EP4687650A1

EP4687650A1 - Methods and devices for monitoring user physiology using photoplethysmography signals

Info

Publication number: EP4687650A1
Application number: EP24781428.8A
Authority: EP
Inventors: Yin Kwee NG; Kei Fong Mark WONG; Weiting HUANG
Original assignee: Nanyang Technological University; Singapore Health Services Pte Ltd
Current assignee: Nanyang Technological University; Singapore Health Services Pte Ltd
Priority date: 2023-03-30
Filing date: 2024-04-01
Publication date: 2026-02-11
Also published as: CN121311170A; WO2024205510A1; JP2026511795A

Abstract

The present disclosure generally relates to a system and computerized method (200) for monitoring physiology of a user using photoplethysmography (PPG) signals (310). The method (200) comprises: receiving (210) a set of PPG signals (310), each PPG signal measured from a different location on the user; identifying a plurality of 2D pulse wavelets (312) from each PPG signal (310); aligning the respective 2D pulse wavelets (312) in 3D; and constructing a set of 3D morphological representations (400) from the aligned pulse wavelets (312), wherein a set of physiological parameters (410) of the user is measurable based on the constructed 3D morphological representations (400).

Description

METHODS AND DEVICES FOR MONITORING USER PHYSIOLOGY USING PHOTOPLETHYSMOGRAPHY SIGNALS

Cross Reference to Related Application(s)

The present disclosure claims the benefit of Singapore Patent Application No. 10202300873U filed on 30 March 2023, which is incorporated in its entirety by reference herein.

Technical Field

The present disclosure generally relates to methods and devices for monitoring user physiology using photoplethysmography signals. More particularly, the present disclosure describes various embodiments of methods and devices for monitoring physiological parameters of users, such as blood pressure, using photoplethysmography signals measured from the users.

Background

Many monitoring systems use a variety of mechanical, electrical, and optical components to measure physiological conditions of users, such as cardiac monitoring systems for measuring cardiovascular parameters. One example is ambulatory sphygmomanometers that incorporate an inflatable cuff that encircles the user's limbs. Another example is electrocardiograms, such as Holter monitors, that require cumbersome and disruptive patch electrodes. The application of such electrodes and cuffs can be quite cumbersome and impractical during prolonged use or heavy physical exertion. In addition, inflation of the cuff causes significant constriction of the user's limbs, making the cuff unsuitable for continuous wear during the day or, more critically, during the night.

Blood pressure can be measured indirectly using the auscultatory or oscillometric methods. The auscultatory method is based on the detection of Korotkoff sounds issued from the acoustic transducer signal, whereas the oscillometric method is based on the detection of volume variations in the cuff. In both auscultatory and oscillometric methods, blood pressure is measured through the occlusion of the artery, making the measurement protocol unsuitable for rapid measurements. Although the brachial cuff method is the clinically accepted standard for blood pressure monitoring, it is not suitable for certain situations where continuous blood pressure monitoring is desirable, such as daily orthostatic hypotension. For example, cuff-based blood pressure monitors cannot be used to monitor sleep patterns without disturbing patients, as the inflation of brachial or wrist cuff pressure along with an increase in systemic blood pressure causes restlessness.

Orthostatic hypotension or postural hypotension is an abnormally severe drop in blood pressure from a sitting or lying position that causes decreased blood flow to the brain, such as shown in Figure 1A. This can lead to symptoms such as dizziness or lightheadedness, unsteadiness, visual disturbances, and sometimes fainting. The risk for adverse outcomes is increased if there are no symptoms as it is closely related to the risk of falling and can affect a person's quality of life.

There are four major subtypes of orthostatic hypotension - initial orthostatic hypotension, delayed blood pressure recovery, classic orthostatic hypotension, and delayed orthostatic hypotension. Clinical symptoms are varied among the subtypes, ranging from cognitive slowing with unconscious hypotension or unexplained falls to classic presyncope and syncope.

Orthostatic hypotension diagnosis is made when the blood pressure drops by 20 mmHg systolic and 10 mmHg or more diastolic, within 3 minutes of standing up after lying supine or at a 60° angle on a tilt table for 5 minutes. The sudden drop in blood pressure may be due to autonomic reflex failure, volume deficiency, or an adverse reaction to medication. Symptoms are usually related to decreased blood flow to the brain, but many patients may be asymptomatic. Because of this disease process, falls are common, resulting in high morbidity and mortality rates and numerous hospitalizations. Diagnosis of orthostatic hypotension is difficult and current practice requires a detailed medical history, examination, and both supine and upright blood pressure measurements. In-clinic assessments require the patient to undergo blood pressure measurements in the following 3 or 4 scenarios - 5 minutes after lying, 1 minute after standing, 3 minutes after standing, and a tilt table measurement may be required for these measurements.

These measurements are time-consuming, labour-intensive, and require a physician to administer. Moreover, orthostatic hypotension could be asymptomatic and may not present any signs in the clinic, resulting in the measurements being inconclusive. A physician could then ask the patient to wear an ambulatory blood pressure monitor for 24 hours, such as shown in Figure 1 B. However, the ambulatory blood pressure monitor, which comprises a blood pressure machine 100 and a cuff 110, is occlusive, uncomfortable, and does not measure posture changes of the patient.

Conventional monitoring devices are used in clinical settings where the user is usually stationary. For example, a patient under medical observation is either sitting or lying down. However, these devices are not suitable for individuals who often need access to blood pressure or cardiac data while performing a range of daily tasks, such as during office work, driving, or sporting activities. Conventional devices designed for clinical assessment of such physiological parameters are ineffective or impractical for continuous monitoring during routine activities or sleep.

Therefore, in order to address or alleviate at least one of the aforementioned problems and/or disadvantages, there is a need to provide improved methods and devices for monitoring physiological parameters of users.

