Devices, systems and methods for blood glucose monitoring

Abstract

Devices, systems, and methods for blood glucose monitoring are disclosed. The apparatus comprises: a light emitter configured to emit a light signal; a light receiver configured to receive a reflected light signal; a controller configured to be operatively connected with the light emitter and the light receiver; and a housing. The light signals comprise a first light signal having a first wavelength of about 940 nm, a second light signal having a second wavelength of about 1350 nm, and / or a third light signal having a third wavelength of about 1500 nm, and wherein the controller comprises an operating module and further comprises or is operatively connected to a data processing system, the data processing system includes a machine learning module that analyzes the data signal to generate output data. The devices, systems, and methods are non-invasive and monitor blood glucose levels in real time with high accuracy.

Description

Cross - Reference to Related Applications

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 388,434, filed on Jul. 12, 2022, under 35 U.S.C. § 119(e), the entire content of which is hereby incorporated by reference herein. Technical Field

[0002] This application relates to devices, systems, and methods for detecting and monitoring glucose levels. This application relates to devices, systems, and methods for detecting and monitoring a user's blood glucose level using non - invasive techniques. Background Art

[0003] Diabetes is a disease that affects more than 34.2 million people worldwide. To effectively manage diabetes symptoms, diabetic patients can monitor their blood glucose levels to monitor the corresponding insulin levels in the body.

[0004] Typically, measuring a user's blood glucose level requires the user to repeatedly draw blood from, for example, a finger in order to apply the blood to a test strip. Alternatively, the user may be required to wear a continuous glucose monitoring device, which includes a sensor implanted in the interstitial fluid space of the user's skin. Both of these methods are invasive to the user.

[0005] There is a great need to establish a reliable, needle - free device, system, and method for continuously detecting and monitoring blood glucose levels in real - time without the need for such invasive procedures. Summary of the Invention

[0006] Devices, systems, and methods for blood glucose monitoring, data processing systems, computer - implemented methods, methods for processing data from a device or its system, and methods for training a machine - learning system for data obtained from a device or its system are provided.

[0007] In some embodiments, a device for monitoring a user's blood glucose is provided. The device includes: (a) a light emitter configured to emit a light signal directed at a target surface of the user to generate a reflected light signal reflected from the target surface; (b) a light receiver configured to receive the reflected light signal; (c) a controller configured to be operatively connected to the light emitter and the light receiver; and (d) a housing configured to house the light emitter, the light receiver, and the controller. The light signal includes a first light signal having a first wavelength of about 940 nm, a second light signal having a second wavelength of about 1350, and / or a third light signal having a third wavelength of about 1500 nm. The controller includes an operation module that controls the operation of the device and converts the reflected light signal into digital data. The controller further includes or is operatively connected to a data processing system that processes the digital data, wherein the data processing system includes a machine learning module that analyzes the data signal to generate output data as the user's blood glucose level. In some embodiments, the light signal includes one or more wavelengths selected from the range of 400 nm - 2000 nm. In some embodiments, the first wavelength is 400 nm - 950 nm, the second wavelength is 1000 nm - 1450 nm, and the third wavelength is 1500 - 2000 nm.

[0008] In some embodiments, a device for blood glucose monitoring of a user is provided. The device includes: a) a light emitter having a plurality of near-infrared LEDs, each configured to emit an optical signal directed to a target surface of the user so as to generate a reflected optical signal reflected from the target surface, wherein the optical signal includes a first optical signal having a wavelength of about 950 nm, a second optical signal having a wavelength of about 1350 nm, and a third optical signal having a wavelength of about 1500 nm; b) a light receiver having a photodetector configured to receive the reflected optical signal; c) a controller configured to be operatively connected to the light emitter and the light receiver; and d) a housing configured to house the light emitter, the light receiver, and the controller. The controller is configured to control the light emitter to turn on and off to sequentially emit the first optical signal, the second optical signal, and the third optical signal one by one, such that a plurality of reflected light groups are formed at defined time intervals, each reflected light group including a first reflected light, a second reflected light, and a third reflected light obtained in each cycle. The controller includes a processor unit coupled to a memory storing an executable software program, the software program including an operation module that controls the operation of the device and converts each reflected light group into a digital data vector. The controller further includes a data processing system for processing the data vector or is operatively connected to the data processing system, wherein the data processing system includes a machine learning module that analyzes the data vector to generate output data as the blood glucose level of the user. In some other embodiments, if a parallel processing controller is used, the sequential signal sampling process is reconfigured to parallel sampling.

[0009] In some embodiments, a system for blood glucose monitoring of a user is provided. The system includes: (a) a device as described in any one of the preceding claims; and (b) a server electrically connected to the device.

[0010] In some embodiments, a glucose monitoring system is provided, comprising: a wearable device capable of measuring a user's blood glucose level, the device including: a housing, an optical sensor, and a processing unit, the optical sensor including a circuit comprising one or more near-infrared light-emitting diodes and a receiver chip and configured to generate a voltage signal, the processing unit being configured to convert the analog voltage signal into a digital voltage signal; and computer software including an algorithm that generates a trained neural network model capable of predicting the user's blood glucose level in real time based on the voltage signal received from the processing unit, wherein the system is configured to measure the user's blood glucose level in a non-invasive manner in real time, and wherein the trained neural network model includes a trained non-linear model and a linear model to perform the following steps: generating a class prediction probability value and a numerical value by inputting the voltage signal into the trained non-linear model and the trained linear model respectively; classifying the class prediction probability value and the numerical value as low, normal, or high; comparing the classification results to determine whether the values are consistent; and if the values are consistent, determining an output blood glucose state and a blood glucose value.

[0011] In some embodiments, a method for monitoring blood glucose level is provided, the method comprising the steps of: (i) obtaining a first reflected light, a second reflected light, and a third reflected light from the device or system as described in any one of the foregoing embodiments; and (ii) calculating the blood glucose level based on the first reflected light, the second reflected light, and the third reflected light.

[0012] In some embodiments, a method for processing data from a device or system for blood glucose monitoring is provided, the method comprising the steps of: a) obtaining a data vector obtained from the device or system as described in any one of the foregoing embodiments; b) preprocessing the data vector; and c) analyzing the data vector through a trained neural network model to generate output data, thereby obtaining the user's blood glucose level.

[0013] In one embodiment, a glucose monitoring system includes a wearable device capable of measuring a user's blood glucose level. In this embodiment, the device includes a housing, an optical sensor having a circuit with one or more near-infrared light-emitting diodes and a receiver chip, wherein the sensor is configured to generate an analog voltage signal. The system may further include a microcontroller unit (MCU) that includes a built-in analog-to-digital converter (ADC) to convert the analog voltage signal into a digital voltage signal and hosts computer software containing an algorithm that generates a trained neural network model capable of predicting the user's blood glucose level in real time based on the digital voltage signal. In one embodiment, the system is configured to measure the user's blood glucose level in a non-invasive manner in real time.

[0014] This disclosure has many advantages. In certain embodiments, the provided devices, systems, and methods are non-invasive and address the technical obstacles associated with the low glucose measurement detection limit and selectivity of existing non-invasive optical glucose monitoring (NIOGM) techniques by detecting single or multiple near-infrared (NIR) signals. For example, a single NIR signal has a wavelength of approximately 940 - 950 nm. For example, multiple NIR signals have a specific combination of three wavelengths of approximately 940 nm, approximately 1350 nm, and approximately 1500 nm. Through multi-wavelength detection, the NIR signals of multiple wavelengths can provide differential vectors for inferring the blood glucose absorption factor.

[0015] In certain embodiments, the provided devices, systems, and methods are based on non-invasive optical glucose monitoring (NIOGM) techniques, where machine learning modules (such as deep meta-learning frameworks) are used for single or multiple NIR detections to address the technical problem of weak spectral signals originating from glucose molecules, thereby improving the signal-to-noise ratio (SNR) of the instrument.

[0016] In certain embodiments, the provided devices, systems, and methods provide an excellent selective measurement signal over background noise relative to other components of the skin, such as membranes, glycosylated structures, and soluble compounds (such as albumin, urea, amino acids, and ascorbic acid) within the ISF matrix, thereby providing a robust basis for measurement accuracy. For example, the subtle differences between different skin components can be obtained and derived by the following calculation based on the detection levels of the reflected light signals at 940 nm, 1350 nm, and / or 1500 nm. The total absorption factor of the light signal traveling through the blood is calculated as: μ(λ) = ε1(λ)c1 + ε2(λ)c2…… + ε n (λ)c n , where λ is the wavelength of the light, ε is the attenuation coefficient of the tissue and blood components, and c is the substance concentration.

[0017] In some embodiments, the provided devices, systems, and methods can reduce the interference or influence of one or more of the following parameters on measurement results: skin pigmentation, surface roughness, skin thickness, respiratory artifacts, blood flow, body movement, and environmental temperature.

[0018] In some embodiments, the provided devices, systems, and methods are based on primary (or direct) glucose sensing and secondary (or indirect) glucose sensing. Primary measurements involve collecting signals directly from glucose molecules, while secondary measurements involve measuring one or more parameters affected by glucose concentration, such as: the change in heart rate obtained by electrocardiogram; the rate of red blood cell aggregation obtained by ultrasound; the blood volume dynamics obtained by photoplethysmography measurement of blood; the dielectric properties of the skin matrix obtained by diffuse scattering or temperature-modulated local reflectance; and / or diastolic dysfunction obtained by electrochemically skin conductivity and sweating asymmetry.

[0019] In some embodiments, the provided devices, systems, and methods reduce the impact of skin pigmentation on the accuracy of clinical pulse oximetry measurements, and various substances that may affect skin pigmentation, skin structure, and the reflection properties of the detected radiation include topical medications, cosmetics, sweat, cosmeceuticals, and estrogen, as well as tobacco and alcohol.

[0020] In some embodiments, the provided devices, systems, and methods use multiple factors other than glucose (such as skin type, sweat, cholesterol, and other blood components) as composite factors. Then, by collecting the intensity of NIR light signals at multiple (e.g., three or more) different wavelengths received and training a deep meta-learning model on data covering key user types (such as those in the composite factors), the influence of these factors can be learned, and thus the blood glucose level can be accurately determined.

[0021] In some embodiments, the provided devices, systems, and methods provide a deep meta-learning architecture that involves separately using multiple different machine learning / deep learning models (scholar learners) to learn on the same task of glucose level prediction, and then using a meta-learning model (meta-learner) to learn how to aggregate their prediction results and obtain the final prediction result. In summary, the provided devices, systems, and methods provide a meta-learner that optimizes and aggregates n basic machine learning / deep learning models. Therefore, the provided devices, systems, and methods provide a deep meta-learning framework that can improve the efficiency and effectiveness of individual machine learning / deep learning algorithms and the accuracy of data even when the training data is limited and / or even when there is significant variability in the data.

[0022] In some embodiments, the provided devices, systems, and methods provide a three-wavelength NIR sensor module (including a transmitter, a light receiver, a refractor, and an optical path) to capture NIR reflected light signals (e.g., 940 nm, 1350 nm, and 1500 nm), which signals contain absorption information of blood glucose and other components (such as skin / vessels / blood flow). Without being bound by any theory, the combination response at 1350 nm and 1500 nm is sensitive enough to capture the slightest differences in spectral response between blood glucose and other blood components (such as cholesterol). However, without being bound by any theory, the combination of the data streams at 940 nm and 1500 nm will characterize the differences between blood glucose and the remaining blood components.

[0023] In some embodiments, the provided devices, systems, and methods provide a deep meta-learning (also known as'meta deep learning') framework for accurate blood glucose level determination based on data streams at multiple wavelengths (such as 940 nm / 1350 nm / 1500 nm). The deep learning framework can be provided online from a backend (e.g., a server) or as a plug-in component of an embedded system of the device. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Further embodiments of the present application can be understood with reference to the drawings.

[0025] Figure 1A is a schematic diagram of an example of a blood glucose monitoring device system.

[0026] Figure 1B is a schematic diagram of a blood glucose monitoring system in use.

[0027] Figure 1C is a schematic diagram of the circuit of a blood glucose monitoring device system.

[0028] Figure 1D is a depiction of the path of light traveling through the user's skin.

[0029] Figure 1E shows a block diagram depicting the process of analyzing and storing blood glucose data within the MCU of the system.

[0030] Figure 1F is a block diagram depicting the process of continuously monitoring blood glucose levels using a trained neural network model.

[0031] Figure 2A is an illustration of an example niCGM device 10 placed on a user's skin.

[0032] Figure 2B shows an example console app 1600 for controlling the niCGM device 10 and receiving blood glucose measurement results.

[0033] Figures 3A to 3EIt is the result of comparing the accuracy of the test data of Volunteer #11 in the Sannuo, Abbott, and the exemplary non-invasive continuous glucose monitoring (niCGM) device.

[0034] Figures 4A to 4E It is the result of comparing the accuracy of the test data of Volunteer #12 in the Sannuo, Abbott, and the exemplary niCGM device.

[0035] Figures 5A to 5E It is the result of comparing the accuracy of the test data of Volunteer #13 in the Sannuo, Abbott, and the exemplary niCGM device.

[0036] Figures 6A to 6E It is the result of comparing the accuracy of the test data of Volunteer #14 in the Sannuo, Abbott, and the exemplary niCGM device.

[0037] Figures 7A to 7E It is the result of comparing the accuracy of the test data of Volunteer #15 in the Sannuo, Abbott, and the exemplary niCGM device.

[0038] Figures 8A to 8B It is the result of comparing the test data between Abbott and the exemplary niCGM device for 5 volunteers (Volunteer #11 - 15).

[0039] Figure 9A It is a schematic diagram of an exemplary system 2000 including an exemplary niCGM device 2100.

[0040] Figure 9B It is a schematic diagram of a printed circuit board assembly PCBA 2200.

