Device and computer program product for detecting atrial fibrillation in an electrocardiogram

Abstract

Device for detecting atrial fibrillation in an electrocardiogram, comprising a recording unit (33) configured to record the electrocardiogram (1, 3, 4) of a predetermined length, a computing unit (34) configured to perform R-wave detection in the electrocardiogram and to calculate an RR interval (2) as the time interval between two adjacent R-waves of the electrocardiogram (1, 3, 4), to calculate an average of the RR intervals (2), to assign a reference symbol to each RR interval (2), whereby different reference symbols are used for different deviations of the RR interval (2) under consideration from the calculated mean value, to form a character string from the assigned reference characters, whereby each of the assigned reference characters is included only once in the character string, and to compare the character string thus formed with several known character strings which are typical for various, previously determined and stored in a memory unit (37) of the computing unit (34) courses of electrocardiograms (1, 3, 4) in atrial fibrillation, and an output unit (35) which is arranged to output a signal upon a detected match of the character sequence formed with the recorded electrocardiogram (1, 3, 4) with one of the known character sequences typical of atrial fibrillation, wherein each of the reference symbols is assigned to a class of RR intervals (2), and wherein each of the classes of RR intervals (2) indicates a deviation from the calculated mean value.

Description

[0001] The present invention relates to a device and a computer program product for detecting atrial fibrillation in an electrocardiogram.

[0002] Atrial fibrillation, also known as absolute arrhythmia or atrial fibrillation (AF), is the most common cardiac arrhythmia. Its incidence increases with age and is between 5% and 10% in people over sixty years of age. The causes of atrial fibrillation are not yet fully understood. Possible causes include atherosclerosis of the cardiovascular system, cardiac inflammation, electrolyte disturbances, and dilatation of the atrium—for example, in heart defects. From a systems theory perspective, signal generation in atrial fibrillation is inconsistent, and several different mechanisms must be captured algorithmically.

[0003] Diagnosis is usually made with an electrocardiogram (ECG), but even then, a high percentage of attacks of atrial fibrillation (paroxysmal atrial fibrillation) go unnoticed. An electrocardiogram reflects the condition and function of the heart and, as an established technique, represents an important tool for medical diagnostics. The electrocardiogram provides a direct link to the condition and function of the heart, is non-invasive, and provides a quantitative assessment. An electrocardiogram amplitude corresponds to a measured voltage value.

[0004] A large proportion of episodes, i.e., segments of the electrocardiogram with paroxysmal atrial fibrillation, are asymptomatic. However, irregular activity of the ventricles results in reduced pumping performance and an increased risk of thromboembolism, such as stroke.

[0005] To protect against such serious consequences and to initiate effective treatment, the detection of atrial fibrillation is therefore of paramount importance. The aim is to use long-term ECG monitoring to detect even paroxysmal atrial fibrillation, which eludes conventional diagnostics. However, this real-time detection requires sophisticated algorithms that cannot be implemented with the limited hardware of body-worn monitors or that rely on statistics of long signal segments, making the medically desired timely alarm impossible.

[0006] To date, algorithms have been published in the scientific literature to address this problem; these algorithms can be used for different objectives in different patient populations. However, the aim should be to develop an algorithm that is tailored to the patient population and the associated monitoring goal. For example, in an initial group of patients being monitored electrocardiographically after cardiac surgery or cardioversion, it is important to detect atrial fibrillation quickly and with high sensitivity and to alert a physician if necessary. In patients with known atrial fibrillation (usually with a chronic course), a physician is primarily interested in a heart rate trend, percentage burden (AF burden), and a temporal relationship between fibrillation phases and clinical symptoms. The monitoring strategy involves the timely detection of even short episodes of atrial fibrillation.In patients in whom atrial fibrillation attacks are suspected or where a rhythm change is to be detected, the decision regarding therapeutic intervention depends on the results of long-term ECG monitoring. Consequently, longer, treatment-relevant periods of atrial fibrillation must be detected with a high degree of certainty.

