WO2021074102A1

WO2021074102A1 - System and method for detection of intermittent claudication in remote sensor data

Info

Publication number: WO2021074102A1
Application number: PCT/EP2020/078690
Authority: WO
Inventors: Sabine Mollus; Salvatore SAPORITO
Original assignee: Koninklijke Philips N.V.
Priority date: 2019-10-16
Filing date: 2020-10-13
Publication date: 2021-04-22
Also published as: EP4044903A1

Abstract

A system and method for detecting a health condition receives data derived from a plurality of sensors, determines an activity pattern of a subject based on the data, determines a vital sign pattern of the subject based on the data, receives location information of the subject, and detects that the subject has the health condition based on the activity pattern, the vital sign pattern, and the location information. The sensors may include an activity sensor and a vital sign sensor located in at least one device worn or carried by the subject. The health condition may be intermittent claudication. In addition, the system and method may transmit a notification of the detected intermitted claudication to one or more devices.

Description

SYSTEM AND METHOD FOR DETECTION OF INTERMITTENT

CLAUDICATION IN REMOTE SENSOR DATA

TECHNICAL FIELD

[0001] This disclosure relates generally to processing information, and more specifically, but not exclusively, to managing the allocation and cost of providing healthcare resources.

BACKGROUND

[0002] Exertion induced pain of the lower limb which resolves with rest is a common clinical problem. Different causes may lead to this clinical symptom called intermittent claudication (IC). Clinicians differentiate vascular (peripheral arterial occlusive disease, venous occlusive disease or spinal stenosis with neurogenic claudication) and nonvascular (orthopedic, rheumatologic, neurologic) disorders. However, peripheral artery disease (PAD) is the most common cause for IC. PAD is an abnormal narrowing of the peripheral arteries commonly found in the elderly. According to studies, every fourth patient over the age of 55 is affected. For some patients, PAD occurs without significant pain, usually in the beginning stages. As PAD worsens, claudication intensifies. Treatments are available, but early detection and prevention is crucial in order to avoid complications and degradation. No approaches that attempt to detect claudication under free-living conditions exist so far.

SUMMARY

[0003] A brief summary of various example embodiments is presented below. Some simplifications and omissions may be made in the following summary, which is intended to highlight and introduce some aspects of the various example embodiments, but not to limit the scope of the invention. Detailed descriptions of example embodiments adequate to allow those of ordinary skill in the art to make and use the inventive concepts will follow in later sections. [0004] In accordance one or more embodiments, a method for detecting a health condition includes receiving data from a first sensor of a subject, determining at least one pattern based on the first sensor data, identifying a candidate based on data from a second sensor on the subject, relating the candidate to the at least one pattern to determine occurrence of an event, generating a score for the event and comparing the score to a predefined threshold, and detecting that the subject has the health condition event based on a result of the comparison, where the health condition is intermittent claudication. The data from the first sensor may be indicative of activity of the subject. The pattern may include a walking pattern of the subject. The first sensor may be an accelerometer, barometric pressure sensor, gyroscope, or another type of physical activity sensor. The second sensor may be a vital sign sensor. Examples include a sensor generating electrocardiogram (ECG) data or photo-plethysmography (PPG) data. In one embodiment, the second sensor may be a heart rate sensor and/ or a sensor that measures other characteristics of a cardiac vital sign (e.g. ECG, PPG) such as pulse rise time, fullwidth half max, interpeak time, time to peak, pulse downslope time, pulse arrival time, and signal baseline level.

[0005] The candidate may be identified based on the data from a cardiac activity sensor and relating the candidate to the pattern may include determining a correspondence between the walking pattern and the candidate determined based on the data from the cardiac vital sign sensor. Identifying the candidate may include generating a predicted heart rate (or other cardiac vital sign signal characteristic) based on the data from the cardiac vital sign sensor and a predetermined activity pattern of the subject stored in a database, computing a similarity value between the expected heart rate (or other cardiac vital sign signal characteristic) and the data from the heart rate sensor, and identifying the candidate based on the similarity value. The score may be indicative of a probability or risk that the subject has intermittent claudication. At least one of the first or second sensor may be on a wearable device of the subject. [0006] In accordance with one or more other embodiments, a method for detecting a health condition includes receiving data from a first sensor of a subject, determining a profile of the subject from the first sensor data, mapping the profile to a pattern derived from second sensor data, identifying a candidate for the health condition based on the mapping, generating a score for the candidate and comparing the score to a threshold, and detecting that the subject has the health condition based on a result of the comparison, where the second sensor is an activity sensor and wherein the health condition is intermittent claudication. The data from the first sensor may be indicative of at least one location of the subject. The profile may be a motion profile of the subject. The pattern derived from the second sensor data may include a walking pattern.

[0007] The method may include determining a similarity between the profile and the pattern, wherein identifying the candidate for the health condition is performed based on the similarity. Determining the profile may include comparing the first sensor data to one or more pre-stored profiles of the subject; and determining the profile of the subject based on the comparison.

[0008] In accordance with one or more other embodiments, a method for detecting a health condition may include receiving data derived from a plurality of sensors, determining an activity pattern of a subject based on the data, determining a vital sign pattern of the subject based on the data, receiving location information corresponding to the subject, and detecting that the subject has the health condition based on the activity pattern, the vital sign pattern, and the location information, where the plurality of sensors includes an activity sensor and a vital sign sensor and a location sensor wherein at least one device worn or carried by the subject and wherein the health condition is intermittent claudication. Determining the activity pattern may include comparing the data of the activity sensor to predetermined activity data of the subject stored in a knowledge base, and determining the activity pattern based on results of the comparison. The vital sign pattern may include a heart rate pattern. [0009] The method may include generating a predicted heart rate based on the data from the activity sensor and predetermined heart rate templates of the subject or of a population stored in the knowledge base or a database, wherein the heart rate pattern is determined based on the expected heart rate and the predetermined reference activity data of the subject stored in the knowledge base. The method may include transmitting a notification of the detected intermittent claudication to at least one predetermined device. The at least one device may be a device of the subject, a device of medical personnel, or device of a guardian or other person known by the subject.

BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying figures, where like reference numerals refer to identical or functionally similar elements throughout the separate views, together with the detailed description below, are incorporated in and form part of the specification, and serve to further illustrate example embodiments of concepts found in the claims and explain various principles and advantages of those embodiments.

[0011] These and other more detailed and specific features are more fully disclosed in the following specification, reference being had to the accompanying drawings, in which:

[0012] FIG. 1 illustrates an embodiment of a system for detecting a health-related condition;

[0013] FIG. 2 illustrates an embodiment of a method for detecting intermittent claudication in (and/or other conditions of) a subject who, for example, may have peripheral artery disease or some other vascular or non-vascular condition;

[0014] FIG. 3 illustrates an example of system and method embodiments described herein;

[0015] FIG. 4 illustrates an embodiment of a method for detecting intermittent claudication events based on a first combination of sensor data;

[0016] FIG. 5 illustrates an embodiment of a method for detecting activity and/ or behavioral patterns for a subject; [0017] FIG. 6 illustrates an embodiment of a method for detecting intermittent claudication based on a different combination of sensor data; and

[0018] FIG. 7 illustrates an embodiment of a method for collecting patterns from historical data indicating physical activity and location data of the subject being monitored.

DETAILED DESCRIPTION

[0019] It should be understood that the figures are merely schematic and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the figures to indicate the same or similar parts.

[0020] The descriptions and drawings illustrate the principles of various example embodiments. It will thus be appreciated that those skilled in the art will be able to devise various arrangements that, although not explicitly described or shown herein, embody the principles of the invention and are included within its scope. Furthermore, all examples recited herein are principally intended expressly to be for pedagogical purposes to aid the reader in understanding the principles of the invention and the concepts contributed by the inventor(s) to furthering the art and are to be construed as being without limitation to such specifically recited examples and conditions. Additionally, the term, “or,” as used herein, refers to a non-exclusive or (i.e., and/ or), unless otherwise indicated (e.g., “or else” or “or in the alternative”). Also, the various example embodiments described herein are not necessarily mutually exclusive, as some example embodiments can be combined with one or more other example embodiments to form new example embodiments. Descriptors such as “first,” “second,” “third,” etc., are not meant to limit the order of elements discussed, are used to distinguish one element from the next, and are generally interchangeable. Values such as maximum or minimum may be predetermined and set to different values based on the application. [0021] A number of wearable devices have been developed with sensors to monitor of activity, mobility, locality, and health parameters (e.g., heart rate (HR), blood pressure, skin conductance). In elderly populations, these devices attempt to perform early detection or the occurrence of health issues. The issues may be known or unknown to the wearer or may be ones that trigger unspecified symptoms that map to slight and subtle changes in gait or mobility characteristics. For example, intermittent claudication causes pain that triggers a rise in HR/ skin conductance/ blood pressure and sudden stops of activity. These incidences may occur in a variety of circumstances, but are easier to detect in free-living conditions during longer steady-state walks in a relatively distraction-free environment because the subject is more attuned to his physical state during these times.

