Diabetes management hearing aid system and diabetes monitoring method based thereon

Abstract

The embodiment of the present invention discloses a diabetes management hearing aid system and a diabetes monitoring method based thereon, and relates to the field of hearing aid devices and algorithm technology. The system includes: a hearing aid; a flexible sweat sensor patch, including a sweat collection hole, an electrochemical electrode, and a microchannel; the outer side of the flexible sweat sensor patch is used to adhere to the skin behind the ear, and the inner side is adhered to the outer surface of the hearing aid; the sweat of the skin behind the ear flows into the microchannel through the sweat collection hole, and the sweat in the microchannel undergoes an electrochemical reaction under the action of the electrochemical electrode, generating an electrical signal that is transmitted to the hearing aid chip, and the hearing aid chip uses an internal integrated analysis algorithm to detect blood sugar concentration; a disease monitoring system is used to record detection data, analyze diabetes risks, and link the hearing aid to perform an in-ear warning when the data is abnormal. This implementation is non-invasive, portable, and can solve the problems of user hearing and diabetes management at the same time.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of hearing aid devices and algorithm technology, and in particular to a diabetes management hearing aid system and a diabetes monitoring method based thereon. Background Art

[0002] With the development of society and the advancement of aging, health monitoring has become a focus of social attention, among which diabetes has become a major problem for middle-aged and elderly people. Diabetes is a group of metabolic diseases characterized by chronic hyperglycemia caused by multiple causes. Diabetes can cause a variety of complications. If it is not properly controlled, it may even cause amputation, uremia, etc. Diabetes can be effectively controlled through blood sugar testing.

[0003] The existing blood sugar detection methods have the following problems:

[0004] 1. Traditional blood sugar testing mostly involves acupuncture to draw blood. Long-term invasive testing increases the user's physical pain and easily creates a resistance to the test.

[0005] 2. Existing blood sugar testing devices are inconvenient to carry and cannot achieve real-time management of diabetes;

[0006] 3. Routine self-examinations lack records and analysis, making it difficult to independently judge the course of the disease. Summary of the invention

[0007] The embodiments of the present invention provide a diabetes management hearing aid system and a diabetes monitoring method based thereon to solve at least one of the above problems.

[0008] In a first aspect, an embodiment of the present invention provides a diabetes management hearing aid system, comprising:

[0009] Hearing aids, including hearing aid chips;

[0010] A flexible sweat sensor patch, comprising a sweat collection hole, an electrochemical electrode and a microchannel; wherein the outer side of the flexible sweat sensor patch is used to adhere to the skin behind the ear, and the inner side is adhered to the outer surface of the hearing aid, and is electrically connected to the hearing aid chip through a contact electrode; after the hearing aid is worn, the sweat of the skin behind the ear flows into the microchannel through the sweat collection hole, and the sweat in the microchannel undergoes an electrochemical reaction under the action of the electrochemical electrode, generating an electrical signal that is transmitted to the hearing aid chip, and the hearing aid chip uses an internal integrated analysis algorithm to detect blood glucose concentration;

[0011] The disease monitoring system is used to record test data, analyze diabetes risks, and link the hearing aid to provide in-ear warning when the data is abnormal.

[0012] The diabetes management hearing aid system combines the advantages of the portability of hearing aids and the non-invasiveness of sweat sensors, and can achieve the following beneficial effects:

[0013] 1. Traditional blood sugar testing is mostly done by needle pricking blood. Long-term invasive testing increases the user's physical pain and easily leads to psychological resistance. This embodiment estimates the human body's glucose level by detecting the glucose concentration in sweat, thereby assisting in diabetes management.

[0014] 2. As the existing blood sugar detection devices are inconvenient to carry, hearing aids are the preferred product to solve hearing problems. They have the advantages of miniaturization, portability and unified audience. Combining sweat sensor detection devices with hearing aids can achieve real-time management of diabetes;

[0015] 3. In view of the problem that conventional self-examination in existing technologies lacks records and analysis, and it is difficult to independently judge the direction of the disease, the real-time warning system equipped with the diabetes management hearing aid system can perform health management, record blood sugar concentration in real time, analyze the disease, assist in treatment, and provide real-time warnings. At the same time, with the help of hearing aids, the user's hearing condition is improved to solve the problem of users not being able to hear or not being able to hear clearly.

