Noninvasive blood glucose concentration detection method based on microwave time domain signal and extreme learning machine
The combination of microwave time domain signals and the ELM algorithm addresses inefficiencies in non-invasive glucose detection, achieving precise blood glucose concentration predictions with minimal error.
Patent Information
- Application Number
- CN202310632355.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-31
- Publication Date
- 2024-12-06
AI Technical Summary
Existing non-invasive blood glucose detection methods using microwave signals suffer from inefficiencies and inaccuracies, particularly in processing and analysis, which are not adequately addressed by traditional neural network learning algorithms.
A method utilizing microwave time domain signals in conjunction with the Extreme Learning Machine (ELM) algorithm for non-invasive blood glucose detection, involving feature extraction and normalization of signals, followed by training with ELM to achieve accurate predictions.
The proposed method achieves high accuracy and reduced errors in predicting blood glucose concentrations, with an average relative error of 1.26% and a root mean square error of 3.02 mg/dL.
Smart Images

Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] The invention relates to a non-invasive blood sugar concentration detection method based on microwave time domain signals and extreme learning machines. Background Art
[0002] Diabetes is a metabolic disorder or endocrine disease caused by a lack of insulin in the body. High blood sugar can lead to damage or functional failure of various organs in the body, and its harm cannot be ignored. Blood sugar testing in daily life and clinical treatment can play a key role in the control and treatment of diabetes. The necessity and importance of blood sugar testing are self-evident. In recent years, non-invasive blood sugar testing methods have received increasing attention due to their potential advantages such as painlessness, convenience, and reduced risk of infection. Compared with other detection methods, microwave detection has the advantages of lower cost, easier use, higher safety, and better compliance with market and medical needs.
[0003] The effective processing and analysis of microwave blood glucose signals requires efficient machine learning algorithms. In this regard, the Extreme Learning Machine (ELM) algorithm has shown good adaptability and superiority in processing microwave blood glucose signals. The ELM algorithm is a single hidden layer feedforward neural network type machine learning algorithm proposed by Huang based on the Moore-Penrose generalized inverse matrix. It mainly solves the problems of slow learning speed, long iteration time, and the need to set parameters such as learning step size and learning rate in advance for single hidden layer feedforward neural networks. Compared with traditional neural network learning algorithms, the ELM algorithm only needs to set the appropriate number of nodes in the hidden layer, randomly generate all the parameters required for the hidden layer, and use the least squares method to determine the weights of the output layer. The entire learning process has only one step and does not need to update the parameters of the hidden layer. Therefore, the ELM algorithm processes data faster and more efficiently.
[0004] The present invention proposes a method for non-invasively detecting blood sugar concentration based on microwave time domain signals and extreme learning machine algorithm. Summary of the invention
[0005] The present invention proposes a non-invasive blood glucose concentration detection method based on microwave time domain signals and extreme learning machines. Compared with other microwave non-invasive detection methods, this method has smaller errors and higher accuracy.
[0006] A non-invasive blood glucose concentration detection method based on microwave time domain signals and extreme learning machines comprises the following steps: Step 1: Prepare glucose aqueous solutions with different concentrations similar to the dielectric properties of blood as the experimental test samples. The concentration range of the experimental solution is 10 mg / dL-500 mg / dL, with an interval of 10 mg / dL. Use a pipette to take equal amounts of glucose solutions of different concentrations and inject them into the test model each time. Use a continuous sinusoidal microwave signal with a resonant frequency of 1 GHz and a power of 10 dBm for detection. Step 2: Use an oscilloscope to record the time domain detection signal results at each concentration; Step 3: Extract time domain features from the received microwave time domain signal and normalize all input features. The data with glucose concentrations of 50 mg / dL, 100 mg / dL, 150 mg / dL, 200 mg / dL, 250 mg / dL, 300 mg / dL, 350 mg / dL, 400 mg / dL, 450 mg / dL and 500 mg / dL are used as test sets, and the rest of the data are used as training sets. Step 4: Use the ELM algorithm to train the training set sample data, compare the predicted concentration results with the test set, and achieve accurate prediction of the unknown concentration of glucose solution. BRIEF DESCRIPTION OF THE DRAWINGS
[0007] Figure 1 Detection model diagram Figure 2 Local time domain received signal Figure 3 Comparison of concentration prediction results based on the ELM algorithm Implementation
[0008] The present invention is further described below in conjunction with a specific embodiment. The present invention proposes a non-invasive blood glucose concentration detection method based on microwave time domain signals and extreme learning machines. The specific implementation scheme is as follows: (1) Prepare glucose aqueous solutions of different concentrations, ranging from 10 mg / dL to 500 mg / dL, with an interval of 10 mg / dL, for a total of 50 groups. Use a pipette to take an equal amount of the test solution each time and inject it into the test model. The test model adopts a fat layer-blood layer-fat layer structure with a size of 80 mm×30 mm×6 mm, of which the thickness of the blood layer is 3 mm and the thickness of the fat layer is 1.5 mm. Use polyoxymethylene resin (POM) material to replace adipose tissue. The dielectric constant of polyoxymethylene resin is about 3.7, which is similar to that of adipose tissue. In addition, the dielectric constant of this material is relatively stable and is not easily affected by changes in humidity and temperature in the environment. It is also less affected by changes in frequency. Therefore, this material is used to replace adipose tissue, and the blood layer is replaced by a glucose aqueous solution with dielectric properties similar to blood.
