BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning

Abstract

The invention provides a BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning, and relates to the technical field of biomedical signal processing, and the method comprises the steps: dynamically adjusting the signal collection sensitivity through a high-sensitivity flexible piezoelectric sensor, optimizing adaptive gain control in combination with the body weight of a user and mattress material parameters, and generating a de-noised BCG time domain sequence; extracting a dynamic signal complexity coefficient (DSCC) and a physiological state confusion index (PSCI), and quantizing a signal nonlinear characteristic and a cardiac and respiratory coupling relationship; constructing a semi-supervised deep learning model fusing a convolutional neural network and a graph attention mechanism, and training a feature extraction network by using limited label samples and augmented unlabeled data; establishing a topological correlation model based on a medical calibration database, and analyzing the time-frequency domain characteristics of the BCG signal in combination with a persistent coherence algorithm; and integrating the electronic medical record data of the user, and generating a personalized health assessment report through an attention mechanism and transfer learning. And the screening precision is improved through dynamic signal adaptation, semi-supervised learning and data optimization.

Description

Technical Field

[0001] The present invention relates to the field of biomedical signal processing technology, and in particular to a BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning. Background Art

[0002] In the field of sleep monitoring and cardiovascular health analysis based on BCG (Ballistocardiogram), existing technologies face multiple limitations. First, during the BCG signal acquisition process, the sensitivity of the sensor is fixed and cannot be dynamically adjusted according to individual differences of users (such as weight changes) and external environmental factors (such as differences in mattress materials), resulting in large fluctuations in signal amplitude and low signal-to-noise ratio. Especially in the case of obese users or soft mattresses, key physiological information is easily submerged by noise. Secondly, traditional signal preprocessing methods (such as fixed threshold filtering or frequency domain filtering) are difficult to effectively handle the non-stationary characteristics of BCG signals. Although technologies such as phase space reconstruction have certain applications, the parameter selection relies on experience, lacks an adaptive adjustment mechanism, and the denoising effect is unstable.

[0003] Furthermore, existing feature extraction methods are mostly limited to manual features in the time or frequency domain (such as heart rate and respiratory rate), and the nonlinear dynamic characteristics of the signal (such as complexity and rhythm coupling) are insufficiently characterized, resulting in limited accuracy in subsequent health status assessments. In addition, the construction of medical calibration models relies heavily on large amounts of labeled data. However, obtaining BCG samples labeled by gold standard equipment (such as polysomnography) is expensive, and the labeling process is time-consuming, which limits the large-scale application of the model. Existing supervised learning algorithms are prone to overfitting in small sample scenarios, making it difficult to balance model generalization ability and prediction accuracy.

[0004] Finally, existing health assessment systems generally lack personalized adaptation capabilities and fail to fully integrate users' historical medical data (such as medical history and medication records in electronic medical records), resulting in deviations between risk assessment results and an individual's actual health status. For example, cardiovascular disease risk models fail to consider the interactive effects of a user's age, gender, and medical history, resulting in overly broad warning thresholds and an inability to accurately identify high-risk groups. Summary of the Invention

[0005] In order to solve the technical problems in the existing technology such as fixed sensor sensitivity leading to unstable signal quality, poor adaptability of traditional denoising methods to non-stationary signals, lack of multi-scale nonlinear characterization capability of feature extraction, scarcity of labeled data restricting model training, insufficient generalization capability of supervised learning, and failure to integrate personalized medical data into health assessment, the present invention provides a medical calibration and analysis method for BCG sleep signals based on semi-supervised deep learning.

[0006] The technical solutions provided by the present invention are as follows:

[0007] The present invention provides a BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning, including:

[0008] S1. The raw BCG signal is collected through a highly sensitive flexible piezoelectric sensor. The sensor is constructed based on PVDF resin and metal electrode layers, and the signal acquisition sensitivity is dynamically adjusted according to the user's weight and mattress material.

[0009] S2. Preprocessing the original BCG signal, including signal denoising based on phase space reconstruction and adaptive gain control, to generate a denoised BCG time domain sequence;

[0010] S3. Extract multi-scale features from the preprocessed BCG signal, including the dynamic signal complexity coefficient (DSCC) and the physiological state confusion index (PSCI). DSCC is used to quantify the nonlinear characteristics of the signal, and PSCI is used to characterize the coupling relationship between heart rate and respiratory rhythm.