[0016] In a second aspect, an embodiment of the present invention provides a diabetes monitoring method based on a diabetes management hearing aid system, which is applied to the above disease monitoring system, and the method includes:

[0017] According to the sound signal obtained by the hearing aid and the preset user schedule, the current scene of the user is identified, and the proportion of the current scene in the future period of time and the next scene to be entered after the current scene ends are predicted; wherein the scenes of the user include sleeping, eating, working, exercising and others, and the future period of time covers at most two different scenes;

[0018] Calling a blood sugar prediction model corresponding to the current scenario, and predicting the first blood sugar time series data of the user in the future under the current scenario according to the proportion and the user's current blood sugar detection time series data;

[0019] Calling a blood sugar prediction model corresponding to the next scenario, and predicting the user's second blood sugar time series data in the next scenario in the future according to the proportion and the first blood sugar time series data;

[0020] The first blood glucose time series data and the second blood glucose time series data are compared with the blood glucose thresholds in the current scenario and the next scenario, respectively, and data anomalies are identified.

[0021] The diabetes monitoring method based on the diabetes management hearing aid system can achieve the following beneficial effects:

[0022] Based on the voice signals collected by the hearing aid and the user's daily schedule, the user's scenario and possible scenario transitions in the future are predicted. The future trend of glucose levels is predicted in different time periods based on the scenario transitions, and threshold comparisons are performed in different time periods to improve the accuracy of glucose predictions and diabetes risk warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 is a functional structure diagram of a diabetes management hearing aid system provided by an embodiment of the present invention;

[0025] Figure 2 It is a main structure diagram of a diabetes management hearing aid system provided by an embodiment of the present invention;

[0026] Figure 3 is a diagram of an installation method of a hearing aid and a sweat sensor provided by an embodiment of the present invention;

[0027] Figure 4 is a schematic structural diagram of a flexible sweat sensor patch provided by an embodiment of the present invention;

[0028] Figure 5 This is a schematic diagram of a method of stimulating sweating by iontophoresis provided by an embodiment of the present invention;

[0029] Figure 6 is a flow chart of a diabetes monitoring method based on a diabetes management hearing aid system provided by an embodiment of the present invention;

[0030] Figure 7 It is a data processing schematic diagram of a scene prediction model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0032] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0033] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0034] Reference Figure 1-Figure 5 The embodiment of the present invention provides a diabetes management hearing aid system. The system includes: a hearing aid, a flexible sweat sensor patch and a disease monitoring system.

[0035] The hearing aid includes at least one microphone, a hearing aid chip and a receiver. The microphone is a high-precision microphone, which is used to receive external sounds into the hearing aid and transmit them to the hearing aid chip. The hearing aid chip integrates a hearing aid algorithm and a blood glucose analysis algorithm. The hearing aid algorithm can optimize the sound and output it (for example, amplify, reduce, denoise, etc. the sound and then output it). The blood glucose analysis algorithm will be described in the flexible sweat sensor patch section. The receiver is used to output the sound into the user's ear to solve the hearing loss problem.

