Method, system, and computer program for detecting cardiac health conditions

Abstract

1. A method for detecting a cardiac health status of a user, comprising: receiving, in a processing means, a measurement signal from an IMU placed on the user's body; determining, in the processing means, a plurality of features based on the measurement signal; and determining, in a machine learning engine of the processing means, the cardiac health status based on the plurality of features.

Description

[Technical field]

[0001] The present invention relates to a method, system and computer program for detecting a user's cardiac health, preferably for detecting congestive heart failure (CHF). [Background technology]

[0002] The health of the human heart is essential for human health. Early detection of heart problems can prevent serious complications in human health. Such problems can be CHF, coronary artery disease, atrial fibrillation, hypertension, etc. Traditionally, cardiac monitoring is done using electrocardiograms (ECGs). Some cardiac monitoring techniques are based on ballistocardiography (BCGs) or seismocardiography (SCGs). SCGs are usually measured at the human chest. BCGs can be measured in a variety of ways. While many SCGs or BCGs measure or use only one or a few dimensions, WO201736887 proposes cardiac monitoring based on a combination of ECGs and BCGs measured at the center of gravity of the body, and finally combined with SCGs. It is proposed to use multi-dimensional BCG and SCG measurements together with the three linear dimensions of an accelerometer to obtain the linear kinetic energy of the heart's motion, and together with the three rotational dimensions of a gyroscope to obtain the rotational kinetic energy of the heart's motion. However, these solutions all require complex and specialized clinical measurement equipment, such as ECG, SCG, and / or BCG equipment, which increases the cost of such cardiac monitoring and makes such cardiac monitoring unsuitable for simple at-home cardiac checks that may be performed without medical assistance.

[0003] Some cutting-edge techniques propose using everyday life techniques to measure cardiac SCGs, and machine learning (ML) techniques to analyze SCGs, to make simple cardiac monitoring available to everyone.

[0004] US20210057101 discloses measuring BCG using a remote sensor placed on a flat surface and analyzing the BCG using an ML engine to detect CHF, but the reliability of the remote detection of BCG is not very high, and the features used for the ML analysis are not disclosed.

[0005] EP3267886 proposes inputting a 6-dimensional SCG signal into an ML engine to determine cardiac dysfunction such as atrial fibrillation. Principal component analysis is proposed to reduce the dimensionality of the 6-dimensional SCG signal. Although principal component analysis can reduce the dimensionality of the signal, it does not solve the problem of a large number of samples in the time dimension, which increases the computational complexity of the ML analysis and reduces the reliability of the output.

[0006] EP3897355 proposes cardiac monitoring to detect internal bleeding in the body. It is disclosed that an ML engine is used to analyze BCG and ECG to determine bleeding in the body. BCG is measured in human body supply. Therefore, this solution also requires specialized equipment for cardiac monitoring.

[0007] EP3895605 discloses monitoring the heart using any kind of cardiac data such as ECG, EMG, BCG, etc. The data is subdivided into many sub-periods of about 30 seconds to 2 minutes. For each sub-period, a feature vector is calculated and fed into an ML engine. The ML engine detects sub-periods that are of interest for further human analysis, but does not detect cardiac health.

[0008] EP3731746 discloses measuring a user's SCG and ECG using a smartphone equipped with capacitive electrodes. The ECG is used to detect cardiac cycles in the measured signals and to segment each cardiac cycle into subsegments such as systolic and diastolic segments, or even finer subsegments. Features are extracted from the different subsegments of the cardiac cycle to feed into an ML engine to detect cardiac anomalies. As features, many possible time or frequency features are proposed. It is mentioned that cardiac anomaly detection can work without an ECG, but it is not explained how cardiac cycles and subsegments are identified without an ECG.

[0009] However, most of the above solutions require specialized measurement equipment, making such cardiac monitoring impractical in daily life. Most solutions require ECG measurements, which are difficult to obtain without an ECG device. Solutions based solely on SCG measurements remain unclear with respect to the feature vectors used to feed the ML engine. Thus, there are currently no known solutions that provide simple and robust detection of cardiac health. Summary of the Invention [Problem to be solved by the invention]

[0010] It is an object of the present invention to provide a method, system and computer program for detecting a user's cardiac health condition that overcomes the problems of the state of the art. [Means for solving the problem]

[0011] According to the present invention, this object is solved by a method, a system and a computer program according to the independent claims.

[0012] According to the present invention, this object is solved by a method for detecting a cardiac health status of a user, comprising the steps of receiving, in a processing means, a measurement signal from an IMU placed on the user's body, determining, in the processing means, a plurality of features based on the measurement signals, and determining, in an ML engine of the processing means, the cardiac health status based on the plurality of features.

[0013] The use of ML makes it possible to detect cardiac health conditions independent of ECG measurements and complex clinical equipment, for example using a simple smartphone.

[0014] The dependent claims refer to further advantageous embodiments.

[0015] In one embodiment, the IMU is placed on a smartphone or a wearable connected to the smartphone, the measurement signal includes a six-dimensional signal having three linear dimensions representing the linear acceleration of the heart measured using the accelerometer of the IMU and three rotational dimensions representing the angular velocity of the heart measured using the gyroscope of the IMU, a cardiac cycle of the user's heart is identified based on the measurement signal, and a plurality of sub-periods in the cardiac cycle are determined, and the plurality of features includes linear features and rotational features, the linear features including values ​​of at least one linear feature variable of the heart's motion in each sub-period calculated based on the three linear dimensions of the measurement signal, and the rotational features including values ​​of at least one rotational feature variable of the heart's motion in each sub-period calculated based on the three rotational dimensions of the measurement signal. The combination of ML, smartphone six-dimensional IMU measurement, ECG independence, and robustness of the rotational and linear parameters to smartphone orientation allows for good cardiac condition detection using a standard smartphone, requiring only software to be run on the smartphone to measure the six-dimensional measurement signal using the smartphone's IMU. Although this combination of features is particularly effective, the features can also be used individually or in other sub-combinations. Preferably, the features are determined independently from an ECG measurement, but it is also possible to determine the features based on additional ECG measurements.

[0016] In one embodiment, the measurement signal includes a six-dimensional signal having three linear dimensions representing the linear acceleration of the heart measured using the accelerometer of the IMU and three rotational dimensions representing the angular velocity of the heart measured using the gyroscope of the IMU, a cardiac cycle of the user's heart is identified based on the measurement signal, and a plurality of sub-periods in the cardiac cycle are determined, the plurality of features includes linear features and rotational features, the linear features include values ​​of at least one linear feature variable of the heart's motion in each sub-period calculated based on the three linear dimensions of the measurement signal, and the rotational features include values ​​of at least one rotational feature variable of the heart's motion in each sub-period calculated based on the three rotational dimensions of the measurement signal. By having a combined / scalar linear value and one combined / scalar rotational value, the dependency of the measurement signal from the IMU's orientation can be removed. The dependency of the measured components on the IMU's orientation has been shown to increase the error rate of the ML engine.