[0011] S4. Build a semi-supervised deep learning model, using limited labeled samples and augmented unlabeled samples to train a feature extraction network, which combines a convolutional neural network with a graph attention mechanism to generate a high-dimensional feature vector;

[0012] S5. Map high-dimensional feature vectors based on a medical calibration database, establish a correlation model between BCG signals and cardiovascular health status through topological analysis, and output sleep quality scores and cardiovascular disease risk levels;

[0013] S6. Optimize the association model based on the user's personalized electronic medical record data and generate a dynamic health assessment report.

[0014] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0015] (1) In this invention, the signal acquisition sensitivity is dynamically adjusted by a highly sensitive flexible piezoelectric sensor, and gain control is optimized in real time based on the user's weight and mattress material parameters, significantly improving the signal-to-noise ratio and stability of the BCG signal. In response to the non-stationary signal characteristics, phase space reconstruction and adaptive filtering techniques are used to effectively eliminate high-frequency noise and baseline drift, solving the signal amplitude fluctuation problem caused by the fixed sensitivity of traditional sensors. This ensures that high-quality physiological signals can be obtained for users of different body shapes and mattress environments, providing a reliable data foundation for subsequent analysis.

[0016] (2) In this paper, the nonlinear dynamic characteristics of the BCG signal are comprehensively characterized by extracting the dynamic signal complexity coefficient (DSCC) and the physiological state confusion index (PSCI), combined with multi-scale permutation entropy and heart rate-respiration coupling modeling. Based on a semi-supervised deep learning framework, the feature extraction network is trained using limited labeled samples and augmented unlabeled data, breaking through the traditional supervised learning's high dependence on labeled data, enhancing the model's generalization ability in small sample scenarios, and significantly improving the accuracy of sleep quality assessment and cardiovascular disease risk identification.

[0017] (3) In this invention, personalized health risk assessment is achieved by integrating the user's personalized electronic medical record data, using an attention mechanism to weightedly fuse medical history features with BCG feature vectors, and optimizing the cardiovascular health association model based on transfer learning. This method overcomes the drawback of existing health assessment models that are disconnected from the user's actual medical background, accurately associates BCG signal features with specific disease risks, and provides customized solutions for sleep disorder monitoring and early screening of cardiovascular disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A schematic diagram of a process for the medical calibration and analysis of BCG sleep signals based on semi-supervised deep learning provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of the calculation process of the dynamic signal complexity coefficient in the BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning provided by an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of the process for generating cardiovascular disease risk levels in the BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning provided in an embodiment of the present invention;

[0022] Figure 4 A schematic diagram of the personalized optimization process in the BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0024] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0025] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0026] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0027] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0028] Reference Manual Figure 1 , which shows a flow chart of the BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning provided by an embodiment of the present invention.

[0029] The embodiment of the present invention provides a BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning. The processing flow may include the following steps:

[0030] S1. The original BCG signal is collected through a highly sensitive flexible piezoelectric sensor. The sensor is constructed based on PVDF resin and metal electrode layers, and the signal acquisition sensitivity is dynamically adjusted according to the user's weight and mattress material.

[0031] It should be noted that the original BCG (Ballistocardiogram) signal is collected by a highly sensitive flexible piezoelectric sensor. The sensor is composed of alternating PVDF (Polyvinylidene Fluoride) resin layers and metal electrode layers. The metal electrode layers are deposited on the surface of the PVDF film using a sputtering process to enhance signal conduction efficiency. The sensor is integrated into the mattress and dynamically adjusts its sensitivity based on the user's weight and mattress material. Specifically, the embedded microcontroller reads the mattress pressure distribution data in real time and adjusts the piezoelectric unit sampling frequency based on a preset weight-sensitivity mapping table. When the user's weight exceeds the threshold or the mattress material density changes, the system automatically switches to high-frequency sampling mode to improve signal resolution.

[0032] S2. Preprocess the original BCG signal, including signal denoising based on phase space reconstruction and adaptive gain control, to generate a denoised BCG time domain sequence.

[0033] In one possible implementation, the adaptive gain control in S2 further includes:

[0034] S201. Calculate the initial gain value G0 based on the user's weight and mattress material parameters. The formula is:

[0035]

[0036] Where W is the user's weight in kg, and D is the mattress density in kg / m 3 , S is the sensor sensitivity level, k1 and k2 are preset calibration coefficients;

[0037] S202 : Dynamically adjust the gain value based on the real-time signal amplitude to ensure that the peak-to-peak value of the signal is within a preset range.

