Personalized baseline correction for detection of infection from physiological signals

Abstract

A system (1) includes a monitoring device (10) including at least one sensor (12). The wearable monitoring device is configured to be worn by an associated patient and the at least one sensor configured to measure physiological sensor data (13) of the associated patient wearing the monitoring device. An electronic processor (16) is integrated with or in wireless communication with the monitoring device. The electronic processor is programmed to: calibrate a personalized normalization process for the associated patient by applying a calibration process to physiological sensor data measured of the associated patient during a baseline time period; apply the personalized normalization process to physiological sensor data measured of the associated patient during an analysis time period to generate normalized physiological sensor data; analyze the normalized physiological sensor data to identify a condition of the associated patient; and output an indication of the identified condition of the associated patient.

Description

PERSONALIZED BASELINE CORRECTION FOR DETECTION OF INFECTION FROM PHYSIOLOGICAL SIGNALS

[0001] The following relates generally to the medical monitoring arts, wearable medical monitor arts, patient condition monitoring arts, and related arts.BACKGROUND

[0002] Health-related unobtrusive sensing systems enable replacement of continued hospitalization with unobtrusive vital signs sensor technologies, centered around the individual, to provide remote monitoring of the subject’s general health condition. Vital signs monitoring typically includes monitoring one or more of the following physical parameters: heart rate (HR), blood pressure (BP), respiratory rate (RR), core body temperature and blood oxygenation (SpCh).

[0003] Outpatient unobtrusive sensing systems are typically used to provide early detection of worsening of chronic medical conditions such as certain heart or respiratory conditions, or to detect episodic events such as periods of atrial fibrillation or heart palpitations. Unobtrusive vital sign monitoring can additionally or alternatively detect an onset of a new health condition. Patients exhibiting symptoms of a disease (e.g., COVID-19), for example, can wear such sensing systems to determine a condition of the patient. In a military context where unit readiness can be critical, early detection of onset of a communicable disease such as COVID can facilitate rapid isolation of infected patients and limit spread of the disease throughout the unit. The effectiveness of this approach relies on detecting the disease in its onset phase before the individual becomes highly contagious. However, these incipient symptoms can be subtle, and can be masked by individual variability of the monitored vital sign(s). Individuals can exhibit a wide range of intra-personal baseline values measured by the sensing systems. This variability, or noise, makes it hard for artificial intelligence (Al) models to train or learn, and accurately predict infections or to recognize early onset symptoms.

[0004] Current techniques for compensating for individual variability include normalizing the individual’s values by subtracting the baseline mean and dividing the by the baseline standard deviation. This standard process is also called z-normalization because it assumes the baseline data is a Gaussian (i.e.: normal) distribution, so that the normalized values are equivalent to a “z- value” used in statistical hypothesis testing.

[0005] The following discloses certain improvements to overcome these problems and others.SUMMARY

[0006] In one aspect, a system includes a monitoring device including at least one sensor. The wearable monitoring device is configured to be worn by an associated patient and the at least one sensor configured to measure physiological sensor data of the associated patient wearing the monitoring device. An electronic processor is integrated with or in wireless communication with the monitoring device. The electronic processor is programmed to: calibrate a personalized normalization process for the associated patient by applying a calibration process to physiological sensor data measured of the associated patient during a baseline time period; apply the personalized normalization process to physiological sensor data measured of the associated patient during an analysis time period to generate normalized physiological sensor data; analyze the normalized physiological sensor data to identify a condition of the associated patient; and output an indication of the identified condition of the associated patient.

[0007] In another aspect, a health monitoring method includes: with a wearable monitoring device including at least one sensor, measuring physiological sensor data of an associated patient wearing the monitoring device; with an electronic processor integrated with or in wireless communication with the monitoring device: calibrating a personalized normalization process for the associated patient by applying a calibration process to physiological sensor data measured of the associated patient during a baseline time period; applying the personalized normalization process to physiological sensor data measured of the associated patient during an analysis time period to generate normalized physiological sensor data; analyzing the normalized physiological sensor data to identify a condition of the associated patient; and outputting an indication of the identified condition of the associated patient.

