METHOD AND DEVICE FOR CARDIAC MONITORING

Abstract

A method for early identification of the presence of coronary heart disease or an arrhythmia in a patient being examined, including the steps of: (i) non-invasive recording of EKG signals at the patient's heart when resting, (ii) filter processing of the recorded EKG signals, (iii) transferring the filtered EKG signals into orthogonalised measurement variables on the basis of vectorcardiography, and (iv) entering the orthogonalised and, in the case of incorrectly applied electrodes, corrected measurement variables into a system based on artificial intelligence in which known findings data from comparative patients are stored and, by comparing these entered orthogonalised measurement variables with the findings data of the comparative patients within the Al system, a diagnosis is obtained for the patient being examined.

Claims

1. A method for early detection of a presence of coronary heart disease (CHD) and/or cardiac arrhythmia (HRD) in a patient to be screened, the method comprising: (i) non-invasive recording of ECG signals at the heart (12) of the patient (11) in the resting state; (ii) filtering processing of the recorded ECG signals; (iii) converting the filtered ECG signals into orthogonalized measured values based on vectorcardiography; and (iv) inputting the orthogonalized measured values into a system based on artificial intelligence, in which already known findings data of reference patients is stored, wherein a diagnosis is made for the screened patient by comparing the entered orthogonalized measured values with the findings data of the reference patients within the Al system; wherein the Al system is trained prior to step (iv), wherein the Al system comprises at least one neural network, wherein, for training the Al system, a number of specific learning values is input therein, the number of specific learning values being between 10 and 30 or the number of specific learning values being 20, wherein the specific learning values are determined by the following sequence of steps: (v) providing measured values of a set (M) of patients with a known finding, wherein these measured values are orthogonalized based on vectorcardiography; (vi) providing a plurality of time series parameters and at least one statistic; (vii) forming a 3D matrix, wherein the orthogonalized measured values of the set of patients define the rows, the time series parameters define the columns and the at least one statistical method defines the depth of this matrix; (viii) classifying all pairs of values of the 3D matrix according to the principle of the Area-under-Curve (AUC) calculation; (ix) selecting a pair of values from the set in step (viii) with the highest AUC value; (x) checking another pair of values from the set in step (viii), and selecting this pair of values, if a limit value for a correlation with the value pair of step (ix) is smaller than 1.65/N, where N=number of the data points or parameter statistics (patients) in step (vi); (xi) repeating step (x) for another pair of values from the set in step (viii), and selecting this pair of values if a limit value for a correlation with the previously selected value pairs is in each case smaller than 1.65/N; and (xii) repeating the steps (ix) to (xi) until a predetermined number of value pairs is reached, which are then defined as specific learning values for training the Al system.

2. The method according to claim 1, wherein in step (i) the ECG signals are recorded at a total of four lead points on the body of the patient.

3. The method according to claim 2, wherein potential differentials are measured in the form of an anterior lead between a first lead point and a fourth lead point, a dorsal lead between a second lead point and the fourth lead point, a horizontal lead between a third lead point and the fourth lead point, a vertical lead between the first lead point and the third lead point, and an inferior lead between the first lead point and the second lead point.

4. The method according to claim 3, wherein the leads between the respective lead points are converted into spherical coordinates.

5. The method according to claim 2, wherein, for recording the ECG signals a t-shirt is used, which has four sensors assigned to a correct position of the four lead points on the body of the patient.

6. The method according to claim 1, wherein in step (v) the measured values of the set of patients are provided in the form of time series, preferably in milliseconds, or in the form of heartbeats.

7. The method according to claim 1, wherein in step (vi) a plurality of statistical methods (mean value, variance, kurtosis, skew, 5% quantile, 95% quantile) are provided.

8. The method according to claim 7, wherein, after step (vii), the standardized matrix is calculated from the data of the 3D matrix using a statistical method, thus achieving a uniform depth.

9. The method according to claim 1, wherein in step (viii) the AUC calculation is performed empirically or according to the principle of Johnson distribution.

