EP3478169A1 - Method for detecting at least one cardiac rhythm disturbance - Google Patents

Abstract

The invention concerns a method for detecting at least one cardiac rhythm disturbance of a human or animal subject from a time series of intervals RR representative of the cardiac rhythm of the subject, extracted from a physiological signal (S). The method comprises a step (S07) of characterising the time series with at least one descriptive variable (Vb), the value of which is calculated from a derivative series (Sj₂, Sj₃, Sj₄, Sj₅, Sj₆, Sj₇) of which the constituent element is a discrete derivative of the order of 1 or greater than 1 of the interval RR of the time series.

Description

 Method for detecting at least one heart rhythm disorder

The present invention relates to a method and a device for detecting a cardiac rhythm disorder of a human or animal subject from a physiological signal containing information on the subject's cardiac rhythm.

 The present invention relates more particularly but not exclusively to the realization of a portable device for ensuring a continuous monitoring of the cardiac activity of a subject.

 To date, there is a real need for a device that makes it possible to record in real time the vital parameters of a person in order to prevent any potential health risks, in particular cardiac risks. In particular, there is a real need for screening for atrial fibrillation. Atrial fibrillation is the most common cardiac arrhythmia, responsible for a large number of hospital admissions for rhythmic problems. It is estimated that one in six people over the age of 40 will experience this arrhythmia, which currently affects several million subjects. Its prevalence in the population increases with age, reaching 1.5% over 50 years and more than 20% over 80 years. Atrial fibrillation is also responsible for 10 to 15% of all stroke and 25% of stroke after 80 years. In almost a third of cases, atrial fibrillation is asymptomatic, behaving like a "silent killer". Due to the aging of the population, screening for atrial fibrillation is a major public health issue.

 There is no commercially available device on the market that provides cardiac monitoring functions in a manner that can be considered medically reliable. Various portable devices, such as watches equipped with means for acquiring a photoplethysmographic signal, make it possible to measure the heart rate but are not able to diagnose in real time the occurrence of a rhythm disorder. Their degree of medical relevance is therefore very low or non-existent. In addition, specialized equipment such as the "Home Health Hub" marketed by the company Freescale or the "Body Guardian ® Heart" developed by the companies Preventice and STMicroelectronics, are addressed to patients with cardiac arrhythmias. Biometric data is transmitted to a physician outside the hospital environment, allowing patients to perform normal daily activities while receiving continuous monitoring.

On subjects at risk, a statement of the electrocardiogram or ECG is generally practiced using a recorder event called "Holter", or a version portable of this one, called "Holter-ECG". A Holter-ECG recording typically lasts 24 hours. It has the advantage of allowing the monitoring of a patient outside the hospital environment but can not be used daily by the population.

 The heart is a muscle that contracts in a relatively regular rhythm. Each normal beat is initiated by an electrical signal generated by the cardiac electrogenic tissue and conveyed by the conduction system of the heart. The ECG signal consists of a succession of electrical depolarizations resulting in the appearance of waves P, Q, R, S, T, U whose appearance is shown in Figure 3. The P wave, which corresponds at depolarization of auricles, presents a small amplitude. The PQ interval reflects the atrioventricular conduction time. The QRS complex reflects the ventricular contraction, and the T wave the ventricular repolarization. In practice, peak R is considered as a marker of ventricular systole, ie, heartbeat. The R wave is the thinnest and largest peak of the ECG signal, and is generally used to mark the moment of heartbeat with very good accuracy.

 To automate the detection and identification of a rhythm disorder and thus replace the eye of the physician, various methods have been proposed such as those of analyzing the morphology of the ECG signal, including the P wave and complex QRS. For example, atrial fibrillation observed on an electrocardiogram is characterized by the replacement of P-waves by rapid oscillations (called fibrillation waves) of varying size, shape, and frequency of occurrence, associated with an often fast irregular ventricular rhythm when atrio - ventricular is intact.

 Other methods propose characterization of the ECG by classification of the RR interval using artificial intelligence methods (Ref 1). The detection of the peaks R makes it possible to form time series whose constituent element is the interval RR, the time between two successive peaks R. To construct a time series of RR intervals, the ECG signal is sampled, digitized and analyzed to detect the R-waves. A time series of RR intervals thus consists of a plurality of successive RR intervals (... RRi 1, RRi, RRi + 1 ...) as illustrated in FIG. 3, each interval RR corresponding to the time interval separating two successive waves R of the ECG signal.

The article "Automatic Detection of Atrial Fibrillation Using RR Interval Signal" (Ref.2) proposes to use the mathematical operators VAI and VLI, SD1 and SD2 to distinguish time series in sinus rhythm from time series with atrial fibrillation. A scatter plot, said Lorenz or Poincaré graph, whose abscissa is the RR interval (RRi) and the ordinate the following RR interval (RRi + 1) of the time series, allows to represent successive RR intervals as scatterplots in which we try to distinguish those related to subjects in sinus rhythm and those attached to subjects with atrial fibrillation.

 But it is also admitted that the analysis of time series alone does not reliably identify a disorder of rhythm. Thus, the document US Pat. No. 7,941,207 proposes to add, at the detection of the intervals RR, a step of detection of the ventricular contractions. Document US Pat. No. 8,155,735 proposes to predict the occurrence of atrial fibrillation by classification of RR intervals resulting from an ECG signal as well as from additional characteristics such as the wavelet transform of the ECG signal and the morphology of the P wave. (width and height).

Still in the context of detecting atrial fibrillation by classification of the RR intervals, US Pat. No. 8,019,407 proposes to compare the RR intervals and the dRR variations of the RR intervals by means of a scatter plot in which they are placed. on the abscissa and the ordinate. Such a graph makes it possible to show clouds of points ("clusters") of different geometries depending on whether the successive RR intervals are in sinus rhythm or have a disturbance of the rhythm. More specifically, this paper suggests a multivariate analysis of the "topography" of these point clouds, for their classification by means of an artificial neural network or a fuzzy logic method. The proposed classification criteria ("metrics") are purely graphical and include the NZCEL number of cells RdR of the graph comprising at least one point, the number NZCOL of columns of cells RdR having at least one point, the number NZROW of rows of cells RdR having at least one point, the NPEAK number of cells having a number of points greater than a threshold, the XDIST trajectory distance of the cloud of points along the RR axis, the maximum difference between two RR intervals, the distance YDIST trajectory of the point cloud along the dRR axis, ie the maximum difference between two dRR variations, the maximum distance DGMAX with respect to a global center of the point cloud, the maximum distance DPMAX with respect to a determined positive center from all positive value dRRs, and the maximum distance DNMAX from a negative center determined from all negative value dRRs. It may, however, be desirable to provide a method for detecting a heart rhythm disorder by time series classification which is simple to implement and provides better accuracy than known methods.

 It may also be desirable to provide a portable device for detecting a cardiac rhythm disorder which is easy to use and not difficult to use, and which can detect a rhythm disorder with good accuracy while offering the possibility of subsequently allowing further monitoring of the subject on which the rhythm disorder has been detected, in a manner that is exploitable by the medical profession.

 Embodiments of the present invention provide a method for detecting at least one cardiac rhythm disorder of a human or animal subject, from a time series of RR intervals representative of the cardiac rhythm of the subject extracted from a physiological signal, comprising the steps of characterizing the time series with at least one descriptive variable whose value is calculated from the RR intervals of the time series, using the value of the descriptive variable, classifying the time series into a first class of time series associated with subjects in sinus rhythm or in at least one second class of time series associated with subjects with cardiac rhythm disorder, in which the value of the descriptive variable is calculated from a derived series whose constituent element is a discrete derivative of order 1 or greater than 1 of the RR interval of the se time.

 According to one embodiment, the constituent element of the derived series is chosen from the group comprising: the rate of variation of the RR interval, ie the discrete time derivative of the RR interval, reflecting the acceleration of the cardiac rhythm, the absolute value of the rate of variation of the interval RR, the rate of variation of the rate of variation of the interval RR, that is the discrete time derivative of the rate of variation of the interval RR, or again the discrete second time derivative of the RR interval, reflecting the jolts of the heart rate, the absolute value of the rate of change of the rate of change of the RR interval, the rate of change of the absolute value of the rate of change of the RR interval, and the absolute value of the rate of change of the absolute value of the rate of change of the RR interval.

According to one embodiment, the step of characterizing the time series also comprises calculating the value of at least one descriptive variable from a time series whose constituent element is chosen from the group comprising the interval RR, the variation of the RR interval, and the absolute value of the variation of the RR interval. According to one embodiment, the value of the descriptive variable is calculated by means of a mathematical operator selected from the group consisting of: the mean value, the median, the standard deviation, the variance, the asymmetry coefficient or Skewness, flattening coefficient or Kurtosis, ULF ultra low frequency power, VLF very low frequency power, LF low frequency power, RF high frequency power, LF / HF ratio, total power, power Normalized LF, the normalized RF power, SD1 or dispersion of the points along the small axis of the ellipse of the Poincaré diagram, SD2 or dispersion of the points along the long axis of the ellipse of the Poincaré diagram, the ratio SD1 / SD2, VAI or the vector angle index, VLI or the vector length index, RMSSD or squared mean of the successive differences of the constituent elements of the series, SDSDD or standard deviation of the absolute value of the ord differentiation re 2 of the constituent elements of the series, the recurrence rate of the recurrence diagram, Lmax or maximum length of the diagonal lines of the recurrence diagram, Lmean or average length of the diagonal lines of the recurrence diagram, determinism, Shannon's entropy of the diagonal length distribution of the matrix of the recurrence diagram, the Shannon entropy, the ApEn entropy, the SampEn entropy, the correlation dimension.

