WO2016193423A1

WO2016193423A1 - A system and wearable sensing device

Info

Publication number: WO2016193423A1
Application number: PCT/EP2016/062639
Authority: WO
Inventors: Guang-Zhong Yang; Benny Lo; Ching-Mei CHEN
Original assignee: Sensixa Limited
Priority date: 2015-06-04
Filing date: 2016-06-03
Publication date: 2016-12-08
Also published as: GB201509792D0

Abstract

The disclosure comprises a wearable device for assessing blood flow in a subject, the device comprising a plurality of photoplethysmography (PPG) sensors comprising a first PPG sensor and a second PPG sensor, each sensor being arranged to carry out measurements indicative of blood flow in the subject, wherein the PPG sensors are disposed such that, when the device is worn on an upper or lower limb of the subject, the first PPG sensor is in contact with the upper or lower limb on which the device is worn, and the second PPG sensor is contactable by the other of the upper or lower limb of the subject.

Description

A SYSTEM AND WEARABLE SENSING DEVICE

The present disclosure relates to a system and wearable device for analyzing blood flow in a subject, in particular to assess diabetic peripheral neuropathy (DPN) or arterial stiffness in the subject.

Peripheral neuropathy occurs in a patient when the peripheral nerves have been ruptured. These nerve disruptions can lead to impairment of tactile perception, pain or undesired stings, peculiar changes in skin colour and blood pressure, weakness, muscle loss and pain, depending on the affected nerves. The most common form of diabetic neuropathy is the peripheral sensorimotor neuropathy which affects the feet and worsens over time due to its negative impact on both the nervous and vascular systems. Patients also suffer deterioration in both temperature and vibratory response.

The time of emergence of peripheral neuropathy varies but it has been found that type 1 diabetic patients do not develop this ailment until after a few years while its appearance in type 2 diabetic patients comes much quicker, sometimes even at the onset of diabetes.

Patients with peripheral neuropathy often are not aware of their condition, and many of the patients with peripheral neuropathy may not have any symptom. The main complication of peripheral neuropathy is the diabetic foot. Typically about a fifth of hospital visits from diabetic patients are foot related making it the leading cause of hospital visits for diabetes.

Neuropathy leads to numbness due to disrupted blood flow to the limbs which in turn causes trauma (often unnoticeably) as well as joint deformity. When such trauma is coupled with large amount of pressure over time, sepsis can occur. Sepsis mainly affects the soles of the feet, which are most vulnerable to pressure during normal daily activities such as walking, running, standing etc. Further development of the sepsis can lead to ulcer development and could lead to complete amputation of the foot; a procedure which has a high mortality rate within the year of amputation. Early detection of DPN is therefore of high importance.

Various sensors have been used for the purpose of neuropathy detection, for example, photoplethysmography (PPG) sensors can be used to estimate arterial blood pressure through calibration techniques involving Pulse Wave Velocity (PWV) (which is obtained by placing sensors at different points on the body of known distances and finding the time difference between pulses from said locations) and hydrostatic pressure variation.

Alternatively, arterial blood pressure can be used to measure the ankle brachial pressure index (ABPI) which is the ratio between the systolic blood pressure of the ankle and upper arm (brachial blood pressure). Typically the ABPI measurement is collected using ultrasound and blood pressure cuffs. ABPI has been widely used as a clinical diagnostic tool for assessing neuropathy.

However, these existing techniques typically require multiple sensors to be applied to a patient and require medical expertise to carry out the measurements necessary to assess DPN. This may mean a patient having to make regular visits to a hospital or doctors' surgery, or at least multiple appointments for a medical professional to take the measurements at the patient's home, in order for DPM monitoring and assessment to be carried out. This is often inconvenient, time consuming and costly.

In a first aspect, a wearable device for assessing blood flow in a subject is provided, the device comprising a plurality of photoplethysmography (PPG) sensors comprising a first PPG sensor and a second PPG sensor, each sensor being arranged to carry out measurements indicative of blood flow in the subject, wherein the PPG sensors are disposed such that, when the device is worn on an upper or lower limb of the subject, the first PPG sensor is in contact with the upper or lower limb on which the device is worn, and the second PPG sensor is contactable by the other of the upper or lower limb of the subject.

A PPG sensor comprises at least one light emitting device (LED) and at least one light receiving device. Light emitted from the LED illuminates the skin and is reflected back from various structures beneath the skin. The reflected light is then detected by the light receiving device. Alternatively, light from the LED may be transmitted through a portion of the subject's body, for example, through a finger, such that the transmitted light, rather than the reflected light, is detected by the photo-detector.