[0007] Typical distribution patterns in the electrocardiogram that would reliably characterize atrial fibrillation or distinguish it from other arrhythmias can usually only be detected after a sufficiently long electrocardiogram recording time and a correspondingly large amount of data, which is disadvantageous for practical application. Furthermore, attacks of atrial fibrillation vary greatly in duration (from 10 seconds to several hours), so sufficient data is not always available for statistical conclusions.

[0008] The present invention is therefore based on the object of proposing a device and a computer program product by means of which atrial fibrillation is reliably detected and the presence of atrial fibrillation is only signaled if it is present with a high degree of certainty.

[0009] This object is achieved according to the invention by a device according to claim 1 and a computer program product according to claim 8. Advantageous embodiments and further developments are described in the dependent claims.

[0010] A method performed by an atrial fibrillation detection device for detecting atrial fibrillation in an electrocardiogram comprises several steps. For this purpose, an electrocardiogram of a predetermined length is recorded, i.e., cardiac activity is recorded over a predetermined time period. R-wave detection is performed in the recorded electrocardiogram, and an RR interval is calculated. Typically, several RR intervals are calculated. The RR interval is defined as the time interval between any two adjacent R waves in the electrocardiogram. Subsequently, an average, typically an arithmetic mean, of the RR intervals is calculated.

[0011] Each of the RR intervals is assigned a reference symbol, with different reference symbols being used for different deviations of the respective RR interval from the calculated mean. Thus, the temporal length of the RR interval currently being considered is compared with the mean. A character string is formed from the reference symbols thus assigned. Each of the assigned reference symbols is included only once in the character string.

[0012] The resulting sequence of characters is then compared with several known sequences of characters. These known sequences are typical of various previously determined and stored electrocardiogram patterns in atrial fibrillation. If a match is found between the sequence of characters generated from the recorded electrocardiogram and the known sequences of characters typical for atrial fibrillation, a signal is output.

[0013] Since the duration of the RR interval normally does not deviate by more than + / - 15% from the duration of the immediately preceding RR interval (sinus rhythm), an irregular sequence of R waves or a high variability of the RR intervals can be inferred from histograms to indicate irregular ventricular activity. The simple approximation method described above allows distribution patterns of the RR intervals to be recorded quickly and efficiently. Atrial fibrillation can be reliably detected by comparing them with known distribution patterns, which do not occur in electrocardiograms of healthy individuals or in known arrhythmias, but are possible due to the stochastic distribution of the RR intervals in atrial fibrillation.Not all of the acquired data is necessary for this; it is sufficient to detect a specific combination of deviations from the mean value determined by the reference symbols, which in turn significantly reduces the amount of data to be processed. The described method can also be executed on body-mounted long-term ECG devices with low computing power, as an algorithm used is resource-efficient, extends the operating time of such a device without recharging, and simultaneously enables real-time monitoring. Due to its sufficiently high relevance, atrial fibrillation is only signaled when it is actually present with a high degree of certainty.

[0014] Using the easy-to-implement approximation method in a sliding window of only a few heartbeats, typical distribution patterns of the RR intervals can be recorded and compared with learned distribution patterns that do not occur in electrocardiograms of healthy individuals or in known arrhythmias, but are possible due to the stochastic distribution of the RR intervals in atrial fibrillation.

[0015] It can be provided that the reference characters of the generated character string are sorted according to a predetermined order before being compared with the known character strings, and that the reference characters of the known character strings are also sorted according to this predetermined order. This facilitates comparison because a sequence of characters is predetermined, allowing matches to be found more quickly. The sequence can, for example, be alphabetically ascending or descending, or include digits sorted in ascending or descending order.

[0016] Typically, the assigned reference symbols are implemented using binary numbers, which are arranged, in particular, as bit masks. For a character string encoding that is easily perceptible to humans, alphanumeric characters, preferably letters of the Latin alphabet and / or Indo-Arabic numerals, are typically used. The resulting character string and the known character strings can thus be alphanumeric character strings.

[0017] Each of the reference symbols is assigned to one of the classes of RR intervals, and each of the classes of RR intervals thus indicates a deviation from the calculated mean, preferably in the form of a percentage deviation between the calculated mean and the RR interval. This allows for a clear assignment of different deviations, which are grouped into classes for better clarity. A difference should also be understood as a percentage value, which indicates a percentage deviation when a difference is calculated.