 According to one embodiment, the step of classifying the time series is performed by means of a supervised learning classifier tool, and the method comprises a step of configuring the classifier tool comprising the steps of providing the classifier tool a plurality of values of the descriptive variable calculated from series derived from a first set of time series from subjects in sinus rhythm, and providing the classifier tool with a plurality of values of the descriptive variable calculated from series derived from a second set of time series from subjects with cardiac arrhythmia.

 According to one embodiment, the method comprises a step of continuous or quasi-continuous production of a plurality of time series from the physiological signal, and the real-time determination of the class to which each time series belongs.

 According to one embodiment, the time series are produced according to a sliding time window, so that two successive time series can comprise common RR intervals.

According to one embodiment, the method includes a step of detecting missing RR intervals and wherein the production of time series is interrupted during periods including missing RR intervals. According to one embodiment, the time series are extracted from a photoplethysmographic signal.

 According to one embodiment, the method comprises a decision step by which the cardiac rhythm disorder is considered to be detected when a determined number greater than 1 of time series does not belong to the first class of time series.

 Embodiments also relate to a device for detecting at least one cardiac rhythm disorder of a human or animal subject from a time series of RR intervals representative of the subject's cardiac rhythm extracted from a physiological signal, the device being configured to characterize the time series with at least one descriptive variable whose value is calculated from the RR intervals of the time series, using the value of the descriptive variable, classifying the time series into a first class of series temporal associated with subjects in sinus rhythm or in at least a second class of time series associated with subjects having the cardiac rhythm disorder, the device being configured to calculate the value of the descriptive variable from a derived series whose constituent element is a 1 or higher order derivative of the RR interval of the t-series emporelle.

 According to one embodiment, the device is configured to calculate the value of the descriptive variable from a derived series whose constituent element is chosen from the group comprising the rate of variation of the RR interval, namely the time derivative. of the RR interval, reflecting the acceleration of the heart rate, the absolute value of the rate of change of the RR interval, the rate of change of the rate of change of the RR interval, the discrete temporal derivative of the rate of variation of the RR interval, again the second derivative of the RR interval, reflecting the jolts of the heart rate, the absolute value of the rate of change of the rate of change of the RR interval, the rate of change of the absolute value of the rate of change of the RR interval, and the absolute value of the rate of change of the absolute value of the rate of change of the RR interval.

 According to one embodiment, the device is configured to also calculate the value of at least one descriptive variable from a time series whose constituent element is selected from the group comprising the RR interval, the variation of the RR interval, and the absolute value of the variation of the RR interval.

According to one embodiment, the device comprises a supervised learning classifier tool configured to classify the time series into the first or second classes of time series. According to one embodiment, the device is configured to produce a plurality of time series continuously or quasi-continuously from the physiological signal, and to classify in real time each time series.

 According to one embodiment, the device is configured to produce the time series according to a sliding time window, so that two successive time series can comprise common RR intervals.

 According to one embodiment, the device is configured to detect missing RR intervals and interrupt the production of time series of RR intervals during periods including missing RR intervals.

 According to one embodiment, the device is configured to extract time series of RR intervals from a photoplethysmographic signal.

 According to one embodiment, the device is configured to perform a decision step by which the cardiac rhythm disorder is considered to be detected when a determined number greater than 1 of successive time series does not belong to the first class of time series. .

 According to one embodiment, the device is configured to activate an acquisition channel of the subject's electrocardiogram when a disturbance of the heart rhythm is considered detected.

 In one embodiment, the device is configured to detect at least one atrial fibrillation.

 According to one embodiment, the device is configured to be fixed around the wrist.

 According to one embodiment, the device comprises a photoplethysmographic signal acquisition channel, at least one electrocardiographic signal acquisition channel, and a processor configured to detect the cardiac rhythm disorder in the photoplethysmographic signal and in the electrocardiographic signal.

According to one embodiment, the device is configured to, in a first mode of operation, monitor only the photoplethysmographic signal for detecting cardiac rhythm disorder, and after detecting the cardiac rhythm disorder in the photoplethysmographic signal, switch to a second mode of operation including monitoring the electrocardiographic signal for cardiac arrhythmia. According to one embodiment, the device is configured to, in the second mode of operation, monitor both the photoplethysmographic signal and the electrocardiographic signal.

 According to one embodiment, the device comprises a housing or fixing means receiving an acquisition electrode of the electrocardiographic signal that the subject must touch with a part of his body so that the electrocardiographic signal is captured by the acquisition channel of the electrocardiographic signal.

 According to one embodiment, the device comprises a multiplexing circuit comprising a first input receiving the photoplethysmographic signal, a second input receiving the electrocardiographic signal, and an output connected to the processor, the device being configured for, in an operating mode, selecting alternately each of the two inputs of the multiplexing circuit with a high switching frequency in front of the subject's heart rate.

 These objects and characteristics will be better understood on reading the following description of embodiments of the present invention, given in a nonlimiting manner in relation to the appended figures among which:

 FIG. 1 represents in block form an embodiment of a device for detecting a cardiac rhythm disorder according to the invention,

 FIG. 2 represents in block form another embodiment of a device for detecting a disturbance of the cardiac rhythm according to the invention,

 FIG. 3 shows an electrocardiographic signal that can be used in one embodiment of the device of FIG. 1 or that of FIG. 2,

 FIG. 4 shows a photoplethysmographic signal that can be used in one embodiment of the device of FIG. 1 or that of FIG. 2,

FIG. 5 is a flowchart showing an embodiment of a method for detecting a cardiac rhythm disorder implemented by the device of FIG. 1 or that of FIG. 2,

 FIG. 6 is a flowchart describing a first embodiment of a time series characterization step appearing in the method of FIG. 5,

FIGS. 7, 8 and 9 illustrate results of studies concerning the characterization of time series,

 FIG. 10 is a flowchart describing a second embodiment of the time series characterization step in the method of FIG. 5,

FIG. 11 is a flowchart describing an embodiment of a time series classification step in the method of FIG. 5, FIG. 12 is a flowchart describing a learning step of a classifier tool used for the classification of time series,

 FIG. 13 represents in block form an embodiment of a portable device for detecting a cardiac rhythm disorder according to the invention,

FIGS. 14A and 14B are respectively views from above and from below of the device of FIG. 13,

 FIG. 15 is a flowchart describing functionalities of an embodiment of the device of FIG. 13,

 FIG. 16 is a flowchart describing an embodiment of a method for detecting a cardiac rhythm disorder implemented by the device of FIG. 13,

FIG. 17 is a flowchart describing another embodiment of a method for detecting a cardiac rhythm disorder implemented by the device of FIG. 13,

FIG. 18 is a flowchart describing an embodiment of an RR interval generation step included in the method of FIG. 16 or that of FIG. 17,

 FIG. 19 is a flowchart describing an embodiment of a time series production step included in the method of FIG. 16 or that of FIG. 17,

 FIG. 20 is a flowchart describing an embodiment of a decision step included in the method of FIG. 16 or that of FIG. 17, and

 FIG. 21 is a flowchart describing other functionalities of an embodiment of the device of FIG. 13.

FIG. 1 represents an embodiment of a device DV1 according to the invention. The device comprises a PI processor, a physiological signal acquisition CH channel S, a Mi l program memory, an M12 data memory, and a CI 1 wireless communication interface. The acquisition channel CH is connected. AT terminal coupled to the body of a subject that can be of any known type, including a cardiac probe, skin electrodes or a photoplethysmography module comprising emitter and receiver diodes. The acquisition channel CH supplies the signal S in digitized form to the processor P1. The processor analyzes this signal to detect a disturbance of the cardiac rhythm, by means of program-algorithms PG1, PG2, PG3, PG4 provided in the memory Mi. . A clock circuit CCT may be provided to provide a clock signal CK usable as a reference time base for the measurement of RR intervals. The DV1 device may be of implantable type, for example subcutaneous, portable type, or fixed type that the subject uses temporarily during a phase of observation of his heart rate. Various other members that can be provided in the device DV1 are not shown, such as a power supply, voltage regulators, a display, physiological data or posture sensors, for example a temperature sensor, an accelerometer, a magnetometer ...

 FIG. 2 represents another embodiment of a device DV2 according to the invention. The device comprises a processor P2, a program memory M21, an HD data storage means, for example a hard disk, a wireless communication interface CI2 and a circuit CCT providing a clock signal CK. The program memory M21 comprises the aforementioned program-algorithms PG1, PG2, PG3, PG4 for the analysis of the physiological signal S and the detection of a disturbance of the cardiac rhythm. The signal S may be prerecorded in the HD data storage means, or received from the DV1 device via the communication interface CI2 for analysis in real time or after being recorded in the HD storage means. The device DV2 can be a personal computer or the computer of a doctor, a workstation in a medical laboratory, a medical server, and generally any device equipped with calculation means for implementing the PG1 programs to PG4.

FIGS. 3 and 4 show two examples of physiological signals S that can be analyzed by the device DV1 or DV2. The signal shown in Fig. 3 is an ECG signal, while the signal shown in Fig. 4 is a photoplethysmographic signal, or PPG signal. The ECG signal shows the QRS complex corresponding to the depolarization of the ventricles, from which we can extract the peak R for the measurement of RR intervals (RRi-1, RRi, RRi + 1 ...). The PPG signal has a roughly sinusoidal appearance and has peaks denoted "R" by analogy with the electrocardiographic signal, the occurrence of which is representative of the subject's cardiac activity. At each heartbeat, the moment of occurrence of a peak "R" of the signal PPG is correlated, with a slight phase shift, at the time of occurrence of the peak R of the ECG signal, because the signal PPG accounts for the pulsatile nature Pulse resulting from cardiac contractile mechanical activity, rheological properties of the blood and mechanical properties of the vessels. Thus the pseudo-period defined by the duration of the peak-to-peak interval of the PPG signal is correlated with the duration of the RR interval. The detection of the "R" peaks of the photoplethysmographic signal and the measurement of the time intervals between these peaks thus makes it possible to produce time intervals which will be considered in the following as RR intervals. The P waves and the fibrillation waves of the ECG signal are against electrical events without macroscopic mechanical counterpart, and are not observable in the PPG signal.