Summary

According to a first aspect of the present disclosure, there is a system and a computerized method for monitoring physiology of a user using PPG signals. The method comprises: receiving a set of PPG signals measured from the user, each PPG signal measured from a different location on the user; identifying a plurality of pulse waves from each PPG signal, , each pulse wavelet defined in 2D by an amplitude axis and a first time axis; for each PPG signal, aligning the respective 2D pulse wavelets in 3D along a second time axis; and constructing a set of 3D morphological representations from aligned pulse wavelets, wherein a set of physiological parameters of the user is measurable based on the constructed 3D morphological representations.

According to a second aspect of the present disclosure, there is a measurement device for monitoring haemodynamic conditions of a user using PPG signals. The measurement device comprises: a plurality of PPG sensors for measuring a plurality of PPG signals from different locations on the user; an inertial measurement unit for measuring motion data from the user; and a processor configured for: calculating a first haemodynamic profile of the user from the PPG signals and a first machine learning model; and calculating data second haemodynamic profile of the user from the first haemodynamic profile, motion data, and a second machine learning model.

According to a third aspect of the present disclosure, there is a system and a computerized method for monitoring haemodynamic conditions of a user using PPG signals. The method comprises: receiving a plurality of PPG signals measured from the user, each PPG signal measured from a different location on the user; receiving motion data measured from the user using an inertial measurement unit; generating a first haemodynamic profile of the user from the PPG signals and a first machine learning model; and generating a second haemodynamic profile of the user from the first haemodynamic profile, motion data, and a second machine learning model.

Methods and devices for monitoring user physiology using photoplethysmography signals according to the present disclosure are thus disclosed herein. Various features, aspects, and advantages of the present disclosure will become more apparent from the following detailed description of the embodiments of the present disclosure, by way of non-limiting examples only, along with the accompanying drawings.

Brief Description of the Drawings

Figures 1A and 1 B are illustrations of an existing method of measuring orthostatic hypotension.

Figures 2A and 2B are illustrations of a method for monitoring physiology of a user using PPG signals, according to embodiments of the present disclosure.

Figures 3A and 3B are illustrations of PPG sensors arranged on a user and PPG signals measured by the PPG sensors.

Figures 4A to 4F are illustrations of identifying pulse wavelets of the PPG signals.

Figures 5A and 5B are illustrations of aligning pulse wavelets of the PPG signals.

Figures 6A to 6D are illustrations of 3D morphological representations constructed from the aligned pulse wavelets.

Figures 7 A to 7D are illustrations of a grid matrix for aligning the pulse wavelets.

Figure 8 is an illustration of another grid matrix for aligning the pulse wavelets.

Figures 9A to 9J are illustrations of another grid matrix for aligning the pulse wavelets. Figures 10A to 10D are illustrations of grid matrices and 3D morphological representations constructed from different numbers of PPG signals.

Figures 11 A to 11 C are illustrations of 3D morphological representations for different users.

Figures 12A and 12B are illustrations of two PPG signals and their pulse wavelets.

Figures 13A to 13E are illustrations of calculating a pulse transit time from two PPG signals.

Figures 14A to 14C are illustrations of estimating blood glucose levels from the 3D morphological representations.

Figures 15A and 15B are illustrations of a system and method for monitoring haemodynamic conditions of a user using PPG signals, according to embodiments of the present disclosure.

Figures 16A and 16B are illustrations of a measurement device for measuring PPG signals and motion data to monitor haemodynamic conditions of a user, according to embodiments of the present disclosure.

Figures 17A to 17C are results of estimated and actual blood pressures for a group of subjects.

Figure 18 is an illustration of a system for monitoring physiology of a user using PPG signals, motion data, and the 3D morphological representations, according to embodiments of the present disclosure.

Detailed Description

For purposes of brevity and clarity, descriptions of embodiments of the present disclosure are directed to methods and devices for monitoring user physiology using photoplethysmography signals, in accordance with the drawings. While aspects of the present disclosure will be described in conjunction with the embodiments provided herein, it will be understood that they are not intended to limit the present disclosure to these embodiments. On the contrary, the present disclosure is intended to cover alternatives, modifications and equivalents to the embodiments described herein, which are included within the scope of the present disclosure as defined by the appended claims. Furthermore, in the following detailed description, specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be recognized by an individual having ordinary skill in the art, i.e. a skilled person, that the present disclosure may be practiced without specific details, and/or with multiple details arising from combinations of aspects of particular embodiments. In a number of instances, well-known systems, methods, procedures, and components have not been described in detail so as to not unnecessarily obscure aspects of the embodiments of the present disclosure.

In embodiments of the present disclosure, depiction of a given element or consideration or use of a particular element number in a particular figure or a reference thereto in corresponding descriptive material can encompass the same, an equivalent, or an analogous element or element number identified in another figure or descriptive material associated therewith.

References to “an embodiment / example”, “another embodiment I example”, “some embodiments / examples”, “some other embodiments I examples”, and so on, indicate that the embodiment(s) / example(s) so described may include a particular feature, structure, characteristic, property, element, or limitation, but that not every embodiment I example necessarily includes that particular feature, structure, characteristic, property, element or limitation. Furthermore, repeated use of the phrase “in an embodiment I example” or “in another embodiment I example” does not necessarily refer to the same embodiment / example.

The terms “comprising”, “including”, “having”, and the like do not exclude the presence of other features / elements I steps than those listed in an embodiment. Recitation of certain features / elements / steps in mutually different embodiments does not indicate that a combination of these features I elements I steps cannot be used in an embodiment.

