation. The method involves extracting ventricular repolarisation information by windowing in continuous time blocks of beats of variable length of a bioelectric heart signal, ECG, where the windowing is effected with a window defined by a periodic adjustable duty-cycle signal.

Description

DEVICE AND METHOD FOR DETECTION OF THE ALTERNANCE OF VENTRICULAR REPOLARIZATION THROUGH WINDING.

TECHNICAL SECTOR The object of the patent is the description of a method that can be implemented in physical devices, both external and implantable in the human body, for the detection of alternating cardiac repolarization. Its use is therefore intended for the clinical environment, being included in the biomedical instrumentation technique sector.

STATE OF THE TECHNIQUE

The invention is a device and method for detecting alternating ^patterns' of Ia repolarization wave. ventricular on electrical signals of the heart or electrocardiogram (ECG). The ECG is the electrical recording of the activity of the heart and is used to establish the diagnosis of heart disease. Each heartbeat is made up of a set of waves that are traditionally; denote by the letters PQRST. The segment of the signal between the S wave and the T wave, referred to as the ST-T complex, corresponds to the ventricular repolarization and is related to various pathologies. T-wave alternation (AOT) or alternation of ventricular repolarization has been proposed as a risk indicator for the diagnosis of ventricular failure and ventricular arrhythmias. Recent studies have concluded that the appearance of AOT is a stratifier of the risk of sudden cardiac death [GehO5]. It has also been concluded that AOT can appear before the cardiac ventricular fibrillation and after the occlusion of the coronary arteries that can also lead to episodes of ventricular fibrillation and sudden cardiac death. Its detection can be valid to predict these episodes and act accordingly, for example in an irriplementation in implantable defibrillators that can complement and improve the detection of positive cases. The AOT consists of small variations in the morphology, amplitude and duration of the ST-T complex that occur between consecutive beats. In most of the cases, the AOT episodes are characterized by amplitudes close to the microvolts making their inspection and visual recognition practically impossible.

Figure 1 shows an example of a T wave alternation episode. Consider that the ST-T complex in the absence of AOT episodes presents a morphology that we can associate with a pattern referenced as A. In case AOT ₁ exists periodically, beat-by-beat variations on the ventricular repolarization segment occur. In this case, a new pattern appears on the ST-T complex that is referred to as B. The consecutive alternation between patterns A and B (Fig. 1) corresponds to an AOT event (ABABABA ...). This phenomenon is reflected in the spectrum of the signal in the frequency components equal to half of the beat frequency.

Among the various analysis techniques of the alternation of the T wave proposed to date, we can mention: '• The spectral method [Ros94], [Smi98]: this technique proposes to analyze the energy variations of the frequency spectrum of the ECG signal of the time series of a group of beats, so that a peak of energy is sought for a revealing frequency of the fluctuation sought.

• The complex demodulation technique [Nea91]: this technique tries to model the fluctuation of the amplitude of the T wave by a sinusoid of frequency 0.5 cycles per beat and of variable amplitude and phase, so as to ensure dynamic tracking of the alternation variations of the T wave.

• The technique of the modified moving average [VerOO]: this method consists in calculating, every two beats, the moving average of the level of the T wave at a given point of the repolarization segment and in quantifying the difference in amplitude between the two means . '

• The stretching technique [Ber97]: in this technique the T wave is superimposed on a template and the temporal component is stretched so as to minimize the difference between the template and the analyzed beat.

• The autocorrelation technique [Bur97]: it is a technique that consists in quantifying, in the temporal domain, the amplitude and morphology variations of the repolarization wave based on a correlation index. Each T wave is correlated with a mean T wave representative of a series of beats, translating an alternation, positive or negative, into an oscillation of the correlation index around the unit value. The intef beat series of coefficients is analyzed using a zero-pass counter in the time domain.

«The approach using wavelets [Cou99]: the ECG signal is broken down as a sum of Gaussians that are subsequently processed, so that the different components of the wave (P, QRS and T) are isolated and thus appear singularities, revealing in particular of an alternation for the T wave.

• Transformed from Karhunen-Loewe [Lag96]: Ia is used. KLT truncated to compact the energy of the ST-T complex in a reduced number of coefficients to subsequently analyze the series spectrally using the correlation method.

• Filtering method Capón [MarOO]: is a variant of complex demodulation. An FIR filter is used that minimizes the power of the output signal while preserving the alternating component. It is applied instead of an invariant low pass filter.

• Poincaré projection method [StrO2]: Poincaré maps are used to analyze dynamic systems that show periodicity. The Poincaré series is obtained by taking samples in the same phase of the ST-T complex of consecutive beats and pairs of differences between the samples of that series are represented. The alternation is identified when two separate point groupings are present on the maps.

• Method of the periodicity transform [SriO2a]: The periodicity transform is applied to the interlaced series of some characteristics of the T wave, such as the peak amplitude, area or variance. With this method, the energy of the orthogonal projection of each series in the subspace of periodicity sequences 2 beats is calculated.

• Method of statistical tests [SriO2b]: different statistical tests are proposed for the detection of AOT. • Method of the Laplacian probability relation [MarO6]: Given a signal model that includes noise and alternation, the maximum probability estimator (MLE) and the generalized probability relation test (GLRT) can be used for Ia alternation estimation and detection. It is shown that the probability density function of physiological noise corresponds to a Laplacian distribution. The MLE and GLRT for this model are based on median filters.

The methods described in the state of the art for the detection of AOT must cope with two main effects inherent in ECG signals. These two effects are the noise of the ECG signal and the variability of the heart rate.

• The noise on the ECG signal is one of the main problems for the detection of AOT, especially when its power is of the same order of magnitude as that of the power of the alternation. In this case, it can happen that the noise conceals the small variations associated with the AOT. This fact is common for all the proposed detection methods, no more solution having been given than the inclusion of preprocessing stages for the elimination of noise.