Some of the light emitted from the LED will be absorbed by blood flowing in the subject's (or patient's) blood vessels. The amount of absorption will vary with the volume of blood present. With each cardiac cycle, the heart pumps blood through the blood vessels of the body creating a pressure pulse which expands the blood vessels in the body. The pressure pulse causes an increase in blood volume which results in increased light absorption, hence a PPG waveform is obtained which reflects the pressure pulse in the blood vessels. Using this waveform, the time taken for the blood to flow from the heart to the location of the PPG sensor measurement location, known as the pulse arrival time (PAT), may be calculated.

The device according to the claimed subject matter is a wearable device that is capable of obtaining a PPG waveform using PPG sensors disposed to carry out measurements at locations on a subject's upper limb and lower limb. Deviations in a pulse arrival time (PAT) between the lower limb location and upper limb location can be captured by comparing the PPG waveforms obtained at the first and second PPG sensors and used as a surrogate measure for detecting and assessing DPN or arterial stiffness. Alternatively, a time difference between the arrival of corresponding pulses captured by the respective PPG sensors at the upper and lower limbs may be determined. There is no requirement for the distance between the PPG sensors measurement location to be known since only the pulse arrival time is required to detect and assess DPN. Arterial stiffness may also be determined which can occur as a result of biological aging and arteriosclerosis and which is associated with an increased risk of heart attack and stroke.

The claimed device may be worn by a subject, for example, in their home, and data from the wearable device may then be uploaded subsequent analysis and diagnosis, for example, by a medical professional at a later time. Obtaining these PPG measurements does not require the subject to visit a medical practitioner and so is a more convenient way of obtaining suitable measurements which may then be used to assess DPN or arterial stiffness.

Any suitable PPG sensor may be used.

The measurements indicative of blood flow may be a measure of light absorption. From the measurements indicative of blood flow a PPG waveform may be produced.

In some embodiments, two PPG sensors are be used.

In some embodiments, the first and second PPG sensors are disposed on opposite sides of the device such that, when worn, the first PPG sensor is directed towards the upper or lower limb on which the device is worn and the second PPG sensor is directed away from the upper or lower limb on which the device is worn. In this arrangement, a user may conveniently place the other of the upper or lower limb (i.e. the limb on which the device is not worn), for example, a finger, against the second PPG sensor to enable PPG measurements of the other of the upper or lower limb to be made.

The wearable device may comprise a band. The PPG sensors may be disposed on opposite sides of the band.

The wearable device may comprise an item of clothing, for example, a glove, a sock or an insole. The PPG sensors may be disposed on opposite sides of the item of clothing e.g.in the case of a glove or sock, for example, one PPG sensor may be disposed on the inside and one on the outside.

The PPG sensor disposed to carry out measurements indicative of blood flow in a subject's lower limb may be disposed to carry out measurements of the subject's lower leg, ankle, knee or foot, for example. The PPG sensor disposed to carry out measurements indicative of blood flow in a subject's upper limb may be disposed to carry out measurements of the subject's arm, hand, finger or thumb, for example.

The device may further comprise an electrocardiogram (ECG) monitor for obtaining an electrocardiogram corresponding to the rhythm of the subject's heart. The ECG monitor and PPG sensors may be configured to carry out measurements simultaneously.

The ECG monitor may comprise a 1-lead ECG monitor. In some embodiments the ECG monitor comprises a plurality of electrodes, for example, 2-10 electrodes, for example 2 or 3. In some embodiments, the ECG monitor comprises a first and second electrode wherein the first and second electrodes are disposed on opposite sides of the device such that, when worn, the first electrode is directed towards the upper or lower limb on which the device is worn and the second electrode is directed away from the upper or lower limb on which the device is worn.

Any suitable ECG monitor may be used. In some embodiments, the second electrode (which is directed away from the limb on which the device is worn), is provided sufficiently close to the second PPG sensor (which is also directed away from the limb on which the device is worn) such that the user may

conveniently place the other of the upper or lower limb (i.e. the limb on which the device is not worn), for example, a finger, against both the second electrode and the second PPG sensor at the same time.

The first electrode may be disposed such that, when the device is worn, the first electrode is in contact with the upper or lower limb on which the device is worn. For example, when the device is worn on the lower limb, the first electrode is disposed such that it is in contact with the lower limb. In this way, when a user places the other of the upper or lower limb (i.e. the limb on which the device is not worn) against the second electrode, the electrical circuit between the two electrodes is complete and an ECG measurement is captured.

When worn, the first PPG sensor and the first electrode may be in contact with the limb on which the device is worn, for example, by virtue of user manipulation of the device to place the first PPG sensor and electrode against the limb, or, for example, by the device being worn sufficiently close to the limb.

The wearable device may comprise a memory component for storing measurements made by the first and second PPG sensors and/or the or an ECG monitor.

The wearable device may comprise a processor for computing a PAT at the lower limb and a PAT at the upper limb based on the measurements made by the first and second PPG sensors. In some embodiments, the processor also uses measurement made by an ECG monitor to compute the pulse arrival time at the respective locations.

