A new non-invasive multi-parameter monitoring device and method

Abstract

The present application relates to a kind of new noninvasive multi-parameter monitoring device and method.The device includes light source generation unit, photoelectric detection unit, calculation unit, the content of different components in blood is calculated by self-learning and optimized neural network model, so as to realize the noninvasive real-time monitoring of blood component.101 is light source generation unit, 102 is finger, 103 is detection unit, 104 is calculation unit, the characteristic quantity corresponding to different transmission wavelengths is calculated by analyzing and calculating the signal collected by detection unit, the characteristic quantity corresponding to all wavelengths is used as the input variable of model training, the actual measured value of blood analyzer is used as the output quantity of neural network, multi-layer neural network is created, and the weight corresponding to different layers and different nodes is obtained, once optimal model training is completed, subsequent only needs to input the characteristic value of the transmission light corresponding to different wavelengths to calculate blood component, so as to realize the function of noninvasive multi-parameter real-time monitoring.

Description

Technical Field

[0001] This invention relates to the field of home health care, and in particular to a device and method for non-invasive real-time monitoring of blood components. Background Technology

[0002] This invention utilizes a humanoid robot for home use, combined with corresponding physiological parameter detection equipment. Employing self-learning and deep learning algorithms in AI technology, it accurately calculates various physiological parameters of the human body, providing early warning. If abnormal results are detected in relevant bodily parameters, prompt medical examination is recommended to prevent potential problems and avoid delays in treatment.

[0003] Humanoid robots are developing rapidly, and they have made great progress in performance, factory work, and conversation. However, they have not yet found a breakthrough in home care and have not been widely used.

[0004] Home care is best achieved using non-invasive monitoring methods, such as pulse oximeters, which use photoplethysmography (PPG) to calculate blood composition based on the different absorption rates of different blood components at different wavelengths. However, blood glucose, glycated hemoglobin, uric acid, creatinine, and other components require blood tests to determine their levels. The idea is to develop a method that sequentially monitors the signals transmitted through the arteries and veins of the fingertip using different wavelengths of light. Then, based on an AI model and deep learning algorithms, this method compares the signals transmitted through different wavelengths with actual blood test results. A backpropagation (BP) algorithm is used to build a model, optimize different weight parameters, find the optimal model and correlation coefficient, train the model with a sufficient amount of clinical data, and then combine the trained model with actual data from different wavelengths transmitted through the finger to predict blood composition. Summary of the Invention

[0005] The purpose of this invention is to provide a device and method for non-invasive testing of blood-related components.

[0006] To achieve the above objectives, the basic units of the present invention include: a light source generating unit that provides light sources of different wavelengths; a detection unit that detects signals after penetrating the fingertips; and a calculation unit that calculates the content of relevant components in the blood based on the detected signals and a model obtained by a deep learning algorithm.

[0007] The aforementioned light source generating unit uses broadband light sources such as incandescent lamps, xenon lamps, and halogen lamps as light sources, and uses gratings or prisms as beam splitting devices to generate monochromatic light sources of different wavelengths. Due to the different contents of different components in the blood, different absorption effects will be produced, thereby affecting the DC and AC components of the transmitted light intensity. The AC component corresponds to the peak-to-peak intensity of the photoplethysmography wave.

[0008] The aforementioned detection unit employs a photodetector with a wide spectral range responsivity to detect the light intensity signal after penetrating the peripheral blood vessels of the finger, and to separate the DC and AC signal intensities from the signal.

[0009] The aforementioned computing unit calculates the characteristic quantities corresponding to different wavelengths based on the DC and AC signal intensities of the transmitted light at different wavelengths detected. Then, it calculates the content of each component in the blood based on the optimal model constructed by deep learning, thereby realizing non-invasive multi-parameter monitoring.

[0010] This invention relates to a non-invasive multi-parameter monitoring device and method for monitoring blood components. Compared with traditional invasive blood component analysis, it can achieve non-invasive real-time monitoring of blood components, reduce the pain of blood drawing for patients, and can be used for home monitoring. It can also provide early warning for patients, and if relevant components exceed the normal value, it can remind patients to go to the hospital for medical treatment as soon as possible.

[0011] This invention utilizes the different absorption values ​​of different components in human blood at different wavelengths. By constructing a large model and employing a deep learning algorithm, the differences between the component content calculated by the model and the results of blood tests are calculated. Then, a backpropagation neural network is used to reverse-correct the weights of each layer of the network. Through continuous accumulation of sample size and self-learning, the calculated blood components are made closer to the test results of a real blood analyzer. After a period of self-learning and training, the optimal model can be obtained. Subsequently, the content of each blood component can be calculated by different feature quantities calculated from the transmitted light intensity of different wavelength tests, achieving the goal of real-time, non-invasive monitoring.