[0036] The flexible sweat sensor patch includes sweat collection holes, electrochemical electrodes (i.e. Figure 4 The outer side of the flexible sweat sensor patch is used to adhere to the skin behind the ear, and the inner side is attached to the outer surface of the hearing aid, and is electrically connected to the hearing aid chip of the hearing aid through the contact electrode. For example, Figure 2 , Figure 3 As shown, the inner side of the flexible sweat sensor patch is connected to the hearing aid host through contact electrodes for power supply and communication, and is fixed to the back of the hearing aid through adhesive. After the hearing aid is worn, the outer side of the flexible sweat sensor patch is attached to the skin behind the ear. Figure 4, the sweat from the skin behind the ear flows into the microchannel through the sweat collection hole, the sweat in the microchannel contacts the electrochemical electrode, and an electrochemical reaction occurs under the action of the electrochemical electrode, generating an electrical signal which is transmitted to the hearing aid chip, and the hearing aid chip uses an internally integrated blood glucose analysis algorithm to calculate the transmitted electrical signal to obtain the blood glucose concentration. At the same time, an exhaust port is provided at the end of the microchannel, and waste liquid is converted into gas through electrochemical reaction and heating in the microchannel, which is discharged through the exhaust port. Optionally, the electrochemical electrode can be a nickel-gold alloy electrode. In the blood glucose analysis algorithm, the glucose concentration in the sweat is first calculated by the electrochemical principle, and then the glucose concentration in the sweat is converted into blood glucose concentration according to the corresponding relationship. The specific process is a prior art and will not be repeated here.

[0037] Furthermore, the main body of the flexible sweat sensor patch is a microfluidic chip system. Microfluidic chip, also known as Lab-on-chip, is a scientific technology that is mainly characterized by the manipulation of fluids in micrometer-scale space. It has good biocompatibility, electrical insulation, heat dissipation, optical properties, etc. The surface has good modifiability. The sensor made based on this principle has the characteristics of high sensitivity, fast response speed and small size. The sensor of the present invention uses polydimethylsiloxane as the base material, depicts microchannels to collect sweat, and detects blood sugar concentration through electrical and chemical reactions. If sweat sensors are divided from a functional perspective, they can be divided into sweat monitoring system, sweat removal system, sweat collection system and sweat analysis system.

[0038] Sweat monitoring system: Since the microchannel substrate is made of polydimethylsiloxane with a rough surface, the different refractive indexes of visible light will cause the microchannel to change color in the empty state and the water-filled state, thereby detecting the sweating of the human body in real time and analyzing whether the detection state is met. When there is a detection demand but the human sweat flow is not enough to support the detection, active sweating will be started. Optional, such as Figure 5 As shown, iontophoresis can be used, where the electrochemical electrodes act as iontophoresis electrodes, and a voltage is applied between the iontophoresis electrodes to generate an electric current under the skin surface, thereby stimulating sweat secretion.

[0039] Sweat discharge system: The liquid in the microchannel undergoes electrochemical reaction and heating at the electrochemical electrode to generate gas, which is then discharged from the sensor through the discharge hole.

[0040] Sweat collection system: Sweat behind the ear enters the sensor through the unidirectional sweat collection hole and is absorbed by the microchannel.

[0041] Sweat analysis system: After being absorbed into the microchannel, sweat reacts with the alkaline solution. The reactants react with the electrochemical electrodes and generate a large number of electrical signals. The electrical signals are transmitted to the hearing aid chip for analysis, and the blood glucose concentration is calculated based on the changes in the level value.

[0042] The disease monitoring system can be used to record and analyze the measured data, and when the data is abnormal, the hearing aid will be linked to an in-ear alarm. The disease monitoring system can be deployed on a mobile phone or other electronic devices. In a specific embodiment, the disease monitoring system can be divided into a data processing system, a data monitoring system, and a disease early warning system. Among them, the data processing system can record and analyze blood sugar data; the data monitoring system can analyze the health status based on the processed data, form a health report, and provide a reference for auxiliary treatment of diabetes; and the disease early warning system will link the hearing aid to an in-ear alarm when the data is abnormal, and alarm the bound relatives or doctors through the mobile phone APP, so that timely examination and treatment can be carried out.

[0043] In summary, this embodiment provides a diabetes management hearing aid system, which combines the advantages of the portability of hearing aids and the non-invasiveness of sweat sensors, and can achieve the following beneficial effects:

[0044] 1. Traditional blood sugar testing is mostly done by needle pricking blood. Long-term invasive testing increases the user's physical pain and easily leads to psychological resistance. This embodiment estimates the human body's glucose level by detecting the glucose concentration in sweat, thereby assisting in diabetes management.