[0022] Example embodiments describe a system and method for detecting health conditions of a subject. The health condition may be, but is not limited to, a clinical symptom and/or another condition or state related to health. One example is intermittent claudication in a person who has peripheral artery disease or some other vascular or non-vascular (e.g. orthopedic) condition. The system and method are performed based on signals from one or more sensor-bearing device, at least one of which may include, for example, a wearable device.

[0023] The system may include an expert system that determines pre-defmed behavior patterns in sensor data, detects intermittent claudication based on those patterns, and then initiates notifications or alerts for the subject being monitored, caregivers, and/ or other personnel or autonomous systems. One implementation of the expert system analyzes acceleration data to detect phases of activity (motion patterns) of the subject over a period of time. In one particular embodiment, a gait temporal or spectral analysis may be performed based on data from a tracker and other types of mobility monitoring devices and patterns may be detected based on multiple channels (that receive activity, heart rate, and/or location data) which are processed to identify intermittent claudication. Since intermitent claudication causes pain, measuring changes in heart rate, skin conductance, and/ or blood pressure may be especially beneficial for at least some embodiments.

Peripheral Artery Disease

[0024] Peripheral artery disease (PAD) is an abnormal narrowing of the peripheral arteries caused by atherosclerosis. Due to arterial occlusion, lower limb muscles do not receive the oxygen required while exercising. This creates pain which often motivates the subject to stop exercising. The pain typically resolves with rest; however, the underlying PAD problem remains. The pain that is often associated with PAD is commonly referred to as intermittent claudication (IC), which limits daily physical activities and impairments in health-related quality of life. PAD is especially common in the elderly, with every fourth patient over the age of 55 affected. Early detection and prevention is crucial to avoid complications and progression of the disease. Moreover, individuals with PAD have an elevated risk for falls and cardiovascular events. The associated mortality risk is high.

[0025] In the case of PAD specifically, early detection is complicated by the fact that the disease progresses with no or very slight symptoms. In the first stage of the disease, only occasional pain in the lower limbs occurs. In the second or subsequent stages, intermittent claudication manifests but often escapes detection because the cause of the pain is mis -interp eted as orthopedic issues. One study has shown that every fifth patient over 65 years visiting the doctor had PAD without knowing. A conservative (medication and lifestyle intervention based) therapy is only successful in the early stages. Then more invasive and more costly (revascularization) therapies have to be performed. The embodiments described herein are effective for screening, early diagnosis of PAD and monitoring of evolution of this disease. However, the embodiments are not to be limited only to detection of a PAD- related conditions. Furthermore, the described techniques can be used for performing fall risk assessment. [0026] FIG. 1 illustrates an embodiment of a system for detecting a health-related condition (e.g., intermittent claudication in a subject who, for example, may have peripheral artery disease or some other vascular or non-vascular (orthopedic) condition. The system includes an expert system 10, a knowledge base 40 (or at least a link to the database), and an output device 50.

[0027] The expert system 10 receives signals from one or more sensors 301 to 30N through one or more corresponding interfaces 20. All or a portion of the sensors may be included in a wearable device. Examples of wearable devices which may include the sensor(s) 301 to 30N include, but are not limited to, Philips GoSafe, FitBit, Apple Watch, etc. The sensors 301 to 30N may also include various types of indoor and/or outdoor location sensors, activity (motion) sensors, vital signs sensors, and/or sensors for tracking a subject of interest. The data from these sensors may be used, for example, to perform plausibility checks.

[0028] In one or more embodiments, the sensor(s) may include an accelerometer, a heart rate monitor, a blood pressure sensor, a skin conductance sensor, and/ or one or more other types of sensors. The signals may be received through corresponding interface(s) 20 in a variety of ways. For example, one of the sensors may be coupled to the expert system through a wire. Another sensor may be wirelessly coupled to the expert system based on a predetermined communication protocol. The protocol may conform, for example, to a Bluetooth standard or another type of wireless protocol. [0029] In one embodiment, one or more of the sensor(s) 301 to 30N may include, or be included in, a wearable device such as a health or activity tracker. The sensor(s) may include activity sensors and/ or contextual sensors. The activity sensors may be one or more accelerometers, barometric pressure sensors, and/ or other sensors that detect data relating to the activity and/ or body orientation of the subject, or even whether the subject is in a stationary or sitting position. The contextual sensors may generate data indicative of location, heart rate, blood pressure, pressure/ skin conductance, and/ or other health related or context related data of the subject. The continuous, periodic, or event-driven monitoring performed by these sensors may play a key role for early detection of health/condition- specific patterns relating to PAD. Analysis of the Philips Lifeline subscribers’ mobility data is one option.

[0030] Especially when activity (mobility) data is coupled with other sensor data (such as location or heart-rate/blood pressure/ skin conductance), subtle health issues evolving over time may be detected which otherwise would not be recognized. Therefore, in accordance with one or more embodiments, no clinical tests may need to be conducted. Instead, when analyzed, the sensor data may provide insight for disease based on the monitored detection of everyday life activities.

[0031] The knowledge database 40 stores the sensor and/ or other types of data for use by the expert system in performing pattern recognition. In one embodiment, at least some of the sensor data may be processed by the expert system (and/ or a processor in a device including an associated one of the sensors) to generate one or more predetermined patterns indicative of a type of activity, body orientation, health condition, symptom or disease, or other status of the subject that may or may not be coincident with a health-related event. Examples include predetermined pattern activity indicative of walk-stop patterns, stop-go patterns, walk-stop-go patterns, condition-specific walking patterns, and body pose patterns. A more detailed discussion of these patterns is provided below.

[0032] The location sensor data may provide supplemental contextual information to the observed activity patterns of subject being monitored. For example, stop-and-go patterns might be common/natural for a person performing chores, cooking in the kitchen, or doing other tasks. The location sensor data may be combined with historical data and/ or individual habit information to generate pattern data or other indicia used to detect an intermittent claudication event or other health- related condition, when, for example, a person is walking alone in a distraction-free environment (e.g., heading alone to the bus stop). [0033] Examples of other data stored in the knowledge base 40 is related to pain. Pain may be an indicator for various types of pathological conditions and, for example, may manifest in vital signs of the subject. When pain occurs, physiological parameters such as blood pressure, heart-rate, breathing rate, skin conductance and other indicia are affected. Changes in these parameters (and/ or patterns derived from them) may provide an indication of a health-related episode or condition. Because dedicated sensors can assess these changes, pain-related and other forms of data stored in the knowledge base may be used by the expert system to perform automatic incident detection.

[0034] The expert system 10 includes a processor 14 and at least one memory 15 for storing instructions for implementing one or more algorithms. When executed, the processor performs various operations including processing the sensor data for purposes of detecting the various patterns described herein, in order to detect intermittent claudication or other health-related conditions. In one embodiment, processor 14 processes the activity and/ or contextual sensor data to identify patterns or other information and then compares the patterns or other information to predetermined patterns or other data stored in the knowledge base 40. These operations may be performed by the processor 14 using pattern recognition, pattern matching, classification, and/ or other techniques embodied in the instructions stored in memory 15. Based on this comparison, the processor 14 may determine the status of the patient, including whether the patient is currently experiencing intermittent claudication and/ or other features related to PAD.

[0035] The output device 50 may be a smartphone, tablet, computer, or another type of device capable of outputting the analysis results of the expert system. In one embodiment, the output device may be the wearable device that includes the one or more sensors which generate the sensor data analyzed by the expert system. The results may be output in a variety of ways, including but not limited to graphics information, textual information, instant or text messages, alarms, notifications, and/ or other forms of output indicating results of the analysis performed by the expert system. [0036] In one embodiment, the results may be output to the patient himself (e.g., to the patient device through interface 20) and/ or may be output to a caregiver, guardian, or other person or entity through the same or another interface 60. In this case, a reporting loop may be formed that notifies the person wearing the wearable device including the sensor(s), family care -givers, formal care -givers (e.g., Home Instead, Guardian Angels, etc.), or other entity (e.g., automated or manned monitoring agency) of the occurrence of an incident relating to a certain evolving condition. Once informed, action may be taken. Several forms of reporting and data aggregation are possible. Examples include risk prediction scores and qualitative notifications as described in greater detail below.