[0017] In one embodiment, the multiple features include values ​​of a linear feature variable at different times of a heartbeat and values ​​of a second feature variable at different times of a heartbeat, the first feature variable being determined based on three linear dimensions and the second feature variable being determined based on three rotational dimensions. By having a combined / scalar linear value and one combined / scalar rotational value, the dependency of the measurement signal from the IMU orientation can be removed. The dependency of the measured components on the IMU orientation has been shown to increase the error rate of the ML engine.

[0018] In one embodiment, the (at least one) linear feature variable is one or more of linear (cardiac) force, linear (cardiac) kinetic energy, and linear (cardiac) work. In one embodiment, the (at least one) rotational feature variable is one or more of rotational (cardiac) force, rotational (cardiac) kinetic energy, and rotational (cardiac) work. These kinetic parameters are independent of sensor orientation and have proven to be very robust in detecting cardiac health conditions.

[0019] In one embodiment, the measurement signal includes a six-dimensional signal having three linear dimensions representing the linear acceleration of the heart measured using an accelerometer of the IMU and three rotational dimensions representing the angular velocity of the heart measured using a gyroscope of the IMU, a cardiac cycle of the user's heart is identified and a number of sub-periods in the cardiac cycle are determined, the number of features including linear features and rotational features, the linear features including values ​​of linear kinetic energy of the heart's motion in each sub-period calculated based on the three linear dimensions of the measurement signal, and the rotational features including values ​​of rotational kinetic energy of the heart's motion in each sub-period calculated based on the three rotational dimensions of the measurement signal.

[0020] In one embodiment, the measurement signal comprises a six-dimensional signal having three linear dimensions representing the linear acceleration of the heart measured using the accelerometer of the IMU, and three rotational dimensions representing the angular velocity of the heart measured using the gyroscope of the IMU. The inventors have discovered that a measurement signal having all six dimensions provides significantly better results, as information cannot be hidden in some unmeasured dimensions. State-of-the-art techniques often only measured or at least considered some dimensions for analysis.

[0021] In one embodiment, a linear velocity signal is calculated based on three linear dimensions, and the plurality of features includes at least one linear feature calculated based on the linear velocity signal, and / or a rotational acceleration signal is calculated based on three rotational dimensions, and the plurality of features includes at least one rotational feature calculated based on the rotational acceleration signal. It has been found that kinematic features based on linear velocity and / or rotational acceleration, which are not directly measured, contain a lot of useful information that is not available from the measured signals. Characterization of the ML engine using these features has proven to be more robust than features that use only signal characteristics of the signals measured in the IMU. In particular, it has been found that linear kinetic energy calculated from the linear velocity signal is of particular interest as a feature of the ML engine.

[0022] In one embodiment, the method comprises the step of determining, in the processing means, a velocity signal representative of a velocity of the kinematic movement of the heart based on the measurement signal, the processing means determining a plurality of features based on the velocity signal.

[0023] In one embodiment, the processing means determines from the velocity signal a processed velocity signal representative of a square of the velocity of the kinematic motion of the heart or a kinetic energy of the cardiac motion, and the processing means determines a plurality of features based on the processed velocity signal.

[0024] In one embodiment, the plurality of features includes values ​​of at least one feature variable at different times during a heart beat.

[0025] In one embodiment, a cardiac cycle of the user's heart is identified, a number of sub-periods in the cardiac cycle are determined, and the number of features includes values ​​of at least one feature variable in each determined sub-period. Defining the sub-periods based (only) on the measurement signal from the IMU (and not on the ECG) allows such segmentation to occur even without an ECG signal, which makes the current feature extraction possible even on a simple smartphone.

[0026] In one embodiment, a reference point in the cardiac cycle is determined, and a sub-cycle is defined based on the reference point and based on a fixed time width of the sub-cycle. Instead of calculating the time width (or start and end) of the sub-cycle for every heartbeat, the time width of each sub-cycle is selected as fixed (between heartbeats), and in this way a simple and reliable determination can be made regarding the reference point. The use of a fixed time width of the sub-cycle allows for a simple, robust and repeatable calculation of the sub-cycle and thus a repeatable and robust calculation of the value of the characteristic variable of the sub-cycle. The fixed time width is preferably defined independently of the physiological parameters, or a time point such as, for example, a specific peak of the signal, or the end or start of the systole or diastole. A sub-cycle defined based on the physiological parameters is computationally complex and prone to errors if only one of the physiological parameters is not detected correctly. The fixed time width is preferably an absolute fixed time width or a relative fixed time width. An absolute fixed time width has an absolute value of the time width that does not change between heartbeats. The relatively fixed time duration has, for example, a relative value (percentage value) with respect to the length of the cardiac cycle that does not change between cardiac cycles.

[0027] In one embodiment, cardiac cycles in the measurement signal are identified based on a matrix profile motif algorithm applied to a signal based on the measurement signal, the matrix profile motif algorithm identifying one or more reference cardiac cycles as motifs in the signal based on the matrix profile. Detection of cardiac cycles in the IMU signal is very difficult without measurement of an ECG signal. The use of the matrix profile allows for robust detection of cardiac cycles at low computational cost.

[0028] In one embodiment, respiratory information of a user is detected based on a measurement signal.

[0029] In one embodiment, the multiple features include the user's respiratory information. Since the heartbeat is highly dependent on the user's respiratory phase / state, this feature as an input of the ML engine has been found to significantly improve the results of the ML engine. This is especially true when the respiratory information is combined with multiple feature variables related to the linear and rotational motion of the heart in different partial cycles. This allows the ML engine to distinguish between various movements of the heart in different respiratory states, and thus avoid erroneous conclusions due to differences due to the user's respiratory state.

[0030] In one embodiment, the features are determined independently from the ECG measurement, making the method feasible on any common smartphone that includes an IMU.

[0031] In one embodiment, the ML engine determines whether the heart exhibits congestive heart failure.

[0032] In one embodiment, the method includes placing an IMU on a user's chest and measuring a measurement signal using the IMU when placed on the user's chest.

[0033] In one embodiment, the IMU is included in a smartphone or a wearable connected to the smartphone, and the processing means requests from an operating system running on the smartphone a type of smartphone or wearable or a user input regarding the type of smartphone or wearable, and the plurality of characteristics depends on the type of smartphone received back, and in particular on the weight of the type of smartphone received back. The type of smartphone / wearable is preferably a model. This allows to obtain any motion parameters calibrated with the correct mass or moment of inertia of the smartphone. Calibration with the type of sensor used makes the measurement signals even more comparable. This embodiment is particularly advantageous for ML engines that determine the health of the heart, since in order to provide good results, different measurements on different smartphones / wearables should be comparable. However, this has always been ignored in the state of the art. However, this embodiment can be advantageous even when detecting the health of the user without using an ML engine.