[0008] One advantage resides in providing a process to reduce data variability in patient monitoring data due to inter-personal differences by applying a transformation that standardizes an individual’s data based on their respective baseline statistics.

[0009] Another advantage resides in providing a process to improve a normalization of patient monitoring data which does not assume the patient monitoring data is Gaussian-distributed.

[0010] Another advantage resides in computational efficiency relative to z-normalization.

[0011] Another advantage resides in providing a process to improve a normalization of patient monitoring data by considering an entire distribution of the baseline values, rather than just the mean and variance.

[0012] Another advantage resides in providing a process to a normalization of patient monitoring data that is faster and more efficient than current similar processes.

[0013] A given embodiment may provide none, one, two, more, or all of the foregoing advantages, and / or may provide other advantages as will become apparent to one of ordinary skill in the art upon reading and understanding the present disclosure.BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The disclosure may take form in various components and arrangements of components, and in various steps and arrangements of steps. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the disclosure.

[0015] FIGURE 1 diagrammatically shows a patient monitoring system in accordance with the present disclosure.

[0016] FIGURE 2 shows an example flow chart of operations suitably performed by the system of FIGURE 1.

[0017] FIGURE 3 shows an example of data collected by the system of FIGURE 1.DETAILED DESCRIPTION

[0018] As used herein, the singular form of “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise. As used herein, statements that two or more parts or components are “coupled,” “connected,” or “engaged” shall mean that the parts are joined, operate, or co-act together either directly or indirectly, i.e., through one or more intermediate parts or components, so long as a link occurs. Directional phrases used herein, such as, for example and without limitation, top, bottom, left, right, upper, lower, front, back, and derivatives thereof, relate to the orientation of the elements shown in the drawings and are not limiting upon the scope of the claimed invention unless expressly recited therein. The word “comprising” or “including” does not exclude the presence of elements or steps other than those described herein and / or listed in a claim. In a device comprised of several means, several of these means may be embodied by one and the same item of hardware.

[0019] Health-related unobtrusive sensing systems disclosed herein provide improved and personalized baseline correction that can correct for non-Gaussian variability in the data. For example, the disclosed approaches for baseline correction can correct for kurtosis and skewness in heart rate variability. This is done by considering the entire distribution of the baseline values, rather than just the mean and variance as done typically in z-normalization. In some illustrative approaches, the Empirical Cumulative Distribution Function (ECDF) of the baseline data is used in performing the baseline normalization. The improved baseline correction may enable the unobtrusive sensing system to detect more subtle vital sign changes due to onset of a condition such as an infection by a communicable disease such as COVID. While described in the illustrative embodiments in the illustrative application of a dedicated wearable medical monitoring device, the approaches disclosed herein can be employed more broadly, for example in conjunction with a smart watch having a heart rate monitor built into its back face that contacts the watchwearer’s wrist. In the event of an outbreak of a highly communicable disease such a smart watch or other wearable monitor worn by healthy individuals can enable early onset detection of illness, thus serving as an early warning system to enable timely quarantining or other measures to mitigate spread of the communicable disease.

[0020] With reference to FIGURE 1, a system 1 for monitoring an associated patient P is shown. As used herein, the term “patient” (and variants thereof) refers to, and includes, an outpatient, a discharged patient, a patient undergoing screening using the monitoring device as part of an annual medical checkup, a healthy person wearing a vital sign monitoring device such as a vital sign measurement-equipped smart watch for precautionary purposes, or other person whose health condition is to be monitored. As shown in FIGURE 1, the system 1 includes a wearable monitoring device 10 that is wearable by, or otherwise attached to, the patient P. The wearable monitoring device 10 can include any suitable monitoring device, such as a Mobile Cardiac Outpatient Telemetry (MCOT)® device (available from Philips ECG Solutions), or a medical wearable device, a torso-worn vital signs health patch, a wrist-worn watch, a chest strap, a smart garment, medical ear buds / over the ear, a forehead or nose sensor, a smart ring, or so forth.