• The variability of the heart rate is another main problem in the detection of AOT in the sense that some of the methods described are designed with the necessary condition that the analyzed ECG signal has stationary conditions in the heart rate. Since the ECG signal is a variable frequency signal, the detection methods require obtaining said conditions of proximity to Ia. periodicity, through actions external to the AOT analysis. The solutions adapted to achieve these conditions consist in the administration of drugs, such as atropine or dobutamine, or in the performance of stress tests monitored by medical personnel. This heart rate variability factor is magnified when the AOT detection method requires a high number of beats for the ECG analysis since it becomes difficult for said stationary conditions to be maintained over time. The techniques described above use frames of high length of at least 64 to 128 beats. These two effects prevail in the case of certain r seu rs as Holter records or stress tests or ergometry records.

The present invention presents a new technique of detection, quantification and calculation of the alternating waveform in the analysis of the alternation of the T wave of an ECG signal. With respect to the techniques referred to above, this invention has the following properties:

- Processes the ECG in the time domain.

- It is based on spectral analysis, but instead of processing time series extracted from the ECG it uses the ECG as the original signal in the analysis.

- It uses a reduced number of beats in the analysis, 8 to 32, reducing the effect of heart rate variability, improving the resolution of the analysis and reducing the computational cost making it valid for its implantation in any existing device, including portable or implantable devices .

- It is robust against noise, being the valid method for the analysis of any type of electrical signal of the heart from any existing source or device, among which we can mention, the long-term Holter registers, signals from stress tests , cardiac monitoring devices or intracavitary signals from electrophysiological studies or implantable devices.

- In addition to estimating the level of alternation of the signal, it allows recovering the waveform of the alternation and defining the temporal moments of the ECG on which it appears.

BIBLIOGRAPHY

[Ada81] D. R. Adams, S. Akselrod, and R. J. Cohen, "Estimation of ventricular vulnerability to fibrillation through T-wave time series analyzes," Computers in Cardiology, vol. 8, pp. 307-310, 1981.

[Ada84] D. R. Adam, J. M. Smith, S. Akselrod, S. Nyberg, A. O. Powell, and R. J. Cohen, "Fluctuation in T-wave morphology and susceptibility to ventricular fibrillation," Journal of Electrocardiology, vol. 17, pp. 209-218, 1984.

[Ada99] K. Adachi, Y. Ohnishi, T. Shima, K. Yamashiro, A. Takei, N. Tamura and

M. Yokoyama, "Determinant of microvolt-level T-wave altérnans in patients with dilated cardiomyopathy, "Journal of the American College of Cardiology, vol. 34, no 2, pp. 374-380, August 1999.

[Ala96] R F. Alajaji, N. Phamdo, and T. Fuja, "Channels codes that exeploit the residual redundancy in CELP-encoded mean," IEEE Transactions on Speech and

Audio Processing, vol. 4, no. 5, pp. 325-336, September 1996.

[Ber97] Berger R et al., Beat-to-Beát QT Interval Variability: Novel Evidence for Repolarizatioh Ability in Ischemic and Non-lschemlc Dilated Cardiomyopathy, Circulation 96: 1557-1565; 1997

[Bur97] L. Buratini, W. Zareba, J. P. Couderc, E. L. Titlebaum, and A. J. Moss, "Computer detection of nonstationary T-wave altemañs using a new correlative method," Computers in Cardiology, vol. 24, pp. 657-660, 1997.

[Bur99] L. Buratini, W. Zareba, and A. J. Moss, "Correlation method for detection of transient T-wave alternans in digital Holter ECG recording," Annals of Noninvasive Electrocardiology, vol. 4, not 4, pp. 416-426, 1999.

[Cou99] Couderc JP et al., Beat-to-Beat Repolarization Variability in LQTS Patients with the SCN5A Sodium Channel Gene Mutation, PACE 22, 1581-1592; 1999

[Gar75] W. A. Gardner and L E. Franks, "Characterization of cyclostationary random signal processes," IEEE Transactions on Information Theory, vol. 21, no. 1, pp. 4-14, Jan. 1975.

[GehO5] A. K. Gehi, R. H. Stein, L. D. Metz, and J. A. Gomes, "Microvolt T-wave alternans for the risk stratification of ventricular tachyarrithmic events," Journal of the American College of Cárdiologíay, vol. 46, no. 1, pp. 75-82, Mar 2005. [Est97] NAM Esteslll, G. Michaud, DP Zipes, N. El-Sherif, FJ Venditti, DS Rosenbaum, P. Albrecht, P. J, Wang, and RJ Cohen, "Electrical Alternans During Rest and Exercise as Predictors of Vulnerability to Ventricular Arrhythmias, "The

American Journal of Cardiology, vol. 80, no 10, pp. 1314-1318, November 1997.

[Hua01] NE Huang, Z Shen, SR Long, MC Wu, EH Shih, Q Zheng, "The empirical mode decomposition method and the Hilbert spectrum for non-stationary time series analysls", Proceedings of the Royal Society of London, 0962- 8452 vol: 454 [Hjo70] B. Hjorth, "EEG analysis based on time domain," Electroencephalography and Clinical Neurolophysiology, vol. 29, pp. 306-310, 1970.

[Hjo73] B. Hjorth, "The physical significance of time domain descriptors in. EEG analysis," Electroencephalography and Clinical Neurolophysiology, vol. 34, pp. 321-

325, 1973..

[JyhO1] Jyh-Jong Wei, Chuang-Jan Chang, Nai-Kuan Chou, and Gwo-Jen Jan, "ECG Data Compression Using Truncated Singular Valué Decomposition" IEEE Trahsactions on Information Technology In Biomedicine, VoI. 5, No. 4, December

2001

[Lag96] P. Laguna ,, M. Ruiz, G. B. Moody, and R. G. Mark, "Repolarization alternans detection using the KL transform and the beatquency spectrum," Computers in Cardiology, vol. 23, pp. 673-676, 1996.

[MarOO] J, P. Martínez, S. Olmos, and P. Laguna, "Simulation study and performance evaluation of T-wave alternans detectors," Proceedings of 22 ^πd Conf. IEEE EMBS (CD-ROM), 2000.

