Method, device and system for assessing the reliability of an estimated metric related to a user's physiological state - Patents.com

Abstract

The present invention relates to a method, device and system for assessing the reliability of an estimated metric related to a physiological state of a user, the method comprising the steps of acquiring one or more sensor signals related to the physiological state of a user, estimating a first metric and a second metric independently of each other using different methods and / or algorithms based on the one or more sensor signals, the first metric being related to the second metric by a first mathematical function, comparing the first value to a bound of the first value, evaluating a reliability assessment for the estimation of the first metric and / or the second metric based on the comparison, and outputting the evaluated reliability assessment.

Description

[Technical field]

[0001] The present invention relates to methods, devices and systems for assessing the reliability of an estimated metric related to a user's physiological state. [Background technology]

[0002] Continuous, unobtrusive and remote monitoring of activity and biometrics is becoming more and more prevalent in order to improve patient care, reduce hospitalizations and enhance patient outcomes. Gait speed, i.e. walking speed, is considered the sixth vital sign and can be used to assess the physical capabilities of patients in clinical practice and to determine frailty and associated health risks. Activity levels (e.g. in the form of pedometry) can help assess the overall level of a patient's mobility in hospital and to describe the context of their physiological response to activity. All these metrics become particularly relevant for devices monitoring ambulatory patients to enhance hospital decision-making processes, e.g. regarding discharge readiness, which so far mostly relies on subjective input from nurses regarding activity and patient condition.

[0003] Several algorithms exist for estimating physiological and anatomical metrics, especially gait metrics. Most of these algorithms use acceleration and cardiovascular signal features (e.g., acquired by on-body accelerometers, photoplethysmography (PPG) and electrocardiogram (ECG)) and data analysis techniques (e.g., linear regression and machine learning) to estimate these metrics (e.g., walking speed, pedometry and walking economy).

[0004] US Patent Application Publication No. 2013 / 0138007 describes a technique for measuring a user's heart rate. The disclosed heart rate measuring device includes an information storage unit for storing unique information on a user's specific heart rate or information on a detected heart rate, and a decision unit for deciding whether a state of the user's heart rate measurement is normal based on the unique information or the information on the heart rate stored in the information storage unit.

[0005] US Patent Application Publication No. 2015 / 0099945 and US Patent Application Publication No. 2013 / 0080255 describe additional methods for estimating physiological or anatomical metrics.

[0006] For all algorithms, the estimation may deviate from the actual values ​​due to unexpected data or inherent limitations of the characteristics and data analysis techniques used. In fact, the estimation of gait metrics is always characterized by a certain degree of error. However, it may not always be possible to improve the estimation results of the above-mentioned estimation methods, which does not guarantee that estimation errors will not occur on new data. Unfortunately, these estimation errors may lead to an incorrect assessment of the patient's condition. Apart from the severe consequences in clinical decisions, incorrect estimation of the user's physiological or anatomical metrics may lead to an incorrect assessment of the user's fitness level and thus to an inappropriate training program. Moreover, such incorrect estimation may result in an inappropriate or even harmful diet plan. Summary of the Invention [Problem to be solved by the invention]

[0007] Therefore, accurate estimation of physiological and anatomical metrics, especially gait metrics, is important to assess the condition of a patient, or the health status of a person, with high (or at least known) confidence. Therefore, these errors need to be minimized. In particular, to ensure the quality of the estimated physiological metrics, it becomes paramount to assess said confidence. [Means for solving the problem]

[0008] It is an object of the present invention to improve the reliability of estimated metrics regarding a user's physiological state and thus to more accurately assess the user's health status, in particular to improve the reliability of estimated metrics regarding a user's walking ability.

[0009] In a first aspect of the present invention, a method for assessing the reliability of an estimated metric related to a physiological state of a user is presented, the method comprising: obtaining one or more sensor signals related to a physiological condition of a user; estimating a first metric and a second metric independently of each other using different methods and / or algorithms based on one or more sensor signals, the first metric being related to the second metric by a first mathematical function; determining a first value of the first mathematical function; comparing the first value to a boundary for the first value; assessing a confidence rating for the estimation of the first metric and / or the second metric based on the comparison; and outputting the evaluated reliability rating. has.

[0010] In a further aspect of the present invention, a device for assessing reliability of an estimated metric related to a physiological state of a user is presented, said device comprising: a sensor input configured to acquire one or more sensor signals related to a physiological condition of the user; 1. A processor comprising: estimating a first metric and a second metric independently from each other using different methods and / or algorithms based on the one or more sensor signals, where the first metric is related to the second metric by a first mathematical function; determining a first value of the first mathematical function; comparing the first value to a boundary of the first value; and assessing a confidence assessment for the estimate of the first metric and / or the second metric based on the comparison; a processor configured to: an information output unit configured to output the evaluated reliability assessment; has.

[0011] In yet another aspect of the present invention, a system for assessing reliability of an estimated metric related to a physiological state of a user is provided, the system comprising: a sensor configured to measure one or more sensor signals related to a physiological condition of a user; A device as disclosed herein, and a user interface configured to provide a trustworthiness assessment of the device; has.

