Biometric monitoring system and method

Abstract

A method of determining blood pressure of the subject, the method comprising: receiving at least one ultrasonic waveform collected using at least one ultrasonic transducer, the at least one ultrasonic waveform being representative of an ultrasonic signal received by the at least one ultrasonic transducer from the subject; processing the at least one ultrasonic waveform to generate at least one ultrasound image representative of blood vessel walls; determining motion data from the at least one image, the motion data being representative of motion of the blood vessel walls; and providing the determined motion data as inputs to at least one model, the at least one model being configured to determine the blood pressure of the subject based on the inputs to the model.

Description

[0001] M&C PG451048GB

[0002] 1

[0003] BIOMETRIC MONITORING SYSTEM AND METHOD

[0004] Field

[0005] The present invention relates to a biometric monitoring system and method, particularly but not exclusively to the monitoring of blood pressure using ultrasound.

[0006] Background

[0007] Conventional blood pressure monitoring methods are typically invasive, provide only periodic measurements, and often require professional oversight, limiting their use to clinical settings. Existing non-invasive technologies do not offer the necessary accuracy and consistency for reliable continuous monitoring. There is a significant need for an advanced solution that can provide precise, continuous monitoring suitable for both clinical and everyday environments.

[0008] Furthermore, there are challenges in providing the required accuracy, particularly in devices having a suitable form factor for unobstrusively fitting into everyday wearable devices such as watches, rings, pads and the like.

[0009] Summary

[0010] Various aspects of the present invention are defined in the independent claims. Some preferred features are defined in the dependent claims.

[0011] According to a first example of the present disclosure is a method of determining at least one biometric parameter of a subject, such as a blood pressure of the subject, the method comprising: receiving at least one acoustic waveform collected using at least one transducer, the at least one acoustic waveform being representative of an acoustic signal received by the at least one transducer from the subject; processing the at least one acoustic waveform to generate at least one image representative of structure within the subject; determining motion data from the at least one image, the motion data being representative of motion of the structure within the subject; and providing the determined motion data as inputs to at least one model, the at least one model being configured to determine the at least one biometric parameter or other data indicative thereof based on the inputs to the model.

[0012] The at least one biometric parameter may comprise a parameter of blood flow in the subject. The at least one biometric parameter may comprise blood pressure of

[0013] 55387831-1 M&C PG451048GB

[0014] 2 the subject. The at least one biometric parameter may comprise at least one of: pulse rate, blood flow, systolic blood pressure, diastolic blood pressure, and / or saturation of peripheral oxygen (SpC>2), of the subject. The subject may be a living creature such as a human or other animal. The structure may be or comprise at least one blood vessel, such as an artery. The structure may be or comprise the walls of the at least one blood vessel, e.g. the arterial walls.

[0015] The at least one acoustic signal may be an ultrasound signal. The at least one acoustic waveform may be an ultrasound waveform. The at least one transducer may be an ultrasound transducer. The ultrasound transducer may be configured to produce and / or receive the ultrasound signal. The acoustic signal may be within a range of sound frequencies that are higher than the range that can be heard by humans, e.g. frequencies greater than 5MHz, e.g. greater than 20kHz and optionally, less than 500MHz, e.g. less than 200MHz. For example, the frequency or frequencies of the acoustic signal may be comprised or consist in a range from 20kHz up to 200MHz. For example, the ultrasonic signal may comprise or consist of one or more signals having a frequency in a range from 100kHz to 20MHz or in a range from 1 to 100MHz. Beneficially, the ultrasonic signals may comprise one or more signals with frequencies in a range from 5MHz to 200MHz. This particular frequency range has been found to enable high-resolution imaging suited for capturing detailed arterial motions.

[0016] The at least one transducer may be integrated into a wearable device. The wearable device may be, comprise or be configured as one of: a smartwatch, a smart ring, a patch, a necklace, an armband, a bracelet, or the like. The at least one transducer may be formed of materials and / or using manufacturing methods that enable continuous and real-time monitoring capabilities.

[0017] For example, the at least one transducer may be a flexible transducer, e.g. a flexible ultrasound transducer. The at least one transducer may be a thin film transducer, e.g. a thin-film ultrasound transducer. The thin film transducer may comprise a layer of piezoelectric material on a flexible substrate. The substrate may be electrically conductive, i.e. it may be an electrical conductor. The substrate may be planar. The substrate may be a film or sheet. The substrate may be metallic, e.g. a metal film. The substrate may be or comprise a metal or metallic foil such as aluminium foil. The substrate may be or comprise a thin film. The substrate may have a thickness in the range of 20 to 200pm. The substrate may be thicker than the layer of piezoelectric material, e.g. by at least a factor of 2 or by a factor of 6 or more.

[0018] The layer of piezoelectric material may be, comprise or be comprised in a film of piezoelectric material. The layer of piezoelectric material may be configured and / or operable to produce ultrasound, i.e. the layer of piezoelectric material may be or

[0019] 55387831-1 M&C PG451048GB

[0020] 3 comprise an ultrasound production layer. The piezoelectric material may be or comprise an inorganic material. The piezoelectric material may be a crystalline, e.g. polycrystalline, piezoelectric material. The piezoelectric material may be a non- polymeric piezoelectric material. The piezoelectric material may be or comprise a metal oxide, such as zinc oxide. The piezoelectric material may be or comprise a metal nitride, such as aluminium nitride. The piezoelectric material may be or comprise a continuous layer of material having piezoelectric properties, e.g. the piezoelectric material may not comprise discrete domains of piezoelectric material having piezoelectric properties within a matrix of non-piezoelectric material. The layer of piezoelectric material may have a thickness in the range of 2 to 50pm. The layer of piezoelectric material may be thinner than the substrate. However, in other examples, the piezoelectric material may be or comprise other forms and materials, such as single crystal piezoelectric materials. In examples, the ultrasound transducer is constructed from materials including, but not limited to, zinc oxide (ZnO), doped zinc oxide, Aluminium Nitride (AIN), or Lead Zirconate Titanate (PZT), or doped versions thereof. These materials may be selected to optimize the acoustic response and durability of the transducer.

[0021] The substrate may be, comprise, or be comprised in an electrical ground electrode. The substrate may be, comprise, or be comprised in a counter electrode to the electrode formed by the additive techniques. The counter electrode may form an electrode pair with the at least one working electrode (e.g. working electrodes of an electrode array), which may be provided on an opposing side of the piezoelectric material to the counter electrode. The substrate may be, comprise, or be comprised in a ground electrode. A surface of the substrate that is opposite to a surface of the substrate upon which the layer of piezoelectric material is disposed may be a radiating surface from which ultrasonic waves are radiated in use.

[0022] The one or more transducers may comprise a plurality of transducers and may comprise or be comprised in a transducer array, such as an ultrasonic transducer array. In some examples, the ultrasonic transducer array may comprise a patterned array of discrete domains of the piezoelectric material on a common substrate, e.g. the flexible substrate. In other examples, the ultrasonic transducer array may comprise a continuous layer of piezoelectric material on the substrate and discrete electrodes on at least one side of the layer of piezoelectric material in order to form the transducer array. Each transducer in the transducer array may be individually addressable, e.g. via its own electrode and / or conduction track.

[0023] 55387831-1 M&C PG451048GB

[0024] 4

[0025] The transducer or transducer array may comprise multiple transducer elements, which may number from 1 to 1024 elements. A transducer array may facilitate variable resolution and / or depth of imaging depending on the application requirements.

[0026] The transducer array may be configured such that its output can be dynamically adjusted to focus and / or steer the emitted acoustic signal, e.g. the ultrasound beam. The focussing and beam steering may comprise at least one of: selectively activating and de-activating transducers or transducer elements of the transducer array, configuring the signals emitted by different transducers or transducer elements of the transducer array to selectively constructively combine or to cancel each other out, or by other techniques that would be apparent. This arrangement may allow for adaptive imaging based on the specific anatomical features of the subject.

[0027] The at least one transducer may be fabricated using Micro-Electro-Mechanical Systems (MEMS) technology, such as Capacitive Micromachined Ultrasonic Transducers (CMUT) or Piezoelectric Micromachined Ultrasonic Transducers (pMUT). These types of transducers or transducer arrays may enhance sensitivity and miniaturization of the device.

