Apparatus, system, and method for determining cardiovascular health information of a subject.

Abstract

The present invention relates to an apparatus, a system and a method for determining health information related to a subject's cardiovascular system. To improve the accuracy of the determined health information, the apparatus comprises an HR input 31 configured to acquire a time-dependent heart rate (HR)-related signal allowing to represent or derive a heart rate of the subject over time, an event selection input 32 configured to acquire an event selection signal allowing to detect an exercise event of the subject, a processing unit 33 configured to determine health information related to the subject's cardiovascular system from the acquired HR-related signal and the acquired event selection signal, and an output 34 configured to output the determined health information. The processing unit 33 is configured to detect the exercise events of the subject based on the event selection signal, extract time segments of the HR-related signal, each time segment covering a different period including a respective exercise event, combine the extracted time segments to calculate an aggregated HR response signal, and extract one or more features from the aggregated HR response signal to determine health information related to the subject's cardiovascular system.

Description

TECHNICAL FIELD

[0001] The present invention relates to, for example, an apparatus, a system and a method for determining health information related to the cardiovascular system of a subject, for determining the so-called heart age (HeartAge) of the subject as an indicator of the health of the cardiovascular system of the subject. BACKGROUND ART

[0002] In recent years, people track their sleep, nutrition, overall health, and many other aspects of life through the use of apps and wearable devices. Cardiovascular monitoring based on, for example, electrocardiogram (ECG) is currently successfully used for disease identification, stress analysis, prognosis, biometric identification and the like. However, ECG requires arranging a plurality of sensors at specific positions, which is inefficient for wearable devices. Photoplethysmography (PPG) sensors solve this problem by measuring heart rate (HR) and heart rate variability (HRV) using a pulse oximeter that emits light onto the skin and measures the intensity of reflected light. This intensity depends on blood flow under the skin which is regulated by the cardiac cycle. Measuring the periodicity in reflection provides estimates of HR and HRV. The fact that this measurement can be performed on various parts of the body, such as the wrist, fingertip or earlobe, makes PPG sensors convenient in daily life and allows them to be integrated into wearable devices such as smart watches. Furthermore, it has been found that PPG sensors can measure changes in the physical and mental state of the body only by monitoring the heart. In addition, it has been found that remote PPG can detect vital signs such as HR, respiratory rate and SpO2 from camera signals of a skin region of a subject.

[0003] Obtaining a clean PPG signal is the biggest challenge in wearable devices, as it involves unpredictable artifacts. Therefore, most research and applications are conducted in controlled environments with minimal physical movement. PPG signal windows contaminated by movement noise are either ignored or filtered out.

[0004] Despite the immense potential of PPG use in monitoring applications, PPG still suffers from numerous problems related to noise and unreliable measurements. These artifacts hinder the use of PPG as a clinical monitoring device.

[0005] U.S. Patent Application Publication No. 2018 / 0249951(A1) discloses a device for accurately estimating a person's energy expenditure, particularly taking into account the effects of cardiovascular drift. The device comprises an input unit for acquiring an exercise signal representing a person's physical activity and a heart rate signal representing a person's heart rate; a cardiovascular drift determination unit for determining a cardiovascular drift phase from the exercise signal, the heart rate signal, and / or one or more cardiovascular drift-related signals carrying information about one or more of a person's sweating, weight loss, body temperature rise, blood lactate concentration, and physical fatigue; a correction unit for correcting the generated heart rate signal representing the heart rate during the cardiovascular drift phase; and an estimation unit for estimating a person's energy expenditure from the corrected heart rate signal. [Overview of the initiative] [Problems that the invention aims to solve]

[0006] The object of the present invention is to provide an apparatus, system, and method for determining health information related to the cardiovascular system of a subject, which preferably operates in an inconspicuous manner and can optionally be used while the subject is resting or sleeping. [Means for solving the problem]

[0007] In a first aspect of the present invention, a device is presented for determining health information related to the cardiovascular system of a subject, the device is An HR input unit configured to acquire a time-dependent heart rate (HR) related signal that enables the representation or derivation of a subject's heart rate over time, An event selection input unit configured to acquire an event selection signal that enables the detection of a subject's motor events, A processing unit configured to determine a subject's cardiovascular health information from acquired HR-related signals and acquired event selection signals, It has an output unit configured to output the determined health information, The processing unit is Based on the event selection signal, the system detects the subject's motor events. Extract time segments of HR-related signals, each covering a different period that includes the respective exercise event. The extracted time segments are combined to calculate the aggregated HR response signal. It is configured to extract one or more features from aggregated HR response signals to determine health information related to the subject's cardiovascular system.

[0008] In a further aspect of the present invention, a system is presented for determining health information related to the cardiovascular system of a subject, and this system is An HR sensor configured to sense time-dependent heart rate (HR) related signals that enable the representation or derivation of a subject's heart rate over time, A selection signal sensor configured to sense an event selection signal that enables the detection of a subject's motor events, Apparatus disclosed herein for determining a subject's cardiovascular health information from sensed HR-related signals and sensed event selection signals, It has an output interface configured to issue determined health information.

[0009] Further aspects of the present invention provide a computer program comprising a corresponding method, a program code means for causing a computer to perform steps of the method disclosed herein when executed on a computer, and a non-transient computer-readable recording medium storing the computer program for causing a processor to perform the method disclosed herein when executed by a processor. The method can be carried out on a computer.

[0010] Preferred embodiments of the present invention are defined in the dependent claims. It is understood that the claimed methods, systems, computer programs and media have preferred embodiments similar to and / or identical to the claimed systems, particularly as defined in the dependent claims and disclosed herein.