[0012] In a further aspect of the present invention, there is provided a corresponding computer program having program code means for causing a computer to perform the steps of the methods disclosed herein, when said computer program is run on a computer, as well as a non-transitory computer readable recording medium storing a computer program product which, when run by a processor, performs the methods disclosed herein.

[0013] Preferred embodiments of the invention are defined in the dependent claims. It is to be understood that the claimed devices, systems, computer programs and media have similar and / or identical preferred embodiments as the claimed methods, particularly as defined in the dependent claims and as disclosed herein.

[0014] Conventional methods used to estimate physiological or anatomical metrics, especially gait metrics, are not as accurate as needed. In particular, estimates that deviate significantly from the actual values ​​often go unrecognized. This is true for many types of estimations of physiological and anatomical parameters, ranging from acceleration estimation to cardiac signal estimation.

[0015] The present invention proposes a method in which the reliability of two or more estimated metrics relative to one another is estimated based on the biomechanical and physiological boundaries of the relationship of these metrics. In other words, it is proposed to assess the reliability of the estimation of physiological (or anatomical) metrics, in particular gait metrics, by cross-checking different metrics relative to one another based on the biomechanical, physiological or anatomical boundaries of their relationship.

[0016] In many clinical decision-making processes, it is important to determine when an estimation is correct, since decisions related to the health status of the user should be based on reliable metrics. Furthermore, only if the reliability of the metric (i.e. the output of the estimation algorithm used, or in other words the estimation of the activity or bio-parameter) is precisely known, metrics that deviate significantly from the actual value are rejected, i.e. screened out and ignored in further consideration. Furthermore, knowledge of the reliability of the estimated metric can help to improve the algorithm to obtain even more reliable results.

[0017] In the proposed method, user activity (and / or bio-parameters) are monitored and at least two boundary-related physiological (or anatomical) metrics are calculated (or more accurately estimated) from the monitoring device's signals. To estimate the metrics, one or more methods, in particular algorithms, are used. Then, it is evaluated whether a metric respects the boundaries determined by the other metric (preferably vice versa). If the boundaries are not respected, the corresponding metric can be flagged / labeled in a specific way so that such unreliable metrics are ignored in further consideration. For example, a metric that does not respect the boundaries of another metric is labeled as "unreliable" and / or a specific reliability score is provided for said metric.

[0018] More precisely, a first metric A is estimated using a method g and a second metric B is estimated using a method h, where g and h are methods (or algorithms) that use one or more features of one or more sensor signals. The feature sets of functions g and h may be the same or different. In particular, parts of each feature set may be the same. However, the metrics are generally estimated independently of each other. The one or more features mentioned may include, for example, any of the periodicity of the acceleration signal, the standard deviation of the acceleration signal, the slope of the pressure signal (e.g. from a barometer), the average heart rate, the maximum heart rate and the respiration rate range.

[0019] A first metric A and a second metric B are related to each other by a first mathematical function. For example, A and B are related to each other via a third metric f(A,B) such that f(A,B)=A / B. By inserting the values ​​of the metrics into the function f(A,B), a first value C can be determined. Then, said first value C is compared with the boundaries of said first value (or more precisely with the boundaries of the possible values ​​of the function) and based on the result of this comparison it is assessed whether the first metric and / or the second metric are reliable.

[0020] The proposed idea stems from the fact that the values ​​of different physiological metrics are bounded against each other (i.e. one metric defines the minimum and / or maximum of another metric). Instead of defining the boundaries for each of the metrics based on the maximum and minimum of the first value C, the estimated first value C is obtained via f(A,B)=A / B and bounded to obtain the confidence of both metrics A and B.

[0021] As an example, the walking metrics walking speed and walking rate (ie, cadence) are related to each other based on stride length. Walking speed [t] = measured walking rate [t] * measured stride length Moreover, both metrics are bounded to each other. As an example, each metric of the first metric and the second metric is extracted from one or more sensor signals. For example, the walking speed is estimated based on a signal provided by an inertial sensor, such as an accelerometer or a gyroscope. At the same time, the walking rate of the user can be estimated from the same sensor signal or from another sensor signal (using the same or other signal features). The walking speed of the first metric is related to the walking rate of the second metric by a stride function from the estimation of both metrics, so that a value of the stride function (i.e., a first value) is calculated. Since the stride represents a bounded value (bounded both above and below), the estimated (i.e., calculated) stride is compared to the bounds.

[0022] In general, the comparison of the first value to the boundary includes a direct comparison and an indirect comparison.

[0023] In a direct comparison, it can be checked whether the first value respects the boundary, i.e. whether the first value is lower, equal or higher than the boundary. A direct comparison can also include calculating the difference between the first value and the boundary. In the above example, it can be checked for example whether the estimated value of the stride length is smaller than the upper stride length boundary. Similarly, the difference between the upper stride length boundary and the estimated value of the stride length can also be calculated.

[0024] On the other hand, in an indirect comparison, the first value may not be compared directly with the boundary, but indirectly by comparing the first value with a function of this boundary. This function of the boundary may for example be a function of the difference between the boundary and the first value. In the above example, a function may be calculated which includes the difference between the estimated value of the stride length and the boundary. Furthermore, the result of the evaluation of said function may be compared with the boundary.