[0034] In one embodiment, the user lies down while the smartphone including the IMU is placed on the user's chest. While the state of the art has mainly proposed sleeves or fixtures to fix the measurement device to the chest or elsewhere on the body, it has been discovered by the inventors that measuring the IMU signal in a lying position has several advantages regarding the measurement quality: The user is not moving, so movements caused by the user's movements do not disturb the measurement; The user is relaxed while lying down, and the measurement signal is always measured in a relaxed state, resulting in a more reproducible signal; Finally, the sleeve holding the device including the IMU dampens the movement and therefore reduces the signal quality. Other embodiments according to the invention are mentioned in the appended claims and in the following description of the embodiments of the invention. [Brief description of the drawings]

[0035] [Figure 1]FIG. 1 illustrates an embodiment of a system for detecting a user's cardiac health, in accordance with the present invention. [Diagram 2] FIG. 1 illustrates an embodiment of a method for detecting a user's cardiac health in accordance with the present invention. [Diagram 3] FIG. 1 illustrates an example embodiment of feature extraction. [Figure 4] FIG. 1 illustrates a first embodiment of a system. [Diagram 5] 1 shows a second embodiment of the system, in which the same reference numbers are assigned to the same or similar elements. [Figure 6] FIG. 2 shows examples of measurement signals, respiratory signals and pre-feature signals. [Figure 7] FIG. 13 is a diagram illustrating an example of some steps of feature extraction. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0036] Other characteristics and advantages of the invention will emerge from the following non-limiting description and by reference to the drawings and examples.

[0037] The system, method and computer program according to the present invention detects the cardiac health condition (heart health condition) of a user 10. The terms body and heart shall refer to the body and heart of the user 10 whose cardiac health condition is to be detected. The user is usually a human being. However, the user can also be an animal, preferably a mammal. The user 10 is referred to as a user and not a patient, since the system, method and computer program can be executed by any person and no medical supervision is required. Obviously, this should not exclude that medical personnel can also execute the system, method and computer program according to the present invention on the user 10 (in this case the patient). The user 10 must always be the person whose cardiac health condition has to be detected. The person controlling the system, method and computer program according to the present invention is preferably the same user, but can be a different person.

[0038] 1 shows an embodiment of a system for detecting a user's cardiac health according to the present invention, comprising an inertial measurement unit (IMU) 1 and a processing means 2.

[0039] The IMU1 is configured to measure a measurement signal of the heart's movement / motion. The measurement signal preferably includes at least six dimensions. The six dimensions include three translational dimensions, also called linear dimensions, and three rotational dimensions, also called angular dimensions. The IMU1 typically comprises an accelerometer for measuring the acceleration of the heart's movement (in three linear dimensions) and a gyroscope for measuring the angular velocity (in three rotational dimensions). The IMU1 typically comprises the accelerometer and the gyroscope on a common chip or device. However, the gyroscope and the accelerometer can also be realized as separate chips or devices and still be considered as an IMU1. The IMU1 is preferably placed on the chest of the user 10 to measure a measurement signal, usually called SCG. Often, SCG refers only to three linear acceleration measurements or even only to one linear and / or rotational dimension. However, in this specification, SCG refers to a six-dimensional measurement signal including three linear and three rotational dimensions. Usually, one, two, or three of the three linear dimensions are referred to as SCG, and one, two, or three of the three rotational dimensions are referred to as Gyrocardiography (GCG). For simplicity, in this specification, a six-dimensional measurement signal including SCG and GCG is referred to as SCG.

[0040] In a preferred embodiment, the IMU 1 is located on the smartphone 5 as shown in the embodiment of Fig. 4. Thus, the standard IMU 1 available on almost all smartphones 4 can be used to detect the cardiac measurement signals required for the present invention. No other or additional sensors are required to perform the cardiac measurements.

[0041] However, it is also possible that the IMU 1 is not located on the smartphone 5. The IMU 1 may for example be located on a wearable 7 that is worn on the user's body like a smart watch, a heart monitor fixed to the chest using a sleeve around the torso, or another wearable. Such a wearable may be connected to the smartphone 5, preferably via a wireless connection, for example a Bluetooth connection. For example, the IMU 1 may be located on a heart monitor fixed to the chest of the user 10 using a sleeve.

[0042] It is also possible that the IMU 1 is located in a dedicated device that is not associated with the smartphone 5.

[0043] The smartphone 5 preferably comprises a smartphone processor, a storage, communication means, input means, preferably a touch screen, and output means, preferably a (touch) screen. The storage of the smartphone 5 is configured to store application programs for performing special functions on the smartphone 5 when the application programs are executed on the smartphone processor. Such application programs are often also called apps. The smartphone 5 in the embodiment of figures 4 and 5 preferably comprises an application program (hereafter called monitoring app) for performing or requesting at least some steps of the inventive method described later. The monitoring app is preferably installed on the smartphone 5, i.e. stored in the smartphone 5 storage. However, the term monitoring app also refers to a front-end program that is downloaded each time a particular website is accessed using the smartphone browser. Preferably, the monitoring app is configured to request the IMU 1 to measure a measurement signal and to receive the measurement signal from the IMU 1 (located on the same smartphone 5 as in figure 4 or connected to the smartphone 5 as in figure 5). Preferably, the monitoring app is configured to communicate with the server 6. The communication means are configured to establish a communication connection to the server 6. The communication means can be a WLAN and / or a mobile phone network communication module. The mobile phone network communication module preferably includes a subscriber identity module (a virtual one (eSIM), a SIM card, or simply a holder of such a SIM card) for receiving data required for communication with the mobile phone network.

[0044] The processing means 2 is configured to carry out all or at least some / most steps (S2-S4) of the present invention as will be explained in more detail below.

[0045] The processing means 2 can be a single processor or a combination of processors. For example, different processors can realize different steps. Preferably, the processing means 2 can comprise a local processor and a remote processor. The local processor can be located in the aforementioned smartphone 5 (smartphone processor) and / or wearable 7 or a dedicated device. The remote processor can be located in, for example, a server 6 connected to the local processor (or a device including the local processor) via the Internet. The remote processor shall be understood not only as one physical processor but also as one that realizes the functions of the remote processor or server 6, for example a server center or a cloud, etc. Steps S2 to S4 described below may be realized all on the local processor, or all on the remote processor, or distributed on both. Preferably, the processing means includes a general-purpose processor (or a plurality of them) on which software is running to realize the functions of the method of the invention described later. This software can realize all functions or only some functions. If a local processor is present, it is preferable that there is local software, for example a monitoring app, running on the local processor. When there is a remote processor, there is preferably remote software running on a remote processor, for example server or back-end software.

[0046] The processing means 2 comprises a feature extractor 3 and an ML engine 4 .

[0047] The feature extractor 3 is configured to implement step S3, described later, for extracting features from the measurement signals for the ML engine 4. The feature extractor 3 may be implemented in a local processor / smartphone 5, or in a remote processor / server 6, or both. Preferably, the processing means comprises a general purpose processor (or several of them) running software to implement the functionality of the feature extractor 3, e.g. a monitoring app and / or server software.

[0048] The ML engine 4 is configured to implement step S4, which will be described later. The ML engine 4 may be implemented on a general purpose processor or on a dedicated ML processor. The general purpose processor can be the same as the processor on which the feature extractor 3 runs. In this case, the ML engine 4 is implemented as software. However, it is also possible that the ML engine 4 is implemented in a dedicated chip or processor dedicated to the ML application. Such a dedicated ML chip is then programmed for this ML application, as will be described later. The ML engine 4 is preferably located in a remote processor / server 6. However, it is also possible that the ML engine 4 is present in a local processor / smartphone 5.