[0011] The present invention is based on the idea of ​​focusing on time segments of HR-related signals (e.g., PPG signals) that are "contaminated" with exercise-based noise. Thus, in contrast to known methods for patient monitoring, the subject's exercise events are detected, and the corresponding time segments of the HR-related signals covering the period containing each exercise event are used for subsequent processing and for determining information about the subject's cardiovascular system.

[0012] Therefore, the present invention overcomes the technical limitations of known monitoring devices, systems, and methods and presents a solution that enables the unobtrusive determination of a subject's cardiovascular vitality by monitoring HR responsiveness, for example, after exercise at night or in other situations, such as when a subject's exercise during the day affects their HR (e.g., when getting out of their car).

[0013] Since this cardiovascular vitality (i.e., determined health information) often correlates with the subject's actual age, this new physiological measure is called “HeartAge,” which may be related to the subject's biological age. Therefore, it can function as a powerful single value as “HeartAge,” i.e., the “age” of the subject’s cardiovascular system in the sense of an indicator of the adaptability / flexibility of the subject’s cardiovascular system. By continuously monitoring the subject’s “HeartAge” without the presence of a healthcare professional (e.g., a nurse or doctor), it may be possible to determine and / or signal the risk to various cardiovascular problems at a very early stage. Thus, the present invention could save lives and greatly contribute to healthcare in the near future. In general, the present invention can be used to determine other parts of health information related to the subject’s cardiovascular system, which can be presented in terms of, for example, weight (“HeartWeight”), BMI (“HeartBMI”), vitality (“HeartVitality”), well-being (“HeartHappiness”), stress levels (“HeartStress”), etc.

[0014] In this context, "detection" of an exercise event means not only actually measuring or identifying an exercise event, but also predicting or estimating one. For example, a time segment of a relevant HR signal can be initially extracted after a certain exercise event, and then this time segment can be temporally extended to include that exercise event.

[0015] In a preferred embodiment, the processing unit is configured to align the time segments extracted before joining based on an event selection signal or an acquired alignment signal indicating the time information of the motor events within each time segment. This further improves the accuracy of determining health information. Generally, the event selection signal can also be used for alignment; that is, the event selection signal can be used not only to select motor events but also to align the extracted time segments, although a separate alignment signal may also be used. For example, an accelerometer (ACC) signal from a body-worn ACC sensor can be used as both an alignment signal and an event selection signal. In another embodiment, an ACC signal can be used as an alignment signal, and a non-ACC signal (e.g., a weight or pressure sensor signal sensing the subject's movement or change in posture) can be used as an event selection signal. In yet another embodiment, an ACC signal can be used as an event selection signal, and a non-ACC signal can be used as an alignment signal.

[0016] The processing unit may be configured to detect motion events by, advantageously, detecting motions of the subject that affect the subject's HR, particularly repetitive motions of the subject, or more frequently occurring motions such as changes in the subject's posture or rotational movements of the subject. One or more sensor signals from one or more body-worn sensors (e.g., ACC sensors) and / or external sensors (e.g., cameras or pressure sensors) may be used for this purpose.

[0017] Preferably, the processing unit can be configured to detect motion events by detecting one or more motion features of the subject's movement, the motion features including motion intensity exceeding an intensity threshold, motion pattern, motion velocity, motion direction, and motion distance.

[0018] In a preferred embodiment, the processing unit is configured to detect, within the selected time segments, one or more unwanted time segments in which the HR-related signal is distorted or, in particular, based on one or more features of each selected time segment, in which the HR is unaffected despite the detection of an exercise event, and to discard the detected unwanted time segments when synthesizing the extracted time segments. This further improves the accuracy of determining health information. In addition to time segments in which the HR is distorted, time segments in which the HR is unaffected (despite the detection of an exercise event) are also removed and not used for further processing to determine the desired health information. The HR may not be affected by the subject's exercise if the exercise is falsely detected due to, for example, signal noise or other sources of distortion. These errors are corrected retrospectively by removing time segments in which the HR remains (substantially or completely) constant, based on the primary assumption that exercise causes changes in HR.

[0019] In another embodiment, the processing unit is configured to extract a time segment of the HR-related signal by determining a characteristic time point in which a specific feature of the detected event appears, and to select time segments before and after the characteristic time point so as to extend a first time period before the characteristic time point and a second time period after the characteristic time point. The characteristic time point may be, for example, the time when a change in the subject's position or posture actually occurs. This characteristic time point may be recognized, for example, in the event selection signal and / or the HR-related signal. Then, a period before (or only after) the characteristic time point can be selected from the HR-related signal as a time segment. This period may be predetermined and fixed, or it may be individually set or selected by the user or the processing unit. A reasonable value for this period may be in the range of 10 to 240 seconds, particularly in the range of 30 to 120 seconds. The actual value may depend on the application of the present invention and may differ from those values, i.e., the actual value may generally be smaller or larger than the minimum and maximum values ​​mentioned.

[0020] The processing unit may be further configured, for the purpose of combining the extracted time segments, to discard the time period in each selected time segment in which a detected motion event occurs, and use only the time period in each selected time segment that is before and / or after the detected motion event. This contributes to a further improvement in accuracy, since it is known that at the moment of a motion event, the HR-related signal may contain noise or be distorted, that is, motion may cause artifacts in the HR-related signal, as a result of which this portion of the HR-related signal should be discarded, and only the portions of the HR-related signal immediately before and / or immediately after the motion event should be used as time segments.

[0021] In general, different features of the aggregated HR response signal can be used to determine desired health information. In one embodiment, the processing unit may be configured to extract one or more of the following as one or more features from the aggregated HR response signal: - the difference between a maximum value and a minimum value of the aggregated HR response signal; - the difference between a maximum value of the aggregated HR response signal and a starting value; - the difference between an ending value and a starting value of the aggregated HR response signal; - the time difference between occurrences of a minimum value and a maximum value of the aggregated HR response signal; - the time difference between occurrences of a starting value and an ending value of the aggregated HR response signal; - a maximum slope of the aggregated HR response signal; and - a duration and / or a rate of increase of the aggregated HR response signal.

