Anea early warning model and device based on blood oxygen and snore

Abstract

The invention relates to the technical field of sleep monitoring, and provides an apnea early warning model and device based on blood oxygen and snore. The method of the model comprises the following steps: acquiring physiological data from a public database, and preprocessing the physiological data to obtain discrete physiological data; performing heterogeneous feature extraction on the discrete physiological data to obtain recognized physiological data; performing space-time fusion on the recognized physiological data to obtain an initial early warning model; performing joint optimization on the initial early warning model to obtain a final early warning model; and measuring physiological data of the patient, and inputting the physiological data of the patient into the final early warning model to obtain an early warning result. According to the invention, the limitation of traditional single parameter monitoring is broken through, the equipment has data processing and analysis capabilities, real-time early warning and long-term health assessment can be provided for users, and a new solution is provided for daily monitoring and early intervention. The technical effect of providing a new solution for daily monitoring and early intervention of the OSAHS is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of sleep monitoring, and provides a sleep apnea warning model and device based on blood oxygen and snoring. Background Art

[0002] Obstructive Sleep Apnea Hypopnea Syndrome (OSAHS) is a common sleep-disordered breathing disorder with a high incidence and high risk of serious consequences. It can lead to serious health consequences and even threaten patients' lives. Therefore, developing an effective sleep apnea monitoring and real-time early warning system has become an urgent issue.

[0003] Among existing clinical diagnostic methods, polysomnography (PSG) is commonly used for the definitive diagnosis of OSAHS. PSG comprehensively evaluates the patient's sleep state and respiratory function by synchronously monitoring multiple physiological parameters including electroencephalogram (EEG), oculogram (EOG), mandibular electromyogram (EMG), nasal and oral airflow, blood oxygen saturation (SpO2), respiratory rate, electrocardiogram (ECG), etc. However, polysomnography equipment is bulky, expensive, and requires monitoring in a professional sleep laboratory. Its complex connection methods and environmental limitations may interfere with the patient's natural sleep state, thereby affecting the accuracy of the monitoring results.

[0004] In recent years, some domestic and international medical companies have been committed to exploring more convenient sleep apnea monitoring methods. These methods typically collect data from single or multiple locations, such as the mouth, nose, chest, or lungs, to simplify the monitoring process and reduce equipment costs. Although such methods have reduced the size of the equipment and improved portability to a certain extent, their detection locations still have significant limitations. For example, sensors at the mouth and nose may obstruct normal respiratory airflow, while chest monitoring data is easily interfered with by heartbeat signals, thus affecting the reliability of the monitoring results.

[0005] Meanwhile, electronic technology companies both domestically and internationally are using technologies such as blood oxygen saturation monitoring and wrist motion detection to predict sleep apnea. However, these methods typically rely on single or limited physiological parameters, failing to fully reflect a patient's apnea status. They also suffer from issues such as delayed results and low accuracy, failing to meet clinical needs for real-time early warning and precise diagnosis.

[0006] Therefore, existing monitoring methods have certain limitations in terms of equipment size, cost, portability and detection accuracy. In order to promote the further development and application of OSAHS monitoring technology, it is imperative to develop an innovative solution that integrates multi-dimensional physiological parameter monitoring with high precision and real-time warning functions. Summary of the Invention

[0007] The present invention aims to address at least one of the technical problems existing in the related art. To this end, the present invention provides a sleep apnea warning method and device based on blood oxygen and snoring. This method overcomes the limitations of traditional single-parameter monitoring. The device combines data processing and analysis capabilities, providing users with real-time warnings and long-term health assessments, offering a new solution for daily monitoring and early intervention.

[0008] The present invention provides an apnea warning model based on blood oxygen and snoring. The training method of the model is as follows: S1: Acquire physiological data, and preprocess the physiological data to obtain discrete physiological data; S2: extracting heterogeneous features from the discrete physiological data to obtain recognized physiological data; S3: performing spatiotemporal fusion on the identified physiological data to obtain an initial warning model; S4: Jointly optimize the initial warning model to obtain a final warning model.

[0009] According to the present invention, a sleep apnea warning model based on blood oxygen and snoring is provided, wherein the physiological data includes electroencephalogram, oculogram, mandibular electromyogram, nasal and oral airflow, blood oxygen saturation, respiratory rate, electrocardiogram and snoring data.

[0010] According to the present invention, a sleep apnea warning model based on blood oxygen and snoring is provided, step S1 includes: S11: Perform standard deviation screening on the physiological data to determine the values exceeding the mean. The data points in the range are marked as outliers, and the outliers are eliminated to obtain the first physiological data; S12: performing singular value decomposition on the first physiological data to obtain second physiological data; S13: Perform dynamic sample adjustment on the second physiological data to obtain discrete physiological data.

[0011] According to the apnea warning model based on blood oxygen and snoring provided by the present invention, step S13 includes: S131: Design default event time; S132: Design sample length; S133: discretizing the second physiological data according to the default event time, and splicing the discretized second physiological data according to the sample length to obtain discrete physiological data.

[0012] According to the present invention, a sleep apnea warning model based on blood oxygen and snoring is provided, and step S2 of the training method includes: S21: extracting low-frequency modal data from the discrete physiological data using a shallow network with a large convolution kernel to obtain first recognized physiological data, wherein the low-frequency modal data includes nasal and oral airflow, blood oxygen saturation, respiratory rate, and electrocardiogram; S22: extracting high-frequency modal data from the discrete physiological data using a deep network with a small convolution kernel to obtain second recognition physiological data, wherein the high-frequency modal data includes electroencephalogram, eye movement graph, mandibular electromyogram, and snoring data; S23: The first recognition physiological data and the second recognition physiological data are combined to obtain recognition physiological data.