[0049] FIG. 2 illustrates an embodiment of a method for detecting a user's cardiac health condition according to the present invention.

[0050] In a first step S1, the IMU1 is placed on the body of the user 10 to measure a measurement signal. The measurement signal is preferably an SCG. The IMU1 is placed on the user's body in such a way that the IMU1 can detect the movement of the heart. Preferably, the IMU1 is placed on the torso, preferably on the upper part of the torso and / or preferably on the chest side of the torso, most preferably on the chest of the user 10. Preferably, the user 10 lies down for the measurement and the IMU1, preferably in the form of a smartphone 5, is placed on the torso. As the user 10 is lying down, the IMU1 or the smartphone 5 must not be fixed to the torso and can freely measure the movements produced by the heart. A sleeve or other means for pressing the MU1 against the torso may dampen the movement of the IMU1. In addition, in this way the user 10 is in a relaxed state and the measurement results are not tampered with by other body movements that may tamper with the measurement results. The user 10 can perform the measurement himself by providing the monitoring app / smartphone 5 / IMU1 with an input command to start the measurement and by placing the smartphone 5 / IMU1 on his torso / chest (in any order). The measurement can thus be performed using any common smartphone 5 in which the IMU1 is included. The user only needs to download a special monitoring app to the smartphone 5. Preferably, the measurement lasts less than a maximum length, preferably less than 10 minutes, preferably less than 5 minutes, preferably less than 2 minutes, preferably less than 1 minute. In this way, the measurement step can be realized quickly by the user 10. The measurement preferably lasts longer than a minimum length, which is preferably longer than one heart beat, preferably longer than 1 second, preferably longer than 5 seconds, preferably longer than 10 seconds, preferably longer than 20 seconds. In this way, the IMU1 records the measurement signal SCG for a certain measurement window. The measurement window is preferably between a minimum length and a maximum length. But obviously other lengths are also possible. The IMU1 / smartphone 5 is placed on the torso of the user 10 and therefore moves with the movement of the heart or heartbeat. The measurement signal SCG therefore represents the heart's motion within a six-dimensional (degree of freedom) measurement window.The measurement signal SCG is measured at a particular sampling rate, which is preferably higher than 10 Hz, preferably higher than 25 Hz, preferably higher than 50 Hz, preferably higher than 60 Hz, preferably higher than 70 Hz, preferably higher than 80 Hz, preferably higher than 90 Hz, preferably higher than 100 Hz.

[0051] Step S1 is optional in practice and the method or computer program of the invention may start with step 2 of receiving the measurement signal SCG as described in step S1. If step S1 is included in the method or computer program, the monitoring app / smartphone 5 / dedicated device preferably receives a user input to start a measurement using the IMU1. After this user input is received, the monitoring app / smartphone 5 / dedicated device sends an instruction to the IMU1 to start the measurement signal SCG. The duration of the measurement is preferably pre-determined or may be configured in the monitoring app. The monitoring app / smartphone 5 / dedicated device preferably issues an indication that the measurement has ended, e.g. an audio signal, when the duration of the measurement has ended.

[0052] In step S2, the system / method / computer program / monitoring application / smartphone 5 / processing means 2 receives the measurement signal SCG from the IMU 1.

[0053] In step S3, the processing means 2 determines a number of features based on the measurement signal SCG. These features are used in the next step as input features (also called feature vector) for the ML engine 4. The features preferably include data features derived from the measurement signal SCG and subject features associated with the user 10.

[0054] The subject features are usually independent of the measurement signal SCG. The subject features preferably include one or more of the following: age, sex, weight, height, body mass index (BMI), body surface area (BSA) of the user 10. The subject features are preferably derived from input information, preferably user input. This user input may be received, for example, in a monitoring app in the smartphone 5. However, the input information may also be stored in a database on the server 6 of the user 10 and retrieved therefrom. For example, this information may be received from a parallel application that already has this information and allows the transfer. The BMI and BSA may be calculated from the weight and height of the user 10. Even if the presence of a subject feature in the features of the ML engine 4 is preferred, it is also possible to perform step S4 without a subject feature or using a different subject feature.

[0055] In a preferred embodiment, the data features are derived (only) from the measurement signal SCG measured using the IMU1. Preferably, the data features are derived only from the measurement signal SCG measured using the IMU1, i.e. independent of other sensors placed on the body, in particular independent of sensors not included in a standard smartphone, in particular independent of ECG measurements, which usually require special equipment. However, the invention can also be applied in more complex embodiments including ECG sensors and / or other sensors, for example using a special wearable 7 including an IMU sensor 1 and an ECG sensor. The wearable can be connected to the smartphone 5. The wearable can be worn by a chest belt or simply placed or held on the chest of the user 10 for the measurement.

[0056] 3 illustrates one preferred embodiment for determining data features based on a measurement signal SCG (hereinafter simply SCG for short). The data features may be obtained from the complete measurement window of the SCG signal or from only a sub-window of the measurement window.

[0057] In step S31, respiratory information is retrieved from the SCG. The SCG also reflects respiratory information, since breathing causes an increase or decrease in the volume of the torso or thorax / chest. Preferably, a respiratory signal is extracted from the SCG. This can be achieved by independent component analysis, low-pass filtering, or other means. The respiratory signal is preferably determined based on a complete measurement window. The respiratory information is preferably based on the respiratory signal. Preferably, the respiratory signal is discretized into a plurality of respiratory states. The respiratory states preferably include at least four respiratory states. The respiratory states preferably include two or more of a high-volume state (HV), a middle volume state during expiration (MV-EX), a low volume state (LV), and a middle volume state during inspiration (MV-IN). It is also possible that there are five or more respiratory states. The high volume state HV can be differentiated into two states (high volume during expiration (HV-EX) and high volume during inspiration (LV-IN)) or into three states (HV-EX, HV, HV-IN). The same can be done for the low volume state LV. Also, the medium volume state MV can be differentiated into smaller states. Thus, one respiratory cycle / period includes a sequence of HV, then MV-EX, then LV, and then MV-IN (or any other complete sequence of respiratory states of a respiratory cycle / period). The discretization of the respiratory signal leads to a respiratory state signal indicating the respiratory state over time (within the measurement window). The discretization of the respiratory signal can be performed only on the complete measurement window or only on a sub-window of interest. The features preferably include at least one respiratory information, preferably the respiratory state at at least one time point of the heartbeat. The respiratory signal / information may be determined from all six dimensions of the SCG, from a subgroup of the SCG dimensions, or from one or more composite signals each determined from one or more (preferably different) subgroups of the SCG signals.FIG. 6 shows an example of a respiratory signal 23 extracted from an exemplary SCG.