[0045] 2. As the existing blood sugar detection devices are inconvenient to carry, hearing aids are the preferred product to solve hearing problems. They have the advantages of miniaturization, portability and unified audience. Combining sweat sensor detection devices with hearing aids can achieve real-time management of diabetes;

[0046] 3. In view of the problem that conventional self-examination in existing technologies lacks records and analysis, and it is difficult to independently judge the direction of the disease, the real-time warning system equipped with the diabetes management hearing aid system can perform health management, record blood sugar concentration in real time, analyze the disease, assist in treatment, and provide real-time warnings. At the same time, with the help of hearing aids, the user's hearing condition is improved to solve the problem of users not being able to hear or not being able to hear clearly.

[0047] Figure 6 1 is a flow chart of a diabetes monitoring method based on a diabetes management hearing aid system provided by an embodiment of the present invention. The method is applicable to the situation of predicting the glucose level in the user's body and risk warning, and can be executed by the disease monitoring system in the diabetes management hearing aid system. Figure 6 As shown, the method specifically includes:

[0048] S110. Identify the current scene of the user based on the sound signal obtained by the hearing aid and the preset user schedule, and predict the proportion of the current scene in a future period of time, as well as the next scene to be entered after the current scene ends; wherein the scenes of the user include sleeping, eating, working, exercising and others, and the future period of time covers at most two different scenes.

[0049] The changing rules of glucose levels in the user's body are not the same in different scenarios. Therefore, when predicting future glucose levels, this embodiment first considers the specific scenario in which the user is in the future. Optionally, the user can preset his or her daily schedule in the disease monitoring system, and use the schedule information as a priori information in the scenario judgment. For example, the schedule for weekdays is 22:00-7:00 sleep, 7:00-8:00 other, 8:00-9:00 diet, 9:00-12:00 work, 12:00-13:00 sleep, ... 19:00-20:00 exercise... etc., and rest days may be another schedule. This preset schedule represents the general rules of the scene in which the user is in each time period, which may not be completely consistent with the actual situation. When the user does not preset the schedule, the system can default to the general schedule of most people.

[0050] In addition to the preset schedule, this embodiment also uses the sound signal obtained by the hearing aid to predict the scene in which the user will be in the future. Specifically, the sound signal includes semantic information, such as the content of the user's conversation (such as what is currently being done and what light will be used in a while), and also includes audio information, such as the user's speaking state and emotions (such as tension, relaxation, panting, etc.), environmental sounds (such as sounds on the court, on the roadside, and indoors), etc. In order to make full use of various types of information, this embodiment constructs a multimodal scene prediction model, which predicts the current scene in which the user is located, the proportion of the duration of the current scene in the future, and the next scene to be entered after the current scene ends by fusing multimodal information. Among them, the future duration is also pre-set, such as the next 2 hours, and the duration includes at most one scene change. Optionally, according to the preset user schedule, the sum of the duration of each two adjacent scenes can be counted; the minimum value of the sum of each duration is taken as the future duration to be predicted.

[0051] In a specific embodiment, the scenario prediction model is as follows: Figure 7 As shown in the figure, it includes semantic branch, audio branch, prior branch, multimodal fusion layer and two classifiers. The structure and function of each part are explained in combination with the specific data processing flow.

[0052] First, the semantic content is extracted from the sound signal acquired by the hearing aid in the recent period through the sound-to-text function, and the semantic content is input into the semantic branch in the form of text for encoding to obtain an embedding vector of the semantic content. Optionally, the semantic branch can adopt the Glove structure or other text vectorization models.

[0053] At the same time, the sound signal is converted into a frequency domain, and the spectrum information is input into the audio branch for encoding to obtain an embedding vector of the audio signal. Optionally, the original audio waveform can be converted into a frequency domain to generate a log-Mel spectrum graph, and the audio branch uses a VGGish structure to convert the log-Mel spectrum graph into an embedding vector.