As used herein, the terms “a” and “an” are defined as one or more than one. The use in a figure or associated text is understood to mean “and/or” unless otherwise indicated. The term “set” is defined as a non-empty finite organization of elements that mathematically exhibits a cardinality of at least one (e.g. a set as defined herein can correspond to a unit, singlet, or single-element set, or a multiple-element set), in accordance with known mathematical definitions. The recitation of a particular numerical value or value range herein is understood to include or be a recitation of an approximate numerical value or value range. The terms “first”, “second”, etc. are used merely as labels or identifiers and are not intended to impose numerical requirements on their associated terms.

In representative or exemplary embodiments of the present disclosure, there is a computer-implemented or computerized method 200 for monitoring physiology of a user using photoplethysmography (PPG) signals 310, as shown in Figures 2A and 2B. The method 200 includes a step 210 of receiving a set of PPG signals 310 measured from the user, each PPG signal 310 measured from a different location on the user. The PPG signals 310 are measured from a set of PPG sensors 300 positioned at one or more locations on the user’s body. For example, the PPG sensors 300 are configured to measure the PPG signals 310 at frequencies of at least 300 or 400 Hz.

In some embodiments, the set of PPG signals 310 includes only one PPG signal 310 measured from a single location on the user’s body. In other embodiments, the set of PPG signals 310 includes a plurality of PPG signals 310. Preferably, an array of PPG sensors 300 is placed at multiple locations on the user to measure multiple PPG signals 310.

The PPG sensors 300 optically measure changes in arterial blood volume and is synchronized with the cardiac cycle. Each PPG sensor 300 has a light-emitting diode (LED) and a photodiode. The LED emits light towards the user’s skin at the respective location and the change in blood volume at that location is measured from the amount of light transmitted or reflected to the photodiode. For example, a PPG sensor 300 placed at the user’s forehead measures light reflection from the forehead, while a PPG sensor 300 placed at the user’s fingertip measures light transmissive absorption by the fingertip.

The PPG sensors can be positioned at various locations on the user. For example as shown in Figure 3A, a PPG sensor 300a is placed at the forehead near the temporal artery, a PPG sensor 300b is placed at the earlobe, a PPG sensor 300c is placed at the clavicle near the axillary artery, a PPG sensor 300d is placed at the inside of the elbow near the brachial artery, a PPG sensor 300e is placed at a proximal wrist site near the radial artery, a PPG sensor 300f is placed at a distal wrist site near the radial artery, a PPG sensor 300g is placed at a fingertip near the palmar digital artery, and a PPG sensor 300h is placed at a toe. Figure 3B shows respective PPG signals 310 respectively measured from the forehead PPG sensor 300a, earlobe PPG sensor 300b, clavicle PPG sensor 300c, proximal wrist PPG sensor 300e, fingertip PPG sensor 300g, and toe PPG sensor 300h.

The method 200 includes a step 220 of identifying a plurality of pulse wavelets 312 from each PPG signal 310. Each pulse wavelet 312 represents a single pulse in the cardiac cycle and exhibits a wave-like oscillation. Each pulse wavelet 312 is defined in two dimensions (2D) by an amplitude axis and a first time axis. Notably, the amplitude axis is the vertical axis and the first time axis is the horizontal axis.

Various ways can be used to identify the pulse wavelets 312. For example, good quality pulse wavelets 312 are identified from the PPG signal 310 while poor quality pulse wavelets 312 are discarded from the PPG signal 310. Figure 4A shows a periodicity windowing algorithm used to identify the good quality pulse wavelets 312. This algorithm detects periodicity in time series analysis and involves 4 steps to recognize patterns that repeat at regular intervals. In the first transformation step, the discrete Fourier transform (DFT) converts the PPG signal 310 from the time domain to the frequency domain, making it easier to recognize periodicities and remove high- frequency domains that typically do not occur biologically, i.e. denoising the PPG signal 310. In the second step, important morphological features 314 of the PPG signal 310 are detected on priority of occurrence. Peaks corresponding to the frequencies of the periodic components are detected in both the time and frequency domains. The important morphological features 314 including the peaks and troughs of the PPG signal 310 are identified and subjected to a series of conditions to classify them as one of the following: systolic peak 314a, trough 314b, dicrotic notch 314c, or diastolic peak 314d (see Figure 13B). Morphological structure measures based on periodic features may be extracted from sequences of the morphological features 314. Periodic distance measures may be used for more meaningful structural clustering and visualisation of sequences of the morphological features 314, regardless of whether they are periodic or not. Once the peaks 314a and troughs 314b are identified, the PPG signal 310 can be split or windowed into individual pulse wavelets 312 with their respective prominent peaks 314a and troughs 314b.

As shown in Figure 4B, the order of occurrence of the morphological features 314 in each pulse wavelet 312 is trough 314b, systolic peak 314a, dicrotic notch 314c, diastolic peak 314d, and trough 314b. An initial windowing is performed indicating that a single pulse wavelet 312 consists of a single systolic peak 314a located between two troughs 314b. The dicrotic notch 314c is identified by localizing the 2nd derived pulse and the frequency range. The diastolic peak 314d is determined by localizing the subsequent gradient change in the 1st derived pulse, which corresponds to the point where the 1 st derived pulse crosses the x-axis or first time axis. Each pulse wavelet 312 should have exactly one systolic peak 314a, one dicrotic notch 314c, and one diastolic peak 314d. The detection of both the dicrotic notch 314c and diastolic peak 314d is acceptable, whereas the absence of either is not acceptable. In this way, pulse wavelets 312 are classified as good quality or poor quality based on the presence of these morphologically features 314 that indicate the pulse wavelet 312 is morphologically sound.