[0021] The illustrative wearable monitoring device 10 includes a single-use electrodes patch 11 with a heart rate (HR) or an ECG sensor 12 (e.g., embodied as electrodes 12) for collecting cardiac data 13. While HR or ECG data is described herein as an illustrate example, the monitored vital sign data could additionally or alternatively be respiratory rate (RR) data, SpChdata, activity data (e.g., acquired by a wearable accelerometer), or so forth, or various combinations thereof. The illustrative monitoring device 10 includes an electronics module 16 that attaches onto the electrodes patch 11. The electronics module 16 includes at least one microprocessor or microchip (not shown) configured (e.g. programmed) to optionally preprocess the ECG signals from the ECG sensor (electrodes) 12 to produce the cardiac data 13. The device 10 including the electrodes patch 11 and the attached electronics module 16 is adhesively secured to the chest or other anatomy of the patient P after suitable preparation (e.g. cleaning and / or shaving) of the skin. The use of the separate electrodes patch 11 and electronics module 16 advantageously enables the single-use electrodes patch 11 to be a low-cost component that can be replaced as needed over the course of a patient monitoring session (which may extend over multiple days or weeks) while re-using the more expensive electronics module 16. However, other arrangements are contemplated such as having the patch 11 and electronics 16 constructed as a unitary single-use unit. While such skin preparation is typical for a medical monitoring device, it may be omitted in other applications. For example, a smartwatch may include backside ECG electrodes for HR monitoring that are applied with no skin preparation at all (i.e., the user simply puts the wristwatch on). Lack of skin preparation may increase variability of the acquired vital sign data, but the disclosed baseline correction approaches can advantageously at least partially compensate for this. Although not shown, it will be appreciated that the wearable monitoring device 10 also includes an on-board battery or other on-board electrical power source to power the electronics module 16. The on-board battery may, for example, be integrated with the electronics module 16, which may include a recharging port connector 17 for recharging the electronics module 16, or the electronics module 16 could be placed on a wireless inductive recharging station to recharge it if needed during a patient monitoring study.

[0022] The wearable monitoring device 10 more generally can include one or more sensors configured to measure physiological sensor data of the patient P wearing the monitoring device 10. The measured physiological sensor data of the patient P is used to detect a health-related event of the patient P. The HR sensor 12 is configured (e.g., comprising skin-contacting electrodes with silver / silver chloride coatings, for example) to measure cardiac data 13 of the patient P, and the health-related event comprises a cardiac event detected by analysis of the cardiac data 13. In addition, the HR sensor 12 can be replaced with (or supplemented by) any other suitable sensor tomeasure a corresponding vital sign of the patient (e.g., SpCh, electroencephalogram (EEG), and so forth).

[0023] The illustrative wearable monitoring device 10 also includes a wireless transmitter or transceiver 18 (referred to hereinafter as a transceiver 18), which may optionally be integrated with the electronics module 16. The transceiver 18 is integrated with or in wireless communication with the monitoring device 10 to transmit the patient data (e.g. the cardiac data 13) to a clinical health information system 20 comprising a server computer, for example transmitted over the Internet via a 3G / 4G / 5G wireless cellular network, Wi-Fi, various combinations thereof, and / or another wireless communication protocol. In some embodiments, the electronic processor 16 is configured to collect and optionally preprocess the cardiac data 13 from the ECG sensor 12, and the transceiver 18 is configured to transfer the cardiac data 13 to the clinical health information system 20. In a typical arrangement, the transceiver 18 is a low-power wireless transceiver (e.g., Bluetooth™, Zigbee™, or the like) that connects with low power to the clinical health information system 20, thus placing a low power draw on the on-board battery of the wireless monitoring device 10. In some examples, this information can be transmitted to a clinical health information system 20.