[0025] The results of this comparison can then be used to evaluate a reliability rating of the estimate for the first metric and / or the second metric. This reliability rating may be output directly or may be filtered before being output. For example, reliability ratings below (or above) a predefined limit may be screened out as unreliable and therefore not output. On the other hand, only reliability ratings below (or above) a predefined limit may be output.

[0026] Generally, multiple first metrics and / or second metrics (values) may be determined using different methods / algorithms. Accordingly, multiple first values of a mathematical function are determined using these metrics, and these multiple first values are compared to a boundary of the first values. Accordingly, multiple reliability assessments for the estimation of the first metric and / or second metric based on said comparison are evaluated. Next, the multiple reliability assessments are analyzed, and the method used to determine the metric is selected according to the reliability assessment. In particular, only the method for estimating the metric whose reliability assessment of the corresponding metric results in a value higher than a predetermined threshold can be further used. Accordingly, the reliability of the metric estimation can be further improved.

[0027] The main use of the present invention relates to health monitoring in hospitals and at home, particularly focusing on sleep and respiratory therapy, biometric detection, and connectivity detection.

[0028] In one embodiment of the method, comparing a first value to a boundary of the first value has the effect of determining whether the first value satisfies a relationship between a first mathematical function and the boundary.

[0029] The relationship between the first mathematical function and the boundary can have an equation or an inequality. Accordingly, it is checked whether the first value satisfies the equation or the inequality. For example, the first value is A / B, where A is a first metric and B is a second metric, and it is checked whether A / B < C, where C represents a boundary value. Generally, the relationship between the first mathematical function and the boundary can have any of the difference between the first mathematical function and the boundary, the sum of the first mathematical function and the boundary, the fraction of the first mathematical function and the boundary, and the product of the first mathematical function and the boundary.

[0030] The boundary may be time-dependent or user-dependent. User-dependence has a dependence on any of age, weight, height, gender, leg length, foot size, or any other physiological or anatomical characteristic of the user.

[0031] In one embodiment, the boundary is a lower boundary and / or an upper boundary.

[0032] For example, a first metric is walking speed and a second metric is walking rate. Both metrics are related by stride length. The value for the user's stride length is bounded by physiological / anatomical constraints such as the leg length of the user (or people in general). Thus, the bound of the first value may be an upper bound of the stride length. Similarly, a lower bound for the stride length value or another first value may be defined.

[0033] In another example, the first metric is the user's heart rate and the second metric is the user's activity (during a particular activity). The relationship between these metrics depends heavily on the user's fitness level. This fitness can be expressed in terms of VO2max (i.e., the maximum amount of oxygen consumption measured during incremental exercise), which is bounded. The lower and upper bounds of VO2max are determined based on the user's physical characteristics, such as weight, BMI, lean mass, and age.

[0034] As an example, the lower bound is set to 0 and the upper bound is set to (positive) infinity.

[0035] In another embodiment, comparing the first value to a boundary for the first value comprises determining whether the first value is higher than a lower boundary and / or lower than an upper boundary.

[0036] Similarly, it may be determined whether the first value is equal to or lower than a lower boundary and / or whether the first value is equal to or higher than an upper boundary. For example, an estimated stride length value may be compared to a lower boundary, including determining whether the estimated stride length value is higher than the lower boundary.

[0037] In another embodiment of the method, the boundary is predefined, or the method further comprises estimating the boundary based on any of a previous first metric, a previous second metric, a previous first value, or one or more sensor signals.

[0038] The boundary, in particular the lower and / or upper boundary, may be determined from the literature. For example, as can be seen from the literature, during walking, the minimum stride of a person is 30 cm and the maximum stride is 90 cm. However, as the data in the literature shows, these boundaries may be different during running.

[0039] However, the boundary may be determined empirically. In particular, the boundary can depend on one or more signals obtained and / or on the estimated metrics based on these signals. In fact, the lower and upper boundaries of the stride may vary based on the accelerometer signal obtained or on the metrics estimated from this signal. The boundary can be extrapolated as a statistical value from the recorded data (for example, for user A, the fraction (walking speed / walking rate) < N for 99%, where N is the upper boundary at this time). In fact, based on empirical knowledge, common sense, or data collected previously, a fixed threshold can be extrapolated.

[0040] Furthermore, the boundary depends on one or more predetermined parameters.

[0041] In any case, in particular for the estimated, i.e., empirically determined, lower and upper boundaries, possible estimation errors can be further taken into account. Thus, if the stride is mainly overestimated, the boundary, in particular the upper boundary for the stride, can be lowered.

[0042] In yet another embodiment of the method, any of the first metric, the second metric, and the first mathematical function can be made time-dependent. For example, the relationship between the first metric, i.e., the estimated walking speed, and the second metric, i.e., the estimated walking rate, changes according to the time elapsed during the day or during walking.

[0043] Additionally, any of the first metric, the second metric and the first mathematical function may be dependent on a physiological parameter of the user.

[0044] In a further embodiment of the method, the method further comprises determining a second value of a second mathematical function relating the first mathematical function to the boundary, wherein assessing a confidence assessment for the estimate of the first metric and / or the second metric is based on the second value.