[MarO5] J. P. Martínez and S. Olmos, "Methodological principies of T wave alternans analisys: A unified framework," IEEE Transactions on Biomedical Engineering, vol. 52, no 4, pp. 599-613, Apr 2005.

[MarO6] JP Martínez, S. Olmos, G. Wagner, and P ^' Laguna, "Characterization of

Repolarization alternans during ischemia: Time-course and spatial analysis, "IEEE Transactions on Biomedical Engineering, vol. 53, no 4, pp. 701-711, Apr 2006.

[Nea91] B. D. Nearing, A. H. Huang, and R. L. Verier, "Dynamic tracking of cardiac vulnerability by complex demodulation of the T-wáve," Science, no 252, pp. 437-

440, 1991.

[Nea93] BD Nearing and RL Verier, "Personal computer system for tracking cardiac vulnerability by complex demodulation of the T-wave," Journal of Applied Physiology, vol. 74, no 5, pp. 2606-2612, 1993. [NeaO2] BD Nearing and RL Verrier, "Modified moving average analysis of T-wave altemans to predict ventricular fibrillation with high accuracy," Journal of Applied Physiology, vol. 92, pp. 541-549, 2002.

[Ram97] A. G. Ramakrishnan, and Supratim Sana, "ECG Coding by Wavelet- Based Linear Prediction", IEEE Transactions on Biomedical Engineering, vol. 44, no. 12, December 1997.

[Ros94] D. S. Rosembaum, L. E. Jackson, J. M. Smith, H. Garan, J. N. Rushin, and R. J. Cohen, "Electrical alternans and vulnerability to ventricular arrhytmias," New England Journal of Medicine, vol. 330, no 4, pp. 235-241, Jan 1994.

[Ros96] D. S. Rosembaum, L. Jackson, and R. J. Cohen, "Predicting sudden cardiac death from T wave alternans of the surface electrocardiogram: promise and pitfalls," Journal of Cardiovascular Electrophisiology, vol. 7, no 11, pp. 1095-1111, Nov 1996.

[SalO7] K. E. Salber, "Ethical issues in defibrillator patients' quaüty of life," IEEE

Engineering in Medicine and Biology Magazine, May / June 2007.

[Smi98] J. M. Smith, E. A. Clancy, C. R. Valeri, J. N. Rushin, and R. J. Cohen, "Electrical alternans and cardiac electrical instability," Circulation, vol. 77, no 1, pp. 110-121, 1998.

[Sor05] L. Sδmmo and P. Laguna, Bioelectrical signal processing ¡n cardiac and neurological applications. Elsevier Academic Press, 2005.

[SriO2a] T. Srikanth, D. Lin, N. Kanaan, and H. Gu, "Estimation of low level alternans using périodicity transform simulation and european ST-T data results," Proceedings of 24th Conf. IEEE EMBS, pp. 1407-1408, 2002.

[Sr¡02b] T. Srikanth, D. Lin, N. Kanaan, and H. Gu, "Presence of T-wave alternans in the statistícal context-Á new approach to low amplitude alternans measurement,"

Computers in Cardiology, vol. 29, pp. 681-684, 2002. [StrO2] P. Strumillo and J. Ruta, "Poincaré mapping for detecting abnormal dynamics of cardiac. Repolarization," IEEE Engineering in Medicine and Biology Magazine, vol. 21, no 1, pp. 62-65, 2002.

[Tap01] JM Tapanainén, AM Still, KE Airaksinen, and HV Huikuri, "Pronostic significance of risk stratifiers of mortality, including T wave alternans, after acute myocardial infarction: Results of a prospective follow-up study," Journal of Cardiovascular Eletrophysiology, vol. 12 no. 6, pp. 645-652, June 2001.

[VerO3] RL Verrier, BD Nearlng, MT Ia Rovere, GD Pinna, MA Mittieman, JT Bigger, and PJ Schwartz, "Ambulatory electrocardiogram-based tracking of T wave alternans in postmyocardial infarctio ^ patients to asses risk of cardiac arrest of arrhytmic death , "Journal of Cardiovascular Electrophysiology, vol. 14, pp.

705-711, June 2003.

EXPLANATION OF THE INVENTION The described invention aims at the detection of T Wave Alternation (AOT) by analyzing the frequency components preferably at half of the heartbeat frequency. The proposed detection methods are applied on ECG signal blocks formed by several beats. To model each of these heart signal blocks, two properties are considered: proximity to periodicity conditions and proximity to stationary conditions. In order to achieve said conditions, the operation of the invention is based on the analysis of blocks with a reduced amount of beats (Fig. 2). On the one hand it is possible to reduce the heart rate variability of the block thus favoring conditions close to periodicity. On the other hand it is achieved that the succession of PQRST waves of each beat is stationary (Fig. 3).

The method proposed in the invention is based on the theoretical relationship between the heart rate and the frequency of the alternating waves. The ECG is considered to consist of a stationary signal. Let f _{c be the} average heart rate at a given time of the ECG. Under these conditions, the occurrence of an AOT event, which is manifested periodically every two beats, will be reflected in the frequency spectrum at half of the average heart rate f _< / 2. The method proposed in the invention performs the ECG treatment directly in the time domain. The analysis is performed on segmented blocks of L beats. The description of the ECG signal, p _or (t), is based on an additive model that is obtained from the periodic repetition of a single beat q (t). In this way block of L beats, for all L even and without loss of generality, it is expressed as: p _o (t) = ∑q (t-ιτ _b ),

; = or where T _b - is the period of the heartbeat block. Since p (t) is a periodic function, its frequency spectrum P ₀ (O)) consists of a train of pulses at the multiple frequencies of the average beat frequency. P _Q (ω) = 2π∑ a _k -δ (ω - £ -), k = -∞ J- b where a _k are the Fourier series development coefficients of p _ϋ (t).