[0022] These features generally have the goal of describing the aggregated heart rate response with a single numerical value. Accordingly, one or more of these features (or alternatively also additional features such as the area under the aggregated HR response signal, the amount of heartbeats before returning to normal) can be used to serve this purpose.

[0023] Preferably, the processing unit can be configured to determine health information related to the subject's cardiovascular system by using a predetermined function, in particular a linear or nonlinear approximation that applies one or more predetermined scalar parameters. The scalar parameters may be predetermined for various types of subjects (e.g., based on age, sex, size, weight, health status, fitness level, etc.) or may be set or determined individually for a specific subject during the examination.

[0024] In one embodiment, the processing unit may be configured to determine a scalar parameter from one or more relationships between the subject's age and one or more features from the aggregated HR response signals. Such a determination is preferably made in advance based on empirical data that enables the identification or derivation of relationships between age and one or more features of the aggregated HR response signals that should be used to make actual determinations about the subject's health information.

[0025] In general, various signals can be used as HR-related signals. Preferably, one of the following is used in actual implementation: contact or remote photoplethysmography (PPG) signal, HR signal, ECG signal, cardiometer signal, intraocular pressure measurement signal, or holographic laser Doppler imaging signal.

[0026] Regarding event selection signals, various options exist, for example, based on which sensors can be used or which signals are available in a particular application scenario. In practice, event selection signals can be one of the following: motion signals indicating the subject's body movement, accelerometer signals, muscle activity signals, brain activity signals, pressure sensor signals, weight sensor signals, vehicle CAN (Controller Area Network) bus signals (or other bus signals), and camera signals.

[0027] Similarly, various options exist for alignment signals, depending on which sensors can be used or which signals can be utilized in a particular application scenario. In practice, alignment signals can be one of the following: motion signals indicating the subject's body movement, accelerometer signals, muscle activity signals, brain activity signals, pressure sensor signals, weight sensor signals, vehicle CAN bus signals (or other bus signals), and camera signals.

[0028] Several options exist for synthesizing (e.g., averaging) multiple time segments of HR-related signals into a final aggregated HR response signal.

[0029] In a preferred embodiment, the processing unit is configured to synthesize the extracted time segments by first converting the extracted time segments into signals representing frequencies over time (e.g., using wavelet transform), and then aggregating these signals to obtain an aggregated HR response signal; that is, the HR-related signals are first converted into signals representing frequencies over time, and then synthesized by aggregating the extracted time segments. In another embodiment, the processing unit is configured to synthesize the aggregated time segments by first converting the extracted time segments into time-frequency matrices, in particular by synthesizing the matrices pixel by pixel, and then taking the maximum edge to obtain an aggregated HR response signal; that is, the corresponding time-frequency matrices are synthesized, and then the maximum edge is obtained as an aggregated HR response signal. [Brief explanation of the drawing]

[0030] These and other aspects of the present invention will become apparent from and be described with reference to the embodiments described below. [Figure 1] A schematic diagram of an embodiment of the system according to the present invention. [Figure 2] A schematic diagram of an embodiment of the apparatus according to the present invention. [Figure 3] A schematic diagram of a first embodiment of the method according to the present invention. [Figure 4] A schematic diagram of a second embodiment of the method according to the present invention. [Figure 5] A diagram showing the time-frequency matrix. [Figure 6] A diagram showing the summation and averaging of time segments of HR-related signals. [Figure 7] A figure showing that the aggregated HR response signal remained stable over multiple nights. [Figure 8] A diagram showing the relationship between the intensity of the aggregated HR response signal and the age of the subject. [Modes for carrying out the invention]

[0031] Figure 1 shows a schematic diagram of an embodiment of System 1 according to the present invention for determining health information related to the cardiovascular system of a subject. While the subject can generally be any human being, the present invention may find specific applications in monitoring patients, the elderly, people with significant cardiovascular issues, athletes, and others.

[0032] System 1 includes a heart rate (HR) sensor 10 configured to detect (i.e., acquire or measure) a time-dependent HR-related signal that enables the representation or derivation of a subject's HR over time. The HR sensor 10 may be a wearable sensor such as a contact photoplethysmography (PPG) sensor, an HR rate sensor, or an ECG sensor. In other embodiments, the HR sensor may be an external sensor such as a remote PPG sensor, an electrocardiogram sensor, an intraocular pressure measurement sensor, or a holographic laser Doppler imaging sensor. Thus, the HR-related signal can, for example, directly indicate the subject's HR over time in the case of a contact PPG signal or an HR rate sensor, or, for example, enable the derivation of the subject's HR in the case of a remote PPG signal or an electrocardiogram signal.

[0033] System 1 further includes a selection signal sensor 20 configured to sense event selection signals, which enable the detection of motion events of a subject. The selection signal sensor 20 may be any wearable or external sensor that provides a signal that can be detected when a subject is performing a movement or attempting to perform a movement, particularly a movement that affects the subject's HR. The signal may directly indicate a movement or allow for the derivation that there is or will be a movement of the subject. Thus, a motion event may be any event in which the subject changes position and / or posture, particularly to some extent or in a certain manner, such that, as a result of the subject's movement, the subject's HR shows a potentially transient change. In a preferred embodiment, the selection signal sensor 20 may be a motion sensor, accelerometer, muscle activity sensor, brain activity sensor, pressure sensor, weight sensor, CAN bus, or camera that provides a motion signal indicating the movement of the subject's body.