[0041] Figure 9C It is a flowchart showing the steps of an exemplary operation when executing the embedded software 2300.

[0042] Figure 9D It is a flowchart of the DMLF module 2400 of a data processing system.

[0043] Figure 9E It depicts the basic model MLP neural network architecture 2600.

[0044] Figure 9F It shows the single-layer MLP of the meta-model.

[0045] Figure 9G It shows the two-layer MLP of the meta-model.

[0046] Figure 10A It is a schematic diagram of another exemplary blood glucose monitoring system 3000, which includes another exemplary niCGM device 3100 and a server 3400.

[0047] Figure 10B is a flowchart showing steps of an example operation when executing the embedded software 3300.

[0048] Figure 11A is a schematic diagram of another example blood glucose monitoring system 4000, which includes another example niCGM device 4100, a mobile device 4600 including a console APP 4610, and a remote server 4400 including a DMLF module 4410.

[0049] Figure 11B is a flowchart showing steps of an example operation when executing the embedded software 4300.

[0050] Figure 11C is a flowchart showing steps of an example overall operation of the console APP 4610. DETAILED DESCRIPTION DEFINITIONS

[0051] Before explaining in detail at least one aspect of the disclosed and / or claimed inventive concept(s), it is to be understood that the disclosed and / or claimed inventive concept(s) are not limited in their application to the construction details and arrangements of components or steps or methods set forth in the following description or shown in the drawings. The disclosed and / or claimed inventive concept(s) can have other aspects or can be practiced or carried out in various ways. Further, it is to be understood that the terminology and terms used herein are for the purpose of description and should not be regarded as limiting.

[0052] As used in accordance with this disclosure, unless otherwise indicated, the following terms shall be understood to have the following meanings.

[0053] Unless otherwise defined herein, technical terms used in connection with the disclosed and / or claimed inventive concept(s) shall have the meanings commonly understood by one of ordinary skill in the art. Further, unless the context otherwise requires, singular terms shall include the plural and plural terms shall include the singular.

[0054] The singular forms "a", "an", and "the" include plural forms unless the context clearly dictates otherwise or the context in which they are mentioned clearly implies the contrary. When ranges are mentioned in the specification, the ranges are understood to include each discrete point within the recited range. For example, 1 to 7 means 1, 2, 3, 4, 5, 6, and 7. The terms "comprising" and "consisting of" include more restrictive claims such as "consisting essentially of" and "consisting of".

[0055] For the purposes of the following detailed description, except in the case of any operating examples or where otherwise indicated, numbers expressing quantities of ingredients, as used in the specification and claims, are to be understood as being modified in all instances by the term "about". The numerical parameters set forth in this specification and the appended claims are approximations that may vary depending upon the desired properties sought to be obtained while carrying out the present invention.

[0056] Unless otherwise indicated, all percentages, parts, ratios, and proportions used herein are by weight of the total composition. All such weights, when referring to the listed ingredients, are based on the active level, and thus; do not include solvents or by-products that may be included in commercially available materials, unless otherwise indicated.

[0057] All publications, articles, papers, patents, patent publications, and other references cited herein are hereby incorporated by reference in their entirety for all purposes to the extent consistent with the disclosure herein.

[0058] The use of the term "at least one" will be understood to include one and any quantity more than one, including but not limited to 1, 2, 3, 4, 5, 10, 15, 20, 30, 40, 50, 100, etc. The term "at least one" can extend to 100 or 1000 or more, depending on the term to which it is attached. Additionally, amounts of 100 / 1000 should not be considered limiting, as lower or higher limits may also produce satisfactory results.

[0059] As used herein, the words "comprising" (and any form of comprising, such as "comprise / comprises"), "having" (and any form of having, such as "have / has"), "including" (and any form of including, such as "includes / include"), or "containing" (and any form of containing, such as "contains / contain") are inclusive or open-ended and do not exclude additional unrecited elements or method steps.

[0060] It should be understood that for each embodiment in which the term “comprising” (or any related form, such as “comprise / comprises”), “including” (or any related form, such as “include / includes”), or “containing” (or any related form, such as “contain / contains”) is used, this disclosure / application also includes alternative embodiments in which the term “comprising”, “including”, or “containing” is replaced with “consisting essentially of” or “consisting of”. These alternative embodiments using “consisting of” or “consisting essentially of” are understood to be embodiments of a narrower scope than the “comprising”, “including”, or “containing” embodiments.

[0061] For clarity, “comprising”, “including”, “containing”, and “having” and any related forms are open - ended terms that allow additional elements or features in addition to the specified essential elements, while “consisting of” is a closed - ended term that is limited to the elements listed in the claim and does not include any elements, steps, or components not specified in the claim.

[0062] “Consisting essentially of” limits the scope of the claim to specific materials, components, or steps (“essential elements”) that do not materially affect the (one or more) essential features of the claimed invention. In some embodiments, the essential features are the (one or more) essential and novel features of the claimed invention.

[0063] For clarity, “characterized by” or “characterized in that” (along with their related forms as described above) does not limit or change whether the list of terms following them is open - ended or closed - ended. For example, in a claim relating to “an apparatus comprising A, B, C and characterized by D, E, and F”, the elements D, E, and F are still open - ended terms, and due to the use of the word “comprising” at the beginning of the claim, the claim is intended to include other elements.

[0064] The term “each independently selected from the group consisting of” means that when a group appears more than once in a structure, the group can be independently selected each time it appears.

[0065] The term “artificial neural network” refers to a computational architecture with programming instructions that can learn from a training dataset to make one or more predictions, such as predictions of the properties of new test objects.

[0066] The term "computer system" refers to an electronic device that includes a memory configured to store encoded instructions, a processor that executes the instructions, an output interface, etc., and is capable of performing the various claimed steps of the present invention.

[0067] In non-limiting embodiments, the present application discloses a wearable or otherwise portable device, such as an armband or a watch, configured to continuously and non-invasively monitor a wearer's blood glucose level (BGL). In one embodiment, the device utilizes a near-infrared (NIR) light-emitting diode (LED) sensor and machine learning algorithms to accurately predict and continuously provide the wearer's BGL in real time.

[0068] As used herein, the term "about" is understood to be within the normal tolerances in the art and not exceeding ±20% of the specified value. By way of example only, about 940 means from 752 to 1128, including all values therebetween. As used herein, the phrase "about" a particular value also includes the particular value, e.g., about 940 includes 940. For example, values of wavelengths described herein include the peak and a ±20% bandwidth coverage.

[0069] As used herein and in the claims, the terms "substantially" or "substantially" or "essentially" or "basically" mean that the recited features, angles, shapes, states, structures, or values need not be exactly achieved, but deviations or variations (including, for example, tolerances, measurement errors, measurement precision limitations, and other factors known to those skilled in the art) may occur in an amount that does not exclude the effect expected to be provided by the feature. For example, an object having a "substantially" cylindrical shape means that the object has an exact cylindrical shape or an almost exact cylindrical shape. In another example, an object "substantially" perpendicular to a surface means that the object is exactly perpendicular to the surface or almost exactly perpendicular to the surface, e.g., with a 5% deviation.

[0070] The term "light emitter" refers to an element or component that can emit light. For example, a light emitter is or includes one or more "light emitters" or "emitters", such as a light-emitting diode (LED). For example, a light emitter is or includes one or more near-infrared light-emitting diodes. For example, a light emitter is or includes a NIR transceiver, the LEDs of which have a nominal wavelength of about 940 nm, 1350 nm, and / or 1500 nm and a bandwidth of ±20%.

[0071] The term "light receiver" refers to an element or component that can receive light or a light signal. In some examples, a light receiver is or includes a photodetector or an infrared receiver chip, which can receive a light signal from an emitter such as an LED.

[0072] The term "controller" refers to an element or component that controls the operation of components or elements in a control device.

[0073] The term "blood glucose monitoring" is the process of measuring a user's blood glucose level over a period of time. For example, the blood glucose level can be measured regularly. The term "device for blood glucose monitoring" means a device that can make a single measurement of a user's blood glucose level at a specific time point and multiple measurements of a user's blood glucose level over a period of time (e.g., regularly or continuously). The term "continuous glucose monitoring (CGM)" is the process of continuously measuring a user's blood glucose level.

[0074] The terms "light", "light wave", and "light signal" are interchangeable and refer to a form of electromagnetic radiation emitted by a light emitter at a specific wavelength. In some embodiments, the light can be emitted as light pulses over a defined period of time. For example, the light can be one or more near-infrared (NIR) lights having one or more defined wavelengths (e.g., about 940 nm, about 1350 nm, and / or about 1500 nm).

[0075] The terms "reflected light", "reflected light signal", or "reflected light wave" are interchangeable and refer to light reflected from a target surface of a user. For example, the light can penetrate the epidermis, encounter glucose molecules in the user's blood vessels, and be reflected back to a light receiver.

[0076] The term "data vector" refers to a numerical (digital) representation of a data point or set of data points generated from a set of reflected light signals and used as an input to a machine learning model. In some examples, the data point or set of data points is converted from a light signal or set of light signals. The term "data vector stream" refers to a continuous or regular stream of multiple data vectors over a period of time.

[0077] The terms "deep meta-learning framework", "meta-deep learning framework", and "DMLF" are used interchangeably herein to describe a framework for machine learning algorithms that involve training multiple artificial neural networks to learn from other learning algorithms.

[0078] Unless otherwise specifically stated, it should be understood that throughout the specification, discussions using terms such as "processing", "computing", "calculating", "determining", and "identifying" refer to the actions or processes of a computing device such as one or more computers or one or more similar electronic computing devices that manipulate or transform data represented as physical electronic or magnetic quantities within the memory, registers, or other information storage devices, transmission devices, or display devices of a computing platform.

[0079] Embodiments of the methods disclosed herein may be performed in the operation of such a computing device. The order of the blocks presented in the above examples may be changed, e.g., the blocks may be reordered, combined, and / or broken down into sub-blocks. Certain blocks or processes may be performed in parallel.

[0080] As used herein, "adapted to" or "configured to" means open and inclusive language that does not exclude a device adapted to or configured to perform additional tasks or steps. Additionally, the use of "based on" means open and inclusive because a process, step, calculation, or other action "based on" one or more recited conditions or values may actually be based on additional conditions or values beyond the recited conditions or values. The headings, lists, and numbers included herein are for ease of explanation only and are not meant to be restrictive.

[0081] It will also be understood that although terms such as "first", "second", etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first node may be called a second node, and similarly, a second node may be called a first node, which changes the meaning of the description, provided that all occurrences of "first node" are consistently renamed, and all occurrences of "second node" are consistently renamed. The first node and the second node are both nodes, but they are not the same node.

[0082] It should be understood that terms such as "top", "bottom", "middle", "side", "length", "inner", "outer", "interior", "exterior", "outer side", "vertical", "horizontal", etc. that may be used herein only describe reference points and do not limit the present invention to any particular orientation or configuration.

[0083] As used herein and in the claims, unless otherwise specified, "coupled" or "connected" means electrically coupled or connected directly or indirectly via one or more electrical devices.

[0084] As used herein and in the claims, the term "one after another in sequence" means steps, processes, or events that occur one after another in a particular order or sequence in one or more repeating cycles. In some embodiments, a control transmitter emits a first optical signal at a first time interval, then emits a second optical signal at a second time interval, and finally emits a third optical signal at the first time interval in a first cycle. After the first cycle, these steps are repeated in a second cycle, and so on.

[0085] The terms used herein are for the purpose of describing particular embodiments only and are not intended to limit the claims. As used in the description of the embodiments and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms as well. It will also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. It will be further understood that when used in this specification, the terms "comprises" and / or "comprising" specify the presence of the stated features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0086] As used herein, depending on the context, the term "if" can be interpreted to mean "when the precondition holds", or "when the precondition is established", or "in response to determining that the precondition holds", or "in accordance with the determination that the precondition holds", or "in response to detecting that the precondition holds". Similarly, depending on the context, the phrases "if it is determined that the precondition holds" or "if the precondition holds" or "when the precondition holds" can be interpreted to mean "when it is determined that the precondition holds", or "in response to determining that the precondition holds", or "in accordance with the determination that the precondition holds", or "when the precondition is detected to hold", or "in response to detecting that the precondition holds".

[0087] The foregoing description and summary of the invention should be understood to be illustrative and exemplary in every respect and not restrictive, and the scope of the invention disclosed herein should not be determined solely from the detailed description of the illustrative embodiments, but rather in accordance with the full breadth permitted by patent law. It should be understood that the embodiments shown and described herein are merely illustrative of the principles of the invention, and that various modifications can be made by those skilled in the art without departing from the scope and spirit of the invention.

[0088] Although the description refers to specific embodiments, the disclosure should not be construed as limited to the embodiments set forth herein. Examples

[0089] Examples are provided herein that more particularly describe certain embodiments of the disclosure. The examples provided herein are for illustrative purposes only and are not meant to limit the scope of the invention in any way. All references given below and elsewhere in this application are hereby incorporated by reference herein. Example 1A

[0090] Figure 1Ais a perspective view of an embodiment of the portable device 10, which includes a housing 12 and a sensor 14 disposed within the housing 12. As Figure 1A shown, the sensor 14 can be connected to a microcontroller unit (MCU) 16. In another embodiment (not shown), the MCU 16 is disposed within the housing 12. In use, as Figure 1B shown, the portable device 10 can be placed on the skin of a user, and in one embodiment, on the inner surface of the user's forearm, where the sensor 14 is positioned over the user's artery, blood vessel, capillary, etc., such that it can obtain readings from the user's blood.