The PAT may be obtained by measuring the time delay from an ECG R-peak to a subsequent PPG pulse at the first or second PPG sensor.

The PAT obtained at the upper limb may be compared to that obtained at the lower limb and a difference between the two determined. This difference may be compared to known values to determine whether or not, and/or to what extent, arterial stiffness or DPN is present in the subject. The wearable device may comprise a communication device for uploading data relating to the measurements made by the first and second PPG sensors and/or the or an ECG monitor and/or results from the or a processor to an external device. The data may be transmitted via any suitable means, for example, wired or wireless communication techniques. The external device may be an external computing or storage device.

The wearable device may also comprise a gait sensor for carrying out measurements indicative of a subject's gait. Such measurements are useful in detecting and assessing DPN since DPN commonly results in nerve disruptions, e.g. pain or numbness, in the subject's foot. These nerve disruptions can affect how a subject places their foot when walking and hence may affect the gait of the subject. Accordingly, analysis of gait together with blood flow provides enhanced detection and assessment of the presence or extent of DPN in a subject.

The gait sensor may comprise an accelerometer, for example, a 3D accelerometer (e.g.

Analog Devices ADXL335).

The gait sensor may be sensitive to shock waves generated by Ground Reaction Forces (GRF). GRFs are the forces exerted by the ground in reaction to the forces that body exerts on the ground through the foot. Whenever the heel strikes, a shock wave is generated and propagated through the skeleton to the sensor position.

By analyzing measurements made by the gait sensor, a signature derived from the sensor acceleration can be determined and compared to known signatures to detect certain types of gait. Hence, a gait indicative of DPN can be identified.

Analysis of the measurements made by the gait sensor may be carried out by a processor provided on the wearable device or external to the wearable device.

In a second aspect a system is provided for assessing blood flow in a subject, the system comprising: a wearable device comprising a plurality of photoplethysmography (PPG) sensors comprising a first PPG sensor and a second PPG sensor, each sensor being arranged to carry out measurements indicative of blood flow in the subject, wherein the PPG sensors are disposed such that, when the device is worn on an upper or lower limb of the subject, the first PPG sensor is in contact with the upper or lower limb on which the device is worn, and the second PPG sensor is contactable by the other of the upper or lower limb of the subject; and a processor arranged to compute a pulse arrival time at the portion of the lower limb and at the portion of the upper limb based on the measurements made by the first and second PPG sensors.

It will be appreciated that each of the features described above in relation to the wearable device may apply to the wearable device of the system. All possible combinations are not provided in detail here for the sake of brevity.

The system may comprise an ECG monitor and the system may be arranged such that the PPG sensors and ECG monitor carry out measurements simultaneously. Such an ECG monitor may be arranged to record the electrical activity of the subject's heart. The ECG monitor may be positioned to take measurements at any location on the subject's body.

In some embodiments, the ECG monitor is disposed on the wearable device as described above.

The system may comprise a memory component for storing measurements made by the first and second PPG sensors and/or the or an ECG monitor.

The system, for example the wearable device, may comprise a communication device for transmitting data relating to the measurements made by the first and second PPG sensors and/or the or a ECG monitor to an external device, for example a storage device or the processor. For example, the wearable device may comprise a communication device for transmitting data to the processor.

Any suitable means of communication may be used, e.g. wired or wireless communication.

The system may comprise a gait sensor as described above. The gait sensor may be provided at any location on the subject's body e.g. the head. Specific embodiments are described below by way of example only and with reference to the accompanying drawings in which:

Figure 1 is a diagram illustrating a wearable device as worn by a subject; Figure 2 is a diagram illustrating the wearable device of figure 1 ; Figure 3 is a diagram illustrating the wearable device of figure 1 ;

Figure 4 is a plot of the signals obtained from a PPG sensor directed towards a subject's ankle and from a PPG sensor directed towards a subject's finger;

Figure 5 is flow diagram illustrating how measurements from the wearable device may be used;

Figure 6 is a plot of raw PPG sensor signals captured on the finger and ankle;

Figure 7 is a plot of PPG signals captured by two sensors, one attached to the knee and one attached to the ankle, while pressures of 80mmHg and lOOmmHg were applied to the upper leg;

Figure 8 is a plot of a PPG waveform from a finger and an ECG reference from the chest;

Figure 9a is a plot of raw sensor signals captured by the ECG sensor and PPG sensor on the ankle, and the rectangle highlights the period when a pressure is applied to the upper leg;

Figure 9b is a plot of raw sensor signals captured before pressure was applied;

Figure 9c is a plot of raw sensor signals captured during pressure application;