[0034] System 1 further includes a device 30 configured to determine a subject's cardiovascular health information from sensed HR-related signals and sensed event selection signals. Details of the device 30 will be described later. The device 30 may be implemented, for example, in software and / or hardware, as a processor or computer, or within one.

[0035] System 1 further includes an output interface 40 configured to output the determined health information. The output interface 40 can generally be any means of outputting the determined health information visually or audibly, for example, in text form, as an image or diagram, as audio or spoken words, etc. For example, the output interface 40 may be a display, a loudspeaker, a touchscreen, a computer monitor, a smartphone or tablet screen, etc.

[0036] Optionally, system 1 may further have an alignment signal sensor 50 configured to sense an alignment signal that enables the alignment of time segments extracted from HR-related signals before they are synthesized. For this alignment, an event selection signal may be used (in which case the alignment signal sensor 50 is omitted). Otherwise, separate rough surfaces and signals can be obtained that indicate the time information of the motion events within each time segment, and the time segments are temporally aligned based on this time information.

[0037] Figure 2 shows a schematic diagram of an embodiment of the apparatus 30 for determining health information related to the cardiovascular system of a subject according to the present invention.

[0038] The device 30 includes an HR input unit 31 configured to acquire time-dependent HR-related signals and an event selection input unit 32 configured to acquire event selection signals. Optionally, an additional alignment signal input unit 35 is provided for acquiring alignment signals when available. The HR input unit 31 and the event selection input unit 32 (and the optional alignment signal input unit 35) can be directly coupled to or connected to the HR sensor 10 and the event signal sensor 20 (and the optional alignment signal sensor 50), respectively, or these signals can be acquired (i.e., read out or received) from a storage device, buffer, network, or bus, etc. Thus, the input units 31, 32, and 35 can be (wired or wireless) communication interfaces or data interfaces, such as a Bluetooth® interface, WiFi interface, LAN interface, HDMI interface, direct cable connection, or any other suitable interface enabling signal transmission to the device 30.

[0039] The device 30 further includes a processing unit 33 configured to determine health information related to the subject's cardiovascular system from acquired HR-related signals and acquired event selection signals. The processing unit 33 is any kind of means configured to process signals and determine health information therefrom. It can be implemented in software and / or hardware, for example, as a programmed processor or computer or application on a user device such as a smartphone, smartwatch, tablet, laptop, PC, or workstation.

[0040] The device 30 further comprises an output unit 34 configured to output the determined health information. The output unit 34 can generally be any interface that provides the determined health information, for example, by transmitting it to another device or making it available for retrieval by another device (e.g., a smartphone, computer, tablet, etc.). Therefore, it can generally be any (wired or wireless) communication or data interface.

[0041] Figure 3 shows a schematic diagram of a first embodiment of method 100 for determining a subject's cardiovascular health information according to the present invention. The steps of the method can be performed by an apparatus, and the main steps of the method are performed by a processing unit. The method can be implemented as a computer program that runs on a computer or processor.

[0042] In the first step 101, a time-dependent HR-related signal is acquired, for example, by being extracted or received from an HR sensor 10. In the second step 102, which is performed before, after, or concurrently with step 101, an event selection signal is acquired, for example, by being extracted or received from an event signal sensor 20. Subsequently, in step 103, health information related to the subject's cardiovascular system is determined from the acquired HR-related signal and the acquired event selection signal. Finally, in step 104, the determined health information is output.

[0043] Step 103, which determines health information, includes a first step 1031 in which exercise events of the subject are detected based on an event selection signal. For each exercise event, in step 1032, a time segment is extracted from the HR-related signal, which is a time segment covering the period in which each exercise event is contained. In step 1033, the extracted time segments are combined (e.g., averaged) to calculate an aggregated HR response signal. From the aggregated HR response signal, in step 1034, one or more features are extracted, and the extracted one or more features are used to determine health information related to the subject's cardiovascular system.

[0044] In another embodiment, the apparatus 30 of the present invention may comprise the following three modules: i) A recording module (including or commonly represented by inputs 31, 32, and 35) for recording PPG signals (as HR-related signals), selection signals, and alignment signals; ii) A processing module (representing processing unit 33) that synthesizes these signals, extracts characteristic aggregated heart rate responses (AHRR), and determines health information from them; iii) A feedback module that summarizes and reports the results (representing output unit 34).

[0045] According to one embodiment, the processing module (or processing unit) extracts the AHRR as follows: i) Preprocess the PPG signal and the selected signal using a bandpass filter; ii) Select relevant motor events using selection signals; iii) Denoise and extract heart rate curves from PPG signals during these exercise events; iv) Detect and remove severely distorted heart rate curves; v) Align and average all remaining heart rate curves using the alignment signal to obtain the AHRR.

[0046] The resulting AHRR is shown to be stable for each individual. This is used to estimate a person's health information, such as their HeartAge. These steps are explained in more detail below.

[0047] The feedback module preferably summarizes and reports all results over a specified period. The feedback can be given to the user or to healthcare professionals such as doctors or nurses. The feedback can be provided in several forms (e.g., text and video).

[0048] In one embodiment, the algorithm for extracting AHRR takes a PPG signal (as an HR-related signal) and a selection signal as inputs. A suitable signal can be collected during sleep, as the heart rate is lower and more stable compared to daytime. A person makes 20 to 80 small (turning over) movements during sleep. These cause a typical transient increase in heart rate (possibly related to the so-called cardiovascular orthostatic reflex). An acceleration signal can be used as a selection signal to detect the onset of turning over movements. Figure 4 shows a schematic diagram of a second embodiment of Method 200 according to the present invention using such an algorithm. This embodiment specifically performs a series of steps to remove noise from the signal and determine the AHRR.