10. A device for early detection of the presence of coronary heart disease and/or cardiac arrhythmia of a patient to be screened, the device comprising: a plurality of sensors positionable at predetermined lead points on the body of the patient so as to non-invasively record ECG signals at the heart of the patient in the patient's resting state; at least one filter with which the recorded ECG signals are filtered; an evaluation device via which the filtered ECG signals are converted into orthogonalized measured values on the basis of vectorcardiography; and a system based on artificial intelligence, in which already known findings data of reference patients is stored, wherein the orthogonalized measured values are entered into the Al system and compared therein with the findings data of the reference patients in order to establish a diagnosis for the patient being screened, wherein a total of four sensors are provided corresponding to a total of four lead points on the body of the patient, wherein a position of at least the sensor assigned to the first lead point is verifiable by the device and that the device is programmed such that a heart-related space vector is determined and displayed, wherein this space vector represents the electrical field vector formed by the activity of the heart, wherein measured values of the heart are recorded on the body at a first lead point, at a second lead point, at a third lead point and at a fourth lead point, wherein potential differences are measurable in the form of an anterior lead between the first lead point and the fourth lead point, a dorsal lead between the second lead point and the fourth lead point, a horizontal lead between the third lead point and the fourth lead point, a vertical lead between the first lead point and the third lead point, and an inferior lead between the first lead point and the second lead point, wherein an orthogonal system is be formed with the relationships:
x=D cos 45I
y=D sin 45+A
z=(VH) sin 45 wherein the measured values and the space vector determined therefrom are mapped in this orthogonal system (x, y, z), wherein the device is adapted to perform the following step sequence: (a) performing a measurement on a patient using the first to fourth lead points on the body of the patient in order to obtain a cardiogram for this patient, (b) extracting the amplitudes of the R wave from the cardiogram of step (a) for each heartbeat in the x, y and z direction, (c) determining mean values x, y, z and standard deviations x, y, z of the respective amplitudes recorded in millivolts from the cardiogram in step (b), wherein these mean values and standard deviations then form a calculation vector, (d) forming a coefficient matrix which is obtained based on a Principal Component Analysis by using different formats for measurements on reference patients, (e) multiplying the calculation vector of step (c) by the coefficient matrix of step (d) to form a resulting vector with a total of six main axes, wherein the coefficient matrix, with which in step (e) the calculation vector is multiplied to form the resulting vector, is a 66 coefficient matrix, (f) extracting the first main axis and the second main axis from the resulting vector of step (e) to form a reference point in the space of the first and second main axis, (g) determining a Euclidean distance of the reference point from a predetermined target point which corresponds to a correct position of the four lead points on the human body, and (h) if the distance of the reference point from the predetermined target point is greater than a predetermined maximum value: performing an angular correction for a first triangle formed by the first lead point, the third lead point and the fourth lead point, and for a second triangle formed by the first lead point, the second lead point and the fourth lead point so that thereby the Euclidean distance between the reference point and the predetermined target point is minimized by adapting the orthogonal system (x, y, z) to the changed geometry, wherein in step (h) the angular correction determines adjustment values with which the non-orthogonal angles of the first triangle and the non-orthogonal angles of the second triangle are corrected, so that at least an incorrect position of the first lead point on the human body is compensated.

11. The device according to claim 10, wherein the evaluation device is programmed in such a way that the orthogonalized measured values obtained with the filtered ECG signals on the basis of vectorcardiography are converted into spherical coordinates.

12. The device according to claim 10, wherein the sensors are integrated in a t-shirt, namely in areas of the t-shirt which are assigned to a correct position of the four lead points on the body of the patient.

13. The device according to claim 10, wherein the Al system is trained, preferably that the Al system is trained with a number of specific learning values which are determined on the basis of a set of patients with a known finding, and wherein the Al system comprises at least one neural network.

14. A method for determining and displaying a space vector related to the heart, which represents the electrical field vector formed by the activity of the heart, wherein measured values of the heart are recorded on the body at a first lead point, at a second lead point, at a third lead point and at a fourth lead point, wherein potential differences are measured in the form of an anterior lead between the first lead point and the fourth lead point, a dorsal lead between the second lead point and the fourth lead point, a horizontal lead between the third lead point and the fourth lead point, a vertical lead between the first lead point and the third lead point, and an inferior lead between the first lead point and the second lead point, wherein an orthogonal system is formed with the relationships:
x=D cos 45I
y=D sin 45+A
z=(VH) sin 45 and the measured values and the space vector determined therefrom are mapped in this orthogonal system, the method comprising: (a) performing a measurement on a patient using the first to fourth lead points on the body of the patient in order to obtain a cardiogram for this patient; (b) extracting the amplitudes of the R wave from the cardiogram of step (a) for each heartbeat in the x, y and z direction; (c) determining mean values x, y, z and standard deviations x, y, z of the respective amplitudes recorded in millivolts from the cardiogram in step (b), wherein these mean values and standard deviations then form a calculation vector; (d) forming a coefficient matrix which is obtained based on a Principal Component Analysis by using different formats for measurements on reference patients; (e) multiplying the calculation vector of step (c) by the coefficient matrix of step (d) to form a resulting vector with a total of six main axes, wherein the coefficient matrix, with which in step (e) the calculation vector is multiplied to form the resulting vector is a 66 coefficient matrix; (f) extracting the first main axis and the second main axis from the resulting vector of step (e) to form a reference point in the space of the first and second main axis; (g) determining a Euclidean distance of the reference point from a predetermined target point which corresponds to a correct position of the four lead points on the human body; and (h) if the distance of the reference point from the predetermined target point is greater than a predetermined maximum value: performing an angular correction for a first triangle formed by the first lead point, the third lead point and the fourth lead point, and for a second triangle formed by the first lead point, the second lead point and the fourth lead point so that thereby the Euclidean distance between the reference point and the predetermined target point is minimized by adapting the orthogonal system to the changed geometry, wherein in step (h) the angular correction determines adjustment values with which the non-orthogonal angles of the first triangle and the non-orthogonal angles of the second triangle are corrected, so that at least an incorrect position of the first lead point on the human body is compensated, or wherein in step (h) the adjustment values are determined for the angular correction by minimizing the Euclidean distance between reference and target values.