Figure 9d is a plot of raw sensor signals captured after a pressure cuff was released; Figure 10 is a plot of the raw ECG and PPG signals captured (upper plot) and the corresponding PAT measurement results (lower plot). The black rectangle highlights the period in which pressure was applied to the subject's upper leg using a pressure cuff;

Figures 11a and 1 lb illustrate an alternative wearable device;

Figure 12 is a diagram illustrating a system comprising a wearable device;

Figure 13 illustrates the subplantar regions used for GRF analysis;

Figure 14 illustrates Bayesian networks for detecting steps (left) and left/right foot strikes (right);

Figure 15 illustrates Bayesian networks for capturing the hindfoot, mid-foot, the metatarsals and phalanges contacts;

Figure 16 illustrates Bayesian Networks for capturing medial and lateral foot contacts; and

Figure 17 illustrates Bayesian networks for capturing the contacts at the subdivided regions HO (left) and HI (right).

A wearable device 2 is now described with reference to Figures 1, 2 and 3. The device 2 is arranged such that it may be worn around a subject's ankle and comprises a band having an outward facing sensor portion 6 and an inward facing sensor portion 8. The device 2 also comprises a clasp 10 for fastening the device around the subject's ankle.

The inward facing sensor portion 8 and outward facing sensor potion 6 are disposed on opposite sides of the band. The outward facing sensor portion 6 is positioned such that it is directed away from the subject's ankle and such that the subject may place their finger against the outward facing sensor portion 6 when the device is worn. The inward facing sensor portion 8 is disposed such that it is directed towards the subject's ankle.

As shown in Figure 2, the outward facing sensor portion 6 comprises a

photoplethysmography (PPG) sensor 12 for taking measurements at the subject's finger. As shown in Figure 3, the inward facing sensor portion 8 comprises a second PPG sensor 16 for taking measurements at the subject's ankle.

The PPG sensors 12, 16 each comprise a pair of light emitting diodes (LED), a red LED and an infrared LED, and a photo-detector. Any suitable PPG sensor may be used. In a preferred embodiment, a Hamlyn St0₂ sensor is used for each of the PPG sensors 10, 14. This sensor was developed by Chen et al. at the Hamlyn Centre for Robotic Surgery at Imperial College, London, and is described in further detail in "C.-M. Chen, R. Kwasnicki, B. Lo, and G. Z. Yang, "Wearable Tissue Oxygenation Monitoring Sensor and a Forearm Vascular Phantom Design for Data Validation," in Wearable and Implantable Body Sensor Networks (BSN), 2014 11th International Conference on, 2014, pp. 64-68" which in incorporated herein by reference.

The Hamlyn St0₂ sensor comprises an infrared LED, a red LED, and a photo-detector. With the low power design, the St0₂ sensor consumes only 6.6mW on average when sampling at 1 kHz. The infrared LED emits light at 880nm and the red LED emits light at 740nm. An example of the PPG signals obtained from the PPG sensor 16 directed towards the ankle and the PPG sensor 12 directed towards the finger is shown in Figure 4.

The device 2 also comprises a 1-lead electrocardiogram (ECG) monitor comprising two electrodes. A first ECG electrode 14a is disposed on the outward facing portion and a second ECG electrode 14b is disposed on the inward facing portion 6. When the subject touches both the first and second ECG electrodes 14a, 14b, the electrical circuit is completed and an ECG corresponding to the rhythm of the subject's heart may be obtained. Hence, the ECG monitor records the electrical activity of the heart when the subject places a finger on the ECG first electrode 14a and the subject's ankle is in contact with the second electrode 14b.

The device 2 comprises a communications unit (not shown) which is configured to transmit data relating to the measurements made by the PPG sensors and ECG monitor to an external device for analysis. The data may be transmitted via any suitable means, for example, wired or wireless communication techniques. The external device may be an external computing or storage device. The device 2 also comprises a memory component (not shown) for storing data relating to the measurements made by the PPG sensors 12, 16 and the ECG monitor 14. This stored data may then be transmitted at desired intervals to the external device, for example, daily or weekly. Alternatively, data may be downloaded from the device as and when required.

In some embodiments, the device 2 comprises a processor (not shown) for carrying out analysis of the measurements made by the PPG sensors and ECG monitor. The results of this analysis may then be transmitted from the device to an external device, for example for further analysis, by any suitable means as described above.

In use, the wearable device 2 is positioned around a subject's ankle and fastened in place with the clasp 10. Referring to Figure 5, the PPG sensor 16 directed towards the subject's ankle measures 18 the amount of light emitted by the LEDs which is absorbed and generates a PPG waveform indicative of the pressure pulse in the blood vessels of the ankle. The ECG electrode 14b and PPG sensor 16 directed towards the subject's ankle is placed in contact with the subject's ankle, for example, by fastening the device or by manipulation of the device such that contact is made.