Alternatively or additionally, the pulse wavelets 312 may be classified as good quality or poor quality based on a time interval-based segmentation method. The PPG signal 310 is segmented based on a fixed time interval. The intervals would be categorized as ‘good’ if a potential pulse wavelet 312 could be found within the interval, and otherwise as ‘noisy’ or ‘poor’. For example, a pulse wavelet 312 could never be shorter than 50 ms or longer than 3 seconds, because that would indicate a heart rate faster than 300 BPM (beats per minute) or slower than 20 BPM, respectively, both of which are humanly impossible.

In the third step, an autocorrelation function is calculated to detect any self-similarity in the time series at different delays that may indicate periodicity. Therefore, the autocorrelation is the inverse Fourier transform of the periodogram, which means that the autocorrelation function can be considered as the dual of the periodogram, from the time domain into the frequency domain. Next, based on predefined selection criteria, each individual pulse wavelet 312 would be classified as good or poor quality, regarding their resemblance to pulse wavelets 312 that have occurred prior. For example, as shown in Figure 4C, an input pulse wavelet 312’ would be classified as good quality if the difference between the input pulse wavelet 312’ and a reconstruction of prior pulse wavelets 312” is small. Conversely as shown in Figure 4D, an input pulse wavelet 312’ would be classified as poor quality if the difference between the input pulse wavelet 312’ and a reconstruction of prior pulse wavelets 312” is large.

In the fourth step, the detected periodicities are clustered and filtered to remove false positives and identify the dominant periodicities. The resultant good quality pulse wavelets 312 are then stitched together to construct a good quality PPG signal 310’ as shown in Figure 4A. The good quality PPG signal 310’ can be used to interpret the recognized periodicities in the context of the original time series of the original PPG signal 310.

In some embodiments as shown in Figures 4E and 4F, alternate pulse wavelets 312 may be identified from the good quality pulse wavelets 312 to construct the good quality PPG signals 310’. For example, even-numbered good quality pulse wavelets 312 are identified and stitched to construct the PPG signals 310’. The method 200 includes a step 230 of, for each PPG signal 310, aligning the 2D pulse wavelets 312 along a second time axis. Notably, the second time axis is perpendicular to the amplitude axis and the first time axis. Aligning the pulse wavelets 312 may be performed using a process known as image alignment or image registration, which involves transforming the image data from each pulse wavelet 312 into a common coordinate system. For example, a feature-based method may be used to image align or register the pulse wavelets 312. The feature-based method establishes a correspondence between a set of morphological features 314 of the pulse wavelets 312, and a geometrical transformation may be determined and thereby establish feature-to-feature correspondence between the pulse wavelets 312. For example, the morphological features 314 may include one or more of the systolic peak 314a, pulse trough 314b, dicrotic notch 314c, and diastolic peak 314d. Figures 5A and 5B show examples of aligning the pulse wavelets 312 based on their morphological features 314 including at least the systolic peak 314a.

The method 200 includes a step 240 of constructing a set of three-dimensional (3D) morphological representations 400 from the aligned pulse wavelets 312. Figures 6A to 6D show exemplary 3D morphological representations 400 constructed from the aligned pulse wavelets 312 which were derived from PPG signals 310 measured over durations of 30 seconds, 3 minutes, 30 minutes, and 2 hours, respectively. Notably, the 3D morphological representation 400 is defined by the vertical amplitude axis and two horizonal time axes. The 3D morphological representation 400 may also be referred to as a photoplethysmorphogram.

In some embodiments as shown in Figures 7A to 7D, aligning the pulse wavelets 312 may include arranging each pulse wavelet 312 over a grid matrix 500 having an array of cells 510. A set of morphological features 314 is identified from each pulse wavelet 312, such as one or more of the systolic peak, pulse trough, dicrotic notch, and diastolic peak. A morphological feature 314 is added to a cell 510 of the grid matrix 500 when it is aligned with the respective cell 510, such as shown in Figure 7A. Subsequent pulse wavelets 312 are successively added to the respective cells 510 when their respective morphological features 314 are aligned to the respective cells 510, as shown in Figures 7B to 7D. The resultant grid matrix 500 is thus a summation of the morphological features 314 from the aligned pulse wavelets 312.

In the example as shown in Figures 7 A to 7D, the cells 510 of the grid matrix 500 are arranged in 5 rows and 5 columns, and 4 pulse wavelets 312 are summed in the grid matrix 500. In another example as shown in Figure 8, 6 pulse wavelets 312 are summed in a second grid matrix 500 having cells 510 arranged in 6 rows and 6 columns. In another example as shown in Figures 9A to 9 J , 10 pulse wavelets 312 are summed in a third grid matrix 500 having cells 510 arranged in 10 rows and 10 columns. The third grid matrix 500 is finer than the first and second grid matrices 500 and can be used to construct a 3D morphological representation 400 with higher resolution.

It will be appreciated that a grid matrix 500 can have any number of cells 510 arranged in any number of rows and columns. It will also be appreciated that any number of pulse wavelets 312 can be added to the grid matrix. For example, Figures 10A to 10D show exemplary grid matrices 500 and 3D morphological representations 400 constructed from pulse wavelets 312 extracted from 5, 10, 20, and 30 PPG signals 310, respectively.

As described above, a set of 3D morphological representations 400 is constructed from the aligned pulse wavelets 312. The set of 3D morphological representations 400 represents a concatenation of the PPG signals 310 measured from one or more locations on the user’s body. The set of 3D morphological representations 400 is unique to the user and a set of physiological parameters 410 of the user is measurable based on the respective set of 3D morphological representations 400. For example, the physiological parameters 410 may include blood pressure, blood glucose, arterial distensibility, vascular perfusion, heart rate variability, respiratory rate, and/or stress parameters.

Figures 11A to 11C show some exemplary 3D morphological representations 400 that are unique to different users. Each 3D morphological representation 400 may be compared to other 3D morphological representations 400 to measure physiological parameters 410 and/or diagnose medical conditions for the user. For example, a user’s 3D morphological representation 400 may be compared to earlier 3D morphological representations 400 for the same user, a group of 3D morphological representations 400 for users with similar health profiles, and/or 3D morphological representations 400 for a larger population of people.

