A non-invasive blood glucose estimation method, system and device based on dual-wavelength channel coupling characteristics, and a storage medium

Abstract

The application discloses a non-invasive blood glucose estimation method and system based on a dual-wavelength channel coupling feature, a device and a storage medium, and belongs to the technical field of non-invasive physiological detection. The method comprises the following steps: acquiring a dual-wavelength photoelectric plethysmogram signal, pre-processing a red light channel signal and an infrared light channel signal respectively, extracting corresponding single-channel pulse wave features, constructing candidate coupling features based on the corresponding features, screening or combining the candidate coupling features to form a target feature set, inputting the target feature set into a blood glucose estimation model, and obtaining a blood glucose estimation value or a blood glucose risk index. The application improves the stability, robustness and generalization ability of non-invasive blood glucose estimation.

Description

Technical Field

[0001] This invention relates to the field of non-invasive physiological testing technology, and in particular to a non-invasive blood glucose estimation method, system, device, and storage medium based on dual-wavelength channel coupling characteristics. Background Technology

[0002] Blood glucose monitoring plays a crucial role in screening for abnormal glucose metabolism, identifying diabetes risk, and long-term health management. Current blood glucose testing methods mostly rely on blood sampling, which presents challenges such as being invasive, inconvenient for high-frequency monitoring, and having low user compliance. Non-invasive blood glucose estimation technology based on photoplethysmography (PPG) signals has attracted widespread attention due to its potential for non-invasiveness, portability, and continuous monitoring.

[0003] Existing non-invasive blood glucose estimation schemes based on dual-wavelength pulse wave signals typically extract features from the red and infrared channels separately and then directly concatenate them into the model, or use only a single ratio index for estimation. These schemes suffer from at least the following problems: First, the correspondence between the red and infrared channels is not fully utilized, making it difficult to effectively express the synergistic changes between the two wavelengths; second, single-channel features are easily affected by subject tissue thickness, local perfusion level, measurement posture, and environmental interference, resulting in poor stability; third, simply concatenating features cannot reduce the impact of single-channel fluctuations and individual differences on the model output, limiting the model's generalization ability.

[0004] Therefore, there is an urgent need to provide a non-invasive blood glucose estimation method based on dual-wavelength channel coupling features to construct a more stable and physiologically relevant feature expression, thereby improving the accuracy, robustness, and cross-sample adaptability of non-invasive blood glucose estimation. Summary of the Invention

[0005] The purpose of this invention is to provide a non-invasive blood glucose estimation method, system, device and storage medium based on dual-wavelength channel coupling features, so as to solve the problem that existing dual-wavelength PPG blood glucose estimation only performs simple splicing of dual-channel features and cannot make full use of the dual-wavelength correspondence.

[0006] To achieve the above objectives, this invention provides a non-invasive blood glucose estimation method based on dual-wavelength channel coupling features, comprising the following steps: acquiring dual-wavelength photoplethysmography (PPG) signals from the target site of the subject, wherein the dual-wavelength PPG signals include at least a red light channel signal and an infrared light channel signal; preprocessing the red light channel signal and the infrared light channel signal respectively; extracting corresponding single-channel pulse wave features from the red light channel signal and the infrared light channel signal; constructing candidate coupling features based on the amplitude relationship, difference relationship, normalized difference relationship, correlation relationship, time offset relationship, or frequency domain response relationship between the corresponding single-channel pulse wave features; screening or combining the candidate coupling features to form a target feature set for blood glucose estimation; and inputting the target feature set into a blood glucose estimation model to obtain a blood glucose estimate or a blood glucose risk indicator.

[0007] Preferably, the center wavelength of the red light channel is approximately 660 nm, and the center wavelength of the infrared light channel is approximately 880 nm. Preferably, the preprocessing includes at least one of DC drift removal, bandpass filtering, median filtering, outlier removal, and baseline correction. Preferably, the single-channel pulse wave features include at least one of time-domain features, frequency-domain features, and morphological features.

[0008] Preferably, the candidate coupling features include at least one of the following: ratio features, difference features, normalized difference features, correlation features, time difference features, and frequency domain coupling features. Preferably, the ratio feature can be composed of the ratio of the AC component to the DC component of the red light channel relative to the ratio of the AC component to the DC component of the infrared light channel; the time difference feature can be composed of the time offset between corresponding peak points, valley points, or other key feature points of the red light channel and the infrared light channel; the frequency domain coupling feature can be composed of the ratio of the main frequency power of the two channels, the ratio of harmonic energy, the ratio of bandpass energy, or the spectral correlation coefficient.

[0009] Preferably, the screening or combination is performed based on at least one of feature stability, feature correlation, feature importance, cross-sample consistency, or model performance evaluation results to form a target feature set. This invention also provides a non-invasive blood glucose estimation system based on dual-wavelength channel coupled feature construction, including a signal acquisition module, a preprocessing module, a single-channel feature extraction module, a coupled feature construction module, a target feature formation module, and an inference module. This invention also provides a non-invasive blood glucose estimation device and a computer-readable storage medium.