[0037] In addition to intermittent claudication, another risk or condition associated with PAD is an increased risk of a patient (especially the elderly) falling. This embodiment might provide further information to compute an individual fall risk score. The information sent to the output device 50 and/ or through the reporting loop may provide sufficient notice to take precautions to prevent the patient from falling during intermittent claudication or other symptomatic episode. The information on the output device may also prompt patients to take preemptive action to prevent progression of the disease and/ or to prevent or better manage potentially severe health events.

[0038] FIG. 2 illustrates an embodiment of a method for detecting intermittent claudication in (and/ or other conditions of) a subject who, for example, may have peripheral artery disease or some other vascular or non-vascular condition. The method may be performed, in whole or part, by the system of FIG. 1 or another system.

[0039] Referring to FIG. 2, in this embodiment the subject 201 is wearing an activity tracking device 205 equipped with one or more sensors as previously described. For illustrative purposes, it is assumed that the sensor(s) include one or more accelerometers that generate data 210 that tracks the activity (motion) and/or orientation of the subject. The activity may be walking (or running) along with associated stop and start states. The sensor data output to the expert system 10 may be in the form of a signal waveform with data points or a signature that tracks the activity. For example, when the subject is walking at a steady pace, the signal waveform may have a substantially regular signature or pattern with peaks that occur at nearly equal intervals, e.g., as shown by curve 215. When the subject slows down or speeds up, the distance between the peaks of the curve may change in a proportional manner. When the subject stops, the signal waveform may reflect this activity, for example, in the form of a lower and/ or substantially constant magnitude with no substantial peaks.

[0040] Another sensor worn by the subject may be a vital signs sensor, such as a heart rate (HR) sensor. This sensor generates HR data 220 and may be included in the same wearable device 205 as the accelerometer(s) or a different device. In this example, the heart rate sensor is included in a sensor- equipped watch 225 worn by the subject. The signal waveform generated by the heart rate sensor is also output to the expert system 10.

[0041] Additional sensor data output to the expert system 10 may include location (or localization) data 230. The location data may include or be based on GPS data, location services data, mobile communication system location data, or other position or location monitoring data. In FIG. 2, the location data is optionally expressed in a digital map 235 in the vicinity of the subject whose activity is being tracked. The digital map may include an indicator 236 corresponding to the current location of the subject. In the favored embodiment, the location data may be expressed in coordinates.

[0042] One purpose of the location data 230 is to establish a pattern of behavior of the subject that can be used as a baseline or reference for comparison to sensor data patterns to detect intermittent claudication. The location data may be stored in the knowledge base 40 in raw form and/ or may be processed (e.g., by the expert system or an external processor) to extract the behavioral pattern(s). The location data may be generated, for example, by sensors on the wearable device 210 or watch 220 or another device or system that operates, for example, based on smartphone or other technology. In one embodiment, the location data may be processed by the expert system to train a model or establish predetermined location and behavior paterns of the subject, and also may be subsequently processed along with the activity data and HR sensor data to detect a claudication event or other health condition. [0043] The expert system 10 may perform pattern detection 240 based on all or a portion of the activity data 210, HR data 220, and localization data 230. In one embodiment, the expert system may perform pattern detection based on this data and additional contextual information 260 corresponding to any of the types described herein. The contextual information may include factors relating to health status of the subject, including an indication of the medications that are being taken. The contextual information may be derived, for example, from electronic medical records derived from hospitals, doctors, and other healthcare-related or medical facilities. The expert system 40 may compare the data (e.g., received in real-time) and the contextual information to the patterns and reference information stored in the knowledge base 40 to identify or predict a health-related episode or condition of the subject, including but not limited to intermittent claudication.

[0044] Once the expert system 10 has rendered a decision on pattern detection, the processor 14 of the expert system may perform one or more actions. These actions include generating a notification or alert 250 to the subject being monitored, to one or more healthcare professionals or caregivers, to a guardian, and/or other persons who have a personal or medical relationship with the subject. The alert may be transmitted to a user device, the activity tracker, or heart rate monitor worn by the subject and/ or to a user device or workstation participating with the system. The notification can be in a graphical, textual, or audio format and may be sent over a wired or wireless network, of if the expert system is included in the activity monitor itself to a display screen of the activity monitor. As previously indicated, the activity monitor can be embodied in a smartphone or other type of device, such as a GoSafe, Fitbit, etc.

[0045] FIG. 3 illustrates an example implementation of system and method embodiments described herein. In this example implementation, the expert system 10 performs pattern detection 240 based on all of the activity data 210, HR data 220, and localization data 230. In an initial operation, the expert system 40 (which may be remotely located from the subject and the sensors), receives data from one or more accelerometer sensors on the subject being monitored. The data is in the form of acceleration data 270, which in this case tracks the movement of the subject as indicated by corresponding waveforms generated in a received data stream.

[0046] In this example, the acceleration data reflect active patterns 271a, 271b, and 271c and inactive patterns 272a and 272b. The active patterns have peaks arranged with substantially similar spacing that exceed a predetermined threshold level. When the peaks are arranged with an approximately regular spacing (as shown), the active patterns may be identified as walking patterns. The identification may be made, for example, with comparison to predefined walking patterns of the subject (or generally recognized) stored in the knowledge base 40. Different walking patterns (with different peak spacings) may be stored in the knowledge base to identify different types of walking activity. The inactive patterns 272a and 272b have a substantially constant level that is significantly below the predetermined threshold level. These patterns may therefore be indicative of periods where the subject is in a stationary position. A stationary position may be, for example, be a position where the subject is either sitting, standing, or standing still. Based on results of the comparison to the knowledge base, the acceleration data corresponding to the active patterns may be labeled by the expert system as “Walking,” “Standing Still,” or “Standing.” Other labels and/ or types of activity or motion of the subject may be detected in other embodiments, as discussed in greater detail below.

[0047] The cardiac vital sign sensor data might be provided in form of a waveform 280. During the conversion, one or more pre-processing (e.g., filtering) operations may be performed to generate smoothed HR data indicated by the waveform. As illustrated in FIG. 3, the amplitude of the waveform is typically related to the walking speed of the subject being monitored. For example, the waveform 280 has a first level during the active walking pattern 271a, a second level during the active walking pattern 27 lb, and another level during the active walking pattern 271c. Times just prior to the inactive patterns 272a and 272b may reflect an initial spike related to a transient heart rate increase at the time of standing or standing still. During the inactive pattern time periods, the heart rate waveform 280 experiences a reduction. The activity data can be processed by the expert system 10 to generate a second waveform 285 indicative of the expected heart rate. Expected heart rate may be determined using physical activity level, for example, in manner discussed in greater detail below.

[0048] The location data received by the expert system 10 may be processed and synchronized with the activity data and heart rate data, to provide an additional basis for detecting a claudication event. During the period of the first active walking pattern 271a, the location data indicates that the subject is at a location 290a. During the period of the second active walking pattern 27 lb, the location data indicates that the subject has advanced to location 290b. And, during the period of the third active walking pattern 271c, the location data indicates that the subject has advanced to location 290c. Based on the activity (acceleration) data, heart rate data, and the location data, the expert system 10 may detect whether the subject is or is not experiencing intermittent claudication (or other health-related event) based on one of the algorithms described below.

[0049] FIG. 4 illustrates an embodiment of a method for detecting intermittent claudication events based on a first combination of sensor data of FIGS. 2 and 3, namely activity data and heart rate data. The method may be implemented, in whole or part, based on instructions stored in memory 15 of the expert system 10. The instructions may be executed by the processor 14 of the expert system to perform detection based on the activity and heart rate data.

[0050] Referring to FIG. 4, the method includes, at 301, the expert system 10 (e.g., processor 14) acquiring and pre-processing data from one or more activity sensors monitoring a subject of interest. In this example, the sensor data is derived from at least one accelerometer that measures movement/physical activity of the subject. The accelerometer data may take various forms. In one embodiment, the accelerometer data includes three-dimensional data indicating movement of the subject in three directions (x, y, and z). In pseudocode, this may be indicated by the notation acc_x,y,z(n), where n is the sampling index of a discrete data input stream Furthermore, the accelerometer data can be represented by the vector norm.

