JP2000509618A - Prediction of cardiac arrhythmia by detecting weak potential - Google Patents

Abstract

(57) Abstract Electrocardiographic signal collection and analysis non-invasively detects and quantifies the presence of abnormal cardiac status patterns in patients at risk for heart disease. Signals from orthogonal (X, Y, Z) surface leads are amplified (44,36,48), digitized (40) and stored for later processing or processed directly. The R-wave can be elicited, matched and globally averaged for patients at ventricular pathological risk with the incoming beat, or P-wave for patients at atrial pathological risk. Can be triggered, matched, and averaged overall. The QRS onset and offset and the P-wave onset and offset are calculated for ventricular and atrial post-analysis applications, respectively. A windowed Fourier transform of the second derivative (acceleration) of the signal averaged ECG is calculated for the region of interest of each lead. The spectral change index calculated from the "acceleration spectrum" for each lead serves to quantify the degree of spectral splitting within a given bandwidth.

Description

Description: BACKGROUND OF THE INVENTION The present invention relates to electrocardiography (ECG), and more particularly to the use of computer-aided analysis of selected signal components in electrocardiogram recordings. An improved electrocardiographic device and method for predicting possible ventricular tachycardia and other forms of cardiac arrhythmia by analyzing all of the ECG segments and intervals, comprising preamplification and impedance An electrocardiographic apparatus and method for amplifying, normalizing, analog-to-digital converting, storing in a memory, and operating with a computer to provide frequency information about selected components of the signal after performing the transformation. Myocardial infarction (M. I. After), the patient is at risk for ventricular tachycardia due to the occurrence of a dangerous disorder of cardiac rhythm. Acute cardiac arrhythmias, especially sudden death due to ventricular tachycardia, are a major risk in the first few hours after a myocardial infarction. During the first few days after myocardial infarction, the incidence of ventricular arrhythmias is about 90%. After the first few days, the incidence of arrhythmias decreases significantly, but there is still a substantial risk for patients with myocardial infarction. 5-10% of post-infarction patients die within one year from cardiac death due to sudden rhythm. Statistically, 50% of myocardial infarction patients die from ventricular tachycardia without treatment. Earlier methods to identify this risk group, such as long or Holter ECG, exercise tests, resting ECG, are either insensitive or ineffective. In the 1970's, weak signal fluctuations (of about 1-5 μV amplitude) were discovered on the ECG surface at the end of the QRS complex during sinus rhythm. It is significant that these so-called delayed potentials occur more frequently in post-infarction patients at risk for dysrhythmia than in patients with a better prognosis. U.S. Pat. No. 4,422,459 to Simson teaches capturing a series of continuous ECG waveforms, converting them to digital form, and averaging the whole (after removing abnormal or unusual waveforms). Teach providing a relatively noise-free composite waveform. Simson filtered the averaged waveform using a high pass filter with a break frequency of 25 Hz, inverted or bi-directional filtration. Simson then calculated the root mean square (RMS) value of the voltage in the filtered tail segment of the QRS complex. Simson also used RMS voltage measurements and QRS width measurements as indicators of whether a patient may have ventricular tachycardia. The Simson method does not seem applicable to patients with leg blocks. This method always balances between significant high frequency components at the trailing edge of the QRS complex, which may indicate a predisposition to ventricular tachycardia, and noise due to power line disturbances or skeletal or smooth muscle, etc. I can't tell. To extract more useful diagnostic information from the ECG signal, researchers have used the Fourier transform to process the ECG waveform. For example, M. E. Cain et al., "Quantification of Differences in Frequency Content of Signal-Averaged Electrocardiograms in Patients With Compared to Those Without Sustained Vent ricular Tachycardia" (American Journal of Cardiology, Vol. 55, 1500, June 1, 1985). Please refer to. It includes frequency domain representations of ECG X, Y, Z lead signals from patients with and without ventricular tachycardia. The energy, expressed as the square of the voltage, is plotted against frequency. Ca in et al.'S analysis showed that the long 40 ms end portion of the QRS complex and the entire ST portion of the average ECG waveform preprocessed by the Blackman-Harris window function to minimize spectral leakage were included. The segment was limited to a single Fourier transform. Cain et al.'S analysis of frequency domain data was limited to the frequency range of 20-50 Hz. Cain et al.'S principle for performing Fourier analysis that treats such long segments as a single unit was "enhancing frequency resolution." Frequency domain methods for positioning small potentials often include ambiguous voltages due to the relative size of the QRS complex in the ECG waveform. According to the Fourier transform method, even small segments containing little information must be windowed, and further reduced in information content to help determine the presence or absence of weak potentials in the selected frequency range. This results in a frequency domain display with too much resolution. A fitted autoregressive mathematical model for calculating the frequency power spectrum was used for ECG signal frequency analysis. These mathematical models can be obtained using the so-called "maximal entropy method" (MEM) or the so-called "adaptive filter decision" (AFD) associated with the so-called "fast adaptive forward / backward least squares method" Enable analysis to be performed. See U.S. Pat. No. 5,211,179 to Haberl et al. Such methods have been developed to calculate the optimal order of the autoregressive model and to remove interfering low frequency fundamental vibrations. The aforementioned signal processing methods are particularly well suited for ECG analysis. U.S. Pat. No. 5,109,862 to Kelen et al. Discloses a system capable of receiving X, Y, Z lead ECG signals, digitizing the signals, and averaging the signals. The second derivative or "acceleration" of the signal is then derived, and a Blackman-Harris window is applied to the resulting vector, which is then subjected to a fast Fourier transform (FFT) to produce a spectrum. Kelen et al. Processed the tail portion of the QRS signal to find the delayed potential and relied on samples of the overlap time segment. In a sense, we performed a “spectral entropy” calculation related to the topography of the pseudo-three-dimensional spectral plot. However, Kelen et al.'S teaching is not directed to measuring the inter-QRS signal itself. U.S. Pat. No. 5,092,341 to Kelen discloses a system related to the system disclosed by Kelen et al., Supra. The Kelen system is aimed at addressing the limitations of traditional delay potential analysis systems, especially time domain systems such as the Simson system. Kelen states that the abnormalities due to arrhythmias are such that the frequency spectrum of the QRS wavefront velocity propagates throughout the ventricle around the region of abnormal conduction, which results in a high degree of spectral turbulence, and the It is disclosed by frequent sudden changes. The high spectral disturbances of the entire QRS signal during sinus rhythm can be dissected for reentrant cardiac tachycardia independently of the detection of delayed potentials in the terminal QRS region of the digitized and averaged ECG signal. It was considered to represent an electrophysiological substrate. Observations, measurements, and calculations were generally performed on the QRS complex as a whole, and not on any part arbitrarily identified by temporal frequency or amplitude characteristics. U.S. Pat. No. 5,271,411 to