[0054] At the same time, according to the preset user schedule, the current scene, the next scene and the proportion of the current scene in the future time period in the schedule of the user are determined, and the three pieces of information are vectorized respectively using one-hot encoding or hash encoding methods, and input into the prior branch of the scene prediction model for encoding. Optionally, the prior branch can adopt a multi-layer perceptron structure.

[0055] After encoding of each branch is completed, the encoding vectors output by each branch are input into the multimodal fusion layer for feature fusion to obtain multimodal fusion features. Optionally, the multimodal fusion layer can adopt the encoder structure of Transformer, and the encoding vectors output by each branch are concatenated and input into the encoder to obtain multimodal fusion features.

[0056] Finally, the multimodal fusion features are simultaneously input into two classifiers, where classifier 1 is used to output the probability vector of the current scene of the user belonging to each scene type, and classifier 2 is used to output the current scene of the user, the proportion of the current scene in the future, and the next scene to be entered after the current scene ends. Optionally, both classifiers can adopt a multi-layer perceptron structure, and classifier 2 can output the probability vector of all combinations of the current scene, the next scene, and the proportion, and the one with the largest probability is the final classification result.

[0057] Furthermore, in the training process of the above-mentioned scene prediction model, the model training can be completed by constraining the difference between the scene type with the highest probability output by classifier 1 and the current scene output by classifier 2 to be minimized, and constraining the embedding vectors of the current scene and the next scene output by classifier 2 to be minimized after weighted summation according to the proportion and the difference with the probability vector output by classifier 1. Under this constraint, the probability vector output by classifier 1 can not only be used to determine the type of the current scene, but also to characterize the component of the future period of time covering the current scene and the next scene in the two scenes, thereby providing more information about the proportion and the next scene, and improving the accuracy of the prediction of the three types of information. Optionally, the following loss function can be constructed to implement the above constraints:

[0058] (1)

[0059] in, represents the loss function value, represents the probability vector output by classifier 1, Indicates the scene type with the highest probability in the probability vector; , and Respectively represent the current scene, next scene and proportion output by classifier 2; , and Respectively , and The embedding vector of (the specific vectorization method can be consistent with that in the prior branch); represents the result of weighted summation of the embedding vectors of the current scene and the next scene output by classifier 2 according to the proportions, Represents the similarity between two vectors; and are the weights of the two loss terms, with values ​​between [0,1]. Further, we can set > , because the consistency of the current scene prediction is the premise for the accuracy of the above proportion and the next scene prediction. The above constraints can be achieved by minimizing .

[0060] Of course, in addition to the above constraints, an appropriate amount of labeled samples are also required to label at least one of the current scene, proportion, and next scene to be predicted. These labeled samples come from user settings. For example, the user sets in advance what is being done, how long to do it, or what to do next, which can be used as labels for the current scene, proportion, or next scene for model training. Of course, not every sample has the standard of three contents. In actual training, whichever label each sample has, the label information is used for supervised training, and the unlabeled samples are unsupervised according to the above constraints.

[0061] Specifically, in the parameter updating process, for unlabeled samples, the parameters of the scene prediction model can be updated by minimizing the difference between the scene type with the highest probability output by constraining classifier 1 and the current scene output by classifier 2, and minimizing the difference between the embedding vectors of the current scene and the next scene output by constraining classifier 2 and the probability vector output by classifier 1 after weighted summation according to the proportion.

[0062] For samples that are labeled with at least one of the current scene, next scene, and proportion, on the basis of the above constraints, the constraint of minimizing the difference between the data item output by the classifier and the corresponding labeled item is added to update the parameters of the scene prediction model. Optionally, the loss function of such samples can be expressed as follows:

[0063]

[0064] in, , and are the label values ​​of the current scene, the next scene and the proportion. , and is the corresponding weight coefficient, and its value is between [0,1]. In actual calculation, , and When there is annotation information in any of the items, the corresponding weight coefficient , and The value is non-zero. The weight coefficient corresponding to the item without annotation information is 0, and , and When it takes a non-zero value, it can be greater than and , increasing the learning intensity of label information. Through this combination of supervised and unsupervised methods, we can make full use of fully labeled samples, partially labeled samples, and unlabeled samples, overcome the problem of difficult sample accumulation, and quickly complete model training.