 FIG. 5 shows steps of the method for detecting a cardiac rhythm disorder implemented by the device DV1, DV2 by means of the programs PG1 to PG4. The method comprises:

 a step S03 for detecting peaks R (or similar) in the physiological signal S, and for calculating intervals RR, executed by the program PG1,

 a step S05 of formation of time series Sj of intervals RR from the intervals RR provided by the program PG1, executed by the program PG2;

a step S07 for characterizing the time series Sj provided by the program PG2, executed by the program PG3, and

 a step S09 of classifying the time series Sj by the program PG4, based on their characterization by the program PG3.

 In step S03, the measurement of the RR intervals by the program PG1 may include a filtering for identifying aberrant RR intervals due for example to false detections and / or non-detections of RR intervals. In step S05, the program PG2 "cuts" the continuous stream of intervals RR in time series Sj of a minimum duration Te, each comprising a number of successive intervals RR which is a function of the heart rate of the subject. In step S07, the program PG3 analyzes the intervals RR constituting the time series Sj to extract a discriminant information necessary for their classification. In step S09, the program PG4 uses the discriminant information to classify the time series Sj in a first class C 1 of time series from subjects in sinus rhythm or in a second class C2 of time series from subjects presenting the disorder the rhythm to be detected, for example atrial fibrillation.

When a disturbance of the rhythm is detected in real time by the device DV1 by classification of one or more time series in the class C2, the classification step S09 is followed by an action performed by the device DV1. This action may for example comprise storing in the data memory M12 the raw signal S and / or the time series Sj in which the disturbance of the rhythm has been detected, and / or transfer these data via the communication interface CH . The device DV 1 can for example establish a communication with the device DV2 and transfer to it the signal S, the time series Sj and their classification. The device DV2 is then able to realize, in deferred time, a verification of the classification provided by the device DV 1, or even to submit the signal S to other types of analyzes to confirm the validity of the classification retained by the device DVl. The action following the classification of one or more time series in class C2 may also include a decision step preceding an action proper, aimed at validating the classification chosen by the program PG4.

 It is known that a time series of RR intervals can be characterized, for classification purposes, by means of descriptive variables whose values are calculated by applying mathematical operators to the RR intervals of the series. The aforementioned article "Automatic Detection of Atrial Fibrillation Using R-R Interval Signal" (Ref.2) suggests using the operators VIA and VLI or SD1 and SD2 phase space analysis or Poincaré diagram. These mathematical operators, as well as others described below, for example the mean value, the median, the standard deviation, the variance of the RR intervals, etc., make it possible to obtain discriminant information that can be used for classifying the series. time.

 FIG. 6 shows a first embodiment S07 (1) of the characterization step S07 implemented using the program PG3. Step S07 (1) comprises a preliminary step S070 of choice of a group of descriptive variables Val, Va2 ... Vak, corresponding here to the choice of a group of mathematical operators intended to be applied to time series d. RR intervals. Step S070 is preferably performed before commissioning the DV1, DV2 device and writing the PG3 program, through studies and trials to determine the best combination of descriptive variables in relation to the disorder. the rhythm to detect. In one embodiment, the DV1, DV2 device is configured to detect several types of rhythm disorders and uses different groups of descriptive variables, each dedicated to the detection of a pathology. In this case, the step S070 can include a step of selecting, in the program memory Mi1, M12, the appropriate group of descriptive variables, or a step of selecting a branch of the program PG3 using the group of descriptive variables dedicated the detection of the target rhythm disorder.

Step S07 then comprises a loop for calculating the values of each descriptive variable, which is initiated after reception, during a step S071, of a new time series Sj provided by the program PG2. The calculation loop comprises a step S072 for calculating the value Valj of the descriptive variable Val, a step S073 for calculating the value Va2j of the descriptive variable Va2, and so on until a step S07k for calculating the value Vakj of the descriptive variable Vak, the value of each variable being calculated from the RR intervals of the series Sj. The program PG3 then supplies the program PG4 with the values Valj, Va2j ... Vakj of the descriptive variables. The effectiveness of this characterization process has been evaluated from RR intervals provided by the Massachusetts Institute of Technology (MIT) Physionet site (http://www.physionet.org/), which is available to the public. ECG signal banks and RR interval time series, including "Sinus Rhythm RR Interval Database", "MIT-BIH Normal Sinus Rhythm Database", "MIT-BIH Atrial Fibrillation Database", "AF Termination Challenge Database" , and "MIT-BIH Arrythmia Database". From these databanks, a time series library of RR intervals has been constituted, including:

 - 56964 reference time series of 60 s each of which 56711 in sinus rhythm and 253 in atrial fibrillation ("atrial fibrillation"), and

 322779 reference time series of 10 s each of which 297269 in sinus rhythm and 25 in atrial fibrillation.

 The study revealed a significant percentage of false negatives in a classification of reference time series by logistic regression or by means of a supervised learning classifier tool. In other words, a significant number of time series in atrial fibrillation have been classified as time series in sinus rhythm.

 It has therefore been desired to improve the time series characterization method implemented in step S07, to obtain a more accurate classification without involving descriptors based on the analysis of the ECG signal morphology, which would not be possible. not usable with another type of signal, in particular the signal PPG. An improvement has been found, which will now be described in connection with FIGS.

11.

According to this improvement, the step S07 of time series characterization includes derived series calculation steps whose constituent elements are discrete derivatives, denoted by ^N RR / dt ^N , RR intervals of the initial time series, N being an integer at least equal to 1. It has indeed been demonstrated that such derived series contain discriminant information that can be added to that which can be extracted from the initial time series, or even replace it, with a view to their classification.

For the sake of clarity, the table of FIG. 7 illustrates, with shades of gray, a degree of correlation between derivatives of RR intervals of increasing order N. This degree of correlation is measured on a scale of 0 to 1 using the Pearson coefficient. The table in Figure 7 is reproduced in Appendix 1, Table 1, with shades of gray replaced by numerical values. The central diagonal from the upper left corner to the lower right corner of the table shows that, trivially, the RR intervals and their derivatives have in themselves a correlation coefficient equal to 1. The diagonals close to the central diagonal show that the derivatives of RR intervals of adjacent orders (for example a second-order derivative and a derivative of order 3) have between them a degree of correlation of less than 1, for example of the order of 0.8. This degree of correlation decreases to tend toward 0 as the rank difference between the derivatives increases. Similarly, there is a low degree of correlation between the RR intervals and their derivatives, from the first-order derivative. These results testify to the existence, in the derived series, of a discriminant information not found in the initial time series.

 Thus, from an initial time series to be characterized for classification, the method comprises the following steps:

a step of calculating at least one derived series of order N whose constituent element is the discrete time derivative of ^N RR / dt ^N of the interval RR of the initial time series, and

a step of characterizing the derived series by means of one or more descriptive variables whose values are calculated by applying a mathematical operator to the constituent elements of the derived series, and are considered as forming discriminant information for the classification of the initial time series.

 For the sake of clarity of the presentation, in what follows:

 derived descriptive variable Vb: a descriptive variable obtained by applying a mathematical operator to a derived series,

 - primitive series: a time series of RR intervals or a time series whose constituent element is the variation of the RR interval or the absolute value of this variation, and

 primitive descriptive variable Vc: a descriptive variable obtained by applying a mathematical operator to a primitive series.

 Thus, the notion of "primitive descriptive variable" encompasses the notion of "descriptive variable" previously mentioned, which refers only to the application of a mathematical operator to a time series of RR intervals.

 On the other hand, in one embodiment, the concept of "derived series" includes series whose constituent element is calculated from the absolute value of the elements of a lower order derived series, for example, as will be see below, the rate of change of the absolute value of the rate of change of the RR interval, or the absolute value of this rate of change.

Thus, if Nmax denotes the maximum degree of derivation used to characterize a time series, ie the number of derived series available, and NVa the number of mathematical operators that can be applied to the derived series, the number of derived descriptive variables that can be used is equal to NVa * Nmax. By adding to the derived series those whose constituent element is calculated from the absolute value of the elements of a lower-order derived series, the number of derived descriptive variables that can be used is even greater.