[0058] In step S32, a pre-processing of the SCG signal is performed. This pre-processing mainly involves the removal of noise, i.e. the removal of signal components that are not related to the heart movement. This pre-processing can be performed, for example, by a frequency filter, for example a high-pass filter or a low-pass filter or a band-pass filter. The pre-processing step preferably removes frequencies above 100 Hz, preferably above 80 Hz, preferably above 70 Hz, preferably above 60 Hz. The pre-processing step preferably removes frequencies below 1 Hz, preferably below 2 Hz, preferably below 3 Hz. Preferably, the pre-processing also removes any trends in the SCG. Pre-processing is a preferred step, but is not essential for the invention. The respiratory signal / information is preferably determined based on the SCG before pre-processing or by a different pre-processing, except for the removal of frequencies below the normal heartbeat frequency. Figure 6 shows a three linear dimension a x (t), a y (t), a z (t) and the three rotation dimensions ω x (t), ω y (t), ω z (t) shows an example of a 6D SCG signal preprocessed using signal a x (t) represents the acceleration in the x direction, and the signal a y (t) is the acceleration in the y direction, and a z (t) shows the acceleration in the z direction. Signal ω x (t) denotes the rotation speed around the x direction, and the signal ω y (t) denotes the rotation speed around the y direction, and the signal ω z(t) denotes the rotation speed around the z-direction. The x-, y-, and z-directions refer to the three orthogonal dimensions of the Cartesian coordinate system of the IMU 1. Preferably, the coordinate system of the IMU 1 coincides with the coordinate system of the user 10. The user's coordinate system preferably has a y-direction in the direction from the user's feet to the head, an x-direction in the direction from one arm to the other, preferably from the right arm to the left arm, and a z-direction from the back of the torso to the chest. Such alignment may be achieved by giving the user instructions on how to position the smartphone 5 on the body. For example, the instructions may display an arrow pointing to the head of the user 10. However, it is also possible that the positioning of the IMU 1 with respect to the user 10 is unknown.

[0059] In step S33, at least one pre-feature signal is calculated based on the (preferably pre-processed) SCG signal. In a preferred embodiment, at least one (scalar) linear pre-feature signal is calculated based on the three rotation dimensions ω of the SCG. x (t), ω y (t), ω z (t) (independent of the three linear dimensions of the SCG a x (t), a y (t), a z (t), and / or at least one (scalar) rotation pre-feature signal is determined based on a combination of (t) (the three linear dimensions of the SCG a x (t), a y (t), a z (t) independent of the three rotation dimensions of the SCG ω x (t)ω y (t), ω z (t) is determined based on a combination of (t). The (linear and / or rotational) pre-features preferably correspond to physical variables describing the cardiac motion, such as kinetic energy, kinetic power, work, force / torque, etc. Physical values ​​describing the cardiac linear motion are, for example:

[0060]

number

[0061] Physical values ​​representing the rotational motion of the heart are, for example, as follows:

[0062]

number

[0063] The measurement or pre-processing step S32 of the IMU1 provides two measurement sub-signals: a three-dimensional linear acceleration vector signal a(t) and a three-dimensional angular velocity vector signal ω(t). Preferably, in step S33, at least one composite linear motion pre-characteristic signal is calculated based on the three-dimensional linear acceleration vector signal a(t) and / or at least one composite rotational motion pre-characteristic signal is calculated based on the three-dimensional rotational velocity vector signal ω(t). The at least one composite linear motion pre-characteristic signal is preferably a scalar value combining the three linear dimensions of the measurement signal into one scalar value and / or the at least one composite rotational motion pre-characteristic signal is preferably a scalar value combining the three rotational dimensions of the measurement signal into one scalar value, such that the composite value is independent of the orientation of the IMU1 when measuring the measurement signal. The at least one composite linear pre-characteristic signal is preferably a linear motion energy signal K lin (t) and / or the linear motion force signal P lin (t). One composite linear pre-feature signal and / or linear kinetic energy signal K lin Preferably, v(t) is calculated for / based on a linear velocity vector v(t). The at least one composite rotational pre-characteristic signal is a rotational kinetic energy signal K rot (t) and / or the rotational force signal P rot (t). One composite linear pre-characteristic signal and / or rotational force signal P rot (t) is preferably calculated for / based on the rotational acceleration vector α(t).

[0064] Herein, the heart motion is approximated by the motion of the IMU1. So the physical variables of the heart motion finally correspond to the (preferably preprocessed) motion of the IMU1 or the device including the IMU1. The value / signal of the linear or rotational kinetic energy and the value of the linear or rotational kinetic force depend on the device in which the IMU1 is located, e.g. the moment of inertia I and / or its mass m. If the device is a dedicated device that is always the same, this value is simply a constant and does not play a major role. However, when using a smartphone 5 or other wearable, the correct mass and / or moment of inertia must be retrieved to calculate the pre-characteristic signal, the characteristic, the linear / rotational kinetic energy and / or the linear / rotational kinetic force. In this case, the processing means, preferably a monitoring app, retrieves from the operating system of the smartphone 5 or the wearable the type of the device including the IMU1. Instead of automatically retrieving the type of smartphone 5 / wearable 7 from the operating system of the smartphone 5, it is also possible that the smartphone 5 information on the type of smartphone 5 or wearable 7 is entered by user input. The processing means may store for each type of smartphone 5 or wearable 7 relevant information for calculating / calibrating the characteristics, preferably the moment of inertia and / or mass of the smartphone 5 / wearable 7. The type of smartphone 5 / wearable 7 preferably refers to the model of the smartphone 5 / wearable 7. Based on the type of device the mass and / or moment of inertia of the device may be retrieved. Thus the kinetic energy and kinetic forces are calibrated using the device characteristics, which measure the heart motion and prevent erroneous results based on the use of different measuring devices. Hereinafter the term kinetic energy signal K(t) refers to the linear kinetic energy signal Klin (t) as a first dimension or component, and a rotational kinetic energy signal K rot A two-dimensional signal K(t)=(K lin (t),K rotHereinafter, the term motion force signal P(t) will refer to the linear motion force signal P(t) or to one of the two components individually. lin (t) as a first dimension or component, and a rotational force signal P rot A two-dimensional signal P(t)=(P lin (t),P rot (t)) or individually to one of the two components.

[0065] At least one heart beat is identified in step S34. Preferably, multiple (e.g. all) heart beats are identified in the SCG. If multiple heart beats are identified, at least one is selected for further analysis. A heart beat is a time window related to one heart beat of the heart. A heart beat is a periodic cardiac cycle of the heart. A heart beat is usually defined in terms of one characteristic point in the heart beat, usually R or Q in the ECG signal. Since the invention is preferably realized without an ECG signal, at least one heart beat is identified based on the SCG, preferably based on at least one pre-feature signal calculated in step S33, preferably based on a kinetic energy signal and / or a kinetic force signal, or a combination of these. In a preferred embodiment, the (data) features are calculated based on the SCG (or pre-feature signal) of one heart beat, i.e. one cardiac cycle. This also includes the case where a representative heart beat is calculated by averaging multiple heart beats (preferably of the same respiratory state). In a preferred embodiment, the cardiac health status is determined based on multiple heart beats by selecting multiple heart beats and detecting the health status for each heart beat in steps S34, S35, and S4. A composite cardiac health status can then be calculated by combining the cardiac health status from the individual heart beats. State-of-the-art techniques usually do not calculate feature vectors based on a single heart beat, which reduces the robustness of the method. Although some state-of-the-art documents propose to calculate feature vectors based on data of one heart beat, those methods always require an ECG signal to detect heart beats. It is one of the innovations of the present invention that heart beats can be identified from SCG (without using ECG signal). Heart beat cycles are determined based on the characteristics of heart beats in SCG and / or pre-feature signals (e.g., kinetic energy or kinetic force) obtained from SCG. While heart beats in SCG and pre-feature signals lack clear characteristics as in ECG signals, robust detection of heart beat cycles is difficult. A preferred embodiment for detecting heart beats in SCG is described below.