[0091] As Figure 1C shown, among other components, the optical sensor 14 includes a circuit having a near-infrared light-emitting diode 18 and an infrared receiver chip 20. In one embodiment, the NIR LED 18 is configured to emit infrared signals within the near-infrared region (780 nm to 2500 nm) of the electromagnetic spectrum. In one embodiment, the wavelength of the NIR LED 18 is approximately 940 nm. Commercially available NIR LEDs can be obtained from China Young Sun LED Technology Company, Ltd. The optical sensor also includes an infrared receiver chip 20, which is configured to receive the signal reflected from the user's arm (as Figure 1D shown). In one embodiment, the infrared receiver chip 20 is a monolithic photodiode with an on-chip transimpedance amplifier, such as the OPT101 from Texas Instruments.

[0092] As Figure 1D shown, the infrared signal from the NIR LED 18 is directed towards the user's body, and in one instance, towards the inner surface of the user's forearm, where the infrared signal penetrates the epidermis 22, encounters glucose molecules 24 within the user's blood vessel, and is reflected back to the infrared receiver chip 20.

[0093] Referring again to Figure 1C , the infrared receiver chip 20 converts the reflected infrared light signal into an analog voltage signal, which is then transmitted to a built-in high-precision analog-to-digital converter (ADC) within the MCU 16 to obtain a digital voltage signal x. Then, the digital voltage signal x is transmitted to a trained neural network model, which produces a predicted BGL. Then, the predicted BGL can be transmitted via Bluetooth or other suitable protocols to an application stored on the user's phone or computer or cloud server.

[0094] Figure 1EFIG. 0 is a block diagram of an example system architecture of an exemplary apparatus (or MCU) configured to train, store, and / or use a neural network. While certain specific features are shown, those skilled in the art will understand from this disclosure that, for the sake of brevity and to not obscure the more relevant aspects of the embodiments disclosed herein, various other features are not shown. To that end, as a non-limiting example, in some embodiments, apparatus 100 includes one or more processing units 102 (e.g., microprocessors, ASICs, FPGAs, GPUs, CPUs, processing cores, etc.), one or more input / output (I / O) devices 104, one or more communication interfaces 106 (e.g., interfaces of types such as USB, IEEE 802.3x, IEEE 802.11x, IEEE 802.16x, GSM, CDMA, TDMA, GPS, IR, Bluetooth, ZIGBEE, SPI, I2C, etc.), one or more programming (e.g., I / O) interfaces 108, a memory 110, and one or more communication buses 112 for interconnecting these components and various other components. In some embodiments, one or more communication buses 112 include circuitry for interconnecting and controlling communication between system components.

[0095] Memory 110 includes high-speed random access memory such as DRAM, SRAM, DDR RAM, or other random access solid-state memory devices. In some embodiments, memory 110 includes non-volatile memory such as one or more disk storage devices, optical disk storage devices, flash memory devices, or other non-volatile solid-state storage devices. Memory 110 optionally includes one or more storage devices located remotely from one or more processing units 102. Memory 110 includes non-transitory computer-readable storage medium. In some embodiments, memory 110 or the non-transitory computer-readable storage medium of memory 110 stores the following programs, modules, and data structures or subsets thereof, including an optional operating system 114 and one or more modules 116. Operating system 114 includes programs for handling various basic system services and for performing hardware-related tasks. Neural network trainer 118 is an example of a module that can be configured to train neural network 120 according to the techniques disclosed herein. Neural network 120 represents a neural network that has been integrated into an application or otherwise trained and then stored in memory 110. In one embodiment, neural network 120 can include both linear models and non-linear models. Simulation engine 122 is an example of a module that can be configured to predict a user's BGL as described herein.

[0096] In some embodiments, neural network 120 is a machine learning model. Alternatively, other machine learning models other than neural networks (e.g., artificial neural networks), such as decision trees, support vector machines, Bayesian networks, etc., can be utilized to predict a user's blood glucose level in real time based on the obtained digital voltage signal. In some embodiments, neural network trainer 118 trains neural network 120 to predict a user's blood glucose level by aggregating and classifying the blood glucose levels of multiple users based on the variability of the obtained digital voltage signal using statistical and / or machine learning techniques. In some embodiments, the blood glucose data is classified based on comparing the variability of the digital voltage signal with a threshold. For example, the techniques described herein can classify the obtained digital voltage signal of a user to determine the blood glucose state as a hypoglycemic state, a normoglycemic state, and / or a hyperglycemic state.

[0097] Figure 1E Rather, it is a functional description of the various features present in a particular embodiment than a structural schematic of the embodiments described herein. As will be appreciated by one of ordinary skill in the art, the items shown separately can be combined and some items can be separated. The actual number of units and the specific functional partitioning and how the features are allocated among them will vary from embodiment to embodiment and, in some embodiments, will depend in part on the particular combination of hardware, software, or firmware selected for a particular embodiment.

[0098] Figure 1F is a flowchart showing an exemplary method 100 for predicting a blood glucose level using a trained neural network model. In some embodiments, method 100 is performed by a device (MCU) (e.g., Figure 1E device 100). Method 100 can be performed at a mobile device, a desktop computer, a laptop computer, or a server device. In some embodiments, method 100 is performed by processing logic that includes hardware, firmware, software, or a combination thereof. In some embodiments, method 100 is performed by a processor that executes code stored in a non-transitory computer-readable medium (e.g., a memory).

[0099] At block 110, the method obtains new input digital voltage data x from an ADC (which converts analog voltage data received from an infrared receiver chip) and transmits the data for processing 112. The data x then undergoes a trained non-linear model 114 and a linear model 116, which produce a class prediction probability value and a numerical value, respectively. The data is then classified 118 as low, normal, or high, and the classification results of the two models are compared 120 to determine if the values are consistent.

[0100] If so, the determination 122 outputs the blood glucose status and the predicted blood glucose value.

[0101] In one example, the non-linear model produces the probabilities for each of the three glucose states, i.e., low glucose level (0.2), normal glucose level (0.6), and high glucose level (0.2). The state with the highest probability is considered the true state. In this example, the normal glucose state would be considered the true state. Then, the linear model is configured to produce a numerical value of the glucose level, such as 5.9. Based on the predefined intervals for the three states (e.g., 0 - 4 represents low, 4 - 8 represents normal, and >8 represents high), the linear model would also result in one of the three states. In this example, 5.9 would be considered normal. Thus, the non-linear model and the linear model produce consistent results.

[0102] The machine learning models 114 and 116 are trained to predict the user's BGL based on digital infrared light reflection signals that are converted from analog infrared light reflection signals collected by the infrared light receiving chip 20. The machine learning models are trained by comparing the BGL predicted based on the voltage signals with the experimentally determined BGL obtained by a professionally certified invasive blood glucose device.

[0103] The linear model 116 can utilize a polynomial regression model. For example, one example of the non-linear mode is an ANN (artificial neural network) model. It should be understood that other models, such as convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory (LSTM), or generative adversarial network (GAN), can also be used.

[0104] During the iterative training process, the machine learning model can make predictions based on such input, compare those output predictions with the experimentally determined BGL, and adjust the configuration of the machine learning model to reduce the difference between the predictions and the experimentally known results in future iterations. Training such a model using multiple collected BGLs can provide an accurate machine learning model for predicting the user's BGL. Example 1B Example apparatus and system architecture

[0105] In this example, the exemplary glucose monitoring portable device 10 described as in Example 1A was used for experiments. The exemplary glucose monitoring portable device includes a light emitter with the wavelength of the NIR LED 18 being about 940 nm, a light receiver (which is the infrared receiver chip 20), a controller (which is the MCU 16), and a housing (which is the housing 12). The exemplary device 10 generates an infrared signal emitted by the NIR LED 18, and the infrared signal is directed at the surface of the user's forearm, where the infrared signal penetrates the epidermis, encounters glucose molecules in the user's blood vessels, and is reflected back to the light receiver (in this example, the infrared receiver chip 20). Then, the reflected infrared light signal is converted into an analog voltage signal (digital signal) in the MCU 16, and the analog voltage signal is transmitted to a data processing system with a trained machine learning module (which is a trained neural network model) as described in Example 1 after preprocessing, and the data processing system finally generates the predicted blood glucose level of the user. For each measurement, the blood glucose level data was captured for about 2 minutes, and the stable value was selected and recorded. In other examples, the blood glucose level data was captured for 1 - 3 minutes. Accuracy analysis

[0106] Fifteen volunteers were recruited (10 volunteers for data collection for neural network model training and 5 volunteers for system performance evaluation) to evaluate the accuracy of the exemplary non-invasive continuous glucose monitor device (hereinafter abbreviated as 'niCGM') and compare its performance with that of the commercially available device, the FreeStyle Libre2 system of Abbott (hereinafter abbreviated as 'Abbott'), which uses a minimally invasive method and a sensor pad to penetrate the skin. The invasive continuous glucose monitoring blood glucose meter of Sinocare Inc. (Safe-Accu) (Changsha Sinocare Inc.) (hereinafter also referred to as 'Sannuo') was used as a reference (as an accuracy standard), and the blood glucose meter uses finger pricking to collect blood and obtains the glucose level through test strips. Figure 2A FIG. shows the exemplary niCGM device 10 placed on the user's skin. Figure 2B FIG. shows an exemplary console APP 1600 for controlling the niCGM device 10 and receiving blood glucose measurement results.

[0107] The blood glucose levels of 15 volunteers were collected 3 - 4 times a day using niCGM, Abbott, and Sannuo respectively for 30 days. The data of 10 volunteers within 30 days were used to train our deep learning framework, and the data of the remaining 5 volunteers (Volunteers #11 - 15) were used for testing. The procedures for determining the blood glucose levels of Abbott and Sannuo were carried out according to the manufacturer's instructions respectively.

[0108] Now referring to Figures 3A to 3E , the comparison results of the accuracy of the test data of Volunteer #11 in Sannuo, Abbott, and the exemplary device niCGM are shown. The test data of Volunteer #11 were measured 3 times a day for 30 days to obtain a total of 90 output data points respectively.

[0109] Figure 3A The comparison of glucose level accuracy using the test data of measuring Volunteer #11 3 times a day for 30 days is shown. The results show that the overall blood glucose level of the exemplary device niCGM is highly correlated with the overall blood glucose level of the reference Sannuo. Figure 3B Shown in Figure 3A is the error rate of the exemplary device niCGM from the test data in Figure 3C Shown in Figure 3A is the error rate of the glucose monitor of Abbott from the test data in Figure 3D . The results show that 95.56% of the error rates are lower than 10%, and 100% of the error rates are lower than 15%, indicating that the exemplary device in this disclosure has a comparable or even lower error rate during the entire 90 - day test period compared to the minimally invasive device Abbott. As Figure 3E, the figure shows the Clarke error grid analysis of the example device niCGM and Sannuo as a reference sensor. Region A is those values within 20% of the reference sensor. Region B contains points outside 20% but does not result in inappropriate treatment. Region C is those points that lead to unnecessary treatment. Region D is those points that indicate a potential dangerous failure to detect hypoglycemia or hyperglycemia. Region E is those points that confuse the treatment of hypoglycemia with hyperglycemia (and vice versa). The results show that the example device niCGM achieved 100% accuracy in Region A of the CEG (Consensus Error Grid) system (i.e., values within 20% of the reference sensor), as Figure 3E shown.

[0110] Now referring to Figures 4A to 4E , the results of the accuracy comparison of the test data of Volunteer #12 in Sannuo, Abbott, and the example device niCGM are shown. Similarly, the test data of Volunteer #12 was measured 3 times a day for 30 days to obtain a total of 90 output data points respectively.

[0111] Figure 4A shows the glucose level accuracy comparison using the test data of measuring Volunteer #12 3 times a day for 30 days. The results show that the overall blood glucose level of the example device niCGM is highly correlated with the overall blood glucose level of the benchmark Sannuo. Figure 4B shows the error rate of the example device niCGM from the test data in Figure 4A . The results show that 98.89% of the error rates are less than 10%, and 100% of the error rates are less than 15%, indicating that the example device in this disclosure has an extremely low error rate during the entire 90-day test period. Figure 4C shows the error rate of the Abbott glucose monitor from the test data in Figure 4A . The results show that 95.56% of the error rates are less than 10%, and 100% of the error rates are less than 15%, indicating that compared with the minimally invasive device Abbott, the example device in this disclosure has a comparable or even lower error rate during the entire 90-day test period. As Figure 4D shown, the average error rate and the standard deviation of the error rate of our glucose monitor are 5.7 and 2.0 respectively, while the average error rate and the standard deviation of the error rate of Abbott are 7.7 and 2.2 respectively, indicating that compared with Abbott, the example device in this disclosure has a significantly lower error rate and the standard deviation of the error rate. Now referring to Figure 4E, the figure shows the Clarke error grid analysis of niCGM using the test data of Volunteer #12 measured three times a day for 30 days and Sannuo as a reference sensor. The results also show that the exemplary device niCGM achieved 100% accuracy in Region A of the CEG (Consensus Error Grid) system (i.e., values within 20% of the reference sensor), as Figure 4E shown.

[0112] Now referring to Figures 5A to 5E , the figure shows the accuracy comparison results of the test data of Volunteer #13 in Sannuo, Abbott, and the exemplary device niCGM. Similarly, the test data of Volunteer #13 was measured three times a day for 30 days to obtain a total of 90 output data points respectively.

