Continuous blood glucose monitoring signal denoising method based on two-factor layered adaptive wavelet threshold

Abstract

The invention provides a continuous blood glucose monitoring signal denoising method based on a two-factor layered adaptive wavelet threshold. The method comprises the following steps: firstly, performing discrete wavelet decomposition on an original noisy blood glucose signal; then, on the basis of noise characteristics of each scale, independently designing a threshold value for each layer of decomposition by adopting a hierarchical adaptive strategy; further, a novel two-factor nonlinear threshold function is introduced to perform contraction processing on the wavelet coefficient, a smooth transition factor controls function continuity near a threshold boundary to suppress a reconstruction ringing effect, and an amplitude retention factor adjusts the attenuation degree of a large-scale wavelet coefficient to retain key physiological features such as rapid change of blood glucose; and finally, reconstructing the de-noised signal through wavelet inverse transformation. According to the method, measurement noise and baseline drift are effectively suppressed, meanwhile, the fidelity capability of dynamic characteristics such as a blood glucose trend turning point is remarkably improved, and the inherent contradiction between excessive smoothness and noise residue in a traditional wavelet threshold method is relieved.

Description

A Denoising Method for Continuous Blood Glucose Monitoring Signals Based on Two-Factor Hierarchical Adaptive Wavelet Thresholding Technical Field

[0001] This invention belongs to the field of biomedical signal processing technology, and in particular relates to a method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold. Background Technology

[0002] Continuous Glucose Monitoring (CGM) technology uses subcutaneously implanted sensors to collect glucose concentrations in interstitial fluid in real time, generating time-series signals to provide dynamic, high-frequency data support for blood glucose management in diabetic patients. Compared to traditional finger-prick blood glucose meters, CGM can reveal blood glucose fluctuation trends, identify potential hyperglycemic and hypoglycemic events, and support the operation of closed-loop insulin infusion systems, making it irreplaceable in clinical and daily blood glucose management.

[0003] Then, CGM signals are inevitably affected by various interference sources during acquisition, resulting in noise in the raw signal, mainly including high-frequency electrochemical noise, low-frequency baseline drift, motion artifacts, and temperature drift. These noise components are non-stationary and multi-scale superposition characteristics, making it difficult for traditional linear filtering methods, such as Butterworth low-pass filtering, moving average, and Savitzky-Golay smoothing, to simultaneously achieve denoising effectiveness and signal fidelity. For example, excessive smoothing can weaken the slope information during rapid rises or falls in blood glucose, affecting clinical early warning; while insufficient filtering can leave a large amount of noise, reducing the accuracy of blood glucose prediction.

[0004] To overcome the aforementioned problems, some researchers have attempted to introduce wavelet transform into CGM signal processing. Wavelet analysis possesses excellent time-frequency localization capabilities, decomposing signals into sub-bands of different scales, thereby achieving the separation of noise and effective components. Among these, wavelet thresholding-based denoising methods are widely used due to their high computational efficiency and solid theoretical foundation. However, traditional wavelet thresholding methods employ a uniform threshold for hard or soft thresholding of wavelet coefficients across all scales, leading to the neglect of differences between scales. Furthermore, hard thresholding functions are discontinuous at the threshold point, easily introducing ringing artifacts; while soft thresholding functions are continuous, they exhibit a constant deviation for large coefficients, resulting in signal amplitude attenuation.

[0005] To address this, subsequent studies have proposed various improvement methods. For example, the SURE (Stein's Unbiased Risk Estimate) criterion is used to adaptively select the threshold, or a smoothing threshold function is used to alleviate discontinuities and amplitude distortion. However, these methods are still mostly designed for general signals and do not fully integrate the temporal characteristics of CGM signals and clinical needs.