As shown in Figure 11 A, the first 3D morphological representation 400A was constructed for a 28-year-old male user using PPG signals 310 measured over a 3- minute duration. The first 3D morphological representation 400A is indicative that the user has a blood pressure of 111/76 mmHg and a blood glucose of 74 mg/dL, and does not have any significant medical conditions.

As shown in Figure 11 B, the second 3D morphological representation 400B was constructed for a 34-year-old female user using PPG signals 310 measured over a 3- minute duration. The second 3D morphological representation 400B is indicative that the user has a blood pressure of 119/68 mmHg and a blood glucose of 66 mg/dL, and does not have any significant medical conditions.

As shown in Figure 11 C, the third 3D morphological representation 400C was constructed for a 46-year-old male user using PPG signals 310 measured over a 3- minute duration. The third 3D morphological representation 4000 is indicative that the user has a blood pressure of 114/74 mmHg and a blood glucose of 97 mg/dL, and does not have any significant medical conditions.

In some embodiments, constructing the set of 3D morphological representations 400 may include constructing, for each PPG signal 310, a respective 3D morphological representation 400 from the respective pulse wavelets 312 identified from the respective PPG signal 310, such that each 3D morphological representation 400 corresponds to one PPG signal 310. If multiple PPG signals 310 were measured from the user, the same number of 3D morphological representations 400 would be constructed for the user. For example as shown in Figure 12A, two PPG signals 310A, 310B are measured from the user at different sites. The pulse wavelets 312A,312B are respectively identified from each PPG signal 310A, 310B. The pulse wavelets 312A,312B from each PPG signal 31 OA, 31 OB are respectively aligned to thereby construct a respective 3D morphological representation 400 corresponding to the respective PPG signal 31 OA, 31 OB.

In some embodiments, constructing the set of 3D morphological representations 400 may include constructing a single 3D morphological representation 400 from all the pulse wavelets 312 identified from all the PPG signals 310. If multiple PPG signals 310 were measured from the user, only one 3D morphological representation 400 would be constructed for the user. For example, as shown in Figure 12B, the pulse wavelets 312A,312B are extracted from the respective PPG signals 310A, 310B. Figure 12B shows two sets of exemplary pulse wavelets 312A,312B from two time windows - indicated as 312A’,312B’ in the first time window, and 312A”,312B” in the second time window. The pulse wavelets 312A,312B from the whole PPG signals 310A, 310B are collectively aligned to thereby construct a single 3D morphological representation 400 for the user.

In some embodiments as shown in Figure 13A, there is a first PPG sensor 300A and a second PPG sensor 300B, positioned at a proximal location and a distal location, on the user for measuring blood volume changes along an artery. The arrow A1 indicates the direction of blood flow along the artery and the arrow A2 indicates the hemodynamic forces acting on the arterial walls. The PPG signals 310 include a first PPG signal 310A measured by the first PPG sensor 300A from the proximal location and a second PPG signal 310B measured by the second PPG sensor 300B from the distal location. The first PPG sensor 300A and second PPG sensor 300B may be separated by a distance of at least 25 mm.

Various morphological features 314 of the pulse wavelets 312 can be determined from the PPG signals 310, some of which are listed below with reference to Figure 13B. 314a - Systolic peak 314b - Trough

314c - Dicrotic notch

314d - Diastolic peak

314e - Time from systolic peak to diastolic peak 314f and 314g - Areas to determine augmentation index

314h - Systolic time

314i - Diastolic time

314j - Pulse transit time

314k - Time from trough to diastolic peak

3141 - Anacrotic limb

314m - Time from dicrotic notch to diastolic peak

314n - Dicrotic limb

314o - Time from diastolic peak to trough

314p - Pulse width

The method 200 may include a step of calculating a pulse transit time between the first PPG signal 310A and second PPG signal 31 OB. As shown by the morphological feature 314j in Figure 13B, the pulse transit time may be defined as the time required for the arterial blood pulse to change between two arterial measurement sites within the same cardiac cycle of a single systole-diastole event. For example as shown in Figures 13B and 13C, the pulse transit time is defined as the duration between the systolic peaks of the first PPG signal 310A and second PPG signal 31 OB. Figure 13D shows the various pulse transit times between corresponding pulse wavelets 312A,312B from the first and second PPG signals 310A, 31 OB.

Multiple PPG sensors 300 may be placed on the user and the pulse transit times may be calculated for any pair of PPG sensors 300 that are spatially separated from each other. For example, with reference to Figure 3A, Figure 13E shows four PPG sensors 300 placed on the forearm - elbow PPG sensor 300d, proximal wrist PPG sensor 300e, distal wrist PPG sensor 300f, and fingertip PPG sensor 300g. Figure 13D also shows the pulse transit times for various pairs of the four PPG sensors 300d,300e,300f,300g.

With reference to the array of PPG sensors 300 shown in Figure 3A, the pulse transit times may be calculated for 6 pairs of the PPG sensors 300.