[0009] (2) Two antennas are fixed on both sides of the model to be tested, respectively, to transmit and receive microwave time domain signals, such as Figure 1 As shown. A vector signal generator is used to transmit a continuous sinusoidal signal with a frequency of 1 GHz to the transmitting antenna fixed on one side of the model to be tested. The microwave signal is transmitted from the antenna on one side of the blood glucose model to be tested. The single-frequency microwave signal is transmitted from the transmitting port to the transmitting antenna on one side of the model to be tested. After penetrating the glucose solutions of different concentrations to be tested, it is received by the receiving antenna fixed on the other side of the test model. The received signal is finally displayed and stored via an oscilloscope. The unprocessed received signal has a poor correlation with the concentration, as shown in Figure 1. Figure 2 shown.
[0010] (3) Feature extraction is performed on the received microwave time domain signal. Six time domain feature parameters are extracted as input features, which are the maximum value of the time domain waveform in each cycle. , minimum , peak-to-peak p , RMS r ,variance v and standard deviation s In order to improve the accuracy and reliability of data processing, the sample data to be processed are normalized. After processing the received signals of 50 groups of glucose solutions for experimental testing, 6×50 groups of sample data sets can be obtained, of which the data with glucose concentrations of 50 mg / dL, 100 mg / dL, 150 mg / dL, 200 mg / dL, 250 mg / dL, 300 mg / dL, 350 mg / dL, 400 mg / dL, 450 mg / dL and 500 mg / dL are used as test sets, and the rest of the data are used as training sets.
[0011] (4) The training set data is processed using the ELM algorithm. Figure 3 is the glucose prediction result based on ELM. It can be seen from the figure that each glucose prediction value (circle) is very close to the corresponding reference value (square). The results show that the method proposed in the present invention can achieve accurate prediction of blood glucose concentration. Table 1 specifically gives the concentration results and relative errors predicted by the ELM algorithm. It can be seen from the table that the average relative error between the predicted concentration obtained by this method and the actual concentration is 1.26%, and the root mean square error of the prediction result is 3.02 mg / dL. The experimental results show that the predicted concentration is consistent with the actual concentration.
[0012] Table 1 Concentrations predicted by ELM algorithm and relative errors
[0013] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
[0014] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.
Claims
1. A non-invasive blood glucose concentration detection method based on microwave time domain signals and extreme learning machines, characterized in that: The method comprises the following steps: Step 1: Use a pipette to take an equal amount of the solution to be tested and inject it into the model; Step 2: Use a continuous sinusoidal microwave signal to non-invasively detect the solution, and use an oscilloscope to record the time domain detection signal results at each concentration; Step 3: Extract features of the received microwave time domain signal and perform normalization processing; Step 4: Use the ELM algorithm to train the training set sample data, compare the predicted concentration results with the test set, and achieve accurate prediction of unknown concentration solutions.
2. The non-invasive blood glucose concentration detection method based on microwave time domain signal and extreme learning machine according to claim 1, characterized in that: The model to be tested in step 1 is a three-layer structure of fat-blood-fat with a size of 80 mm×30 mm×6 mm, of which the thickness of the blood layer is 3 mm and the thickness of the fat layer is 1.5 mm. Polyoxymethylene resin (POM) material with a dielectric constant similar to that of fat is used to replace adipose tissue. Its dielectric constant is 3.7 and is relatively stable. Glucose solution with a dielectric constant similar to that of blood is used to replace blood samples.
3. The non-invasive blood glucose concentration detection method based on microwave time domain signal and extreme learning machine according to claim 1, characterized in that: The concentration range of the glucose aqueous solution prepared in step 1 is 10 mg / dL-500 mg / dL, with an interval of 10 mg / dL, for a total of 50 groups.
4. The non-invasive blood glucose concentration detection method based on microwave time domain signal and extreme learning machine according to claim 1, characterized in that: The continuous sinusoidal microwave signal in step 2 has a resonant frequency of 1 GHz and a power of 10 dBm, and a planar monopole antenna is used to transmit and receive the signal.
5. The non-invasive blood glucose concentration detection method based on microwave time domain signal and extreme learning machine according to claim 1, characterized in that: The input features extracted from the time domain signal in step 3 are the maximum value of the time domain waveform in each cycle. , minimum , peak-to-peak p , RMS r ,variance v and standard deviation s .
6. The non-invasive blood glucose concentration detection method based on microwave time domain signal and extreme learning machine according to claim 1, characterized in that: In step 3, the time domain detection signals received for each concentration are processed to obtain 6×50 groups of data samples, of which the data with glucose concentrations of 50 mg / dL, 100 mg / dL, 150 mg / dL, 200 mg / dL, 250 mg / dL, 300 mg / dL, 350 mg / dL, 400 mg / dL, 450 mg / dL and 500 mg / dL are used as test sets, and the rest of the data are used as training sets.
7. The non-invasive blood glucose concentration detection method based on microwave time domain signal and extreme learning machine according to claim 1, characterized in that: The average relative error between the predicted concentration obtained by this method and the actual concentration was 1.26%, and the root mean square error of the predicted results was 3.02 mg / dL.
Citation Information
Patent Citations
Non-invasive blood glucose concentration detection method based on microwave spectrum of earlobe blood layer
CN108095734A
Blood glucose prediction method based on variational mode decomposition and extreme learning machine
CN109859850A
Microwave time domain signal non-invasive blood glucose concentration detection method based on earlobe blood layer
CN110074790A
Multivariable blood glucose prediction algorithm based on transfer learning and glycosylated hemoglobin
CN115482932A