[0051] At 303, the processor 14 detects episodes of physical activity (PA) and steady state walking of the subject based on the accelerometer data. The processor may perform these detection operations, for example, based on comparing the accelerometer data to one or more predetermined thresholds in the time or frequency domain, as indicated by the accelerometer data corresponding to the active patterns in FIG. 3. Once the physical activity or walking episodes (or other active or inactive periods) have been detected, the processor may label the activity data. In pseudocode, this may be indicated by the notation label_PA_episodes(n).

[0052] At 305, the processor detects state and feature changes within the detected activity and walking episodes. The state and features changes may be caused, for example, by variations in walking speed, intensity, regularity, and/or other associated characteristics or parameters. The state and features changes may be detected, for example, based on an analysis of the speed, intensity, regularity, symmetry of gait etc., of the accelerometer data. In pseudocode, this may be indicated by the notation label_walks(n), which may identify, for example, the types of walking of the subject.

[0053] At 307, the processor detects patterns in sequences of labelled walking episodes. The patterns may be detected, for example, by grouping data points of the accelerometer data to form signal waveforms, which, for example, may correspond to the active patterns 271a to 271c and inactive patterns 272a and 272b in FIG. 3. The amplitudes, spacings, and/or other features of the signal waveforms gathered in the output of label_walks(n) maybe compared with corresponding thresholds in order to detect the occurrence and types of patterns. In one embodiment, a clustering technique may be used to group the accelerometer data for purposes of generating signal waveforms corresponding to the detected patterns. In pseudocode, this may be indicated by the notation lab el_patterns (n) .

[0054] At 309, the expert system acquires and pre -preprocesses heart rate (HR) data from another sensor monitoring the subject. The HR sensor may be, for example, photoplethysmogram (PPG) sensor. The PPG shines light on the skin. An optical sensor measures the reflection of the light. The heart rate is determined by measuring how much blood passes through the illuminated spot as the heart beats. . The HR data may be acquired and streamed (or otherwise sent) to the expert system 10 at the same time the accelerometer data is streamed (or sent) to the expert system. An example of the HR data which has been processed (e.g., smoothed) into a continuous waveform is indicated by 280 in FIG. 3. In pseudocode, the HR data may be indicated by the notation HR(n).

[0055] At 311, the processor calculates the expected heart rate of the subject based on the HR data received from the PPG sensor and the accelerometer data acquired from (and/or corresponding walking patterns ol) the subject previously generated and stored in the knowledge base 40, which is labeled as a pattern database in FIG. 3. The processor may compute the expected heart rate of the subject based on a function that includes walking pattern and accelerometer data as inputs. For example, the prediction may be generated by the following equation:

HR_expected(n) = f(walking_^state(n-l), walking_^state(n))

[0056] This equation takes changes between different walking states into account. Examples of changes in walking characteristics include but are not limited to:

• from brisk walking with moderate intensity towards slow walking with low intensity

• from walking to walking downwards/ stairs climbing

• from walking to standing still

• from walking to sit [0057] These and/ or other walking characteristics are key to model, since a correlation may be drawn between one or more motion patterns and intermittent claudication. In many cases, a physiological decline in a vital sign (e.g., heart rate or so-called heart rate recovery after exercise) may be expected. This recovery may be modelled, for example, with a mono -exponentially decay function such as:

HR_decay(t) = HR_0 + HR_A *exp(-t/HRR_tau), where

HR_0 is the asymptotic value of HR as t- > ¥, HR_A is the difference between HR_peak and HR_0, and HRR_tau is a time-constant. The parameters of the decay function may be learned, for example, from historical data of the individual. Also, in one example, HR_0 may be set to the rest HR measured in times without activity or in the morning just before wake-up. In one embodiment, population-based standard values may be used.

[0058] From physiology, different relations between HR and physical activity are known. Heart rate recovery is one example. A rise of HR and the establishment of steady-state as response to aerobic exercise may be modelled analogously. In one embodiment, the modeling may be based on information indicative of physiological hear rate regulation, such as but not limited to observed heart rate response to aerobic training. The article at https:/ /www.ncbi.nlm.nih.gov/pmc/articles/PMC5447093/pdf/ fphys-08-00301.pdf is illustrative. This article is incorporated herein by reference.

[0059] Each time an expected HR value is generated, the prediction and/or its associated accelerometer data may be used to produce additional pattern recognitions for storage in the knowledge base 40. In pseudocode, the HR data may be indicated by the HR_expected(n). One technique that may be used to make HR predictions involves implementing a method that is based on a machine-learning or neural network model. Such a method may predict heart rate from previous HR and accelerometer data. In accordance with one or more embodiments, an expected HR curve may be computed.

[0060] Another technique may be performed based on statistical methods. Such a method may involve comparing the characteristics of the currently considered walk (including variability during the event) with previous walks stored in the knowledge base 40, identifying a group of walks in the knowledge base which are similar to the current walk (e.g., see later similarity criterion), and combining the HR (and/ or any other PPG features) measured from the group of identified walks from knowledge base using one or more statistical indicators, e.g., mean, median, standard deviation, etc. In one embodiment, one or more combined estimates may be used as prediction (e.g. median) or prediction band (e.g. mean +/2).

[0061] At 313, the processor determines the similarity between the expected HR and real data, which, for example, may correspond to or be derived from the smoothed HR data actually acquired from the PPG sensor in operation 309. The similarity (denoted Similarity(n)) may be determined, for example, by calculating a correlation measure (e.g. Pearson correlation coefficient) between the expected and the measured HR time series, which function may be based on the following equation:

HR * HR_expected

where HR(t_k) is a particular value of the HR time series at time t_k, and HR is the average value of HR. Prior to the comparison, a transformation may be applied to one or both of the time series, including time shift (e.g., resulting correlation measure would be the cross-correlation between the two series for one or multiple time shift) . [0062] At 315, the processor compares the similarity Similarity(n)to a predetermined threshold value. This threshold value may be computed or selected to indicate a match or mismatch between expected and measured HR time series. In one embodiment, the similarity between the PPG predictions for walks with similar walk patterns are calculated. The similarity data is gathered in a histogram. Well- known methods of histogram thresholding are used to select a similarity threshold that discriminates between similar walking patterns and other, dislike walking patterns. Histogram thresholding can make use of potentially bimodal characteristics in the similarity distribution or can use preset acceptance intervals.

[0063] In one case, the threshold value may be a default value. For example, the normalized cross correlation coefficient may range between -1 and 1, as previously indicated. Values below 0.5 may be considered indicative of low correlation. A negative correlation between expectation and observation might be a good indicator for an incident.

[0064] In one embodiment, the threshold value may be determined by accessing historical walk patterns in the knowledge base 40. Then, for each of the possible values of the threshold, and for each walk indicated in knowledge base 40, the number of the walks indicated the knowledge base 40 may be calculated, which are similar for the currently considered value of the threshold. The average number of similar walks may then be stored for a given threshold. A threshold value may then be selected which results in a preset desired fraction of walks resulting as similar in the historical knowledge base 40 (e.g. 1%).

[0065] At 316, if the computed similarity is equal to or lower than the predetermined threshold value, then the processor marks the episode corresponding to the acquired actual HR data as an intermittent claudication (IC) incident candidate. [0066] At 317, if the computed similarity is greater than the predetermined threshold value, then the processor marks the episode corresponding to the acquired actual HR data as being a no IC incident candidate, e.g., not a candidate for intermittent claudication (IC).

[0067] At 318, irrespective of the comparison performed in operation 315, the processor may filter the incident annotation. The filter may be, for example, a low-pass filter applied for the purpose of eliminating spurious signals. The spurious signals may include, for example, single incident annotations most likely caused by signal noise or artifacts. Multiple/ grouped labels may not be filtered out.

[0068] This operation may also include reducing fluctuations in the IC incident score in the individual HR data. A specific HR data might be corrupted by noise, or specific operational conditions. For example, missing contact between user and PPG could cause HR to be incorrectly classified as 0, and the HR data to be very different from the predicted HR data, resulting in large IC score. This mechanism may be used to reduce or control the number of false alarms, at the individual user level or in a group of users. An example of a filter that may be used to perform the operations of 318 may be one that computes a sliding mean to the incident annotation, which, for example, may be represented as annotated_HR(n).