[0006] In recent years, some studies have attempted to introduce deep learning models for end-to-end denoising. While these efforts have yielded some results, they rely on large amounts of labeled data, suffer from poor model interpretability, and are difficult to deploy in resource-constrained CGM systems. In contrast, wavelet-based analytical methods offer advantages such as fewer parameters, lower computational cost, and clear physical meaning, making them more suitable for integration into real-time CGM systems.

[0007] In summary, there is an urgent need for a denoising method that can effectively suppress multi-source noise while accurately preserving the dynamic characteristics of blood glucose. Summary of the Invention

[0008] To overcome the shortcomings of existing continuous glucose monitoring (CGM) signal denoising methods in terms of noise suppression capability, preservation of physiological details, and algorithm adaptability, this invention provides a CGM signal denoising method based on a two-factor hierarchical adaptive wavelet threshold. This method achieves precise suppression of multi-source noise in CGM signals by constructing a hierarchical dynamic threshold mechanism and an improved two-factor nonlinear threshold function, while maximizing the preservation of key dynamic features of blood glucose changes. It is particularly suitable for resource-constrained real-time CGM processing scenarios.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold, comprising the following steps:

[0010] Step S1: Acquire the raw continuous blood glucose monitoring signal S(k);

[0011] Step S2: For the original signal in step S1 Perform a 7-level discrete wavelet decomposition to obtain a set of wavelet coefficients. This includes the high-frequency wavelet coefficients from layer 1 to layer 7 and the low-frequency wavelet coefficients from layer 7.

[0012] Step S3: Based on the high-frequency wavelet coefficients of each layer, a hierarchical threshold estimation strategy is used to dynamically calculate the adaptive threshold for each layer, where the first... Adaptive threshold of the layer Calculate using the following formula:

[0013]

[0014] in, This is an estimate of the noise standard deviation of the high-frequency coefficients in the j-th layer. The sampling length of the blood glucose signal. ;

[0015] Step S4: Utilize the adaptive threshold from step S3 Using an improved two-factor nonlinear threshold function, the high-frequency wavelet coefficients of each layer are shrunk to obtain estimated wavelet coefficients. The improved two-factor nonlinear threshold function is defined as follows:

[0016]

[0017] in, For symbolic functions, As a smooth transition factor, As the amplitude retention factor, To prevent extremely small positive numbers from being divided by zero;

[0018] Step S5: Perform inverse wavelet transform on the processed high-frequency wavelet coefficients from layer 1 to layer 7 and the unprocessed low-frequency wavelet coefficients from layer 7 to reconstruct the denoised blood glucose signal. And truncated to the original signal Same length.

[0019] As a further aspect of the present invention, in step S1, the original continuous blood glucose monitoring signal is acquired by a micro-sensor implanted under the skin. The sensor can continuously detect the glucose level in the interstitial fluid of human tissue and generate a signal sequence, which includes high-frequency electrochemical noise, low-frequency baseline drift, motion artifacts and temperature drift.

[0020] As a further aspect of the present invention, in step S2, the wavelet decomposition employs the db4 wavelet basis function for the blood glucose signal. Perform a 7-level discrete wavelet transform.

[0021] As a further aspect of the present invention, in step S3, the hierarchical threshold estimation strategy makes the threshold vary with the layer number. The increase in frequency is monotonically increasing, thus applying stronger noise reduction to low-frequency components and being more tolerant of high-frequency components while retaining more details.

[0022] As a further aspect of the present invention, in step S4, the smooth transition factor Amplitude preservation factor used to control the continuity and smoothness of the function at threshold boundaries. The attenuation of large-scale wavelet coefficients is adjusted, and both are preset based on the physiological signal characteristics.

[0023] As a further aspect of the present invention, in step S5, the reconstructed signal Length and original signal It is strictly consistent, and the entire denoising process is completed independently within a single continuous monitoring segment of a single subject, without relying on external data or cross-segment information.

[0024] As a further aspect of the present invention, a mobile terminal device is provided, wherein a computer program is stored thereon, characterized in that: when the program is executed by a processor, it implements the aforementioned two-factor hierarchical adaptive wavelet denoising method.