The appearance of AOT is modeled as an additive component ε (t) of an alternating wave function that is assumed to arbitrary morphology, duration and amplitude. This function is repeated every couple of beats, so its frequency is half of the heart rate. Adding said function to the proposed ECG signal model, the following function is obtained:

L L 12

_J p (0 = ∑ _? (Í- / 2;) + ∑6 (í-2 / r _é X - I e Z, L par,

/ = 0 Z = O whose Fourier transform is:

Where b _k are the coefficients of the Fourier series development of the periodic extension of the alternating wave function. The AOT is manifested in the spectrum as impulse functions that are repeated at frequencies multiple of half of the beat frequency. The objective of the method proposed in the invention is to extract the information of the coefficients b _k of the alternating wave with respect to the rest of the ECG signal together with the noise that could exist. For this, the method of the invention must increase the significance of the harmonics of the AOT in relation to the main harmonics of the signal and the noise.

The method proposed in the invention adapts to different modes of operation or implementations. These implementations can be fixed, portable or implantable material devices. Three types of modes of operation applicable to the three types of implementation of the invention. These are, operation in normal mode, operation in noisy mode and operation in real time. The operation in normal mode is proposed for the analysis of signals in stable conditions such as ECG records taken in medical centers under low noise conditions. The clinical results of this method are the estimated alternating power and the alternating waveform.

The operation in noisy mode is proposed for the analysis of signals in unstable conditions with high noise level, such as signals from long-term Holter registers or stress tests. The clinical result of this mode of operation is the estimated alternating power.

The real-time operation is proposed for the implementation of devices in which the detection of the AOT must be carried out in a known time interval that is sufficiently small for decision making. This mode is indicated in implementations of implantable devices, such as the case of implantable defibrillators and pacemakers, in which the noise level of the intracavitary signal is low. The detection of AOT is used as an added parameter in the detection of ventricular fibrillation phenomena and sudden cardiac death.

The possible implementations of the invention may include one or more of the following characteristics:

- A mechanism for the extraction of ECG signals implemented in the device itself or a mechanism that allows obtaining the signals from another device capable of doing so (195).

- One or several graphic presentation devices that allow the visualization of the physiological signal captured with one or more leads as well as the results of the analysis, such as a screen or any printing medium (215, 220, 210, 225, 235 ). - One or several communication mechanisms that allow the sending of control commands of the device or the transmission of information to another device based on any known wireless or cable communication system (235).

- One or more data storage devices based on solid or magnetic state disks (185), removable or not, that allow storing signals, results, or execution data necessary in the operation of the device. - An operation interface based on a keyboard (200), mouse (205) or operation buttons (230).

The results shown for the estimation of the AOT can be one or more of the following:

The value of the power estimate, which can be given in voltage units such as volts or power such as watts (220).

The zone of the ECG in which the alternation appears superimposed on the ECG in the time axis. Since the amplitude of the alternating wave is negligible with respect to the power of the signal, this superposition will be represented with a different color. To have the information on the power of the wave, color gradients are included that allow differentiating the temporal instants in which there is greater power of the alternating wave from those with a lower power (225).

The estimated waveform of the corresponding alterating wave (215).

The presentation of the results commented on the device can be combined with other physiological data of the signal or of the patient that are considered appropriate (210). Among them, by way of example, the estimation of the noise level of the signal, the heart rate, the respiratory rate, the values of cardiac variability, blood pressure, data referring to the gasometry tests, etc. are included.

DISCUSSION OF THE FIGURES: Figure 1: a schematic example with the sequence of ABABA patterns is shown ... on the repolarization segment that illustrates an AOT episode. Figure 2: in order to characterize the heart rate variability, the histograms of the standard deviation of the heart rate are shown for the same group of ECG signals from ergometry tests for blocks of lengths of 8, 16, 32, 64 and 128 beats. In the ordinate axis the standard deviation values of the heart rate are shown and in the abscissa axis the number of times said deviation values appear. The scales in the axis of ordinates of the previous graph are different to be able to appreciate the distributions of the cases of 64 and 128 beats. As can be seen, for a smaller heartbeat block the standard deviation of the heart rate is smaller. Under these conditions of analysis with the use of heartbeat blocks with lengths of 8, 16 e even 32 beats, lower frequency variabilities are obtained by what. conditions of proximity to the periodicity are assumed.

Figure 3: 128 isolated beats from a Holter register aligned from the QRS complex are shown. As can be seen, the morphology of the waves does not have a great variation. Therefore, it can be assumed that the cardiac signal consists of a waveform that is repeated at certain time points to give rise to a heartbeat. ^•

Figure 4: an example of a window used for the extraction of AOT information is shown. This window is synthesized from the periodic repetition of a

T -, square pulse with a variable duty cycle (20, 25) of - - - 100%.

Figure 5: the spectrum corresponding to a block of 32 beats with a frequency of 71.2 beats per minute (1.8 Hz) is shown. As can be seen, the harmonics that reflect the AOT are located in the odd multiples of half of the beat frequency of 0.6 Hz (40, 45, 50, 55). The signal used in the simulation has been created using additive physiological noise. For the example, a periodic periodic pulse has been used at the beat frequency with a 25% duty cycle. Figure 6: the sequential operation algorithm is shown. The stages of Extraction and / or storage of the ECG (80), adaptation of the signal _, and elimination of noise and artifacts (85), extraction of the AOT information by means of the synthesis of a periodic and poisoned window of the signal are differentiated (90), post-processing for the improvement of AOT detection (95), detection of AOT from the calculation of RAOT. And to finalize the stages that can be implemented jointly or separately for the presentation of results (110) and the sending of data to the output interface (105). Figure 7: the algorithm that allows the extraction of information related to the AOT through it poisoned from a periodic function is shown. For this, the average frequency of the heartbeat block (115, 120) is calculated and a periodic window that will be multiplied by the block (125, 35, 20, 25, 140) is synthesized. In order to increase the resolution and the sensitivity in the detection, the poisoning is performed with a variable number of windows consisting of versions displaced in time from the original (135). The number of windows used will depend on the conditions of the signal as well as the characteristics, processing and memory of the physical device where the procedure is implemented. Figure 8: a design scheme of the stage for the improvement of AOT detection is shown. This stage increases the significance of the information that identifies the AOT. The actions taken are the following: elimination of noise, artifacts and the background EQG (145, 155) and estimation of the trend of the signal (150). Figure 9: the design scheme of the stage destined to the detection of AOT is shown, based on the estimation of the power spectral density of the. Poisoned signals (160), the T wave alternation ratio (RAOT) (165, 170) on which the decision of existence or not of AOT (175) is calculated is calculated. .