[0065] S120, calling the blood glucose prediction models corresponding to the current scenario and the next scenario, predicting the user's future blood glucose time series data, and comparing it with the blood glucose threshold under the corresponding scenario to determine whether there is a risk of diabetes.

[0066] This embodiment predicts the user's glucose level time series data in the future period of time based on the current scene, the next scene, and the proportion of the current scene in the future period of time predicted in S110. Since the blood glucose prediction models in different scenes are different, the process can include the following two optional implementations according to the results of the scene prediction model:

[0067] The first optional implementation is applicable to the case where the current scene is different from the next scene and the ratio is between (0, 1). This indicates that the user will have a scene transition in the future, so first multiply the ratio by the future time to obtain the future duration of the current scene.

[0068] Next, the blood sugar prediction model corresponding to the current scenario is called, and the user's blood sugar time series data within the duration is predicted based on the user's blood sugar detection time series data s1 at the current moment and a period of time before it. Optionally, the blood sugar prediction model in each scenario can adopt an LSTM structure and be trained with the user's own blood sugar data. The input of the model is the blood sugar time series data s1 at the current moment and a period of time before it, and the output is the blood sugar time series data s2 within the future period of time t (for example, 2 hours). After the prediction is completed, the time series data s3 within the duration t1 (for example, 1.5 hours) can be obtained.

[0069] Then, subtract the duration of the current scene from the future duration to obtain the duration t2 (for example, 0.5 hours) of the next scene. A new set of time series data is then formed from the most recent time series data in s3 and s1, that is, the time window is rolled forward by t1 duration, so that the total time series length satisfies the model input dimension. The new time series data is input into the blood glucose prediction model corresponding to the next scene to predict the user's second blood glucose time series data in the next scene in the future. Similarly, the output of the model is blood glucose time series data of duration t, from which the time series data of duration t2 can be taken. For ease of distinction and description, in this embodiment, the time series data with a duration of t1 predicted according to the current scene is referred to as the first blood glucose time series data, and the time series data with a duration of t2 predicted according to the next scene is referred to as the second blood glucose time series data.

[0070] Finally, the first blood sugar time series data and the second blood sugar time series data are compared with the blood sugar thresholds in the current scene and the next scene, respectively, and data anomalies are identified. Specifically, the healthy thresholds of blood sugar in different scenes are also different, so the two sets of time series data are compared with the thresholds of the corresponding scenes, and those that exceed the respective threshold ranges are abnormal data. This method can filter abnormal data separately in different time periods, further improving the accuracy of screening and risk warning.

[0071] The second optional implementation is applicable to the case where the current scene is the same as the next scene, or the proportion = 0 or 1. This situation indicates that the user is in a single scene for a period of time in the future, wherein when the current scene is the same as the next scene or the proportion = 1, it indicates that the user will be in the current scene for a period of time in the future; when the proportion = 0, it indicates that the user will be in the next scene for a period of time in the future. Regardless of which scene you are in, you only need to call the blood glucose prediction model for the single scene, and predict the user's blood glucose time series data for the future period of time based on the user's blood glucose detection time series data s1 at the current moment and the period of time before it, and compare the time series data with the glucose threshold value in the single scene. Data exceeding the threshold range is abnormal data, and risk warnings can be issued.