[0113] Figure 5A presents the glucose level accuracy comparison using the test data of Volunteer #13 measured three times a day for 30 days. The overall blood glucose level of the exemplary device niCGM is highly correlated with the overall blood glucose level of the reference Sannuo. Figure 5B shows the error rate of the exemplary device niCGM from the test data in Figure 5A . The results show that 97.78% of the error rates are less than 10%, and 100% of the error rates are less than 15%, indicating that the exemplary device in this disclosure has an extremely low error rate during the entire 90-day test period. Figure 5C shows the error rate of Abbott from the test data in Figure 5A . The results show that 95.56% of the error rates are less than 10%, and 100% of the error rates are less than 15%, indicating that the exemplary device in this disclosure has a comparable or even lower error rate during the entire 90-day test period compared to the minimally invasive device Abbott. As Figure 5D shown, the average error rate and the standard deviation of the error rate of the exemplary device niCGM are 5.85 and 2.03 respectively, while the average error rate and the standard deviation of the error rate of Abbott are 7.99 and 2.01 respectively, indicating that the exemplary device in this disclosure has a significantly lower error rate and standard deviation of the error rate compared to Abbott. Now referring to Figure 5E , the figure shows the Clarke error grid analysis of the exemplary device niCGM using the test data of Volunteer #13 measured three times a day for 30 days and Sannuo as a reference sensor. The results show that the exemplary device niCGM achieved 100% accuracy in Region A of the CEG (Consensus Error Grid) system (i.e., values within 20% of the reference sensor), as Figure 5E shown.

[0114] Now referring to Figures 6A to 6E, which shows the accuracy comparison results of the test data of Volunteer #14 in Sannuo, Abbott, and the example device niCGM. Similarly, the test data of Volunteer #14 was measured three times a day for 30 days to obtain a total of 90 output data points respectively.

[0115] Figure 6A presents a comparison of glucose level accuracy using the test data of measuring Volunteer #14 three times a day for 30 days. The overall blood glucose level of the example device niCGM is highly correlated with the overall blood glucose level of the reference Sannuo. Figure 5B shows the error rate of the example device niCGM from the test data in Figure 6A . The results show that 98.89% of the error rates are lower than 10%, and 100% of the error rates are lower than 15%, indicating that the example device in this disclosure has an extremely low error rate during the entire 90-day test period. Figure 6C shows the error rate of Abbott from the test data in Figure 6A . The results show that 95.56% of the error rates are lower than 10%, and 100% of the error rates are lower than 15%, indicating that compared with the minimally invasive device Abbott, the example device in this disclosure has a comparable or even lower error rate during the entire 90-day test period. As shown in Figure 5D , the average error rate and the standard deviation of the error rate of the example device niCGM are 6.07 and 2.11 respectively, while those of Abbott are 8.23 and 2.00 respectively, indicating that compared with Abbott, the example device in this disclosure has a significantly lower error rate and the standard deviation of the error rate. Now referring to Figure 6E , the figure shows the Clarke error grid analysis of the example device niCGM and Sannuo as a reference sensor using the test data of measuring Volunteer #14 three times a day for 30 days. The results show that the example device niCGM achieved 100% accuracy in Region A of the CEG (Consensus Error Grid) system (i.e., values within 20% of the reference sensor), as shown in Figure 6E .

[0116] Now referring to Figures 7A to 7E , which shows the accuracy comparison results of the test data of Volunteer #15 in Sannuo, Abbott, and the example device niCGM. Similarly, the test data of Volunteer #15 was measured three times a day for 30 days to obtain a total of 90 output data points respectively.

[0117] Figure 7AThe comparison of glucose level accuracy using the test data of measuring Volunteer #15 three times a day for 30 days is presented. The overall blood glucose level of the example device niCGM is highly correlated with the overall blood glucose level of the benchmark Sannuo. Figure 7B The error rate of the example device niCGM from the Figure 7A test data therein is shown. The results show that 98.89% of the error rates are below 10%, and 100% of the error rates are below 15%, indicating that the example device in this disclosure has an extremely low error rate during the entire 90-day test period. Figure 7C The error rate of Abbott from the Figure 7A test data therein is shown. The results show that 95.56% of the error rates are below 10%, and 100% of the error rates are below 15%, indicating that compared with the minimally invasive device Abbott, the example device in this disclosure has comparable or even lower error rates during the entire 90-day test period. As Figure 7D shown, the average error rate and the standard deviation of the error rate of the example device niCGM are 5.77 and 2.25 respectively, while those of Abbott are 8.21 and 2.11 respectively, indicating that compared with Abbott, the example device in this disclosure has similar or even lower error rates and the standard deviation of the error rate. Now referring to Figure 7E , the figure shows the Clarke error grid analysis of our niCGM and Sannuo CGM as a reference sensor using the test data of measuring Volunteer #12 three times a day for 30 days. The results show that the example device niCGM achieved 100% accuracy in Region A of the CEG (Consensus Error Grid) system (i.e., values within 20% of the reference sensor), as Figure 7E shown.

[0118] Now referring to Figures 8A to 8B , the comparison results of the foregoing test data between 5 volunteers (Volunteer #11 - 15) for Abbott and the example device niCGM are shown. According to the results presented above, it is clearly shown that the example device niCGM and its system achieved better accuracy than Abbott's minimally invasive FreeStyle Libre 2. Specifically, the average error rate and the standard deviation of the error rate of the example device niCGM for Volunteer #11, #12, #13, #14, and #15 are 5.83 and 2.03 respectively, while those of Abbott's FreeStyle Libre 2 for Volunteer #11, #12, #13, #14, and #15 are 7.99 and 2.01 respectively. Figure 8A The comparison results are summarized therein. Now referring to Figure 8B, the figure shows the Clarke error grid analysis of the exemplary device niCGM using test data from volunteers #11 - 15 and Sannuo as a reference sensor. As Figure 8B shown, the exemplary device niCGM achieved 100% accuracy in region A of the CEG (Consensus Error Grid) system.

[0119] In summary, the results show that compared with the Abbott's minimally invasive glucose monitor FreeStyle Libre 2, which uses the performance of Sannuo's minimally invasive glucose monitor as a benchmark reference sensor, the exemplary non - invasive continuous glucose monitoring device (niCGM) and its system perform excellently in terms of accuracy. The results show that the average error rate and the standard deviation of the error rate of the exemplary device and system niCGM are 5.83 and 2.03 respectively, while those of Abbott's FreeStyle Libre 2 are 7.99 and 2.01 respectively, indicating that the exemplary device in this disclosure has significantly lower error rate and standard deviation of the error rate compared with Abbott. In addition, the exemplary device and system niCGM achieved 100% accuracy in region A of the CEG (Consensus Error Grid) system. Example 2 Example blood glucose monitoring system and apparatus

[0120] Now referring to Figure 9A , a schematic diagram of an exemplary system 2000 including an exemplary non - invasive continuous glucose monitoring (niCGM) device 2100 is shown. In this example, the device 2100 is a wearable device. The device 2100 generally includes a housing, a light emitter, a light receiver, and a controller. In this example, the housing is or includes a case 2500 to house other components, the light emitter is or includes an emitter 2101, the light receiver is or includes a photodetector 2102, and the controller is or includes a signal - processing PCBA (Printed Circuit Board Assembly) 2200. The emitter 2101, the photodetector 2102, and the PCBA are disposed within the case 2500. The PCBA 2200 includes a processor unit coupled to a memory storing an executable software program. In this example, the processor unit is a microprocessor, and the software program can be named embedded software 2300. In this example, the embedded software 2300 further includes an operation module for controlling the operation of the device 2100 and a data - processing system for processing data signals, wherein the data - processing system includes a machine - learning module for analyzing the data signals to generate output data as the blood - glucose level of the user. In this example, the machine - learning module includes a deep meta - learning framework (DMLF) module 2400, which is a meta - learning software plug - in of the device 2100 and will be in Figure 9AShown and described in more detail later. In this example, the emitter 2101 and the photodetector 2102 are adjacent to each other near the lower surface of the housing 2500 at a defined distance. The emitter 2101 is configured to emit a first light signal, a second light signal, and a third light signal directed at a target surface (e.g., skin) of a user such that a first reflected light signal, a second reflected light signal, and a third reflected light signal (collectively referred to as reflected light signals or a reflected light signal group) reflected from the target surface (skin) can be substantially received by the photodetector 2102.

[0121] In some examples, the light emitting portion of the emitter and / or the signal receiving portion of the photodetector are positioned at a defined angle towards each other such that most or substantially all of the reflected light signals can be received by the photodetector 2102. In this example, the emitter 2101 and the photodetector 2102 are positioned on the lower side of the housing 2500 such that the emitter 2101 and the photodetector 2102 are close to the target detection area (such as the skin of the forearm or wrist) of the subject or user.

[0122] In some examples, the device 2100 further includes one or more optical refraction lens components for guiding the NIR optical path to the PCBA 2200.

[0123] The emitter 2101 and the photodetector 2102 are in electrical communication with the PCBA 2200. In other words, the PCBA 2200 is operatively connected to the emitter 2101 and the photodetector 2102. One of the functions of the PCBA 2200 is to control the emission of light from the emitter 2101 and capture the response of the reflected light signals of the photodetector 2102. Example NIR LED emitter

[0124] In some instances, the emitter 2101 is an array of multiple NIR LEDs that emits NIR signals onto a target detection area of a user or subject. In some instances, the emitter 2101 is configured as an array of more than one near-infrared light-emitting diode (NIR LED) for emitting at least one infrared signal within the near-infrared region (from 780 nm to 2500 nm) of the electromagnetic spectrum. In some instances, the emitter 2101 is an array of 3 NIR LEDs that separately and individually emit a first light with a wavelength of 752 - 1125 nm, a second light with a wavelength of 1080 - 1620 nm, and a third light with a wavelength of 1200 - 1800 nm. In this instance, the emitter 2101 is an array of 3 NIR LEDs that separately and individually emit a first light signal with a wavelength of approximately 940 nm, a second light signal with a wavelength of approximately 1350 nm, and a third light signal with a wavelength of approximately 1500 nm. For ease of description, the first light signal, the second light signal, and the third light signal (and more light signals) are collectively referred to as "light signals", and the first reflected light signal, the second reflected light signal, and the third reflected light signal (and more reflected light signals) reflected from the target surface are collectively referred to as "reflected light signals". For clarity, the reflected light signals reflected from the target surface include light components that penetrate downward and are reflected from skin tissue, blood vessels, blood, etc.

[0125] The light emitter and the light receiver of the device 2100 are configured to be positioned above a user's target surface that may have arteries, blood vessels, capillaries, etc. below, such that the emitted light signals can be directed at the user's arteries, blood vessels, capillaries, etc. and reflected to the light receiver. The emitted light waves can penetrate the epidermis, the dermis layer, and / or the blood vessel wall to reach the user's blood. Then, a certain amount of light will be absorbed, and at least a certain remaining amount of light will be reflected by the blood components and the body tissues of the skin and blood vessels. The reflected light signals will be substantially received by the light detector 2102.

[0126] In this instance, the NIR LEDs of the emitter 2101 are turned on and off one by one in different time slots to separately generate light signals so that the light detector 2102 can better capture the absorption factor and the reflection factor of each wavelength. In some instances, the switching of the NIR LEDs is driven and controlled by the PCBA 2200. Example NIR photodetector

[0127] The photodetector 2102 is configured to receive the reflected optical signal. In this example, the photodetector 2102 covers three detection wavelengths of approximately 940 nm, approximately 1350 nm, and approximately 1500 nm. In another example, the photodetector 2102 covers detection wavelengths ranging from 940 nm to 1500 nm. This arrangement will enable the photodetector to receive signals reflected from each of the plurality of NIR transmitters. In another example, the photodetector 2102 covers detection wavelengths ranging from 940 nm to 950 nm. Without being bound by any theory, the attenuation of each reflected optical signal will reflect the absorption caused by each constituent blood component. For the attenuation level on each waveform, the levels of blood glucose and other blood components can be deduced. In some examples, the photodetector 2102 is configured to receive NIR reflected optical signals in the range of approximately 900 - 1700 nm with a relative spectral sensitivity of at least 40%. In this example, the photodetector 2102 is a monolithic photodiode with an on-chip transimpedance amplifier, such as the OPT101 (Texas Instruments). Example printed circuit board assembly

[0128] Now referring to Figure 9B , a schematic diagram of the controller printed circuit board assembly PCBA 2200 is shown. The PCBA 2200 is a control system configured to be operatively connected to an optical transmitter (transmitter 2101) and an optical receiver (photodetector 2102) and is intended to operate the transmitter and the photodetector. The PCBA 2200 is further configured to control the transmitter 2101 to turn on and off to sequentially emit a first optical signal, a second optical signal, and a third optical signal one by one in a plurality of cycles, such that a plurality of groups of optical signals are formed at defined time intervals, and each group of optical signals includes a first reflected optical signal, a second reflected optical signal, and a third reflected optical signal obtained in each cycle. The PCBA 2200 includes an operation module that controls the operation of the device and converts the reflected optical signal into digital data. The two-dimensional array consists of three columns and multiple rows of data vectors of reflected optical signals, and in this example, the PCBA 2200 includes a battery management system 2201, a low dropout regulator (LDO) 2202, a voltage converter 2203, a transmitter connector 2204, a photodetector connector 2205, a wireless module 2206, a user interface indicator 2207, a thermometer 2208, a memory 2209, and a microcontroller unit (MCU) 2210 loaded with embedded software 2300.

[0129] The battery management system 2201 is electrically connected to the LDO 2202 and the voltage converter 2203 to ensure the power supply of the PCBA 2200. In some instances, the battery management system 2201 includes a battery (such as a rechargeable battery) and a charger. In some instances, the voltage converter 2203 is a DC / DC converter. In some instances, the voltage of the PCBA 2200 is regulated to 3.3V (3V3).

[0130] The voltage converter 2203 is electrically connected to the transmitter connector 2204 and the optical detector connector 2205. The transmitter connector and the optical detector connector are respectively connected to the transmitter 2101 and the optical detector 2102 to supply a regulated voltage to the transmitter 2101 and the optical detector 2102. The MCU 2210 is connected to the transmitter connector 2204 via a general-purpose input / output (GPIO), and the optical detector connector 2205 is connected to the MCU 2210 via an analog-to-digital converter (ADC) interface.