Ripley et al. Discloses a system for detecting premature ventricular contractions and quantifying cardiac arrhythmias. The Ripley et al. System includes a plurality of channel connectors connected to a multiplexer that supplies analog ECG signals to a filter bank. In general, by extracting morphological information from electrocardiogram signals to classify heart beats into similar grades, information on morphologically similar beats is used to determine that certain beats, such as early ventricular contractions, are normal. It was possible to assist in judging whether the object was abnormal or abnormal. U.S. Pat. No. 5,199,438 to Pearlman discloses a microcomputer-based system for measuring cardiac power. Pearlman measured the pumping power of the heart by the time-varying rate of mechanical energy generated by the left ventricle of the human heart. The system includes a closed cuff blood pressure monitoring system, a gamma camera, an ultrasound cardiography system, an electrocardiogram system, and a Doppler ultrasound blood flow sensor that supplies all signals to the microcomputer. Pearlman defined the force of the heart as the derivative of the product of the volume of the heart with the cardiac or aortic pressure over time. The cardiac power index is the slope of the portion of the time versus force curve from the onset of systole to the moment when maximum force occurs. The cardiac power index is used to measure ventricular performance during the ejection portion of the cardiac cycle. U.S. Pat. No. 4,680,708 to Ambos et al. Discloses a frequency domain method for delayed potential analysis that uses a Fourier transform on signal segments in the terminal QRS domain. The area under the high and low frequency bands was calculated from the resulting spectral curve and the area ratio was measured. More recently, Ambos et al. And Cain et al. Attempted to study single-segment FFT results of electrocardiogram (SAECG) signals averaged with a selective bandpass filter from the entire cardiac cycle. They have addressed the problem of extracting information about weak potentials within the QRS complex, which is generally obscured by the relatively large and high swirl rate QRS configuration, by applying a digital bandpass filter. However, the frequency range they concern was limited to 70-128 Hz and lower. US Pat. No. 5,215,099 to Haberl et al. Discloses a spectral time mapping (STM) of a windowed Fourier transform for overlapping segments of a terminal QRS, and an early method for delayed potential analysis in the frequency domain. The ST area is taught. The result was a representation of a pseudo three-dimensional time-frequency-force spectral density (PSD). When STM is used with the standard time domain method, the Simson method, it has been found that STM improves the sensitivity and specificity of SAECG. However, STM may be less reproducible due to the high noise sensitivity and the dependence of the resulting STM on accurate QRS terminal measurements. In spectral turbulence analysis (STA), Kelen calculated the speed of the SAECG data and then used a short-time Fourier transform on multiple overlapping segments, including segments in the QRS complex. As in Haberl's STM method, the result was again a representation of the three-dimensional time-frequency-force spectral density (PSD). However, this technique has the disadvantage that the frequency resolution is limited by a short time length, especially after multiplication by the window function. Furthermore, the calculations required to quantify "spectral disturbances" are relatively complex, and the resulting indices are difficult to understand when viewing the resulting three-dimensional map. Slurs and nodules in the QRS complex were found to be strongly correlated with the presence of myocardial infarction and coronary artery disease. Flowers et al. Strengthened this association by correlating QRS nodal records with anatomical confirmation of infarct scars. For example, Flowers et al., "The Anatomic Basis of High Frequency Components in the Electrocardiogram" (Circulation, 1969; 39: 531) and Flowers et al., "Localization of the Site of Myoccardial Scarring in Man by High-Frequency Components" , 1969; 40: 927). Historically, narrowband analog filters have been used in an attempt to characterize the frequency bands of normal, nodular QRS complexes. Atrial fibrillation is thought to be due to the reentry route to the atria. Patients with atrial flutter or atrial fibrillation can have cerebrovascular disorders or strokes that can be disabling or even fatal. For atrial fibrillation to occur, the atrial contractile matrix also requires a slow conduction area to initiate and maintain the reentry circuit. A non-invasive risk assessment of atrial fibrillation was recently examined using a P-wave-induced or aligned SAECG. Due to the lack of commercial signal processing packages for frequency domain analysis of P-waves, only the time domain approach was used. The main criterion has been reported by Guidera and Steinberg, Fukunami et al., And other separate researchers to be the extended and signal averaged total filtered P-wave duration. For example, A. Guidera and J.S. S. Steinberg, "The Signal-Averaged P Wave Duration: A Rapid and Noninvasive Marker of Risk of Atrial Fibrillation" (J Am Col Cardiol, 1993; 21: 164 5-51) and M.S. See Fukunami et al., "Detection of Patients at Risk for Paroxysmal Atrial Fibrillation During Sinus Rhythm by P Wave-Triggered Signal-Ave ragged Electrocardiogram" (Circulation, 1991; 83: 162-169). Since the time domain criteria proposed by these authors are specific to their technique used, no filtering standard has yet been established. The extended P-wave duration in the time domain is considered to be a non-invasive indicator of the presence of delayed activation in the atria. In patients with mitral stenosis or Wolf Parkinson-White syndrome, it is difficult to determine the end of the P-wave for time domain analysis. Thus, Yamada and Fukunami et al. Report a spectral method for P-wave analysis using area ratios, similar to the method used by Cain and Ambos for delayed potential analysis. T. See Yamada et al., "Characteristics of Frequency Content of Atrial Signal-Averaged Electrocardiograms During Sinus Rhythm in Patients with Paroxysmal Atrial Fibrillation" (J Am Col Cardiol, 1992; 19: 559-63). They did not perform the analysis above 50 Hz. However, their technique was not used extensively and had lower sensitivity and specificity than time domain P-wave analysis. This method of delayed potential analysis utilizes both a time domain method and a frequency domain method. The existing frequency-domain method is a sensitive and sensitive measurement of the QRS end, or the derivative of some statistical exponents derived from complex mathematical algorithms including multiple-slice correlation statistics of "spectral perturbation" information. Need. However, all such techniques are as unacceptable as the usual time-domain Simson method when comparing the reproducibility, sensitivity, and specificity of the results. This time domain method also has its limitations. It cannot be used to analyze SAECG signals in patients with conduction delay problems and has poor positive predictive accuracy. Delayed activation and abnormal recovery are well documented during the "delay potential" test. These are described as superficial manifestations of changes in conduction patterns through gangrenous or fibrotic heart tissue in patients after myocardial infarction with the risk of sudden cardiac death. Since healthy heart tissue forms a more homogeneous conductive matrix compared to damaged tissue, SAECG signals obtained from patients without previously myocardial damage should not reveal fragmentary conduction patterns . In the case of damaged tissue, the layer of damaged or gangrenous tissue disrupts or “disperses” the conduction wavefront, forming a small amplitude, high frequency “interference pattern” along the conduction path. Having this type of tissue matrix can lead a patient to a life at risk for re-entrant ventricular infarction. Atrial flutter or fibrillation also derives from matrix support in the reentry pathway. Such patients have a high