[0025] As a further aspect of the present invention, a continuous glucose monitoring system is provided, characterized in that it comprises:

[0026] (1) An implantable blood glucose sensor for acquiring noisy continuous blood glucose signals;

[0027] (2) Bluetooth communication module, used to wirelessly transmit the signal to the mobile terminal;

[0028] (3) The mobile terminal is configured to execute the denoising method and output the denoised blood glucose signal.

[0029] Compared with existing methods, the present invention has the following advantages:

[0030] (1) By introducing a hierarchical thresholding strategy, the threshold increases monotonically with the increase of the wavelet decomposition layer. Experiments show that in CGM signals, low- and mid-frequency detail particles mainly carry sensor baseline drift, temperature drift, etc., and their interference energy is often higher than that of high-frequency white noise. This invention applies stronger threshold suppression to these layers in a targeted manner, effectively eliminating trend distortion. At the same time, because the highest layer approximation coefficient does not participate in thresholding, the true main trend of blood glucose is completely preserved.

[0031] (2) The proposed improved threshold function uses a smooth transition factor Eliminate discontinuities at the threshold boundaries of the hard threshold to avoid ringing artifacts in the reconstructed signal; use an amplitude preservation factor. Ensuring that the amplitude wavelet coefficients are fully preserved during the contraction process helps to address the constant bias problem of the soft threshold. The two factors work synergistically to effectively preserve key dynamic features such as the postprandial blood glucose rise and nocturnal trough while providing strong denoising, thus improving clinical interpretability.

[0032] (3) The entire denoising process relies only on the statistical characteristics of the wavelet coefficients of the current signal segment, without relying on external annotations or cross-subject models. The core of the algorithm is discrete wavelet transform, simple algebraic operations and exponential function calculation, which has low time complexity and can meet the engineering requirements of CGM system for low latency and high energy efficiency. It is especially suitable for resource-constrained embedded devices or mobile terminal deployments. Attached Figure Description

[0033] Figure 1 is a flowchart of a continuous blood glucose monitoring signal denoising method based on a two-factor hierarchical adaptive wavelet threshold.

[0034] Figure 2 shows a comparison of blood glucose signal denoising using hard thresholding, soft thresholding, compromise Garrote thresholding, and improved thresholding functions.

[0035] Figure 3 is a schematic diagram of the high-frequency coefficient distribution at different decomposition levels.

[0036] Figure 4 is a schematic diagram of the improved two-factor nonlinear threshold function curve of the present invention. Detailed Implementation

[0037] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the illustrated embodiments are merely one possible form of the present invention and are not intended to limit its scope of protection. Any equivalent changes or modifications derived by those skilled in the art based on conventional technical means without departing from the core concept of the present invention should be considered to fall within the scope of protection of the claims of the present invention.

[0038] In this specification, unless otherwise specified, all technical terms and expressions use their common meanings in the field of biomedical signal processing. Furthermore, the use of open-ended expressions such as "comprising" and "including" is solely for indicating the presence of the stated features, steps, or components, but does not preclude the presence or addition of one or more other technical features, steps, or components.

[0039] As shown in Figure 1, this embodiment discloses a method for denoising continuous blood glucose monitoring signals. The overall process is based on a two-factor hierarchical adaptive wavelet thresholding mechanism, and the method specifically includes the following steps:

[0040] Step S1: Acquire the raw continuous blood glucose monitoring signal S(k);

[0041] Step S2: For the original signal in step S1 Perform a 7-level discrete wavelet decomposition to obtain a set of wavelet coefficients. This includes the high-frequency wavelet coefficients from layer 1 to layer 7 and the low-frequency wavelet coefficients from layer 7.