[0131] The LDO 2202 is connected to the flash memory 2209 and the MCU 2210 to ensure a regulated voltage to the MCU 2210. The MCU 2210 and the flash memory 2209 are connected via a serial peripheral interface (SPI) bus.

[0132] The MCU 2210 is further electrically connected to the wireless module (Bluetooth 5.0) 2206 via BLE 2.4G respectively, and is connected to the user interface indicator (UI LED) 2207 via GPIO. The thermometer (NTC) 2208 is electrically connected to the MCU 2210 via an ADC.

[0133] The wireless module 2206 streams the processed signal (digital data vector) to the backend console via a mobile application (APP), or streams the processed signal to the backend server via Bluetooth, WiFi, or any other wireless means in the art. In some instances, the wireless module 2206 is Bluetooth 5.0 using a short-range wireless communication protocol (such as those operating in the 2.4GHz band). In other instances, the wireless module 2206 is replaced by a connection module (e.g., a USB connector) that alternatively allows a direct wired connection.

[0134] In some instances, the user interface indicator 2207 is connected to the MCU 2210 via a general-purpose input / output (GPIO) and indicates the battery status and the detection status of the PCBA 2200. The user interface indicator 2207 is controlled by the GPIO from the MCU 2210.

[0135] In some instances, the thermometer 2208 is connected to the MCU 2210 via an analog-to-digital converter (ADC) interface, which is a negative temperature coefficient (NTC) thermometer that detects the system environment to ensure it has a normal operating range. In some instances, the thermometer also captures the body temperature of the subject and then converts the analog signal into additional digital data vectors for data analysis.

[0136] The flash memory 2209 stores the detected and processed optical detection data (digital data, data vectors, digital data vector streams, data vector streams, and / or output data, etc.). In some instances, the flash memory 2209 is an 8M-byte SPI flash memory. In other instances, other types and memory sizes of memory components available in the art may be alternatively used.

[0137] The microcontroller unit (MCU) 2210 is loaded with the embedded software 2300 and executes the control program or the embedded software 2300. The embedded software 2300 is a set of software programs designed to perform the operations of the device 2100.

[0138] Now refer to Figure 9C , the flowchart shows the steps of an example operation when executing the embedded software 2300. The embedded software 2300 includes operation modules that control the operations of the device 2100. In step 2310, the embedded software 2300 controls the timing of each NIR LED of the transmitter 2101 to emit signals with wavelengths of 940 nm, 1350 nm, or 1500 nm on the target. In some instances, the NIR LEDs of the transmitter 2101 are turned on / off one after another in 3 different time slots in sequence to capture the absorption factor or reflection factor of each wavelength.

[0139] In step 2320, the embedded software 2300 controls the reception timing of the light detector 2102 to receive the reflected signal from the target. In some instances, the reflected signal from the target is 940 nm, 1350 nm, or 1500 nm.

[0140] In step 2330, the NIR signal is fed to the MCU 2210 for processing.

[0141] In step 2340, the embedded software 2300 processes the optical signal group into a digital data vector stream and outputs it to the built-in data processing system in the device with a deep meta-learning framework module.

[0142] In step 2350, the digital data vector stream is processed by the embedded DMLF module 2400.

[0143] In step 2360, the data analysis results are displayed, for example, on a wearable user interface (UI). Additionally or alternatively, the data analysis results can be reported via wireless communication to other parties, such as a hospital or a caregiver.

[0144] In other instances, steps 2350 - 2360 can be replaced with alternative instances, which will be described in more detail in Examples 4 - 5 below (e.g., the data vector stream of the 3 - wavelength detection will be streamed via 3206 Bluetooth / WiFi to the console). Example deep meta - learning framework module

[0145] Now referring to Figure 9D , a flowchart of the DMLF module 2400 of the data processing system is shown. Figure 9D A software block diagram of the DMLF module 2400 is depicted. The module employs a deep meta - learning framework (also referred to as 'DMLF') to analyze and predict blood glucose levels in one example of the present invention. Referring to this figure, one or more processed digital data vectors (or near - infrared sensor data) are passed to the input data vector module 2410. In some instances, the input data vector further includes or encompasses other key user demographics or parameters, such as gender, age, ethnicity, blood glucose status, cholesterol level, and / or cross - circadian rhythm timestamps. Due to uncontrollable and inevitable environmental variations, the infrared light vector data may be noisy and contain abnormal data points. Thus, the input data vector is passed to the data pre - processing module 2420 for data pre - processing or cleaning. Thereafter, it is fed into the deep learning ensemble model. In one example, a stacked ensemble model is employed. The stacked ensemble model includes at least one base model 2430. Each base model is selected from a set of machine - learning models, and they can be different from each other. The outputs of these base models are sent to the meta - model 2450 via respective model weights represented as w1 to w Figure 9D as shown in n (2440). The meta - model 2450 takes the weighted output values as inputs and calculates an aggregated value indicating the user's glucose level.

[0146] In one example, the input data module 2410 combines near-infrared sensor data into a two-dimensional array consisting of three columns and multiple rows. Each column corresponds to a reading of the near-infrared sensor data at a specific wavelength. The device is designed to collect samples of the near-infrared sensor data at regular intervals. In the example, the sampling frequency is about 60 per minute. Each sample consists of three values corresponding to the readings of the three near-infrared sensor data outputs; and they are stored as a row of the above two-dimensional array. Therefore, the array has many rows, and each row has three columns. In some examples, a sliding window (or sampling time window) can be set to collect a set (3 columns) of data vectors, such that there will be 3 means and 3 medians, each calculated based on the corresponding column of the data vectors collected within a defined time period. For example, in the case where the sampling "time window" is set to 2 minutes, there will be 120 rows of data, the data will be preprocessed, and the mean and median of each column will be calculated. In other examples, the time window is set to about 1, 3, 4, 5, 6, 7, 8, 9, 10 or more minutes.

[0147] The two-dimensional array (and optionally the corresponding timestamps) is passed to the data preprocessing module 2420 to clean noisy and outlier data points. In some examples, this module 2420 performs three subtasks: data cleaning, data processing, and / or training data preparation. These subtasks are detailed below.

[0148] The purpose of data cleaning is to remove certain anomalies or outliers from the collected data. These anomalies may occur due to errors in the collected data or because they deviate significantly from the normal data cluster. These anomalies may potentially interfere with the normal operation of the deep learning stacked ensemble model during training. In one example, three data cleaning methods are employed. They are: (1) Isolation Forest algorithm (iForest), (2) One-Class Support Vector Machine (SVM), and Local Outlier Factor (LOF) algorithm.

[0149] In Isolation Forest (iForest), anomalies are defined as "outliers that are easy to isolate". This means they can be regarded as points far from the densely populated clusters. In the feature space, regions with sparse population indicate a lower probability of events occurring in that region. Therefore, data points falling within these regions are considered anomalies. Isolation Forest is an unsupervised anomaly detection method suitable for continuous data, which means it does not require labeled samples for training, but the features need to be continuous. In addition, Isolation Forest is based on decision trees, which allows it to be easily parallelized and scaled to handle large datasets. Isolation Forest uses an efficient strategy to identify easily isolable points by recursively partitioning the dataset until all samples are isolated. Using this random partitioning strategy, outlier points usually have shorter paths and can then be filtered out.

[0150] The single-class SVM algorithm shares similar principles and mathematical models with the SVM algorithm. It treats the entire data set as a single class and seeks a hyperplane that separates normal data from abnormal data. This hyperplane is defined as the one that maximizes the margin between the boundary and the observations in the class. During the test phase, new data points are evaluated to determine whether they are inside or outside the boundary. Points that fall outside the boundary are classified as outliers.

[0151] The Local Outlier Factor (LOF) is a probability model commonly used for clustering or density estimation of data. In the context of outlier detection, LOF can identify outlier points by estimating the probability density distribution of the data. Outlier points usually have a lower probability density, so their status as outliers can be determined by calculating the probability of data points under each Gaussian distribution.

[0152] The performance of three data cleaning methods is evaluated and shown in Table 1 below. These experiments were conducted under the condition that all other parameters in the data preprocessing module and the stacked ensemble model (including the base model and the meta-model) are fixed except for the internal parameters of these three methods. The same data set was used to evaluate these three cleaning methods. 1240 volunteers were recruited to evaluate the accuracy of the exemplary non-invasive continuous glucose monitoring system 2000 and device 2100 (hereinafter abbreviated as 'niCGM') and compare its performance with that of the standard hospital clinical device BS-350E (Mindray), where the standard hospital clinical device obtains the blood glucose level by drawing blood from the volunteers (or using a finger prick to collect blood) and analyzing the blood sample in the laboratory to obtain information (hereinafter abbreviated as 'control'). Test data were measured from volunteers #1 - 1240. 70% of the data was used to train the DMLF module, and 30% of the data was used for system performance evaluation. Therefore, the performance metric is the accuracy between the predicted value and the verified data. The accuracy of the exemplary device was calculated by the Mean Absolute Relative Difference (MARD). For example, if the comparison value is x and the reference value is y, the absolute relative difference is |x - y| / y * 100%, and the MARD is the average of all absolute relative differences.

[0153] In these experiments, the following parameters were used for each data cleaning method. For Isolation Forest: the number of estimators was 100; the sample size was 256; the maximum number of features was 1, and the outlier ratio was 0.1. For single-class SVM, the outlier ratio was set to 0.1; and for LOF, it was set to detect 20 Local Outlier Factor points with an outlier ratio of 0.2 using the Euclidean distance metric. Using this set of parameters, each data cleaning method was evaluated individually and in combination with other methods. As shown in Table 1, the best result was obtained by combining Isolation Forest and single-class SVM. It produced an accuracy of 0.882. Table 1. Experimental results of data cleaning

[0154] Based on the experimental results, both iForest and one-class SVM are used for the data cleaning subtask. In one implementation, the two modules process the data from the data vector module 2410 sequentially. If a data point is identified as an outlier or anomalous by the first method, it is discarded from the data set for subsequent processing. In other instances, the modules are executed simultaneously. If any data point is identified as an outlier or anomalous by one or both methods, it is discarded from the data set for subsequent processing.

[0155] After data cleaning, the purpose of data processing is to merge the data into a table. Common practices involve using the mean or median of each dimension (i.e., each column of the data set table). In one instance, the mean and median of each column of the data table are calculated. In additional instances, a method of representing the data as the average of the median and the mean is employed. Table 2 shows the results of three experiments. Namely, the mean alone, the median alone, and the average of the mean and the median. Additionally, the experimental conditions are the same as before and will not be repeated here. Table 2. Experimental results of data processing methods

[0156] Table 2 shows that by adopting the average of the mean and the median, the accuracy is improved. Therefore, this method is adopted in one implementation of the present invention.

[0157] The third subtask of the data preprocessing module 2420 is training data preparation. In one instance, the bootstrap sampling method is adopted. Substantially, after data cleaning and adopting the average of the mean and the median, new data subsets are generated for each of the base models 2430. These data subsets are randomly generated by the sampling method with replacement. The size of each subset is the same as the original data set, but some instances may be replicated while other instances may be omitted. Therefore, each data subset is unique and different from other data subsets.

[0158] Then the data subsets are passed to a stacked ensemble model composed of at least two layers. In the instance depicted in Figure 9D , the first layer includes a plurality of base models 2430, and the second layer is the meta-model 2450. Each base model in the first layer is connected to the meta-model via the model weights 2440. The stacked ensemble model is a powerful and flexible machine learning method that can combine the prediction results of individual base models to achieve better and more robust predictions.

[0159] In an example of blood glucose level prediction, each base model acts as a regressor that takes a subset of the data assigned to that base model as input and calculates a value representing the blood glucose level. This value is a continuous variable. The base models can be selected from various deep learning regressors. In one example, candidates for the base models include, but are not limited to, recurrent neural network (RNN) regressors, long short-term memory (LSTM) regressors, convolutional neural network (CNN) regressors, support vector regression (SVR), multi-layer perceptron (MLP) regressors, k-nearest neighbor (KNN) models, ElasticNetCV regressors, Catboost regressors, XGBoost (extreme gradient boosting) regressors, gradient boosting regressors, LGBM regressors, bagging regressors, decision trees, etc.

[0160] In one example, the same deep learning regressor can be deployed to one or more base models 2430. Since the data preprocessing module 2420 prepares different subsets of data for each of these base models, even when different base models select the same deep learning regressor, these base models are trained to capture different aspects of the entire dataset. Additionally, the stacking ensemble employs a flexible framework whereby different base models can select different deep learning regressors. In some cases, the base model itself can be a stacking ensemble model. Thus, the ensemble of base models is able to cover all aspects of the infrared sensor data. By combining multiple predictions at the meta-model 2450, the overall prediction can be significantly improved.

[0161] Many of the above deep learning techniques need to be trained before being put into use. Training is a process by which a learning algorithm specifically designed for that deep learning regressor can adjust internal parameters to improve its prediction performance by repeatedly presenting a set of training data to the base model. In one example, the subset of data is divided into a training set and a test set. In additional examples, 70% of the subset of data is used for training and the remaining 30% is used for testing. After the base model is properly trained, the internal parameters are frozen and the base model can be readily used to perform regression predictions.

[0162] The outputs of each base model 2430 are sent to the input of the meta-model 2450 via the corresponding model weights 2440. In one instance, the model weights 2440 can be the same. In another instance, the model weights can be randomly assigned. In yet another instance, the model weights can be trained. In one implementation, after the base models are trained and their corresponding internal parameters are frozen, the same set of training data is fed into the base models. The output of each base model serves as the input to these model weights. In a further implementation, the PyTorch machine learning library is used to define the neural network model and its trainable parameters. The nn.Parameter subclass in this library is used to train the model weights 2440. This is done by first initializing all model weights to one (1) using torch.ones. Then it is wrapped with nn.Parameter. This enables the model weights to be trained synchronously with the gradient descent of the model during the training process.