[0024] The processing of the patient data 13 can be variously distributed between the electronic module 16 of the wearable monitoring device 10 and the clinical health information system 20 including an electronic module or processor 21. Hence, an electronic processor 16, 20 is referred to herein, to indicate the combined processing capacity of the electronic module 16 of the wearable monitoring device 10 and the clinical health information system 20. In general, processing described as being performed by the electronic processor 16, 20 may be performed entirely by the electronic module or processor 16 of the wearable monitoring device 10, or entirely by the electronic module or processor 21 of the clinical health information system 20, or the processing may be shared between them.

[0025] As further diagrammatically indicated in FIGURE 1, the electronic processor 16, 21 can be configured to perform a health status monitoring method 100 of monitoring the patient P by analyzing the cardiac data 13.

[0026] With reference now to FIGURE 2, an illustrative embodiment of the health status monitoring method 100 is shown by way of a flowchart. To begin the method 100, the monitoringdevice 10 is atached to the patient P. At an operation 102, physiological sensor data (e.g., the cardiac data 13 from the heart rate sensor 12) is measured.

[0027] At an operation 104, a personalized normalization process is calibrated by applying a calibration process to physiological sensor data 13 during the baseline time period. It is typically assumed that the patient P is healthy during this baseline time period, and that this baseline time period is representative to the patient’s P normal healthy behavior.

[0028] To do so, the calibration process includes computing an ECDF of the physiological sensor data 13 measured during the baseline time period to generate a personalized baseline ECDF for the patient P.

[0029] At an operation 106, the personalized normalization process is applied to the physiological sensor data 13 during the analysis time period to generate normalized physiological sensor data. To do so, the personalized normalization process also includes performing a nonlinear and non-Gaussian normalization of the physiological sensor data 13 measured during an analysis time period that is different from the baseline time period using the personalized baseline ECDF to generate the normalized physiological sensor data. The personalized normalization process then includes applying the personalized baseline ECDF for the patient P to the physiological sensor data 13 during the analysis time period. For example, the personalized baseline ECDF is applied to the physiological sensor data 13 measured during the analysis time period to generate a transformed distribution, and computing the normalized physiological sensor data by applying a transform 22 to the transformed distribution.

[0030] In some embodiments, the transform 22 includes computing Shannon’s information entropy for the transformed distribution, and can be stored in the clinical health information system 20. This transformed distribution should be uniform, and therefore its entropy should be a maximum value possible. Any deviations from this baseline distribution will have a lower entropy value. All data from all patients P can be in a standardized range from 0 to 1 , with personalized transformation applied to each patient P. Moreover, if a patient’s P data is similar to its baseline behavior, its distribution is Uniform and its entropy close to baseline values. The Shannon’s entropy can be used to quantify the distance or deviation between the current data from the patient’s P baseline data.

[0031] In other embodiments, the physiological sensor data 13 measured of the patient P during a baseline time period as a set of n measurements {X}n= X1, X2, ... , Xnmeasured duringthe baseline time period. The baseline physiological sensor data has a cumulative distribution function F(x) where x is a measured physiological sensor value and F(x) is the statistical probability that a randomly acquired measurement X will be less than or equal to the actual measurement x, i.e. F(x) = P(X < x). The empirical cumulative empirical distribution (ECDF) for the baseline data {X}nmeasured during the baseline time period is:where lXi<xis the indicator of whether the baseline sample Xtis less than or equal to x. This is merely one formulation, and other variant formulations of the ECDF are contemplated, e.g. in the1 above definition the factor - could be replaced and / or the output of the above equationcould be smoothed, or the ECDF could be approximated by bins at fixed set regions (xb). Various statistical data can optionally be determined for the ECDF, such as the zeroth order moment (mean), first order moment (variance, or equivalently the standard deviation), and / or any higher-order moment of the ECDF (i.e., kurtosis, skewness etc.).