 Examples of mathematical operators and their corresponding formulas, usable for defining primitive or derived descriptive variables, are described in Appendix 2. There are:

 statistical operators, such as:

 - the average value,

 - the median,

 - the standard deviation,

 - the variance,

 - the dissymmetry coefficient or Skewness,

 - the coefficient of flattening or urtication,

 spectral analysis operators (see Ref 3), such as:

 - Ultra Low Frequency Power (ULF) or ultra low frequency power [0 - 3 mHz],

 - Very Low Frequency Power (VLF) or very low frequency power [3 mHz - 40 mHz],

 - Low Frequency Power (LF) or low frequency power [40 mHz

- 150 mHz],

 - High Frequency Power (HF) or high frequency power [1 0 mHz 400 mHz],

 - the ratio LF / HF,

 - the total power,

 - the normalized LF power [40 mHz - 150 mHz],

 - the normalized HF power [150 mHz - 400 mHz],

 phase phase analysis operators (Poincaré diagram, see Ref.2) such as:

 SD1, or dispersion of the points along the minor axis of the ellipse,

 SD2, or dispersion of the points along the long axis of the ellipse,

 the ratio SD1 / SD2,

 - VAI, or Vector Angle Index,

- VLI, or Vector Length Index, - RMSSD, or quadratic mean of the successive differences of the constituent elements of the series to be characterized,

 SDSDD, or Standard deviation of the absolute value of the differentiation of order 2 of the constituent elements of the series to be characterized,

 - recursion pattern analysis operators RP (Recurrence Plot) such as:

 - the recurrence rate (recurrence rate),

 - Lmax, or Maximum length of the diagonal lines of the RP recurrence diagram,

 - Lmean, or Average length of the diagonal lines of the recurrence diagram RP,

 - Determinism,

 - Shannon entropy of the distribution of the lengths of the diagonal lines of the matrix of the RP diagram,

 operators for analyzing the regularity of the series to be characterized, such as:

 - Entropy of Shannon,

 - ApEn entropy (Approximate Entropy),

 - Entropy SampEn (Sample Entropy), and

 operators for analyzing the chaotic and fractal properties of the series to be characterized, such as the correlation dimension.

 The characterization of a time series according to the method currently described can be implemented according to several variants:

 - using exclusively one or more derived descriptive variables Vb, or

using several derived descriptive variables Vb and one or more primitive descriptive variables Vc.

Studies have been carried out to evaluate the best combinations of derived descriptive variables Vb and primitive descriptive variables Vc making it possible to obtain a good accuracy of classification ("accuracy") without complicating in a crippling manner the characterization step, and this in the perspective of implementation of the method in a device of a reduced cost, offering an average computing power. These studies show that it may be preferable, in some embodiments, to use only primitive or derived descriptive variables formed from different mathematical operators, in other words not to apply the same mathematical operator to different primitive or derived series. It also appears that the use of primitive descriptive variables may not be necessary with some derived descriptive variables, while with other descriptive variables derived the addition of one or more Primitive descriptive variables can improve the accuracy of the classification. Finally, it appears that this choice also depends on the disorder of the rhythm to be detected.

 For example, the table in Figure 8 illustrates the result of a study to identify the best combination of primitive and derived descriptive variables using as mathematical operators the mean value and the standard deviation ( "σ"). The table in Figure 8 is also reproduced in Appendix 1, Table 2, the shaded areas being replaced by crosses. The study was conducted under the following conditions

- by fixing the maximum degree of derivation at 4 (Nmax = 4), or by using derived series of order 1, 2, 3 and 4,

 using the mean value and the standard deviation to form combinations of primitive or derived descriptive variables comprising from 1 to 10 variables,

 - in the context of a classification of time series in a class of time series associated with subjects in sinus rhythm (class NSR) or in a class of time series associated with subjects having atrial fibrillation (class AF),

- using a logistic regression function to distinguish the NSR and AF classes, and

 using the aforementioned reference time series, generated from the Physionet data, whose class of membership is known.

Each column of the table corresponds to a combination of variables comprising from 1 to 10 variables. Gray cells (or, in Appendix 1, those with crosses) indicate the best choice of variables. When two variables are used (column 2), the offeror the best classification accuracy is ^σ the ^difference -tyP ^e of the constituent elements of the second derivative and the average value of the constituent elements of the first derivative. With these two derived descriptive variables, a classification accuracy of 99.9% was obtained without using primitive descriptive variables. From column 5 (ie 5 variables used) the standard deviation of the RR intervals of the time series is part of the optimal combination of variables. From column 9 (ie 9 variables used) the average value of RR intervals is also part of the optimal combination of variables. These two primitive descriptive variables have therefore been preferred to certain derived descriptive variables, for example

The study of all possible combinations of primitive and derived descriptive variables showed that the accuracy of the classification could vary between 69.6% and 99.9% depending on the combination chosen, hence the importance of the choice of variables. With a result as satisfactory as 99.9% obtained with only two derived descriptive variables judiciously chosen, it appears that the choice of a large number of primitive and / or derived descriptive variables may not be necessary. Thus, the shaded cells in columns 3 to 10 of the table include the two derived descriptive variables retained in column 2 of the table, which alone provide a classification accuracy of 99.9%, so that the addition of the other variables in these columns does not improve the accuracy of the classification.

 Other studies have been conducted to evaluate the influence of the duration of the observation window, that is to say the minimum duration Te of each time series (step S05, Fig. 5) on the relevance of the classification. FIG. 9 illustrates one of the results of these studies and shows a curve C1 corresponding to the variation of the accuracy of the classification, expressed as a percentage, as a function of the duration of the observation window, expressed in seconds, in using the pair of derived descriptive variables

Other curves C2, C3. ... Cn appearing in Figure 9

correspond to the variation in the accuracy of the classification for other combinations of variables. The curve C l shows that the duration of the observation window can be reduced to 5 seconds, which represents only 5 heart beats with a subject whose heart rate is 60 beats per minute, while maintaining an accuracy classification higher than 95%. However, the reduction of the duration of the observation window makes it possible to reduce the time of classification of each time series and thus to increase the reactivity of the device DV1, DV2 during the appearance of a disturbance of the rhythm, in the real-time monitoring of the subject's cardiac activity.

 It will be noted that the analyzes, conclusions and results of the studies which have just been described are purely indicative and can not be considered as reducing the scope of the method currently described.

FIG. 10 shows an embodiment S07 (2) of the characterization step S07 according to this method. Step S07 (2) comprises a preliminary step S0700 of choosing at least one derived descriptive variable Vb. As before, this choice is preferably made before the commissioning of the DV 1 or DV 2 device, through prior studies, and depending on the rhythm disorder to detect. This is for example the average value of the first derivative and the standard deviation of the second derivative, the discriminant performance of which has been highlighted above in relation to FIG. 8. Optionally, at least one variable descriptive primitive Vc can also be detention. Finally, in the context of detecting different rhythm disorders, this choice can be made dynamically by the device DV1, DV2 as a function of the rhythm disorder to be detected, by selecting a group of variables from among several predetermined groups or PG3 program branches configured to use these variables.

 After receiving a time series Sj during a step S0701, the program PG3 performs all or some of the following steps:

 step S0702: calculation of a series Sjo whose constituent element is the variation Vi of the RR interval of the series Sj,

 step S0703: calculation of a series Sji whose constituent element is the absolute value I Vi | the variation of the RR interval,

step S0704: calculation of a derived series Sj ^' 2 whose constituent element is the rate of variation Ai of the RR interval (acceleration of the cardiac rhythm),

step S0705: calculation of a derived series Sj ^' 3 whose constituent element is the absolute value | Ai | the rate of change Ai of the RR interval (acceleration in absolute value of the cardiac rhythm),

step S0706: calculation of a derived series Sj ₄ whose constituent element is the rate of variation ACi of the variation rate Ai of the RR interval (jerk of the cardiac rhythm),

step S0707: calculation of a derived series Sj ^' 5 whose constituent element is the absolute value | ACi | the rate of variation ACi of the variation rate Ai of the RR interval (jerk in absolute value of the cardiac rhythm),

 step S0708: calculation of a derived series Sjô whose constituent element is the rate of change Gi of the absolute value | Ai | the variation rate Ai of the RR interval,

step S0709: calculation of a derived series Sj ₇ whose constituent element is the absolute value | Gi | the rate of change Gi of the absolute value | Ai | the rate of change Ai of the RR interval.

 Examples of calculation formulas for the various derived series just mentioned are given in Annex 3. Once these calculation steps are executed, or as they are executed, the program PG3 performs the following steps:

 step S0720: determining the value Vclj of at least one primitive descriptive variable Vcl calculated from the series Sj,

 step S0730: determination of the value Vc2j of at least one primitive descriptive variable Vc2 calculated from the Sjo series,

step S0740: determination of the value Vc3j of at least one primitive descriptive variable Vc3 calculated from the Sji series, step S0750: determination of the value Vbj of at least one derived descriptive variable Vb from a derived series Sj ^' 2 to Sj ₇ .

 Steps S0720, S0730 and S0740 are optional and are not executed if the choice made in step S0700 does not include a primitive descriptive variable. Thus, the method may comprise only two of these steps, only one of these steps, or none of these steps. Similarly, the number of derived descriptive variables whose values are calculated in step S0750 depends on the choice made in step S0700. Finally, some of the steps S0704 to S0709 may not be executed if the derived derived descriptive variables do not require the calculation of the corresponding derivatives.

 The program PG3 then supplies the program PG4 with the value or values Vbj of one or more derived descriptive variables Vb, and optionally the value or values Vcj of one or more primitive descriptive variables Vc, and returns to the step S0701 to wait to receive a new time series Sj.

 In one embodiment of the detection method, the program PG4 is a classifier tool comprising a network of artificial neurons (RNA) vector quantization and supervised learning, called "LVQ" ("Learning Vector Quantization"). It has a hidden layer called competitive layer, followed by a classification layer. The competitive layer contains hidden neurons, the classification layer contains output neurons, each of which is representative of a membership class of vectors of a learning base. The results provided by this classifier tool architecture are of the order of 100% in learning and 99.9% in cross validation. The robustness of the model has also been proven in the presence of various special cases such as the presence of ectopic beats, a sinus rhythm with a strong respiratory sinus arrhythmia, etc.