[0012] The non-invasive multi-parameter monitoring device and method for monitoring blood components proposed in this invention, once the feasibility of the scheme is verified, will effectively reduce patient suffering. When used in home care, it can detect abnormal blood component levels in advance, allowing patients to go to the hospital for further treatment earlier and avoid delaying their condition. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the architecture of the non-invasive multi-parameter monitoring system for blood components proposed in this invention.

[0014] Figure 2 This is a schematic diagram of the device for generating a monochromatic light source in the light source generation unit.

[0015] Figure 3 This is a schematic diagram of a transmissive, non-invasive fingertip detection method.

[0016] Figure 4 This is a schematic diagram of the BP neural network algorithm and the reverse correction of network node weights.

[0017] Figure 5 This is a schematic diagram of the pulse wave detected penetrating the fingertip and related characteristic values.

[0018] Figure 6 It is a table that records the different characteristic values ​​of transmitted light at different wavelengths.

[0019] Figure 7 This is a record table of the differences between the calculated values ​​of different blood components and the actual measured values ​​by a blood analyzer. Figure 8 This is a schematic diagram illustrating the combined application of a home care humanoid robot and a non-invasive multi-parameter monitor. Detailed Implementation

[0020] The following description describes embodiments of the present invention with reference to the accompanying drawings. These embodiments are illustrative examples and do not imply that the invention is limited to these specific embodiments. The present invention can be implemented in various ways without departing from its essential principles.

[0021] Embodiment 1 of the invention Figure 1 The diagram shows the architecture of the non-invasive blood component monitoring system proposed in this invention. 101 is the light source generating unit, which achieves the function of spectral dispersion through a prism or grating similar to that in a monochromator, generating monochromatic light sources of different wavelengths. 102 is the finger, which allows monochromatic light of different wavelengths to irradiate the capillaries at the fingertips in sequence. Different blood component contents will affect the intensity of transmitted light from different wavelength light sources. Unit 103 is the detection unit, which uses a wide-spectrum detection range to detect the light intensity transmitted through the fingertip, separating DC and AC signals. The AC signal corresponds to the pulse wave in the photoplethysmography. Unit 104 is the calculation unit, which analyzes and calculates the characteristic quantities corresponding to different transmission wavelengths by analyzing the signals collected by the detection unit. The characteristic quantities corresponding to all wavelengths are used as input variables for model training, and the actual measured values ​​of the blood analyzer are used as the output of the neural network. A multi-layer neural network is created, and the weights corresponding to different nodes in different layers are obtained. In the actual model creation, a cooperative relationship with medical institutions is established. While the patient is having blood drawn, signals of multiple monochromatic light penetrating the finger of the other arm are collected and stored. These calculated characteristic values ​​are used as input to the model. After the patient's blood analysis results are obtained, the actual test results are used as the output reference values. Through the collection of a large number of patient samples, the model will continuously learn and optimize the corresponding weights, so that the model calculation results are closer to the real test results. Once the optimal model training is completed, the blood components can be calculated by inputting the characteristic values ​​of the transmitted light corresponding to different wavelengths.

[0022] Figure 2 The diagram shows a schematic of the device for generating a monochromatic light source in the light source generating unit. 201 is the collimating lens optical unit, 202 is the input end of the monochromator broadband light source, 203 is the dispersive element, which can be realized by a prism or grating, and 204 is the emitted monochromatic light source.

[0023] Figure 3 This is a schematic diagram of a non-invasive fingertip transmissive ...

[0024] Figure 4 This is a schematic diagram of the BP neural network algorithm and the weights of the reverse correction network nodes. 401 is the input terminal of the neural network. In this invention, the DC and AC components of multiple wavelengths of transmitted light, as well as multiple feature values ​​obtained from the AC components, are used as input variables to the neural network model. 402 is the output terminal of the multi-layer neural network model. In this invention, it is used to generate numerical values ​​for different blood component contents. 403 represents the real data obtained from the subject's actual blood through a clinical blood analyzer during model training. By inputting the feature quantities calculated from the transmitted light detection signals of different wavelengths from the subject's fingertip and the data obtained from the actual subject's blood through the blood analyzer, and using a large AI model and NAS (Neural Architecture Search) technology, a multi-layer neural network model can be automatically generated. Through continuous training with hundreds or even thousands of actual subject data, a neural network model close to the optimal solution will eventually be generated. Subsequently, by inputting the feature quantities calculated from the transmitted light detection signals of different wavelengths from the subject's fingertip, the optimal neural network model can output the values ​​for different blood component contents, thereby achieving the function of non-invasive monitoring.