[0028] The ultrasound transducer array may be configured in a linear format, e.g. with the transducers or transducer elements of the transducer array being arranged linearly, which may result in an elongate transducer array, that is longer than it is wide, e.g. 2, 3, 5 or more times longer than it is wide. This linear arrangement may facilitate the acquisition of elongated, narrow field of view images particularly suitable for tracking longitudinal arterial motion. In alternative examples, the transducer array may be configured in a matrix or 2D format, e.g. a plurality of the transducers or transducer elements of the transducer array are distributed in at least two orthogonal directions. This arrangement may enable broader and more comprehensive imaging coverage, beneficial for capturing complex anatomical details in a single scan. However, the present disclosure is not limited to these arrangements of transducer array and the transducer array may be configured in other specialized form factors, including but not limited to phased array or curved array. In examples, the specific form of the transducer array may be selected, e.g. from the above, to optimize imaging based on specific clinical or non-clinical use cases.

[0029] The at least one acoustic waveform may be collected at high frame rates which may be frame rates above 10 frames per second (fps), preferably 12 fps or above. The collection frame rates may be 500 fps or below, e.g. 100fps or below. Frame rates in this range may be high enough to accurately collect blood pressure measurements without requiring unduly high processing and / or data storage requirements.

[0030] 55387831-1 M&C PG451048GB

[0031] 5

[0032] The at least one acoustic waveform may be collected in a defined period of time, such as a set or pre-set period of time. The set or pre-set period of time may be at least 2 seconds and may be less than 30 seconds, e.g. between 6 and 20 seconds such as from 6 to 16 seconds, e.g. 10 seconds. The at least one acoustic waveform may comprise a plurality of waveforms distributed over time, e.g. over the defined period of time. At least one or each waveform of the plurality of waveforms may be for a later time than at least one or each other of the plurality of waveforms. The plurality of waveforms may collectively represent motion or evolution of the structure over time.

[0033] The method may comprise locating the at least one transducer on the subject. The method may comprise operating the at least one transducer to collect the at least one acoustic waveform. Operating the at least one transducer may comprise operating the at least one transducer to emit at least one acoustic signal, e.g. the at least one ultrasonic signal, into the subject. Operating the at least one transducer may comprise operating the at least one transducer to receive a reflection of the emitted at least one acoustic signal from the subject. The reflections of the emitted at least one acoustic signal may be reflections from interfaces between layers of the sample.

[0034] The transducer may be configured to capture dynamic motion within the subject. The dynamic motion within the subject may comprise motion of the at least one blood vessel, e.g. arterial pulses. The motion of the blood vessels may be pulsing of the at least one blood vessel, e.g. expansion and constriction of the at least one blood vessel. The motion of the blood vessels may comprise moving of walls of the at least one blood vessel, e.g. during expansion and constriction of the at least one blood vessel.

[0035] The processing the at least one acoustic waveform to generate at least one image may comprise generating one or more B-mode images. That is, the at least one image may be or comprise a B-mode image. The at least one image may be representative of the structure, e.g. representative of walls of the at least one blood vessel, such as the arterial walls of the at least one blood vessel. The processing the at least one acoustic waveform to generate at least one image may comprise generating a plurality of images, such as a plurality of B-mode images, which may optionally be generated from a plurality of acoustic waveforms. The plurality of images may be representative of the motion of the structure within the subject, e.g. the motion of the at least one blood vessel.

[0036] The determining of the motion data from the at least one image may comprise determining motion date from the plurality of images. The determining of the motion data from the at least one image may comprise tracking the structure over the plurality of images.

[0037] 55387831-1 M&C PG451048GB

[0038] 6

[0039] The determining of the motion data from the at least one image may comprise identifying the structure in the at least one image, e.g. identifying the blood vessel, such as the blood vessel walls. The identifying the structure in the at least one image may comprise identifying the structure using object recognition. The determining of the motion data from the at least one image may comprise determining motion of the structure, e.g. motion of the blood vessels such as motion of the blood vessel walls, which may be indicative of the expansion and constriction of the at least one blood vessel. The determining of the motion data from the at least one image may comprise applying object tracking, e.g. tracking of the blood vessel walls.

[0040] The determining of the motion data from the at least one image may comprise speckle recognition and / or tracking. Speckle is a known artefact in ultrasound images, sometimes regarded as noise, but may not be random, but may be correlated with motion of the structure. As such, changes in the speckle in the ultrasound images may be representative of motion of the structure, such as motion of the blood vessel, e.g. of the blood vessel walls, particularly of pulsing of the blood vessels. The determining of the motion data from the at least one image may comprise pixel tracking, e.g. tracking signature changes in pixels of the at least one ultrasound image, such as repetitive or cyclical patterns in motion of the pixels. Although various examples of determining motion data representative of motion of the structure within the subject are given above, the present disclosure is not limited to these and other ways to determine the motion data may be apparent from the present disclosure.

[0041] The motion data may be representative of cyclical motion of the structure. The motion data may be or comprise one or more waveforms. The one or more waveforms may be representative of the cyclical motion of the structure. The one or more waveforms may be representative of the pulsing, e.g. the expansion and constriction, of the blood vessels. The one or more waveforms may be representative of flow of blood in the at least one blood vessel. The one or more waveforms may be representative of one or more systolic peaks. The one or more waveforms may be representative of one or more valve openings. The one or more waveforms may be representative of one or more diastolic states. The one or more waveforms may be indicative of diameter of the at least one blood vessel, e.g. of a part of at least one blood vessel.

[0042] The at least one model may be a machine learning model. The machine learning model may be a trained model. The model may have been trained on labelled data sets, e.g. manually labelled data sets. The labelled data sets may comprise labelled motion data, e.g. labelled waveforms. The labelled data sets may be labelled with and / or may include at least one of: standard biometric parameter, e.g. blood pressure, measurements, personal data and / or metadata. The standard biometric

[0043] 55387831-1 M&C PG451048GB

[0044] 7 parameter, e.g. blood pressure, measurements may comprise “ground truth” or gold standard measurements, e.g. using clinical blood pressure measurement apparatus. The labelled data sets may correlate the labelled motion data with values of the biometric parameters, e.g. values of blood pressure. Although a trained machine learning model is described, other types of models could be used, e.g. generative adversarial networks or other generative or self-learning models. The machine learning model may comprise any suitable machine learning model, such as an artificial neural network (ANN), e.g. a recurrent neural network, convolutional neural network, feed-forward neural network or the like. The machine learning model may comprise an input layer, one or more convolutional layers and an output layer. The input layer may be configured to receive at least the motion data as inputs. The output layer may output at least the at least one biometric parameter. Any intervening layers, e.g. the one or more convolutional layers, may map the inputs to the outputs. The model may predict the values of the one of more biometric parameters, e.g. the blood pressure, for that subject based on at least the motion data. The model may employ algorithms capable of analyzing the motion data, e.g. the extracted waveform data, to determine detailed physiological characteristics, which may comprise a value of the at least one biometric parameter. The use of a model, such as a machine learning model, may avoid or mitigate the need to recalibrate in use.

[0045] The method may comprise extracting principal components of the motion data, e.g. of the waveforms representing the motion data. The principal components may comprise extracting systolic peak data, such as systolic peak height and / or position or the variation therein, extracting diacritic notch height or variation therein, extracting diastolic depth or position or variations therein, and / or the like. The method may comprise providing the principal components of the waveforms as inputs to the model instead of, or in addition to, the raw motion data. The method may comprise using an autoencoder, such as a trained autoencoder, or other trained or configured model or algorithm, configured for feature extraction of the principle components of the motion data. In this way, the principle components may be automatically extracted.

[0046] The method may comprise obtaining metadata. The metadata may comprise one or more metrics or other data of the subject. The metadata may comprise at least one or more or each of: height, weight, gender, age, BMI, resting heart rate, VO2max, SpO2or other metrics or data for that subject. The metadata may be provided as inputs to the model to determine the at least one biometric parameter of the subject, e.g. in addition to the motion data and / or the principal components thereof.

[0047] The method may comprise obtaining other measurements of the subject, which may be measurements collected using the same of different wearable devices to the

[0048] 55387831-1 M&C PG451048GB

[0049] 8 wearable device in which the one or more transducers is comprised. The other measurements of the subject may comprise at least one of: photoplethysmography (PPG) measurements, skin impedance or conductivity, electrocardiogram (ECG), electroencephalogram (EEG), accelerometer data, gyroscope data, other inertial data, temperature data, moisture level data, and / or the like. The other measurements of the subject may be provided as inputs to the model to determine the at least one biometric parameter of the subject, e.g. in addition to the motion data and / or the principal components thereof and / or the metadata.