[0038] It should be noted that the original BCG signal is preprocessed. First, phase space reconstruction technology is used to remove noise. The specific method is as follows: the original signal is embedded in a high-dimensional phase space using the time delay method. The time delay τ is calculated using the mutual information method, and the embedding dimension m is determined using the false neighborhood method. The reconstructed phase space trajectory is then subjected to local projection filtering to eliminate high-frequency interference. Adaptive gain control is then performed. The initial gain value G0 is calculated based on the user's weight W, mattress density D, and sensor sensitivity level S. In the calculation formula, k1 and k2 are determined to be 0.78 and 0.15 through calibration experiments. During the real-time signal processing stage, if the peak-to-peak value of the signal exceeds the preset range (0.5V to 2.0V), the gain coefficient is adjusted every 10ms, with the adjustment amplitude proportional to the degree of deviation from the current amplitude, until the signal stabilizes within the target range.

[0039] S3. Extract multi-scale features from the preprocessed BCG signal, including the dynamic signal complexity coefficient (DSCC) and the physiological state confusion index (PSCI). DSCC is used to quantify the nonlinear characteristics of the signal, and PSCI is used to characterize the coupling relationship between heart rate and respiratory rhythm.

[0040] In a possible implementation, S3 further includes:

[0041] The calculation of the dynamic signal complexity coefficient DSCC adopts a multi-scale permutation entropy algorithm based on information entropy. The permutation entropy formula is defined as follows:

[0042]

[0043] Among them, m is the embedding dimension, p i is the probability distribution of the arrangement pattern.

[0044] In one possible implementation, Figure 2 As shown, the calculation of the dynamic signal complexity coefficient DSCC further includes:

[0045] S301, performing multi-scale wavelet decomposition on the pre-processed BCG signal;

[0046] S302, calculating the permutation entropy value at each scale and normalizing it;

[0047] S303 : Generate a final DSCC value through weighted summation, where the weight is positively correlated with the scale frequency.

[0048] Specifically, the calculation of the dynamic signal complexity coefficient DSCC includes the following steps:

[0049] The signal is decomposed by multi-scale wavelet, using Daubechies 4 wavelet basis function and 5 decomposition levels, corresponding to the frequency band of 0.5Hz to 40Hz.

[0050] Calculate the permutation entropy value H for each scale subband signal p (m), the embedding dimension m is set to 6, the time window length is 5 seconds, the sliding step is 1 second, and the arrangement pattern probability distribution p i It is obtained by counting the frequency of occurrence of the arrangement pattern of each sub-band signal segment;

[0051] The entropy values of each scale are normalized and then weighted and summed to generate the DSCC. The weight coefficient is positively correlated with the scale frequency, with the high-frequency sub-band weight being 0.4 and the low-frequency sub-band weight being 0.1.

[0052] In a possible implementation, S3 further includes:

[0053] The calculation formula of the Physiological State Confusion Index (PSCI) is:

[0054]

[0055] Among them, HRV SDNN is the standard deviation of heart rate variability, which is calculated by the standard deviation of 30 consecutive cardiac cycles, in milliseconds; ΔR is the respiratory rhythm variability, which is the root mean square of the difference between adjacent respiratory cycles, R avg is the average respiratory rate in the sliding window in BPM, α and β are normalized weight coefficients, and the weights are determined to be 0.6 and 0.4 respectively through principal component analysis.

[0056] S4. Build a semi-supervised deep learning model, use limited labeled samples and augmented unlabeled samples to train the feature extraction network, and combine the convolutional neural network with the graph attention mechanism to generate high-dimensional feature vectors.

[0057] In one possible implementation, the semi-supervised deep learning model in S4 further includes:

[0058] The semi-supervised deep learning model adopts the Mean Teacher framework, in which the teacher model updates the student model parameters through exponential moving average, and the consistency regularization loss function is as follows:

[0059]

[0060] Where x is the input signal, f student 、f teacher Outputs for the student model and teacher model respectively.

[0061] Specifically, the feature extraction network consists of a three-layer one-dimensional convolutional neural network (CNN) and a graph attention network (GAT). The number of filters in the CNN layer is 64, 128, and 256, respectively, the convolution kernel length is 7, and the step size is 2. The graph attention mechanism regards the signal segments as graph nodes, the node features are CNN output vectors, and the edge weights are determined by the cosine similarity between the signal segments. The model training adopts the Mean Teacher framework, and the teacher model parameters are updated by the exponential moving average of the student model with an update coefficient of 0.99. The loss function includes cross entropy loss and consistency regularization loss. The consistency loss is calculated as the mean square error of the output feature vectors of the teacher model and the student model. The optimizer uses Adam, and the initial learning rate is 1e-4

[0062] S5. Map high-dimensional feature vectors based on the medical calibration database, establish a correlation model between BCG signals and cardiovascular health status through topological analysis, and output sleep quality scores and cardiovascular disease risk levels.