[0163] In a similar manner, the meta-model 2450 can be trained after the base models 2430 and the model weights 2440 are trained and frozen. The same training dataset with the same split of 70% training and 30% testing is used to train the internal parameters of the meta-model. In one instance, candidates for the meta-model include, but are not limited to, multi-layer perceptron (MLP), recurrent neural network (RNN), convolutional neural network (CNN), etc.

[0164] A set of experiments is conducted to evaluate the performance of the MDLP module. The experimental setup is the same as before and will not be repeated here. In these experiments, we focus on the performance of the stacked ensemble model (i.e., the development of the base models and the meta-model), while keeping the parameters in the data preprocessing module unchanged. Table 3 below shows the results. Table 3. Experimental results of the MDLP module implementing the deep meta-learning framework

[0165] As shown in Table 3, the first layer of the stacked ensemble model consists of three or four base models. They are selected from SVR, MLP, Catboost, XGBoost, ElasticnetCV, and KNN. The parameters of each of these regression models are as follows:

[0166] For SVR, a Gaussian kernel function with a control coefficient of 0.4 for the number of support vectors is used. The tolerance stop criterion is set to 0.0001, while the error term penalty parameter is 0.2. The specified kernel cache size is 1999, and the maximum iteration count is set to 38000.

[0167] For Catboos, the iteration count is set to 600, and the learning rate is 0.2. The maximum depth of each tree is 8, and the root mean square error (RMSE) loss function is selected. The bagging positive coefficient is 0.2, and the overfitting detection type is set to 'Iter' while the overfitting detection waiting time is set to 100.

[0168] For XGBoost, the learning rate is set to 0.2, while the maximum depth of each tree is set to 6. The number of learners is 150. The 'gbtree' tree model is used as the base estimator, and the penalty coefficient is set to 1.

[0169] For ElasticnetCV, the L1 norm is set to 0.5, and the path length is set to 1x10 -3 . For each L1 regularization path, 100 α values are tried. The maximum interaction count is set to 1000, and the tolerance stopping criterion is set to 0.00001.

[0170] For KNN, the number of classes is set to 17. In other words, the output value is calculated as the average of the nearest 17 values.

[0171] For the base model MLP, Figure 9E The complete architecture is shown. In one instance, one or more of the base models are multi-layer perceptron (MLP) regressors. Figure 9E Depicts the base model MLP neural network architecture 2600. It includes an input layer 2610, zero or more hidden layers, and an output layer 2680. In Figure 9EIn [the figure], a network with two hidden layers 2630 and 2650 is shown. Each of the input layer 2610, the hidden layers 2630 and 2650 has a plurality of nodes, while the output layer 2680 has one node, because this MLP model is configured as a regressor. In this instance, there are three nodes in the input layer 2610, and six nodes respectively in each of the hidden layers 2630 and 2650. In this basic model MLP, the outputs of the first hidden layer 2630 and the second hidden layer 2650 are added together at the summation point 2660. The nodes between corresponding layers are fully connected, which means that each node in the first hidden layer 2630 is connected to all the nodes in the input layer 2610. Similarly, each node in the second hidden layer 2650 is connected to all the nodes in the first hidden layer 2630; and the output node 2680 is fully connected to the output of the summation 2660. The labels 2620, 2640, and 2670 represent the corresponding fully connected networks. Each connection is associated with a trainable weight parameter. In addition to the input nodes, all hidden nodes and output nodes can also have trainable bias parameters. In operation, the data vector is first fed into the input layer, and then propagated through the hidden layers by calculating the weighted sum of the input and the connection weights and applying an activation function to produce the output of each node, until the output layer produces the final output. To train the MLP regressor, the output node value is then compared with the expected value, and then the error is backpropagated to the input layer. Then, the connection weights and bias parameters are adjusted to minimize the error. By repeating this process multiple times, the MLP regressor is trained to produce regression values with high accuracy.

[0172] Regarding the meta-model, a single-layer MLP as shown in Figure 9F or a two-layer MLP as shown in Figure 9G is selected. In Figure 9F , the input layer 2710 includes three input nodes. There is no hidden layer. The output node 2730 is connected to each input node via the connection 2720. In Figure 9G , both the input layer 2740 and the hidden layer 2760 have three nodes. All these nodes are fully connected to each other via the connection 2750. The single output node 2780 is also fully connected to the hidden node 2760 via the connection 2770. In addition, each hidden node and output node is also embedded with an activation function. Various activation functions can be used, including but not limited to the rectified linear unit (ReLU), leaky ReLU, Tanh (hyperbolic tangent), and Sigmoid function, etc.

[0173] According to the experimental results shown in Table 3, the stacked ensemble model that produces the highest accuracy includes a first layer of three base models selected from SVR, MLP, and Catboost; and a single-layer MLP regressor using the ReLU activation function as the meta-model. Based on the experimental dataset, this combination achieved the highest accuracy of 0.888.

[0174] It will be understood that the third subtask in the data preprocessing module - training data preparation only needs to generate a data subset for training the base models, model weights, and the meta-model. Once these models are being trained and the DMLF module 2400 is put into actual use, this subtask does not need to be executed, so it can be skipped. Example 3 Example blood glucose monitoring system and apparatus

[0175] Now refer to Figure 10A , which shows a schematic diagram of another example blood glucose monitoring system 3000, the system includes another example non-invasive continuous glucose monitor (niCGM) device 3100 and a server 3400. Similar to Example 2, the device 3100 includes a housing 3500, a transmitter 3101, a light detector 3102, and a signal processing PCBA 3200 disposed within the housing 3500. In this example, the PCBA 3200 includes embedded software 3300, which is indirectly connected to a remote server 3400 including a DMLF module 3410 via wireless communication.

[0176] Now refer to Figure 10B , the flowchart shows the steps of an example operation when executing the embedded software 3300. The embedded software 3300 includes an operation module that controls the operation of the device 3100. In step 3310, the embedded software 3300 controls the timing of each NIR LED of the transmitter 3101 to emit signals with wavelengths of 940nm, 1350nm, and 1500nm on the target. In some examples, the NIR LEDs of the transmitter 3101 are turned on / off one after another in 3 different time slots to capture the absorption factor or reflection factor of each wavelength.

[0177] In step 3320, the embedded software 3300 controls the reception timing of the light detector 3102 to receive the reflected signal from the target. In some examples, the reflected signal from the target is 940nm, 1350nm, and 1500nm.

[0178] In step 3330, the NIR signal is fed to the MCU 3210 for processing.

[0179] In step 3340, the embedded software 3300 processes the optical signal group into a digital data vector stream and outputs it to a data processing system having a deep meta-learning framework module in a server via wireless communication (e.g., via WiFi).

[0180] In step 3350, the digital data vector stream is processed by the embedded DMLF module 3410 in the backend server.

[0181] In step 3360, the data analysis results are pushed from the server and displayed to a display element, for example, on a wearable user interface (UI). Additionally or alternatively, the data analysis results can be reported to other parties, such as a hospital or a caregiver, via wireless communication. Example 4 Example blood glucose monitoring system and apparatus

[0182] Reference now Figure 11A , shows a schematic diagram of another example blood glucose monitoring system 4000, the system includes another example non-invasive continuous glucose monitor (niCGM) device 4100, a mobile device 4600 including a console APP 4610, and a remote server 4400 including a DMLF module 4410. Similar to the previous example, the device 4000 includes a housing 4500, a transmitter 4101, a light detector 4102, and a signal processing PCBA 4200 disposed in the housing 4500, similar to the device 2100 described in Example 2. In this example, the PCBA includes embedded software 4300, which is indirectly connected to the console APP 6410 in the mobile device 4600 via wireless communication (e.g., Bluetooth). The mobile device 4610 is further connected to the DMLF 4410 of the server 4400 via wireless communication (e.g., WiFi). The mobile device 4610 can be a mobile phone, a laptop computer, or replaced by a desktop computer.

[0183] Reference now Figure 11B , the flowchart shows the steps of an example operation when executing the embedded software 4300. The embedded software 4300 includes an operation module for controlling the operation of the device 4100. In step 4310, the embedded software 4300 controls the timing of each NIR LED of the transmitter 4101 to emit light signals with wavelengths of 940nm, 1350nm and 1500nm on the target surface. In some examples, the NIR LEDs of the transmitter 3101 are turned on / off one by one in sequence at different time slots to capture the absorption factor or reflection factor of each wavelength.

[0184] In step 4320, the embedded software 4300 controls the light detector 4102 to receive the reflection signal from the target surface at a reception timing. In some instances, the reflected light signal from the target surface is about 940 nm, about 1350 nm, and about 1500 nm.

[0185] In step 4330, the NIR light signal is fed into the MCU 4210 for processing into a digital data vector stream in the PCBA 4200 and outputting it to the console APP 4610 in the mobile device 4600 via wireless communication (e.g., Bluetooth).

[0186] In step 4340, the embedded software 4300 in the PCBA 4200 processes the optical signal group into a digital data vector stream and outputs it to the mobile device 4600 (e.g., the console APP 4610) via wireless communication (e.g., via WiFi).

[0187] In step 4350, the digital data vector stream received by the mobile device 4600 (e.g., the console APP 4610) is sent to the server 4400 via wireless communication.

[0188] In step 4360, the digital data vector stream is analyzed by the embedded DMLF module 4410 in the backend server 4400.

[0189] In step 4370, the data analysis result from the server is pushed to and displayed by a display element, e.g., displayed on a wearable user interface (UI) and / or a mobile device. Additionally or alternatively, the data analysis result can be reported to other parties, such as a hospital or a caregiver, via wireless communication.

[0190] Now refer to Figure 11C , the flowchart shows the steps of an example overall operation where the console APP 4610 of the blood glucose monitoring system continuously or periodically provides updated blood glucose levels.

[0191] In step 4611, the device 4100 is paired with the console APP 4610 via wireless communication (e.g., Bluetooth).

[0192] In step 4612, the paired mobile device 4600 receives the vector data and its corresponding timestamp from the device 4100. In one embodiment, the data vector is processed by the device 4100 before being sent to the console APP 4610.

[0193] In step 4613, the received vector data and its corresponding timestamp are transmitted to the server 4400 via wireless communication (e.g., WiFi).

[0194] In step 4614, the transmitted vector data (including timestamp information) is analyzed by the DMLF module 4410 in the server 4400 to generate an output data result of the blood glucose level.

[0195] In step 4615, based on the output data result, the display element (such as the UI in the device or console APP) is updated to display the output data result of the blood glucose level. In one embodiment, the UI in the device 4100 is used to display the latest output data result of the blood glucose level. Example 5 Apparatus and system operation procedure

[0196] In this example, the general procedures for operating a blood glucose monitoring device and its system as described in the previous examples 1 - 4 are described below.

[0197] To measure the user's blood glucose level, the blood glucose monitoring device is placed on the user's target surface (e.g., forearm), where the NIR emitter and the light detector face the target surface. To obtain better results, ensure that the NIR emitter and the light detector are in good contact with the target surface (e.g., firmly adhered to it).

[0198] Turn on the device so that the NIR emitter emits a light signal onto the target surface, and the reflected light signal is received by the light detector. In some examples, multiple light signals (or signal streams) are emitted one after another in sequence so that multiple reflected light signals are received. In some examples, the device is turned on to capture data for about 2 minutes for data collection. In some examples, the device sends a warning alert when the data (reflected light signals) collection is completed. In some examples, certain data can be selected manually or automatically for subsequent analysis. For example, certain data can be selected after the readings are stable.

[0199] The reflected light signal is processed into a digital data vector in the operation module in the controller. In some examples, the reflected light signal is received periodically or continuously and processed into a data vector stream. Then, the data vector (or data vector stream) is sent directly to the built - in data processing system in the controller, or indirectly sent to a computer or a backend server with a data processing system for data processing. The data processing system includes a trained machine learning module (e.g., DMLF module) to analyze the data vector into an output data as the user's blood glucose level. In some examples, the data vector is sent indirectly to the backend server via a mobile device such as through wireless communication.

[0200] After data analysis, the output data can be sent directly or indirectly to a mobile device, a device, or other display devices to report the user's blood glucose level.

[0201] Exemplary embodiments of the present invention have thus been fully described. Although the description refers to specific embodiments, those skilled in the art will appreciate that the present invention can be implemented with variations of these specific details. Therefore, the present invention should not be construed as limited to the embodiments set forth herein.

[0202] The devices / methods / methods discussed in different figures can be added to the devices / methods / methods in other figures or exchanged with the devices / methods / methods in other figures. Further, data values of specific numerical values (such as specific quantities, numbers, categories, etc.) or other specific information should be construed as illustrative for discussing example embodiments. Such specific information is not provided to limit the example embodiments.

[0203] For example, in certain embodiments, the light emitter is configured to emit three light signals (with wavelengths of approximately 940 nm, approximately 1350 nm, and approximately 1500 nm), but different numbers (e.g., one, two, three, four, five, six, seven, eight, nine, ten, or more) of light signals with different wavelengths, different emission orders, and / or different time intervals / frequencies can be used according to actual needs. For example, the light emitter can be configured to additionally emit one or more (such as a fourth, fifth, sixth, seventh, eighth, ninth, tenth, or more) light signals, each additional light signal having a wavelength selected from the range of approximately 400 nm - 2000 nm. For example, the wavelengths are 400 nm, 450 nm, 500 nm, 550 nm, 600 nm, 650 nm, 700 nm, 750 nm, 800 nm, 850 nm, 900 nm, 950 nm, 1000 nm, 1050 nm, 1100 nm, 1150 nm, 1200 nm, 1250 nm, 1300 nm, 1350 nm, 1400 nm, 1450 nm, 1500 nm, 1550 nm, 1600 nm, 1650 nm, 1700 nm, 1750 nm, 1800 nm, 1850 nm, 1900 nm, 1950 nm, or 2000 nm.