[0050] The model may be a trained model trained using augmented data, which may be instead of or in addition to labelled training data. The labelled training data may comprise motion data, e.g. waveforms representing the motion date, labelled with labels associated with the respective motion data. The labels may comprise values of the at least one biometric parameter and / or metadata associated with the respective motion data. The augmented data may be used to enhance the size and / or diversity of the training data. The augmented data may comprise labelled motion data, e.g. labelled waveforms, that have been labelled with values for the at least one biometric parameter and / or metadata. In the augmented data, the values of the at least one label, e.g. the one or more biometric parameter and / or the metadata, may be dithered. That is, the augmented data may correspond to the training data described elsewhere, with dithering applied to at least some or all of the label values. For example, a dithering value, which may be a set, pre-set or randomised value, e.g. randomised within a dithering range, may be added to or subtracted from some or all of the values of the at least one biometric parameter and / or metadata used for labelling the augmented data. For example, the dithering value could be in a range from 0% to ±20%, e.g. from ±5% to ±15%. The dithering value could be selected at random from values within the range. In examples, the dithering value may be added or subtracted from a set, preset or random proportion of the values used to label the motion data used to form the augmented data. The present inventors have discovered that applying dithering to at least some of the values used to label the training data has a surprisingly significant improvement in the accuracy of the model outputs in predicting the values of the at least one biometric parameter. Examples of the values of the labels to which the dithering could be applied include blood pressure values, pulse rate values, and / or metadata representing personal data of the subject such as height, weight, age, BMI and / or the like.

[0051] In the augmented data, the variance and / or distribution of the labels for the motion data may be varied and / or adjusted, e.g. to create synthetic data points. The synthetic data points may cover the training data or augmented data associated with

[0052] 55387831-1 M&C PG451048GB

[0053] 9 label values that would be otherwise under-represented. The motion data used for the augmented data may be denoised, e.g. using an autoencoder, which may improve signal fidelity.

[0054] The method may comprise using a dry-couplant for the transducer, e.g. between the transducer and the subject. The dry-couplant may comprise a polymer, biogel, adhesive, or silicone solution or the like. The dry-couplant may be configured to enhance acoustic coupling without compromising user comfort or device flexibility.

[0055] The method may comprise using a secure communication system for transmitting the determined at least one biometric parameter, e.g. blood pressure, and / or any other determined physiological characteristic. The method may comprise employing security measures to ensure data integrity and protect against unauthorised access, e.g. by using encryption, key encoding and / or by storing the data on a blockchain, directed acyclic graph (DAG) or other immutable data structure.

[0056] According to a second example of the present disclosure is a system for determining at least one biometric parameter of a subject, the system comprising at least one processor or processing system configured to implement the method of the first example. The system may be implemented on a computer, which may be a portable computing device such as a smart phone, smart watch or the like. The processor or processing system may be configured to receive at least one acoustic waveform collected using at least one transducer, the at least one acoustic waveform being representative of at least one acoustic signal received by the at least one transducer from the subject; process the at least one acoustic waveform to generate at least one image representative of structure within the subject; determine motion data from the at least one image, the motion data being representative of motion of the structure within the subject; and provide the determined motion data as inputs to at least one model, the at least one model being configured to determine the at least one biometric parameter or other data indicative thereof based on the inputs to the model.

[0057] The system may comprise the at least one transducer. The at least one transducer may comprise or be comprised in a transducer array. The at least one transducer may be configured to output an electrical signal representative of the at least one acoustic waveform. The at least one processor or processing system may be in a unitary device with the at least one transducer, e.g. housed in a common housing. The at least one processor or processing system may be physically connected to the at least one transducer to receive the output therefrom, e.g. via electrically conductive connectors. The at least one processor or processing system may be configured to receive the at least one acoustic waveform wirelessly from the at least one transducer. The at least one processor or processing system may be separate to or remote from

[0058] 55387831-1 M&C PG451048GB

[0059] 10 the at least one transducer, and may be in data communication with the at least one wireless transducer, e.g. via a local wireless connection such as Bluetooth, ZigBee or the like, or via a wide area wireless connection, e.g. via the internet or a cloud service or the like.

[0060] The at least one acoustic signal may be an ultrasound signal. The at least one acoustic waveform may be an ultrasound waveform. The at least one transducer may be an ultrasound transducer. The ultrasound transducer may be configured to produce and / or receive the ultrasound signal. The acoustic signal may be within a range of sound frequencies that are higher than the range that can be heard by humans, e.g. frequencies greater than 5MHz, e.g. greater than 20kHz and optionally, less than 500MHz, e.g. less than 200MHz. For example, the frequency or frequencies of the acoustic signal may be comprised or consist in a range from 20kHz up to 200MHz. For example, the ultrasonic signal may comprise or consist of one or more signals having a frequency in a range from 100kHz to 20MHz or in a range from 1 to 100MHz. Beneficially, the ultrasonic signals may comprise one or more signals with frequencies in a range from 5MHz to 200MHz. This particular frequency range has been found to enable high-resolution imaging suited for capturing detailed arterial motions.

[0061] The at least one transducer may be integrated into a wearable device. The wearable device may be, comprise or be configured as one of: a smartwatch, a smartring, a patch, a necklace, an armband, a bracelet, or the like. The at least one transducer may be formed of materials and / or using manufacturing methods that enable continuous and real-time monitoring capabilities.

[0062] For example, the at least one transducer may be a flexible transducer, e.g. a flexible ultrasound transducer. The at least one transducer may be a thin film transducer, e.g. a thin-film ultrasound transducer. The thin film transducer may comprise a layer of piezoelectric material on a flexible substrate. The substrate may be electrically conductive, i.e. it may be an electrical conductor. The substrate may be planar. The substrate may be a film or sheet. The substrate may be metallic, e.g. a metal film. The substrate may be or comprise a metal or metallic foil such as aluminium foil. The substrate may be or comprise a thin film. The substrate may have a thickness in the range of 20 to 200pm. The substrate may be thicker than the layer of piezoelectric material, e.g. by at least a factor of 2 or by a factor of 6 or more.

[0063] The layer of piezoelectric material may be, comprise or be comprised in a film of piezoelectric material. The layer of piezoelectric material may be configured and / or operable to produce ultrasound, i.e. the layer of piezoelectric material may be or comprise an ultrasound production layer. The piezoelectric material may be or comprise an inorganic material. The piezoelectric material may be a crystalline, e.g.

[0064] 55387831-1 M&C PG451048GB

[0065] 11 polycrystalline, piezoelectric material. The piezoelectric material may be a non- polymeric piezoelectric material. The piezoelectric material may be or comprise a metal oxide, such as zinc oxide. The piezoelectric material may be or comprise a metal nitride, such as aluminium nitride. The piezoelectric material may be or comprise a continuous layer of material having piezoelectric properties, e.g. the piezoelectric material may not comprise discrete domains of piezoelectric material having piezoelectric properties within a matrix of non-piezoelectric material. The layer of piezoelectric material may have a thickness in the range of 2 to 50pm. The layer of piezoelectric material may be thinner than the substrate. However, in other examples, the piezoelectric material may be or comprise other forms and materials, such as single crystal piezoelectric materials. In examples, the ultrasound transducer is constructed from materials including, but not limited to, zinc oxide (ZnO), doped zinc oxide, Aluminium Nitride (AIN), or Lead Zirconate Titanate (PZT), or doped versions thereof. These materials may be selected to optimize the acoustic response and durability of the transducer.

[0066] The substrate may be, comprise, or be comprised in an electrical ground electrode. The substrate may be, comprise, or be comprised in a counter electrode to the electrode formed by the additive techniques. The counter electrode may form an electrode pair with the at least one working electrode (e.g. working electrodes of an electrode array), which may be provided on an opposing side of the piezoelectric material to the counter electrode. The substrate may be, comprise, or be comprised in a ground electrode. A surface of the substrate that is opposite to a surface of the substrate upon which the layer of piezoelectric material is disposed may be a radiating surface from which ultrasonic waves are radiated in use.