[0066] It is proposed to detect heartbeat cycles in the SCG (also including the pre-feature signal determined from the SCG) based on the matrix profile. The matrix profile m(t1) of the signal s(t) at sample time t1 indicates the distance of s(t) to the nearest neighbor in the signal s(t) within the entire time window of the signal s(t). The distance is preferably a Euclidean distance measure, but obviously other distance measures can also be used. The matrix profile m(t1) allows to detect similar motifs in the signal s(t). Since each person's heartbeat looks a little different, this allows to detect repetitive motifs of heartbeats in the signal used to determine the matrix profile m(t). Preferably, the matrix profile m(t) is calculated based on the SCG (which includes the meaning that the matrix profile m(t) is calculated based on the (pre-feature) signal calculated from the SCG), so that the heartbeat cycles are determined computationally very quickly and with high robustness. The matrix profile m(t) provides not only the location of the motifs, i.e. regular heartbeats, but also the location of data anomalies, called dissonances, and the quality of each motif. The quality of each motif is a similarity measure with the most common motif in the signal s(t). Preferably, a multidimensional signal s(t) is used to detect motifs / heartbeat cycles. For details of an example of an algorithm for detecting motifs in a multidimensional signal based on a matrix profile, reference can be made to the article "Meaningful Multidimensional Motif Discovery" by Chin-Chia Michael Yeh, Nickolas Kavantzas, and Eamonn Keogh, published in ICDM 2017, which is incorporated by reference for exemplary details of the matrix profile. The length of the motif is preferably set to a value smaller than 1 second, preferably smaller than 800 milliseconds (ms), preferably smaller than 700 ms. The length of the motif is preferably set to a value larger than 200 ms, preferably larger than 400 ms, preferably larger than 500 ms.The length of the motif is preferably set to a value smaller than 1800 ms, preferably smaller than 1500 ms, preferably smaller than 1000 ms, preferably smaller than 800 ms. In step S34, the linear kinetic energy K is calculated. lin (t) and the rotational kinetic energy K rot In (t) an example of a motif 22 detected using the matrix profile is shown. In one embodiment, one motif 22 is detected. However, it is also possible to detect more than one motif 22 in the underlying signal, as shown in FIG. 7.

[0067] The cardiac cycles are preferably identified based on a pre-characteristic signal, preferably based on a kinetic energy signal and / or a kinetic force signal.

[0068] In step S35, in the identified cardiac cycle, a cardiac sub-cycle 24 is identified. Preferably, the sub-cycle 24 is determined based on one characteristic point of the SCG in the cardiac cycle (including the meaning of a pre-characteristic signal determined based on the SCG). Preferably, the characteristic point is a maximum rotational kinetic energy or a maximum rotational kinetic force in the cardiac cycle. Preferably, the sub-cycle 24 is determined based on the time of the characteristic or reference point and a fixed time relationship / width between the sub-cycles 24 of the cardiac cycle. The fixed time relationship / width of the sub-cycles 24 means that the time width of each sub-cycle 24 does not change from one cardiac beat to the next. This is not the case when the start and end points of the sub-cycle are defined by physiological time points that need to be detected for each cardiac beat. The fixed time relationship / width can be an absolute fixed time relationship / width or a relative fixed time relationship / width. An absolute fixed time relationship is meant to mean that the widths of the sub-periods 24 and the time distances between the sub-periods 24 all have a fixed absolute time relationship (fixed absolute time value) and the positions of the sub-periods 24 with the fixed absolute time relationship between the sub-periods 24 are simply correctly positioned based on the time of the characteristic point. A relative fixed time relationship is meant to mean that the widths of the sub-periods 24 and / or the time distances between the sub-periods 24 all have a fixed relative time relationship (fixed percentage). Therefore, in addition to the correct positioning of the sub-periods 24, a scaling must also be adapted. The scaling can for example depend on the heartbeat frequency or on the length of the selected heartbeat. The sub-periods 24 preferably all have the same (absolute or relative) width. The width of each sub-period 24 is preferably less than 200 ms, preferably less than 150 ms, preferably less than 140 ms, preferably less than 130 ms, preferably less than 120 ms, preferably less than 110 ms, preferably less than 105 ms, and / or preferably greater than 20 ms, preferably greater than 40 ms, preferably greater than 50 ms, preferably greater than 60 ms, preferably greater than 70 ms, preferably greater than 80 ms, preferably greater than 90 ms, preferably greater than 95 ms. Preferably, the width of each sub-period 24 is 100 ms.Preferably, the width of each sub-period 24 is preferably less than 30%, preferably less than 25%, preferably less than 20%, preferably less than 15%, preferably less than 10% of the cardiac cycle / length and / or preferably more than 2%, preferably more than 5%, preferably more than 7%, preferably more than 8%, preferably more than 9% of the cardiac cycle / length. The sub-periods 24 are preferably non-overlapping, i.e. no sample time occurs in two sub-periods 24. The sub-periods 24 are preferably arranged next to each other such that no sample time occurs between two sub-periods 24 that does not belong to any sub-period 24. Preferably, more than 5, preferably more than 6, preferably more than 7, preferably more than 8, preferably more than 9, preferably more than 10 sub-periods 24 are identified. Preferably, less than 20, preferably less than 15, preferably less than 14, preferably less than 13, preferably less than 12 sub-periods 24 are identified. Preferably, eleven partial periods 24 are identified. Preferably, a first partial period 24 is arranged to cover the characteristic point. A first group of partial periods 24 is arranged before the first partial period 24 and a second (different) group of partial periods 24 is arranged after the first partial period 24. The first group preferably includes less than half of the partial period 24 and / or the second group preferably includes more than half of the partial period 24. FIG. 7 shows an example of step S35 for an arrangement of eleven partial periods 24 of 100 ms each, with four partial periods 24 before the characteristic point, one partial period 24 at the characteristic point and six partial periods 24 after the characteristic point. The cumulative duration of all partial periods 24 can be longer or shorter than a cardiac cycle, i.e. it can already extend into the previous and / or next cardiac cycle or it cannot completely cover a cardiac cycle.