[0032] In this embodiment, the transform 22 includes analyzing the physiological sensor data 13 at one or more selected regions of the ECDF. For example, the transform 22 can include analyzing the physiological sensor data 13 ata selected region of 0-10% and 90-100% of the values of the physiological sensor data 13. In a more particular example, the transform 22 can include analyzing the physiological sensor data 13 at a selected region of 0-5% and 95-100% of the values of the physiological sensor data 13. Typically, the values of the physiological sensor data 13 at extreme regions (for example, 0-10% and 90-100%) are likely to be more indicative of a presence of a disease or condition or anomaly of the patient P relative to baseline values of the physiological sensor data 13. In some examples, the selection regions of the ECDF can be tuned by a calibration process using training data and performed by the electronic module or processor 16 of the wearable monitoring device 10, or entirely by the electronic module or processor 21 of the clinical health information system 20.

[0033] In a particular example, one or more selected regions of the personalized baseline ECDF are determined. A ratio of the one or more selected regions over the personalized baseline ECDF is calculated. A deviation of the calculated ratio is determined from a predetermined threshold. An indication of the identified condition of the patient P is output based on the determined deviation. To do so, the values of the physiological sensor data 13 in these regionscan be summed over a predetermined time interval (e.g., an hour, twenty-four hours, one or more days, and so forth). For a first patient, the physiological sensor data 13 is collected over predetermined time range (i.e., 10 days) and the regions of the ECDF can be selected to be 0-10% and 90-100%. Over this time period, the first patient has 50 heart beats in a range of 30-40 beats per minute (bpms). At a time after the predetermined time range (i.e., day 11), the physiological sensor data 13 of the first patient has two beats in a range of 30-40 bpms. The transform 22 processes this data as a ratio of the two beats in the range of 30-40 bpms over a total number of data points in the physiological sensor data 13 (i.e., 50) to determine a region of 5% (i.e., 2 / 50). This calculated ratio of 5% is compared to a predetermined threshold (i.e., a value of 1) to determine whether the calculated ratio comprises a deviation (i.e., is in the selected regions of the ECDF). In another example, the predetermined threshold can be determined by the Al model 24 using training data or set manually by a user. If the deviation exists, then the patient P is determined to have a condition, and an indication of the condition of the patient P is output. Advantageously, this allows the processing of the physiological sensor data 13 by the transform 22 to be performed more quickly (i.e., by analyzing fewer data points) and more efficiently (i.e., by without performing complex computations, and by analyzing those values in data regions sensitive to anomalies or diseases or conditions of the patient P).

[0034] At an operation 108, the normalized physiological sensor data is analyzed to identify a condition of the patient P, and at an operation 110 an indication of the identified condition is output (e.g., a displayed message, an audio message, a push notification to a mobile device, and so forth). To do so, a pre-trained artificial intelligence (Al) model 24 (stored in the clinical health information system 20) is configured to analyze to the normalized physiological sensor data. In some examples, the Al model 24 is pre-trained to detect the condition comprising an illness and / or onset of a viral or bacterial infection of the patient P.

[0035] A training method 200 of training and generating the Al model 24 can be performed by a training processor (e.g., the electronic module or processor 21 of the clinical health information system 20 or the electronic module or processor 16 of the wearable monitoring device 10. At an operation 202, a plurality of training physiological datasets from a plurality of training subjects are received at the training processor 16, 21. At an operation 204, each training physiological dataset is converted to a normalized training physiological dataset using the calibration process described above for the operation 104, so that the data of the trainingphysiological dataset from each training individual undergoes the same personalized normalization process (personalized to that specific individual). At an operation 206, an Al model is trained on the normalized training physiological datasets to generate the pre-trained Al model 24. After the calibration period, any new data from the patient P is passed through the learned transform 22 so that all data from all patients are normalized to the same range and distribution. This makes the training of the Al model 24 more reliable and its deployment more accurate.