Figure 10: A heartbeat block is shown with an arrhythmia in the fourth heartbeat that causes a deviation from the average frequency in the block. It can be seen that the periodic window w (t) correctly adjusts with the ST-T complexes of the first beats but does not synchronize from said pathological event. In this case, the extraction of the alternation information is not done correctly because of which failures can occur in the detection. Figure 11: the method for frequency normalization per truncated period (180) is shown, from which a periodic ECG signal is obtained from a chosen frequency. This method is used to improve the detection of AOT when the heart rate variability of the heartbeat block is high. Figure 12: the possible physical implementations of the device of the invention and the elements that compose them are shown. There are two types of implementations, a device with a screen for data analysis (190) and another version valid for an implantable device (240) such as a defibrillator. Figure 13: a graphical representation interface of the alternation based on the visualization of the AOT information in time is shown. Different colors are used on the repolarization segment in the representation of the ECG without changing its morphology (255). This result is obtained from the parallel design algorithms and its resolution depends on the number of windows used. The representation is obtained by averaging and temporarily weighing the signal blocks that have obtained the highest RAOT. High powers are identified, with one color and low with another (245, 250).

Figure 14: the summary scheme of operation of the device is shown, adaptive modes of operation are included to the conditions of the signal and the actions taken regarding cardiac variability (250) and noise (265). I also know shows the recovery phase of the alternation signal (295) and together with the phases ^" of presentation and / or transmission of the results (300).

MODE OF EMBODIMENT No noise has been included in the ECG model described above. In the case of a noisy scenario, the introduction of additive noise V (Y) of arbitrary probability and power density function is considered. Including the noise, the proposed ECG model is then x (i) = p (t) + v (f), and its spectrum:

X (ώ) = P (ώ) + V (ώ). Prior to the AOT analysis, the method performs a preprocessing stage with the objective of extracting said information. First a poisoning is performed on the ST-T complexes of. The function x (f) through a periodic window w (f) with a cycle

T _w of arbitrary work of —-100%. This window is synthesized from the repetition

periodic of an arbitrary waveform pulse. The chosen period of the window w (t) is equal to the average period of the ECG signal, that is, T _b . The pulse width is defined by the variable T _w . Choosing different width values you get different work cycles. The power spectral density function S _w (ω) of the periodic window ^~ w (i) in the case of the periodic repetition of a rectangular window (Fig. 4) is:

where . sin (kω _b T _w ) _ c _k -, K e ^¿ , kπ y

b τ _b . As a result of the poisoning produced by the multiplication of the periodic ECG with additive noise x (t) and the periodic pulse w {f), the poisoned ECG x _w (t) is obtained, which can be expressed as: χ _w (0 = ^χ (t) ^• Ht) = P (0 ^• w (t) + v (0 ^• w (t).

In order to eliminate redundant or unnecessary information, such as background ECG or low frequency noise artifacts, such as baseline noise, a post-processing stage is applied immediately after poisoned Among the possible actions, a stage consisting in the subtraction of the consecutive repolarization segments extracted is obtained, thus obtaining the differential ECG entangled:

_. x _wι (t) = x _y , (t) - x _y , (t - T _b ) = p _wd W + Vr _t (t),. where,

Y - , . v _wd (O = v _d (Ow (O = [v (Ov (í-7¡)} w (O.

The noise is the variable that can mainly mask the detection of AOT. Once the background ECG has been eliminated, the dependence of the poisoned differential signal on the noise is analyzed in order to evaluate the performance of the proposed method. For this, the statistical properties of the poisoned differential noise function described above are analyzed. If a stationary process is assumed broadly, the autocorrelation function R ₁ ^ (T) of the differential noise v _d {f) is as shown below:

.R _v / r) = 2R, (r) -R, (τ + r _i ) -R, (rr _δ ), where R _v (τ) is the noise autocorrelation function. The power spectral density function is obtained from the spectral density of the noise S _v (ω) as the Fourier transform of R _v (τ)

S _Vd (α>) = 2S _v (ω) - S _v (ω) (e ^JωT> + e ^laT ^. = 2 ^ (O) - (I - COsCiOr ₄ ))

= 4 S _v (ω) • without ² [ω ^ J.

The autocorrelation of the poisoned differential noise function R _wd (t, t - τ) can be calculated from the differential function taking into account c \ μev _wd {f) = v _d (t) - w {t). ^v After wrapping, the resulting process is not stationary but cyclo-stationary [Gar75]. In this way:

R _Vwd (t, t- τ)

τ)} w {t) -w (t- τ)

and the power spectral density function is calculated as the Fourier transform of the average autocorrelation over the time variable.

which in the Fourier domain is equivalent to the convolution:

⁵ U ^ω) = ¿2ΛΓ ^' { ⁵ v> ⁾ * ^S _w ^(β > ⁾ ]

= 4 ^ | c _t | -S _v (ω - kω _b ) -sϊn ² [- -ω - kπ \.

The numerator of the previous expression imposes zeros in the origin and in all the harmonics of the beat frequency. On the other hand, the work cycle of the periodic pulse w (t) can be chosen according to certain criteria that allow fine-tuning the detection of AOT, with values between 5% and 45%. A good compromise solution, by way of example, may be the use of a 25% duty cycle (T _w = T _b / 4). In this case, all odd coefficients of the poisoned differential ECG are canceled:

+ l -, where n is an integer.