[0072] In summary, this embodiment provides a diabetes monitoring method based on a diabetes management hearing aid system. According to the voice signals collected by the hearing aid and the user's daily schedule, it predicts the scene the user will be in for a period of time in the future and the possible scene transitions; and predicts the future glucose level trend in different time periods according to the scene transitions, and performs threshold comparisons in different time periods to improve the accuracy of glucose predictions and diabetes risk warnings.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A diabetes monitoring method based on a diabetes management hearing aid system, characterized in that: Applied to a diabetes management hearing aid system, wherein the diabetes management hearing aid system comprises: Hearing aids, including hearing aid chips; A flexible sweat sensor patch, comprising a sweat collection hole, an electrochemical electrode and a microchannel; wherein the outer side of the flexible sweat sensor patch is used to adhere to the skin behind the ear, and the inner side is adhered to the outer surface of the hearing aid, and is electrically connected to the hearing aid chip through a contact electrode; after the hearing aid is worn, the sweat of the skin behind the ear flows into the microchannel through the sweat collection hole, and the sweat in the microchannel undergoes an electrochemical reaction under the action of the electrochemical electrode, generating an electrical signal that is transmitted to the hearing aid chip, and the hearing aid chip uses an internal integrated analysis algorithm to detect blood glucose concentration; A disease monitoring system for recording test data, analyzing diabetes risk, and linking the hearing aid to provide in-ear warnings when data is abnormal; Specifically, the method is applied to the disease monitoring system, and the method includes: Extracting semantic content from the sound signal acquired by the hearing aid, and inputting the semantic content into the semantic branch of the scene prediction model for encoding; Performing frequency domain conversion on the sound signal, and inputting spectrum information into the audio branch of the scene prediction model for encoding; According to the preset user schedule, determine the current scene, next scene and the proportion of the current scene in the future time period in the schedule of the user, and input them into the prior branch of the scene prediction model for encoding; wherein the scenes of the user include sleeping, eating, working, exercising and others, and the future time period covers at most two different scenes; The encoding results of each branch are input into the multimodal fusion layer to obtain multimodal fusion features; The multimodal fusion features are simultaneously input into two classifiers, wherein classifier 1 outputs a probability vector of the current scene of the user belonging to each scene type, and classifier 2 outputs the current scene of the user, the proportion of the current scene in a future period of time, and the next scene to be entered after the current scene ends; Calling a blood sugar prediction model corresponding to the current scenario, and predicting the first blood sugar time series data of the user in the future under the current scenario according to the proportion and the user's current blood sugar detection time series data; Calling a blood sugar prediction model corresponding to the next scenario, and predicting the user's second blood sugar time series data in the next scenario in the future according to the proportion and the first blood sugar time series data; Compare the first blood glucose time series data and the second blood glucose time series data with the blood glucose thresholds in the current scenario and the next scenario, respectively, and identify data anomalies; Wherein, during the training process of the scene prediction model, For unlabeled samples, the parameters of the scene prediction model are updated by constraining the difference between the scene type with the highest probability output by the classifier 1 and the current scene output by the classifier 2 to be minimized, and constraining the embedding vectors of the current scene and the next scene output by the classifier 2 to be minimized after weighted summation according to the proportions and the probability vector output by the classifier 1; For samples that are labeled with at least one of the current scene, next scene, and proportion, based on the constraints of unlabeled samples, the constraint of minimizing the difference between the data items output by the classifier and the corresponding labeled items is added to update the parameters of the scene prediction model.

2. The method according to claim 1, characterized in that The microchannel substrate is made of polydimethylsiloxane material. Different refractive indexes and transmittances of visible light can make the microchannel present different colors in an empty state and a water-filled state, thereby judging whether the amount of sweat meets the detection state.

3. The method according to claim 1, characterized in that When there is a detection requirement but the amount of sweat is insufficient for detection, a voltage is applied between the electrochemical electrodes to generate a non-inductive current under the skin surface, thereby stimulating sweat secretion.

4. The method according to claim 1, characterized in that The hearing aid also includes a microphone and a receiver. The microphone is used to acquire external sounds and transmit them to the hearing aid chip. The hearing aid chip is used to optimize the sounds. The receiver is used to output the optimized sounds to the user's ears.

5. The method according to claim 1, characterized in that The end of the microchannel comprises an exhaust hole, and the liquid in the microchannel generates gas through electrochemical reaction and heating, and is discharged from the exhaust hole.

6. The method according to claim 1, characterized in that The disease monitoring system is also used to issue an early warning through the mobile phones of relatives, friends or doctors when the data is abnormal.

Images