[0069] At 320, the processor relates incident IC candidates (determined in operation 316) time -wise to one or more walking patterns detected in operation 307. The incident IC candidates may be related to the one or more detected walking patterns using, for example, a pattern matching technique. When an incident IC candidate is determined to be related to one or more of the detected walking patterns, an IC candidate may be turned into an IC event. The related incident IC candidates may be marked with the notation label_IC_event(n). The pattern matching technique may be performed by a production rule based expert system. An example of the pattern matching technique is disclosed in https://en. Wikipedia. org/wiki/Production_system_(computer_science), the contents of which are incorporated herein by reference. In another embodiment, a statistical method based on Pearson correlation coefficients or one that measure vector distances may be used.

[0070] At 322, the processor computes a confidence score for each of the intermittent claudication events that were found to relate to one or more of the walking patterns in operation 320. The higher the confidence score, the higher is the likelihood that a real IC has been detected. Conversely, the lower the confidence score, the lower is the likelihood for an IC event. In one embodiment, the confidence score may be computed as a weighted mean of the cross-correlation between expected and measured HR and the matching quality of the accelerometer signal towards the event pattern. The confidence score may be indicated with the notation probability_IC_event(n).

[0071] At 324, the processor aggregates the number and probability of incident IC claudication events over a predetermined time period in order to generate an IC risk score. The predetermined time period may be, for example, one week before the latest considered event or another time period. The IC risk score may be calculated, for example, based on the ratio between number of detected IC events (e.g., in one week) and number of steady-state walking events (e.g., in one week).

[0072] At 326, the processor compares the risk score to another predetermined threshold. This threshold may be calculated or selected to a predetermined level of confidence, e.g., 99%. In one embodiment, population level may be set over historical data (e.g. % level) which gives a certain number of generated notifications for the entire population in a predetermined period of time (e.g., 10 notifications/day).

[0073] At 327, when the risk score is greater than the predetermined threshold, the processor may generate a notification indicating that the subject is likely suffering from intermittent claudication symptoms. The notification may be in the form of an alarm and/ or one or more of a variety of other types of notifications. The notification may be sent to the monitor or other device of the subject himself and/or to a guardian, medical service, or other responsible party. In one embodiment, information indicative of the IC event may be recorded in a database for viewing by a doctor or other medical personnel, who, for example, may access the database to determine patterns and/ or make other diagnoses or treatment decisions based on this information.

[0074] At 328, when the risk score is less than or equal to the predetermined threshold, the alarm may not be generated. In this case, the IC risk score may be reported by the processor to a database or service responsible for storing electronic medical records and/ or providing care for the subject. [0075] FIG. 5 illustrates an embodiment of a method for detecting activity and/or behavioral (e.g., various types of walking) patterns for a particular subject for storage in the knowledge base 40, which may also be referred to as a pattern database. The patterns may be determined from historical physical activity (PA) and heart rate/photoplethysmogram (PPG) data stored in the knowledge base 40 of a subject being monitored. Determination of the patterns may be performed, for example, as part of the pre-processing operations performed by the processor 14 of the expert system. Once stored in the knowledge base, the patterns may be accessed by processor 14 for purposes of generating subject heart rate predictions in operation 311.

[0076] Referring to FIG. 5, the method includes, at 410, acquiring and pre-processing accelerometer data from the physical activity sensor ((acc_x,y,z(n)). This operation may correspond to operation 301 of FIG. 3. In one embodiment, operation 410 may be repeated for different signal windows such as different days of a program introduction/ system training week. In another embodiment, a different method may be used to divide the accelerometer data from the physical activity sensor into subgroups. [0077] At 420, physical activity and steady state walking episodes are detected, as indicated by the notation label_PA_episodes(n). This operation may correspond to operation 303 in FIG. 3.

[0078] At 430, changes in the state and/or features of the subject (e.g., walking speed, intensity, regularity, symmetry of gait etc.) are detected within each of the walking episodes. This may be indicated by the notation label_walks(n)and may correspond to operation 305 of FIG. 3. [0079] At 440, patterns in sequences of the labelled walking episodes may be detected, as indicated by the notation label_patterns(n)and may correspond to operation 307 of FIG. 3.

[0080] At 450, one or more of the patterns in the sequences of labelled walking episodes may be selected and aggregated. This operation may involve, for example, selecting and aggregating only those patterns that reoccur a predetermined number of times, and thus may be considered to be typical patterns. This operation may be indicated by the notation select_patterns(n). According to one option, operation 450 may include selecting and aggregating one or more pre-defmed patterns 445 in place of or in addition to the patterns detected in operation 440 that are selected. The underlying assumption is that during this training phase no IC typical events systematically occur.

[0081] At 460, heart rate data is acquired and pre-processed from one or more sensors on the subject. The one or more sensors may be PPG sensors and the data may be indicated by the notation HR(n) . This operation may correspond to operation 309 in FIG. 3.

[0082] At 470, one or more heart rate sequences may be selected based on the heart rate data acquired and pre-processed in operation 460 for each of the walking patterns generated in operation 450. Once the heart rate sequences are selected, outliers and/ or other spurious or extraneous signals may be removed, for example, using statistical methods in order to generate a final set of heart rate sequences for the selected walking patterns.

[0083] At 480, the individual walking patterns are stored in association with respective ones of the heart rate sequences (or patterns) in the knowledge base 40, for future use in detecting (e.g., in real time) intermittent claudication and/or other conditions of the subject being monitored. For example, the walking patterns and associated heart rate patterns may be used to generate the prediction(s) in operation 311 of FIG. 3.

[0084] FIG. 6 illustrates a method for detecting intermittent claudication (IC) events based on a different combination of sensor data, namely physical activity sensor data and location data. The location, or localization, data may be GPS data and/ or another type of location data. The method may be implemented, for example, by the system of FIG. 1.

[0085] Referring to FIG. 6, at 501, the processor 14 acquires and pre-processes location data, which in this example is geolocation (e.g., GPS) data expressed by the notation geolocation(n).The location data may be received, for example, by the activity or tracking monitor carried by the subject or a smartphone carried by the subject. If the expert system is located on a device carried by the subject being monitored, then the location data may not be transmitted or, alternatively, may be transmitted to a remote server or other device maintained by a caregiver, doctor, or other interested party. When the expert system is located on a remote server, workstation, or other device, the location data may be transmitted in the aforementioned manner.

[0086] At 503, the processor maps actual geolocation of the subject to recorded common locations, for example, which the subject has been tracked and recorded to occupy or visit in the past. The common recorded locations may be determined, for example, with reference to pattern and other forms of data stored in the knowledge base 40 which indicate habitual patterns where the subject has followed in the past regarding location. The pseudocode notation for operation 503 may include collect_habit(n), where n indicates the time or sample number of the time series data and h indicates geo-location data (e.g., Tatitude, Tongitude, Altitude).

[0087] At 505, the actual motion profile of the subject is mapped to stored walking profiles or patterns. The actual motion profile may be based on sensor data, including but not limited to one or more of the accelerometers as previously discussed. The walking profiles may be determined in accordance with operations discussed in other embodiments described herein, in association with or based on pattern data stored in the knowledge base 40. Thus, operations 525 to 528 may be performed, for example, in the same manner as operations 410 to 440, as explained with reference to FIG. 4. [0088] The mapping performed in operation 505 may involve, for example, comparing the activity/motion data derived from the accelerometer(s) to known habitual patterns detected and stored in the knowledge base. In one embodiment, the activity/motion data may be processed to determine a specific pattern or pattern format before the comparison or mapping is performed. The pseudocode for operation 505 may include collect_walk_from_habit(n).

[0089] At 507, an indication of the similarity is computed between at least one of the walking profiles/patterns/behaviors corresponding to the habit information stored in the knowledge base and the actual behavior or activity of the subject, as indicated by the received sensor data (either in raw or processed form). The similarity (similarity(n)) may be computed as a value, for example, based on the following equation:

S — wl* similarity_walk patterns + w2* similarity location, where wl and w2 are relative weights that may correspond to the similarity measures which, for example, may be of different orders of magnitude. Similarity of walk patterns may be determined as previously discussed. Similarity location may be given, for instance, by distance in kilometers between two locations or by a binary indication (equal/ different) for the place category, which may be derived by querying a reverse geocoding database on two locations. Furthermore, location similarity can be derived from location series data/ location tracks. In one embodiment, the similarities may be based on cross-correlation coefficients.