When the subject places their finger against the outward facing PPG sensor 12 and the outward facing ECG electrode 14a, the PPG sensor 12 directed towards the subject's finger measures the amount of light emitted by the LEDs which is absorbed and generates a PPG waveform indicative of the pressure pulse in the blood vessels of the finger, and the ECG monitor measures the electrical activity of the heart and generates an electrocardiogram 18. These measurements are then used by the processor to calculate the Pulse Arrival Time (PAT) at the finger and at the ankle.

The Pulse Arrival Time is the time taken for the blood to flow from the heart to a given location, and is determined by measuring the time difference from a R-peak of the ECG signal obtained from the ECG monitor to the subsequent pulse captured on the finger and ankle respectively, using the PPG sensors. As such, to accurately measure the PAT, a robust ECG and PPG peak/valley detection algorithm is used.

To detect the R-peaks and pulse valleys from the ECG and PPG signals respectively 20, a fixed window based median filter is used to remove the DC components (i.e. noise) of the signals, and the respective peaks and valleys of the signal are then detected by finding the maximum or minimum values within the time window. The PAT is then obtained by measuring the time delay from an ECG R-peak to the subsequent PPG pulse 22 at each PPG sensor of the finger and ankle.

The PAT obtained at the finger is compared to that obtained at the ankle and a difference between the two is determined 24. This difference is then compared to known values 26 to determine whether or not, and to what extent, arterial stiffness or DPN is identified in the subject. Such known values may comprise values obtained experimentally and for which there is a known relationship between the value and the existence or extent of arterial stiffness or DPN in a subject.

An experiment demonstrating performance of a system representative of the wearable device will now be described.

A. EXPERIMENTAL SETUP

The basic experimental setup involved the placement of the Hamlyn PPG sensors at the chosen sites on the body, the ankle and the finger, and fitting a BSN ECG sensor on the chest to capture the R-peaks as reference for measuring PAT. Further detail on the BSN ECG sensor used is given in "B. P. Lo, S. Thiemjarus, R. King, and G.-Z. Yang, Body sensor network-a wireless sensor platform for pervasive healthcare monitoring: na, 2005" which is incorporated herein by reference. To prove the concept of using multiple PPG sensors to detect DPN, a simulated study was conducted with two healthy volunteers. In the study, subjects had the PPG sensors attached onto the knee, finger or ankle, and the ECG sensor attached onto the chest. A pressure cuff was used to simulate arterial stiffness or DPN.

B. PPG SIGNAL AT FINGER AND ANKLE

The first experiment was designed to verify the proposed PPG sensor's ability in capturing pulse rate from the ankle and finger. Figure 4 shows the output waveforms obtained from this procedure. The difference in the shape of the waveforms is clear with the finger signal denoting distinguishable peaks per cycle while the ankle signal has peaks of similar amplitude. This is mainly due to the fact that the fingertip has a much denser vascular structure than the ankle. A blood pressure cuff was wrapped around the lower leg. Arterial stiffness or DPN was simulated by applying different pressures to block the venous blood flow in the leg.

Experimental results are plotted in Figure 6 which plots raw PPG sensor signals captured at the thumb and ankle while a pressure of 80mmHg was applied onto the lower leg. Black arrows in Figure 6 denote a period (30 seconds) where a pressure of 80mmHg was applied at the lower leg. 80mmHg was found to be enough to disrupt venous blood flow but not arterial blood flow. It can be seen that there is a sudden spike of the local signals in that region because of the applied pressure. The sudden spike in the finger signals suggests that application of pressure at a certain body part has a corresponding effect on blood flow at other parts of the body even when blood flow towards the heart (venous flow) has been disrupted.

C. PPG SIGNAL AT ANKLE AND KNEE UNDER DIFFERENT PRESSURE

CONDITIONS

To verify that PPG sensors can detect the variation in arterial blood flow, an experiment was set up where the pressure cuff was wrapped around the upper leg, and PPG sensors were attached onto knee and ankle. Two situations were evaluated: i) Blood vessels were slightly blocked with pressure of 80mmHg ii) Blood vessels are seriously blocked by lOOmmHg, and in both cases, blood flow time delay between knee and ankle were measured. As the distance between the knee and ankle are relatively short, it is difficult to see time delays from the AC components of PPG signal. As shown in Figure 7, there was a small time difference in detecting the pulse wave between the two sensors, and the difference appeared to slightly increase over time. The delay is expected to stabilise after a certain period of time. One obvious variation between signals at both sites is the signal intensity. The knee signal intensity was much higher than the ankle signal intensity. This is most likely due to the fact that the knee was closer to the pressure cuff where applied pressure can lead to vessel narrowing at that point and subsequently lead to turbulence. Most likely, the knee still experienced the effects of this turbulent flow and the ankle experiences steady, laminar flow as the turbulence might have died down before reaching this site so there was less variation in the magnitude of the result signal. D. MEASUREMENT OF PULSE ARRIVAL TIME

An experiment was designed to verify the measurement of pulse arrival time. A chest worn ECG sensor was used as a reference to measure the pulse arrival time at the ankle and the finger. Figure 8 illustrates the raw signals captured by the ECG sensor and the PPG sensor on the finger. The time difference between the R-peak of the ECG signal to the PPG signal is known as the pulse transit time (or pulse arrival time) which is the time taken from the pulse to reach the finger after the ventricular depolarization of the heart. The pulse arrival time was found to be 0.1917s for this instance.