incidence of cerebral vascular embolism or stroke. Over the past decade, SAECG has become a valuable non-invasive tool for identifying patients with or suspected of having ventricular venous vein. Simson first described the use of SAECG to determine the presence of a delayed potential as a small amplitude, high frequency signal in the terminal QRS and early ST segments of the SAECG. These signals are believed to be consistent with delayed cardiac activation and are believed to indicate the presence of the underlying myocardial matrix for reentrant ventricular venules. Simson first found a delayed potential in a patient after myocardial infarction with persistent ventricular tachycardia. The presence of the delayed potential was found to be indicative of evocable ventricular tachycardia during electrophysiological testing, as well as other post-myocardial arrhythmia events. Signal averaging was also used to examine patients with unexplained syncope. In addition, signal averaging was used with left ventricular ejection fraction (LVEF) testing to manage survivors of acute myocardial infarction. If the patient's LVEF is greater than 40% and the SAECG time domain result is negative, further expensive electrophysiological tests are not recommended for the patient. How accurate is the SAECG test in predicting what will happen if the patient is positive for sudden, provocative or persistent ventricular tachycardia? Is a problem. Although the SAECG time domain results have very high negative predictive values, these result values also unfortunately have low positive predictive values. SUMMARY OF THE INVENTION In the presence of damaged myocardial matrix, weak potentials should be detected not only in the terminal QRS and early ST regions or late potentials, but also during the early stages of ventricular depolarization, ie, throughout the QRS complex. It can be assumed that there is. The detection and analysis of all high-frequency weak potentials of ECG segments and intervals has proven useful, but the identification of very small potentials in waveforms has reached the limits of technical feasibility. The steep upslope and downslope of the QRS waveform morphology has prevented accurate inspection of the entire QRS signal by conventional frequency domain methods. Techniques limited to the analysis of terminal QRS and early ST segments do not detect diagnostically significant signals in other ECG intervals, such as signals within the QRS complex itself. Thus, a robust but simple frequency that addresses the weaknesses of the current frequency domain method while at the same time analyzing the delayed potential domain as well as other ECG segments corresponding to ventricular depolarization and repolarization There is a need for a domain method. Once the task of enabling analysis of the region within the QRS has been met, analysis of other ECGs, such as the region within the P-wave, is then performed using the same basic method. While both of the above prior art approaches to this problem confirm that the presence of high frequency components in various parts of the ECG waveform can identify patients susceptible to cardiac arrhythmias, these approaches are easy to understand and use. The practical problem of providing physicians with clear, unambiguous and reproducible information in a form that is not clearly addressed. Thus, all ECG segments and intervals are analyzed to determine weak potentials to screen and identify cardiac patients at risk for potentially atrial and ventricular arrhythmias, including those with leg blocks. It would be desirable to indicate a cardiac arrhythmia by detecting. One aspect of the present invention provides an apparatus and method that can analyze not only the terminal QRS and the early ST region for a delayed potential, but also the region within the QRS of an R-wave-induced SAECG due to the presence of a high-frequency weak potential. It is. These signals indicate the presence of electrical fission in the myocardium and indicate a patient at risk for ventricular tachycardia or other arrhythmias. It is a further object of the present invention to enable the acquisition of P-wave induced SAECG and the frequency domain analysis of P-waves for examining patients at risk for atrial fibrillation and flutter; Instead, it is intended to allow spectral fluoroscopy and analysis of the QRS region instead. In addition, the device and method are useful for analyzing all other SAECG regions, including the HIS bundle (PR interval) and T-wave regions. Besides the recognition that frequencies of 50 Hz and above are useful and meaningful, the analysis of such high frequency components requires the determination of the background levels of such components and the tracking of the QRS complex of these background levels. And a comparison with the ST portion, whose presence can be indicative of tachycardia. The presentation of the information to the physician is prepared for easy understanding and use without the need for extensive training. One embodiment according to the present invention derives a high frequency signal "sign" of the damaged myocardium throughout the QRS complex. Instead of measuring the area underneath the spectral curve and calculating the area ratio as by Cain, or calculating the normality factor as by Haberl, or interslice correlation as by Kelen, this is called the `` spectral change index '' Is used to quantify the increase or decrease in the presence of spectral peaks in the range of 50-300 Hz. This does not require precise terminal or other reference point determination or pre-filtering of the coarse signal, and therefore provides the necessary tools for analysis both within the P-wave and within the QRS. It uses a large window size (eg, 300 points for intra-QRS analysis) to optimize the performance of the Fourier transform. A novel frequency-domain method for non-invasive detection and analysis of myocardial fission is to examine large spans of the digitized signal in a particular ECG segment. Use the second derivative (acceleration), thereby preserving the frequency resolution of the transform. Thus, this technique does not limit the analysis to the delayed potential region of the SAECG, but allows spectral fluoroscopy and analysis of the QRS region, the P wave region, and other ECG intervals. The index of the degree of spectrum splitting called the spectrum change index is an integer k = 50 to 299, for example, 50 to 300 Hz, and the sum of the absolute spectrum amplitude (SA) at the k frequency point minus the SA at the k + 1 frequency point is calculated as , Divided by the maximum SA in that band, and then multiplied by 100. The advantage of having a frequency domain method for P-wave analysis is that time domain measurements, especially filtered P-wave durations, can be used to improve the risk stratification of patients prone to atrial fibrillation and flutter. It can be used with. In brief summary, the present invention provides an apparatus and method for acquiring and analyzing electrocardiographic signals to non-invasively detect and quantify the presence of abnormal cardiac conduction patterns in a patient at risk for heart disease. About. A windowed Fourier transform of the second derivative (acceleration) of the signal average ECG is calculated for a particular region of interest for each lead, including the QRS region, ST region, T-wave region, and P-wave region. A quantitation index, or spectral change index, is calculated from the resulting "acceleration spectrum" for each lead as well as the composite (X + Y + Z) lead, and the spectrum "split" is quantified to determine the potential risk of a fatal heart condition. Classify a patient hierarchically. Further details of preferred embodiments of the present invention, further objects and advantages of the present invention are set forth in the drawings, the following detailed description, and the appended claims, which are incorporated in and constitute a part of this specification. Constitute. The invention itself is defined in detail in the claims. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a general block diagram of a device embodying the present invention and having a signal collecting device for collecting ECG signals from a human body and a computer for processing these signals. 