[0045] The second mathematical function may in particular comprise the difference between a first value boundary (when the first and second metrics are inserted) and the value of the first mathematical function. For example, if the first mathematical function represents a stride length, the second mathematical function may comprise the difference between a lower stride length boundary MIN_SL and an estimated stride length value. Similarly, the second mathematical function may comprise the difference between an upper stride length boundary MAX_SL and an estimated stride length value. For example, the second function may be:

number

number

[0046] As can be seen from this example, the second mathematical function can be time-dependent. In particular, the second mathematical function can be a function of a time-dependent boundary of the first value. However, the second mathematical function may include the first mathematical function that is time-dependent. However, the second mathematical function may not be time-dependent as well.

[0047] In another embodiment, evaluating the confidence assessment comprises assigning a binary or continuous scale score to the estimate of the first metric and / or the second metric.

[0048] For example, the reliability assessment of a metric can be expressed as a binary score of "reliable" and "unreliable". For example, if the two metrics are estimated speed and estimated walking rate, a first function is defined as the estimated stride length. A first value of the stride length is compared to a minimum stride length MIN_SL and a maximum stride length MAX_SL. In other words, the following relationship:

number

[0049] On the other hand, if the first value is higher than an upper bound for the first value, then the confidence is set to 0, otherwise it is set to 1. Furthermore, if the first value is lower than a lower bound for the first value, then the confidence is set to 0, otherwise it is set to 1, where a confidence rating of 0 corresponds to an unreliable estimated first value and a confidence rating of 1 corresponds to a reliable estimated first value.

[0050] As another example, two quadratic functions, specifically the first quadratic function that represents the “lower stride length reliability”

number

number

[0051] If the first second function or the second second function is less than 0, the first second function or the second second function, respectively, may be set to 0. On the other hand, if the first second function or the second second function is greater than 1, the first second function or the second second function, respectively, may be set to 1. A reliability assessment may be set to the smaller of the values ​​of the first and second second functions, i.e. the reliability assessment is

number

[0052] The reliability rating is then represented by a score ranging from 0 to 1.

[0053] In yet another (different) example, the reliability assessment can be any score between 0 and 1 (or any other number), depending on, for example, the difference between the first value and the boundary. If the difference between the first value and the upper boundary and the difference between the first value and the lower boundary are equal, i.e., if the first value is halfway between the lower boundary and the upper boundary, the reliability is set to 1. On the other hand, the score of the reliability assessment decreases as the first value moves away from the lower boundary and closer to the upper boundary, and vice versa, where if the first value is lower than the lower boundary or higher than the upper boundary, the score of the reliability assessment is set to 0, suggesting that such a first value is not reliable.

[0054] In yet another alternative, any first value that falls between a lower boundary and an upper boundary can be considered reliable, and first values ​​that fall outside either of these boundaries can be assigned a reliability rating that decreases with increasing distance from the boundaries.

[0055] In a further embodiment, the method further comprises adapting or modifying the method for estimating the first metric and / or the second metric based on the reliability assessment. In other words, the reliability assessment is used to update the estimation of the first and / or second metric. In general, more than one estimation method is used for the estimation of one (or both) of these metrics. For example, method A may be used to estimate the walking speed, whereas methods A and B may be used to estimate the walking rate, or methods X and Y may be used for the speed estimation, and methods A and B (as well as C, etc.) may be used for the walking rate estimation. In such a case, the corresponding reliability assessment can be used to select a predicted, i.e. estimated metric (e.g., the walking rate estimation of algorithm A or B) or a combination of estimated metrics (e.g., the walking rate estimation of algorithm A combined with the speed estimation of algorithm X) that can be used for further consideration, in particular for further reliability assessment. Furthermore, based on the reliability assessment, it can be decided which metric to output directly or to use as a further input for the health assessment. In this way, the accuracy of the metric can be improved while still maintaining high coverage.

[0056] As another example, algorithm A1 can be used for speed estimation resulting in estimated speed a1, algorithm A2 can be used for speed estimation resulting in estimated speed a2, and algorithm A3 can be used for speed estimation resulting in estimated speed a3, etc. Given that the algorithms are different from each other, not all of these algorithms will work equally well for the various possible walking speeds or walking patterns applied. For example, for low speeds, algorithm A1 is the most accurate algorithm, while for higher speeds, algorithm A2 is more appropriate, and for walking with a walking aid, etc., algorithm A3 may be the best choice. It is clear that if the speed estimate a1 is not reliable (based on the method described above), it should not be used to evaluate the user. Therefore, algorithm A1 is rejected (excluded) for speed estimation. Thus, the exclusion of algorithm A1, and therefore of estimate a1, reduces the coverage of the speed estimate (i.e., one data point less). However, this can be overcome by using other algorithms (A2, A3, ... An) in parallel. These methods are checked for reliability and the method with the best reliability is used instead of algorithm A1.