15. A heart monitoring method in which ECG signals are recorded at the heart and on the basis of which a space vector is determined using vectorcardiography, wherein this space vector represents the course of the sum vector of the electrical field of the heart and has a direction corresponding to the field direction and a length corresponding to the potential, wherein a quotient is formed from the areas covered by a length of the space vector (=radius vector) as a function of time during the R wave and during the T wave, respectively, wherein this quotient is then subjected to further evaluation taking into account at least one predetermined limit value, in order to verify the presence of coronary heart disease and/or cardiac arrhythmia for a patient, wherein coronary artery disease is detected for the screened heart if the quotient formed from the areas covered by the space vector during the R wave and during the T wave, respectively, $\frac{\mathrm{Area}\left(R-\mathrm{wave}\right)}{\mathrm{Area}\left(T-\mathrm{wave}\right)}=a$ is located outside a range defined by a lower limit value a.sub.0,CHD and an upper limit value a.sub.1,CHD, preferably, that the limit values a.sub.0,CHD and a.sub.1,CHD are determined as a function of a training set, wherein the lower and upper limit values a.sub.0 and a.sub.1 are respectively determined by a training set of test person data, comprising the steps: (i) calculating the quotient area (R wave)/area (T wave) for all test persons to obtain time series ratios; (ii) calculating a mean value and an associated standard deviation from the time series ratios of step (i); and (iii) determining the lower limit a.sub.0 taking into account the mean value and the standard deviation of step (ii), by the relationship: a.sub.0=mean value ()standard deviation (), and/or determining the upper limit al taking into account the mean value and the standard deviation of step (ii), by the relationship: a.sub.1=mean value ()+standard deviation (), wherein in step (ii) the mean value and the associated standard deviation are calculated only for time series ratios of healthy patients.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The present invention will become more fully understood from the detailed description given hereinbelow and the accompanying drawings which are given by way of illustration only, and thus, are not limitive of the present invention, and wherein:

[0059] FIG. 1 is a simplified schematic sketch of a device according to the invention for the early detection of CHD or HRD,

[0060] FIG. 2A is a view of a patient from the front and FIG. 2B is a view of a patient from the back,

[0061] FIG. 3A is a view of a t-shirt from the front and FIG. 3B is a view of a tshirt from the back, which t-shirt can be used by the device according to FIG. 1,

[0062] FIG. 4 shows a front view of a patient, to illustrate an orthogonal system (x, y, z) in which a space vector is mapped according to the present invention,

[0063] FIGS. 5A-5B are schematic simplified views of a first triangle (FIG. 5A) and a second triangle (FIG. 5B), based on which, in respect of the orthogonal system according to FIG. 4, an angular correction is carried out according to the invention,

[0064] FIG. 6 is a generally simplified view of a system architecture according to the invention, within which the device of FIG. 1 according to the invention is used,

[0065] FIG. 7A is a generally simplified representation of a space vector formed during activity of the human heart, FIG. 7B shows a front view of a patient, to illustrate an orthogonal system (x, y, z) in which a space vector is mapped according to the present invention,

[0066] FIG. 8 is a flow chart in which a method according to the present invention is carried out,

[0067] FIG. 9 is a flow chart in which a method according to the present invention is carried out according to a further embodiment,

[0068] FIGS. 10A-10B are a basically simplified representation of a 3D matrix (FIG. 10A) of measured values or a standardized version thereof (FIG. 10B), with the statistics derived from the time series parameter or parameters, which are used in a method according to FIG. 8 or FIG. 9,

[0069] FIGS. 11A-11D show exemplary time series parameters applicable to a method according to FIG. 9,

[0070] FIG. 12 shows exemplary statistical methods applicable to a method according to FIG. 9,

[0071] FIG. 13 is a flow chart in which a method according to the present invention is carried out,

[0072] FIG. 14 is a further flow chart in which a method according to the present invention is carried out according to a further embodiment,

[0073] FIG. 15 shows the multiplication of a calculation matrix with a calculation vector according to a further method of the present invention,

[0074] FIG. 16 is a basically simplified Principal Component Analysis for the method of FIG. 15,

[0075] FIG. 17 is a simplified illustration of the formation of a quotient from the areas covered by a length of the space vector (=radius vector) as a function of time during the R wave and during the T wave, respectively, and

[0076] FIG. 18 is a simplified illustration of a plurality of neural networks used in an application or implementation of the present invention.

DETAILED DESCRIPTION

[0077] In the following, with reference to FIGS. 1-18, preferred embodiments of the heart monitoring method according to the invention and of a device 10 used for this purpose are shown and explained in detail.

[0078] The device 10, shown in basically simplified form in FIG. 1, is used in the screening of a patient 11 (cf. FIGS. 2A-2B) for early detection of the presence of coronary heart disease (CHD) or cardiac arrhythmia (HRD). For this purpose, the device 10 comprises four sensors or electrodes S1, S2, S3, S4, a data filter 16, an evaluation device 18 and a system 20 on the basis of artificial intelligence (Al). The evaluation device 18 is equipped with a memory element (not shown) so that measured signals or values can be stored therein, at least for a short time. The Al system 20 may have at least one neural network 20.sub.N (cf. also FIG. 18) or a plurality of such neural networks or be formed from such neural networks.

[0079] The sensors S1-S4 are connected for signaling (e.g. via a cable connection, or via a wireless radio link) to the device 10 in such a way that its measured values first pass through the filter 16 and then reach the evaluation device 18. The evaluation device 18 is data-technically connected to the Al system 20 in such a way that the measured values, which are suitably processed by means of the evaluation device 18 or converted into orthogonalized data based on the vectorcardiography, as further specified below, can be entered into the Al system 20. This is done for the purpose of making a diagnosis for the patient 11 being screened using the device 10.