[0067] The one or more transducers may comprise a plurality of transducers and may comprise or be comprised in a transducer array, such as an ultrasonic transducer array. In some examples, the ultrasonic transducer array may comprise a patterned array of discrete domains of the piezoelectric material on a common substrate, e.g. the flexible substrate. In other examples, the ultrasonic transducer array may comprise a continuous layer of piezoelectric material on the substrate and discrete electrodes on at least one side of the layer of piezoelectric material in order to form the transducer array. Each transducer in the transducer array may be individually addressable, e.g. via its own electrode and / or conduction track.

[0068] The transducer or transducer array may comprise multiple transducer elements, which may number from 1 to 1024 elements. A transducer array may facilitate variable resolution and / or depth of imaging depending on the application requirements.

[0069] 55387831-1 M&C PG451048GB

[0070] 12

[0071] The transducer array may be configured such that its output can be dynamically adjusted to focus and / or steer the emitted acoustic signal, e.g. the ultrasound beam. The focussing and beam stearing may comprise at least one of: selectively activating and de-activating transducers or transducer elements of the transducer array, configuring the signals emitted by different transducers or transducer elements of the transducer array to selectively constructively combine or to cancel each other out, or by other techniques that would be apparent. This arrangement may allow for adaptive imaging based on the specific anatomical features of the subject.

[0072] The at least one transducer may be fabricated using Micro-Electro-Mechanical Systems (MEMS) technology, such as Capacitive Micromachined Ultrasonic Transducers (CMUT) or Piezoelectric Micromachined Ultrasonic Transducers (pMUT). These types of transducers or transducer arrays may enhance sensitivity and miniaturization of the device.

[0073] The ultrasound transducer array may be configured in a linear format, e.g. with the transducers or transducer elements of the transducer array being arranged linearly, which may result in an elongate transducer array, that is longer than it is wide, e.g. 2, 3, 5 or more times longer than it is wide. This linear arrangement may facilitate the acquisition of elongated, narrow field of view images particularly suitable for tracking longitudinal arterial motion. In alternative examples, the transducer array may be configured in a matrix or 2D format, e.g. a plurality of the transducers or transducer elements of the transducer array are distributed in at least two orthogonal directions. This arrangement may enable broader and more comprehensive imaging coverage, beneficial for capturing complex anatomical details in a single scan. However, the present disclosure is not limited to these arrangements of transducer array and the transducer array may be configured in other specialized form factors, including but not limited to phased array or curved array. In examples, the specific form of the transducer array may be selected, e.g. from the above, to optimize imaging based on specific clinical or non-clinical use cases.

[0074] The at least one processor or processing system may comprise one or more programmable processors, e.g. general purpose processors, executing a computer program to perform functions of the invention by operating on input data and generating output. The at least one processor or processing system may be or comprise special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit) or other customised circuitry. Processors suitable for the execution of a computer program include CPUs and microprocessors, and any one or more processors. Generally, a processor will receive instructions and data from a read-only memory or a random access memory or both.

[0075] 55387831-1 M&C PG451048GB

[0076] 13

[0077] The system may also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. Information carriers suitable for embodying computer program instructions and data include all forms of non-volatile memory, including by way of example semiconductor memory devices, e.g. EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in special purpose logic circuitry.

[0078] To provide for interaction with a user, the system may comprise a device having a screen, e.g., a CRT (cathode ray tube), plasma, LED (light emitting diode) or LCD (liquid crystal display) monitor, for displaying information to the user and an input device, e.g., a keyboard, touch screen, a mouse, a trackball, and the like by which the user can provide input to the computer. Other kinds of devices can be used, for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0079] The system may be implemented on a device, such as a mobile or network enabled device, comprising or configured to implement the controller or processing system of the first aspect. The device may be or comprise or be comprised in a mobile phone, smartphone, PDA, tablet computer, laptop computer, and / or the like. The controller or processing system may be implemented by a suitable program or application (app) running on the device. The device may comprise at least one processor, such as a central processing unit (CPU), maths co-processor (MCP), graphics processing unit (GPU), tensor processing unit (TPU) and / or the like. The at least one processor may be a single core or multicore processor. The device may comprise memory and / or other data storage, which may be implemented on DRAM (dynamic random access memory), SSD (solid state drive), HDD (hard disk drive) or other suitable magnetic, optical and / or electronic memory device. The at least one processor and / or the memory and / or data storage may be arranged locally, e.g. provided in a single device or in multiple devices in in communication at a single location or may be distributed over several local and / or remote devices. The device may comprise a communications module, e.g. a wireless and / or wired communications module. The communications module may be configured to communicate over a cellular communications network, Wi-Fi, Bluetooth, ZigBee, near field communications (NFC), IR, satellite communications, other internet enabling networks and / or the like. The communications module may be configured to communicate via Ethernet or other wired network or connections, via a telecommunications network such as a POTS,

[0080] 55387831-1 M&C PG451048GB

[0081] 14

[0082] PSTN, DSL, ADSL, optical carrier line, and / or ISDN link or network and / or the like, via the cloud and / or via the internet, or other suitable data carrying network. The communications module may be configured to communicate via optical communications such as optical wireless communications (OWC), optical free space communications or Li-Fi or via optical fibres and / or the like. The device and / or the controller or the at least one processor or processing unit may be configured to communicate with the remote server or data store via the communications module. The controller or processing unit may comprise or be implemented using the at least one processor, the memory and / or other data storage and / or the communications module of the device.

[0083] According to a third example of the present invention is a computer program product, such as but not limited to a tangible, non-transient computer program product, that is configured such that, when implemented by a processor or processing system, such as that described above in relation to the second example of the present disclosure, causes the processor or processing system to implement the method of the first example of the present disclosure.

[0084] According to a fourth example of the present disclosure is a method for data augmentation in training machine learning models for predicting at least one biometric parameter, e.g. blood pressure. The method may be a method for augmenting or otherwise forming training data for use in training machine learning models for predicting at least one biometric parameter, e.g. to form an augmented dataset. The model may be trained using the augmented dataset, which may be instead of or in addition to the labelled training data. The augmented dataset can be used to enhance the size and / or diversity of the training data. The method may comprise receiving at least one of: motion data extracted from one or more images of structure in a subject; acoustic waveforms representative of at least one acoustic signal received by the at least one transducer from the subject; and / or at least one image representative of structure within the subject. The method may comprise receiving numerical labels, e.g. in a dataset. The method may comprise dithering the numerical labels, e.g. in the dataset. The dithering may be applied to at least some or all of the values of the numerical labels. The labels may comprise blood pressure values or measurements or values of other biometric parameters. The labels may comprise metadata relevant to the subject, such as personal metrics of a subject or group of subjects, such as but not limited to at least one or each of: age, height, gender and / or weight. The method may comprise denoising of acquired data (e.g. the motion data, the numerical labels, and / or the personal metrics), e.g. via usage of convolutional autoencoders, for example, to improve signal fidelity. The numerical labels and / or the personal data may

[0085] 55387831-1 M&C PG451048GB

[0086] 15 be associated with (or the method may comprise associating the numerical labels and / or personal data with) one or more of: particular motion data or acoustic waveforms and / or at least one image. The method may comprise adjusting a variance and / or distribution of said labels to create synthetic data points. The method may comprise training a machine learning model on the augmented dataset, e.g. to improve its predictive performance under varied conditions.

[0087] The dithering of the numerical labels may comprise employing dithering techniques on the labels. For example, a dithering value, which may be a set, pre-set or randomised value, e.g. randomised within a dithering range, may be added to or subtracted from some or all of the values of the at least one biometric parameter and / or metadata used for labelling the augmented data. For example, the dithering value could be in a range from 0% to ±20%, e.g. from ±5% to ±15%. The dithering value could be selected at random from values within the range. In examples, the dithering value may be added or subtracted from a set, preset or random proportion of the values used to label the motion data used to form the augmented data. The method may comprise augmenting the data, e.g. the motion data, with gold standard blood pressure measurements obtained using different blood pressure measurement apparatus and / or personal data. The method may comprise using the augmented dataset to enhance the diversity and size of the training data for the machine learning model. This may improving the accuracy and robustness of the model, enhancing its ability to generalise across different populations and conditions.

[0088] The model, e.g. the machine learning model, may be the model described in relation to the first example and / or the second example of the present disclosure.

[0089] The method may be carried out on a processing apparatus, such as a computing system or device, such as a portable computing device or by a cloud based server in communication with the computing device.