[0204] For example, in certain embodiments, WiFi and Bluetooth are used for wireless communication between different components, but other wireless communication means such as infrared, zig-bee, near-field communication, etc. can alternatively be used.

[0205] For example, in certain embodiments, the battery management system includes a battery (such as a rechargeable battery) and a charger, but it can include a connector (such as a USB connector) to alternatively connect externally to an external power source (e.g., a DC power source). Numbered embodiments Group 1

[0206] Example 1. A glucose monitoring system, comprising: a wearable device capable of measuring a user's blood glucose level, the device comprising: a housing, an optical sensor and a processing unit, the optical sensor comprising a circuit including a near-infrared light-emitting diode and a receiver chip and configured to generate a voltage signal, the processing unit being configured to convert the analog voltage signal into a digital voltage signal; and computer software, the computer software comprising an algorithm for generating a trained neural network model capable of predicting the user's blood glucose level in real time based on the voltage signal received from the processing unit, wherein the system is configured to measure the user's blood glucose level in a non-invasive manner in real time.

[0207] Example 2. The glucose monitoring system according to any one of the preceding examples, wherein the optical sensor is connected to the processing unit through the circuit.

[0208] Example 3. The glucose monitoring system according to any one of the preceding examples, wherein the trained neural network model is part of the processing unit.

[0209] Example 4. The glucose monitoring system according to any one of the preceding examples, wherein the trained neural network model is stored in a device or network separate from the device. Group 2

[0210] Example 1: A glucose monitoring system, comprising: a wearable device capable of measuring a user's blood glucose level, the device comprising: a housing, an optical sensor and a processing unit, the optical sensor comprising a circuit including a near-infrared light-emitting diode and a receiver chip and configured to generate a voltage signal, the processing unit being configured to convert the analog voltage signal into a digital voltage signal; and computer software, the computer software comprising an algorithm for generating a trained neural network model capable of predicting the user's blood glucose level in real time based on the voltage signal received from the processing unit, wherein the system is configured to measure the user's blood glucose level in a non-invasive manner in real time, wherein the trained neural network model includes a trained non-linear model and a linear model to perform the following steps: Generating class prediction probability values and numerical values by inputting the voltage signal into the trained non-linear model and the linear model respectively; Classifying the class prediction probability values and the numerical values as low, normal or high; Comparing the classification results to determine whether the values are consistent; and If the values are consistent, determining the output blood glucose status and blood glucose value.

[0211] Example 2. The glucose monitoring system according to any one of the foregoing embodiments, wherein the optical sensor is connected to the processing unit through the circuit.

[0212] Example 3. The glucose monitoring system according to any one of the foregoing embodiments, wherein the trained neural network model is part of the processing unit.

[0213] Example 4. The glucose monitoring system according to any one of the foregoing embodiments, wherein the trained neural network model is stored in a device or network separate from the device. Group 3

[0214] Example 1. A device for monitoring a user's blood glucose, comprising: (a) a light emitter configured to emit a light signal directed at a target surface of the user to generate a reflected light signal reflected from the target surface; (b) a light receiver configured to receive the reflected light signal; (c) a controller configured to be operatively connected to the light emitter and the light receiver; and (d) a housing configured to house the light emitter, the light receiver, and the controller, wherein the light signal includes a first light signal having a first wavelength of about 940 nm, a second light signal having a second wavelength of about 1350 nm, and / or a third light signal having a third wavelength of about 1500 nm, wherein the controller includes an operation module that controls the operation of the device and converts the reflected light signal into digital data, and wherein the controller further includes a data processing system that processes the digital data or is operatively connected to the data processing system, wherein the data processing system includes a machine learning module that analyzes the data signal to generate output data as the user's blood glucose level.

[0215] Example 2. The device according to any one of the foregoing embodiments, wherein the light emitter includes three near-infrared LEDs having the first wavelength, the second wavelength, and the third wavelength, respectively.

[0216] Example 3. The device according to any one of the foregoing embodiments, wherein the device further includes an NTC thermometer to obtain the ambient temperature and / or the user's body temperature.

[0217] Example 4. The apparatus according to any one of the foregoing embodiments, wherein the controller is further configured to control the optical transmitter to turn on and off to sequentially emit the first optical signal, the second optical signal, and / or the third optical signal one by one in a plurality of cycles, so that a plurality of optical signal groups are formed at a defined time interval, and each optical signal group includes a first reflected optical signal, a second reflected optical signal, and / or a third reflected optical signal obtained in each cycle.

[0218] Example 5. The apparatus according to any one of the foregoing embodiments, wherein the time interval is about 60 times per minute.

[0219] Example 6. The apparatus according to any one of the foregoing embodiments, wherein the controller includes a processor unit coupled to a memory storing an executable software program, and the software program includes an operation module for controlling the operation of the apparatus. The operating system performs the following steps: a) controlling the optical transmitter to turn on and off to time the emission of the optical signal; b) controlling the optical receiver to receive the reflected optical signal to time the acquisition of a plurality of reflected optical groups; c) processing each reflected optical signal group into a digital data vector; and d) transmitting the data vector to a data processing system to analyze the data vector.

[0220] Example 7. The apparatus according to any one of the foregoing embodiments, wherein the machine learning module includes a deep meta-learning framework (DMLF) module, and the DMLF module processes the data vector obtained from the apparatus to generate output data.

[0221] Example 8. The apparatus according to any one of the foregoing embodiments, wherein the data processing system performs the following steps: a) obtaining the data vector from the apparatus; b) preprocessing the data vector to produce a processed data vector; and c) analyzing the processed data vector through a trained DMLF module to produce output data as the blood glucose level of the user.

[0222] Example 9. The apparatus according to any one of the foregoing embodiments, wherein step b) includes the following steps: b1) cleaning the data vector to remove any outliers.

[0223] Example 10. The apparatus according to any one of the foregoing embodiments, wherein step b1) is performed by using Isolation Forest (iForest), one-class SVM, and LOF, and combinations thereof.

[0224] Example 11. The apparatus according to any one of the foregoing embodiments, wherein step b1) is performed by a combination of Isolation Forest and one-class SVM.

[0225] Example 12. The device according to any one of the foregoing examples, wherein the step b) comprises the following steps: b2) combining the data vectors into representative data by the following equation: representative data = (y + z) / 2, where y is the mean of the data vectors and z is the median of the data vectors.

[0226] Example 13. The device according to any one of the foregoing examples, wherein the DMLF module uses a deep meta-learning framework to analyze the data vectors, the deep meta-learning framework comprising a first hierarchical structure layer and a second hierarchical structure layer, the first hierarchical structure layer comprising a plurality of base models, and the second hierarchical structure layer comprising a meta-model, wherein each base model is configured to receive the processed data vectors and generate intermediate data points, and the meta-model is configured to receive weighted values of the intermediate data points to generate the output data.

[0227] Example 14. The device according to any one of the foregoing examples, wherein the base models are selected from the group consisting of: RNN, LSTM, CNN, support vector regression (SVR), MLP, KNN, ElasticNetCV, Catboost, XGBoost, gradient boosting regressor, LGBM regressor, bagging regressor, decision tree, XGBoost, and combinations thereof.

[0228] Example 15. The device according to any one of the foregoing examples, wherein the meta-model is selected from MLP, RNN, CNN, and any combination thereof.

[0229] Example 16. The device according to any one of the foregoing examples, wherein the weighted values of the intermediate data points are calculated by multiplying the intermediate data points by model weights, wherein each model weight is configured to be a same value, a random value, or an optimized value obtained by a learning algorithm.

[0230] Example 17. The device according to any one of the foregoing examples, wherein the first hierarchical structure layer comprises base models of SVR, MLP, and Catboost, and the meta-model is configured to be a single-layer MLP using ReLU as an activation function.

[0231] Example 18. The device according to any one of the foregoing examples, wherein the DMLF module is pre-trained by a set of training data using a bootstrap sampling and sampling with replacement method.

[0232] Example 19. A device for blood glucose monitoring of a user, comprising: a) a light emitter having a plurality of near-infrared LEDs, each configured to emit an optical signal directed to a target surface of the user to generate a reflected optical signal reflected from the target surface, wherein the optical signal includes a first optical signal having a wavelength of about 940 nm, a second optical signal having a wavelength of about 1350 nm, and a third optical signal having a wavelength of about 1500 nm; b) a light receiver having a photodetector configured to receive the reflected optical signal; and c) a controller configured to be operatively connected to the light emitter and the light receiver; and d) a housing configured to accommodate the light emitter, the light receiver, and the controller, wherein the controller is configured to control the light emitter to turn on and off to sequentially emit the first optical signal, the second optical signal, and the third optical signal one by one, such that a plurality of reflected light groups are formed at defined time intervals, each reflected light group including first reflected light, second reflected light, and third reflected light obtained in each cycle, wherein the controller includes a processor unit coupled to a memory storing an executable software program, the software program including an operation module that controls the operation of the device and converts each reflected light group into a digital data vector, and wherein the controller further includes a data processing system for processing the data vector or is operatively connected to the data processing system, wherein the data processing system includes a machine learning module that analyzes the data vector to generate output data as the blood glucose level of the user.

[0233] Example 20. The device according to any one of the foregoing examples, wherein the software program includes an operation module for controlling the operation of the device, and wherein the operation module performs the following steps: a) controlling the timing of turning on and off the light emitter to emit the first light, the second light, and the third light; b) controlling the timing of the light receiver to receive the first reflected light, the second reflected light, and the third reflected light reflected from the target surface to generate a plurality of reflected light groups; c) processing each reflected light group into a digital data vector; and d) transmitting the data vector to a data processing system including a neural network to process the data vector.

[0234] Example 21. The device according to any one of the foregoing examples, wherein the machine learning module includes a deep meta-learning framework module, the deep meta-learning framework module includes a first hierarchical structure layer and a second hierarchical structure layer, the first hierarchical structure layer includes a plurality of basic modules, and the second hierarchical structure layer includes a meta-learning module. Each basic module is configured to receive a processed data vector and generate an intermediate data point, and the meta-learning module is configured to receive a weighted value of the intermediate data point to generate the output data. The first hierarchical structure layer includes basic modules such as SVR, MLP, and Catboost, and the second hierarchical structure layer includes meta-learning modules such as ReLU.

[0235] Example 22. A system for blood glucose monitoring of a user, comprising: (a) the device according to any one of the foregoing examples; and (b) a server electrically connected to the device.

[0236] Example 23. The system according to any one of the foregoing examples, wherein the server includes a server processor unit coupled to a server memory storing an executable server software program, and the server software program includes a data processing system that processes a data vector obtained from the device to calculate the blood glucose level of the user. The data processing system includes a neural network.

[0237] Example 24. The system according to any one of the foregoing examples, wherein the data processing system performs the following steps: a) obtaining the data vector obtained from the device; b) preprocessing the data vector; and c) analyzing the data vector through a trained neural network model to generate output data, thereby obtaining the blood glucose level of the user.

[0238] Example 25. The system according to any one of the foregoing examples, wherein the step b) includes the following steps: b1) cleaning the data vector to remove any outliers.

[0239] Example 26. The system according to any one of the foregoing examples, wherein the step b1) is performed by using isolation forest (iForest), one-class SVM, and LOF and combinations thereof.

[0240] Example 27. The system according to any one of the foregoing examples, wherein the step b1) is performed by a combination of isolation forest and one-class SVM.

[0241] Example 28. The system according to any one of the foregoing examples, wherein step b) comprises the following steps: b2) combining the vector data into representative data by the following equation: representative data = (a + b) / 2, where x is the average value of the vector data and y is the median value of the vector data.

[0242] Example 29. The system according to any one of the foregoing examples or claim 23, wherein the neural network is a deep meta-learning framework, the deep meta-learning framework includes a first hierarchical structure layer and a second hierarchical structure layer, the first hierarchical structure layer includes a plurality of basic modules, the second hierarchical structure layer includes a meta-learning module, each basic module is configured to receive a data vector and generate an intermediate data vector, and the meta-learning module is configured to receive the intermediate data to generate the output data.

[0243] Example 30. The system according to any one of the foregoing examples, wherein each basic module is a machine learning module or a deep learning model.

[0244] Example 31. The system according to any one of the foregoing examples, wherein the basic modules are selected from the group consisting of: RNN, LSTM, CNN, support vector regression (SVR), MLP, KNN, ElasticNetCV, Catboost, XGBoost, gradient boosting regressor, LGBM regressor, bagging regressor, decision tree, XGBoost, stacking, and combinations thereof; and the meta-learning modules are selected from ReLU, leaky ReLU, Tanh, Sigmoid, and combinations thereof.

[0245] Example 32. The system according to any one of the foregoing examples, wherein each basic module includes model weights, and each model weight is configured to be a same value, a random value, or optimized using the nn.Parameter function in Pytorch.

[0246] Example 33. The system according to any one of the foregoing examples, wherein the first hierarchical structure layer includes basic modules such as SVR, MLP, and Catboost, and the second hierarchical structure layer includes meta-learning modules such as ReLU.

[0247] Example 34. The system according to any one of the foregoing examples, wherein the neural network is pre-trained using bootstrap sampling through training data.

[0248] Example 35. The system according to any one of the foregoing examples further includes a mobile device electrically connected between the device and the server, configured to receive a data vector obtained from the device, transmit the data vector to the server, and optionally display the blood glucose level.