[0057] Furthermore, after excluding the speed estimate a1, the speed estimates of other algorithms can be mixed (e.g., by averaging the speed estimates). In fact, walking speed, as well as other parameters that represent the physiological or anatomical characteristics of the user when measured intra- and inter-individual, show similar boundaries and relationships with each other, and thus typically remain within certain boundaries, but there are often certain parameters that determine the type of relationship. Latt, MD, Menz, HB, Fung, VS, & Lord, SR: “Walking speed, cadence and step length are selected to optimize the stability of head and pelvis accelerations”, Experimental brain research, 184(2), 201-209, for example, discloses various relationships between self-selected walking speed (i.e., the walking speed experienced by the user) and several stride-related metrics. As can be seen from the literature, the relationship between the measured speed and walking rate is linear up to about 2.0 m / s, which corresponds to a walking rate of about 150 steps / min. However, the error bars at this point are very large. In fact, for faster walking speeds, the relationship between the two metrics changes. This is due to the fact that at this point, the user's activity has changed from walking to jogging. The relationship between the two metrics is modulated by several human-dependent biomechanical characteristics and contextual variables. One of the main factors is the variability of stride length. In fact, stride length is not constant from person to person and within a person even at the same speed. For this reason, the estimation of speed via walking rate can have a significant error (even at speeds below 2 m / s). This "weak" relationship does not allow accurate calculation of walking rate based on speed by knowing the stride length (and vice versa). However, it is possible to set bounds for each of said metrics based on the minimum and maximum possible stride length values. Furthermore, the disclosed data can be used to adapt bounds for the reliability estimation.Furthermore, the algorithm used may be modified according to the value of a previously estimated metric, or more precisely according to the parameters on which that metric depends, or according to the first value.

[0058] In a further embodiment, the one or more sensor signals include an acceleration signal, a gyroscope signal, a heart rate signal, a respiration signal, oxygen saturation (SpO 2 ) signal, a movement signal, a force signal, a pressure signal, an electromyography (EMG) signal, a temperature signal, a calorie consumption signal, a Global Positioning System (GPS) signal, a camera signal, a radar signal, a speed measurement signal, a torque measurement signal, an electrocardiogram (ECG) signal, an electroencephalography (EEG) signal, a photoplethysmography (PPG) signal, a barometer signal, a muscle tension signal, a sweat level signal, a stress level signal, a skin conductance signal, a galvanic skin response signal, an inertial measurement signal and a blood pressure signal, and / or the first metric and / or the second metric comprise any of walking speed, walking rate, stride length, heart rate, energy consumption and type of activity.

[0059] Any other metric relating to the physiological or anatomical condition of the user is contemplated as well.

[0060] In one embodiment of the system, the sensors include an accelerometer, a gyroscope, a heart rate sensor, a respiration sensor, and oxygen saturation (SpO 2 ) sensor, motion sensor, force sensor, pressure sensor, electromyography (EMG) sensor, temperature sensor, calorie consumption sensor, global positioning system (GPS) sensor, camera, radar, speed measurement sensor, torque measurement sensor, electrocardiogram (ECG) sensor, electroencephalography (EEG) sensor, photoplethysmography (PPG) sensor, barometer, muscle tension sensor, sweat level sensor, stress level sensor, skin conductance sensor, galvanic skin response sensor, inertial measurement sensor, and blood pressure sensor.

[0061] In another embodiment of the system, the sensor is a remote sensor or a wearable sensor. The sensor may in particular be a contact sensor. Remote sensors include, for example, a camera (recording the user's activity) or a shoe sensor. Wearable sensors may be, for example, integrated into a wearable object, such as a pendant or a trouser pocket, or may be attached to the wrist, fingertip, earlobe, in-ear portion, torso, etc., and may include, for example, any of a smartphone, a smartwatch, and an in-ear device.

[0062] The system may include one or more sensors. In the case of a system with multiple sensors, one or more of the sensors may be wearable sensors. Wearable sensors are fairly unobtrusive and particularly convenient for long-term use. In particular, gait metrics are estimated by using wearable sensors. [Brief description of the drawings]

[0063] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. [Figure 1] FIG. 1 shows a flow chart of an embodiment of the method according to the invention. [Figure 2A] FIG. 2A shows a plot of the relationship between measured stride length and walking speed. [Figure 2B] FIG. 2B shows a plot of the ratio of stride length to measured height versus walking speed. [Figure 2C] FIG. 2C shows a plot of the relationship between measured stride length and walking speed. [Figure 2D] FIG. 2D shows a plot of the ratio of stride length to measured height versus walking speed. [Figure 2E] FIG. 2E shows a plot of estimated walking speed versus measured walking speed. [Figure 2F] FIG. 2F shows a plot of estimated gait rate versus measured gait rate. [Diagram 3]FIG. 3 is a schematic diagram of an embodiment of a device according to the invention. [Figure 4] FIG. 4 is a schematic diagram of an embodiment of a system according to the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0064] 1 shows a flow chart of an embodiment of a method according to the invention, the steps of which can be executed by a device 10, where the main steps of the method are executed by a processor 14. The method may for example be implemented as a computer program executed on a computer or processor.