[0080] FIGS. 2A-2B showfor an improved understanding of the inventiona patient 11, namely in a front view (FIG. 2A) and in a rear view (FIG. 2B). On the body 14 of the patient 11, a total of four lead points are provided, namely a first lead point E1, a second lead point E2, a third lead point E3 and a fourth lead point E4. The lead points E1, E3 and E4 are each located in the chest region of the patient 11, wherein the lead point E2 is located on the patient's 11 back. In relation to these four lead points E1 to E4, it should be noted that the first sensor S1 is placed at the first lead point E1, the second sensor S2 is placed at the second lead point E2, the third sensor S3 is placed at the third lead point E3, and the fourth sensor S4 is placed at the fourth lead point E4 on the body 14 of the patient 11. Potential differences are measured between these lead points, as explained separately below. For further details on the positions of these individual lead points E1-E4 on the patient's body 14, reference is made to the disclosure according to EP 86 429 B1, the contents of which are hereby referred to in their entirety.

[0081] FIGS. 3A-3B show a t-shirt 22 in simplified form. The aforementioned sensors S1-S4 of the device 10 may be integrated into the t-shirt 22, for example by being woven into its textile structure. Such a t-shirt 22 has the advantage that a patient 11 merely puts on or slips over this t-shirt 22 in preparation for an examination, in which case the sensors S1-S4, which are integrated in the t-shirt 22, automatically reach their intended position adjacent to the four lead points E1-E4. The use of such a t-shirt 22 eliminates the need for time-consuming and possibly error-prone manual application of the individual sensors S1-S4 to the body 14 of the patient 11. As an alternative to the t-shirt 22, it is also possible to use a (not shown) chest strap to which the sensors or electrodes S1-S4 are attached.

[0082] To screen a patient 11 or to obtain a set of test data for training the Al system 20, the four sensors S1-S4 are positioned on the human body 14 at the assigned four lead points E1-E4. Subsequently, in the resting state of the patient 11, ECG signals are recorded at the heart 12 of the patient 11 with the aid of the sensors S1-S4 brought into position. The ECG signals are then passed through the filter 16 and subsequently converted in the evaluation unit 18 into orthogonalized measured values (in the axes x, y, z) according to the Sanz system per EP 86 429 B1.

[0083] FIG. 4 illustrates the axes x, y, z according to the Sanz system, in relation to the body 14 of a patient 11 and his heart 12. In this respect, reference is also made to FIGS. 5A-5B, wherein in FIG. 5A a first triangle 31 is schematically illustrated, which is located between the lead points E1, E3 and E4, and FIG. 5B illustrates a second triangle 32, which is selected between the lead points E1, E2 and E4. The meaning of these two triangles 31, 32 will be explained separately below.

[0084] As already explained, potential differences are recorded between the individual lead points E1-E4. In detail, these are an anterior lead A between the first lead point E1 and the fourth lead point E4, a dorsal lead D between the second lead point E2 and the fourth lead point E4, a horizontal lead H between the third lead point E3 and the fourth lead point E4, a vertical lead V between the first lead point E1 and the third lead point E3, and finally an inferior lead I between the first lead point E1 and the second lead point E2. The first lead point E1, the third lead point E3 and the fourth lead point E4as shown in FIG. 5Aform a first triangle 31, wherein a second triangle 32as shown in FIG. 5Bis formed from the first lead point E1, the second lead point E2 and the fourth lead point E4. The mentioned leads with their designations A, D, H, V and I are also shown in FIG. 5A and FIG. 5B. For further correlations on these leads, reference is made to the content of EP 86 429 B1.

[0085] The meaning of the first and second triangles 31, 32 is explained separately elsewhere below in connection with a so-called correction method according to the present invention.

[0086] When an ECG measurement is performed, the electrical measured values of the leads A, D, H, V and I mentioned above enter the device 10 and are further processed therein accordingly, as already explained above.

[0087] With reference to FIG. 6, further details of the device 10 and its integration into an overall architecture in accordance with the invention for carrying out the present invention are explained below. In detail:

[0088] The architecture according to FIG. 6 provides the following components: sensing device 100, data recorder 102, Cardisio device 106, and server 113. The data recorder 102 includes a signal receiver 103, a signal converter 104, and a signal memory 105. The Cardisio device 106 includes a signal reader 107, a vector data generator 108, a vector data memory 109, a vector data evaluator 110, a vector data synchronizer 111 and a vector data display 112. The server includes 113 a vector data memory 114, a vector data analyzer 115, and a vector data evaluation memory 116.

[0089] The sensing device 100 is made up by the four sensors or electrodes S1-S4 of the device 10 mentioned above.

[0090] The sensing device 100 and the data recorder 102 are the components or parts of the device 10 that are used to measure or record the ECG signals. Hereby, the analog ECG signals are received, processed and suitably converted into digital signals. As already explained, the digital signals can be stored at least briefly in the memory element of the evaluation device 18in this case in the form of the signal memory 105.

[0091] The Cardisio device 106 reads the digital signals from the data recorder 102 using the vector data generator 108 to determine vector data based on the digital signals. The vector data generated in this way is then stored in the vector data memory 109. Based on this, the vector data evaluator 110 generates a representation of this vector data, e.g. in the form of a three-dimensional curve, wherein this representation is then shown or visualized by means of the vector data display 112.