[0090] The augmented dataset may be used to train the model of the first example and / or the second example.

[0091] According to a sixth example of the present disclosure is a method of training a machine learning model for predicting at least one biometric parameter, the method comprising receiving labelled motion data that comprises motion data labelled with a value of at least one biometric parameter and / or metadata represented by or otherwise associated with the respective motion data; dithering at least some of the values associated with at least some of the at least one biometric parameter and / or metadata to form augmented training data; and training the machine learning model using the augmented training data.

[0092] 55387831-1 M&C PG451048GB

[0093] 16

[0094] The at least one biometric parameter may comprise blood pressure. The motion data may comprise arterial motion data, e.g. motion data indicative of motion of arterial walls. The motion data may be derived from ultrasound images of a subject, such as B-mode images. The trained model may be the model used in the method of the first example or by the system of the second example.

[0095] According to a sixth example of the present disclosure is a framework for automated feature extraction for use in training of at least one model being configured to determine at least one biometric parameter or other data indicative thereof based at least on motion data indicative of motion of structure within a body, the motion data being derived from ultrasound images. The model may be the model described in relation to the first example and / or the second example. The framework may comprise reducing the motion data, such as waveforms representative of motion of the structure, into principal components indicative of at least one biometric parameter of the subject such as blood pressure. The principal components may comprise at least one of: dicrotic notch height, height variation, position or the like; systolic peak height, position or variation, and / or the like; diastolic position, height, variation and / or the like. The framework may comprise feature extraction, e.g. via usage of a trained autoencoder to facilitate automatic feature extraction. The framework may comprise supplementing the trained machine learning model with additional metadata in combination with features extracted.

[0096] The at least one biometric parameter may comprise blood pressure. The motion data may be derived from ultrasound images derived from one or more ultrasound waveforms received by at least one ultrasonic transducer positioned on the subject. The one or more ultrasound waveforms received by at least one ultrasonic transducer may be ultrasound waveforms received over a period of time and representing the structure at different times.

[0097] According to a seventh example of the present disclosure is a computer program product configured such that, when implemented by a processor or computer, causes the computer to perform any of the methods described above.

[0098] The individual features and / or combinations of features defined above in accordance with any example of the present disclosure or below in relation to any specific embodiment of the invention may be utilised, either separately and individually, alone or in combination with any other defined feature, in any other aspect or embodiment of the invention.

[0099] Furthermore, the present invention is intended to cover apparatus configured to perform any feature described herein in relation to a method and / or a method of using or producing, using or manufacturing any apparatus feature described herein.

[0100] 55387831-1 M&C PG451048GB

[0101] 17

[0102] Brief Description of the Drawings

[0103] For a better understanding of the present disclosure and to show how embodiments may be put into effect, reference is made to the accompanying drawings in which:

[0104] Figure 1 shows a plan view of an example of an ultrasonic transducer array;

[0105] Figure 2 shows a schematic of an ultrasonic device comprising an ultrasonic transducer array;

[0106] Figure 3 shows a system for determining a biometric parameter of a subject, the system comprising the ultrasonic device of Figure 2;

[0107] Figure 4 shows an alternative ultrasonic transducer array;

[0108] Figure 5 shows a side view of the ultrasonic transducer array of Figure 3;

[0109] Figure 6 shows an implementation of the system of Figure 4 in a smartwatch;

[0110] Figure 7 shows an implementation of the system of Figure 4 in a skin patch;

[0111] Figure 8 shows an implementation of the system of Figure 4 in a wrist band;

[0112] Figure 9 shows an implementation of the system of Figure 4 in a smart ring;

[0113] Figure 10 shows an alternative implementation of the smart ring of Figure 9;

[0114] Figure 11 is a flowchart of a method of determining at least one biometric parameter of a subject;

[0115] Figure 12 is a flowchart showing a more detailed implementation of the method shown in Figure 11 ;

[0116] Figure 13 shows motion data of a blood vessel represented as a waveform of blood vessel diameter with time;

[0117] Figure 14 shows extended motion data of a blood vessel represented as a waveform of blood vessel diameter with time; and

[0118] Figure 15 shows even further extended motion data of a blood vessel represented as a waveform of blood vessel diameter with time.

[0119] Detailed Description of the Drawings

[0120] Figures 1 to 3 show an exemplary embodiment of a system 5 for determining at least one biometric parameter of a subject (not shown). In the examples described herein, the subject is an animal, particularly a human and the at least one biometric parameter is a parameter of blood flow, such as blood pressure, pulse rate and / or or the like. The system is configured to determine the at least one biometric parameter from ultrasound signals emitted into, and received from, the subject, which are used to image a structure of the subject, such as a blood vessel in the subject. In particular, the system is configured to image a cross section through one or more blood vessels of

[0121] 55387831-1 M&C PG451048GB

[0122] 18 the subject. However, the present disclosure is not limited to these implementations and other ultrasonic transducer arrangements and materials could be used. In many examples, the system comprises or is implemented in, or using, a wearable device that comprises or is configured to communicate with computing resources in order to determine the values of the at least one biometric parameter of the subject.

[0123] The system 5 comprises an ultrasonic device 1000 that comprises an ultrasonic transducer array 100. Figure 1 shows a schematic planar view of an example ultrasound transducer 100 according to the present disclosure and Figure 2 shows an ultrasonic device 1000 comprising the ultrasonic transducer 100 and including a cross sectional view of the ultrasound transducer 100 along line A-A’ attached to a subject 60 to be monitored. The ultrasound transducer 100 is provided only as an example of an ultrasound transducer which may be comprised in the system 5.

[0124] The ultrasound transducer 100 comprises an electrically conductive substrate 10 in the form of a metal foil, in this case an aluminium foil, and a layer of crystalline piezoelectric material 20 disposed on one planar surface of the substrate 10. The substrate 10 acts to support the layer of piezoelectric material 20 and also functions as a ground electrode. In this example, the piezoelectric material 20 is vanadium doped ZnO but it will be appreciated that other suitable piezoelectric materials such as AIN and / or other dopants, particularly other transition metal dopants, could be used.

[0125] A central transmit electrode 40 at common point X, is connected to an electrically conductive track 48 and in turn to the control system 50. A plurality of receive electrodes 30 are collocated on the same surface layer of the piezoelectric material 20 as the transmit electrode 40. These are arranged into 4 receive electrode arrays 35, each receive electrode array 35 comprising six elongate receive electrodes 30 distributed over a radial direction perpendicular to a longitudinal axis of the receive electrodes 30. Each of the receive electrode arrays 35 are circumferentially arranged in regular intervals around the common point (marked X in Figure 1) on the ultrasound transducer 100. Alternative example transmit electrode arrangements could be used.

[0126] The transmit electrode 40 and the receive electrodes 30 are provided on a surface of the layer of piezoelectric material 20 that is on an opposite side of the layer of piezoelectric material 20 to the substrate 10. Each of the receive electrodes 30 is connected to a corresponding electrically conductive track 38, that is in turn electrically connected to a control system 50. The receive electrodes 30 in this example are thin metallic electrodes and could be provided by deposition or coating of a metallic layer onto the piezoelectric material 20, for example.

[0127] Alternative example electrode connection arrangements are shown in Figure 3 and described in more detail below.

[0128] 55387831-1 M&C PG451048GB

[0129] 19

[0130] The ultrasonic transducer 100 can beneficially be a flexible thin film transducer based on a non-polymeric, polycrystalline piezoelectric material, such as ZnO or AIN, deposited onto a conductive, flexible substrate, such as a metallic foil. Examples of suitable ultrasonic transducers 100 are described in WO 2019 / 166805, WO 2019 / 166815 and PCT / GB2020 / 050468 in the name of the present applicant, the contents of which are incorporated by reference as if set out in full herein. This type of transducer has been found to be of particular benefit in wearable applications, as they can operate at very high frequencies, for good resolution and accuracy, are low cost to manufacture, which is particularly important in consumer electronics, and can be easily conformed into appropriate shapes for incorporating into wearable devices such as watches, rings, patches and arm or leg bands. Furthermore, it is easy to form transducer arrays, which can be configured for scanning, depth variation, focussing and other operations.

[0131] The transmit electrode 40 is a thin metallic electrode, and could also be formed by deposition or coating of a thin metal layer onto the surface of the piezoelectric material 20. The transmit electrode 40 is electrically connected to a conductive track 48 that is in turn electrically connected to the control system 50.