[0018] It can be specified that the character string is only transmitted when a determined comparison value exceeds a predefined threshold with respect to the mean. This increases the reliability of detecting atrial fibrillation, since small fluctuations can always occur. Typically, the comparison value is defined as the sum of absolute differences between the RR intervals and the calculated mean.

[0019] The length of an electrocardiogram evaluation window, which comprises the electrocardiogram data to be analyzed step by step, is selected such that a time window with a specified number of RR intervals is considered. A minimum number of RR intervals considered ensures the reliability of the method. Typically, at least or exactly nine RR intervals are considered. The generated character string has a maximum length that corresponds to the specified number of analyzed RR intervals. This avoids overdetermination and allows the method to be performed more quickly. Typically, however, the length of the character string is smaller than the number of RR intervals analyzed in the window.

[0020] Alternatively or additionally, it can be provided that if a match is found between the formed character string and one of the known character strings, a check is carried out to determine whether at least one further match occurred within a predetermined period of time prior to the detected match, and the signal is only output if the minimum number of detected matches was detected within the predetermined period. Preferably, a check is carried out to determine whether at least five matches occurred, and the signal is only output if all five matches are detected. A predetermined minimum number of matches ensures that individual irregular signals are not detected as atrial fibrillation, and the reliability of the method is increased. A multi-stage verification process can thus increase the reliability of detecting atrial fibrillation.Preferably, only episodes of atrial fibrillation are taken into account and a signal is issued accordingly if the episodes last two minutes or longer, since shorter episodes do not have any therapeutic consequences.

[0021] Real-time processing is preferably provided, which means that all events are processed until the next R-wave occurs, i.e., the method runs once during the duration of an average RR interval. Typically, the described method therefore runs automatically within 800 ms, preferably within 500 ms, and particularly preferably within 300 ms.

[0022] Typically, R-wave detection and the calculation of the mean value, as well as the assignment of the reference symbols and comparison with the known character sequence, occur in parallel with the recording of the electrocardiogram in order to ensure rapid data processing and to be able to continuously monitor patients for atrial fibrillation. Data is recorded continuously and simultaneously subjected to the described analysis. Alternatively, sequential processing can also be provided, in which the electrocardiogram is recorded first and the data obtained is then analyzed using the described method. The recorded electrocardiogram is therefore evaluated continuously with each subsequent recording. An analysis window or evaluation window with electrocardiogram data to be analyzed slides forward for each newly detected beat.

[0023] The signal is typically only output if the electrocardiogram is at least two minutes long and atrial fibrillation is detected during this time. This also increases the reliability of the method, since the presence of atrial fibrillation can only be determined after a sufficiently long observation time.

[0024] R-wave detection is typically performed using a wavelet transform of the electrocardiogram, preferably using modulus maximum pairs. R-wave detection is easy and reliable, even with non-standardized ECG acquisition and varying electrode placement, and is immune to interference from muscle potentials or other signal disturbances.

[0025] Typically, the described procedure is integrated into a higher-level procedure for analyzing electrocardiograms.

[0026] A device for carrying out the described method comprises a recording unit configured to record the electrocardiogram, a computing unit configured to perform R-wave detection, determine the RR intervals, determine the character string, and compare the character string with predetermined character strings, and an output unit configured to output the determined atrial fibrillation. The computing unit also comprises a storage unit in which the known character strings are stored.

[0027] The device is typically a device for performing a long-term ECG or a cardiac pacemaker.

[0028] A computer program product contains a sequence of instructions stored on a machine-readable medium for carrying out the described method and / or for controlling the described device when the computer program product runs on a computing unit.

[0029] Embodiments of the invention are illustrated in the drawings and are described below with reference to Fig. 1 to 6 explained.

[0030] They show: Fig. 1 a schematic electrocardiogram; Fig. 2 a real electrocardiogram in atrial fibrillation; Fig. 3 a table with classes and reference symbols indicating a difference between a length of an RR interval and a calculated mean of all RR intervals; Fig. 4 a schematic representation of a generation of a character string from reference characters; Fig. 5 a flowchart of a method for detecting atrial fibrillation and Fig. 6 is a schematic representation of a device attached to a body for performing the Fig. 5 described procedure.