[0092] The architecture of FIG. 6 further illustrates that within the Cardisio device 106, the vector data evaluator 110 is signal-connected to the vector data synchronizer 111, and the latter is connected to the server 113 via a signal path or link. Hereby, the vector data generated can first be read into the vector data memory 114 on the server 113 via the vector data synchronizer 111. Subsequently, by means of the vector data analyzer 115, it is possible to carry out a targeted evaluation of large data volumes or the generated vector data, and to perform statistical analyses in this respect. Finally, the manual and/or automatic evaluations of the data sets are stored in the vector data evaluation memory 116, from which these evaluations can be uploaded back to the device 106, e.g. for display on or with the vector data display 112.

[0093] It is understood that the components and parts of the sensing device 100, the data recorder 102, the Cardisio device 106 and the server 113 explained in FIG. 6 are each parts of the device 10 according to the invention. In particular, the vector data evaluation memory 116 is part of the Al system 20 or of a neural network 20N, wherein already known findings data of reference patients is stored herein, in respect of which a clear diagnosis (healthy or sick) is known.

[0094] Based on the ECG measured data recorded using the four sensors S1-S4 at the heart 12 of a patient 11, a space vector 24, which shows the electrical activity of the heart 12, can be generated by the evaluation device 18. Specifically, this space vector 24 forms the course of the sum sector of the electrical field of the heart 12 and has a direction corresponding to the field direction and a length corresponding to the potential. An example of such a space vector 24 is shown in FIG. 7A, which is preferably mapped in an orthogonal system, which is formed from the axes x, y, z according to the Sanz system (per EP 86 429 B1). FIG. 7B again showsin the same way as FIG. 4in a simplified manner the heart 12 of a patient 11, in connection with the axes x, y, z according to the Sanz system.

[0095] A method according to the present invention is explained below with reference to FIG. 8, which shows a flow chart of the steps of such a method. In detail:

[0096] At the beginning of the method, the sensors S1-S4 of the device 10 described above (cf. FIG. 1) are placed on the upper body of a patient 11, namely in accordance with the four lead points E1, E2, E3 and E4 (cf. FIG. 2A, FIG. 2B). The t-shirt 22 of FIGS. 3A and 3B can be used for this purpose. In this way, ECG signals are recorded in a non-invasive manner at the patient's 11 heart 12, namely in the resting state of the patient 11. This corresponds to step (i) of the method according to FIG. 8.

[0097] Subsequently, in step (ii) of the method shown in FIG. 8, the recorded ECG signals are filtered, namely, as explained with reference to FIG. 1, by the filter 16. Such filtering serves to eliminate high-frequency noise and low-frequency interference (e.g. caused by the patient's breathing). Examples of filter types are notch filters, high pass filters, Savitzky-Golay low-pass filters.

[0098] In the subsequent step (iii) of the method shown in FIG. 8, the filtered ECG signals are then converted into orthogonalized measured values on the basis of vectorcardiography by means of the evaluation device 18, which is signal-connected to the filter 16. Significant areas in the signal of a heartbeat are localized, e.g. beginning and end of the QRS complex, beginning, maximum and end of the T wave. To determine the actual times of the physiological extreme points, it is advantageous when said leads A, D, H, I and V obtained between the lead points E1-E4 (cf. FIG. 2a, FIG. 2b) are each converted into spherical coordinates.

[0099] Finally, in a further step (iv) of the method according to FIG. 8, the orthogonalized measured values are entered into the Al system 20 or a neural network 20.sub.N. It should be noted that the Al system 20 is based on already known findings data of reference patients for whomas already explained abovethere is clear knowledge with regard to their state of health (healthy or sick). On the basis of this, it is then possible, by means of the vector data analyzer 115 and the vector data evaluation memory 116, to provide a diagnosis for the screened patient 11 by comparing the entered orthogonalized measured values with the findings data of the reference patients within the Al system 20.

[0100] For step (i) of the method of FIG. 8, it is recommended that a non-invasive recording of ECG signals be performed at the heart 12 of the patient 11 at exactly the four lead points E1-E4 described above with reference to FIG. 2A and FIG. 2B.

[0101] By means of an advantageous further development or supplementation of the method of FIG. 8, it is possible to train the Al system 20 or the neural network 20N prior to step (iv), namely by inputting specific learning values f. This is shown in FIG. 9 and explained in detail below:

[0102] The number of specific learning values f with which the Al system 20 or the neural network 20.sub.N is trained prior to step (iv) can be between 10 and 30, and e.g. assume the value of 20. These specific learning values f are determined by the following step sequence:

(v) Providing measured values of a set M (cf. FIGS. 10A-10B) of patients 11 with a known finding, wherein these measured values are orthogonalized on the basis of vectorcardiography,
(vi) Providing a plurality of time series parameters (cf. FIG. 11A and FIG. 11B) and at least one statistical method (cf. FIG. 12),
(vii) Forming a 3D matrix 25 (cf. FIG. 10A), wherein the orthogonalized measured values of the set (M) of patients define the rows, the time series parameters define the columns and the time series length defines the depth of this matrix (25), wherein in the case of scalar parameters, the depth is equal to one,
(viii) Classifying all pairs of values of the 3D matrix (25) according to the principle of theArea-under-Curve (AUC) calculation,
(ix) Selecting a pair of values from the set according to step (viii) with the highest AUC value,
(x) Checking another pair of values from the set in step (viii), and selecting this pair of values if a limit value for a correlation with the pair of values in step (ix) is smaller than 1.65/N, where N=number of data points or parameter statistics (patients) according to step (vi)
(xi) Repeating step (x) for another pair of values from the set in step (viii), and selecting this pair of values if a limit value for a correlation with the previously selected value pairs is smaller than 1.65N in each case, and (xii) Repeating steps (ix) to (xi) until a predetermined number of e.g. 20 value pairs is reached, which are then defined as specific learning values f and are entered into the Al system (20) for purposes of training.