[0090] At 509, the processor 14 of the system compares the similarity value to a predetermined threshold. The predetermined threshold is computed or selected to indicate match or mismatch. [0091] At 510, when the similarity value is less than or equal to the predetermined threshold, the processor of the system marks the actual behavior or activity of the subject as corresponding to an incident intermittent claudication episode. [0092] At 511, when the similarity value is greater than the predetermined threshold, the processor of the system marks the actual behavior or activity of the subject as not corresponding to an incident intermittent claudication episode and thus, more similar to a normal walking behavior [0093] At 515, candidates for a group of possibly valid ID events are identified. In one embodiment, the identification may be performed using thresholds and heuristics such as at least 2 IC candidates in a predetermined time window, e.g., a window of two minutes. The notation for this operation may be given as label_IC_event(n).

[0094] At 517, confidence scores are computed for the intermittent claudication events identified in operation 515. In one embodiment, the scores may indicate a probability that the corresponding event is a valid IC event. For example, the confidence score(s) may be computed as the weighted mean of the correlation between geolocation and habit data and the matching quality of the accelerometer data with the assigned walking behavior patterns. The notation for this operation may be given as probability_IC_event(n) .

[0095] At 519, the number and probability of IC events are aggregated over a predetermined period of time to yield an IC risk score. The number and probability of IC events may be aggregated to yield an IC risk score, for example, in the aforementioned manner discussed relative to HR-based risk computation.

[0096] At 521, the IC risk score is compared to a predetermined threshold. This comparison may be performed, for example, in a manner analogous to the HR-based risk assessment and thresholding previously described. The predetermined threshold is selected to filter out false positive alarms. The threshold may be set to control notifications rate on population level.

[0097] At 522, when the risk score is greater than the predetermined threshold, the processor generates and/or transmits an alarm or other notification indicating that the subject is likely to be experiencing an intermitent claudication event based on the analysis of the current sensor data in view of the stored information in the knowledge base.

[0098] At 523, when the risk score is less than or equal to the predetermined threshold, the processor may report the IC risk score, for example, in a record of the knowledge base and/ or an electronic medical record of the subject for later training and comparison purposes of a model used to implement IC detection for the subject.

[0099] FIG. 7 illustrates an embodiment of a method for collecting patterns from historical data indicating physical activity and location data of the subject being monitored. The operations of this method may be performed, for example, as part of a pre-processing operation before actual activity data begins to be received for the subject. The collected pattern data may be stored, for example, in the knowledge base 40 of the system for performing individualized IC risk assessment in accordance with the embodiments for detecting intermittent claudication, as described herein.

[00100] Referring to FIG. 7, at 610, the method includes acquiring and pre-processing accelerometer data from the physical activity sensor ((acc_x,y,z(n))for different time windows/ several days of a training week. This operation may correspond to operation 301 of FIG. 3.

[00101] At 620, the processor 14 acquires and pre-processes location data, which in this example is geolocation (e.g., GPS) data expressed by the notation geolocation(n). This operation may be similar to operation 501 in FIG. 5. For example, the location data may be received by the activity or tracking monitor carried by the subject or a smartphone carried by the subject. If the expert system is located on a device carried by the subject being monitored, then the location data may not be transmitted or, alternatively, may be transmitted to a remote server or other device maintained by a caregiver, doctor, or other interested party. When the expert system is located on a remote server, workstation, or other device, the location data may be transmitted in the aforementioned manner. [00102] At 630, once the geolocation data has been obtained, the processor may extract from the geolocation data locations the subject has visited or occupied on a reoccurring basis. This operation may be performed, for example, relative to a predetermined threshold. For example, only locations which the subject has visited for more than a certain number of times may be extracted. Also, patterns of location (e.g., relative to certain days of the week, certain hours of the day, etc.) may also be extracted for purposes of generating pattern-based location information.

[00103] At 640, the processor may also extract reoccurring trips from the geolocation data. This may be accomplished, for example, by not only determining the location of the subject but also tracking the movement of the subject. Such movement may be considered to be included in the types of trips extracted in operation 640.

[00104] At 650, the processor clusters the reoccurring locations and trips of the subject and selects a subset of location data from this clustered information. Such information may be considered to correspond to certain habitual behavior of the subject, which may be used to form a baseline for performing intermittent claudication detection when compared with sensor data. The clustering and selecting operations may be associated with the notation select_habits(n).

[00105] At 660, the processor assigns sensor (e.g., accelerometer) data to the habits selected in operation 650. This may be accomplished, for example, by storing the location information with the sensor data of the subject acquired at that time for each of the selected ones of the habitual locations and/or trips.

[00106] At 670, the processor clusters and aggregates the sensor (e.g., accelerometer) data using the cross-correlation function to pair-wise assess similarity of accelerometer signals and to exclude outliers and artifacts that do not correlate well. This may be accomplished in a variety of ways. In one embodiment, walk pattern characteristics may be clustered based on location data, and statistical indicators may be applied to the characteristics of the walks belonging to the same cluster, e.g., mean duration of walks in certain location or variability in walk regularity over walks associated with certain outdoors location categories, e.g., a park.

[00107] At 680, walking episodes and characteristics of the subject are detected based on the clustered and aggregated sensor data.

[00108] At 690, outliers and other spurious or extraneous data points are removed using a filter to generate a final set of patterns generated based on the location data and walking profiles of the subject. [00109] At 699, the pattern information is then stored in the knowledge base 40 for use in performing intermittent claudication detection in accordance with the embodiments described herein.

[00110] The methods of the aforementioned embodiments may be combined into a single method for detecting intermittent claudication based on three data streams - heart rate (PPG) sensor data, location (GPS) data, and activity/ mobility (accelerometer) data - using an analogous and integrated application of operations performed in the aforementioned embodiments.

[00111] In accordance with one or more of the aforementioned method embodiments, various types of walking or activity profiles of a subject may be monitored. In performing this monitoring operation, the subject may have a wearable device equipped with one or more sensors as previously described. [00112] In a particular embodiment, the wearable device is equipped with a suite of sensors for collecting activity (mobility), location, and vital signs data of the subject. The activity (mobility) sensor may include a three-axis accelerometer. The location sensor may include a barometric air pressure sensor and/or GPS sensor. The vital signs sensor may include an optical PPG sensor for taking heart- rate measurements. In one case, the subject may wear a plurality of devices that are equipped with one or more of the aforementioned sensors, and thus which are collectively used by the expert system to detect a health episode such as intermittent claudication. The wearable device(s) may be configured to output the sensor data as (calibrated) raw data and/ or pre-processed data. All sensor data streams received by the expert system may be synchronized and aligned (in various pre-processing operations) with respect to predetermined sampling rate(s) of the sensor data, spurious data segments, and/or other parameters or indicia.

[00113] The wearable device may also include a communication circuit to wirelessly transmit information (including the sensor data) to a remotely located expert system and to wireless receive information from the expert system and or other entities. In one example, the expert system may be located at a medical facility, processing center of a monitoring services, and/ or other locations, which are equipped with personnel and/ or processing resources to autonomously detect and notify the subject of a health condition (e.g., intermittent claudication) or to perform such detection at least partially with the help of skilled technicians and/ or medical personnel. In one embodiment, the expert system may perform data analysis at a remote location, such as, for example, within a cloud-based system or network.

[00114] The processing resources may include one or more data processors that manage and synchronize data streams received from the wearable device. The data streams may originate from respective ones of the sensors for detecting the activity (mobility), location, and vital signs data of the subject. The wearable device may also include one or more processors (e.g., controlled by device firmware) that are able to perform pre-processing operations, for example, as previously described. In one embodiment, the expert system and its processor(s) may only be located in the wearable device.

Activity (Mobility) Sensor Data

[00115] The expert system may detect one or a variety of types of activity or motion of the subject. For example, the processor(s) of the expert system may detect various types (or patterns) of activity or movement (e.g., walking-standing, still -walking, etc.) from other activities of daily life (ADTs) based on different, temporally aligned data streams from the sensors on the wearable device. Once detected, episodes of activity may be further analyzed to detect walks that fulfill the requirement of steady state walking for apre-defmed duration. Therefore, dedicated thresholds and heuristically defined rules may be used by the expert system. A set of quantitative and qualitative mobility features (e.g., start time, duration, regularity, intensity, speed, symmetry of gait etc.) may be extracted from the walk data. These mobility features may then be used to further select walks of a certain intensity and length and/ or activity or movement patterns (e.g., walking-standing, still -walking, etc.).