[0025] Figure 5This is a schematic diagram of the detected pulse wave penetrating the fingertip and related characteristic values. 501 corresponds to the AC signal in the transmitted light, representing the peak-to-peak value of the pulse wave in the photoplethysmography, denoted by H0. 502 represents the time interval between adjacent pulse wave peaks, denoted by T0, corresponding to the heart rate (60 / T0). 503 represents the area of ​​the pulse wave during the rising phase corresponding to the systolic phase, denoted by S1. 504 represents the area of ​​the pulse wave during the falling phase corresponding to the diastolic phase, denoted by S2. The diastolic-systolic ratio is represented by S2 / S1. 505 represents the maximum slope of the pulse wave during the rising phase corresponding to the systolic phase, denoted by the straight line L0, and its corresponding slope is represented by KL0. Figure 5 This describes the signal of a single wavelength light source penetrating a finger. This signal can generate multiple characteristic quantities corresponding to 501 to 505 mentioned above, and the intensity of transmitted light is also one of these characteristic quantities.

[0026] Figure 6 This is a table recording the different characteristic values ​​of transmitted light at different wavelengths. Using the data in this table, the characteristic values ​​corresponding to different wavelengths of transmitted light are calculated, and these data will serve as the input values ​​for a neural network model.

[0027] Figure 7 This table records the differences between the model's calculated values ​​and the actual blood analyzer's measured values ​​for different blood components. The model's output value is calculated based on the differences between the model's output value and the actual blood analyzer's output. These differences are then combined with a BP neural network algorithm, which continuously optimizes the relevant parameters of each node in different layers through self-learning, making the output infinitely close to the actual blood analyzer test results.

[0028] As mentioned earlier, Method 1 calculates relevant feature values ​​by measuring the light intensity of multiple wavelength light sources after penetrating the finger. Then, referring to the actual test results of the subject's blood analyzer, a neural network model is trained. After training with multiple subject samples, a neural network model close to the optimal solution can be found, thereby achieving non-invasive real-time monitoring of the content of blood-related components.

[0029] In practical clinical applications, many diabetic patients endure invasive blood glucose testing for extended periods, and continuous monitoring is not always possible. The large neural network model constructed using this invention may eliminate the need for numerous different wavelengths of monochromatic light source illumination for diabetic patients. It may even allow for the calculation of only a limited number of specific wavelengths, either using the aforementioned model or by recreating the model based on the methods described in this invention, focusing solely on blood glucose and glycated hemoglobin indicators of interest to diabetic patients. By reducing the calculation of other blood components, the model's computation time will be shortened. This concept and algorithm for non-invasive, real-time blood glucose display will bring convenience to a wide range of diabetic patients.

[0030] Embodiment 3 of the invention Figure 8This is a schematic diagram illustrating the combined application of a home care humanoid robot and a non-invasive multi-parameter monitor. The non-invasive multi-parameter monitor's emission light source and detector, as proposed in this invention, can be placed inside the robot's mouth. During testing, the home care robot simply opens its mouth, the subject inserts their finger into the robot's mouth, and then the mouth is closed for measurement. 801 represents the multi-wavelength emission light source and detection unit installed in the home care robot's mouth. 802 represents the arm of the humanoid home care robot, and 803 represents the foot of the humanoid home care robot. Through continuous self-learning and optimization by the humanoid robot, a neural network model that closely approximates the true solution is eventually obtained. Once the optimized model is developed, relevant variable parameters are obtained by inserting the subject's finger into the care robot's mouth. Using this neural network model, the levels of relevant blood components can be immediately determined.

[0031] In addition to being installed at the mouth, the light source emission and detection unit of the non-invasive detection device of this invention can also be installed in other parts of the care robot's body.

Claims

1. A non-invasive multi-parameter monitoring method, characterized in that... It includes a light source generating unit that can generate multiple monochromatic wavelengths, a detection unit that detects light intensity, and a computing unit.

2. The light source generating unit according to claim 1, characterized in that... The light source generating unit can produce monochromatic light of multiple wavelengths, wherein the dispersive element can be a prism or a grating.

3. The light intensity detection unit according to claim 1 can be used to detect the light intensity after penetrating the fingertip.

4. The computing unit according to claim 1. During neural network model training, it is used to generate a neural network model based on the basic characteristics of transmitted light input to the model and the actual values ​​of blood components. During the actual measurement stage, the detected signals are input into the model to calculate the content of different blood components.

5. A non-invasive multi-parameter monitoring device, characterized in that... It includes a light source generating unit, a detection unit, and a computing unit.

6. The light source generating unit according to claim 5, characterized in that... The light source generating unit includes a light source generating unit that can generate multiple monochromatic wavelengths. The light source generating unit can generate monochromatic light of multiple wavelengths, wherein the dispersive element can be a prism or a grating.

7. The detection unit according to claim 5, characterized in that... The detection unit can be used to detect the intensity of light after it penetrates the fingertip.

8. The calculation unit according to claim 5 is used to calculate the content of each component of blood based on a neural network created from data obtained by a blood analyzer from the blood of the subject.

Images