[0036] In one example of the personalized normalization to provide the baseline correction, during a calibration period for each user, statistics are collected related to each user’s baseline period. It is assumed that the user is healthy during this calibration period (or at least that sick periods are very small relative to the baseline window), and that the calibration period is representative of the user’s normal healthy behavior.

[0037] After the calibration period is over an efficient normalizing / standardizing transformation is created for each user, and this is stored in a database. A nonlimiting illustrative example of a suitable normative / standar dizing transformation creation for a user is as follows. The user’s ECDF of their baseline data is computed and stored as ranked bins, allowing for easy computation and updates. The baseline data should have a uniform distribution after it is passed through the ECDF. Next, the patient’s Shannon’s information entropy for the baseline data after it is passed through the ECDF is computed and stored. This transformed distribution should be uniform, and therefore its entropy should be a maximum value possible. Any deviations from this baseline distribution will have a lower entropy value.

[0038] In one nonlimiting illustrative approach, the current daily data distribution is passed through the user's personalized baseline ECDF. If the current daily data is similar to the baseline data, the transformed distribution should be uniform. Next, the Shannon’s information entropy is computed after the data is passed through the personalized ECDF. Any deviation from the baseline values will yield results that are smaller than those obtained for the baseline data acquired in the calibration period. An example of this normalization is plotted in the third graph of FIGURE 3.

[0039] As the ECDF has a range of 0-1, all data from all users is in the standardized range from 0 to 1, with personalized transformation applied to each user. This facilitates comparison between individuals, for example in operation of the Al model 24. Moreover, if a user’s data is similar to its baseline behavior, its distribution is uniform and its entropy close to baseline values. The Shannon’s entropy step quantifies the distance or deviation between the current data from theuser’s baseline data. Other approaches can be used instead of Shannon’s entropy for this quantification, such as a non-linear, non-Gaussian normalization applied after passing the data through the personalized ECDF.

[0040] A reset mechanism is optionally provided, in which these steps are repeated on an on-going basis as desired to re-establish the baseline personalized ECDF. This allows the system to reset and recompute the baseline if a drastic or switch-point event is noticed for the user, or on an ongoing basis to minimize baseline drifts.

[0041] FIGURE 3 shows an example of the output of the method 100 applied to real heart rate variability data. The top graph shows daily heart rate variability for the patient P. The patient P is given a vaccine at day 0. The baseline data encompasses all days up to day 0. The second graph shows the baseline distribution (designated with 2 and 3), which is clearly not Gaussian (high skewness). The curve on the second plot (designated as 4) is the ECDF. This is the transform 22 for the patient P that will normalize their data relative to the baseline. The third graph shows the data passed through the ECDF and summarized through its entropy statistics. That last graph on the bottom shows the standard classical approach using z-normalization for comparison purposes.

[0042] The disclosure has been described with reference to the preferred embodiments. Modifications and alterations may occur to others upon reading and understanding the preceding detailed description. It is intended that the exemplary embodiment be construed as including all such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.

Claims

CLAIMS:

1. A system (1) comprising: a monitoring device (10) including at least one sensor (12), the wearable monitoring device configured to be worn by an associated patient and the at least one sensor configured to measure physiological sensor data (13) of the associated patient wearing the monitoring device; an electronic processor (16) integrated with or in wireless communication with the monitoring device, the electronic processor programmed to: calibrate a personalized normalization process for the associated patient by applying a calibration process to physiological sensor data measured of the associated patient during a baseline time period; apply the personalized normalization process to physiological sensor data measured of the associated patient during an analysis time period to generate normalized physiological sensor data; analyze the normalized physiological sensor data to identify a condition of the associated patient; and output an indication of the identified condition of the associated patient.