2A to 2F are diagrams showing ECG waveforms, particularly attributes of a single QRS complex, for explaining the operation of the method and apparatus shown in FIG. 3A and 3B are flowcharts of some operations performed by the computer shown in FIG. FIGS. 4A and 4B show normal X-lead ECG time-frequency plots for a person with a healthy heart, respectively. FIGS. 5A and 5B respectively show abnormal X-lead ECG time and frequency plots for patients after myocardial infarction, ie, patients at potential risk of fatal cardiac arrhythmia. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Referring to the drawings, and in particular to FIG. 1, an apparatus embodying the present invention is shown generally and designated by the numeral 10. Apparatus 10 includes an IBM PC compatible digital computer 12 to which an electrocardiogram signal collection device 14 is connected. Computer 12 includes a system bus 16 for carrying data, address, control, and timing signals between the various modules of computer 12, all of which are known to those skilled in the art. A microprocessor 18 is connected to the system bus 16 for communication over the system bus 16. A keyboard device 20 and a mouse device 21 are connected to the system bus 16 for communication with the system bus 16, and a storage mechanism 22 is also connected. The storage mechanism 22 includes a read-only memory (ROM) and a random access memory. (RAM), including a video display card or graphics adapter 23 and display 24, and a printer 26. A disk controller 28 is connected for communication with the system bus 16, and a floppy disk 30, an optical disk 31, and a hard disk 32 are connected to the disk controller 28 for communication with the disk controller 28 as storage devices. All of these above elements constitute a conventional microcomputer system 12. The system bus 16 is also connected to a portable signal collection device 14 via a detachable serial parallel PCMCIA data link input / output port 35, the signal collection device 14 comprising a real-time collection module, such as a model 1200EPX or a computer peripheral card, For example, Arrhythmia Research Technology, Inc. of Austin, Texas. Models LP-Pac Q, Predictor I and Predictor IIc No. averaging systems available from Alternatively, signal acquisition device 14 may include an analog or digital storage device in an off-line mode, such as a Holter recorder data playback device. The signal collection device 14 includes a microprocessor 41, and the microprocessor 41 has a storage device and a RAM memory 43 for storing the contents of the storage device for a long time using a battery as a power source. The device 14 also includes an analog-to-digital converter 40 connected to a multiplexer 42, both controlled by a microprocessor 41. Multiplexer 42 is connected to an X-lead bipolar ECG amplifier 44, a Y-lead bipolar ECG amplifier 46, and a Z-lead bipolar ECG amplifier 48, which are connected to the patient as known to those skilled in the ECG art. Connected to the patient via respective bipolar leads X, Y, Z and a ground lead G to sense an electrocardiographic signal from the body of the patient. In practice, a nurse carries the device 14 to the patient, records the patient's ECG, and then plugs the device 14 into the input / output port 35 of the computer and commands the two microprocessors 18, 41 to be digitized. The device 14 is connected to the computer 12 by transferring the ECG signal from the RAM 43 to the RAM portion of the memory 22 and finally to the hard disk 32 under software control. Each amplifier 44, 46, 48 is a low noise electrocardiographic amplifier that amplifies and passes (at half the sampling rate or less) signals from near DC to an upper limit in the range of about 300-500 Hz. These amplifiers 44, 46, 48 are also designed to electrically isolate the conductors X, Y, Z from all power sources and from each other. The multiplexer 42 and the analog-to-digital converter 40 sample the voltage levels at the outputs of all three amplifiers 44, 46, 48 1000 to 2000 times per second, and digitally output from each output of the three amplifiers 44, 46, 48. Signal samples can be created. In the signal collection device 14 used in the above embodiment, for example, the above-mentioned LP-Pac Q, the 0.1. 625 μV LSB, resolution 16 bits, dynamic range +/− 20. 28 mV (analog-digital conversion). The digital samples collected in this manner exhibit a continuous ECG waveform. These samples are first stored in RAM memory 43, and each of the continuous ECG waveforms is QRSed in a manner similar to that fully described in US Pat. No. 4,442,459 to Simson. Samples that can be matched to the template and fall into a waveform that does not match the template or other abnormal waveforms are not retained. The remaining digital samples representing the continuation of the input ECG waveform are then split into a plurality of digitized signal arrays, each array containing samples for one ECG waveform. The digitized signal in each of these arrays is placed in phase with the signal from the other array, and then the amplitude of the signal value in the array representing the waveform amplitude at that time (relative to the QRS waveform). Is averaged over many waveforms (eg, over 200 waveforms) to minimize noise from muscle potentials, power lines, etc. This averaging process is preferably performed at a resolution of at least 12 data bits. The resulting single array of averaged signals is ultimately transferred from RAM 43 and stored on the surface of optical disk 31 or hard disk 32 through input / output port 35. The array of averaged signals may simply include only the X-lead waveform, or simply the Y-lead waveform, or simply the average of the Z-lead waveform, or alternatively, these individual monitored X, Y, and Z waveforms. It may include the sum of all of the waveforms. An array representing the time average of a series of ECG waveforms taken from a single patient is then analyzed according to the teachings of the present invention. The array of averaged signals includes, for example, 800 12-bit samples, each sample being spaced apart from each other by one millisecond in time. This preferred type of embodiment includes amplification (using a low noise amplifier), filtering (with a low pass filter having a cutoff frequency in the range of 300 Hz to 500 Hz), and (at a rate of 1000 12-bit digital samples per second). It begins by digitizing and averaging a large number (eg, 100) of ECG waveforms to eliminate abnormal waveforms. The data acquisition method employed uses a bipolar orthogonal XYZ lead system that provides signals for signal averaged ECG data acquisition. A conventional bipolar orthogonal XYZ (Prank) lead system is used for SAECG data collection. However, other lead arrangements are advantageous and can be used, for example, to detect small amplitude P-waves that are difficult to grasp "Lewis leads". The signals are digitized and summed for signal averaging to provide a time-varying signal averaged electric dipole signal for the X, Y, and Z directions. The rate of change of each of the electrical vectors is equal to the first derivative of each component with respect to time. The "acceleration" of the signal averaged electric field vector is equal to the second derivative of each component with respect to time. When a sampling rate of 1 kHz is employed, the region within the QRS is generally defined by a lower limit of about 150 ms before the QRS offset and an upper limit of about 150 ms after the QRS offset, and a window of about 300 ms. Make width. By applying a Fourier transform to the second derivative of each lead signal with respect to time, the acceleration spectrum can be seen within the relevant time window. Apply a Fourier transform with zero pad points over 512 samples or points. Blackman-Harris tapered windows can be used to reduce spectral leakage. Once the spectral acceleration has been calculated, an undisclosed algorithm checks the relevant bandwidth from 50 Hz to 300 Hz. The following steps make up the signal acquisition. a. There is no X, Y, Z quadrature read preamplification and ECG