The detection of AOT events is subject to the search for the frequency peaks in the harmonics of the multiple frequencies of the alternating wave (Fig. 5), that is, f, / 2 (40), where f _c

and all its odd multiples that compare with the noise in its vicinity. As a result of this analysis, a T Wave Alternation Ratio (RAOT) is obtained on which the criteria that differentiate the existence or not of AOT in the analyzed heartbeat block will be established. This detection can be performed by thresholding, peak detection techniques, use of neural networks ,. etc.

Once the existence or not of AOT is established, the results presentation phase will be carried out where, the device of the invention can calculate a Estimated waveform of the alternating wave by an inverse transformation and filtering of the calculated power spectral density. The design scheme proposed in the invention occurs in five different sequential stages (Fig. 6), which are listed in order of application: i Extraction and / or storage of the ECG (80).

2 Adaptation of the signal and elimination of noise. and artifacts (85).

3 Extraction of the AOT information by means of the synthesis of a periodic and poisoned window of the signal (90).

4 Postprocessed for the improvement of AOT detection (95). 5 Detection of AOT from the calculation of RAOT (100).

6 Presentation of results (110).

7 Output interface (105).

Each of the described steps and the design blocks that comprise them are described below: 1 Extraction and / or storage of the ECG (80).

Its function is to capture the ECG signal that can consist of one or more leads. This capture mechanism is implemented in the device itself (195). Otherwise, it will be responsible for adapting the signals taken by any other device capable of doing so by means of the physical reading of the data or its reception via any means of transmission.

2 Adaptation of the signal and elimination of noise and artifacts (85):

Its function is to adapt the input signal to facilitate its analysis, eliminates noise and artifacts such as baseline noise. It can be implemented by any method existing in the state of the art such as linear filtering, nonlinear filtering or multitasking processing.

^v 3 Extraction of the AOT information by poisoning w (t) (90): Its function is to process the heartbeat block at the input to discriminate

The spectral information outside the AOT. This stage is carried out in parallel in order to improve the resolution of the method. The design scheme is based on the average of the spectra through the poisoning in different areas of the signal. When the window's duty cycle is small (values below 25%), the scanning area on the ST-T complex decreases and it may be the case to cover areas of the ventricular repolarization segment in which there is no AOT being able to exist in adjacent areas. With said objective, the present invention analyzes the signal by means of different displaced versions of the periodic window, calculating the resulting average spectrum of the product of each of them with the ECG block. This scheme, in addition to providing greater robustness against noise, increases the resolution of the detection, eliminating the. effect that can cause the use of windows with small duty cycle. This stage is subdivided into the following blocks to be implemented:

• QRS detection (115):.

Its mission is to detect the QRS complexes of the heartbeat block. Any method existing in the state of the art can be used in its implementation, such as the use of adaptive filters or wavelet transform.

• Calculation of r ₆ (120): -. ^, '

The frequency and standard deviation of the heart rate is calculated. • Synthesis of the window (125):

Its function is to synthesize the window that is used to extract the information from the AOT. For this, a periodic signal is synthesized from a pulse with a period equal to that of the average heartbeat frequency T _b (25) and a duty cycle between 5% and 45%. The choice of the duty cycle is determined based on the required resolution by choosing a value of T _w

(twenty). The poisoned ECG ^" with small work cycles extracts a smaller amount of information from the ST-T complex and is affected by noise to a lesser extent, while the poisoned with larger work cycles is more affected by it. The pulse morphology (3p) from which the window is synthesized, has an arbitrary waveform, such as, for example, hamming, hanning, triangular, rectangular, Kaiser windows, etc. Or non-parametrized waveforms can also be used as is the case of windows adapted with waveforms similar to the ST-T complex. • Delay (130):

Its function is to generate a displaced version of the periodic window synthesized. This block is responsible for introducing a delay r of arbitrary duration. For small r delays, the resolution of the method and its robustness against noise are increased. The choice of the number of displaced versions of the window will depend on the conditions of the signal and the characteristics of the central processing unit (CPU) and memory of the physical device to be implemented. • Wrapped (140):

It is a block destined to realize the product of the signal with the displaced version of the corresponding periodic window.

Post-processing to improve the detection of AOT (95):

This stage has the mission of increasing the significance of the alternation on the noise of the signal by means of a processing so that the detection of AOT is favored. The blocks to be implemented will be one or more of the following: • Elimination of the background ECG (145).

The background ECG is eliminated together with low frequency noise artifacts, by means of the subtraction of the consecutive repolarization segments extracted.

• Estimation of the trend (150): This sub-stage calculates the trend of the signal by partially discriminating the changes that the signal has caused by noise artifacts, so that the information of the AOT is enhanced in the next stage detection. To this end, skillful methods are used for this purpose, as is the case of the empirical mode decomposition (EMD: Empirical Mode Decomposition) [HuaO1] that manages to separate the useful information corresponding to the ventricular repolarization of noise and artifacts. The present invention proposes a method of estimating the trend from the separation that is carried out by means of the partial reconstruction of the signal obtained as the sum of the intrinsic mode functions (MFIs: Intrinsic Mode Functions) that have not been identified as noise . The main problem in carrying out this separation is to identify the useful components. In the present invention, useful information is determined in the EMD domain using the descriptors Hjorth [Hjo70, Hjo73] and the spectral purity index (SPI: Spectrai Purity Index) [Sor05]. Consider x [n], n = 1, ..., N-1, an isolated ST-T complex. The model that describes this part of the heartbeat is considered additive: x [n] = s [n] + v [n], O ≤ n ≤ N- 1 where s [n] is the valid ST-T complex that contains the information corresponding to the ventricular repolarization and v [n] groups the rest of undesirable components. The signal s [n] represents the tendency of the original ST-T complex and corresponds to a slow variation signal, while v [n] is characterized as a rapidly varying signal, associated with higher frequency components. The original signal is represented by the EMD decomposition technique:

L x [n] = ∑c, [/ i] + é¿ [n] _;