[0103] The above-mentioned steps (v) to (xii) of the further development of the method according to FIG. 9 are each symbolized in simplified form by blocks in the associated flow chart. Here, the arrow f, which after step (xii) opens out from below into the sequence between steps (iii) and (iv), is understood to mean that the predetermined number of specific learning values f defined in step (xii) are input to the Al system 0 or the neural network 20.sub.N.

[0104] With regard to the advantageous further development of the method according to FIG. 9, it may be pointed out by way of explanation that in step (v) the measured values of the set M of patients 11 are provided in the form of time series, preferably in milliseconds, or in the form of heartbeats. The formation of the 3D matrix 25 in step (vii) is shown symbolically in FIG. 10A. The set M of all patients is plotted as an ordinate and can be divided, e.g. seen from top to bottom, first into a group of healthy patients (e.g. without CHD findings) and then into a group of sick patients (e.g. with CHD findings). The 3D matrix includes the set M of the total 284 time series as well as the scalar personal parameters, which are shown in FIGS. 11A-11D. However, only the 282 time series parameters are included in the training of the neural network.

[0105] With respect to the 292 parameters shown in FIGS. 11A, 11B, 11C, and 11D, it is separately noted that there are 284 time series parameters. 8 personal scalar parameters, which refer to a respective screened patient, are considered for the probability calculations using Bayes theorem.

[0106] At this point, it is separately noted that in the above method, in steps (x) and (xi) of which, the individual limit values for the respective correlations with the pair of values of step (ix) do not assume a constant fixed value, but instead are each dependent on the number of heartbeats or the majority of time series parameters in step (vi). Thus, higher correlations are allowed for short time series (and thus smaller values of N), and vice versa.

[0107] FIG. 10B shows the 3D matrix in a standardized version with uniform depth, which is achieved by linking or offsetting their columns (=patient data) and rows (=time series parameters) with a plurality of statistical methods, six of which are shown as examples in FIG. 12, namely:

[0108] mean value

[0109] variance

[0110] kurtosis,

[0111] skew,

[0112] 5% quantile,

[0113] 95% quantile.

[0114] The possible application of these six methods is indicated by the entry of 6 (in the image area on the right) in the illustration of FIG. 10B. In this connection it is pointed out that in step (vii) at least one of these statistical methods, or even several such methods, can be used to define both the depth of the 3D matrix 25 and, if necessary, to achieve a uniform depth for the realization of a standardized matrix 25.sub.N.

[0115] Furthermore, for the method shown in FIG. 9, it may be pointed out by way of clarification that in step (viii) the AUC calculation can be performed empirically or according to the principle of Johnson distribution. The principle of Johnson distribution is known per se state of the art, but could be used for the first time with the present invention in connection with the evaluation of patient data or ECG signals for the purpose of early detection of the presence of CHD and/or HRD.

[0116] As already explained, for evaluating the data for the purpose of creating a diagnosis, it is advantageous if the Al system 20, into which the specific learning values f are input, comprises at least one neural network 20.sub.N or a plurality of such networks 20.sub.N.

[0117] By means of the present invention it is possible, as explained, on the one hand to perform a method for early detection of the presence of CHD and/or HRD of a patient 11 to be screened, as is shown and explained in the flow chart of FIG. 8. On the other hand, in order to improve the diagnosis of the patient 11, it is possible to subject the Al system 20 (or a neural network 20.sub.N) to training before the actual measurement, as shown and explained in the flow chart of FIG. 9. Such training is carried out on the basis of the data of such patients of whom there is a clear knowledge with regard to their state of health (healthy or sick). In this respect, such training in the sense of the previous invention is also called supervised learning.

[0118] These two possibilities, namely both an upstream training for the Al system 20 (or a neural network 20.sub.N) and the actual performance of the measurement of a patient 11 for the purpose of creating a desired diagnosis, are shown again below in the flow charts of FIG. 13 and FIG. 14. The flow chart of FIG. 13 shows with its blocks 13.1-13.5 a step sequence for the purpose of training the Al system 20. Such training serves to optimize the actual examination or diagnosis of a patient 11, which is carried out by the step sequence of the flow chart in FIG. 14.

[0119] In the flow chart of FIG. 13, step 13.1 is substantially the same as step (i) of FIG. 8, with step 13.2 being substantially the same as step (ii) of FIG. 8. The same applies to step 13.3, which essentially corresponds to step (iii) of FIG. 8. To avoid repetition, reference may therefore be made to the explanations of FIG. 8 for these steps 13.1, 13.2 and 13.3.

[0120] The following step 13.4 in the flow chart of FIG. 13 corresponds to a parameter extraction, in whichcorresponding to steps (v) to (vii) of FIG. 9time series parameters are combined or offset with the measured values of a set M of patients 11. Here too, it proves to be an advantage if this is done on the basis of spherical coordinates into which the leads A, D, H, I and V are suitably converted.