[0042] Step S3: Based on the high-frequency wavelet coefficients of each layer, a hierarchical threshold estimation strategy is used to dynamically calculate the adaptive threshold for each layer, where the first... Adaptive threshold of the layer Calculate using the following formula:

[0043]

[0044] in, This is an estimate of the noise standard deviation of the high-frequency coefficients in the j-th layer. The sampling length of the blood glucose signal. ;

[0045] Step S4: Utilize the adaptive threshold from step S3 Using an improved two-factor nonlinear threshold function, the high-frequency wavelet coefficients of each layer are shrunk to obtain estimated wavelet coefficients. The improved two-factor nonlinear threshold function is defined as follows:

[0046]

[0047] in, For symbolic functions, As a smooth transition factor, As the amplitude retention factor, To prevent extremely small positive numbers from being divided by zero;

[0048] Step S5: Perform inverse wavelet transform on the processed high-frequency wavelet coefficients from layer 1 to layer 7 and the unprocessed low-frequency wavelet coefficients from layer 7 to reconstruct the denoised blood glucose signal. And truncated to the original signal Same length.

[0049] As a further aspect of the present invention, in step S1, the original continuous blood glucose monitoring signal is acquired by a micro-sensor implanted under the skin. The sensor can continuously detect the glucose level in the interstitial fluid of human tissue and generate a signal sequence, which includes high-frequency electrochemical noise, low-frequency baseline drift, motion artifacts and temperature drift.

[0050] As a further aspect of the present invention, in step S2, the wavelet decomposition employs the db4 wavelet basis function for the blood glucose signal. Perform a 7-level discrete wavelet transform.

[0051] As a further aspect of the present invention, in step S3, the hierarchical threshold estimation strategy makes the threshold vary with the layer number. The increase in frequency is monotonically increasing, thus applying stronger noise reduction to low-frequency components and being more tolerant of high-frequency components while retaining more details.

[0052] As a further aspect of the present invention, in step S4, the smooth transition factor Amplitude preservation factor used to control the continuity and smoothness of the function at threshold boundaries. The attenuation of large-scale wavelet coefficients is adjusted, and both are preset based on the physiological signal characteristics.

[0053] As a further aspect of the present invention, in step S5, the reconstructed signal Length and original signal It is strictly consistent, and the entire denoising process is completed independently within a single continuous monitoring segment of a single subject, without relying on external data or cross-segment information.

[0054] As a further aspect of the present invention, a mobile terminal device is provided, wherein a computer program is stored thereon, characterized in that: when the program is executed by a processor, it implements the aforementioned two-factor hierarchical adaptive wavelet denoising method.

[0055] As a further aspect of the present invention, a continuous glucose monitoring system is provided, characterized in that it comprises:

[0056] (1) An implantable blood glucose sensor for acquiring noisy continuous blood glucose signals;

[0057] (2) Bluetooth communication module, used to wirelessly transmit the signal to the mobile terminal;

[0058] (3) The mobile terminal is configured to execute the denoising method and output the denoised blood glucose signal.

[0059] Compared with existing methods, the present invention has the following advantages:

[0060] (1) By introducing a hierarchical thresholding strategy, the threshold increases monotonically with the increase of the wavelet decomposition layer. Experiments show that in CGM signals, low- and mid-frequency detail particles mainly carry sensor baseline drift, temperature drift, etc., and their interference energy is often higher than that of high-frequency white noise. This invention applies stronger threshold suppression to these layers in a targeted manner, effectively eliminating trend distortion. At the same time, because the highest layer approximation coefficient does not participate in thresholding, the true main trend of blood glucose is completely preserved.

[0061] (2) The proposed improved threshold function uses a smooth transition factor Eliminate discontinuities at the threshold boundaries of the hard threshold to avoid ringing artifacts in the reconstructed signal; use an amplitude preservation factor. Ensuring that the amplitude wavelet coefficients are fully preserved during the contraction process helps to address the constant bias problem of the soft threshold. The two factors work synergistically to effectively preserve key dynamic features such as the postprandial blood glucose rise and nocturnal trough while providing strong denoising, thus improving clinical interpretability.