[0249] Example 36. A glucose monitoring system includes: a wearable device capable of measuring a user's blood glucose level, the device including: a housing, an optical sensor, and a processing unit, the optical sensor including a circuit comprising a near-infrared light-emitting diode and a receiver chip and configured to generate a voltage signal, the processing unit configured to convert the analog voltage signal into a digital voltage signal; and computer software including an algorithm for generating a trained neural network model capable of predicting the user's blood glucose level in real time based on the voltage signal received from the processing unit, wherein the system is configured to measure the user's blood glucose level in a non-invasive manner in real time, wherein the trained neural network model includes a trained non-linear model and a linear model to perform the following steps: generating a class prediction probability value and a numerical value by inputting the voltage signal into the trained non-linear model and the linear model respectively; classifying the class prediction probability value and the numerical value as low, normal, or high; comparing the classification results to determine whether the values are consistent; and if the values are consistent, determining the output blood glucose status and blood glucose value.

[0250] Example 37. The glucose monitoring system according to any one of the foregoing examples, wherein the optical sensor is connected to the processing unit through the circuit.

[0251] Example 38. The glucose monitoring system according to any one of the foregoing examples, wherein the trained neural network model is part of the processing unit.

[0252] Example 39. The glucose monitoring system according to any one of the foregoing examples, wherein the trained neural network model is stored in a device or network separate from the device.

[0253] Example 40. A method for monitoring blood glucose level, the method includes the following steps: (i) obtaining first reflected light, second reflected light, and third reflected light from the device according to any one of the foregoing examples or the system according to any one of the foregoing examples; and (ii) calculating the blood glucose level based on the first reflected light, the second reflected light, and the third reflected light.

[0254] Example 41. The method according to any one of the foregoing examples, further including the following step before step (ii): processing the first reflected light, the second reflected light, and the third reflected light.

[0255] Example 42. A method for processing data from a device or system for blood glucose monitoring, the method comprising the steps of: a) obtaining a data vector from a device as described in any of the foregoing examples or a system as described in any of the foregoing examples; b) preprocessing the data vector; and c) analyzing the data vector by a trained machine learning module to generate output data, thereby obtaining the blood glucose level of a user.

Claims

1. A device for blood glucose monitoring of a user, the device comprising: a) A light emitter configured to emit an optical signal directed at a target surface of the user to generate a reflected optical signal reflected from the target surface; b) A light receiver configured to receive the reflected optical signal; c) A controller configured to be operatively connected to the light emitter and the light receiver; and d) A housing configured to house the light emitter, the light receiver, and the controller, wherein the optical signal includes a first optical signal having a first wavelength of about 940 nm, a second optical signal having a second wavelength of about 1350 nm, and / or a third optical signal having a third wavelength of about 1500 nm, wherein the controller includes an operation module that controls the operation of the device and converts the reflected optical signal into digital data, and wherein the controller further includes or is operatively connected to a data processing system that processes the digital data, wherein the data processing system includes a machine learning module that analyzes the data signal to generate output data as the blood glucose level of the user.

2. The device according to any one of the preceding claims, wherein, The light emitter includes three near-infrared LEDs, each having the first wavelength, the second wavelength, and the third wavelength, respectively.

3. The device according to any one of the preceding claims, wherein, The device further includes an NTC thermometer to obtain the ambient temperature and / or the body temperature of the user.

4. The device according to any one of the preceding claims, wherein, The controller is further configured to control the light emitter to turn on and off to sequentially emit the first optical signal, the second optical signal, and / or the third optical signal one after another in a plurality of cycles, such that a plurality of optical signal groups are formed at a defined time interval, and each optical signal group includes a first reflected optical signal, a second reflected optical signal, and / or a third reflected optical signal obtained in each cycle.

5. The device according to claim 4, wherein, The time interval is about 60 times per minute.

6. The device according to claim 4 or claim 5, wherein The controller includes a processor unit coupled to a memory storing an executable software program, the software program including an operation module that controls the operation of the device, wherein the operating system performs the following steps: a) Controlling the light emitter to turn on and off for timing of emitting the optical signal; b) Controlling the light receiver to receive the reflected optical signal for timing of obtaining a plurality of reflected light groups; c) Processing each reflected optical signal group into a digital data vector; and d) Transmitting the data vector to the data processing system to analyze the data vector.

7. The device according to any one of the preceding claims, wherein, The machine learning module includes a deep meta-learning framework (DMLF) module that processes the data vector obtained from the device to generate output data.

8. The device according to any one of the preceding claims, wherein, The data processing system performs the following steps: a) Obtaining the data vector from the device; b) Preprocessing the data vector to produce a processed data vector; and c) Analyzing the processed data vector by a trained DMLF module to produce output data as the blood glucose level of the user.

9. The device according to claim 8, wherein, The step b) includes the following steps: b1) Cleaning the data vector to remove any outliers.

10. The device according to claim 9, wherein, The step b1) is performed by using Isolation Forest (iForest), One-Class SVM, LOF, and combinations thereof.

11. The device according to claim 9, wherein, The step b1) is performed by a combination of Isolation Forest and One-Class SVM.

12. The apparatus according to claim 8, wherein, The step b) includes the following steps: b2) Combine the data vectors into representative data by the following equation: Representative data = (y + z) / 2, where y is the mean of the data vectors and z is the median of the data vectors.

13. The device according to claim 7 or claim 8, wherein, The DMLF module employs a deep meta-learning framework to analyze the data vectors. The deep meta-learning framework includes a first hierarchical structure layer and a second hierarchical structure layer. The first hierarchical structure layer includes a plurality of base models, and the second hierarchical structure layer includes a meta-model. Each base model is configured to receive the processed data vectors and generate intermediate data points, and the meta-model is configured to receive the weighted values of the intermediate data points to generate the output data.

14. The device according to claim 13, wherein, The base models are selected from the group consisting of: RNN, LSTM, CNN, Support Vector Regression (SVR), MLP, KNN, ElasticNetCV, Catboost, XGBoost, Gradient Boosting Regressor, LGBM Regressor, Bagging Regressor, Decision Tree, XGBoost, and combinations thereof.

15. The apparatus according to claim 13, wherein, The meta-model is selected from MLP, RNN, CNN, and any combination thereof.

16. The apparatus according to claim 13, wherein, The weighted values of the intermediate data points are calculated by multiplying the intermediate data points by model weights, where each model weight is configured to be the same value, a random value, or an optimized value obtained by a learning algorithm.

17. The apparatus according to claim 13, wherein, The first hierarchical structure layer includes base models such as SVR, MLP, and Catboost, and the meta-model is configured to be a single-layer MLP using ReLU as the activation function.

18. The apparatus according to claim 7, wherein, The DMLF module is pre-trained by using bootstrap sampling and sampling with replacement methods through a set of training data.

19. A device for blood glucose monitoring for a user, the device comprising: a) A light emitter having a plurality of near-infrared LEDs, each configured to emit an optical signal directed at a target surface of the user to generate a reflected optical signal reflected from the target surface, wherein the optical signal includes a first optical signal having a wavelength of about 940 nm, a second optical signal having a wavelength of about 1350 nm, and a third optical signal having a wavelength of about 1500 nm; b) A light receiver having a photodetector configured to receive the reflected optical signal; and c) A controller configured to be operatively connected to the light emitter and the light receiver; and d) A housing configured to accommodate the light emitter, the light receiver, and the controller, wherein the controller is configured to control the light emitter to turn on and off to sequentially emit the first optical signal, the second optical signal, and the third optical signal one by one, such that a plurality of reflected light groups are formed at a defined time interval, and each reflected light group includes a first reflected light, a second reflected light, and a third reflected light obtained in each cycle. Wherein, the controller includes a processor unit coupled to a memory storing an executable software program, the software program includes an operation module, the operation module controls the operation of the device and converts each reflected light group into a digital data vector, and Wherein, the controller further includes a data processing system for processing the data vector or is operatively connected to the data processing system, wherein the data processing system includes a machine learning module, and the machine learning module analyzes the data vector to generate output data as the blood glucose level of the user.

20. The device according to claim 19, wherein, The software program includes an operation module for controlling the operation of the device, wherein the operation module performs the following steps: a) Controlling the timing of turning on and off the light emitter to emit the first light, the second light, and the third light; b) Controlling the light receiver to receive the first reflected light, the second reflected light, and the third reflected light reflected from the target surface to generate a plurality of reflected light groups at a timing; c) Processing each reflected light group into a digital data vector; and d) Transmitting the data vector to a data processing system including a neural network to process the data vector.

21. The device according to claim 19, wherein, The machine learning module includes a deep meta-learning framework module, the deep meta-learning framework module includes a first hierarchical structure layer and a second hierarchical structure layer, the first hierarchical structure layer includes a plurality of basic modules, the second hierarchical structure layer includes a meta-learning module, wherein each basic module is configured to receive the processed data vector and generate an intermediate data point, and the meta-learning module is configured to receive the weighted value of the intermediate data point to generate the output data, wherein the first hierarchical structure layer includes basic modules such as SVR, MLP, and Catboost, and the second hierarchical structure layer includes meta-learning modules such as ReLU.

22. A system for blood glucose monitoring of a user, the system comprising: a) A device as described in any one of the preceding claims; And b) A server, the server being electrically connected to the device.

23. The system according to claim 22, wherein The server includes a server processor unit coupled to a server memory storing an executable server software program, the server software program includes a data processing system, and the data processing system processes the data vector obtained from the device to calculate the blood glucose level of the user, wherein the data processing system includes a neural network.

24. The system according to claim 22, wherein The data processing system performs the following steps: a) Obtaining the data vector obtained from the device; b) Preprocessing the data vector; And c) Analyzing the data vector through a trained neural network model to generate output data, thereby obtaining the blood glucose level of the user.

25. The system according to claim 24, wherein, The step b) includes the following steps: b1) Cleaning the data vector to remove any outliers.

26. The system according to claim 25, wherein, The step b1) is performed by using isolation forest (iForest), one-class SVM, and LOF and combinations thereof.

27. The system according to claim 26, wherein, The step b1) is performed by a combination of isolation forest and one-class SVM.

28. The system according to any one of claims 24 to 27, wherein, The step b) includes the following steps: b2) Merging the vector data into representative data through the following equation: Representation data = (a + b) / 2, where x is the mean value of the vector data, and y is the median value of the vector data.

29. The system according to claim 23 or claim 24, wherein The neural network is a deep meta - learning framework. The deep meta - learning framework includes a first - level structure layer and a second - level structure layer. The first - level structure layer includes a plurality of basic modules, and the second - level structure layer includes a meta - learning module. Each basic module is configured to receive a data vector and generate an intermediate data vector, and the meta - learning module is configured to receive the intermediate data to generate the output data.

30. The system according to claim 29, wherein, Each of the basic modules is a machine - learning module or a deep - learning model.

31. The system according to claim 30, wherein, The basic modules are selected from the group consisting of: RNN, LSTM, CNN, Support Vector Regression (SVR), MLP, KNN, ElasticNetCV, Catboost, XGBoost, Gradient Boosting Regressor, LGBM Regressor, Bagging Regressor, Decision Tree, XGBoost, Stacking, and combinations thereof; and the meta - learning module is selected from ReLU, Leaky ReLU, Tanh, Sigmoid, and combinations thereof.

32. The system according to claim 29, wherein, Each basic module includes model weights, and each model weight is configured to be a same value, a random value, or optimized using the nn.Parameter function in Pytorch.

33. The system according to claim 29, wherein, The first - level structure layer includes basic modules such as SVR, MLP, and Catboost, and the second - level structure layer includes meta - learning modules such as ReLU.

34. The system according to claim 29, wherein, The neural network is pre - trained using bootstrap sampling with training data.

35. The system according to claim 22, further comprising a mobile device, the mobile device being electrically connected between the device and the server, configured to receive a data vector obtained from the device, transmit the data vector to the server, and optionally display the blood - glucose level.

36. A glucose monitoring system, comprising: A wearable device, the wearable device being capable of measuring a user's blood - glucose level, the device comprising: A housing, an optical sensor, and a processing unit. The optical sensor includes a circuit comprising a near - infrared light - emitting diode and a receiver chip and is configured to generate a voltage signal. The processing unit is configured to convert the analog voltage signal into a digital voltage signal; and computer software, the computer software including an algorithm for generating a trained neural - network model capable of predicting the user's blood - glucose level in real time based on the voltage signal received from the processing unit. Wherein, the system is configured to measure the user's blood - glucose level in a non - invasive manner in real time. Wherein, the trained neural - network model includes a trained non - linear model and a linear model to perform the following steps: Generating a class - prediction probability value and a numerical value by inputting the voltage signal into the trained non - linear model and the trained linear model respectively; Classifying the class - prediction probability value and the numerical value as low, normal, or high; Comparing the classification results to determine whether the values are consistent; and If the values are consistent, determining the output blood - glucose status and blood - glucose value.

37. The glucose monitoring system according to claim 36, wherein, The optical sensor is connected to the processing unit through the circuit.

38. The glucose monitoring system according to claim 36 or claim 37, wherein The trained neural network model is part of the processing unit.

39. The glucose monitoring system according to any one of claims 36 to 37, wherein, The trained neural network model is stored in a device or network separate from the device.

40. A method for monitoring a blood glucose level, the method comprising the steps of: (i) obtaining first reflected light, second reflected light, and third reflected light from the device according to any one of claims 0 to 21 or the system according to any one of claims 22 to 39; and (ii) calculating the blood glucose level based on the first reflected light, the second reflected light, and the third reflected light.

41. The method according to claim 40, further comprising the following steps before step (ii): processing the first reflected light, the second reflected light, and the third reflected light.

42. A method for processing data from a device or system for blood glucose monitoring, the method comprising the steps of: a) obtaining a data vector from the device according to any one of claims 0 to 21 or the system according to any one of claims 22 to 39; b) preprocessing the data vector; and c) analyzing the data vector by a trained machine learning module to generate output data, thereby obtaining the blood glucose level of the user.

Images