[0069] In step S36, (data) features (used as input in the ML engine 4) are calculated. The (data) features are preferably calculated using at least one feature K iwhere i indicates the index of each sub-period 24 from sub-period i=1 to sub-period i=n, ​​and n is the number of sub-periods in the cardiac cycle. Preferably, the same physical (motion) variables are calculated for each sub-period. Preferably, the data features include one or more of the following for each sub-period: linear kinetic energy, rotational kinetic energy, linear kinetic force, rotational kinetic force, respiratory information or state, rotational cardiac work, linear cardiac work. The linear / rotational kinetic energy of a sub-period can be one or more of the following: accumulated linear / rotational kinetic energy in the sub-period, linear / rotational kinetic energy density in the sub-period, maximum value of linear / rotational kinetic energy in the sub-period, time sample of maximum value of linear / rotational kinetic energy in the sub-period, minimum value of linear / rotational kinetic energy in the sub-period, time sample of minimum value of linear / rotational kinetic energy in the sub-period, or other values ​​such as the mean, median, variance, etc. of linear / rotational kinetic energy in the sub-period. The accumulated linear / rotational kinetic energy of a sub-period refers to the integral of linear / rotational kinetic energy over the sub-period. The linear / rotational kinetic energy density in a partial period is the cumulative linear / rotational kinetic energy in a partial period divided by the width (duration) of the partial period. The linear / rotational kinetic power in a partial period can be one or more of the cumulative linear / rotational kinetic power in a partial period, the maximum linear / rotational kinetic power in a partial period, a time sample of the maximum linear / rotational kinetic power in a partial period, the minimum linear / rotational kinetic power in a partial period, a time sample of the minimum linear / rotational kinetic power in a partial period, or other values ​​such as the mean, median, variance, etc. of the linear / rotational kinetic power in a partial period. The cumulative linear / rotational power in a partial period means the integral of the linear / rotational power over the partial period. The linear / rotational kinetic power density in a partial period is the cumulative linear / rotational kinetic power in a partial period divided by the width (duration) of the partial period. For brevity, linear / rotational is meant in this paragraph to mean that linear / rotational applies to linear variables such as linear kinetic energy, linear kinetic power, etc., and / or rotational variables such as rotational kinetic energy, rotational kinetic power, etc.These (data) features are calculated based on the SCG signal in the cardiac cycle (or each partial cycle), preferably based on a prior feature signal in the cardiac cycle (or each partial cycle), preferably based on a linear / rotational motion energy signal in the cardiac cycle (or each partial cycle), and / or based on a linear / rotational motion motive force signal in the cardiac cycle (or each partial cycle). Step S33 for the calculation of the prior feature signal can also be realized later, i.e. after steps S33 and S35, if it is not required for steps S33 and S35. Step S33 can also be omitted and directly included in step S36 of the calculation of the (data) features.

[0070] The respiratory state for each sub-cycle is calculated based on the respiratory signal or respiratory state in the sub-cycle determined in step S31. It is also possible that the respiratory information / state is entered only once for a complete cardiac cycle instead of every sub-cycle. In this case, it is preferable to select a cardiac cycle that corresponds completely to (only) one respiratory state.

[0071] The (data) features may further include the motif quality obtained in step S34. The motif quality provides a kind of deviation of the normal pattern of the heartbeat of the user 10 and may be used as a further interesting input feature for the ML engine 4.

[0072] FIG. 3 illustrates a preferred embodiment for determining the data features that are fed to the ML engine 4. Obviously, alternative embodiments for (data) feature extraction are possible. For example, a sub-cycle can span multiple heartbeats (two or more). For example, different feature variables are calculated for different sub-cycles. The start and end points of a sub-cycle can be detected based on characteristic points in the data. Instead of a sub-cycle, one, some or all features can be calculated for a complete heartbeat cycle. Many alternatives are possible.

[0073] Returning to FIG. 2, the features determined in step S3 are provided to the ML engine 4 of the processing means 2. The ML engine 4 processes the features input and outputs a cardiac health status of the user 10. The cardiac health status preferably refers to one medical condition / problem of the heart. In a preferred embodiment, the cardiac health status refers to CHF. However, the described method can also be applied to detect other medical conditions / problems of the heart, such as atrial fibrillation, valvular disease, arrhythmia, heart murmur, myocardial infarction, drug induced cardiotoxicity. Preferably, the cardiac health status is binary, i.e. the cardiac health status corresponds to a first cardiac health status indicating that the heart of the user 10 is not suffering from a medical condition / problem, e.g. CHF, or corresponds to a second cardiac health status indicating that the heart of the user 10 is actually suffering from or at risk of suffering from a medical condition / problem, e.g. CHF. Instead of a binary output of the ML engine 4, it is also possible to output a cardiac health status that can have more than two states. The second cardiac health condition may be differentiated, for example, into different stages of a medical condition / problem, or different risk ranges for having a medical condition / problem, and / or even into different medical conditions / problems. In the case of CHF, the second cardiac health condition may be subdivided into different second sub-stages indicating stages of CHF, such as stage A, stage B, stage C, stage D. Thus, the output of the ML engine 4 indicates not only the presence of CHF, but also its stage. It is further possible that the second cardiac health condition may additionally represent a score indicative of the likelihood of having this medical condition / problem.

[0074] The ML engine 4 can follow simple ML techniques such as random forest models, or more complex ML techniques.

[0075] The ML engine 4 is preferably trained by a supervised training model. The ML engine 4 can operate in a detection mode and in a training mode.

[0076] In detection mode, features calculated for user 10 (without prior knowledge of the user's 10 cardiac health state) are input to ML engine 4 and a cardiac health state of user 10 is output by the trained ML engine 4. At a higher level, IMU measurement signals of user 10 (with an unknown cardiac health state that needs to be estimated) are used to calculate features of user 10 which are input to ML engine 4. Preferably, features are calculated from (only) IMU measurement signals and input to the feature-trained ML engine 4.

[0077] In the training mode, the features calculated for the user 10 (the same features used in the detection mode) are input to the ML engine 4 together with the cardiac health of the user 10, i.e. the features annotated with the actual cardiac health of the user 10 are input to the ML engine 4. This cardiac health of the user 10, annotated on the features, is used for training and may be pre-determined by a diagnosis from a doctor. This diagnosis may be based only on the same features, but preferably on additional measurements such as ECG, echocardiogram, cardiovascular magnetic resonance imaging, phonocardiogram, impedance electrocardiogram, MRT, CT, etc. Preferably, the additional measurements also depend on the estimated cardiac health. For example, if the cardiac output health is the user's health with regard to congestive heart failure, the diagnosis of the user 1 may be based on ECG, blood pressure measurements, and SCG measurements. Other cardiac problems may require other additional measurements. Obviously, the cardiac status of the users for training is based on a computerized estimation (instead of a doctor), but it can also be estimated based on a more complete set of data, which results in a very high certainty of the estimated cardiac status of the users input in the ML engine 4 for training. This training is performed using a plurality of users 10 of at least two kinds: a first kind of users with a first health condition, and a second kind of users with a second health condition. To train the ML engine 4, the same features that are usually used as input for the ML engine 4 are used for the output, i.e. the users 10 whose cardiac health conditions are known. To train the ML engine 4, this known cardiac health condition is also input to the ML engine 4 together with the features of the users 10. This is done in order to obtain a large number of users 10 to obtain a well-trained ML engine 4.