amplification, nodal filtering. b. Digitization of the signal by an analog-to-digital converter with any one of 12, 14 or 16 bit resolution. The data can be stored on Holter tape and then played back to an amplifier for digitization, or stored as a digitized signal on a digital Holter device, or on a non-readable medium such as a readable or writable optical disk or hard disk. Note again that the data can be stored on a portable digital storage device. c. Evoking either an R-wave or a P-wave, respectively, depending on whether the end user wants to analyze for ventricular or atrial arrhythmias. When examining P-waves for the presence or absence of atrial pathology, SAECG should be acquired using the P-wave trigger-P-wave template method. This is necessary for preserving the weak potential of the atria. If the preservation is not performed, the weak potential is set to P-R when the analysis of the P wave is performed on the signal averaged ECG in which the R waves are arranged. It may be erased by jitter. d. Form the best template from the first few incoming beats. e. Subsequent incoming beats are selected and matched using a matching (eg, cross-correlation) technique. f. Population averaging is performed until the stopping criterion is met, that is, the noise level is sufficiently low or a certain number of beats are averaged. g. If the desired population average is triggered on the R-wave, then the QRS onset and offset are found. If the population mean is triggered on the P-wave, then the P-wave onset and offset are found. Such an algorithm was developed as a variant of the algorithm used to find the QRS onset and offset. h. The population averaged patient data, onset and offset parameters, and other patient information are then stored in storage on the personal computer. It should be noted that the hardware configuration shown in FIG. 1 in which computer 12 is used with signal collection device 14 is only one of many possible configurations available when implementing the present invention. For example, all of the above could be implemented by the computer 12 rather than by the device 14, and the device 14 could be a simple accessory of the computer 12. The signal array could be transported from the device 14 onto a floppy disk of the computer 12, or could be sent over a parallel data transmission channel, an infrared optical link, or a wireless or magnetic coupling link. If the signal collection device 14 were sophisticated enough to perform the remaining steps and had a suitable printer and / or display device, the computer 12 would have been completely eliminated. Turning now to FIGS. 2A-2F, myocardial activity is illustrated with ECG waveform attributes, particularly those of the QRS complex, to illustrate the weak potential in the QRS resulting from myocardial activity of damaged heart tissue. Indicated by 2E. 2A and 2B show atrial and ventricular contractions, respectively, caused by myocardial contractions triggered by electrical stimulation to cause blood to flow alternately into and out of the heart. FIG. 2C shows the myocardium, or the middle muscular layer of the heart wall, as a schematic diagram of the action potentials traveling down the heart wall, the heart wall showing some of the ion currents indicated by the ion flow circles, Causes depolarization of the heart wall in the direction indicated by the arrow. Thus, the propagation of action potentials in the heart wall results in contraction of the myocardium through electrical stimulation, as shown. The action potential action constitutes a current flowing through the circulatory path through the heart tissue, causing a voltage drop due to the polarity shown. The area ahead of the propagating action potential is positive relative to the area behind it, resulting in a potential difference in the tissue. At the same time that the current flows, the action potential propagates mainly during the QRS wave of the ECG or during the recovery period of the T wave. As will be appreciated, conventional ECG devices measure such potentials on the surface of the body. FIG. 2D illustrates and identifies ECG segments and intervals of a waveform cycle during one cardiac cycle. For purposes of illustration, the P-wave (atrial depolarization) and QRS segment (ventricular depolarization) include the atrial and ventricular contractions described by FIGS. 2A and 2B above. FIG. 2E shows the ECG waveform for a single heart beat. However, here the ECG waveform represents an episode that is indicative of a patient at risk of developing a dangerous disturbance of cardiac rhythm (ventricular tachycardia). Here, the delay potential (L. P. s) is shown in the ST segment found after ventricular depolarization of the QRS segment. Of particular interest in connection with the embodiments described above, the weak potential within the QRS is also shown in the QRS complex shown in FIG. 2E and is represented as notches and slurs. The underlying QRS signal has a lower frequency but a much higher magnitude / amplitude than the weak potential signal. As discussed further, these relative magnitudes often obscure or impede the ability to identify weak potentials in the associated QRS. Turning to FIG. 2F, a schematic block diagram illustrates electrical conduction through healthy tissue 80 and damaged tissue 82 representing heart tissue. As modeled, electrical conduction through healthy heart tissue results in an unattenuated signal through the tissue, while electrical conduction through damaged tissue results in a weaker potential, e.g., smaller amplitude and lower frequency. Higher signal. As mentioned above, delayed activation and abnormal recovery result in the surface manifestation of changes in conduction patterns through gangrenous or fibrous tissue in patients after myocardial infarction or at risk of sudden cardiac death. Become. This is manifested as a fragmented conduction pattern in a representative frequency spectrum due to the decomposition or diffraction of electrical conduction through damaged tissue. Thus, quantification of the degree of this spectral fragmentation can help to grade patients with a potentially fatal cardiopathic risk. The signal processing methods and apparatus of the described embodiments detect fragmentation of conduction patterns in gangrenous or fibrous tissue by various signal artifacts during both the depolarization and repolarization phases of the cardiac cycle. It is desirable to extract low amplitude, high frequency, intra-QRS signals, such signals having a relatively large QRS form, abrupt QRS waveforms that in the past prevented close inspection of the QRS signal by conventional frequency domain methods. The rising and falling slopes are often obscured, so the increase or decrease in the presence of spectral peaks in the 50-300 Hz range is quantified using the spectral change index. This method does not require the determination of an accurate terminal or reference point, nor does it require pre-filtering of the coarse signal. This methodology is implemented by the principle of detecting a diffraction or decomposed conduction wave tip that manifests itself in the form of small amplitude, high frequency "interference fringes" along the conduction path. Weak potentials are the result of diffraction or delay occurring in damaged or gangrenous heart tissue. Spectral analysis of the intra-QRS analysis region or the intra-P-wave region is typically performed in system default mode by a 300 ms time segment that starts 150 ms before the QRS offset and ends 150 ms after the QRS offset. Defined. The end user can redefine the position and duration of the analysis window with a simple keyboard interaction with the device. In the same way, there are defaults for P-wave and ST-segment analysis, but the end user has the freedom to change these settings at all times. Since this technique depends on well-established mathematical principles, as will be discussed later, it will be used as a "universal tool" for the frequency analysis of all ECG polarizations, including the P-wave and T-wave domains. Can be used. This device allows the user to move the "start" and "end" electronic cursors to demarcate the polarization to be analyzed. Naturally, the spectral change