[0063] In a possible implementation, S5 specifically includes:

[0064] The topological analysis uses the persistent homology algorithm to calculate the topological invariance characteristics of BCG signals in the time-frequency domain and establish a Betti number mapping relationship related to cardiovascular disease.

[0065] In one possible implementation, the medical calibration database specifically includes:

[0066] The sleep apnea event signature, validated by gold-standard equipment, was associated with the phase mutation characteristics of the BCG signal, and the association threshold was determined by receiver operating characteristic (ROC) curve analysis.

[0067] In one possible implementation, Figure 3 As shown, the generation of cardiovascular disease risk levels further includes:

[0068] S501, detecting abnormal cardiac cycle patterns according to the topological analysis results;

[0069] S502. Construct a Cox proportional hazard model based on user age and gender;

[0070] S503. Output the probability of cardiovascular events in the next five years and risk intervention recommendations.

[0071] Specifically, high-dimensional feature vectors were mapped based on a medical calibration database. The database contains 3,000 polysomnography (PSG)-verified BCG samples, each labeled with sleep apnea events and cardiovascular abnormality labels. Topological analysis uses the persistent homology algorithm to extract topological features in the time-frequency domain of the BCG signal, including the 0-dimensional Betti number (number of connected components) and the 1-dimensional Betti number (number of ring structures). The feature vectors are associated with cardiovascular disease labels using a support vector machine (SVM) classifier. The classification threshold is determined by ROC (Receiver Operating Characteristic) curve analysis, and the sensitivity is set at 95%.

[0072] S6. Optimize the association model based on the user's personalized electronic medical record data and generate a dynamic health assessment report.

[0073] In one possible implementation, Figure 4As shown, the personalized optimization in S6 specifically includes:

[0074] S601, extracting medical history keywords and medication records from electronic medical records;

[0075] S602: Weighted fusion of medical history features and BCG feature vectors through attention mechanism;

[0076] S603: Fine-tune the association model parameters using transfer learning.

[0077] Specifically, the model is optimized based on the user's personalized electronic medical record data. First, medical history keywords (such as hypertension, diabetes) and medication records (such as beta-blockers) are extracted from the electronic medical record and converted into a 128-dimensional vector through a word embedding model. Then, a multi-head attention mechanism is used to fuse the medical record features with the BCG feature vector, with 4 attention heads and 64 key-value dimensions. The fused features are input into the pre-trained Cox proportional hazard model, and the model parameters are fine-tuned through transfer learning. During fine-tuning, the underlying network weights are frozen, and only the fully connected layer parameters are optimized. Finally, a dynamic health assessment report is output, including a sleep quality score (0 to 100 points) and the risk level of cardiovascular events in the next 5 years (low, medium, and high).

[0078] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0079] (1) In this invention, the signal acquisition sensitivity is dynamically adjusted by a highly sensitive flexible piezoelectric sensor, and gain control is optimized in real time based on the user's weight and mattress material parameters, significantly improving the signal-to-noise ratio and stability of the BCG signal. In response to the non-stationary signal characteristics, phase space reconstruction and adaptive filtering techniques are used to effectively eliminate high-frequency noise and baseline drift, solving the signal amplitude fluctuation problem caused by the fixed sensitivity of traditional sensors. This ensures that high-quality physiological signals can be obtained for users of different body shapes and mattress environments, providing a reliable data foundation for subsequent analysis.

[0080] (2) In this paper, the nonlinear dynamic characteristics of the BCG signal are comprehensively characterized by extracting the dynamic signal complexity coefficient (DSCC) and the physiological state confusion index (PSCI), combined with multi-scale permutation entropy and heart rate-respiration coupling modeling. Based on a semi-supervised deep learning framework, the feature extraction network is trained using limited labeled samples and augmented unlabeled data, breaking through the traditional supervised learning's high dependence on labeled data, enhancing the model's generalization ability in small sample scenarios, and significantly improving the accuracy of sleep quality assessment and cardiovascular disease risk identification.

[0081] (3) In this invention, personalized health risk assessment is achieved by integrating the user's personalized electronic medical record data, using an attention mechanism to weightedly fuse medical history features with BCG feature vectors, and optimizing the cardiovascular health association model based on transfer learning. This method overcomes the drawback of existing health assessment models that are disconnected from the user's actual medical background, accurately associates BCG signal features with specific disease risks, and provides customized solutions for sleep disorder monitoring and early screening of cardiovascular disease.