[0049] First, in step 201, preprocessing is performed on the first derivative of the PPG signal using, for example, a bandpass third-order Butterworth filter with a cutoff frequency of 0.5–2 Hz. The first derivative is used to offset baseline fluctuations and facilitate peak detection. The 0.5–2 Hz cutoff frequency corresponds to the expected HR range of 30–120 BPM (Beats Per Minute), enabling the extraction of the fundamental frequency of HR. Furthermore, both the selected signal and the PPG signal may be normalized to a signal-dependent constant range using a sliding time window around the sample extending to [-0.5, 0.5] seconds. The signal may be zero-padded at the beginning and end to ensure correct normalization of the first and last samples. To handle large spikes and negative portions in the signal, x / (max-min) normalization may be used to ensure a constant positive signal range.

[0050] Secondly, a sleep-related toss-and-turn event is detected (as an example of a “motor event”) (step 202), and the PPG signal is segmented into multiple windows of interest (step 203). These event windows (“time segments”) are defined as specific time intervals during which each event occurs. A motor event can generally be any event that triggers a sudden HR response, i.e., a change in HR. Events may be selected to have a clear start, a clear end, and a substantial pause between them. The start and end of a toss-and-turn event window can be defined based on a temporary increase in the acceleration signal above a predetermined threshold, and for each event, the end of the acceleration increase may be used. The start of an event window can be defined as a first time period before the end of the acceleration increase (e.g., a range of 10-90 seconds before the acceleration increase, e.g., the 40 seconds prior), and the end of an event window can be defined as another second time period after the end of the acceleration increase (e.g., a range of 10-90 seconds after the acceleration increase, e.g., the 60 seconds following). In this way, the acceleration signal (used as a selection signal) can also function as an alignment signal to align different event windows before further processing. Other feature points (different from the end of the acceleration increase) can also be used to define the event windows, as well as the first and second periods of the event windows described above.

[0051] Thirdly, the PPG signal between the windows of interest is transformed into a time-frequency matrix (e.g., a continuous wavelet transform matrix) as shown in Figure 5 (step 304), where each column describes the frequency spectrum at time t. The PPG signal during the event is preferably ignored because it is often too distorted at the moment of the event. This matrix is ​​then compressed by, for example, a (samplefreq x 1) maximal pool filter. The resulting matrix has a resolution of one column per second and is called the heart rate response (HRR). The vertical position of the maximum value in these columns corresponds to the frequency. These frequencies can be extracted to obtain the maximal ridgeline (MR) describing the HRR in BPM, as shown in the lower part of Figure 5.

[0052] Since the PPG signal is highly sensitive to distortion, the maximum pixel value in the time-frequency matrix sometimes corresponds to a frequency caused by noise rather than a frequency caused by HR. Therefore, in step 205, anomalies can be detected by subtracting the filtered MR, e.g., the 9s median filtered MR, from the original MR. The residuals are compared to a threshold to find large deviations indicating anomalous behavior. An asymmetric threshold can be used to favor high HR anomalies. Changes in HR tend to be asymptotic near the minimum and maximum HRs. Therefore, it is expected that the resting HR is more likely to increase than decrease. Residuals greater than 30 percent or less than 10 percent of the previous HR may be omitted from further processing. The resulting gaps shorter than the mean event duration may be linearly interpolated.

[0053] This embodiment allows for the removal of erroneous measurements (outliers). The number of outliers can provide clues about HRR quality. However, it is not the only measure to be used. The process of removing overly distorted HRRs can include two steps: in the first step, features of the HRR signal are measured, and in the second step, a selection is made based on those features whether to remove the HRR curve. For example, in the first step, several features are extracted from the HRR curve. These features describe the morphology, stability, variance, number of bad measurements, etc. Essentially, these features describe all the information a person needs to make their own decision about HRR quality. However, while a person may be asked to make the decision, it is preferable not to be asked to make the decision. Instead, an algorithm can be trained to make this decision automatically. For this purpose, in the second step, after the features have been extracted, the algorithm is trained to determine whether the HRR curve has good quality or whether the HRR curve is overly distorted to be used for anything. Too much distortion means that meaningful HRRs cannot be extracted. As described above, this binary decision task can be performed using a random forest algorithm. Such a random forest algorithm can be trained using multiple (e.g., hundreds) manually labeled HRRs. This labeling and training of the algorithm can be done once for many applications, or especially for any application, and can be considered a kind of calibration. These HRRs are manually labeled as good or poor quality. Based on these HRR curves and all extracted features, the random forest algorithm learns to detect whether the quality of the curves is good or poor. However, other classifiers may be used instead of the random forest algorithm.

[0054] Next, in step 206, the interpolated HRR is analyzed and temporal and morphological features are computed to serve as a feature space for the discrimination algorithm. The discrimination algorithm (e.g., a random forest classifier with a maximum depth of, for example, 10 levels and containing, for example, 100 trees) learns to detect and remove HRRs that do not describe the actual heart rate response (e.g., are too flat) or are too skewed for further processing.

[0055] Finally, the remaining HRR signals from multiple (preferably all) event windows are normalized and aligned (step 207), then synthesized together (e.g., averaged) (step 208) to highlight recurrence patterns, thereby obtaining the AHRR. This is shown in Figure 6, which illustrates multiple HRRs and the resulting AHRR. Since the event windows are aligned, combining separate event windows can reveal recurring HR behavior and cancel out random errors and variability. As described above, HRRs can be aligned by event end time based on a selection signal where the primary interest is on HR behavior following the event. As a result, the AHRR characteristics were found to be very stable from person to person over several nights (intraclass correlation = 0.7-0.8). Then, for example, features can be extracted from these AHRRs obtained over several nights to determine subject health information related to the subject's cardiovascular system, e.g., arterial vitality and state, also referred to herein as HeartAge (step 209).