[0132] The control system 50 comprises a digitizer in the form of an ADC 52 that is operable to convert the analogue electrical signal produced by the ultrasound transducers 100 into a digital signal that can be processed by an on-board processing system 55. The analogue electrical signal is representative of received ultrasound signals received by the receive electrodes. The control system 50 further comprises a DAC 54 that is operable to covert digital signals produced by the on-board processing system 55 into an analogue electrical signal applied to the transmit electrode 40.

[0133] In order to generate the ultrasonic signals 32, the control system 50 applies an alternating electrical driving current to the transmit electrode 40 via electrically conductive track 48. The deposited layer of piezoelectric material 20 generates ultrasonic signals 32 responsive to, and representative of, the electrical signal generated by the control system 50 e.g. dependent on the frequency, amplitude and / or duration of the electrical signal. The ultrasonic signals 32 are emitted from the ultrasound transducer 100 by periodic displacement of the piezoelectric material 20 as longitudinal ultrasonic waves i.e. sound waves propagating in a direction parallel to a plane in which the direction of displacement of the piezoelectric material 20 lies and perpendicular to the planar surface of the substrate 10. The ultrasonic signals 32 are emitted in a direction away from the common point X into the subject 60.

[0134] Any reflections 42 of the ultrasonic signals 32 are received by the piezoelectric material 20. The piezoelectric material 20 generates electrical signals responsive to,

[0135] 55387831-1 M&C PG451048GB

[0136] 20 and representative of, the received ultrasonic signal 42. The electrical signals are transmitted by the receive electrodes 30 along the electrically conductive tracks 38 to the control system 50.

[0137] The on-board processing system 55 receives digitized versions of the electrical signal from the ADC 52 and processes the digitized electrical signal to determine properties of the subject 60 to which it is attached. It will be appreciated that the electrical signal produced by the ultrasound transducer 100 is representative of properties of the received ultrasonic signal 42, such as amplitude and frequency over time. The on-board processing system is configured to derive parameters from the signal, such as time of flight I time between transmission of the emitted ultrasonic signals 32 from the ultrasound transducer 100 and / or from other interconnected ultrasound transducers 100 attached to the entity or object 60 and reception of the receive ultrasonic signals 42, Doppler shift, and / or the like. The on-board processing system is configured to produce an image of the subject 60 and determine at least one biometric property of the subject, particularly blood flow properties such as blood pressure and pulse rate, or changes in biometric properties of the subject 60 from the received ultrasonic signal 42.

[0138] The values of the at least one biometric property of the subject 60 determined by the on-board processing system 55 can be temporarily or persistently stored in onboard data storage 56 that is comprised in the ultrasonic device 1000. Additionally or alternatively, the values of the properties of the entity or object determined by the onboard processing system 55 can be transmitted, preferably wirelessly transmitted, directly or indirectly by a communications system 58 comprised in the ultrasonic device 1000 to an external processing system (such as the external processing system 1050 of Figure 3) for further storage, processing, distribution, display and / or for raising an alarm if the values of the biometric properties of the subject are indicative of an alarmworthy condition. In this way, one or more ultrasonic devices 1000 can be mounted on the subject 60 to perform ultrasound measurements over time on the subject 60, determine the corresponding values of the one or more biometric properties of the subject 60 on board the ultrasonic device 1000 and then transmit the values of the one or more properties of the subject 60 to an external processing system.

[0139] Figure 3 illustrates an illustrative example of the system 5 as a biometric parameter monitoring system 2000 comprising a single ultrasonic device 1000. The ultrasonic device 1000 comprises four ultrasound transducers 100 fixed to a subject 60 (see Figure 2) to be monitored, and a control system 50 to control the ultrasound transducers 100, digitise the ultrasonic signal(s) received by the ultrasound transducers 100, and process, store and / or communicate the digitised ultrasound signals or data

[0140] 55387831-1 M&C PG451048GB

[0141] 21 produced therefrom. In examples, the control system 50 can be configured to digitise the received ultrasonic signal(s) separately, multiplexed or simultaneously. The control system may be the control system 50 of the ultrasonic device 1000 illustrated in Figure 2. It would be understood that the monitoring system 2000 may comprise a plurality of ultrasonic devices 1000 and each ultrasonic device 1000 may be fixed or mounted to the subject 60.

[0142] The monitoring system 2000 further comprises an external processing system 1050 configured to process the data communicated from the control system 50. The data may be representative of the digitised version (e.g. digitised ultrasonic waveform) of the received ultrasonic signal(s) and / or may comprise ultrasonic images derived from the reflected ultrasonic signals received from the subject 60. The monitoring system 2000 further comprises external data storage 1060 configured to store the data communicated from the control system 50, output data from the processing system 1050 or the like. The monitoring system 2000 optionally further comprises I / O devices 1070 which may be used to display a status of the monitoring process, raise an alert or alarm based on the monitoring process and / or allow input from a user to control the monitoring process.

[0143] The external processing system 1050 can receive the data and / or the values of the properties of the entity determined by the control system 50 or each ultrasound transducer 100. In this way, the external processing system 1050 can add additional benefits to the processing performed on the ultrasonic device 1000. For example, the external processing system 1050 can perform further processing on the data and / or the values of the properties of the entity or entities, e.g. to determine data, trends and parameters. In some examples, the external processing system 1050 is configured to raise an alert or alarm in response to the determined values of the biometric properties of the subject meeting an alarm condition. This may allow remote workers or monitoring services and the like to be made aware of determined conditions that may require action. In examples, the alert may comprise a message, an email, a pop-up notification, operation of a light or other visual indicator, an audible indicator, a haptic indicator and / or the like in, on or using one of the I / O devices 1070. The external processing system may raise the alert by automatically electronically signalling a user device over a network or internet when the determined data meets the alert or alarm conditions. The data and / or the values of the biometric properties of the subject being monitored is optionally stored on external storage 1060 so that the subject can view the changes in their biometric properties over time, or with events such as exercise and / or the like.

[0144] 55387831-1 M&C PG451048GB

[0145] 22

[0146] Figure 4 shows an alternative ultrasonic device 400 that could be used as part of the ultrasonic device 1000 of Figure 2 or the system 512000 of Figure 3.

[0147] In contrast to the 3D array of transducers shown in Figure 1, the ultrasonic device 400 of Figure 4 comprises a transducer array 400 that is linear, comprising a plurality of ultrasonic transducers 410 arranged side by side and distributed over a single direction. Each ultrasonic transducer 410 acts for both transmission of ultrasonic signals into the subject 60 and also receives the reflections of the emitted ultrasonic signals from the subject 60. Each ultrasonic transducer 410 is addressable via its own dedicated conductive track 415.

[0148] Figure 5 shows a cross section through the ultrasonic device 400 of Figure 4. This shows a flexible metallic foil substrate 510. A thin layer 515 of polycrystalline piezoelectric material such as ZnO is deposited directly onto the substrate 510. A layer 520 of electrically insulating dielectric material such as Sll-8 is applied over most of the piezoelectric layer 515, except for openings 525 in the insulating layer 520 through which working electrodes 530 corresponding to each transducer 410 extend, wherein the working electrodes 530 are connected to an electrically conducting metallic track 535 which allows electrical signals representative of the emitted and received ultrasonic signals to be applied to, and read from, the working electrodes 530. The ultrasonic signals 540 from each transducer 410 are emitted from an emitting portion of the substrate 510 corresponding to the respective working electrode 530 under the action of excitation of the corresponding portion of piezoelectric layer 515 by electrical waveforms provided by the working electrode 530 and conductive track 535.

[0149] Beneficially, the ultrasonic system of determining biometric properties of a subject 60 can be incorporated into a wearable device worn by the subject 60. The wearable device could incorporate any of the ultrasound transducer 100 of Figure 1 , the ultrasonic device 1000 of Figure 2, the system 5 I 2000 of Figure 3 and / or the transducer array 400 of Figures 4 and 5, although any suitable ultrasonic device or transducer array could be used, such as a PZT, or single crystal based ultrasonic transducer or transducer array.

[0150] Figure 6 shows an example of an ultrasonic device incorporated into a smart watch, where the ultrasonic device is configured to emit ultrasonic waves into the subject 60 and receive reflections of the ultrasonic waves from the subject 60. Some processing of the received ultrasonic waves can be carried out on the smartwatch and / or processing or additional processing can be distributed to a connected computing device such as a smart phone or other mobile computing device and / or a cloud based computing resource.