[0065] In a first step S2, a number of sensor signals are obtained from a number of sensors. These sensor signals relate to the physiological state of the user. In particular, they relate to the walking ability of the user. In a second step S4, a first metric and a second metric are extracted from the signals. In other words, the sensor signals are used to estimate a first metric and a second metric. In this embodiment, the first metric is estimated using algorithm A and algorithm B, and the second metric is estimated using algorithm A and algorithm B. Thus, in practice, two values ​​of the first metric and two values ​​of the second metric are estimated. The first metric and the second metric are selected such that they are related to each other by a mathematical function whose values ​​are bounded. In a next step S6, a first value of the mathematical function is derived (calculated) by inserting the first and second metrics estimated by algorithm A. Furthermore, another first value of the mathematical function is derived by inserting the first and second metrics estimated by algorithm B. Then, in step S8, the first values ​​are compared with a boundary for each of these values. In this embodiment, the boundary is pre-determined. Depending on the result of the comparison, in step S10, a reliability assessment is evaluated for each of the estimated first values, and thus for the estimation of the first and second metrics. Thus, two reliability assessments are obtained, one for the metric based on algorithm A and one for the metric based on algorithm B. In this embodiment, the reliability assessment is binary. The first metric is either labeled as "trusted" or as "untrusted". The same applies to the second metric. In optional step S12, the obtained reliability assessments are filtered using the specific assessment on which the culling is performed. More precisely, the estimated metric that corresponds to the assessment "untrusted" is culled.In step S14, the evaluated reliability ratings (and corresponding metrics) are output, where if "unreliable" metrics were filtered out in step S12, only "reliable" ratings (and metrics) are output.

[0066] The reliability assessment (and corresponding metric) assessed for the first value in step S10 may (optionally) be used as input in step S4. In particular, if the reliability assessment for the metric estimated using algorithm A is higher than the reliability assessment for the metric estimated using algorithm B, then future estimations using algorithm B are omitted.

[0067] 2A and 2B respectively show the (actual) relationship between walking speed and stride length, and between walking speed and the ratio of stride length to the user's height, where height means the user's height, measured for a first data set. In other words, the data points shown are points corresponding to measured values ​​and are not estimated using an algorithm. In particular, the diagram in FIG. 2A shows the measured user's stride length in meters (m) on the y-axis and the measured user's walking speed in meters per second (m / s) on the x-axis. The diagram in FIG. 2B shows the measured ratio of stride length to the user's height on the y-axis and the measured user's walking speed in meters per second on the x-axis. From these (reference) measurements, likelihood boundaries are determined. The solid lines shown in FIG. 2A and FIG. 2B show the 95% estimation interval of the linear regression, i.e. there is a 95% probability that the true best fit line of these data points lies within the confidence interval marked by the solid line. In this embodiment, the 95% prediction interval of the linear regression is used as the speed / walking rate dependent boundary.

[0068] Similarly, the dashed lines indicate upper thresholds for the data shown. In particular, as can be seen from Fig. 2A, an upper boundary for the absolute estimated stride length can be set to 0.9m (Max_SL=90cm). Similarly, an upper boundary for the ratio of stride length to the user's height can be set to 0.5 (Max_SL / height=0.5) (see dashed lines in Fig. 2B).

[0069] 2C and 2D respectively show the relationship between predicted or estimated speed (using an algorithm that utilizes accelerometer data, i.e., accelerometer signals, as input) and predicted or estimated stride length, and between said speed and stride length to height ratio (also using an algorithm that utilizes accelerometer data as input) for a second data set (different from the first data set used to determine the boundaries). In particular, the diagram of FIG. 2C shows the estimated user stride length in meters on the y-axis and the estimated user walking speed in meters per second on the x-axis. The tabular diagram of FIG. 2D shows the estimated user stride length to height ratio on the y-axis and the estimated user walking speed in meters per second on the x-axis.

[0070] The estimation of step length and step length to height ratio can be based on the fact that the estimated step length is expressed as a ratio between an estimated walking speed and an estimated walking rate. The accelerometer used for measuring the walking speed and the step length to height ratio in this embodiment is a pendant-mounted accelerometer. The boundaries shown in Figures 2C and 2D correspond to the boundaries of the measured data shown in Figures 2A and 2B. As can be seen from Figures 2C and 2D compared to Figures 2A and 2B respectively, the estimated values, except for some outliers, have a similar relationship to each other as the measured values, which can be considered as an indicator that the algorithms used for estimating the step length and the step length to height ratio (or more precisely, the algorithms for walking speed and walking rate) are working as desired. On the other hand, if the predicted values ​​have a significantly different relationship to each other than the measured values, this is an indicator that at least one of the algorithms used (either the algorithm for estimating the walking speed or the algorithm for estimating the walking rate) produces erroneous predictions / estimations and is therefore defective.

[0071] In Figures 2E and 2F, the estimated speed and the estimated walking rate are plotted against the (reference) speed and the walking rate, respectively. More precisely, the diagram in Figure 2E shows the estimated walking speed [m / s] (y-axis) against the measured walking speed [m / s] (x-axis), and the diagram in Figure 2F shows the estimated walking rate [step / s] (y-axis) against the measured walking rate [step / s] (x-axis). The diagrams shown in Figures 2E and 2F are based on the second data set.