[00116] In one embodiment, the expert system may receive barometric pressure data from a sensor, which, for example, is in or coupled to one or more devices worn or carried by the subject. With the help of the barometric pressure data transitions of body position, activities such as sit-to-stand, walking-and-standing still, and certain types of walking (e.g., walking uphill or downhill) may be detected. Therefore, dedicated pattern matching techniques may be used that match expected sensor behavior to measured sensor data. It is noted, however, that the use of barometric pressure data is not needed in order to determine these or other types of activity. Rather, these activities may be determined by processing the sensor (e.g., accelerometer) signals with or without the barometric pressure data, location data, and HR data.

[00117] Based on these operations, the expert system may generate a continuous data stream of labelled mobility data segments that hold meta-information of the activity. Therefore, a set of predefined labels or attributes that account for the type of activity, characteristic of activity, and trends may be used. Examples labels for the activity may include, but are not limited to, the following:

Example Types of Mobility (Activity) Patterns

Walking with high intensity Walking with low intensity Walking with decreasing intensity Walking with decreasing regularity Walking-standing still transition Standing still Unknown activity [00118] In detecting and classifying the mobility (activity) patterns of the subject, the expert system may, in one embodiment, use statistical methods to assign to a mobility episode (characterized by a set of features) a likelihood that each of the patterns belong to a pre-defmed class/ label. For filtering and regularization purposes, similar, subsequent episodes may be fused or clustered to one episode, with the same label using, for example, a sliding mean.

[00119] Once the patterns are detected and compared to the information in the knowledge base, additional sensor data (e.g., location data, HR data, etc.) may be analyzed by the expert system, as described herein, to detect intermittent claudication and/or another type of health-related condition. Once detected, the expert system may send notifications through the output device when, for example, intermittent claudication is detected for the subject. The output device may include, for example, one or more of a smartphone, tablet, computer, etc., that is able to give indications of a PAD/ IC risk or event score, e.g., as part of a displayed health and mobility dashboard.

Example Algorithms

[00120] The expert system may detect episodes of activity and episodes of non-activity based on the sensor (accelerometer) data. Either all three x,y,z channels of the raw accelerometer or the vector norm be used. In one embodiment, this may be accomplished by implementing a peak detection algorithm based on signal waveforms generated by the accelerometer data. The accelerometer sensors may be located in the wearable device and/ or located in a pendant or other sensor attachment on the body. As the accelerometer signals are received, the processor of the expert system generates a waveform or spectrum over time. Any peaks (e.g., local maxima) in the waveform that exceed one or more predetermined threshold values may be considered as corresponding to movement (e.g., walking, running, or other activity). The amplitude levels of the predetermined threshold values may be selected to correspond to different types of activity. Similar thresholds can be used in the frequency domain. [00121] In another embodiment, clusters of samples in the acceleration signals are identified and one or more maximum value(s) (or other values in one or more other predetermined ranges) in each cluster are identified as a step boundary. In applying this algorithm, the signal may be analyzed to identify regions of samples over which at least some of the samples in the acceleration signal are above or below a threshold.

[00122] For example, a cluster may correspond to a set of samples that exceed a threshold value (e.g., 2 ms— 2 above gravity, i.e. ^~12 ms— 2). Additionally, or alternatively, a cluster may correspond to a set of samples that exceed one or more threshold values, with a small gap of samples that do not exceed the threshold(s) being permitted, provided the size of the gap is below a predetermined limit, thresholdgap. The predetermined limit may be less than a typical time between steps, e.g., 0.5 seconds. [00123] In one embodiment, instead of using a single threshold to identify samples that form part of a cluster, two thresholds may be used. This introduces a hysteresis — a sample exceeding the first threshold indicates the start of a cluster and the first sample falling below the second threshold indicates the end of the cluster. A minimum time duration (possibly related to the step time — the time between two successive steps) between the start and end of the cluster can be applied if required. The use of two thresholds is particularly useful when the jerk of the acceleration signal is analyzed to identify the step boundaries.

[00124] Instead of steps, the algorithm may be designed in terms of strides. This may be advantageous in case of asymmetric walking patterns (so-called period-2 signals). For example, a step may be indicated by the displacement of one foot, and a stride may be indicated by the displacement of both feet (one step with the left foot, one step with the right foot). When the accelerometer is located at the foot or ankle, strong peaks may be evident in the accelerometer measurements when the corresponding foot is moving, and a little ‘cross talk’ when the other foot is moving. When the accelerometer is located at the torso, movements of both feet may produce similar accelerometer signals. An accelerometer located on the side of the body may show some asymmetry in the accelerometer measurements.

[00125] In some cases, the acceleration signals may contain multiple peaks per step (e.g., per heel strike), or random peaks that occur between steps. If it is assumed that the subject has a steady cadence (e.g., a steady rate of steps), then these multiple peaks and random peaks can be filtered out from the resulting step series, for example, by removing steps that severely influence the overall step time variability. Once step boundaries have been identified in the accelerometer signal, a ‘step time’ can be derived, which is the time between two consecutive identified steps when the accelerometer measurements are collected on the torso of the subject or two consecutive identified strides when the accelerometer measurements are collected. For clarity, the moment of a single step may be referred to as a ‘step peak’ or ‘heel strike’, so the step time in some cases is the time between two step peaks. [00126] The found step times may be tested for regularity. For example, each step time may be analyzed to determine if the gap between the steps or strides (represented by the step time) is too long and should in fact be considered as a gap between different walking parts. In one particular implementation, if a central stepping time (e.g. an average (mean or median) of the step times) is Tstep seconds, then, if there is no step detected in a time (for example given by Thresholdstep*Tstep) seconds, the end boundary of a walking part may be found at that last step. Use of this time threshold (Thresholdstep*Tstep) may be advantageous as it adapts the boundary to walking speed, although a fixed, user-independent, time threshold may be used if desired. A similar method may be used to delineate the start of a walking part. An example value for Thresholdstep is 4.

[00127] In one embodiment, the threshold (or multiple thresholds for a cluster) used to identify extrema (e.g., maxima or minima) corresponding to step boundaries (e.g., depending on the position of the accelerometer on the body) is adapted according to characteristics of the acceleration (or jerk) signal. Adapting the threshold(s) in this way to the signal allows step boundaries to be detected more reliably. This is because, for example, the maximum or minimum acceleration experienced by the accelerometer during a heel strike may change over time due to changes in the positioning of the wearable device relative to the body. Adapting the threshold also allows the algorithm to better adapt to characteristics of a particular subject or the surface that is walked on. While these are discussed here as examples, it is appreciated that different algorithms for processing the sensor (e.g., accelerometer) data to detect episodes and types of activity and motion of the subject may be used in other embodiments. One example is disclosed in US Patent Publication No. 2017/0000384, the contents of which are incorporated herein for all purposes by reference. Examples of other algorithms that may be used are ones based on the processing of spectral features or ones performing template matching.

Heart Rate Sensor Data

[00128] The heart rate data from the HR sensors may be processed by the expert system to generate a corresponding waveform, which alone or in combination, with the other data streams (e.g., activity/mobility, location, etc.) may be used as a basis for detecting patterns and intermittent claudication episodes. In one embodiment, the heart rate (HR) behavior of the subject may be predicted based on the continuous stream of labelled mobility data. This may be accomplished, for example, by applying control loop rules of physiology. In order to increase accuracy, further knowledge about the subject being monitored (e.g., age, weight, height, maximum heart rate, heart rate at rest, etc.) may also be taken into consideration.

[00129] The prediction of heart rate behavior may be related to or confirmed by the heart rate sensor data. Moments of big discrepancies may be detected using subtraction and thresholding techniques. Discrepancies of a certain length and height are labelled to be “potential pain incident” moments indicative for an IC event. [00130] Additionally, as dedicated features in an app or any other, related sensor management system, the subject may maintain a pain diary. Based on information in the pain diary, a subject may provide input on pain incidents and may, for instance, report time, context, and level of experienced pain. This data may be used as a further input data stream to relate subject-reported pain incidences to the HR data and the mobility data.