[0121] The following step 13.5 aims at feature evaluation and corresponds essentially to a sequence of steps (viii) to (xii) of FIG. 9, which has already been mentioned and explained above. This means that the specific learning values f are determined or identified using this feature evaluation according to step 13.5.

[0122] Subsequently, in the flow chart of FIG. 13, namely in step IV.sub.T, the specific learning values f are input into the Al system 20 (or to a neural network 20.sub.N) in accordance with step (iv) of FIG. 8. As a result, the Al system 20 is suitably trained by the input of the specific learning values f. At this point it should be pointed out again that it is of great advantage in the context of this training that the number of specific learning values f is relatively low and can assume the value 20, for example. Alternatively, the number of specific learning values f may be fewer or greater than 20, and may be, for example, 15 or 25. In any case, with respect to the training of the Al system 20 discussed herein, it is to be understood that these learning values f are always obtained from the data of a patient 11 for whom there is a clear knowledge regarding their state of health (healthy or sick).

[0123] The flow chart of FIG. 14 illustrates with its sequence of steps the diagnosis actually carried out for a patient 11. Here, steps 14.1, 14.2 and 14.3 essentially correspond to steps 13.1-13.3, so that, in order to avoid repetition, reference can be made to the explanation of steps 13.1-13.3.

[0124] Step 14.5 in the flow chart of FIG. 14 provides a feature selectionhere, only the data of a screened patient 11 are used on the basis of those recorded ECG signals that correspond to the specific learning values f previously determined in step 13.5. These selected features or values are then compared with the predetermined specific learning values fin step 14.6 (trained network). Subsequently, in step IV.sub.D, the actual diagnosis IVD is made for the patient 11 with the help of the trained Al system 20, in accordance with step (iv) of FIG. 8.

[0125] The diagnosis of a patient 11 explained above, performed with a device 10 of FIG. 1 and by a method according to the flow chart of FIG. 8 or FIG. 14, as well as the training of an Al system 20 (or a neural network 20.sub.N) which is carried out according to the flow chart of FIG. 9 and FIG. 13, respectively, are always based on the fact that the sensors S1-S4 are place on the body of the patient 11, e.g. by using the t-shirt of FIG. 3. Now, according to a further method according to the present invention, it is possible to compensate for a possibly incorrect fit of these sensors S1-S4 on the body of the patient 11, so that a realistic diagnosis for the patient 11 continues to be ensured. Such a method, hereinafter briefly referred to as correction method, first provides that a space vector 24 (cf. FIG. 7a) related to the heart 12 of the patient 11, which represents the vector of the electrical field generated by the activity of the heart 12, is determined and suitably displayed. Here, measured values of the heart 12 are recorded on the patient's body 14 at a total of four lead points E1-E4, as has already been explained above in connection with FIG. 2. In the case of the aforementioned correction method, based on the leads A, D, H, I and V (cf. FIG. 5A, FIG. 5B), an orthogonal system with the relationships

x=D cos 45I

y=D sin 45+A

z=(VH) sin 45

[0126] is formed, wherein the measured values for the patient 11 and the space vector 24 determined therefrom are mapped to this orthogonal system x, y, z.

[0127] Thus, with reference to the illustrations in FIG. 5A, FIG. 5B, FIG. 15 and FIG. 16, the correction method then in particular provides the following steps:

(a) Performing a measurement on a patient using the first to fourth lead points E1-E4 on the patient's body 14 to thereby obtain a cardiogram for that patient,
(b) Extracting the amplitudes of the R wave from the cardiogram of step (a) for each heartbeat in the x, y and z direction,
(c) Determining mean values x, y, z and standard deviations x, y, z of the respective amplitudes recorded in millivolts from the cardiogram in step (b), wherein a calculation vector 26 is then formed with these mean values x, y, z and standard deviations x, y, z,
(d) Forming a matrix of coefficients 28 obtained on the basis of a Principal Component Analysis for measurements in reference patients by using different formats,
(e) Multiplying the calculation vector 26 of step (c) by the coefficient matrix 28 of step (d) to form a resulting vector 30 with a total of six main axes PC.sub.1-PC.sub.6,
(f) Extracting the first main axis PC.sub.1 and the second main axis PC.sub.2 from the resulting vector 30 of step (e) to form a reference point PC.sub.1, PC.sub.2 in the space of the first and second main axis,
(g) Determining a Euclidean distance of the reference point PC.sub.1, PC.sub.2 from a predetermined target point PC.sub.1.sup.fit, PC.sub.2.sup.fit, which corresponds to a correct position of the four lead points E1-E4 on the human body 14, and
(h) if the distance of the reference point PC.sub.1, PC.sub.2 from the predetermined target point PC.sub.1.sup.fit, PC.sub.2.sup.fit is greater than a predetermined maximum value: Performing an angular correction for a first triangle 31 formed by the first lead point E1, the third lead point E3 and the fourth lead point E4, and for a second triangle 32 formed by the first lead point E1, the second lead point E2 and the fourth lead point E4, so that the Euclidean distance between the reference points PC.sub.1, PC.sub.2 and the predetermined target point PC.sub.1.sup.fit, PC.sub.2.sup.fit is minimized by adapting the orthogonal system x, y, z to the changed geometry.