[0056] The use of contextual alignment in the noise reduction process enables detailed HR behavior analysis at a resolution previously unseen in wearable applications. This increased resolution opens up new possibilities for discreet and continuous monitoring of vitality and HeartAge during the night. A more detailed example (HeartAge example) is described below.

[0057] As mentioned above, many different signals can be used as selection signals and / or alignment signals (e.g., torso acceleration and leg muscle activity). Data may be acquired in full or partial nighttime unoccupied home recordings. However, in other situations, data may be acquired, for example, in an office where a person repeatedly changes posture between sitting and standing, when a person is performing sports with exercise or changes in posture, or when a person leaves a vehicle (i.e., standing up from a seat and leaving the vehicle). The fact that multiple sensors can be used and that AHRR was stable for each person indicates that the determination and use of AHRR is a versatile, robust, accurate, and stable new measurement technique.

[0058] Alignment of time segments of various HRR or HR-related signals can also be done in other ways, for example, by using cross-correlation. The HRR or time segment itself can then function as the alignment signal. For example, two HRRs can be correlated with each other over the entire range of delays, and the delays that produce the highest correlation are selected. This is done for all pairs of two HRR curves (e.g., all pairs of two HRR curves for the night), and each HRR signal is shifted by its mean delay amount before the procedure is restarted and the AHRR is calculated from the shifted HRR curves. This method also makes it possible to identify different types of HRR curves from the degree of correlation between individual HHR curves: HRR curves that correlate well with each other (for a given delay) belong to the same cluster. For each cluster, a specific AHHR can then be constructed. Perhaps different clusters are associated with different types of nocturnal tossing and turning movements.

[0059] Generally, many movements, such as each turn during sleep, generate a body-specific HRR. Averaging these HRRs yields a more stable AHRR per night, which is characteristic of a particular body and does not change significantly from night to night, as shown in Figure 7, which shows multiple AHRRs acquired over different nights (each plot shows three AHRR curves for a single subject acquired over three nights). Using the AHHR signals calculated above, "HeartAge" can be calculated as follows: HeartAge = a + b*(AHRR increase) In the formula, a and b are preferably empirically determined scalars, and AHRR is the aggregated heart rate response as described above. The increase in AHHR can be determined by subtracting the AHHR at t=0 from the maximum HR in the AHHR signal. This increase in AHRR may be related to the so-called orthostatic reflex, i.e., the increase in HR when changing to an upright posture (see, for example, G. Cybulski and W. Niewiadomski, "Influence of age on the immediate heart rate response to the active orthostatic test," J Physiol Pharmacol, vol. 54, no. 1, pp. 65-80, 2003).

[0060] The above embodiment calculates HeartAge from AHRR increase using linear approximation. This is the simplest method and generally fits well with available data. Given sufficient available data for the relevant situation (e.g., getting out of a car or getting out of a chair), any other function F(x) (where x is a feature of AHRR and F is desired health information such as HeartAge) can generally be fitted to the data, for example, a quadratic function F(x) = a + bx + cx 2 Therefore, in some applications, various functions can be applied to available datasets when data is collected for particularly different situations (such as getting out of a car).

[0061] Figure 8 shows the relationship between the subjects' AHRR increase and their actual age: on average, the older the subject, the lower their AHRR increase. This overall population relationship helps in estimating the linear dependence of HeartAge and parameters a and b. Thus, a person's HeartAge can be considered the age at which the entire population has the same AHRR increase as theirs.

[0062] The fact that AHRR can be extracted at home without a doctor's assistance opens up new possibilities for optimizing physical training or drug therapy. This allows for easy, time-series monitoring of vitality and heart age.

[0063] In Figure 8, the relationship between age and AHRR increase is shown to be determined from a specific dataset. As can be seen from the graph, it follows the linear formula AHRRn = C1 + C2 * age (C1 = 25.5 bpm, C2 = -0.275 bpm / year). Determining HeartAge from AHRR increase follows the inverse function: HeartAge = (AHRR increase - C1) / C2 = a + b*AHRR increase. In other words, the scalars a and b mentioned above in approaches using linear approximation can be determined from the linear equation of the fitted line shown in Figure 8, such that b = 1 / C2 = 3.64 years / bpm and a = -C1 / C2 = 92.7 years. Other scalars used in other approximations to determine HeartAge (or other health information) can be determined using the same or similar methods based on empirical data, based on the relationship between age and one or more features of AHRR predetermined for a larger population. Furthermore, such scalars can be predetermined for specific types of subjects based on typical patient parameters such as age, sex, size, weight, fitness level, fitness status, and medication.

[0064] In general, one or more features can be extracted from AHRR to detect HeartAge. In the embodiments described above, an increase in AHRR is used as a feature, but one or more other features can be used instead or in addition. In the case of a toss-and-turn event during sleep, the posture changes somewhat during tossing and turning, but then returns to a horizontal position. As a result, the HR before tossing and turning is (sufficiently) almost the same as the HR after tossing and turning, and a temporary increase in HR during and immediately after tossing is expected. When getting out of a vehicle, there is a sitting position before the change in posture and a standing position after the change in posture, and it is known that the HR is higher when standing than when sitting. Therefore, a continued increase in HR is expected, but it may be accompanied by an overshoot during the process of standing. In this case, both the continuous increase in HR and the height of the overshoot are parameters that can indicate health status or HeartAge.

[0065] Other features of AHRR, such as the following, can also be used: - The difference between the maximum and minimum values ​​of the aggregated HR response signal. - The difference between the maximum value and the starting value of the aggregated HR response signal. - Difference between the end and start values ​​of the aggregated HR response signal - The time difference between the appearance of the minimum and maximum values ​​of the aggregated HR response signal. - The time difference between the start and end values ​​of the aggregated HR response signal. - The maximum slope of the aggregated HR response signal, and - Duration and / or rate of increase of the aggregated HR response signal.