[0031] In Fig. Figure 1 shows an idealized electrocardiogram 1, a time series of electrical signals from cardiac muscle fibers. Voltages of varying intensity are recorded along a time axis, with peaks usually designated by the letters P, Q, R, S, and T for easier differentiation. Such a complex of peaks repeats periodically in electrocardiogram 1.

[0032] QRS complexes, also called ventricular oscillations, whose dominant amplitude excursion is referred to as the R wave, follow one another, with intervals between two consecutive ventricular oscillations being referred to as RR intervals 2. If the heart beats with a sinus rhythm, the duration of each of the RR intervals 2 usually deviates by no more than 15 percent from the duration of an immediately preceding RR interval 2.

[0033] During physiological cardiac activity, the atria of the heart are rhythmically excited by the sinus node. They transmit the excitation via the atrioventricular node (AV node) to the ventricles, so that an orderly excitation front runs through all cardiac segments. In electrocardiogram 1, the P wave precedes the QRS complex, reflecting atrial excitation. It is less prominent and more difficult to detect algorithmically than the QRS complex. This is especially true for an ambulatory long-term electrocardiogram, which is typically distorted by muscle potentials and motion artifacts. Even after subtracting the QRST components from electrocardiogram 1 and targeted atrial pacing, P wave detection remains problematic.

[0034] As in Fig. 2 using a real electrocardiogram 3, the baseline in atrial fibrillation is irregular, but no P waves are visible. Recurring features are marked with identical reference symbols in the figures. In atrial fibrillation, abnormal activation centers and circulating excitations cause uncoordinated excitation of the atrial muscles with rapidly successive stimulation of the AV node. Due to its refractory behavior, the AV node can only transmit individual excitations to the ventricles; others are blocked. This results in stochastic ventricular excitation, the so-called absolute arrhythmia. In the electrocardiogram 3, the main ventricular fluctuations (R waves) then occur arrhythmically and without relation to the atrial activities due to the fluctuating AV conduction.

[0035] Atrial fibrillation can be triggered by various mechanisms, which is why typical distribution patterns are rare for all forms of atrial fibrillation. However, the method described in more detail below can reliably detect atrial fibrillation, even with devices with limited computing power.

[0036] For an exemplary implementation, a sampling rate of 250 sps (samples per second) is assumed for electrocardiogram 1, 3. R-wave detection in electrocardiogram 1, 3 is performed using "modulus maximum pairs" of a wavelet transform of electrocardiogram 1, 3. An R-wave detection trigger is therefore suitable for measuring the instantaneous heart rate. The method, described in more detail below as an example, can be used to determine in real time whether and at what point in time a long-term ECG exhibits longer, potentially therapy-relevant episodes of atrial fibrillation, thus confirming the diagnosis of "paroxysmal absolute arrhythmia." This makes it suitable for tasks in which ECG analysis, including monitoring of atrial fibrillation, must be performed under time-critical conditions using simple hardware while conserving space, mass, and energy.The method relates to a computational analysis of electrocardiograms 1, 3, particularly under time-critical conditions, i.e. real-time or time-lapse signal processing.

[0037] Such fields of application arise particularly in implantable pacemakers, Holter devices, or devices worn in clothing for monitoring cardiac activity. Application is also possible in computing technology integrated into medical patient monitoring devices, functional diagnostic devices, or everyday objects with the detection function for physiological signals.