[0128] The above steps (a) to (h) of the correction method are explained as meaning that the amplitudes with which the calculation vector 26 is formed in step (c) are those measured values which were previously measured non-invasively with the ECG signals at the heart 12 of the patient 11 in step (a) or, in the case of the method of FIG. 8, in its step (i), in the case of the method of FIG. 13, in its step 13.1 and/or in the case of the method of FIG. 14, in its step 14.1. Step (e), according to which the calculation vector 26 is multiplied by the 66 coefficient matrix 28 of step (d), resulting in the resulting vector 30, is illustrated in FIG. 15. This resulting vector 30 is present in the form of a 61 matrix. In step (f), the third to sixth rows of the resulting vector 30, as shown in FIG. 15 marked with the strikethrough, are deleted or ignored for the further calculation, wherein the first main axis PC.sub.1 and the second main axis PC.sub.2 are extracted from the resulting vector 30 to form the reference point PC.sub.1, PC.sub.2. Furthermore, the Euclidean distance, which is defined in step (g) of the correction method, is illustrated in the diagram of FIG. 16.

[0129] The diagram of FIG. 16 also shows an exemplary target area, here symbolized by a rectangle 34 with dashed lines. In this regard, it is noted that prior to performing step (h), the predetermined target point PC.sub.1.sup.fit, PC.sub.2.sup.fit is selected from this target area, which is formed on the basis of standard deviations of a plurality of verified correct measurements of the target point on the human body 14.

[0130] In step (h) of the correction method, the angular correction defined herein can be used to determine adjustment values , with which the non-orthogonal angles of the first triangle 31 and the non-orthogonal angles of the second triangle 32 can be corrected. Thus it is possible, to compensate for an, in particular, incorrect position of the first lead point E1 at the human body 14 in order to consider, if necessary, any non-orthogonal triangles. In this way, the adjustment value represents an adjusted value which, in the case of correctly applied electrodes, is identical to the angle of the first triangle 31. The same applies to the adjustment value , which in the case of correctly applied electrodes corresponds to the angle of the second triangle 32. This relationship is also shown graphically in FIG. 5A and FIG. 5B.

[0131] The above adjustment values , , which can be used for the angular correction of step (e), can be determined by minimizing the Euclidean distance between reference and target point.

[0132] The coefficient matrix 28 by which the calculation vector 26 is multiplied in step (e) is formed by a Principal Component Analysis of a 236 matrix based on 23 test measurements and the mean values and standard deviations of the measured values determined therefrom.

[0133] The correction method explained above is based on the principle of a Principal Component Analysis, with which, as a result, an incorrect fit of electrodes or sensors S1-S4 on the human body 14 can be compensated. This applies in particular to the position of the sensor S1, which is assigned to first lead point E1, and is advantageous e.g. for the case that the measured values of the heart 12 are acquired with the t-shirt 22 of FIGS. 3A-3B.

[0134] It has already been pointed out above in the discussion of FIG. 7A, that the space vector 24 shown therein, which is determined on the basis of vectorcardiography, represents the electrical activity of the heart 12. For the present invention, this also defines a heart monitoring method in which a quotient of the areas covered by a length of the space vector 24 (=radius vector) as a function of time is calculated during the R wave and during the T wave, respectively, wherein this quotient is then fed to be further evaluated. These areas covered by the space vector 24 as a function of time are shown by way of example in the diagram in FIG. 17.

[0135] For the present invention, it has been found that for the screened heart 12 of a patient 11, an ischemia or coronary artery disease (CAD) is detected if the quotient formed by the areas covered by the space vector 24 respectively during the R wave and during the T wave is

[00003]$\frac{\mathrm{Area}\left(R-\mathrm{wave}\right)}{\mathrm{Area}\left(T-\mathrm{wave}\right)}=a$

[0136] outside an interval between the limit values a.sub.0,CHD and a.sub.1,CHD.

[0137] A special case of the above findings is the case when a cardiac arrhythmia (HRD) is detected for the screened heart 12, if the quotient formed by the areas covered by the space vector 24 during the R wave and during the T wave, respectively, satisfies the condition

[00004]$\frac{\mathrm{Area}\left(R-\mathrm{wave}\right)}{\mathrm{Area}\left(T-\mathrm{wave}\right)}=a_{0,\mathrm{HRD}}$

[0138] or the condition

[00005]$\frac{\mathrm{Area}\left(R-\mathrm{wave}\right)}{\mathrm{Area}\left(T-\mathrm{wave}\right)}=a_{1,\mathrm{HRD}}$

[0139] With regard to the limit values a.sub.0,CHD and a.sub.1,CHD or a.sub.0,HRD and a.sub.1, HRD, with which the quotient formed by the area (R wave)/area (T wave) in the screening for the presence of CHD and/or HRD in the actual examination of a patient is compared or correlated in each case, reference may be made, in order to avoid repetition, to the explanations in the introductory part of the present patent application, according to which these limit values can also be determined or optimized as a function of a training set.

[0140] The various findings, which for the latter method according to the present invention, based on the ratio of the areas which are covered by the space vector 24 as a function of time during the R wave and during the T wave, respectively, can of course also be applied or taken into account in the aforementioned methods according to the invention, which are shown and explained by means of the flow charts according to FIG. 8, FIG. 9, FIG. 13 and FIG. 14, respectively.

[0141] The invention being thus described, it will be obvious that the same may be varied in many ways. Such variations are not to be regarded as a departure from the spirit and scope of the invention, and all such modifications as would be obvious to one skilled in the art are to be included within the scope of the following claims.