Location (Localization) Data

[00131] In one embodiment, the expert system may detect intermittent claudication or another health- related condition based on location (or localization) data, taken in combination with one or both of the activity sensor data and the heart rate data. Location data adds contextual information to the activity data and, for example, may help to detect activities of daily life (ADLs). The location data may be, for example, derived from a GPS sensor carried or worn by the subject to be monitored. The location data from such a sensor may be used by the expert system to tag the data-stream with location- specific labels. Examples of these labels include indoor, at home, outdoor, garden, shopping mall, in bus, etc.

[00132] While GPS sensors may have proficient accuracy for outdoor locations, in some cases GPS sensors may lack some accuracy for indoor localization. Thus, in one embodiment, one or more other types of real-time localization sensors or systems may be used, alone or in tandem, with the GPS sensor data. One example of another type of real-time localization sensors include Bluetooth Low Energy (BLE) beacons and locators. The expert system may use data from these sensors, for example, when the subject to be monitored is in an indoor location. When compared to patterns and other predetermined reference data stored in the knowledge base, the expert system may calculate a likelihood that the subject is experiencing intermittent claudication based on a certain detected activity/health pattern depending on the locality of the subject. Also, natural causes of certain behavior may be discriminated from pathological causes more accurately.

[00133] Instead of pattern matching and the types of classification algorithms previously described, one embodiment of the expert system may use end-to-end signal classification based on deep-learning methods for performing pattern identification. In such alternative embodiment, a relatively large amount of data may have to be labelled beforehand, in order to train a machine-learning classifier/neural network model.

[00134] In one embodiment, the processor 14 of the expert system may be a configurable data processor that captures or detects different health/behavior patterns. To be extendable to patterns other than PAD/ IC patterns such as intermittent gait abnormalities due to neurological or orthopedic issues, the firmware or other instructions executed by the processor may incorporate rules that define how detection of these other types of patterns may be performed. For example, a care -giver of the subject being monitored or a system administrator may use a dedicated user interface may assign behavior descriptors to detectable and characteristic patterns in an easy (e.g., visual) way. In a different context, tools for visual programming and structured definition of patterns may be implemented. [00135] In one embodiment, actual behavior patterns may be related to or determined by behavior history data. Such data may be used by the expert system as a basis for detecting changes of the subject being monitored from what is considered to be normal behavior. What is considered “normal” may change from person to person, as determined based on predetermined activity or behavior patterns stored in the knowledge base. For example, running may be considered normal for a man under age 40 but may be considered unusual for a man over the age of 70. This data may be used by the expert system as an additional basis for detecting PAD/ IC incidents and/ or may be used to detect other (e.g., even more subtle) health incidents. Examples include the onset of edema (e.g. due to heart failure) or change in pain/ flexibility due to orthopedic exacerbations. [00136] The methods, processes, and/or operations described herein may be performed by code or instructions to be executed by a computer, processor, controller, or other signal processing device. The code or instructions may be stored in the non-transitory computer-readable medium as previously described in accordance with one or more embodiments. Because the algorithms that form the basis of the methods (or operations of the computer, processor, controller, or other signal processing device) are described in detail, the code or instructions for implementing the operations of the method embodiments may transform the computer, processor, controller, or other signal processing device into a special-purpose processor for performing the methods herein.

[00137] The expert systems, processors, detectors, models, or other signal, pattern, or data detection, signal generating, or signal processing features of the embodiments disclosed herein may be implemented in logic which, for example, may include hardware, software, or both. When implemented at least partially in hardware, expert systems, processors, detectors, models, or other signal, pattern, or data detection, signal generating, or signal processing features may be, for example, any one of a variety of integrated circuits including but not limited to an application-specific integrated circuit, a field-programmable gate array, a combination of logic gates, a system-on-chip, a microprocessor, or another type of processing or control circuit.

[00138] When implemented in at least partially in software, the expert systems, processors, detectors, models, or other signal, pattern, or data detection, signal generating, or signal processing features may include, for example, a memory or other storage device for storing code or instructions to be executed, for example, by a computer, processor, microprocessor, controller, or other signal processing device. The computer, processor, microprocessor, controller, or other signal processing device may be those described herein or one in addition to the elements described herein. Because the algorithms that form the basis of the methods (or operations of the computer, processor, microprocessor, controller, or other signal processing device) are described in detail, the code or instructions for implementing the operations of the method embodiments may transform the computer, processor, controller, or other signal processing device into a special-purpose processor for performing the methods herein.

[00139] Although the various exemplary embodiments have been described in detail with particular reference to certain exemplary aspects thereof, it should be understood that the invention is capable of other example embodiments and its details are capable of modifications in various obvious respects. As is readily apparent to those skilled in the art, variations and modifications can be affected while remaining within the spirit and scope of the invention. Accordingly, the foregoing disclosure, description, and figures are for illustrative purposes only and do not in any way limit the invention, which is defined only by the claims.

Claims

WE CLAIM:

1. A method for detecting a health condition, comprising: receiving data from a first sensor of a subject; determining at least one pattern based on the first sensor data; identifying a candidate based on data from a second sensor on the subject; relating the candidate to the at least one pattern to determine occurrence of an event; generating a score for the event and comparing the score to a predefined threshold; and detecting that the subject has the health condition event based on a result of the comparison, wherein the health condition is intermittent claudication.

2. The method of claim 1, wherein the data from the first sensor is indicative of physical activity of the subject.

3. The method of claim 2, wherein the pattern includes a walking pattern of the subject.

4. The method of claim 3, wherein the second sensor is a vital sign sensor.

5. The method of claim 4, wherein: the vital sign sensor is a cardiac activity sensor; the candidate is identified based on the data from the cardiac activity sensor; and relating the candidate to the pattern includes determining a correspondence between the walking pattern and the candidate determined based on the data from the cardiac activity sensor.

6. The method of claim 4, wherein identifying the candidate includes: generating a predicted cardiac activity based on the data from the cardiac activity sensor and a predetermined physical activity pattern of the subject stored in a database; computing a similarity value between the predicted cardiac activity and the data from the cardiac activity sensor; and identifying the candidate based on the similarity value.

7. The method of claim 1, wherein the score is indicative of a probability or risk that the subject has intermittent claudication.

8. The method of claim 1, wherein at least one of the first sensor or the second sensor is on a wearable device of the subject.

9. A method for detecting a health condition, comprising: receiving data from a first sensor of a subject; determining a profile of the subject from the first sensor data; mapping the profile to a pattern derived from second sensor data; identifying a candidate for the health condition based on the mapping; generating a score for the candidate and comparing the score to a threshold; and detecting that the subject has the health condition based on a result of the comparison, wherein the health condition is intermittent claudication.

10. The method of claim 9, wherein the data from the first sensor is indicative of at least one location of the subject.

11. The method of claim 10, wherein the profile is a motion profile of the subject.

12. The method of claim 11, wherein the pattern derived from the second sensor data includes a walking pattern.

13. The method of claim 12, further comprising: determining a similarity between the profile and the pattern, wherein identifying the candidate for the health condition is performed based on the similarity.

14. The method of claim 9, wherein determining the profile includes: comparing the first sensor data to one or more pre-stored profiles of the subject; and determining the profile of the subject based on the comparison.

15. A method for detecting a health condition, comprising: receiving data derived from a plurality of sensors; determining a physical activity pattern of a subject based on the data; determining a vital sign pattern of the subject based on the data; receiving location information corresponding to the subject; and detecting that the subject has the health condition based on the physical activity pattern, the vital sign pattern, and/or the location information, wherein the plurality of sensors includes a physical activity sensor and a vital sign sensor located in at least one device worn or carried by the subject and wherein the health condition is intermittent claudication.

16. The method of claim 15, wherein determining the physical activity pattern includes: comparing the data of the physical activity sensor to predetermined physical activity data of the subject stored in a knowledge base, and determining the physical activity pattern based on results of the comparison.

17. The method of claim 16, wherein the vital sign pattern includes a cardiac activity pattern.

18. The method of claim 17, further comprising: generating a predicted cardiac activity based on the data from the physical activity sensor and predetermined cardiac activity information of the subject stored in the knowledge base or a database, wherein the cardiac activity pattern is determined based on the predicted cardiac activity and the predetermined physical activity data of the subject stored in the knowledge base.

19. The method of claim 15, further comprising transmitting a notification of the detected intermittent claudication to at least one device.

20. The method of claim 19, wherein the at least one device is a device of the subject, a device of medical personnel, or device of a guardian or other person known by the subject.