[0010] Compared with existing technologies, this invention has at least the following advantages: First, this invention does not simply splice together dual-channel features, but constructs coupled features based on the correspondence between dual-wavelength channels, thus improving feature representation capabilities; second, by introducing features such as ratio, difference, normalized difference, correlation, time offset, and frequency domain coupling, it can more fully characterize the synergistic change patterns between dual-wavelength signals; third, compared with single-channel features or simple stacking features, dual-wavelength coupled features can reduce the impact of single-channel fluctuations and individual differences on blood glucose estimation results; fourth, by screening or combining candidate coupled features to form a target feature set, it is beneficial to improve model stability and generalization ability; fifth, it is compatible with signal quality gating and end-side inference technologies, facilitating the formation of a complete non-invasive blood glucose patent layout. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0012] Figure 2 This is a schematic diagram of the system structure of the present invention.

[0013] Figure 3 This is a schematic diagram illustrating the construction of candidate coupling features in this invention.

[0014] Figure 4 This is a schematic diagram illustrating the formation of the target feature set of the present invention. Detailed Implementation

[0015] The present invention will be further described below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited thereto.

[0016] In one embodiment, a finger-clip dual-wavelength optical detection device is used to acquire photoplethysmography (PPG) signals from the finger area of ​​the subject. The dual wavelengths include red light and infrared light, with the center wavelength of the red light being approximately 660 nm and the center wavelength of the infrared light being approximately 880 nm. The sensor outputs PPG signals from both the red and infrared channels.

[0017] Subsequently, the red light channel signal and the infrared light channel signal are preprocessed separately. The preprocessing may include one or more of the following: DC drift removal, bandpass filtering, median filtering, outlier removal, and baseline correction, to reduce the impact of ambient light, baseline drift, and high-frequency noise on subsequent analysis. In a preferred embodiment, the preprocessing parameters can be set according to the sampling rate, signal amplitude range, or target feature type to ensure the comparability of the corresponding features extracted subsequently.

[0018] After preprocessing, corresponding single-channel pulse wave features are extracted from the red light channel and the infrared light channel, respectively. These single-channel pulse wave features may include AC component amplitude, DC component, AC to DC component ratio, rise time, fall time, pulse period, peak position, waveform area, dominant frequency power, spectral centroid, harmonic energy, peak-to-valley difference, dominant wave slope, and waveform width. Preferably, for similar features, their corresponding values ​​are extracted from both the red light channel and the infrared light channel to facilitate subsequent construction of coupled features.

[0019] In a preferred embodiment, candidate coupling features are constructed based on the corresponding single-channel pulse wave characteristics described above. The candidate coupling features may include: 1. Ratio features, such as the ratio of the AC to DC component of the red light channel to the ratio of the AC to DC component of the infrared light channel; 2. Difference features, such as the difference between the peak amplitude of the red light channel and the peak amplitude of the infrared light channel; 3. Normalized difference features, such as the feature value formed by dividing the difference between the corresponding features of the red light channel and the infrared light channel by their sum; 4. Correlation features, such as the correlation coefficient, cross-correlation peak, or synchronization change index of the red light channel and the infrared light channel within a sampling window; 5. Time difference features, such as the time offset between the corresponding peaks, valleys, or key feature points of the red light channel and the infrared light channel; 6. Frequency domain coupling features, such as the ratio of the dominant frequency power, harmonic energy, bandpass energy, or spectral correlation coefficient between the red light channel and the infrared light channel.

[0020] In one embodiment, to further improve model performance, the constructed candidate coupled features are screened or combined. The screening can be performed based on at least one of feature stability, feature relevance, feature importance, cross-sample consistency, or model performance evaluation results; for example, removing candidate features that are too volatile, have high repetition, or contribute little to the model. The combination can include combining different categories of coupled features according to preset rules to form a target feature set for blood glucose estimation.

[0021] After forming the target feature set, the target feature set is input into the blood glucose estimation model to obtain the blood glucose estimate or blood glucose risk index. The blood glucose estimation model can be a linear regression model, a support vector machine model, a random forest model, a lightweight neural network model, or other models suitable for edge deployment. Preferably, the target feature set is mainly composed of coupled features, but it can also be combined with a small number of stable single-channel features to input into the model, so as to balance expressive power and model interpretability.

[0022] In one embodiment, the method can be used in conjunction with a signal quality gating method, i.e., first perform quality assessment on the dual-wavelength PPG signal, and then perform corresponding single-channel feature extraction, candidate coupled feature construction, and target feature set formation after the signal meets the gating conditions. In another embodiment, the method can be used in conjunction with an edge-side inference system, enabling the target feature set to be extracted, constructed, and input into the model at the edge, thereby reducing communication load and improving real-time performance.

[0023] Combination Figure 1 11 represents dual-wavelength PPG signal acquisition, 12 represents signal preprocessing, 13 represents single-channel feature extraction, 14 represents candidate coupled feature construction, 15 represents target feature set formation, 16 represents blood glucose estimation model, and 17 represents blood glucose estimation value or risk indicator output.