index is thus distinguished for different analysis areas, by separating normal data from abnormal data. At any instant, the signal-averaged electrical vector E (t) of the heart recorded on the body surface can be represented as the result of the synthesis of three signal-averaged directed electric dipoles. This is electrically equivalent to what is referred to as a “composite lead” in conventional SAECG analysis. The change speed (speed) of the signal-averaged electric vector is as follows. Therefore, the acceleration of the signal-averaged electric vector is as follows. The SAECG system uses a sampling rate of either 1 kHz or 2 kHz. At the 800 ms and 1 kHz sampling rates of the signal averaged ECG file format of this embodiment, the QRS region typically has an upper limit of 600 ms and a lower limit of 300 ms, as discussed above. , For example, a start at 150 ms before the QRS offset and an end at 150 ms after the QRS offset. The frequency spectrum or "acceleration spectrum" of this acceleration vector can then be found by calculating the Fourier transform of the composite lead bounded by the time window of interest. Therefore, in theory, the operation of the discrete Fourier transform can be expressed as follows. In practical terms, acceleration spectra are calculated and analyzed for each orthogonal and composite lead. For improved directional sensitivity, the amplitude and frequency of these high-frequency weak potentials recorded on any of the X, Y, and Z leads depends on the extent of tissue damage in a particular direction. For each lead, perform the following steps before the Fourier transform. 1. Adjust the DC bias of E "(t). 2. Then multiply the resulting zero-mean data segment by a window function, such as a Blackman-Harris window. Helps taper the segment, thereby minimizing "leakage" when performing the Fourier transform. Then, 3. Normalize the resulting zero-mean windowed data. Thus, after DC bias adjustment, multiplication by a tapered window function, and normalization, the acceleration spectrum of the X-lead is calculated as follows: However, T = 512. Note that the remaining points of the Fourier transformed array are zeroed. Thus, if the end user selects 300 points for analysis, the last 212 points are set to zero. The acceleration spectra of the X lead and the Y lead are calculated similarly. Upon calculating the acceleration spectrum, the algorithm automatically checks the bandwidth of interest preset at 50-300 Hz. Thus, the index of the degree of spectral “split” or spectral change index is formulated using discrete mathematics. The numerator is the calculation of the sum of the absolute spectral gradients, and the denominator S _A , _max Is the maximum amplitude of all spectral peaks in the bandwidth of interest. The determination of the spectral change index is embodied as part of the software list described below as Appendix A. The method of implementing the spectral change index is described in more detail in code written in the BASIC programming language as follows. CHANCALC: CHANO = 0 FOR QQ = 25 TO 150 'FROM 50 to 300 Hz YYI = FFTOUT (QQ, FFTN) YY2 = FFTOUT (QQ + 1, FFTN) grad = ABS (YY2-YY1) CHANO = GRAN + CHANOQ ^* CHANO / MAXVEC Thus, the spectral change index is a measure of the change in spectral components within a particular bandwidth that is not deviated by the spectral voltage. For example, a relatively "flat" acceleration spectrum within a particular bandwidth will result in a small value exponent, and an acceleration spectrum having many different frequency spectral components will result in a large value exponent. In addition to the spectral change index, the number of spectral peaks within the bandwidth of interest is also counted (manually or automatically). This adds yet another quantitative value that may be of diagnostic significance for this test. Turning now to FIGS. 3A and 3B, the operation of the program flow chart is performed to obtain a peak count of (1) the spectral change index, and (2) the number of spectral peaks within the bandwidth of interest. Shows some of them. The implementation of the illustrated flow chart is based on the interpretation of a physician to determine whether a patient is at risk through the detection of a weak potential by quantifying the degree of spectral splitting within the bandwidth of interest. Provide indicators that can be done. As best seen in FIG. 3A, the microprocessor 18 begins at a start step 100, and finally from the information provided in step 130 or 132 shown in FIG. 3B, the physician can determine whether the patient is at risk or not. A determination can be made. In step 102, a wave detection and trigger block is used for detecting and triggering the R wave or the P wave. Detection and triggering is prepared using template beat formation to detect and trigger on the R-wave of the ECG waveform or, alternatively, the P-wave. Thus, the region of interest of the electrocardiogram signal of one heart beat is selected by the detection of the weak potential according to the present embodiment. Also, for this purpose, the selection, matching, and signal averaging are performed as described in step 104 to average the noise and increase the signal-to-noise ratio for the collected data. The criterion for stopping signal averaging is determined in step 106 either by a sufficiently low noise level or by averaging a certain number of beats. If the stop criteria in step 106 are not met, the program flow returns to step 102, the wave detection and trigger block. If the stop criteria in step 106 are met, the program flow proceeds to step 108 to select a segment of interest. When the R-wave is triggered, the onset and QRS offset are determined in defining the segment of interest. When the P-wave is triggered, the P-wave onset and offset are seen when obtaining the segment of interest. At this point, the ECG data can be stored at step 110 for later use. Proceeding to step 112, the second derivative of the segment is derived to generate a profile waveform indicative of the temporal attributes of the segment of interest. The temporal attribute, ie the need for the second derivative, derives from the underlying QRS signal which has a lower frequency but a larger magnitude than the weak potential signal, and therefore does not emphasize low frequency components in the QRS complex Requires techniques to emphasize ingredients. Time domain methodologies, such as measurement of elapsed time and RMS values, cannot "focus" or "extract" a weak potential from a rapid up or down slope of the QRS, so the frequency domain approach is preferred. . Nevertheless, the QRS complex itself exhibits rapid traversing speeds of frequency change, and also creates difficulties in the frequency domain, and thus the generation of profile waveforms indicating the temporal attributes of the segment of interest facilitates frequency domain analysis. To Referring to some of the theorems developed for waveforms and their spectra, the Fourier transform relationship of the second derivative effectively increases the magnitude of the spectrum as a function of frequency. Thus, the magnitude of the spectrum increases as a function of frequency. A constant multiplier and a frequency-variable multiplier are provided as "amplification terms" and thus increase with increasing frequency, leading to a frequency domain spectral representation of the profile waveform, and thus the frequency domain representation is such that the high frequency weak potential is It indicates the segment of interest amplified therein. Thus, the spectral energy at higher frequencies is selectively enhanced or boosted. The "derivative" (first derivative) theorem for frequency domain transformation is as follows. Where V ′ (t) is the time derivative of the signal voltage as a function of time t, S (f) is the relevant frequency domain representation as a function of frequency f, and the constants in equation (7) and the following equations are: Includes imaginary number i and constant pi. Equation (7) thus expresses the velocity as the temporal attribute of the segment of interest, as opposed to the "acceleration" for the temporal attribute. Additional derivatives, of course, increase or increase