[0151] 55387831-1 M&C PG451048GB

[0152] 23

[0153] Figure 7 shows an example of an ultrasonic device incorporated into a skin patch, where the ultrasonic device is configured to emit ultrasonic waves into the subject 60 and receive reflections of the ultrasonic waves from the subject 60. Processing of the received ultrasonic waves can be carried out by a connected computing device such as a smart phone or other mobile computing device and / or a cloud based computing resource.

[0154] Figure 8 shows an ultrasonic device incorporated into a wrist band, where the ultrasonic device is configured to emit ultrasonic waves into the subject 60 and receive reflections of the ultrasonic waves from the subject 60. Processing of the received ultrasonic waves can be carried out by a connected computing device such as a smart phone or other mobile computing device and / or a cloud based computing resource.

[0155] Figures 9 and 10 show an ultrasonic device incorporated into a smart ring, where the ultrasonic device is configured to emit ultrasonic waves into the subject 60 and receive reflections of the ultrasonic waves from the subject 60. Processing of the received ultrasonic waves can be carried out by a connected computing device such as a smart phone or other mobile computing device and / or a cloud based computing resource. Smart rings could be worn on one or any finger or multiple smart rings could be provided on different fingers.

[0156] Figure 11 is a flowchart 1100 showing a method of determining at least one biometric parameter of a subject, such as a blood pressure of the subject. The method could be implemented by any suitable processing resource, such as the on-board processing system 55 of Figure 2, external processing system 1050 of Figure 3 or by cloud based remote processing resource or processing could be distributed among any combination of these. The method comprises, at 1105, receiving at least one acoustic waveform collected using at least one transducer, the at least one acoustic waveform being representative of at least one acoustic signal received by the at least one transducer from the subject. At 1110, the at least one acoustic waveform is processed to generate at least one image representative of structure within the subject. At 1115, motion data is determined from the at least one image, the motion data being representative of motion of the structure within the subject. At 1120, the determined motion data is provided as inputs to at least one model, the at least one model being configured to determine the at least one biometric parameter or other data indicative thereof based on the inputs to the model at 1125.

[0157] A spreadsheet providing a more detailed overview of the method of Figure 11 is shown in Figure 12. At 1205, fast frame rate ultrasound signals from the subject 60 is captured. Fast frame rate in this example, comprises frame rates of greater than 20frames per second, e.g. 24 frames per second or more. This provides suitable

[0158] 55387831-1 M&C PG451048GB

[0159] 24 resolution for this application. The ultrasound signals can be captures by the device 100 of Figure 2, or the system 5 12000 of Figure 3 or using the ultrasound transducer arrays of Figures 1, 4 or 5, or using any other suitable ultrasound transducer device. The ultrasound signals are provided by the ultrasonic transducers as electrical waveforms representative of the received ultrasonic signals. The ultrasonic transducers are configured to capture motion of, e.g. a section through, at least one structure such as a blood vessel of the subject 60. Beneficially, the ultrasonic transducer can be comprised in a wearable device such as any of those of Figures 6 to 10. The wearable device is worn by the subject.

[0160] The electrical waveforms representative of the ultrasonic signals are pre- processed at 1210, e.g. by filtering unwanted frequencies and / or by de-noising, which can be carried out by any suitable method.

[0161] The filtered waveforms from 1210 are converted to ultrasound images and used to track the walls of the blood vessels, e.g. the arterial walls, at 1215. The ultrasound images can be processed using suitably configured object recognition and tracking algorithms or models, for example, that are configured to identify and / or track the arterial walls in the ultrasound images.

[0162] Another way of applying the arterial wall tracking is to apply speckle recognition and tracking. Speckle is a known artefact in ultrasound images, sometimes regarded as noise. However, speckle isn’t random but is correlated with motion of the blood vessels. As such, changes in the speckle in the ultrasound images is representative of motion of the arterial walls, particularly that due to pulsing of the arteries. Thus my monitoring the changes in speckle, particularly cyclical changes, motion of the arterial walls can be tracked.

[0163] Another way of applying arterial wall tracking is by pixel tracking, e.g. tracking signature changes in pixels of the at least one ultrasound image, such as repetitive or cyclical patterns in motion of the pixels. Although various examples of determining motion data representative of motion of the arterial walls within the subject are given above, the present disclosure is not limited to these and other ways to determine the motion of the arterial walls form the ultrasound images may be apparent from the present disclosure.

[0164] At 1220, waveforms representing motion of the arterial walls are determined from the arterial wall tracking performed in 1215. Examples of such waveforms are shown in Figures 13 to 15. Figure 13 shows a waveform representing variation of arterial diameter (on the y-axis) with time (on the x-axis. The waveform in Figure 13 is over a single pulse, the waveform in Figure 14 is over three pulses and the waveform in Figure 15 is over five pulses, but any suitable number of pulses over time could be

[0165] 55387831-1 M&C PG451048GB

[0166] 25 covered in a waveform. The waveforms each comprise principal components including a systolic peak 1305, a dicrotic notch 1310 and a diastolic trough 131 for each pulse.

[0167] At 1225, the waveforms are analysed to extract principle features of the waveforms, such as systolic peak data, e.g. systolic peak height and / or position or the variation therein, diacritic notch height or variation therein, diastolic depth or position or variations therein, and / or the like. This feature extraction could be done automatically via use of a trained autoencoder or the like.

[0168] At 1230, the data, including the principle features form 1225, is normalised to ensure comparability and reproducibility.

[0169] At 1235, the normalised data, e.g. based on the motion waveforms from 1220 and / or the principle features form 1225 are fed as inputs into a model, such as a machine learning model, that has been trained or otherwise configured to map the inputs (e.g. the motion waveforms of the arterial walls and / or the extracted principle features) to outputs that include the one or more biometric parameters such as blood pressure. Additional inputs to the model optionally include metadata indicating subject data specific to that subject, such as weight, height, gender, waking or sleeping time, BMI, age, activity, or the like.

[0170] The model in certain examples is trained using suitable meta data 1240 and / or associated training data 1245 in the form of labelled motion waveforms or principle component values or other data representative of the inputs. The training data is optionally augmented by additional data. The additional data can be created, for example, by dithering numerical labels in a dataset, said labels including blood pressure measurements and personal metrics such as age, height, and weight. The variance and distribution of said labels can be adjusted to create synthetic data points. The machine learning model can then be trained on the augmented dataset to improve its predictive performance under varied conditions. This process could optionally also comprised denoising of acquired data via usage of convolutional autoencoders to improve signal fidelity.

[0171] The output prediction of the machine learning model is output at 1250 and provided as a biometric parameter output at 1255, .g. a blood pressure and / or pulse rate value. This can be provided as stored data or displayed on a screen of a smartwatch or user device such as a smartphone or personal computer. In some examples, this data can be used to raise an alarm, e.g. to carers or medical professionals if the blood pressure or pulse rate drops below a threshold or meets some other alarm criteria.

[0172] Some specific examples of the disclosure are provided as clauses of the description below.

[0173] 55387831-1 M&C PG451048GB

[0174] 26

[0175] In some examples, there is provided a method for predicting blood pressure in a subject using an ultrasound system that captures high frame rate acoustic waveforms, processes these waveforms to create ultrasound images, and applies machine learning techniques to analyze the motion data extracted from these images to predict blood pressure levels accurately.

[0176] Clause 1 : A method for predicting blood pressure in a subject, the method comprising steps of: a. acquiring high frame rate acoustic waveforms from a subject using an ultrasound transducer, wherein the transducer is adapted to capture the dynamic motion within the body reflective of arterial pulses; b. processing these acoustic waveforms to generate ultrasound images, specifically B-mode images, which visually represent the motion of structures within the body such as arterial walls; c. applying tracking methods to these ultrasound images to extract motion data of the observed structures, hereafter referred to as "waveforms"; d. analyzing these waveforms using a machine learning model, which has been trained on labeled datasets including standard blood pressure measurements and personal data; e. predicting the subject’s blood pressure using the machine learning model, which employs algorithms capable of analyzing the extracted waveform data to determine detailed physiological characteristics.