The same experiment was repeated with the PPG sensor placed at the ankle, but this time, a pressure cuff was introduced and 80mmHg pressure was applied to the upper leg. Fig. 9a shows the raw sensor signals captured during the experiment with the time when pressure was applied highlighted with the black rectangle. To allow for proper signal comparison, the graph was divided into three regions: before pressure was applied (Fig. 9b), during pressure application (Fig. 9c) and after pressure cuff was released (Fig. 9d), and the signals are scaled for visual comparison.

The pulse arrival time at the ankle is highlighted in Fig. 9b, 9c and 9d. As shown in Fig. 9b, when no external pressure was applied, the PAT was 0.1533s which is fairly similar to the value of 0.1917s obtained at the finger. During the time when 80mmHg pressure was applied onto the upper leg, there was a significant increase in PAT as shown in Fig. 10c. The pulse arrival time obtained was 0.6786s which is significantly higher than the corresponding values obtained from the finger. In Fig. 9d, the PPG signal after the cuff was released is shown. Three different points where chosen to illustrate the behaviour of the pulse arrival times after the cuff was released. The PAT for point 1, 2 and 3 were 0.6429s, 0.4500s and 0.3571s respectively. The PAT was then measured and the results are shown in Figure 10. The raw ECG and PPG signals captured on the subject's ankle are shown in Figure 10 (Top), and the corresponding PAT is shown in Figure 10 (Bottom). As shown in Figure 10, the PAT during the period when pressure was applied onto the upper leg (highlighted period) is higher than the other two periods - before introducing pressure/after the pressure was released.

It is clear from these results that the deviations in PAT at the ankle compared to the finger can be used to provide an indication of DPN or arterial stiffness. In an alternative embodiment shown in Figures 1 1a and 1 lb, the wearable device 20 comprises the same features as outlined above but does not comprise an ECG monitor. In this case, measurements from an ECG monitor external to the device are used to calculate deviations in PAT at the ankle compared to the finger. These calculations may be carried out at the wearable device itself. In this case, ECG data is transmitted to the wearable device from the external ECG monitor. Alternatively or in addition, PAT calculations may be carried out by a processing device external to the wearable device, as described below.

In another embodiment as shown in Figure 12, a system 200 is provided which comprises a wearable device 202. The wearable device comprises the same features as outlined above in relation to either the embodiment with or without the ECG monitor, but does not comprise a processor. Instead, a processor 204 is provided external to the wearable device 202.

Measurements made by the PPG sensors and, if present, on the wearable device, the ECG monitor 202 are stored in a memory component on the wearable device 202. At desired intervals, for example weekly, or on demand, data from the storage device is transmitted to the external processor 204 via a communications unit 206 provided on the wearable device. The data may be transmitted via any suitable means, for example, wired or wireless communication techniques. The processor 204 then analyses the received data to calculate the Pulse Arrival Time (PAT) at the finger and at the ankle using the techniques described above. Alternatively, data may be transmitted directly to the external processor without storage on the wearable device.

In some embodiments, the system or wearable device comprises a gait sensor. Typically, the gait sensor is a 3D accelerometer e.g. Analog Devices ADXL335.

An example gait sensor, the e-AR (ear- worn Activity Recognition) sensor, will now be described. It is noted that this is by way of example only and that any suitable gait sensor worn at any location on the body may be used e.g. on the ankle as part of the wearable device.

The e-AR (ear-worn Activity Recognition) sensor is a bio-inspired sensor designed to emulate the main functions of the human vestibular system. It mainly consists of an 8051 processor with a 2.4 GHz transceiver (Nordic nRF24El), a 3D accelerometer (Analog Devices ADXL335), a 2MB EEPROM (Atmel AT45DB161), and a 55mAhr Li-Polymer battery.

In addition to capturing posture and activities, the e-AR sensor is also sensitive to shock waves generated by GRF. GRFs are the forces exerted by the ground in reaction to the forces that body exerts on the ground through the foot. Whenever the heel strikes, a shock wave is generated and propagated through the skeleton to the cranium, which may be picked up from the superior and posterior auricular regions. The axial skeleton does not dissipate the transients or affect their interference where matching signals were found from the

accelerometer mounted on the forehead and the accelerometer attached to the tibial tuberosity. This means the axial skeleton is a good conductor for the transient of both high and low-frequency waves generated by the GRFs.