higher frequencies. However, the acceleration spectrum has been found to enhance higher frequency weak potentials appropriately for this detection. Performing the Fourier transform of the second derivative gives: Thus, the "acceleration spectrum" is enhanced at higher frequencies, attenuating lower frequencies and suppressing the zero frequency or DC component. In this way, the lower frequency components of the QRS are suppressed and the higher frequency components of the weak potential are greatly increased. DC bias adjustment provided to obtain the normalized segment data as discussed above after performing the second derivative in step 112 to generate a profile waveform for use in this embodiment. -The DC bias adjustment, e.g. averaging, is prepared in step 114 by the normalization block. At step 116, the acceleration spectrum is converted to a Fourier transform or other frequency transform, such as adaptive filter determination (AFD), maximum entropy (MEM), autoregressive (AR), moving average (MA), or autoregressive moving. Determined using the average (ARMA) method. When using the Fourier transform method, an appropriate window function should be used to window the data set. Once the acceleration spectrum has been calculated, the frequency bandwidth in the spectral representation is defined. Although the preferred embodiment utilizes 50 Hz to 300 Hz, this band can be varied or made variable for user input. In step 118, a peak count is performed to count the number of spectral peaks in the bandwidth of interest, which is determined by detecting the weak potential in the segment of interest of the electrocardiogram signal. Advantageously, it can be used to characterize the degree of change in the bandwidth of interest within the bandwidth of the spectral representation to quantify the degree of spectral splitting within that bandwidth. When identifying multiple spectral peaks from a spectral representation of frequencies within the bandwidth of interest, use criteria for maximum and minimum deviations to determine peaks and valleys, e.g., 65% to 70% deviation And make a judgment. Once the peaks are identified, the peaks are counted and the spectral peak counts are used to detect weak potentials in the region of interest of the electrocardiogram signal to quantify the degree of spectral splitting within the bandwidth of interest. Steps 120, 122 simplify the calculation of the spectral change index in step 124. At step 120, the slope of the spectral curve at the plurality of bandwidths of interest is determined from the spectral representation of each frequency within the bandwidth of interest. In this embodiment, the absolute slope of the spectral curve is summed using each frequency representation from 50 Hz to 300 Hz as indicated above. At step 122, "MAXVEC" is programmatically determined as the maximum amplitude at the spectral peak in the bandwidth of interest to determine a representative maximum amplitude of the spectral representation of the frequency within the bandwidth of interest. The spectral change index (CI) is then used to compare the representative maximum amplitude to the slope or sum of absolute slopes (SAG) from each of the spectral representations of the plurality of frequencies to determine the weak potential in the region of interest of the electrocardiogram signal. Detect and quantify the degree of spectral splitting within the bandwidth of interest. As discussed above, according to the spectral change index, the sum of the absolute gradients is divided by MAXVEC and multiplied by 100 to determine the spectral change index. In the described embodiment, the spectral change index threshold 15 is used to distinguish normal attributes from abnormal attributes, with a lower spectral change index indicating a normal patient condition. At step 126, the spectral change index is compared to a threshold to determine if the patient under examination is at risk, step 130 indicates that the patient is not at risk, and step 132 is at risk. Indicates a state. Similarly, at step 128, the peak count is compared to a threshold to determine if the patient is at risk. The spectral change index threshold can also be fixed or varied under user control, and instead two thresholds are used to determine intermediate spectral change index counts, for example, for borderlines that require further inspection. By preparing 16 to 20 as indices to be indicated, it is also possible to distinguish a patient at the boundary from a patient in an abnormal state. Examples: The device was tested using data previously collected from patients with evidence of affected myocardium, such as patients with persistent ventricular tachycardia. The pilot study discussed here showed the results of two cases, but further testing would be appropriate. Subjects without heart disease generally exhibit low spectral change indices, as seen, for example, in FIGS. 4A and 4B (spectral change indices = 4 in healthy individuals) and have a delayed potential detected by the time domain Simson method. Patients exhibiting myocardial damage, such as patients, exhibit a high spectral change index of 15 or more, as seen in, for example, FIGS. 5A and 5B (for abnormal patients, the spectral change index = 39). The result in the example is that when the patient has evidence of an affected myocardium, such as a patient with persistent ventricular tachycardia, there are many individual spectral peaks within the 50-300 Hz bandwidth of the acceleration spectrum. Indicates that See FIG. 5B. The increase in these high frequency spectral components is due to the splitting of the wave of electrical activation in the damaged myocardium after infarction. This example shows that patients without heart disease generally have a low spectral change index (CI), but patients with myocardial dysfunction, such as patients with delayed potentials found by the time domain Simson method, 15 or more C.I. I. Has been shown. This study analyzed 26 patients (6 with leg block) after myocardial infarction with idiopathic sustained ventricular tachycardia (VT) and 20 control subjects without VT. In the idiopathic VT group, 19 out of 26 patients underwent electrophysiological testing (EPS), all of whom were able to attract persistent VT. C. I. Reproducibility (R) was defined as the subject in a "normal" (or "abnormal") state each time data from the subject was collected and analyzed. R was studied on eight subjects from the entire population. C. in identifying (1) patients at risk of sudden persistent VT and (2) patients with attractable VT during EPS. I. Are summarized in Table I. Table I: Sensitivity, specificity, positive and negative predictive accuracy, and reproducibility of the spectral change index in the initial population of 26 post-MI patients and 20 control subjects. These results show that embodiments of the device and method have robust techniques based on cardiac electrophysiology principles, as well as direct but well established mathematical signal processing techniques. This methodology provides results consistent with the results of the time domain method and provides excellent reproducibility. Investigations on post-MI patients show that this new method has high specificity and positive predictive accuracy, as well as good risk classification for post-MI patients with VT propensity and also for those with leg block It shows that it has reproducibility. Since the negative predictive value of the time domain SA ECG test is very high, but the positive predictive accuracy is low (Simson method), the use of this new method helps to meet the need in the area of signal averaging ECG. Although the preferred embodiment of the present invention has been illustrated and described in terms of an apparatus and method for detecting cardiac potential and detecting cardiac arrhythmia by analysis of all ECG segments and intervals, the present disclosure discloses the present invention and Other embodiments of the invention will be readily apparent to those skilled in the art from consideration of practice. It is intended that the specification and examples be considered as exemplary only, with a true scope and intention of the invention as indicated by the following claims.