[0066] In general, only a small window of interest around a motor event needs to be analyzed, making real-time execution of the present invention possible. Motor events must be separated by the duration of at least one window of interest. Since the algorithm consists only of inexpensive computations, this analysis can be performed immediately after the event's window of interest has ended. Extracted time segments of consecutive HRR or HR-related signals can be added to a temporary AHRR and finally divided by the number of events to obtain the final AHRR. This processing pipeline enables real-time calculation of AHRR in remote applications where computational power and battery resources are limited.

[0067] With the introduction of smartwatches, the use of wearable devices has increased exponentially in recent years and is expected to increase further. Wearable wristband devices often connect to smartphones via Wi-Fi or Bluetooth, enabling remote processing of PPG data (or generally HR-related signals) on the smartphone. Since one embodiment of AHRR estimation uses only PPG and accelerometer sensors, AHRR estimation can ideally be performed using a smartwatch and a user-friendly smartphone app. Due to the great interoperability and frequent use of smartphones, the use cases of AHRR in daily life are endless.

[0068] One potential application of the present invention is to measure HeartAge or other health information via a wristband (or any other wearable device) used by a vehicle driver or passenger. Each time the driver leaves the vehicle, sensors in the seat and door register this activity and can therefore function as selection signals. Upon leaving the vehicle, the driver changes from a seated to a standing position, triggering a true orthostatic reflex in the cardiovascular system, and the heart rate increases to adapt to the standing position (and often temporarily overshoots when doing so). These heart rate changes are measured by PPG sensors in the wristband. The seat and door CAN signals can be used as selection signals to select a range of moments in the PPG data during which the driver changes from a seated to an upright position, and the present invention enables the calculation of AHRR and HeartAge for this particular situation. The same applies to a combination of the passenger seat and door and a wristband of the person sitting there.

[0069] Therefore, in such an embodiment, the CAN bus signal of a vehicle with the driver's door open can be used as a selection signal indicating that the driver is expected to change from a seated to a standing position in the next minute or so (i.e., the CAN bus signal can be used to predict future motion events and estimate the timing of motion events), and as a result, it can be assumed that the driver's HR will increase. The ACC signal can then be used as a precise indicator of when the driver stands up, i.e., as an alignment signal. The advantage of such a combination is that the selection signal may be more precise in the correct selection of the appropriate motion event, and the alignment may be better when determining (and aligning) the correct timing of multiple motion events and the relevant time segments of HR-related signals during these motion events.

[0070] In another embodiment, the ACC signal can be used as a selection signal, and the non-ACC signal can be used as an alignment signal. For example, if an activity sensor (e.g., an ACC sensor) is attached to the subject's chest and an EMG sensor (which senses muscle tension; a non-ACC sensor) is attached to the subject's triceps, the activity sensor will detect that the subject has started to move and the subject's chest is moving upward. This makes it possible to select an HR trace that captures the fact that the chest is rising (and an HR response is expected). The EMG sensor on the triceps will indicate the exact moment when the subject begins to use the muscles to stand up, and the EMG sensor signal can be used for alignment.

[0071] In general, the present invention can be used to determine different health information. One example is HeartAge, as described above. Another example is a person's weight as another indicator of a person's health, as is generally the case with age. The primary information obtained by using the present invention is information regarding the flexibility / adaptability of the cardiovascular system. This information regarding the flexibility / adaptability of the cardiovascular system can then be converted back into something that the general public can better understand, such as age, weight, or BMI.

[0072] To determine body weight, a HeartWeight similar to HeartAge can be constructed, providing an index of cardiac adaptability / flexibility by comparing one or more features of AHRR, such as AHHRincrease, with those of other people of varying body weights. Therefore, HeartWeight can tell us that even if a person is overweight (e.g., weighing over 100kg), their cardiac response may be similar to that of a leaner person, meaning their HeartWeight could be only 80kg. This is actually the same health information as HeartAge, namely information about cardiovascular flexibility / adaptability; it would only be presented in a different manner.

[0073] The same applies to other cardiac information such as BMI ("HeartBMI"), vitality ("HeartVitality"), happiness ("HeartHappiness"), and stress levels ("HeartStress").

[0074] This invention enables discreet, walkable, and stable measurement of people's cardiovascular vitality. For example, a person's "HeartAge" can be determined using wearable PPG measurements during sleep. This invention can identify people's movements, such as turning over during sleep, by utilizing different signals, such as the accelerometer signal of a wearable device. Then, a time-frequency transformation method (e.g., continuous wavelet transform) is used to filter out HR changes around these moments. Thus, this invention takes the opposite approach to conventional practice, utilizing event detection signals, such as the accelerometer signal of a wearable device, to extract specific signal windows of HR-related signals, which would otherwise be omitted as motion-based noise.

[0075] This invention can find specific applications in home health and wellness monitoring, as well as in sleep and respiratory care. Furthermore, the invention provides an easily diagnoseable means. Another application area is passenger health in automotive applications.

[0076] Although the present invention has been illustrated and described in detail in the drawings and the foregoing description, such illustrations and descriptions should be considered descriptive or illustrative and not limiting. The present invention is not limited to the disclosed embodiments. Other variations of the disclosed embodiments can be understood and practiced by those skilled in the art in carrying out the claimed invention, from the examination of the drawings, disclosure and the appended claims.

[0077] In the claims, the word “have” does not preclude other elements or steps, and the indefinite article “a” or “an” does not preclude plurality. A single element or other unit may fulfill the function of several items enumerated in the claims. The mere fact that certain means are described in different dependent claims does not imply that combinations of these means cannot be used advantageously.