Among different parameters of the GRF, we have focused on the plantar force distribution in this study to demonstrate the concept of using the e-AR sensor for gait analysis. Previous research has shown the significance of the plantar force distribution in analyzing pathological gaits. In this study, instead of estimating the foot pressures, the instances of ground contact at different regions of the foot are inferred by using the e-AR sensor.

Figure 13 schematically illustrates the sub-plantar regions used for this study. The forefoot is divided into phalanges (toes) (T) and metatarsals (F), whereas M represents the midfoot (the five tarsal bones) and H represents the hindfoot (the talus and the calcaneus or the heel). Each region is further separated into lateral (O) and medial (I) halves. For example, HO means the lateral hindfoot region.

In this study, a Bayesian Network approach has been chosen to analyse the eAR sensor signals. Its graphical and causal structure representation enables visual and statistical modelling of the underlying problem domain.

A hierarchical Bayesian network is designed to estimate the plantar force distribution. The network is designed to match different phases of the foot strike (i.e. three layers of networks are designed to detect foot step, heel (hindfoot) strike, and lateral hindfoot strike). The parameters of the network are learned from the sensor data (20% of all the sensor data was used for the learning phase). At the bottom layer of the network, a naive Bayesian is constructed to detect steps where the three orthogonal signals of the e-AR sensor (i.e., e-ARx, e-ARy, e-ARz correspond to the lateral, vertical and fore-aft accelerations) are represented as the child nodes as shown in Figure 14 (left).

Each step is detected by calculating the maximum a-posteriori probability (MAP):

Step = arg max P(Step \eAR_x, eAR_y, eAR_z ^{^})

arg max a P(eARi \Step)P(Step)

i=x,y,z

(1) where a is a normalizing constant. As the 3D accelerometer captures the acceleration in three orthogonal directions, conditional independence can be assumed among these sensor signals.

The output of the step detection together with the accumulated signal Ac are then fused to determine whether it is a left or right foot strike as shown in Figure 14 (right). The posterior probability can therefore be calculated as follows:

P(LR \ Ac, Step) = P(Ac | LR)P(Step | LR)P(LR)

(2)

where variable Ac is the accumulated value of the lateral acceleration (i.e. e-ARx).

_{A (} , ₌ (eAR_x(t) + eAR_x(t - 1)^< eAR_y > 0

0 ' else

(3)

To capture the stance phase of the gait (i.e., from heel strike to toe off), the vertical acceleration signal is fused with the output of the step-detection network to infer the force contact at different areas of the foot. This is depicted in Figure 15. To detect heel-strike, a high pass filter is used to extract the high frequency shock waves and the posterior probability of the heel (hindfoot) strike is formulated as follows:

P(H I HP, Step) = aP(HP | H)P(Step | H)P(H)

where HP is the high pass filtered signal of the vertical acceleration (i.e., e-ARy) signal from the e-AR sensor.

It can be assumed that the stance phase is Markovian, to enhance the inference of the midfoot detection, the past states of the hindfoot contact is fused with the vertical acceleration and the posterior probability is calculated as follows:

(5) where H_t.i represents the state of the hindfoot detection in the previous time step. Likewise, the posterior probabilities of the metatarsals (F) and phalanges (T) contact detections are formulated as follows:

k=M_t-₁,eAR_y,Step

(6)

P^F_t^ F^ eARy. Step) =

a P(k\T P(T

k=F_t-₁,F_t-₂,eAR_y,Step

(V)

The medial (I) and lateral (O) foot contacts are then inferred by fusing the lateral acceleration with the step detection, and the corresponding Bayesian network is depicted in Figure 16. Accordingly, the posterior probabilities are calculated as follows:

P(I I eAR_x, Step) aP (eAR_x \ I) P(Step 1 I) P (I)

P(0 I eAR_x, Step) aP (eAR_x \ O) P(Step \ O) P (O)

(9)

The detection in the subdivided regions, such as HO, is inferred by fusing the results from the corresponding regions, such as H and O. Figure 17 depicts the Bayesian networks for detecting the HO and HI contacts, and the posterior probabilities for the subdivided regions are formulated as follows:

(10)

where

ψ = {HO, HI, MO, MI, FO, FI, TO, Tlj

ξ = {H H M, M, F, F, T, T}

ς = {O, I, O, I, O, I, 0, 1} ψ represents the series of subdivided regions, ξ represents the series of the hindfoot to phalanges regions, and ς represents the series of medial and lateral regions.