Claims (1)

[Claims] 1. A method for detecting a weak potential to identify a cardiac arrhythmia,   Selecting a segment of interest of the electrocardiogram signal for one cardiac cycle;   Generating a profile waveform indicating a temporal attribute of said segment of interest Steps   Deriving a frequency domain spectral representation of said profile waveform. Thus, the frequency domain representation is of interest with a high frequency weak potential amplified therein. Indicating the segment;   Defining a bandwidth of interest for frequencies in the spectral representation; ,   The spectral representation for each of a plurality of frequencies within the bandwidth of interest Determining a gradient from   Take the representative maximum amplitude of the spectral representation of frequencies within the bandwidth of interest. Obtaining,   The representative maximum amplitude and the previous magnitude from each of the spectral representations of a plurality of frequencies. The weak potential in the segment of interest of the ECG signal. Steps to issue,   The degree of spectral splitting within the bandwidth of interest is determined by the representative maximum amplitude Quantifying from the comparison with the gradient to identify a cardiac arrhythmia A method characterized by comprising: 2. The selecting step includes reducing noise levels in the representative waveform. 2. The method of claim 1, comprising signal averaging of a plurality of cardiac cycles. 3. The generating step includes the steps of: Taking the second derivative of said segment of interest of the electrogram signal, further comprising: The method of claim 2, wherein: 4. When performing the frequency domain transform using the Fourier transform, That the step further comprises a step of windowing the profile waveform. The method according to claim 3, characterized in that: 5. The step of defining may increase the bandwidth of interest from about 50 Hz to about 300 Hz. 5. The method of claim 4, wherein the method is defined as: 6. The determining step is performed before each of a plurality of frequencies in the bandwidth of interest. 5. The method of claim 4 including determining a gradient from the spectral representation. The method described. 7. The determining step being performed prior to displaying the spectrum within the bandwidth of interest. The method of claim 4, comprising summing the absolute value of the slope. 8. The step of obtaining includes determining the spectrum for frequencies within the bandwidth of interest. Identifying a plurality of spectral peaks from the vector representation; and Recognizing the representative maximum amplitude of the display as an identified maximum spectral peak The method of claim 4, comprising: 9. The comparing step sums the absolute values of the slopes of the spectral representation Dividing the sum determined in the determining step by the representative maximum amplitude Calculating an index quantifying the degree of spectral splitting within said bandwidth of interest. The method of claim 7, comprising: 10. A method for identifying a cardiac arrhythmia by detecting a weak potential,   Selecting a segment of interest of the electrocardiogram signal for one cardiac cycle;   Generating a profile waveform indicating a temporal attribute of said segment of interest Steps   Deriving a frequency domain spectral representation of said profile waveform. Thus, the frequency domain representation is of interest with a high frequency weak potential amplified therein. Indicating the segment;   Defining a bandwidth of interest for frequencies in the spectral representation; ,   Characterizing the degree of change in the bandwidth of interest in the spectral representation within the bandwidth In addition, a step for detecting a weak potential in a segment of interest of the electrocardiogram signal. Step With   The characterization step includes the degree of spectral splitting within the bandwidth of interest. Identifying cardiac arrhythmias by quantifying Method. 11. The generating step may include, as the profile waveform indicating a temporal attribute, Taking the second derivative of the segment of interest of the electrocardiogram signal, further comprising: The method of claim 10, wherein the method comprises: 12. The characterization step includes counting the spectral peaks, The weak potential in the frequency range of about 50 Hz to about 300 Hz in the diagram signal is detected. And quantifying the degree of spectrum splitting therein. The described method. 13. The characterization step comprises:   Multiple spectra from the spectral representation for frequencies within the bandwidth of interest Identifying a turtle peak;   Count the spectral peaks to find the segment of interest in the ECG signal The weak potential of the signal to determine the degree of spectral splitting within the bandwidth of interest. Weighing step The method of claim 10, comprising: 14． The step of defining may include increasing the bandwidth of interest from about 50 Hz to about 300 H 14. The method of claim 13, wherein z is defined as z. 15. An apparatus for detecting a weak potential to identify a cardiac arrhythmia,   Computer and   A signal storage device that can be handled by the computer;   To select the segment of interest of the ECG signal for one cardiac cycle, Storing the segments in the signal storage device and the segments of interest The computer to generate a profile waveform that indicates the temporal attributes of the A signal collection module that can be operated by   Characterize the degree of change in frequency of interest in the spectral representation within the bandwidth of interest To detect the weak potential in said segment of interest of the electrocardiogram signal Before the interest by the high-frequency weak potential amplified within the bandwidth of interest To derive a spectral representation of the profile waveform indicating the segment, Frequency domain conversion program used by the computer With   The frequency domain transforming program is used to identify cardiac arrhythmias prior to interest. Detects weak potentials, characterized by quantifying the degree of spectral splitting within the bandwidth Equipment for doing. 16. The frequency domain conversion program used by the computer,   The spectral representation for each of a plurality of frequencies within the bandwidth of interest Means for determining the gradient from   Take the representative maximum amplitude of the spectral representation of frequencies within the bandwidth of interest. Means for obtaining;   The representative maximum amplitude and the previous magnitude from each of the spectral representations of a plurality of frequencies. The weak potential in the segment of interest of the ECG signal. Means for determining the degree of spectral splitting within said bandwidth of interest The apparatus of claim 15, comprising: 17． The signal collection module, which can be operated by the computer, As the profile waveform indicating the attribute, the segment of interest of the electrocardiogram signal 16. The apparatus according to claim 15, including means for taking the second derivative of G. 18. The comparing means determines the absolute value of the slope of the spectral representation, Sum the absolute value of the slope within the bandwidth with The sum of the absolute values of the slopes is divided by the representative maximum amplitude to obtain the band of interest Including means for calculating an index that quantifies the degree of spectral splitting within the width. The apparatus according to claim 16, characterized in that: 19. The frequency domain conversion program used by the computer, Characterize the degree of change in frequency of interest in the bandwidth in the spectral representation To detect the weak potential in the segment of interest of the ECG signal, Including means for quantifying the degree of spectral splitting within said bandwidth. 16. The device according to claim 15, wherein the device comprises: 20. Said device for characterizing comprises:   Multiple spectra from the spectral representation for frequencies within the bandwidth of interest Means for identifying a turtle peak;   Count the spectral peaks to find the segment of interest in the ECG signal The weak potential of the signal to determine the degree of spectral splitting within the bandwidth of interest. Means of weighing 20. The apparatus of claim 19, comprising: 21. A method for detecting a weak potential to identify a cardiac arrhythmia,   Selecting a segment of interest of the electrocardiogram signal for one cardiac cycle;   Generating a profile waveform indicating a temporal attribute of the segment of interest Steps and   Deriving a frequency domain spectral representation of said profile waveform. Thus, the frequency domain representation is of interest with a high frequency weak potential amplified therein. Indicating the segment;   Defining a bandwidth of interest for frequencies in the spectral representation; ,   The spectral representation for each of a plurality of frequencies within the bandwidth of interest Determining a gradient from   Obtaining a representative spectral change index of frequencies within the bandwidth of interest. And   The representative spectral change index and the spectral table for each of a plurality of frequencies. Comparison with the slope from the reading to find the finer segment in the segment of interest of the ECG signal. Detecting a weak potential;   The degree of spectral splitting within the bandwidth of interest is determined by the representative spectrum Quantifying from the comparison of the change index to the slope to identify a cardiac arrhythmia A method characterized by comprising: 22. The above selecting step reduces the noise level in the representative waveform 22. The method of claim 21 including signal averaging of a plurality of cardiac cycles. Law. 23. The generating step may include, as the profile waveform indicating a temporal attribute, Taking the second derivative of said segment of interest of the electrocardiogram signal. 23. The method of claim 22, wherein the method comprises: 24. When performing the frequency domain transform using the Fourier transform, Wherein the step includes a step of windowing the profile waveform. 24. The method of claim 23, wherein: 25. The step of defining may include the step of controlling the bandwidth of interest from about 500Oz to about 300H. 25. The method of claim 24, wherein z is defined. 26. The determining step includes determining each of a plurality of frequencies in the bandwidth of interest. The method of claim 24, further comprising determining a gradient from the spectral representation. The method described in. 27. The determining step comprises the step of determining the spectral representation within the bandwidth of interest. The method of claim 24, comprising summing the absolute value of the gradient. Law.