[0062] (3) The entire denoising process relies only on the statistical characteristics of the wavelet coefficients of the current signal segment, without relying on external annotations or cross-subject models. The core of the algorithm is discrete wavelet transform, simple algebraic operations and exponential function calculation, which has low time complexity and can meet the engineering requirements of CGM system for low latency and high energy efficiency. It is especially suitable for resource-constrained embedded devices or mobile terminal deployments.

[0063] It is worth emphasizing that, as shown in Figure 2, for the same noisy blood glucose signal, the reconstruction results differ significantly after denoising using hard thresholding, soft thresholding, Garrote compromise thresholding, and the improved two-factor thresholding function proposed in this invention. Hard thresholding causes significant oscillations in the output signal due to discontinuous jumps at the threshold point; while soft thresholding ensures function continuity, it inevitably introduces amplitude compression, weakening the dynamic characteristics of rapid blood glucose changes; Garrote thresholding achieves a compromise between the two, but still exhibits slight distortion in peak or trough regions; in contrast, the two-factor thresholding function designed in this invention not only provides a smooth transition within the threshold neighborhood but also effectively preserves the energy of large-scale coefficients, enabling the denoised signal to highly restore the true blood glucose trend while suppressing noise.

[0064] Figure 3 illustrates the distribution of high-frequency detail subbands after performing a 7-level db4 wavelet decomposition on the actual CGM signal. It can be observed that while the detail coefficients in lower levels (such as levels 1 and 2) exhibit relatively sparse fluctuations, they display large amplitudes at abrupt changes, corresponding to impulsive interference such as instantaneous motion artifacts or sensor jumps. Higher levels (such as levels 6 and 7) show dense but small-amplitude oscillations, reflecting the high-frequency remnants of slowly varying interferences such as baseline drift and temperature perturbations in the wavelet domain. This phenomenon indicates that the noise in the CGM signal not only contains sparse large pulses but also widely distributed small-amplitude high-frequency perturbations. If a uniform threshold is used, the former is easily over-suppressed, leading to signal distortion, while the latter may remain due to insufficient thresholding. Therefore, this invention introduces a hierarchical adaptive thresholding strategy, combined with a two-factor nonlinear function, allowing the threshold to monotonically increase with the level number to eliminate dense small perturbations, while utilizing the amplitude preservation factor in the two-factor approach. This protects the large coefficients in the lower layers from excessive attenuation, thereby achieving synergistic suppression of different types of noise.

[0065] Furthermore, as shown in Figure 4, the improved two-factor nonlinear threshold function constructed in this invention has a smooth transition near the threshold point, and its overall shape is between hard threshold and soft threshold. This avoids the discontinuity of hard threshold and alleviates the excessive attenuation of large coefficients by soft threshold, thereby making the processed wavelet coefficients closer to the effective components of the real signal.

[0066] To quantitatively evaluate the denoising performance, mean square error (MSE), mean absolute error (MAE), and signal-to-noise ratio (SNR) were used as evaluation indicators to compare the effects of different denoising methods on the same test data. The results are shown in Table 1.

[0067] Table 1 Comparison of denoising effects of different algorithms

[0068] Denoising methods: MSEMAESNR soft threshold wavelet 0.970061 0.6308321 9.851136 hard threshold wavelet 0.971760 0.6316841 9.836788 Garrote compromise threshold 0.939311 0.6164692 0.105337 two-factor hierarchical adaptive wavelet (this invention) 0.888873 0.5963832 0.488643 surface

[0069] As shown in Table 1, compared with traditional soft thresholding, hard thresholding and Garrote compromise thresholding methods, the two-factor hierarchical adaptive wavelet denoising method proposed in this invention has lower MSE and MAE indices and higher SNR indices, indicating that it can better preserve the effective information of the original blood glucose signal while suppressing noise, and has better overall denoising performance.