Fig. 11 shows an embodiment of the classification step S09. Prior to being commissioned, the PG4 classifier tool is subjected to a supervised learning configuration step S0900 using two sets NI, N2 of reference time series. Once the learning is finished, the program is able to classify time series in two classes C1, C2 of time series corresponding respectively to sets NI and N2. In one embodiment, the classifier tool PG4 is configured to distinguish a class C1 of time series associated with subjects in sinus rhythm and a class C2 of time series associated with subjects in atrial fibrillation. In a step S091, the classifier tool PG4 receives one or more values Vbj of derived descriptive variables Vb calculated by the program PG3 at the step S07 (2) for a time series Sj, and optionally one or more values Vcj of primitive descriptive variables Vc. At a step S092, the classifier tool PG4 evaluates, from these variables, the membership class of the series Sj. The classifier tool PG4 provides this classification and returns to step S091 to receive new values of these variables associated with a next time series.

 Fig. 12 shows an embodiment of the step of configuring the classifier tool PG4. This can be performed using a version of the classifier tool installed on a workstation and aims to obtain a set of MC4 configuration matrices. It includes the following preparatory steps:

 a step S0901 for preparing a set I of time series derived from subjects in a smusal rhythm, for example a reference time series of one minute each from subjects in sinus rhythm and produced from the database of Massachussetts Institute of Technology,

 a step S0902 for characterizing the time series of the set NI. Here we use the same variables that will be used during the characterization step S07 (2). This step therefore comprises, for each series Sj, the computation of values Vbj of derived descriptive variables Vb, and if necessary the computation of values Vcj of primitive descriptive variables Vc,

 a step S0904 for preparing a set N2 of time series from subjects presenting the disturbance of the rhythm to be detected, for example a reference time series of one minute each from subjects in atrial fibrillation and produced from the database of the Massachusetts Institute of Technology,

 a step S0905 for characterizing the time series of the set N2. The same variables as those used during the characterization step S07 (2) are also used. This step therefore comprises, for each series Sj, the computation of values Vbj of derived descriptive variables Vb, and where appropriate the computation of values Vcj of primitive descriptive variables Vc.

The values Vbj, Vcj of the variables Vb, Vc calculated for the time series of the set NI are then supplied to the classifier tool PG4 with the status "class C1" during a step S0903. Likewise, the values Vbj, Vcj of the variables Vb, Vc calculated for the time series of the set N2 are supplied to the classifier tool PG4 with the status "class C2" during a step S0906. From all the values of variables provided for the class C 1 and those provided for the class C2, the classifier tool learns, during a step S0907, to distinguish the classes C1 and C2. This is a supervised learning since classes are imposed on the classifier tool when providing the values of the variables in steps S0903 and S0906. The configuration of the classifier tool resulting from the learning phase, taking the form of a set of configuration matrices MC4, is then saved during a step S0908. During a step S0909, the configuration matrices MC4 are loaded in the memory Mi1, M21 of the device DV1, DV2 (FIGS. 1 and 2).

 It will be apparent to those skilled in the art that the method of characterizing and classifying time series which has just been described is capable of various variants, embodiments and applications. Although the method was originally designed to monitor the cardiac activity of a human subject, it is also applicable to the detection of cardiac disorders in animals.

 In addition, the PG4 classifier tool can be configured to provide a classification of rhythm disorders in several classes C2 (1), C2 (2), C2 (3) ... each corresponding to a particular rhythm disorder (atrial fibrillation , atrial fiutter, tachyarrhythmia, supraventricular tachycardia, sinus tachycardia, atrial or ventricular extrasystole, ventricular tachycardia, ventricular fibrillation ...). As a variant, several classifying tools, each configured to distinguish two classes, namely the class C1 and one of the classes C2 (1), C2 (2), C2 (3), can be provided and intervene with one another. after the others during the classification step S09. A classification conflict arbitration algorithm may be provided in the case where several classifiers provide different classifications.

 Also, the classification can be implemented with other types of classifying tools than that previously described, in particular a logistic regression classifier tool or other types of supervising neural networks such as those designated "LVQ2", " LVQ3 "," OLVQ3 "in the literature. Specific learning for an individual with a risk of rhythm disorder may also be provided. In this case, the configuration of the classifier tool includes a first supervised learning from an existing database relating to a population, as described above, then a personalized learning in relation to the subject, under medical supervision. .

It will also be apparent to one skilled in the art that the derived series formulas provided in Appendix 3 are not limiting and result from a choice consisting of an observation direction, on the right, corresponding to the direction of flow of the time. Thus, from a moment Ti, we observe what happens at a time Ti + n, n ranging from 1 to the maximum degree of derivation chosen. Conversely, one could, in another embodiment, study what happened before the instant Ti at the moment Ti-n, the derivatives at the right (at i + n) not being equal to the derivatives on the left (in in). This results from the fact that the time series observed are not continuous functions and that the derivatives considered here are discrete time derivatives. Other embodiments may include both forward-looking derivatives (right-derivative) and past-derived derivatives (left-derivative), to obtain even more information.

 It will also be apparent to those skilled in the art that the calculation steps just described can be carried out from the instantaneous heart rate which is calculated by means of the formula 1 / RRi (number of beats per minute). second) or the formula 60 / RRi (number of beats per minute), the interval RRi then to be expressed in seconds. In this case, the series of RR intervals are replaced by series of instantaneous cardiac frequencies, and the derivatives of the series of intervals RR are replaced by derivatives of series of instantaneous heart rates, which are calculated in the same way as that indicated in Annex 3 by replacing the RR interval with the instantaneous frequency. Thus, in the present description and in the claims, the term "RR interval" should be understood as also referring to the instantaneous heart rate, the term "RR time series" should be understood as also referring to instantaneous heart rate series. , and the term "derived series" is to be understood as also referring to a series derived from a series of instantaneous cardiac frequencies.

 FIG. 13 represents an embodiment of a device DV3 according to the invention, intended to be worn by a user. The device comprises a channel CH1 for acquiring a photoplethysmographic signal SI, a channel CH2 for acquiring an electrocardiographic signal S2 (ECG) and optionally other CHn channels for acquiring signals Sn, for example a second channel method of acquiring an electrocardiographic signal, a temperature acquisition channel, a signal acquisition channel provided by an accelerometer, and a signal acquisition channel provided by a magnetometer.

 The channel CH1 is coupled to at least one light emitting diode ED and at least one PD photodiode. The CH2 channel is coupled to two dry electrodes El, E2. The signals S1, S2 ... Sn provided by the different acquisition channels are applied to inputs of a multiplexer MUX whose output is connected to a processor P3 via an ADC analog-to-digital converter. A LPF low-pass filter can be provided between the output of the multiplexer and the ADC converter, in order to remove noise components that may be present in the signals SI and S2 or in one of these signals.

The DV3 device also includes an M31 program memory, an M32 data memory, an IC3 wireless communication interface, a DS display and a clock circuit CCT providing a clock signal CK forming a time base for the measurement of intervals RR. The multiplexer MUX receives a selection signal SEL supplied by the processor P3 and transfers thereto the corresponding signal SI, S2,... Sn selected on one of its inputs. Other peripheral means of the processor P3, schematized by a PD block, may include a battery or any other power source, a power management circuit, regulators providing different bias voltages, a USB port, a piezoelectric pager. (buzzer), a buzzer, an inertial microcentral, etc.

 FIGS. 14A and 14B show the device DV3 respectively by a view from above and a view from below. The device is mounted in a protective case 10 equipped with a bracelet 11. The top of the housing 10 receives the DS display. The underside of the housing 10 receives the electrode El, an auxiliary electrode 12 (reference potential electrode), and a micromodule 13 of photoplethysmography. The photoplethysmography module 13 comprises, for example, three electroluminescent electrodes ED and a photodiode PD. The electrode E2 is arranged here on an outer face of the bracelet 11, FIG. 14A, but could also be arranged on one face of the case 10.

 For the acquisition of the signal ECG S2, the electrode E being in permanent contact with the skin, the user must touch the electrode E2 with a finger of the hand opposite to that which receives the device DV1, or with any part from this hand, for example the top of the hand. The difference in bioelectrical potential for the acquisition of the electrocardiographic signal is thus optimal since measured between two ends of the body.

 In other embodiments of the device DV3, the electronic means shown in FIG. 13 can be integrated in a bracelet, the assembly then being devoid of a case. These electronic means can also be integrated in the case of a watch, or in the wristband of a watch. In a variant, the electrodes E1, E2 are cutaneous electrodes connected to the device DV3 by wires and microconnectors. This embodiment may be provided for subjects who are already known to have a disturbance of the rhythm, to be closely monitored. In another embodiment, the electrodes E1, E2, instead of being connected to the device DV3, are connected to an electronic module configured to transfer the ECG signal S2 to the device DV3 via a wireless communication channel.

An advantageous general characteristic of the device DV3 is that it combines the ease of acquisition of the photoplethysmographic signal SI (no skin electrodes to be placed on the body of the user) while allowing to acquire the electrocardiographic signal S2 when this is done. 'proves necessary, here by a simple press on the electrode E2, if not by means of wound skin electrodes or connected to a wireless transmitter.

 Fig. 15 shows an operating configuration of the DV3 device based on this feature. The DV3 device has a "continuous sleep" operating mode and an "alert" operating mode. In the "continuous sleep" mode, the processor P3 activates the channel CH1 and continuously analyzes the photoplethysmographic signal SI. The processor switches to "alert" mode when a rhythm disorder is detected in the photoplethysmographic signal. The processor then activates the channel CH2, selects the electrocardiographic signal S2 by means of the multiplexer MUX and requests the user, by any means provided for this purpose (display DS, ringing, buzzer, voice message ...) to touch the electrode E2 for a determined time, for example for one minute. The electrocardiographic signal is then duly analyzed and if the rhythm disorder is confirmed, the device can ask the user for an urgent measure and / or himself conduct one or more actions to protect him.