'= 1 where c¡ [r \ \ are the MFIs in the EMD domain and q _L [n] the residue. The partial sum of MFIs is used to separate the signal x [n] into its two additive components as follows:

where:

Y-

The MFIs Cj [n] of faster oscillations, from the first order, C ₁ In], to the order (P-1), c _P .i [n], ^• are considered as non-relevant components to describe Ia ventricular repolarization. The procedure for determining the P index is based on the analysis of the spectral purity index. This parameter is calculated as:

SPI = ^ - m ₀ m ₄ where m¡ represents the spectral / -th moment of the power spectrum of the signal x [n]. The SPI index indicates the extent to which one. Signal can be described by a single frequency. In our case, the P index is identified as the first MFI that does not have oscillatory characteristics, starting with higher order MFIs: In other words, the P order of the P-th MFI is obtained when the value of the spectral purity index for That mode takes a certain value.

• Noise Elimination (155):

Within this block, noise is eliminated by filtering techniques.

AOT Detection: RAOT calculation and decision (100): • Estimation of the power spectral density (160): Its. mission is to calculate the total spectrum of the signal, in particular in the vicinity of the heartbeat frequency (0-5 Hz). Any method existing in the state of the art, both parametric and non-parametric, is valid for implementation. • Search and analysis of peaks in the harmonics of / ¿/ 2 (165):

It is the block responsible for extracting the parameters that identify the AOT of the frequencies f¿2 (40), 3f </ 2 (45), 5f _< / 2 (50), 7f _< / 2 (55), ..., and noise in its vicinity (60, 65, 70, 75).

• Calculation of RAOT (170): Based on the extracted parameters, the RAOT is calculated.

> existence of AOT.

• Decision of AOT (175):

The decision on the detection of the AOT is taken from the value of RAOT

. calculated and can be taken through thresholding criteria or through the use of any pattern discrimination technique such as support vector machines, expert systems, genetic algorithms, neural networks, etc.

Presentation of results (110): The results shown for the estimation of the AOT can be one or more of the following:

• The value of the power estimate, which can be given in voltage units such as volts or power such as watts (220).

• The zone of the ECG in which the alternation appears superimposed on the ECG in the time axis (250,255). Since the amplitude of the alternating wave is negligible with respect to the power of the signal. This overlay will be represented with a different color. To have the information of the power of the wave, color gradients (245) are included that allow differentiating the temporal instants in which there is greater power of the alternating wave from those that have a lower power

(225).

• The estimated waveform of the corresponding alternating wave (215).

Output Interface: It is the interface that transmits the information to a user; to another processing stage or to a device, about the existence or non-existence of AOT in the signal. ^'

The proposed design schemes are based on ECG conditions close to the periodicity for the detection of AOT. These conditions are difficult to achieve in some cases. A signal with high heart rate variability can cause a lack of synchronization between the ECG signal and the window w (t) making the extraction of information from the AOT unsuccessful (Fig. 10). It is necessary to generalize the presented signal model, at the beginning in order to correct this possible lack of synchronization effect between the window and the ECG. In this extension the conditions of proximity to the stationarity of each beat are maintained. The variability of the heart rate is introduced to the model from the sequence of interlaced times T _k , k = 1, 2, ..., L, formed by the series of times elapsed between two consecutive R waves {T _RR ). The elements of the sequence consist of the duration of each beat. Assuming that the beats have a standing waveform, the ECG is defined as:

This generalized ECG model corresponds to the model described above in the case of samples with little dispersion over the heart rate, that is, with a small standard deviation with respect to the average heart rate. In this case, conditions close to the periodicity and by

Therefore, the ECG model corresponds to the initially described model.

With the aim of solving the detection problems that are caused by the lack of synchronization between the window and the signal block due to the variability of the heart rate, the present invention includes a further modification of the method by means of the application of a process of signal period normalization

ECG '

There are period normalization techniques in order to keep the period of the signal constant from compressions and temporary expansions. Among those described in the state of the art are [Ram97] and [JyhO1] based on multitasking transformations. Both have the advantage of being invertible being able to recover the original signal from the normalized one by means of the application of an inverse multitasking transformation. Multitask transformations in the frequency spectrum correspond to widening and compression transformations over time and, if they affect ST-T complexes, they can lead to distortions that mask or introduce false AOT episodes. Therefore, these techniques may not be appropriate for the detection of AOT. For example, the appearance of an alternation on the width of the. ST-T complex could disappear if all the complexes are normalized with the same temporal width. The solution proposed in this invention is carried out by normalizing the heart rate by truncating the period (180), which consists in the imposition of a constant period for the heartbeat block. This truncation changes the frequency components of the signal but does not modify the temporal information of. The alternation since it does not modify the morphology of the ST-T complex that remains intact. Let T _{max be} the period imposed in the heartbeat block taken as the maximum of. the interlaced times of the sequence of times T _\ ¿

T „= max {T _k }. m ^aX SAW € [1, L] ^1k -> - ^•

From which the following signal can be constructed with a constant period:

1 = 0

In this signal the effect of the heart rate variability has been eliminated, affecting only the noise component. The procedures and steps described above apply to this standardized period signal. First, the information of the AOT is extracted by means of the poisoning: x _Wwm (t) = x (t) -w (t) = p _nom (t) -w (t) + v (t) -w (t) Subsequently, in order to eliminate artifacts, the consecutive ventricular repolarization segments are subtracted:

The decision about the existence or non-existence of AOT is made in the same way by analyzing the spectrum of x _wd (t). In this case the harmonics of Ia

alternation will appear in multiples of Hz.