[0038] A shift register is filled with nine consecutive RR intervals 2. Each RR interval 2 describes a time period whose beginning is determined by one of the R waves and whose end is determined by an R wave following the originally detected R wave. The shift register is updated with each new R wave detection. After the update, the rounded arithmetic mean of the nine RR intervals 2 within the observation interval is calculated. Each RR interval 2 in the window is then assigned one of thirteen possible symbols, depending on its deviation from the current mean. Fig. 3 shows an assignment table in which a difference between the length of the respective RR interval 2 under consideration and the rounded mean is specified as a percentage based on the rounded mean. In further embodiments, an exact rather than a rounded mean can of course also be used. Depending on the sign, each percentage value is assigned a letter of the Latin alphabet as a reference symbol for the respective RR interval 2, whereby the order of the assigned letters is predetermined by their order in the alphabet. The largest deviation with a negative sign is therefore assigned the letter "A" and the further letters are assigned in classes of five percent until the class corresponding to a deviation of + / - 2.5 percent is assigned the letter G.As the deviation increases and the sign is positive, the letters are assigned until the letter "M" stands for the largest deviation with a positive sign. The class boundaries used are given only as examples and are based on the usual medical limits for ECG diagnostics, such as sinus rhythm, sinus arrhythmia, extrasystole, and escape systole.

[0039] In Fig. 4 shows schematically how the allocation of the Fig. 3 shown letters to the RR intervals 2. In the top line of Fig. Figure 4 shows an electrocardiogram 4 with nine RR intervals 2 of different lengths. The second line shows the calculated percentage deviation of each RR interval 2 from the mean value calculated from the nine RR intervals 2 in numerical form. The third line shows the percentage deviation according to the table in Fig. 3 is represented as a reference symbol of a character sequence to be formed. As in the fourth line of the Fig. As shown in Figure 4, each of the nine letters is registered only once. For example, the letter "C" appears twice in the third row, while it is recorded only once in the fourth row, which represents a bin activation. The letters remaining after the bin activation are sorted according to their occurrence in the Latin alphabet, i.e., in alphabetical order, and form a Fig. 4 shows the distribution word as a character sequence of reference symbols. Instead of letters of the Latin alphabet, Indian-Arabic numerals or any other characters can be used as reference symbols in further embodiments, alternatively or additionally. Likewise, the resulting reference symbols do not need to be sorted; they can also be passed on to subsequent processing unsorted by permuting the sequence, provided the appropriate computing power is available. For programming purposes, implementation as a bit pattern, i.e., with binary numbers, is preferable.

[0040] With the determined characters, a string is created for each detected beat in the electrocardiogram 4, namely the distribution word. Due to the selected window length, a maximum of nine letters from the set A to M can be used. Fig. 3, however, the length of the window, i.e., the number of heartbeats to be recorded and analyzed, can also be chosen arbitrarily. Since with each beat-by-beat update of the distribution word, occupied classes with one or more representations are displayed only once with the corresponding letter, the distribution word is usually shorter. The distribution word is thus a one-dimensional projection of a histogram of the nine RR intervals 2 of the observation window.

[0041] Subsequently, each newly detected QRS complex is compared with so-called AF words, which are known character strings. These AF words were determined in an empirical learning process using various atrial fibrillation databases and preferably consist of five letters. In further embodiments, the AF word can also have fewer than five letters, for example, only three letters, or more than five letters, for example, seven letters. However, the AF word is no longer than the distribution word. The AF words were determined from the RR intervals 2 of the electrocardiograms stored in the databases, analogous to the described procedure for forming the distribution word.As an example, thirty AF words typical for different types of atrial fibrillation are selected, representing RR interval distribution patterns that do not occur or are very rare in the electrocardiogram of healthy individuals and in known arrhythmias and are therefore typical for atrial fibrillation.

[0042] The subsequent comparison of the current distribution word of a beat with the five letters of one of the thirty AF words used in the illustrated embodiment has a filtering effect that eliminates the usual preprocessing steps of other methods used to detect atrial fibrillation and circumvents their disadvantages. The described method automatically adapts to different heart rates and, since only five letters are tested for this embodiment, is insensitive to outliers and interference. Likewise, false detection of recurring arrhythmia forms such as bigeminy or trigeminy is prevented. While such recurring arrhythmia forms also cause a high variance in the RR intervals 2, they result in typical forms of the distribution word that expressly do not occur in the selected AF words.