It will be understood that the above description is of specific embodiments by way of example only and that many modifications and alterations will be within the skilled person's reach and are intended to be covered by the scope of the appendent claims. For example, in some embodiments, the wearable device comprises a processor for computation of a difference in the pulse time of arrival at each of the PPG sensor measurement locations. In some embodiments, ECG measurements may be obtained from a monitor separate to the wearable device and transmitted to an external device for computation of the pulse arrival time at each of the PPG sensor measurement locations. In some embodiments, the device is arranged to be worn on a subject's upper limb e.g. a wrist or finger, such that the subject may bring the device into contact with his or her lower limb e.g. an ankle.

Claims

A wearable device for assessing blood flow in a subject, the device comprising a plurality of photoplethysmography (PPG) sensors comprising a first PPG sensor and a second PPG sensor, each sensor being arranged to carry out measurements indicative of blood flow in the subject, wherein the PPG sensors are disposed such that, when the device is worn on an upper or lower limb of the subject, the first PPG sensor is in contact with the upper or lower limb on which the device is worn, and the second PPG sensor is contactable by the other of the upper or lower limb of the subject. A wearable device according to claim 1 , wherein the first and second PPG sensors are positioned on opposite sides of the device such that, when worn, the second PPG sensor is directed away from the upper or lower limb on which the device is worn. A wearable device according to claim 1 or 2, wherein the device further comprises an electrocardiogram (ECG) monitor, wherein the PPG sensors and ECG monitor are configured to carry out measurements simultaneously.

A wearable device according to claim 3, wherein the ECG monitor comprises a first ECG electrode and a second ECG electrode, wherein the first ECG electrode is disposed such that, when the device is worn, the first ECG electrode is in contact with the upper or lower limb on which the device is worn and the second ECG electrode is contactable by the other of the upper and lower limb of the subject.

A wearable device according to claim 4, wherein the second ECG electrode is disposed adjacent the second PPG sensor.

A wearable device according to any preceding claim, comprising a memory component for storing measurements made by the first and second PPG sensors and/or the or an ECG monitor.

A wearable device according to any preceding claim, wherein the device further comprises a processor for computing a pulse arrival time at the lower limb and at the upper limb based on the measurements made by the first and second PPG sensors. A wearable device according to claim 7, wherein the processor also uses the measurements made by the or a ECG monitor to compute the pulse arrival time at the lower limb and at the upper limb.

A wearable device according to any preceding claim, comprising a communication device for transmitting data relating to the measurements made by the first and second PPG sensors and/or the or a ECG monitor to an external device.

10. A wearable device according to any preceding claim, wherein the lower limb is the subject's ankle, knee or foot, and wherein the upper limb is the subject's hand, finger or thumb.

11. A wearable device according to any preceding claim, further comprising an additional sensor configured to carry out measurements indicative of a gait of the subject.

12. A wearable device according to claim 11, wherein the additional sensor comprises an accelerometer.

13. A system for assessing blood flow in a subject, the system comprising:

a wearable device comprising a plurality of photoplethysmography (PPG) sensors comprising a first PPG sensor and a second PPG sensor, each sensor being arranged to carry out measurements indicative of blood flow in the subject, wherein the PPG sensors are disposed such that, when the device is worn on an upper or lower limb of the subject, the first PPG sensor is in contact with the upper or lower limb on which the device is worn, and the second PPG sensor is contactable by the other of the upper or lower limb of the subject; and

a processor arranged to compute a pulse arrival time at the lower limb and at the upper limb based on the measurements made by the first and second PPG sensors.

14. A system according to claim 13 wherein the first and second PPG sensors are

disposed on opposite sides of the device such that, when worn, the second PPG sensor is directed away from the upper or lower limb on which the device is worn.

15. A system according to claim 13 or 14, further comprising an ECG monitor, wherein the PPG sensors and ECG monitor are configured to carry out measurements simultaneously.

16. A system according to claim 15, wherein the device comprises the ECG monitor.

17. A system according to claim 16, wherein the ECG monitor comprises a first ECG electrode and a second ECG electrode, wherein the first ECG electrode is disposed such that, when the device is worn, the first ECG electrode is in contact with the upper or lower limb on which the device is worn and the second ECG electrode is contactable by the other of the upper and lower limb of the subject.

18. A wearable device according to claim 17, wherein the second ECG electrode is

disposed adjacent the second PPG sensor.

19. A system according to any of claims 13 to 18, the system further comprising a

memory component for storing measurements made by the first and second PPG sensors and/or the or a ECG monitor.

20. A system according to any of claims 13 to 19, wherein the wearable device comprises a communication device for transmitting data relating to the measurements made by the first and second PPG sensors and/or the or a ECG monitor to the processor or other external device.

21. A system according to any of claims 13 to 20, further comprising an additional sensor configured to carry out measurements indicative of a gait of the subject.

22. A system according to claim 21, wherein the additional sensor comprises an

accelerometer.