1 . Fingertip PPG sensor 300g and toe PPG sensor 300h

2. Forehead PPG sensor 300a and fingertip PPG sensor 300g 3. Forehead PPG sensor 300a and toe PPG sensor 300h

4. Forehead PPG sensor 300a and proximal wrist PPG sensor 300e

5. Proximal wrist PPG sensor 300e and fingertip PPG sensor 300g

6. Proximal wrist PPG sensor 300e and toe PPG sensor 300h

Experimental results showed that the pair of fingertip PPG sensor 300g and toe PPG sensor 300h had a pulse transit time of -50ms. This means that the pulse at the fingertip arrived 50ms before the pulse at the toe, which is logical since the fingertip is closer to the heart. The pair of forehead PPG sensor 300a and fingertip PPG sensor 300g had a pulse transit time of -50ms. This means that the pulse at the forehead arrived 50ms before the pulse at the fingertip. The pair of forehead PPG sensor 300a and toe PPG sensor 300h had a pulse transit time of -100ms. The pair of forehead PPG sensor 300a and proximal wrist PPG sensor 300e had a pulse transit time of - 30ms. The pair of proximal wrist PPG sensor 300e and fingertip PPG sensor 300g had a pulse transit time of -10ms, which is expected due to the short distance between the wrist and fingertip. The pair of proximal wrist PPG sensor 300e and toe PPG sensor 300h had a pulse transit time of -100ms.

The set of physiological parameters 410 is measurable based additionally on the pulse transit time. For example, the physiological parameters 410 may be measured based on the pulse transit time and the spatial distance between the first and second PPG sensors 300A,300B. For example, the pulse transit time may complement the 3D morphological representation 400 to measure the physiological parameters 410 for the user.

The computerized method 200 for monitoring physiology of a user using one or more PPG sensors 300 may be performed on a system having a processor and various steps of the computerized method 200 are performed in response to non-transitory instructions operative or executed by the processor. The non-transitory instructions are stored on a memory and may be referred to as computer-readable storage media and/or non-transitory computer-readable media. Non-transitory computer-readable media include all computer-readable media, with the sole exception being a transitory propagating signal per se. The system for performing the method 200 may include the PPG sensors 300, such as an array of PPG sensors 300 as shown in Figure 3A. The PPG sensors 300 are separated from and communicative with the processor, such as for communicating the PPG signals 310 to the processor for analysis. Alternatively, the PPG sensors 300, such as the first PPG sensor 300A and second PPG sensor 300B spatially separated from each other, may be integrated in a wearable device, such as a watch, for the user. The wearable device may include the processor configured for performing the method 200. Alternatively, the processor is remote from the wearable device and the wearable device is communicative with the processor. For example, a separate computer device or server includes the processor, and the wearable device communicates the PPG signals 310 to the computer device or server.

The system and method 200 provide for non-invasive continuous monitoring of the physiology, including haemodynamic and cardiometabolic conditions, of a user through the processing of PPG signals 310 measured by PPG sensors 300, without the use of an occlusive cuff. PPG signals 310 are parameters that can be easily measured, and PPG sensors 300 are relatively cheap and commonly used in pulse oximetry. The method 200 processes the pulse wavelets 312 of the PPG signals 300 to construct one or more 3D morphological representations 400 that are unique to the user. The entire time series of the PPG signals 300, which can be about 3 minutes for example, can thus be compressed into a 3D morphological representation 400 which can be easily visualized for monitoring the user physiology.

Specifically, the morphology of the 3D morphological representation 400 can be analysed to measure various physiological parameters 410 and diagnose various medical conditions. For example, the user’s blood pressure and blood glucose may be measured from the 3D morphological representation 400 without wearing a cuff or drawing blood. The 3D morphological representation 400 improves diagnostic capabilities and enables comprehensive assessment of various physiological abnormalities, including but not limited to blood pressure fluctuations (e.g. orthostatic hypotension in elderly and vasovagal syncope), blood glucose fluctuations, gastrointestinal perfusion, peripheral vascular disease in diabetes mellitus, and arterial elasticity to inform clinical drug efficacy and durability.

For example, Figure 14A shows the differences between the pulse wavelets 312 in relation to different blood glucose levels 420. For example, Figure 14B shows the different pulse wavelets 312 used to construct respective 3D morphological representations 400 at various timepoints to estimate the blood glucose levels 420 at these timepoints. Notably, the morphology of the pulse wavelets 312 and the 3D morphological representations 400 changes depending on the blood glucose levels of the individual. The blood glucose levels of an individual changes throughout the day due to multi-kinetic effects of blood on the vessel walls. Figure 14C shows the various pulse wavelets 312 and the corresponding blood glucose levels at timepoints 0, 30 minutes, 90 minutes, and 120 minutes.

Individual users can monitor their unique 3D morphological representations 400 over an extended period to manage their personal health. For example, the 3D morphological representation 400 can be used to monitor autonomic innervation of the heart that allows for diagnosis and long-term cardiometabolic diseases monitoring and enable improved therapeutic approaches of cardiometabolic diseases in individuals. For individuals with chronic diseases, the 3D morphological representations 400 provide periodic measurements of cardiometabolic readings, whereby changes over time can be used to monitor peripheral vascular disease in diabetics and orthostatic hypotension in the elderly.

The use of these 3D morphological representations 400 for diagnosis opens opportunities for the development of advanced remote monitoring technologies that circumvent the need for electrocardiographic input and mechanical occlusion. The 3D morphological representations 400 are compressed from lengthy time series data of the PPG signals 310, with minimal data loss and reduced storage requirements and hardware complexity. These 3D morphological representations 400 serve to improve the monitoring and treatment of medical conditions, such as hypertension, diabetes, and peripheral vascular disease. In some embodiments, two or more spatially-separated PPG sensors 300 are used to measure multiple PPG signals 310 which are then processed to construct one or more 3D morphological representations 400. For example, two temporally and morphologically differentiated PPG signals 310 can provide insights into blood perfusion and pulse dynamics that are not possible with conventional electrocardiograms or sphygmomanometers. Blood perfusion is the local fluid flow through the capillary network and is a critical determinant of organ health.

In representative or exemplary embodiments of the present disclosure, there is a computer-implemented or computerized method 600 for monitoring haemodynamic conditions of a user using PPG signals 310, as shown in Figures 15A and 15B. For example, the method 600 can be used to monitor haemodynamic conditions such as blood pressure, e.g. orthostatic hypotension.