[0043] If the distribution word, as a formed character string, contains all five letters of one of the thirty AF words as a known, predetermined character string and at the same time the sum of the deviations from the current mean value of this cycle lies above a threshold value, this cycle (hereinafter referred to as the beat) is classified as a so-called AF beat. In further embodiments, it is possible for a signal to be output even when an AF beat is present. The presence of an episode of atrial fibrillation is inferred from the density of AF beats in the course of the electrocardiogram. In the illustrated embodiment, an entry threshold for the AF word counter in the forty-beat window is set to five, while a corresponding exit threshold is zero. In the paroxysmal absolute arrhythmia detectable by the method, attacks with irregular ventricular activity occur, and consequently an irregular sequence of R waves orhigh variability of the RR intervals 2, which can be captured using histograms.

[0044] In Fig. Figure 5 shows a flowchart of the method, which in the illustrated embodiment is integrated into a higher-level method for analyzing electrocardiograms 1, 3, 4 and is executed automatically within 300 ms, i.e., in real time. The electrocardiogram 1, 3, 4 has therefore already been recorded over a duration of at least two minutes or is being continuously recorded, and R-wave detection is also performed continuously. The method begins at point 11, when a QRS detector transmits the time 12 of a detected R-wave of the electrocardiogram 1, 3, 4, thus updating the window of the last nine RR intervals 2. In step 13, the temporal average is calculated over these nine RR intervals 2.Furthermore, the difference from this mean value is calculated for each of the nine RR intervals, and in a further step 15, a comparison value (sum value) is formed, which represents the sum of absolute deviations from the current mean value. In further embodiments, the method can also be carried out without determining a comparison value.

[0045] In step 16, it is checked whether the determined comparison value lies above an empirically determined, predetermined threshold. If this is the case, in step 17 the distribution of the nine RR intervals 2 is used to generate the distribution word as a string of assigned reference symbols, as in connection with Fig. 3 already explained. If the comparison value is below the empirically determined, predefined threshold, the process continues with step 24. As in the Fig. 3 and Fig. As already shown in Figure 4, in the illustrated example, thirteen intervals with a fixed percentage deviation from the current mean are always formed and labeled with a letter. The ordered sequence of the letters of single- or multiply-occupied classes results in the distribution word of the current window.

[0046] In step 18, the comparison with the AF words is performed as a direct word comparison. If a match is found, an AF word counter is incremented 21 for the window under consideration. By testing in step 24 whether an AF word leaves the sliding window and, if so, a subsequent decrement 27 of the AF word counter, the counter for the current window is updated, and the method continues with step 25, in which a decision is made regarding the status.

[0047] To achieve hysteresis behavior, the threshold for the density of AF pulses described above is split into an entry threshold and an exit threshold. The exit threshold is examined in step 22: If an AF episode is active and the counter has fallen below the exit threshold, the episode is set as inactive in step 20, and the time of termination 19 is reported as event 14.

[0048] If the exit threshold is not exceeded, the process proceeds to step 23 with the status "episode active." If no AF episode was active in step 25, the episode is set active in step 29 if the entry threshold 28 is exceeded, and the time of onset 26 is reported as event 14. If the entry threshold is not exceeded in step 28, the process proceeds to step 30 with the status "episode inactive." The process is processed for incoming R-wave detections and updates the status accordingly.

[0049] The method can be stored as a computer program on a machine-readable medium, for example, a compact disc (CD), digital versatile disc (DVD), a universal serial bus (USB) flash drive, or an internal memory of a device, for example, a hard disk, and run on the computer or machine. Likewise, the computer program can control a device, shown in more detail below, for carrying out the method.

[0050] The procedure can be carried out on a computer or microcontroller and serves for a step-by-step processing of the externally or intracardially derived, digitized electrocardiogram 1, 3, 4. This involves a continuous pattern search in a physiologically defined interval.

[0051] In Fig.Figure 6 shows a schematic view of a Holter ECG device 31, which is worn on the body by a user 32 and supplied with electrical energy by a battery, but in further embodiments can also be implanted. The Holter ECG device 31 enables ECG analysis under time-critical conditions, which is carried out using simple hardware in a space-, mass-, and energy-saving manner. The illustrated Holter ECG device 31 is integrated into the clothing of the user 32 and can operate autonomously for extended periods (more than 24 hours during normal operation) without an external power supply. The reliability of a biosignal analysis determines whether the signal is permanently stored, as well as the frequency and length of necessary radio transmissions to a base station. A recording unit 33, a computing unit 34, and an output unit 35 are arranged in the Holter ECG device 31 and are electrically connected to one another.