[0078] Preferably, steps S3 and S4 are repeated for multiple heart beats such that the cardiac health state is estimated for multiple heart beats. The cardiac health state results of multiple heart beats can be combined to obtain a combined cardiac health state. This combination can be, for example, counting the number of first cardiac health states and the number of second cardiac health states, and if the percentage of the second cardiac health state is above a threshold, the second cardiac health state is given as the combined cardiac health state. If a score is output together with the second cardiac health state, this score can be used to improve the combination of different cardiac cycles, for example by weighted averaging. Step S3 does not have to be repeated completely for every cardiac cycle. Steps S31-S34 may be performed once for a complete measurement window, after which only steps S35 and S36 are repeated for each cardiac cycle that needs to be analyzed using the ML engine 4.

[0079] In a preferred embodiment, the feature extractor 3 and the ML engine 4 are located on a server 6 or on a remote processor. Preferably, the local processor or smartphone 5 transmits the measurement signal SCG received from the IMU 1 to the server 6 or remote processor (via a communication connection such as the Internet) for further processing (e.g. steps S31-S36 and S4). The output of the ML engine 4, i.e. the cardiac health status, is preferably sent back to the local processor or smartphone 5 and output to the user 10 via the output means of the smartphone 5. Obviously, it is also possible for the cardiac health status to be sent elsewhere, e.g. to a medical professional (who requested the measurement), to the user's account, or something else. However, it is also possible that some or all of these steps are performed on the local processor or smartphone 5. If all steps are performed on the smartphone 5 or on the local processor, there is no data confidentiality issue and the output from the ML engine 4 should not be sent back to the smartphone 5 or anywhere else.

[0080] It should be understood that the invention is not limited to the described embodiments and that modifications can be applied without departing from the scope of the claims.

Claims

1. A method for detecting cardiac health of a user (10), comprising: receiving (S1) in a processing means (2) measurement signals (17, 18) from an IMU (1) placed on the body of the user (10); In the processing means (2), a plurality of features (K ｌｉｎ，ｉ , K ｒｏｔ，ｉ ) (S3), In the machine learning engine (4) of the processing means (2), ｌｉｎ，ｉ , K ｒｏｔ，ｉ and (S4) determining the health status of the heart based on the measured heart rate.

2. The IMU (1) is placed in a smartphone (5) or a wearable, and the measurement signal includes a six-dimensional signal having three linear dimensions (17) representing the linear acceleration of the heart measured using an accelerometer of the IMU (1) and three rotational dimensions (18) representing the angular velocity of the heart measured using a gyroscope of the IMU (1), and a cardiac cycle of the heart of the user (10) is identified based on the measurement signal, and a plurality of sub-periods (24) in the cardiac cycle are determined, and the plurality of features (K) are calculated. ｌｉｎ，ｉ , K ｒｏｔ，ｉ ) is a linear feature (K ｌｉｎ，ｉ ) and rotation features (K ｒｏｔ，ｉ ), and the linear feature (K ｌｉｎ，ｉ ) includes the value of at least one linear parameter of the cardiac motion in each partial period (24) calculated based on the three linear dimensions (17) of the measurement signal, and the rotational feature (K ｒｏｔ，ｉ 2. The method of claim 1, wherein the rotational parameters of the heart motion for each sub-period are calculated based on the three rotational dimensions of the measurement signal.

3. The at least one straight line feature variable (Klin,i,K ｒｏｔ，ｉ 3. The method of claim 2, wherein the at least one rotational characteristic variable is one or more of a linear force, a linear kinetic energy, and a linear work, and the at least one rotational characteristic variable is one or more of a rotational force, a rotational kinetic energy, and a rotational work.

4. a linear velocity signal is calculated based on the three linear dimensions, and the at least one linear characteristic variable is calculated based on the linear velocity signal; and / or The method of claim 2 , wherein a rotational acceleration signal is calculated based on the three rotational dimensions, and the at least one rotational parameter comprises at least one rotational feature calculated based on the rotational acceleration signal.

5. 3. The method of claim 2, wherein a reference point within the cardiac cycle is determined, and the sub-period (24) is defined based on the reference point and based on a fixed time duration of the sub-period.

6. A cardiac cycle of the heart of the user (10) is identified, a plurality of sub-periods (24) in the cardiac cycle are determined, and the plurality of features (Klin,i,K ｒｏｔ，ｉ 2. The method of claim 1, wherein the cardiac cycle includes a value of at least one characteristic variable in each determined sub-period, a reference point within the cardiac cycle is determined, and the sub-period is defined based on the reference point and on a fixed time width of the sub-period.

7. 7. The method of claim 6, wherein the fixed time span can be an absolute fixed time span or a relative fixed time span, the absolute fixed time span being a fixed absolute time value that does not change from one cardiac cycle to another, and the relative fixed time span of the partial period (24) being a relative value with respect to the cardiac cycle that does not change from one cardiac cycle to another.

8. The method of claim 5 , wherein the fixed time duration is defined independently of the physiological time instant of the heartbeat.

9. 3. The method of claim 2, wherein the cardiac cycles in the measurement signal are identified based on a matrix profile motif algorithm applied to a signal based on the measurement signal, and the matrix profile motif algorithm identifies one or more reference cardiac cycles as motifs in the signal based on the matrix profile.

10. The method of claim 1 , wherein the machine learning engine (4) determines whether the heart exhibits congestive heart failure.

11. The method according to any one of claims 2 to 9, wherein respiratory information of the user (10) is detected based on the measurement signal, and the plurality of features includes respiratory information of the user (10).

12. The method of claim 11 , wherein the machine learning engine (4) determines whether the heart exhibits congestive heart failure.

13. The method of claim 1 , wherein the plurality of features are determined independently from ECG measurements.

14. 10. The method of claim 1, comprising the steps of: placing the IMU (1) on the chest of the user (10); and measuring the measurement signal using the IMU (1) when placed on the chest of the user (10).

15. 15. The method of claim 14, wherein the IMU is included in either a smartphone or a wearable connected to the smartphone, and the processing means requests a type of the smartphone or wearable from an operating system running on the smartphone or requests user input regarding the type of the smartphone or wearable, and the plurality of characteristics depend on the type of smartphone or wearable received in return.

16. 16. The method of claim 15, wherein the user (10) lies down while a smartphone including the IMU (1) is placed on the chest of the user (10).

17. A system for detecting cardiac health of a user (10), comprising: an IMU (1) configured to measure measurement signals (17, 18) when placed on the body of the user (10); and processing means (2) comprising a machine learning engine (4) configured to perform the steps of the method according to any one of claims 1 to 9 based on the measurement signals (17, 18) received from the IMU (1).

18. 18. The system of claim 17, wherein the system comprises a smartphone (5) and a server (6), the IMU (1) is included in either the smartphone (5) or a wearable (7) connected to the smartphone (5), an application program running on a processor of the smartphone (5), the application program configured to connect with the server (6), and the processor of the smartphone (5) and / or the processor of the server (6) function as the processing means.

19. A computer program comprising instructions arranged to carry out the steps of the method according to any one of claims 1 to 9 when executed on a processing means.

Images