[0082] The above content is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0083] There are a few points to note:

[0084] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0085] (2) For the sake of clarity, the thickness of layers or regions in the drawings used to describe the embodiments of the present invention are exaggerated or reduced, that is, these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region, or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or intervening elements may be present.

[0086] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to form new embodiments.

[0087] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning, characterized by: include: S1. The raw BCG signal is collected through a highly sensitive flexible piezoelectric sensor. The sensor is constructed based on PVDF resin and metal electrode layers, and the signal acquisition sensitivity is dynamically adjusted according to the user's weight and mattress material. S2. Preprocessing the original BCG signal, including signal denoising based on phase space reconstruction and adaptive gain control, to generate a denoised BCG time domain sequence; S3. Extract multi-scale features from the preprocessed BCG signal, including the dynamic signal complexity coefficient (DSCC) and the physiological state confusion index (PSCI). DSCC is used to quantify the nonlinear characteristics of the signal, and PSCI is used to characterize the coupling relationship between heart rate and respiratory rhythm. S4. Build a semi-supervised deep learning model, using limited labeled samples and augmented unlabeled samples to train a feature extraction network, which combines a convolutional neural network with a graph attention mechanism to generate a high-dimensional feature vector; S5. Map high-dimensional feature vectors based on a medical calibration database, establish a correlation model between BCG signals and cardiovascular health status through topological analysis, and output sleep quality scores and cardiovascular disease risk levels; S6. Optimize the association model based on the user's personalized electronic medical record data and generate a dynamic health assessment report.

2. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1 is characterized in that: The adaptive gain control in S2 further includes: S201. Calculate the initial gain value G0 based on the user's weight and mattress material parameters. The formula is: Where W is the user's weight in kg, and D is the mattress density in kg / m 3 , S is the sensor sensitivity level, k1 and k2 are preset calibration coefficients; S202 : Dynamically adjust the gain value based on the real-time signal amplitude to ensure that the peak-to-peak value of the signal is within a preset range.

3. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1 is characterized in that: Said S3 further comprises: The dynamic signal complexity coefficient DSCC is calculated using a multi-scale permutation entropy algorithm based on information entropy, and the permutation entropy formula is defined as follows: Among them, m is the embedding dimension, p i is the probability distribution of the arrangement pattern.

4. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1, characterized in that: Said S3 further comprises: The calculation formula of the physiological state confusion index PSCI is: Among them, HRV SDNN is the standard deviation of heart rate variability, in ms, ΔR is the respiratory rhythm variability, R avg is the average respiratory rate in BPM, and α and β are normalized weight coefficients.

5. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1 is characterized in that: The semi-supervised deep learning model in S4 further includes: The semi-supervised deep learning model adopts the Mean Teacher framework, in which the teacher model updates the student model parameters through exponential moving average, and the consistency regularization loss function is as follows: Where x is the input signal, f student 、f teacher Outputs for the student model and teacher model respectively.

6. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1, characterized in that: The S5 specifically includes: The topological analysis adopts a persistent homology algorithm and establishes a Betti number mapping relationship related to cardiovascular disease by calculating the topological invariance characteristics of BCG signals in the time-frequency domain.

7. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1, characterized in that: The personalized optimization in S6 specifically includes: S601, extracting medical history keywords and medication records from electronic medical records; S602: Weighted fusion of medical history features and BCG feature vectors through attention mechanism; S603: Fine-tune the association model parameters using transfer learning.

8. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 3 is characterized in that: The calculation of the dynamic signal complexity coefficient DSCC further includes: S301, performing multi-scale wavelet decomposition on the pre-processed BCG signal; S302, calculating the permutation entropy value at each scale and normalizing it; S303 : Generate a final DSCC value through weighted summation, where the weight is positively correlated with the scale frequency.

9. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1, characterized in that: The medical calibration database specifically includes: The sleep apnea event signature, validated by gold-standard equipment, was associated with the phase mutation characteristics of the BCG signal, and the association threshold was determined by receiver operating characteristic (ROC) curve analysis.

10. The BCG sleep signal medical calibration and analysis method based on semi-supervised deep learning according to claim 1, characterized in that: The generation of the cardiovascular disease risk level further includes: S501, detecting abnormal cardiac cycle patterns according to the topological analysis results; S502. Construct a Cox proportional hazard model based on user age and gender; S503. Output the probability of cardiovascular events in the next five years and risk intervention recommendations.

Images