[0070] In summary, this invention effectively resolves the contradiction between noise suppression and physiological detail preservation in existing methods by integrating a hierarchical thresholding strategy oriented towards CGM signal characteristics with an adjustable two-factor thresholding function. It is particularly suitable for real-time application scenarios with single independent monitoring and no external training data dependency.

[0071] The above embodiments are only used to illustrate the technical concept and features of the present invention, and their purpose is to enable those skilled in the art to understand and implement the present invention. Any modifications, equivalent substitutions, or improvements made in accordance with the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold, characterized in that, The sequence of steps is as follows: Step S1: Acquire the raw continuous glucose monitoring signal. Step S2: For the original signal in step S1 Perform a 7-level discrete wavelet decomposition to obtain a set of wavelet coefficients. This includes high-frequency wavelet coefficients from layers 1 to 7 and low-frequency wavelet coefficients from layer 7; Step S3: Based on the high-frequency wavelet coefficients of each layer, a hierarchical threshold estimation strategy is used to dynamically calculate the adaptive threshold for each layer, where the first layer... Adaptive threshold of the layer Calculate using the following formula: in, This is an estimate of the noise standard deviation of the high-frequency coefficients in the j-th layer. The sampling length of the blood glucose signal. Step S4: Utilize the adaptive threshold from step S3 Using an improved two-factor nonlinear threshold function, the high-frequency wavelet coefficients of each layer are shrunk to obtain estimated wavelet coefficients. The improved two-factor nonlinear threshold function is defined as follows: in, For symbolic functions, As a smooth transition factor, As the amplitude retention factor, To prevent extremely small positive numbers from being divided by zero; Step S5: Perform inverse wavelet transform on the processed high-frequency wavelet coefficients of layers 1 to 7 and the unprocessed low-frequency wavelet coefficients of layer 7 to reconstruct the denoised blood glucose signal. And truncated to the original signal Same length.

2. The method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold according to claim 1, characterized in that: In step S1, the original continuous blood glucose monitoring signal is acquired by a micro-sensor implanted under the skin. The sensor can continuously detect the glucose level in the interstitial fluid of human tissue and generate a signal sequence, which includes high-frequency electrochemical noise, low-frequency baseline drift, motion artifacts and temperature drift.

3. The method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold according to claim 1, characterized in that: In step S2, the wavelet decomposition uses the db4 wavelet basis function to analyze the blood glucose signal. Perform a 7-level discrete wavelet transform.

4. The method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold according to claim 1, characterized in that: In step S3, the hierarchical threshold estimation strategy makes the threshold vary with the layer number. The increase in frequency is monotonically increasing, thus applying stronger noise reduction to low-frequency components and being more tolerant of high-frequency components while retaining more details.

5. The method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold according to claim 1, characterized in that: In step S4, the smooth transition factor Amplitude preservation factor used to control the continuity and smoothness of the function at threshold boundaries. The attenuation of large-scale wavelet coefficients is adjusted, and both are preset based on the physiological signal characteristics.

6. The method for denoising continuous blood glucose monitoring signals based on a two-factor hierarchical adaptive wavelet threshold according to claim 1, characterized in that: In step S5, the reconstructed signal Length and original signal It is strictly consistent, and the entire denoising process is completed independently within a single continuous monitoring segment of a single subject, without relying on external data or cross-segment information.

7. A mobile terminal device having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the two-factor hierarchical adaptive wavelet denoising method as described in any one of claims 1 to 6.

8. A continuous glucose monitoring system, characterized in that, include: (1) An implantable blood glucose sensor for acquiring a noisy continuous blood glucose signal; (2) A Bluetooth communication module for wirelessly transmitting the signal to a mobile terminal; (3) The mobile terminal is configured to perform the denoising method as described in any one of claims 1 to 6 and output a denoised blood glucose signal.

Images