[0052] The recording unit 33 has several electrodes attached to the body of the user 32 and is used to record the electrocardiogram 1, 3, 4. The recorded data is evaluated by the computing unit 34, in this case a microcontroller on which the computer program product also runs, using the described method and output on a display. The display can also be configured to show the entire recorded electrocardiogram 1, 3, 4. In further embodiments, a data transmission device for wireless or wired transmission of the data can also be provided on the long-term ECG device 31. This data transmission device can also be a connection for a memory card. The computing unit 34 also has a storage unit 37 in which the AF words are stored as known or predetermined character strings.In addition, a ring buffer 36 is provided in which various data can be temporarily stored, for example wavelet coefficients used to determine R-waves.

[0053] In further embodiments, the output unit 35 may also have only individual diodes instead of the display or in addition to the display.

[0054] Only features of the various embodiments disclosed in the exemplary embodiments can be combined with one another and claimed individually.

Claims

[1] Device for detecting atrial fibrillation in an electrocardiogram, comprising a recording unit (33) configured to record the electrocardiogram (1, 3, 4) of a predetermined length, a computing unit (34) configured to perform R-wave detection in the electrocardiogram and to calculate an RR interval (2) as the time interval between two adjacent R-waves of the electrocardiogram (1, 3, 4), to calculate an average of the RR intervals (2), to assign a reference symbol to each RR interval (2), whereby different reference symbols are used for different deviations of the RR interval (2) under consideration from the calculated mean value, to form a character string from the assigned reference characters, whereby each of the assigned reference characters is included only once in the character string, and to compare the character string thus formed with several known character strings which are typical for various, previously determined and stored in a memory unit (37) of the computing unit (34) courses of electrocardiograms (1, 3, 4) in atrial fibrillation, and an output unit (35) which is arranged to output a signal upon a detected match of the character sequence formed with the recorded electrocardiogram (1, 3, 4) with one of the known character sequences typical of atrial fibrillation, wherein each of the reference symbols is assigned to a class of RR intervals (2), and wherein each of the classes of RR intervals (2) indicates a deviation from the calculated mean value. [2] Device according to claim 1, characterized bythat the computing unit is designed to sort the reference symbols of the formed character string according to a predetermined order before comparing them with the known character strings, wherein reference symbols of the known character strings are also sorted according to this predetermined order. [3] Device according to claim 1 or claim 2, characterized by that the associated reference characters are alphanumeric characters, preferably letters of the Latin alphabet and / or Indian-Arabic numerals, and the formed character string and the known character strings are each an alphanumeric character string. [4] Device according to one of the preceding claims, characterized by that the computing unit (34) is arranged to determine the character string only when a predetermined threshold value of a sum of absolute differences from the mean value is exceeded. [5] Device according to one of the preceding claims, characterized bythat the length of the electrocardiogram (1, 3, 4) is selected such that a time window with a predetermined number of RR intervals (2) is considered, preferably nine RR intervals (2), wherein the formed character string has a maximum length corresponding to the predetermined number of analyzed RR intervals (2). [6] Device according to one of the preceding claims, characterized by that the computing unit (34) is designed, when a match is found between the formed character string and one of the known character strings, to check whether at least one further match, preferably at least five further matches, has / have occurred within a predetermined period of time before the match was found, and the signal is only output if the at least one match found was detected within the predetermined period of time. [7] Device according to one of the preceding claims, characterized by in that the computing unit (34) is designed to carry out the R-wave detection, the calculation of the mean value, the assignment of the reference symbols, the formation of the character sequence and the comparison of the formed character sequence in parallel with the recording of the electrocardiogram, and the output unit (35) is designed to carry out the output of the signal in parallel with the recording of the electrocardiogram. [8] Computer program product containing a sequence of instructions stored on a machine-readable carrier, which is designed to control the device according to one of claims 1-7 when the computer program product runs on a computing unit (33).

Images