The method 600 includes a step 610 of receiving a plurality of PPG signals 310 measured from the user, each PPG signal 310 measured from a different location on the user. Preferably, the PPG signals 310 include a first PPG signal 310A measured from a proximal location on the user and a second PPG signal 310B measured from a distal location on the user. The method 600 includes a step 620 of receiving motion data measured from the user using an inertial measurement unit (IMU) 320. The method 600 includes a step 630 of generating a first haemodynamic profile of the user from the PPG signals 310 and a first machine learning model 710. For example, the first haemodynamic profile may include blood pressure and/or blood glucose data. The method 600 includes a step 640 of generating a second haemodynamic profile of the user from the first haemodynamic profile, motion data, and a second machine learning model 720. For example, the second haemodynamic profile may include blood pressure and/or blood glucose data. Preferably, the second haemodynamic profile includes orthostatic blood pressure data, and more preferably orthostatic hypotension data.

The computerized method 600 can be performed on a system 700 having a processor and various steps of the computerized method 600 are performed in response to non- transitory instructions operative or executed by the processor. The non-transitory instructions are stored on a memory and may be referred to as computer-readable storage media and/or non-transitory computer-readable media. Non-transitory computer-readable media include all computer-readable media, with the sole exception being a transitory propagating signal per se.

The system 700 for performing the method 600 may include the PPG sensors 300 and IMU 320 which are separated from and communicative with the processor. Alternatively, the PPG sensors 300, such as the first PPG sensor 300A and second PPG sensor 300B spatially separated from each other, and the IMU 320 may be integrated in a measurement device 800, such as shown in Figure 16A. The measurement device 800 may include the processor configured for performing the method 600, specifically to generate the first and second haemodynamic profiles using the machine learning models 710,720. Alternatively, the processor is remote from the measurement device 800 and the measurement device 800 is communicative with the processor. For example, a separate computer device or server includes the processor, and the measurement device 800 communicates the PPG signals 310 and IMU motion data to the computer device or server.

The measurement device 800 may be in the form of a wearable patch that the user can wear discreetly under clothing, similar to Holter patch electrodes. For example, the measurement device 800 can be placed at the deltopectoral area of the user’s chest to measure PPG signals 310 from the axillary artery.

The IMU 320 is configured to measure motion data, such as posture orientation and speed. The IMU 320 may include a 3-axis accelerometer for measuring acceleration, a 3-axis gyroscope for measuring angular velocity, and a 3-axis magnetometer for measuring magnetic orientation. The IMU 320 can thus measure motion data in 9 degrees of freedom, and at frequencies such as up to 20 Hz. For example, the motion data can be used to determine the posture of the user, such as standing, sitting, or lying supine as shown in Figure 16B.

As described above, the method 800 uses two machine learning models 710,720 to process the PPG signals 310 from the PPG sensors 300 and motion data from the IMU 320 to generating the haemodynamic profiles. With reference to Figure 15B as an example, the first machine learning model 710 includes an artificial neural network (ANN) 712, and the second machine learning model 720 includes a long short-term memory (LSTM) network 722. The PPG signals 310 are input to the ANN 712 to generate the first haemodynamic profile, which may include a classification of the user’s blood pressure. For example, the first machine learning model 710 includes a softmax layer for classifying the blood pressure. The motion data and the blood pressure classification are then concatenated in the LSTM network 722. The LSTM network 722 includes a series of regression layers to generate the second haemodynamic profile, which may include a prediction of the user’s orthostatic blood pressure such as orthostatic hypotension.

The machine learning models 710,720 thus combine the ANN 712 and LSTM network 722 in a hybrid machine learning model that is trained to recognize the different body postures and consider the physiological data from the PPG signals 310 and thereby, adjust the prediction output accordingly. Further, the classification and regression tasks can be learned in parallel and the hybrid machine learning model is able to learn feature representations from the intermediate layers, allowing for optimization of its internal parameters according to gradient descent convergence on the minimum error of the prediction output. The hybrid machine learning model advantageously improves learning effectiveness, efficiency, and prediction accuracy.

In the method 600 and system 700, the PPG signals 310 and IMU motion data are combined to monitor haemodynamic conditions of a user, such as to predict orthostatic hypotension due to posture awareness from the motion data. The measurement device 800 is designed to continuously measure the PPG signals 310 and motion data without an occlusive cuff. The measurement device 800 is unobtrusive to regular lifestyles, leading to improved awareness and acceptance of the measurement device 800 by various users. For example, the user can wear the measurement device 800 on the chest to continuously measure the data and monitor his/her health, such as for diagnosis, management, and treatment of orthostatic hypotension. In some embodiments, the method 600 may include a step of calculating a pulse transit time between the first PPG signal 310A and second PPG signal 31 OB. Haemodynamic parameters may be calculated from the pulse transit time. For example, the first haemodynamic profile, such as including blood pressure data, may be calculated from the pulse transit time.

In some embodiments, the first machine learning model 710 includes a support vector machine (SVM) to generate the first haemodynamic profile. Specifically, a set of morphological features 314 of the PPG signals 310A, 31 OB are extracted and, together with the calculated pulse transit time, are fed to the SVM. The SVM then predicts the blood pressure (systolic and diastolic) based on the morphological features 31 and pulse transit time of the PPG signals 310A,31 OB.