[0024] Combination Figure 2 1 represents the signal acquisition module, 2 represents the preprocessing module, 3 represents the single-channel feature extraction module, 4 represents the coupled feature construction module, 5 represents the target feature formation module, and 6 represents the inference module.

[0025] Combination Figure 3 31 represents the characteristic corresponding to the red light channel, 32 represents the characteristic corresponding to the infrared light channel, 33 represents the ratio characteristic, 34 represents the difference characteristic, 35 represents the normalized difference characteristic, 36 represents the correlation characteristic, 37 represents the time difference characteristic, and 38 represents the frequency domain coupling characteristic.

[0026] Combination Figure 4 , 41 represents the candidate coupled feature set, 42 represents feature selection, 43 represents feature combination, and 44 represents the target feature set.

[0027] The present invention can also be embedded in a terminal device in software form, or stored as a program in a computer-readable storage medium, and executed by a processor to complete the above method.

Claims

1. A non-invasive blood glucose estimation method based on dual-wavelength channel coupling characteristics, characterized in that, The process includes the following steps: S1, acquiring a dual-wavelength photoplethysmography (PPG) signal of the target area of ​​the object to be tested, wherein the dual-wavelength PPG signal includes at least a red light channel signal and an infrared light channel signal; S2, preprocessing the red light channel signal and the infrared light channel signal respectively; S3, extracting the corresponding single-channel pulse wave features from the red light channel signal and the infrared light channel signal. S4. Construct candidate coupling features based on the amplitude relationship, difference relationship, normalized difference relationship, correlation relationship, time offset relationship, or frequency domain response relationship between the corresponding single-channel pulse wave features; S5. Filter or combine the candidate coupling features to form a target feature set for blood glucose estimation; S6. Input the target feature set into the blood glucose estimation model to obtain the blood glucose estimate or blood glucose risk index.

2. The method according to claim 1, characterized in that, The center wavelength of the red light channel is approximately 660 nm, and the center wavelength of the infrared light channel is approximately 880 nm; the preprocessing in step S2 includes at least one of DC drift removal, bandpass filtering, median filtering, outlier removal, and baseline correction.

3. The method according to claim 1, characterized in that, The single-channel pulse wave characteristics in step S3 include at least one of time-domain characteristics, frequency-domain characteristics, and morphological characteristics; wherein, the time-domain characteristics include at least one of AC component amplitude, DC component, AC component to DC component ratio, rise time, fall time, pulse period, peak position, and waveform area; the frequency-domain characteristics include at least one of dominant frequency power, spectral centroid, harmonic energy, spectral width, and bandpass energy ratio; and the morphological characteristics include at least one of peak-to-valley difference, dominant wave slope, diphtheria wave correlation characteristics, waveform width, and waveform symmetry index.

4. The method according to claim 1, characterized in that, The candidate coupling features in step S4 include at least one of the following: the ratio feature of the red light channel corresponding feature to the infrared light channel corresponding feature, the difference feature, the normalized difference feature, the correlation feature, the time difference feature, or the frequency domain coupling feature.

5. The method according to claim 4, characterized in that, The ratio feature includes the ratio of the AC component to the DC component of the red light channel to the ratio of the AC component to the DC component of the infrared light channel; the normalized difference feature includes the feature value formed by dividing the difference between corresponding features of the red light channel and the infrared light channel by the sum of corresponding features; the correlation feature includes at least one of the correlation coefficient, cross-correlation peak value, or synchronous change index between the red light channel and the infrared light channel within a preset sampling window; the time difference feature includes the time offset between corresponding peak points, valley points, or other key pulse feature points between the red light channel and the infrared light channel; the frequency domain coupling feature includes at least one of the main frequency power ratio, harmonic energy ratio, bandpass energy ratio, or spectral correlation coefficient between the red light channel and the infrared light channel.

6. The method according to claim 1, characterized in that, In step S5, when screening or combining candidate coupling features, a target feature set is formed based on at least one of the following: feature stability, feature correlation, feature importance, cross-sample consistency, or model performance evaluation results.

7. The method according to claim 1, characterized in that, The blood glucose estimation model is a regression model or classification model built based on machine learning. The blood glucose estimation model includes at least one of the following: linear regression model, support vector machine model, random forest model, lightweight neural network model, or model suitable for edge deployment.

8. A non-invasive blood glucose estimation system based on dual-wavelength channel coupling characteristics, characterized in that, include: The signal acquisition module is used to acquire dual-wavelength photoplethysmography (PPG) signals. The preprocessing module is used to preprocess the red light channel signal and the infrared light channel signal respectively; A single-channel feature extraction module is used to extract features from the red light channel and the infrared light channel; A coupling feature construction module is used to construct candidate coupling features based on the red light channel features and the infrared light channel features; The target feature formation module is used to filter or combine the candidate coupled features to form a target feature set; the inference module is used to input the target feature set into the blood glucose estimation model and output the blood glucose estimate or blood glucose risk index.

9. A non-invasive blood glucose estimation device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of any one of claims 1 to 7.

Images