 In one embodiment of the "alert" operating mode, the processor P3 alternately selects the photoplethysmographic signal SI and the electrocardiographic signal S2 by rapidly switching the inputs of the multiplexer MUX by means of the signal SEL, for example with a frequency of the order of 1000 Hz or more, about 1000 times the average heart rate of a person at rest. In this case, the multiplexer MUX provides the ADC converter signals SI, S2 in a pseudo-sampled form which is digitized by the ADC converter before being analyzed by the processor P3. Other Sn signals can also be simultaneously provided to the processor in the "alert" mode. In particular, a second channel for acquiring the ECG signal, using another acquisition circuitry to be chosen from among various known circuits, connected to the electrodes E1, E2 or using other electrodes, can make it possible to reinforce the reliability of the acquisition and the analysis of the ECG signal by analysis of the two ECG signals and arbitration of the results obtained.

 The analysis of the signal S1 or S2 by the processor P3 is provided by various program-algorithms loaded into the memory M31, in particular:

 a program PG01 for digital filtering and suppression of noise in the signal SI,

a program PG02 for digital filtering and noise suppression in the signal S2,

a program PG11 for detecting "R" peaks in the signal SI (peaks PPG) and measurement of the intervals RR, equivalent to the program PG1 previously described, a program PG12 for detecting the peaks R in the signal S2 (QRS complex) and for measuring the intervals RR, equivalent to the program PG 1 previously described,

 the program PG2 of time series formation Sj already described,

 the program PG3 for characterizing the time series Sj, preferably in its improved version using the derived descriptive variables Vb and primitives Vc,

the program PG4 for classifying time series, for example the classifier tool previously described and its configuration matrices MC4.

 A PG5 decision program is also planned. The program PG5 receives the time series classifications provided by the classifier tool PG4 and decides, in view thereof, whether it can be considered that a rhythm disorder has been detected. Finally, a PAP application program is provided to manage the general features of the device and its modes of operation.

 Figure 16 shows steps of a method for detecting a cardiac rhythm disorder performed by the DV3 device using the aforementioned programs. The method comprises an SOI channel selection step CH1 and / or CH2 and steps specific to each of the signals SI and S2, namely:

 a step S02 (1) for filtering the signal S1 by the program PG01 followed by a step S03 (1) for detecting the "R" peaks (PPG peaks) and for measuring the RR intervals by the program PG11, and

a step S02 (2) for filtering the signal S2 by the program PG02 followed by a step S03 (2) of detecting the peaks R (QRS complex) and measuring the intervals RR by the program PG12.

 The method then comprises signal processing steps common to each of the signals S1, S2, but applied separately thereto, including:

the previously described step S05 for forming time series Sj from the intervals RR provided by the program PG11 or PG12, executed by the program PG2,

the previously described step S07 (2) (FIG. 10) for characterizing the time series Sj, executed by the program PG3, and

 step S09 previously described (FIGS. 5, 11) of classification of the time series Sj by the classifier tool PG4, in the class C1 (normal subject) or class C2 (subject presenting a disturbance of the cardiac rhythm).

Step S09 is followed by a decision step SI 1 executed by the program PG5, based on the classification information provided by the classifier tool PG4. In one embodiment, the program PG5 confirms the detection of a disturbance of the rhythm when a determined number "D" of successive time series have been attached to the class C2. Thus, an isolated classification of a time series in the C2 class is not considered sufficient to consider that the subject has a disorder of the rhythm, as well as several classifications in the C2 class of time series that are not successive.

 In one embodiment, the program PG5 may equivalently be configured to confirm the detection of a rhythm disorder when "D" successive time series have not been classified in the class Cl. This embodiment applies especially when the classifier tool is configured to provide a classification of rhythm disorders in several classes C2 (1), C2 (2), C2 (3) ... each corresponding to a disorder of the rhythm. the decision then taken without investigating whether the detected disorders were attached to the same class among all available C2 classes.

 Finally, when the decision step S 11 indicates that a rhythm disorder has been detected, the method comprises a step S 12 of initiating a specific action, carried out by the PAP application program, aimed at protecting the subject and / or the collection of information enabling the medical profession to diagnose the event.

 Fig. 17 shows another embodiment of a method for detecting a rhythm disorder performed by the DV3 device. This embodiment differs from the previous one by the fact that, when the electrocardiographic signal S2 is analyzed, the step of characterization S07 (2) of the time series is replaced by a characterization step S07 (3) carried out by a program PG32 which replaces the PG3 program. This step S07 (3) includes, in addition to the characterization of the time series in the manner previously described, a step of characterizing the morphology of the ECG signal, which comprises, for example, the analysis of the shape of the P wave and the QRS complex, for example their amplitude and duration. In this case, the classification step S09 is replaced by a classification step S09 (2) conducted by a second classifier tool PG42 which replaces the program PG4. The program PG42 has MC42 configuration matrices obtained at the end of a learning phase based both on the characterization of the time series and the characterization of the morphology of the signal S2.

Thus, when a disturbance of the rhythm has been detected by analysis of the photoplethysmographic signal SI, the switchover in the "alert" mode enables the device DV3 to acquire the electrocardiographic signal S2 and to conduct an accurate analysis thereof, allowing the PG5 program to provide very reliable confirmation of the presence of a rhythm disorder. It will be noted, however, that the refinement proposed above for the characterization of time series using variables derived descriptive Vb, already allows to obtain a reliable detection based on the observation only of the photoplethysmographic SI signal.

 The algorithm executed by the processor P3 under the control of the program PG11 or PG12 in the step S03 (1) or S03 (2) is preferably designed to detect and eliminate the aberrant RR intervals and thus to further improve the reliability of the method detecting a rhythm disorder. FIG. 18 shows an embodiment of this algorithm, which can also be used to implement the step S03 of FIG. 5. The algorithm comprises two steps S030 and S031 executed in background tasks and a calculation loop of FIGS. RRi intervals. The step S030 consists in receiving the discrete values of the digitized signal SI or S2 and the step S031 consists in the analysis of this signal for the detection of the peaks R or the like (peaks PPG).

 The calculation loop is initiated after detecting, at a step S0302, a peak Ri (Ti), "Ti" denoting the instant when the peak Ri is detected. It comprises steps S0300 of initialization of a loop variable i (i = 0) and S0301 of initialization in the low state of an IFR interrupt flag. After detection of the peak Ri (Ti), the processor P3 verifies during a step S0303 whether a previous peak Ri-1 (Ti-1) of instant Ti-1 has been memorized. If so, the processor determines, during a step S0304, the value of the interval RRi, such that RRi = Ti (Ti-1), and then verifies during a step S0305 whether the interval RRi is below a threshold Tmin. If not, the processor stores the peak Ri (Ti) during a step S0306, and then determines during a step S0307 whether the interval RR is greater than a threshold Tmax. If not, the processor supplies the program PG2 with the interval RRi during a step S0308. The RRi interval is accompanied by the IFR interrupt flag, which may be in the low or high state depending on the previously executed steps. In a next step S0309, the processor forces the interrupt flag IFR in the low state, then carries out an optional step S0310 of adjusting or "resetting" the thresholds Tmin, Tmax of admissibility of the intervals RRi. This step consists of redefining the thresholds Tmin, Tmax as a function of the increase or decrease of the subject's heart rate related to its activity, and involves detection of slow variations of the interval RRi. The processor then increments the loop variable i (i = i + 1) during a step S0311 and returns to step S0302 to wait for the occurrence of the next peak Ri.

When it appears in step S0303 that no peak Ri has been memorized, the processor stores the current peak Ri (Ti) during a step S0312 then goes to step S0311 to increment the index of loop before returning to step S0302. When it appears in step S0305 that the interval RRi is less than Tmin, the current peak Ri is considered as aberrant and the processor returns directly to step S0301 to wait for a new peak, without storing the current peak. When it appears in step S0307 that the interval RRi is greater than Tmax, the processor considers that one or more previous peaks have not been detected due to an interruption in the reception of the signal SI, S2 or non-peak detection. The current peak Ri is considered to be the first peak received after the presumed interruption and the processor goes to a step S0314 where it puts the interrupt flag IFR high and then erases the previous peak Ri-1 (T 1). l) during a step S0315. The processor then proceeds to step S0311 to increment the loop variable and returns to step S0302 to wait for a new peak to occur.

 Thus, the programs PG1, PG11, or PG12 provide RRi intervals devoid of outliers and accompanied by the IFR interrupt flag enabling the program PG2 to know, when this flag is in the high state, that the corresponding interval RRi is the first RR interval detected after a presumed interruption in the reception of the R peaks.

 FIG. 19 shows an example of an algorithm executed by the processor P3 during the step S05, under the control of the program PG2. The algorithm comprises a formation loop of a time series Sj which is initiated after each reception, at a step S0502, of an interval RRi, and is preceded by a step S0500 for initializing a loop variable j (j = 0) and an initialization step S0501 in the low state of an IFS interrupt flag distinct from the IFR interrupt flag.

The formation loop of a time series Sj comprises a step S0503 during which the processor checks whether the IFR interrupt flag associated with the interval RRi received is in the high state. If not, the processor adds the interval RRi to the series Sj during a step S0504, then goes to a step S0505 where it determines whether the cumulative duration of the intervals RRi of the series is greater than or equal to a threshold Te corresponding to the minimum duration of the time series previously mentioned. If not, the processor returns to step S0502 to wait for a new interval RRi. If the cumulative duration of the intervals RRi is greater than or equal to the threshold Te, the processor goes to a step S0506 where it provides the series Sj to the characterization program PG3 with the IFS interrupt flag in the low or high state according to the previously executed steps. The processor then inclines the loop variable j during a step S0507 (j = j + 1) then goes to a step S0508 where it initializes a new series Sj by deleting the first Np intervals RRi from the previous series to form the new series Sj. A sliding window of time series formation is thus defined, each time series comprising RRi intervals present in the previous time series. The The processor then lowers the IFS interrupt flag during a step S0509, then returns to step S0502 to wait for a new interval RRi.