2. The system (1) of claim 1, wherein the calibration process includes: computing an empirical cumulative distribution function (ECDF) of the physiological sensor data measured of the associated patient during the baseline time period to generate a personalized baseline ECDF for the associated patient; wherein the personalized normalization process includes applying the personalized baseline ECDF for the associated patient to the physiological sensor data measured of the associated patient during an analysis time period.

3. The system (1) of claim 2, wherein the personalized normalization process further includes: performing a non-linear and non-Gaussian normalization of the physiological sensor datameasured of the associated patient during an analysis time period using the personalized baseline ECDF to generate the normalized physiological sensor.

4. The system (1) of either one of claims 2 and 3, wherein the personalized normalization process includes: applying the personalized baseline ECDF to the physiological sensor data measured of the associated patient during an analysis time period to generate a transformed distribution; and computing the normalized physiological sensor data by applying a transform (22) to the transformed distribution.

5. The system (1) of claim 4, wherein the transform (22) includes computing: determining one or more selected regions of the personalized baseline ECDF; calculating a ratio of the one or more selected regions over the personalized baseline ECDF; determining a deviation of the calculated ratio from a predetermined threshold; and outputting an indication of the identified condition of the associated patient based on the determined deviation.

6. The system (1) of any one of claims 1-5, wherein the normalized physiological sensor data is analyzed to identify the condition of the associated patient by applying pre-trained artificial intelligence (Al) model (24) to the normalized physiological sensor data.

7. The system (1) of claim 6, further comprising: a training electronic processor (16, 21) programmed to generate the pre-trained Al model (24) by operations including: receiving a plurality of training physiological datasets from a plurality of training subjects; converting each training physiological dataset to a normalized training physiological dataset using the calibration process; and training an Al model on the normalized training physiological datasets to generate the pre-trained Al model.

8. The system (1) of either one of claims 6 and 7, wherein the pre-trained Al model (24) is pre-trained to detect the condition comprising an illness of the associated patient.

9. The system (1) of either one of claims 6 and 7, wherein the pre-trained Al model (24) is pre-trained to detect the condition comprising onset of a viral or bacterial infection.

10. The system (1) of any one of claims 1-9, wherein the at least one sensor (24) comprises a heart rate sensor.

11. A health monitoring method (100) comprising: with a wearable monitoring device (10) including at least one sensor (12), measuring physiological sensor data (13) of an associated patient wearing the monitoring device; with an electronic processor (16) integrated with or in wireless communication with the monitoring device: calibrating a personalized normalization process for the associated patient by applying a calibration process to physiological sensor data measured of the associated patient during a baseline time period; applying the personalized normalization process to physiological sensor data measured of the associated patient during an analysis time period to generate normalized physiological sensor data; analyzing the normalized physiological sensor data to identify a condition of the associated patient; and outputting an indication of the identified condition of the associated patient.

12. The method (100) of claim 11, wherein the calibration process includes: computing an empirical cumulative distribution function (ECDF) of the physiological sensor data measured of the associated patient during the baseline time period to generate a personalized baseline ECDF for the associated patient; wherein the personalized normalization process includes applying the personalized baseline ECDF for the associated patient to the physiological sensor data measured of the associated patient during an analysis time period.

13. The method (100) of claim 12, wherein the personalized normalization process further includes: performing a non-linear and non-Gaussian normalization of the physiological sensor data measured of the associated patient during an analysis time period using the personalized baseline ECDF to generate the normalized physiological sensor.

14. The method (100) of either one of claims 12 and 13, wherein the personalized normalization process includes: applying the personalized baseline ECDF to the physiological sensor data measured of the associated patient during an analysis time period to generate a transformed distribution; and computing the normalized physiological sensor data by applying a transform (22) to the transformed distribution.

15. The method (100) of claim 14, wherein the transform (22) includes computing: determining one or more selected regions of the personalized baseline ECDF; calculating a ratio of the one or more selected regions over the personalized baseline ECDF; determining a deviation of the calculated ratio from a predetermined threshold; and outputting an indication of the identified condition of the associated patient based on the determined deviation.

Images