An experiment was done to evaluate the SVM’s performance in predicting blood pressures. Blood pressure estimation was done for a group of 78 participants and compared with actual blood pressure measurements. The 78 participants ranged from 21 years old to 79 years old with an average age of 36.7 years. Their systolic and diastolic blood pressures were estimated using the SVM and compared to the actual blood pressures that were measured using a sphygmomanometer. Figure 17A shows the results of the estimated and actual systolic blood pressures, and Figure 17B shows the results of the estimated and actual diastolic blood pressures. Figure 17C shows the errors or deviations between the estimated and actual systolic blood pressures. It was found that the estimated blood pressures were mostly within 5 mmHg of the actual blood pressures, which is within the global standard for blood pressure estimation established by the Association for the Advancement of Medical Instrumentation (AAMI). The results show that the SVM can predict blood pressures with reasonable accuracy using PPG signals 310.

In some embodiments as shown in Figure 18, the PPG signals 310 from the two or more PPG sensors 310A,310B, etc and the motion data from the IMU 320 may be used together with the 3D morphological representations 400 constructed from the PPG signals 310 to monitor physiological parameters 410 of the user. In the foregoing detailed description, embodiments of the present disclosure in relation to methods and devices for monitoring user physiology using photoplethysmography signals are described with reference to the provided figures. The description of the various embodiments herein is not intended to call out or be limited only to specific or particular representations of the present disclosure, but merely to illustrate non-limiting examples of the present disclosure. The present disclosure serves to address at least one of the mentioned problems and issues associated with the prior art. Although only some embodiments of the present disclosure are disclosed herein, it will be apparent to a person having ordinary skill in the art in view of this disclosure that a variety of changes and/or modifications can be made to the disclosed embodiments without departing from the scope of the present disclosure. Therefore, the scope of the disclosure as well as the scope of the following claims is not limited to embodiments described herein.

Claims

1. A computerized method for monitoring physiology of a user using photoplethysmography (PPG) signals, the method comprising: receiving a set of PPG signals measured from the user, each PPG signal measured from a different location on the user; identifying a plurality of pulse wavelets from each PPG signal, each pulse wavelet defined in 2D by an amplitude axis and a first time axis; for each PPG signal, aligning the respective 2D pulse wavelets in 3D along a second time axis; and constructing a set of 3D morphological representations from the aligned pulse wavelets, wherein a set of physiological parameters of the user is measurable based on the constructed 3D morphological representations.

2. The method according to claim 1 , wherein constructing the set of 3D morphological representations comprises constructing, for each PPG signal, a respective 3D morphological representation from the respective pulse wavelets identified from the respective PPG signal, such that each 3D morphological representation corresponds to one PPG signal.

3. The method according to claim 1 , wherein constructing the set of 3D morphological representations comprises constructing a single 3D model from all the pulse wavelets identified from all the PPG signals.

4. The method according to any one of claims 1 to 3, wherein the PPG signals comprise a first PPG signal measured from a proximal location on the user and a second PPG signal measured from a distal location on the user.

5. The method according to claim 4, further comprising calculating a pulse transit time between the first and second PPG signals, wherein the physiological parameters are measurable based additionally on the pulse transit time.

6. The method according to any one of claims 1 to 5, wherein the physiological parameters comprise blood pressure, blood glucose, and/or arterial distensibility.

7. A system for monitoring physiology of a user using PPG signals, the system comprising: a processor configured for performing the computerized method according to any one of claims 1 to 6.

8. A non-transitory computer-readable storage medium storing computer- readable instructions that, when executed, cause a processor to perform the computerized method according to any one of claims 1 to 6.

9. A measurement device for monitoring haemodynamic conditions of a user using photoplethysmography (PPG) signals, the measurement device comprising: a plurality of PPG sensors for measuring a plurality of PPG signals from different locations on the user; an inertial measurement unit for measuring motion data from the user; and a processor configured for: calculating a first haemodynamic profile of the user from the PPG signals and a first machine learning model; and calculating data second haemodynamic profile of the user from the first haemodynamic profile, motion data, and a second machine learning model.

10. The measurement device according to claim 9, wherein the first haemodynamic profile comprises blood pressure and/or blood glucose data, and wherein the second haemodynamic profile comprises orthostatic blood pressure data.

11. The measurement device according to claim 9 or 10, wherein the plurality of PPG sensors comprises: a first PPG sensor for measuring a first PPG signal from a proximal location on the user; and a second PPG sensor for measuring a second PPG signal from a distal location on the user.

12. The measurement device according to claim 11 , wherein the computer processor is configured for calculating a pulse transit time between the first and second PPG signals.

13. The method according to claim 12, wherein the first machine learning model comprises a support vector machine for generating the first haemodynamic profile comprising blood pressure data from the PPG signals and the pulse transit time.

14. A computerized method for monitoring haemodynamic conditions of a user using photoplethysmography (PPG) signals, the method comprising: receiving a plurality of PPG signals measured from the user, each PPG signal measured from a different location on the user; receiving motion data measured from the user using an inertial measurement unit; generating a first haemodynamic profile of the user from the PPG signals and a first machine learning model; and generating a second haemodynamic profile of the user from the first haemodynamic profile, motion data, and a second machine learning model.

15. The method according to claim 14, wherein the plurality of PPG signals comprises a first PPG signal measured from a proximal location on the user and a second PPG signal measured from a distal location on the user.

16. The method according to claim 15, further comprising calculating a pulse transit time between the first and second PPG signals.

17. The method according to claim 16, wherein generating the first haemodynamic profile comprises, using a support vector machine of the first machine learning model, predicting blood pressure from the PPG signals and the pulse transit time.

18. The method according to any one of claims 14 to 17, wherein the first machine learning model comprises an artificial neural network and the second machine learning model comprises a long short-term memory (LSTM) network.

19. A system for monitoring haemodynamic conditions of a user using PPG signals, the system comprising: a processor configured for performing the computerized method according to any one of claims 1 to 18.

20. A non-transitory computer-readable storage medium storing computer- readable instructions that, when executed, cause a processor to perform the computerized method according to any one of claims 14 to 18.