[0072] If the algorithms used to estimate walking speed and rate work as desired, all data points shown will be on the solid black line. Thus, the further a data point is from this black line, the worse the estimated value will be. Based on the boundaries determined by Figures 2A and 2B, data points that do not adhere to the boundaries are marked as unreliable. In other words, in this embodiment, a binary reliability assessment approach is applied to the walking activity metric.

[0073] For example, in Figures 2C to 2F, data points indicated with open circles correspond to data points that correspond to a stride length that is too large, i.e., data points that are outside the upper stride length boundary. Estimated data points that correspond to an estimated stride length-height ratio that is too large are indicated with triangles. Estimated data points that correspond to stride length estimates that are outside the boundary are indicated with crosses, and estimated data points that correspond to estimated stride length to height ratios that are outside the boundary are indicated with stars. In this way, the four stride length related boundaries can be used to determine unreliable data points.

[0074] As can be seen in Figures 2E and 2F, the data points marked as unreliable are for the estimated walking rates that are (most of the time) quite far from the black line, and therefore are indeed inaccurate, whereas for the estimated walking speed this is less clear. This result is also confirmed by the performance metrics shown in the table below. [Table 1]

[0075] The table shows the results of the speed and gait rate estimation when all data are retained (line 3) and after removing data points that are outside the respective imposed boundaries (lines 4-7). The abbreviation "MAE" stands for "Mean Absolute Error", the abbreviation "RMSE" stands for "Root Mean Square Error", and the abbreviation "R" stands for "Pearson's Correlation" (where "R" stands for correlation between measured and predicted / estimated values).

[0076] As can be seen from the table, the reliability of said metrics shows an improvement when removing data based on estimated stride length (ratio), especially for step rate (see the right part of the table). Indeed, if all predictions / estimations were exactly as they were, all errors shown in the table would be 0 and "R" would be 1. Removing only random points would not have improved the accuracy of the data much. An improvement comes at the expense of not estimating the results for certain parts of the data, but this effect appears to be relatively limited, at most 4% (see the first column).

[0077] In the above example, both estimated metrics, i.e., estimated stride length and estimated walking rate, are considered unreliable if the first values ​​of stride length and stride length to height ratio are not inside their imposed boundaries, while other embodiments are conceivable in which the estimated metric considered most reliable (e.g., based on previously calculated results) can be used as a guideline for the other metric, and thus only the latter is considered unreliable if the boundaries are not respected. For example, in the above example, the speed estimate is generally more accurate than the walking rate estimate. Thus, only the estimated walking rate is considered unreliable, so that the estimated speed is maintained. Furthermore, the walking rate estimate can be updated based on the speed estimate and the stride length and / or stride length to height ratio boundaries.

[0078] 3 shows a schematic diagram of an embodiment of a device 10 according to the invention. The device 10 comprises a sensor input 12, a processor 14 and an information output 16.

[0079] The sensor input 12 is configured to acquire at least one sensor signal from at least one sensor, the sensor signal being related to a physiological condition of the user. The sensor input 12 may be directly coupled or connected to a sensor, or may acquire (i.e., capture or receive) the sensor signal from a storage device, buffer, network, bus, or the like. Thus, the sensor input 12 may be a communication or data interface (wired or wireless), such as, for example, a Bluetooth interface, a WiFi interface, a LAN interface, an HDMI interface, a direct cable connection, or any other suitable interface that allows for signal transfer to the device 10.

[0080] The processor 14 applies at least one algorithm to the signal and its features to estimate a first metric and a second metric, the metrics being related to each other by a mathematical relationship. In particular, the processor 14 is configured to determine a first value of a first mathematical function by inserting the metric to be estimated into this function. The first value is then compared with a bound for this first value. Based on the result of this comparison, the processor 14 evaluates a confidence assessment for the estimation of the first metric and / or the second metric.

[0081] Subsequently, the information output unit 16 provides the assessed trustworthiness rating for further processing. The information output unit 16 can generally be any interface that provides the assessed trustworthiness rating, for example, transmits this rating (to another device) or provides this rating for searching (by another device, for example, a smartphone, a computer, a tablet, etc.). It can therefore generally be any (wired or wireless) communication or data interface.

[0082] FIG. 4 shows a schematic diagram of an embodiment of a system 100 according to the invention. The system comprises two sensors 20a and 20b, a device 10 and a user interface 40. Both sensors 20a and 20b are configured to measure at least one sensor signal related to a physiological state of a user. In this embodiment, the sensor 20a is a smartwatch configured to be worn by a user and the sensor 20b is a smartphone configured to be worn by a user. The sensor 20a is configured to measure at least one signal related to the heart rate of the user and the sensor 20b is configured to measure a signal related to the activity amount of the user during a certain activity. The device 10 is configured to obtain the signals of both sensors 20a and 20b and to estimate the heart rate and the activity amount of the user based on these signals. The relationship between the heart rate metric and the activity amount metric depends heavily on the fitness level of the person. This fitness level can be expressed in terms of VO2max, which generally cannot be exceeded or exceeded by a certain boundary value. Thus, when using VO2max boundaries, the device 10 of the system 100 is configured to assign a confidence rating to the estimated metrics of heart rate and activity. This confidence rating is then provided to a user interface 40 configured to provide a confidence rating of the device 10. The user interface 40 can generally be any means for outputting the confidence rating in visual or audible form, e.g., in text form, as an image or diagram, as sound or speech, etc. For example, the user interface 40 can be a display, a speaker, a touch screen, a computer monitor, a smartphone or tablet screen, etc.