The choice of the truncated period of T _max has been proposed in order to ensure that all beats are adjusted for duration within the normalized period and to ensure that the periodic pulse poisoning covers a greater part of the. ST-T complex. Alternatively, depending on the operating specifications of the invention, the value of the truncated period can be chosen arbitrarily, for example taking a fixed and constant value for the entire analysis, and it can be a user configurable value of the device. . For example, in the case of real-time operation, the period used in the normalization must be a default value entered as an input parameter, of the device or as an input value entered by the user of the device depending on the characteristics of the patient in any range of values that allow high enough heart rates. The normalization stage of the described period is applied in cases in which the ECG heart rate variability is high (250) and is performed. subsequent to Ia (120) to the poisoned, synthesizing w (t) (125) from the period T _max chosen. This extension of the method of the invention, by eliminating the effect of the heart rate variability, achieves greater robustness in the detection of AOT and allows to accommodate and solve the different operating scenarios and implementations described.

Claims

1. Method for the detection of the alternation of the cardiac ventricular repolarization characterized by the extraction of the information of the ventricular repolarization by means of the continuous time poisoning of heartbeat blocks of variable length of a bioelectric signal of the heart, ECG, where the poisoning it is performed with a window defined by a periodic signal of adjustable duty cycle.

2. Method, according to claim 1, characterized in that the poisoning process (140) is defined as the product of the periodic window (30) and the signal block.

3. Method, according to any of the preceding claims, wherein the poisoning process of the same ECG block is carried out with different versions of the displaced periodic window (30) (130) with the following delays: 0, T ₁ 2τ,

3τ, ..., mr, where r and m are preset values or parameters entered by the user and cannot exceed the maximum delay value of the repolarization segment duration.

4. Method according to any of the preceding claims, characterized by a periodic window (30) with the following aspects:

• The period of the window (25) can be a fixed value chosen or it can be the average period of the heartbeat block that is poisoned.

• The duty cycle of the periodic window is variable and is between 15-45% (20, 25)

• The periodic window (30) is synthesized as the periodic repetition of a pulse

(35).

• The duty cycle of the periodic window is calculated from the average period of the ECG signal, T _b (20), and the pulse time of the window, T _w (25), as

- ^ -100%.

T _b

5. Method according to claim 4, wherein the pulse waveform (35) can use any parameterized waveform as a window to extract the information, such as the use of hamming, hamming, triangular, rectangular, kaiser or also non-parametrized waveforms can be used as is the case of windows adapted with the waveforms similar to the ST-T complex.

6. Method, according to any of the preceding claims, characterized by a technique to reduce the effect of the heart rate variability by means of a procedure for the normalization of heart rate by means of the use of a truncated period (180).

7. Method according to any of the preceding claims, where a series of post-processing actions of the poisoned signal (95) are carried out in order to highlight the information corresponding to the alternation in relation to the noise.

Method according to any of the preceding claims, in which a method that subtracts consecutive beats (145) is used in order to eliminate noise

Background ECG and low frequency artifacts.

9. Method according to any of the preceding claims, in which the trend of the ST-T segment (150) is calculated to separate the information of the alternation of the T wave, AOT, from the unwanted information.

10. Method according to any of the preceding claims, in which a method is used that by filtering the signal eliminates noise artifacts (155).

11. Method according to any of the preceding claims, wherein the power spectral density (160) of the poisoned signal (90) and postprocessed (95) is estimated. , ^:

12. Method according to (to claim 11, in which from the search and analysis of "peaks in the harmonics of / i / 2 (165) the coefficient of alternation ratio of the T wave, RAOT (170) is obtained ), where f _{c is the} average frequency of the ECG signal.

1.3. Method, according to claims 11 and 12, in which from the relationship between the odd harmonics of / j / 2 (40, 45, 50-, 55) and the noise in its vicinity (60, 65, 70, 75) RAOT coefficient is calculated.

14. Method, according to any of the preceding claims, which determines the existence or not of alternating ventricular repolarization from the RAOT coefficient (175).

15. Method, according to any of the preceding claims, which calculates the results on the existence of alternating repolarization from the power spectral density of the poisoned and post-processed signal (165) and the RAOT coefficient (170).

16. Method according to claim 15, characterized in that the estimation of the signal corresponding to the alternating waveform, ε (t), is calculated and visualized. obtained by filtering the poisoned spectrum from the odd harmonics (40, 45, 50, 55) located in multiples of half of the frequency chosen in the periodic window (20).

17. Method according to any of the preceding claims, wherein when the duration of the RR intervals, that is to say the intervals between consecutive R waves, of the beats of a block exceeds a certain percentage of the average value of the duration of the all the intervals, the normalization of the period of the heartbeat block (180) is performed.

18. Method according to claim 17, wherein the percentage from which normalization is performed is 10%.

19. Method according to claims 17 and 18, wherein the normalization period is chosen as the maximum of the RR intervals of the heartbeat block or as a fixed value that must be greater than 0.65 seconds and which can be entered by the device user (180).

20. Method according to any of the preceding claims, in which an empirical decomposition technique is used that allows the components that characterize the AOT to be separated from the noise components (140).

21. Method, according to claim 20, in which the characterization of the signal is operated as the sum of intrinsic mode functions.

22. Method according to claim 20, wherein the signal is separated from the noisy components by means of the Hjorth descriptors and the spectral purity index.

23. Device for the detection of the alternation of the cardiac ventricular repolarization characterized by including means for the extraction of the information of the ventricular repolarization by means of the continuous time poisoning of heartbeat blocks of variable length of a bioelectric signal of the heart, ECG, where the poisoning is done with a window defined by a periodic signal of adjustable work cycle.

24. Device for detecting the alternation of cardiac ventricular repolarization according to claim 23, characterized by including means for performing the steps of the method according to any of claims 2-22.

25. Device, according to claim 23, characterized by operating in an adaptive or predetermined manner depending on the operating scenario in which it is located and its operating requirements, the operation modes being: ^•

• Normal mode operation, intended for the analysis of signals or signal blocks with low noise level and stable conditions of heart rate variability, in which the periodic window is generated at the average heartbeat frequency. '• Operation in noisy mode, intended for the analysis of signals or signal blocks with high noise level, in which the period normalization (180) is performed taking the normalization frequency as the maximum RR interval of the analyzed signal block.

• Operation in real time mode.