[0177] Clause 2: A framework for automated feature extraction used to further enhance the training of the blood pressure model. The framework consists of: A. reducing waveforms into principal components indicative of blood pressure such as dicrotic notch height variation; B. usage of a trained autoencoder to facilitate automatic feature extraction; C. supplementing the trained machine learning model with additional metadata in combination with features extracted.

[0178] Clause 3: A method for data augmentation in training machine learning models for blood pressure prediction, the method comprising: a. dithering numerical labels in a dataset, said labels including blood pressure measurements and personal metrics such as age, height, and weight; b. adjusting the variance and distribution of said labels to create synthetic data points; c. training a machine learning model on the augmented dataset to improve its predictive performance under varied conditions; d. Denoising of acquired data via usage of convolutional autoencoders to improve signal fidelity.

[0179] Clause 4: The method according to any preceding clause, further comprising: a. employing dithering techniques on the model labels, including gold standard blood pressure measurements and personal data, to augment the dataset; b. using the augmented dataset to enhance the diversity and size of the training data for the

[0180] 55387831-1 M&C PG451048GB

[0181] 27 machine learning model; c. thereby improving the accuracy and robustness of the model, enhancing its ability to generalise across different populations and conditions.

[0182] Clause 5: The method of any preceding clause, further comprising the use of a dry-couplant for the ultrasound transducer when conventional gels are unsuitable, wherein the dry-couplant includes a polymer, biogel, adhesive, or silicone solution to enhance acoustic coupling without compromising user comfort or device flexibility.

[0183] Clause 6: The method of clause 1 , wherein the ultrasound transducer is integrated into a wearable device configured as one of a smartwatch, smartring, patch, necklace, armband, or bracelet, employing materials and manufacturing methods that enable continuous and real-time monitoring capabilities.

[0184] Clause 7: The method of clause 1 or 5, further including a secure communication system for transmitting the predicted blood pressure data, employing security measures to ensure data integrity and protect against unauthorised access.

[0185] Clause 8: The method of clause 1, wherein the ultrasound transducer operates within a frequency range of 5 MHz to 200 MHz, enabling high-resolution imaging suited for capturing detailed arterial motions.

[0186] Clause 9: The method of any preceding clause, wherein the ultrasound transducer is constructed from materials including, but not limited to, zinc oxide (ZnO), doped zinc oxide, Aluminium Nitride (AIN), or Lead Zirconate Titanate (PZT), which are selected to optimize the acoustic response and durability of the transducer.

[0187] Clause 10: The method of any preceding clause, wherein the ultrasound transducer comprises multiple elements ranging from 1 to 1024 elements, facilitating variable resolution and depth of imaging depending on the application requirements.

[0188] Clause 11 : The method of any preceding clause, wherein the ultrasound transducer is fabricated using Micro-Electro-Mechanical Systems (MEMS) technology, including Capacitive Micromachined Ultrasonic Transducers (CMUT) or Piezoelectric Micromachined Ultrasonic Transducers (pMUT), to enhance sensitivity and miniaturization of the device.

[0189] Clause 12: The method of any preceding clause, wherein the ultrasound transducer includes an array configuration that can be dynamically adjusted to focus and steer the ultrasound beam, allowing for adaptive imaging based on the specific anatomical features of the subject.

[0190] Clause 13: The method of any preceding clause, wherein the ultrasound transducer array is configured in a linear format, facilitating the acquisition of elongated, narrow field of view images suitable for tracking longitudinal arterial motion.

[0191] Clause 14: The method of any preceding clause, wherein the ultrasound transducer array is configured in a matrix or 2D format, enabling broader and more

[0192] 55387831-1 M&C PG451048GB

[0193] 28 comprehensive imaging coverage, beneficial for capturing complex anatomical details in a single scan.

[0194] Clause 15: The method of any preceding clause, wherein the ultrasound transducer array may be configured in other specialized form factors, including but not limited to phased array or curved array, to optimize imaging based on specific clinical or non-clinical use cases.

[0195] Described herein is a system and method for non-invasive, end-user calibration-free, continuous blood pressure monitoring utilising high frame rate ultrasound imaging and sophisticated machine learning algorithms to predict blood pressure accurately. This invention provides significant advancements in medical technology, enabling reliable blood pressure monitoring in both clinical and non-clinical settings.

[0196] 55387831-1

Claims

M&C PG451048GB29CLAIMS:

1. A method of determining at least one biometric parameter of a subject, such as a blood pressure of the subject, the method comprising: receiving at least one acoustic waveform collected using at least one transducer, the at least one acoustic waveform being representative of an acoustic signal received by the at least one transducer from the subject; processing the at least one acoustic waveform to generate at least one image representative of structure within the subject; determining motion data from the at least one image, the motion data being representative of motion of the structure within the subject; and providing the determined motion data as inputs to at least one model, the at least one model being configured to determine the at least one biometric parameter or other data indicative thereof based on the inputs to the model.

2. The method of claim 1, wherein the motion data comprises waveforms of motion and the method comprises comprising extracting principal components of the motion data waveforms.

3. The method of claim 2, wherein the principle components comprise at least one of: systolic peak height, position or the variation therein, diacritic notch height or variation therein, diastolic depth or position or variations therein.

4. The method of any of claims 2 to 3, comprising automatically extracting the principle features using an autoencoder or other trained or configured model or algorithm.

5. The method of any preceding claim, comprising providing metadata representing the subject as additional inputs to the model.

6. The method of any preceding claim, wherein the model has been trained using training data and dithering has been applied to values of at least some numerical labels associated with at least some of the training data.55387831-1M&C PG451048GB307. The method of any preceding claim, wherein the acoustic waveforms are collected whilst a dry-couplant is provided between the transducer and the subject.

8. The method of any preceding claim, wherein the at least one transducer is integrated into a wearable device.

9. The method of any preceding claim, comprising using a secure communication system for transmitting the determined at least one biometric parameter.

10. The method of any preceding claim, wherein the at least one transducer operates within a frequency range from 5 MHz to 200 MHz.

11. The method of any preceding claim, wherein the ultrasound transducer is constructed from materials comprising one of: zinc oxide (ZnO), doped zinc oxide, Aluminium Nitride (AIN), or Lead Zirconate Titanate (PZT).

12. The method of any preceding claim, wherein the one or more transducers comprise a plurality of transducers comprised in a transducer array, and each transducer is individually addressable.

13. The method of claim 12, wherein the transducer array is configured such that its output can be dynamically adjusted to focus and / or steer the emitted acoustic signal.

14. The method of any preceding claim, wherein the at least one transducer has been fabricated using Micro-Electro-Mechanical Systems (MEMS) technology, such as Capacitive Micromachined Ultrasonic Transducers (CMUT) or Piezoelectric Micromachined Ultrasonic Transducers (pMUT).

15. The method of any preceding claim, wherein one of: the ultrasound transducer array is configured in a linear format with the transducers of the transducer array being arranged linearly so as to have an elongated, narrow field; or the transducer array is configured in a matrix or 2D format; or the transducer array is configured in a phased array or curved array.55387831-1M&C PG451048GB3116. A system comprising a processor and a memory configured to perform the method of any preceding claim.

17. A computer program product configured such that, when implemented by a processor or processing system, causes the processor or processing system to implement the method of any of claims 1 to 15.

18. A framework for automated feature extraction for use in training of at least one model being configured to determine at least one biometric parameter based at least on motion data indicative of motion of structure within a body, the motion data being derived from one or more ultrasound images, wherein the framework comprises reducing the motion data into principal components indicative of at least one biometric parameter of the subject; performing feature extraction using a trained autoencoder to facilitate automatic feature extraction; and supplementing the trained machine learning model with additional metadata in combination with features extracted.

19. A method for data augmentation in training machine learning models for predicting at least one biometric parameter, the method comprising receiving at least one of: motion data extracted from one or more images of structure in a subject; acoustic waveforms representative of at least one acoustic signal received by the at least one transducer from the subject; and / or at least one image representative of structure within the subject; receiving numerical labels in a dataset, the labels comprising personal metrics of a subject or group of subjects; dithering the numerical labels; and denoising acquired data.

20. A method of training a machine learning model for predicting at least one biometric parameter, the method comprising receiving labelled motion data that comprises motion data labelled with a value of at least one biometric parameter and / or metadata represented by or otherwise associated with the respective motion data; dithering at least some of the values associated with at least some of the at least one biometric parameter and / or metadata to form augmented training data; and training the machine learning model using the augmented training data.55387831-1

Images