[0083] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description should be considered as exemplary or explanatory and not restrictive, i.e., the invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

[0084] In the claims, the word "comprising" does not exclude other elements or steps, nor does it exclude a plurality of them. A single element or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0085] The computer program may be stored / distributed on a suitable non-transitory medium, such as an optical storage medium or a solid-state medium, provided, for example, together with or as part of other hardware, but may also be distributed in other forms, such as, for example, via the Internet or other wired or wireless communication systems.

[0086] Any reference signs in the claims should not be construed as limiting the scope.

Claims

1. 1. A method for assessing the reliability of an estimated metric related to a physiological state of a user, the method comprising: acquiring one or more sensor signals related to a physiological condition of the user; estimating a first metric and a second metric independently of each other using different methods and / or algorithms based on the one or more sensor signals, wherein the first metric is related to the second metric by a first mathematical function; determining a first value of the first mathematical function; comparing the first value to a boundary of the first value; assessing a confidence rating for the estimation of the first metric and / or the second metric based on the comparison; and outputting the evaluated reliability rating. A method comprising:

2. 2. The method of claim 1, wherein comparing the first value to a boundary of the first value comprises determining whether the first value satisfies a relationship between the first mathematical function and the boundary.

3. The method according to claim 1 or 2, wherein the boundary is a lower boundary and / or an upper boundary.

4. 4. The method of claim 3, wherein comparing the first value to a boundary of the first value comprises determining whether the first value is higher than the lower boundary and / or lower than the upper boundary.

5. 2. The method of claim 1, wherein the boundary is predetermined, or the method further comprises estimating the boundary based on any of a previous first metric, a previous second metric, a previous first value, or the one or more sensor signals.

6. The method of claim 1 , wherein any of the first metric, the second metric, and the first mathematical function is time dependent.

7. 2. The method of claim 1, further comprising determining a second value of a second mathematical function relating the first mathematical function to the boundary, wherein assessing a confidence rating for the estimation of the first metric and / or the second metric is based on the second value.

8. The method of claim 1 , wherein assessing the reliability assessment comprises assigning a binary or continuous score to the estimate of the first metric and / or the second metric.

9. The method of claim 1 , further comprising adapting or modifying a method for estimating the first metric and / or the second metric based on the reliability assessment.

10. the one or more sensor signals comprise any of an acceleration signal, a gyroscope signal, a heart rate signal, a respiration signal, an oxygen saturation signal, a movement signal, a force signal, a pressure signal, an electromyogram signal, a temperature signal, a calorie expenditure signal, a global positioning system signal, a camera signal, a radar signal, a speed measurement signal, a torque measurement signal, an electrocardiogram signal, an electroencephalogram signal, a photoplethysmography signal, a barometer signal, a muscle tension signal, a sweat level signal, a stress level signal, a skin conductance signal, a galvanic skin response signal, an inertial measurement signal, and a blood pressure signal; and / or The first metric and / or the second metric include any one of walking speed, walking rate, stride length, heart rate, energy expenditure, and activity type; The method of claim 1.

11. 1. A device for assessing the reliability of an estimated metric related to a physiological state of a user, the device comprising: a sensor input configured to acquire one or more sensor signals related to a physiological condition of the user; 1. A processor, comprising: estimating a first metric and a second metric independently from each other using different methods and / or algorithms based on the one or more sensor signals, the first metric being related to the second metric by a first mathematical function; determining a first value of the first mathematical function; comparing the first value to a boundary of the first value; and assessing a confidence assessment for the estimation of the first metric and / or the second metric based on the comparison; a processor configured to: an information output unit configured to output the evaluated reliability assessment; A device having:

12. 1. A system for assessing the reliability of an estimated metric related to a physiological state of a user, the system comprising: a sensor configured to measure one or more sensor signals related to a physiological condition of the user; A device according to claim 11, and a user interface configured to provide a trustworthiness assessment of the device; A system having:

13. 13. The system of claim 12, wherein the sensor comprises any of an accelerometer, a gyroscope, a heart rate sensor, a respiration sensor, an oxygen saturation (SpO2) sensor, a movement sensor, a force sensor, a pressure sensor, an electromyography (EMG) sensor, a temperature sensor, a calorie expenditure sensor, a global positioning system sensor, a camera, a radar, a speed measurement sensor, a torque measurement sensor, an electrocardiogram (ECG) sensor, an electroencephalography (EEG) sensor, a photoplethysmography (PPG) sensor, a barometer, a muscle tension sensor, a sweat level sensor, a stress level sensor, a skin conductance sensor, a galvanic skin response sensor, an inertial measurement sensor, and a blood pressure sensor.

14. The system of claim 12 or 13, wherein the sensor is a remote sensor or a wearable sensor.

15. 2. A computer program comprising program code means for causing a computer to carry out the steps of the method according to claim 1 when said computer program is run on said computer.

Images