 When the IFR interrupt flag is found in the high state at step S0503, the processor goes to a step S0510 where it increments the loop variable j (j = j + 1) and then goes to a step S0511 where it initializes a new empty series Sj marked with the IFS interrupt flag in the high state, then adds the current RRi interval to the new series Sj during a step S0512 and returns to the step S0502 to wait for a new interval RRi . Thus, the new time series formed from step S0511 is accompanied by the flag IFS in the high state indicating that the time series Sj is formed after a presumed interruption in the reception of signal SI, S2 or a non-signal. peak detection. This series will be supplied to the program PG3 at step S0506 after having received a number of intervals RRi sufficient for the time Te to be reached, if no other interruption intervenes in the meantime.

The flag IFS is used during the decision step SU executed by the processor P3 under the control of the program PG5, and makes it possible to know if time series provided by the program PG2 are successive or not. In an embodiment shown in FIG. 20, the decision step SU comprises a step S11 in which the processor waits to receive a classification of a series Sj provided by the classifying tool PG4 (or PG42, FIG. 17), this step being preceded by a step SI 10 of initializing a count variable "d" (d = 0). After receiving the result of the classification, the processor determines during a step SI 12 if the series Sj is attached to the class C2, namely whether it has a disturbance of the heart rhythm. If not, the processor returns to step S10 to reset count variable d, then goes to step S11 to wait for the classification of the next time series Sj. When it appears in step S12 that the series Sj is attached to the class C2, the processor goes to a step S13 where it checks whether the interrupt flag IFS is in the high state. If so, the processor sets the count variable "d" during a step SI 17 and then returns to step SI 11 to wait for the classification of the next time series Sj. If not, the processor goes to step SI 14 where it increments the count variable (d = d + 1) by 1, then goes to step S11 where it checks whether the count variable "d" is equal to a threshold D greater than 1. If the threshold D is not reached, the processor returns to step S 111 to wait for the classification of the next time series Sj. If the threshold D is reached, the processor goes to a step SI 16 where it indicates to the PAP application program that a rhythm disorder has been detected. The threshold D is determined so that the duration of a global observation window encompassing D series sequential time, which is less than the sum of the respective durations of the time series because they are generated according to a sliding time window, is sufficiently short depending on the degree of urgency possible of the management of the disturbance of the detected rhythm, and long enough for the rhythm disorder to be medically relevant. In one embodiment, D is 20, the duration of the time series is of the order of 10 seconds and the sliding window is regenerated with each heart beat. In another embodiment, a count of a time corresponding to the desired duration of the observation window is provided instead of counting the number of successive series.

 In a variant of the method, the processor, in step S12, determines whether the classification received consists of an attachment of the time series Sj to the class C 1 of the normal subjects. If the answer is positive, the processor returns to step SI 10, otherwise goes to step S 113.

 As indicated above, the device DV3 is initially placed in the "continuous standby" operating mode in which the processor has activated the channel CH1 and selected it by means of the multiplexer MUX, and the application program PAP switches it into the operating mode "alert" when the decision program PG5 indicates that a rhythm disorder has been detected in the SI signal. Figure 21 shows, purely illustrative and not limiting, an example configuration of the DV3 device in the "alert" mode. This configuration comprises an initial step S 120 during which the processor P3:

 - archive in the data memory M32 the previous SI signal that caused the switchover in the "alert" mode, as well as the corresponding time series Sj (Sl) and their classification by the classifier tool PG4,

- asks the user to put the finger on the electrode E2,

 - activates the channel CH2 and selects it via the multiplexer MUX, or activates the two channels CH1, CH2 and selects them alternately via the multiplexer MUX,

 - analyzes the signal S2 and optionally the signal S 1,

 if the channel CH1 is active, stores in the memory M32 the signal SI received, the corresponding time series Sj (Sl) and their classification by the classifier tool PG4,

in the memory M32 stores the signal S2, the corresponding time series Sj (S2) and their classification by the classifier tool PG4.

During a step S 121, the processor waits for a confirmation of a detection, in the signal S2, of the rhythm disorder that caused the switchover to the "alert" mode. During a step S 122, it verifies that a time Ta has elapsed since the switchover in the "alert" mode did not reach a Tmax threshold. If the threshold Tmax is reached without a disturbance of the rhythm has been found in the signal S2, or without the signal S2 has been received (if the user has not responded to the request that him has been addressed to touch the electrode E2), the processor returns to the operating mode "continuous standby", to possibly reboot a few moments later in the "alert" mode if the disorder of the rhythm is again detected in the signal S 1.

 If, on the contrary, the rhythm disorder is confirmed by the analysis of the signal S2 before the time Tmax is reached, the processor goes to a step S 122 during which it attempts to connect to an SRV server and / or or at a workstation WS shown in Fig. 13, via the wireless communication interface CI3 and an NTW computer or telephone network. If the connection is established, the processor transmits an alert to the remote device and then forwards archived data to it. The user can be informed of the success of the data transfer, for example via an information display.

 In one variant, the step of connecting to an SRV server and / or to a WS workstation is initiated as soon as the switchover to the "alert" mode, and the aforementioned data are transferred to the remote device without waiting for the confirmation provided for in FIG. Step S 121. The device DV3 can then return to the operating mode "continuous standby" or continue to manage the mode of operation "alert" in any way conceivable by those skilled in the art, for example taking into account the number times when the rhythm disorder has been detected, the user's behavior in the face of electrocardiographic signal acquisition requests, etc.

 Also, the user can be offered the possibility to ask the DV3 device to leave the "alert" mode until further notice, if a contact has already been made with the medical profession, or to switch to a mode of operation " silent alert "where the device DV3, while remaining connected to the remote device and / or while continuing to record data from the signal SI, no longer solicits the user to capture the electrocardiographic signal. In the variant of the device DV3 equipped with skin electrodes, the signal S2 can be captured and stored or transferred for the duration of the "silent alert" mode. The user may also be offered the possibility of initiating a cardiac recording himself even if the device has not detected a rhythm disorder by the analysis of the PPG signal.

It will be apparent to those skilled in the art that the DV1, DV2 and DV3 devices just described are susceptible to various variants, embodiments and applications. Notably, although it has been indicated in relation to Figures 16 and 17 that the device DV3 detects a disturbance of the rhythm in the photoplethysmographic signal by characterization and classification of time series, the general functionalities of the device, in particular the prediction of a mode of operation "continuous watch" where the photoplethysmographic signal is used for the monitoring of the rhythm cardiac, and an "alert" mode of operation where the electrocardiographic signal is captured with or without the contribution of the user, are independent of the method implemented to detect a rhythm disorder in each of these signals.

References cited:

Ref. 1: M.G. Tsipouras, D. I. Fotiadis, and D. Sideris, "An Arrhythmia Classification System Based on the RR-Interval Signal," Artificial Intelligence in Medicine, vol. 33, pp. 237-250, 2005.

Ref. 2: "Automatic Detection of Atrial Fibrillation Using RR Interval Signal" Xiuhua Ruan, Liu Changchun, Chengyu Liu, Xinpei Wang, Li Peng, School of Science Control and Engineering, Shandong University, Jinan, Shandong Province, PR China, 250061, 2011, 4th International Conference on Biomedical Engineering and Informatics (BMEI).

Ref. 3: "Heart rate variability, Standards of measurement, physiological interpretation, and clinical use". Task Force of the European Society of Cardiology and The North American

Society of Pacing and Electrophysiology. European Heart Journal (1996) 17, 354-381.

Annex 1 (forming part of the description)

Table 1 (correlation between series and derived series)

 Table 2

ANNEX 2

 (part of the description)

 Math Operators

Analysis Operators Formulas

 spectral (see Ref 3)

ULF (ULF Ultra Low Frequency = psd _x (f) .df

 Power) or ultra power

Where: psd _x is the spectral power density of the low frequency [0 - 3 mHz]

series of resampled X elements

mHz - 400 mHz] P _tot - ULF - VLF SD2 or scatter points

SD2 = ^ 2a¾ - SD l ² along the major axis of the ellipse

SD1

 Ratio SD1 / SD2 Ratio = - - SD2

N-

 VAI or Angle Index

 ï = l

Vector (Vector Angular Index)

 VLI or Length Index from Where:

Vector (Vector Length Index)

RMSSD or Medium

quadratic differences RMSSD =

successive elements X

SDSDD or standard deviation of the

 SDSDD

absolute value of the

2nd order differentiation of Where:

X elements; = M

Analysis Operators

Recurrence diagram RP Formulas

(Recurrence Plot)

JV-m + l

Recurrence rate (Recurrence

 REC = - - ^ Y RPÇj. ) Rate) j, = l

Or :

¹¹ -ini .n ^ Ι '

Analysis operators of the

 formulas

 regularity of the series

Entropy of Shannon "= - / _P Wln ( _P (*)) dx

 Or :

p (x) is the probability density of elements X

ApEn {m, r, N) = 0 ^m (r) - (p ^{m + 1} (r)

 Or:

4> ^m (r)

 N- m

 Entropy ApEn (Approximate J = l

 Entropy)

nbr {it _fe | d (ii _j , u _fe ) <r}

Nm + ld (u-, u _ft ) = m {| _{+ n} - ¾ _{+ n} || n = 0, ..., m-1]

SampEn (m, r, N) = ln (C ^m (r / C ^{m + 1} r))

Or :

 N-m + l

Entropy SampEn (Sample C ^m r)

 N- m

Entropy) en nr { _fe | d (ity, ii _k ) <r}

 (R)

 N- m

With:

 X: constituent element of the series considered,

u = (Xi_, ¾) ■■■)

 N = number (Xi): length of the vector u,

hist _w (u): histogram of 10 sections of the vector u,

Uj = (X _j , X _{j + 1} , ..., X _{j + m} _ ₁ ), j = 1, 2,.,., Ν -m + 1,

m: dimension of subdivision of the vector u, fixed at 2,

r: Euclidean distance tolerated between 2 points (Xi, Xi ₊ i) and (X _j , X _{j +} i), i and j G [1, N], E: expectation.