[0078] Computer programs can be stored / distributed on suitable non-transient media such as optical or solid-state media supplied together with or as part of other hardware, but they can also be distributed in other forms, such as via the Internet or other wired or wireless communication systems.

[0079] No reference numeral in a claim should be construed as limiting in scope.

Claims

1. A device for determining the cardiovascular health information of a subject, An HR input unit configured to acquire a time-dependent heart rate (HR) related signal that enables the representation or derivation of the subject's heart rate over time, An event selection input unit configured to acquire an event selection signal that enables the detection of the subject's motor events, A processing unit configured to determine the health information relating to the subject's cardiovascular system from the acquired HR-related signals and the acquired event selection signals, An output unit configured to output the determined health information, The processing unit has, Based on the event selection signal, the subject's motor event is detected. Of the time segments of the HR-related signals, time segments are extracted that each cover a different period including each motor event. The extracted time segments are synthesized to calculate the aggregated HR response signal. One or more features are extracted from the aggregated HR response signals to determine the health information relating to the subject's cardiovascular system. A device configured in such a way.

2. The apparatus according to claim 1, wherein the processing unit is configured to align the extracted time segments before synthesis based on the event selection signal or an acquired alignment signal indicating time information of motion events within each time segment.

3. The apparatus according to claim 1 or 2, wherein the processing unit is configured to detect movement events by detecting movements of the subject that affect the subject's HR, such as changes in the subject's posture or the subject's turning over in bed, in particular recurring movements of the subject.

4. The apparatus according to any one of claims 1 to 3, wherein the processing unit is configured to detect an exercise event by detecting one or more exercise characteristics of the subject's movement, and the exercise characteristics include one or more exercise intensity exceeding an intensity threshold, exercise pattern, exercise speed, exercise direction, and exercise distance.

5. The apparatus according to any one of claims 1 to 4, wherein the processing unit is configured to detect, from selected time segments, in particular, based on one or more characteristics of each selected time segment, one or more unwanted time segments during which the HR-related signal is distorted or the HR during which is not affected despite the detection of a motion event, and discard the detected unwanted time segments in order to synthesize the extracted time segments.

6. The apparatus according to any one of claims 1 to 5, wherein the processing unit is configured to extract a time segment of the HR-related signal by determining a characteristic time point in which a specific feature of the detected event appears, and to select time segments before and after the characteristic time point so as to extend a first time period before the characteristic time point and a second time period after the characteristic time point.

7. The apparatus according to any one of claims 1 to 6, wherein the processing unit is configured to discard the time period in each selected time segment during which a detected motion event occurred, and to use only the time periods in each time segment before and / or after the detected motion event, in order to synthesize the extracted time segments.

8. The processing unit selects one or more features from the aggregated HR response signal, The difference between the maximum and minimum values ​​of the aggregated HR response signals, The difference between the maximum value and the start value of the aggregated HR response signal, The difference between the end value and the start value of the aggregated HR response signal, The time difference between the appearance of the minimum and maximum values ​​of the aggregated HR response signal, The time difference between the appearance of the start and end values ​​of the aggregated HR response signal, The maximum slope of the aggregated HR response signal, and The duration and / or rate of increase of the aggregated HR response signal. The apparatus according to any one of claims 1 to 7, configured to extract one or more of the following.

9. The apparatus according to any one of claims 1 to 8, wherein the processing unit is configured to determine the health information relating to the cardiovascular system of the subject by using a predetermined function, in particular a linear or nonlinear approximation to which predetermined scalar parameters are applied.

10. The apparatus according to claim 9, wherein the processing unit is configured to determine the scalar parameter from one or more relationships between the age of the subject and one or more features from the aggregated HR response signals.

11. The HR-related signal is one of the following: contact or remote photoplethysmography (PPG) signal, HR signal, ECG signal, cardiomyograph signal, intraocular pressure measurement signal, and / or holographic laser Doppler imaging signal. The apparatus according to any one of claims 1 to 10, wherein the event selection signal is one of a motion signal indicating the movement of the subject's body, an accelerometer signal, a muscle activity signal, a brain activity signal, a pressure sensor signal, a weight sensor signal, a vehicle's CAN bus signal, and a camera signal.

12. The aforementioned processing unit i) First convert the extracted time segments into signals representing frequencies over time, and then aggregate these signals into an aggregated HR response signal, or ii) First convert the extracted time segments into a time-frequency matrix, synthesize the matrix for each pixel, and then obtain the maximum ridgeline to obtain the aggregated HR response signal. The apparatus according to any one of claims 1 to 11, configured to synthesize the extracted time segments by means of the following.

13. A system for determining the cardiovascular health information of a subject, wherein the system is An HR sensor configured to detect time-dependent heart rate (HR) related signals that enable the representation or derivation of the subject's heart rate over time, A selection signal sensor configured to detect an event selection signal that enables the detection of the motor events of the subject, An apparatus according to any one of claims 1 to 12 for determining the health information relating to the cardiovascular system of a subject from the detected HR-related signals and the detected event selection signals, and An output interface configured to emit the determined health information, A system that has

14. A computer program executed by a processor of a device for determining a subject's cardiovascular health information, which causes the processor to execute a method for determining the health information, The method described above is The steps include obtaining a time-dependent heart rate (HR) related signal that enables the representation or derivation of the subject's heart rate over time, The steps include obtaining an event selection signal that enables the detection of the subject's motor events, A step of detecting the subject's motor event based on the acquired event selection signal, The steps include: extracting time segments from the acquired HR-related signals, each time segment covering a different period including each exercise event; The steps include: synthesizing the extracted time segments to calculate the aggregated HR response signal; A step of determining the health information relating to the cardiovascular system of the subject by extracting one or more features from the aggregated HR response signals, A step of outputting the determined health information, A computer program that has [a certain characteristic].

Images