Annex 3

 (part of the description)

Primitive series:

 Sj: RR interval

 RRi = T (¾) - Τ (¾_

 Sjo: variation Vi of the RR interval:

Vi = dRR _; = RR _{i + 1} - RR _j

 Sji: absolute value | Vi | of the Vi variation of the RR interval:

Derived series:

Sj ^' 2: rate of change (Ai) of RR interval (heart rate acceleration):

Sj ^' 3: absolute value (| Ai |) of the variation rate (Ai) of the interval RR (acceleration in absolute value of the cardiac rhythm): Sj ^' 4: rate of change (ACi) of the rate of change (Ai) of the RR interval (pulse rate):

Sjs: absolute value (| ACi |) of the rate of change (ACi) of the variation rate (Ai) of the RR interval (jerk in absolute value of the cardiac rhythm):

Sjô: rate of change (Gi) of the absolute value (| Ai |) of the variation rate (Ai) of the RR interval:

dt) i + ll

 RRi + i RRi + i

Sj ^' 7: absolute value (| Gi |) of the variation rate (Gi) of the absolute value (| Ai |) of the variation rate (Ai) of the interval RR:

Claims

claims

A method for detecting at least one cardiac rhythm disorder of a human or animal subject from a time series of RR intervals representative of the subject's cardiac rhythm extracted from a physiological signal (S, S1, S2) ), comprising the steps of:

- characterizing (S07, S07 (I), S07 (2)) the time series with at least one descriptive variable (Vb, Vc) whose value is calculated from the intervals RR of the time series, - by means of the value of the descriptive variable, classifying the time series into a first class (Cl) of time series associated with subjects in sinus rhythm or in at least a second class (C2) of time series associated with subjects presenting the cardiac rhythm disorder, characterized in that the value of the descriptive variable (Vb) is calculated from a derived series (Sj _2, SJ3, Sj ₄ Sjs, SJO, Sj _7), the component is a discrete derived from order 1 or greater than 1 of the RR interval of the time series.

The method of claim 1, wherein the constituent element of the derived series is selected from the group consisting of: - the rate of change (Ai) of the RR interval, ie the discrete time derivative of the RR interval, reflecting the acceleration of the heart rate,

the absolute value (| Ai |) of the rate of variation of the RR interval,

the variation rate (ACi) of the variation rate (Ai) of the interval RR, that is the discrete time derivative of the rate of variation of the interval RR, or again the discrete second time derivative of the interval RR, reflecting the jolts of the heart rhythm,

the absolute value (| ACi |) of the variation rate (ACi) of the variation rate (Ai) of the interval RR, the variation rate (Gi) of the absolute value (| Ai |) of the rate of variation of the interval RR, and

the absolute value (| Gi |) of the rate of variation (Gi) of the absolute value (| Ai |) of the rate of variation of the interval RR.

3. Method according to one of claims 1 and 2, wherein the step of characterizing the time series also comprises calculating the value of at least one descriptive variable (Vc) from a time series of which the constituent element is selected from the group consisting of: - the RR interval,

the variation (Vi) of the RR interval, and

the absolute value (| Vi |) of the variation of the interval RR.

4. Method according to one of claims 1 to 3, wherein the value of the descriptive variable is calculated by means of a mathematical operator selected from the group consisting of: the average value, the median, the standard deviation, the variance, Skewness coefficient, Kurtosis coefficient, ULF ultra low frequency power, VLF very low frequency power, LF low frequency power, RF high frequency power, LF ratio / HF, the total power, the normalized LF power, the normalized HF power, SDl or dispersion of the points along the minor axis of the ellipse of the Poincaré diagram, SD2 or dispersion of the points along the long axis of the ellipse of the diagram of Poincaré, the ratio SD1 / SD2, VAI or the vector angle index, VLI or the vector length index, RMSSD or root mean square of the successive differences of the constituent elements of the series, SDSDD or standard deviation of the abs value olue of the differentiation of order 2 of the constituent elements of the series, the recurrence rate of the recurrence diagram, Lmax or maximum length of the diagonal lines of the recurrence diagram, Lmean or average length of the diagonal lines of the recurrence diagram, determinism , Shannon's entropy of the diagonal length distribution of the matrix of the recurrence diagram, the Shannon entropy, the ApEn entropy, the SampEn entropy, the correlation dimension.

5. Method according to one of claims 1 to 4, wherein the step of classification of the time series is performed by means of a classifier tool (PG4) supervised learning, and comprising a step (S0900) configuration of the classifier tool comprising the steps of:

- providing (S0903) to the classifier tool a plurality of values (Vbj) of the descriptive variable (Vb) calculated from series derived from a first set (NI) of time series from subjects in sinus rhythm, and

- providing (S0906) to the classifier tool a plurality of values (Vbj) of the descriptive variable (Vb) computed from series derived from a second set (N2) of time series from subjects with the cardiac rhythm disorder .

6. Method according to one of claims 1 to 5, comprising a step (S05) of continuous or quasi-continuous production of a plurality of time series from the physiological signal (S, S1, S2), and the determination of real time of the class (C1, C2) to which each time series belongs.

The method of claim 6, wherein the time series are produced (S05, S0 08) according to a sliding time window, so that two successive time series may comprise common RR intervals.

Method according to one of claims 6 and 7, comprising a step (S0307) for detecting missing RR intervals and in which the production of the time series is interrupted (S0315, IFS) during periods including missing RR intervals.

9. Method according to one of claims 6 to 8, wherein the time series are extracted from a photoplethysmographic signal (SI).

10. Method according to one of claims 6 to 9, comprising a decision step (SU) by which the cardiac rhythm disorder is considered to be detected when a determined number (D) greater than 1 of time series does not belong to the first class of time series.

11. Device (DV1, DV2, DV3) for detecting at least one cardiac rhythm disorder of a human or animal subject from a time series of RR intervals representative of the subject's cardiac rhythm extracted from a signal physiological (S, SI, S2), the device being configured to: - characterize (S07, S07 (1), S07 (2)) the time series with at least one descriptive variable (Vb, Vc) whose value is calculated at from the RR intervals of the time series,

- using the value of the descriptive variable, classify the time series into a first class (Cl) of time series associated with subjects in sinus rhythm or in at least one second class (C2) of time series associated with subjects presenting the cardiac rhythm disorder, characterized in that it is configured to calculate the value of the descriptive variable (Vb) from a derived series (Sj ₂ , Sj ₃ , Sj ₄ , Sjs, Sje, Sj ₇ ) whose constituent element is a derivative of order 1 or greater than 1 of the RR interval of the time series.

12. Device according to claim 11, configured to calculate the value of the descriptive variable (Vb) from a derived series whose constituent element is chosen from the group comprising: the rate of change (Ai) of the RR interval, the discrete time derivative of the RR interval, reflecting the acceleration of the heart rate,

the absolute value (| Ai |) of the rate of variation of the RR interval,

the variation rate (ACi) of the variation rate (Ai) of the interval RR, that is the discrete time derivative of the rate of variation of the interval RR, or again the second derivative of the interval RR, reflecting the heartbeat, the absolute value (| ACi |) of the variation rate (ACi) of the variation rate (Ai) of the interval RR,

the variation rate (Gi) of the absolute value (| Ai |) of the variation rate of the interval RR, and - the absolute value (| Gi |) of the variation rate (Gi) of the absolute value (| Ai |) of the rate of change of the RR interval.

13. Device according to one of claims 11 and 12, configured to also calculate the value of at least one descriptive variable (Vc) from a time series whose constituent element is selected from the group comprising:

- the RR interval,

the variation (Vi) of the RR interval, and

the absolute value (| Vi |) of the variation of the interval RR.

Apparatus according to one of claims 11 to 13, comprising a supervised learning classifier tool (PG4) configured to classify the time series into the first or second classes of time series.

15. Device according to one of claims 11 to 14, configured to produce (S05) continuously or quasi-continuously a plurality of time series from the physiological signal (S, SI, S2), and classify in real time each time series.

Apparatus according to claim 15, configured to produce (S05, S0508) the time series in a sliding time window, so that two successive time series may comprise common RR intervals.

Device according to one of claims 15 and 16, configured to detect (S0307) missing RR intervals and interrupt (S0315, IFS) the generation of time series of RR intervals during periods including missing RR intervals.

18. Device according to one of claims 15 to 17, configured to extract time series of RR intervals of a photoplethysmographic signal (SI).

19. Device according to one of claims 1 to 18, configured to perform a decision step (SU) by which the cardiac rhythm disorder is considered to be detected when a determined number (D) greater than 1 of successive time series n does not belong to the first class of time series.

20. Device (DV3) according to one of claims 11 to 19, configured to activate a channel (CH2) acquisition of the electrocardiogram of the subject when a disorder of the heart rhythm is considered to be detected.

21. Device according to one of claims 11 to 20, configured to detect at least one atrial fibrillation.

22. Device (DV3) according to one of claims 11 to 21, configured to be fixed around the wrist.