[0013] According to the apnea warning model based on blood oxygen and snoring provided by the present invention, step S3 includes: S31: Aligning the first recognized physiological data with the second recognized physiological data to obtain unified input data; S32: Input the unified input data into a bidirectional two-layer LSTM to perform time series modeling to obtain an initial warning model.

[0014] According to the present invention, a sleep apnea warning model based on blood oxygen and snoring is provided, and step S4 of the training method includes: S41: Design a localization head, which is used to make the initial warning model more focused on capturing changes in event boundaries, thereby improving localization accuracy by biasing gradient weights to channels related to waveform mutations; S42: Design a classification head that enables the initial warning model to more accurately identify event types, thereby improving classification performance by shifting gradient weights to channels related to the blood oxygen decline trend; S43: Designing an optimization function based on the positioning head and the classification head : in, is the first hyperparameter, is the positioning head loss function, is the second hyperparameter, is the classification head loss function, when the optimization function The initial warning model corresponding to the minimum is the final warning model.

[0015] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the apnea warning model based on blood oxygen and snoring as described above are implemented.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of any of the above-mentioned apnea warning methods based on blood oxygen and snoring are implemented.

[0017] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: The sleep apnea warning method and device based on blood oxygen and snoring provided by the present invention, first, uses two-dimensional data of blood oxygen saturation and snoring for monitoring, breaking through the limitations of traditional single parameter monitoring; second, the monitoring device is designed to be worn on the ear, which not only avoids the interference of traditional devices on sleep, but also can more accurately collect blood oxygen data through the posterior auricular artery; finally, the device has both data processing and analysis capabilities, can provide users with real-time warnings and long-term health assessments, and provides a new solution for daily monitoring and early intervention of OSAHS.

[0018] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 This is a training flow chart of the apnea warning model based on blood oxygen and snoring provided by the present invention.

[0021] Figure 2 It is a structural schematic diagram of the electronic device provided by the present invention.

[0022] Figure 3 Graph of the loss function for training with the localization head and the classification head.

[0023] Figure 4 This is a model index diagram of the final model corresponding to the method provided by the present invention.

[0024] Reference numerals: 810 , processor; 820 , communication interface; 830 , memory; 840 , communication bus. DETAILED DESCRIPTION

[0025] To make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below. Obviously, the embodiments described are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0026] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0027] The following combination Figures 1 to 4 The present invention is described.

[0028] Example like Figure 1 As shown, Figure 1 The training flow chart of the apnea warning model based on blood oxygen and snoring provided by the present invention includes the following steps: S1: Acquire physiological data, and preprocess the physiological data to obtain discrete physiological data; S2: extracting heterogeneous features from the discrete physiological data to obtain recognized physiological data; S3: performing spatiotemporal fusion on the identified physiological data to obtain an initial warning model; S4: Jointly optimize the initial warning model to obtain a final warning model.

[0029] Specifically, physiological data includes but is not limited to electroencephalogram, oculogram, mandibular electromyogram, nasal and oral airflow, blood oxygen saturation, respiratory rate, electrocardiogram and snoring data.

[0030] Specifically, step S1 includes: S11: Perform standard deviation screening on the physiological data to determine the values exceeding the mean. Data points within the specified range are marked as outliers, and these are removed to obtain the primary physiological data. This statistically based approach effectively identifies and removes extreme outliers caused by sudden interference or equipment failure. After removing outliers, signal stability and consistency are significantly improved, laying the foundation for subsequent noise reduction.

[0031] S12: Perform singular value decomposition on the first physiological data to obtain second physiological data.

[0032] The embodiments of the present invention use a singular value decomposition (SVD) method to reduce noise. SVD is a linear algebra-based signal processing method that decomposes the original signal into multiple singular values and corresponding singular vectors. It effectively separates the useful information in the signal from the noise components, retaining the main singular values and discarding smaller singular values. After noise reduction, high-quality data input is provided for subsequent analysis and model training. By reconstructing the signal, a smoother and more accurate physiological signal can be obtained.

[0033] S13: Perform dynamic sample adjustment on the second physiological data to obtain discrete physiological data.

[0034] Wherein, step S13 includes: S131: Designing default event time ; S132: Design sample length ; S133: According to the default event time Discretize the second physiological data according to the sample length The discretized second physiological data are spliced to obtain discrete physiological data.

[0035] The embodiment of the present invention further designs a model training strategy based on the characteristics of sleep apnea events. According to the observation of the labeled data, the duration of each sleep apnea event has a large uncertainty, usually ranging from 10 seconds to 100 seconds. In order to adapt to this characteristic, the embodiment of the present invention will seconds is defined as a default event and every The default event (i.e. A time window of 100 seconds (100 seconds) is fed into the model for training. This segmentation strategy not only captures the dynamics of apnea events but also improves the model's sensitivity to short-duration events. During training, the default number of events in the sample can be dynamically adjusted based on the model's output to further optimize the model's accuracy and generalization.

[0036] Specifically, step S2 includes: S21: extracting low-frequency modal data from the discrete physiological data using a shallow network with a large convolution kernel to obtain first recognized physiological data, wherein the low-frequency modal data includes nasal and oral airflow, blood oxygen saturation, respiratory rate, and electrocardiogram; In this embodiment of the present invention, a three-layer shallow network with large convolution kernels is used for feature extraction. This shallow network, while not overly focused on short-term fluctuations, has a relatively simple structure and can effectively extract the overall trend of low-frequency signals. For blood oxygen saturation signals, the third layer of the shallow network alone can reach 30 seconds, sufficient to cover a complete cycle of blood oxygen saturation changes. This design prevents the model from being bogged down by short-term noise, instead focusing on the overall downward trend of blood oxygen saturation, thereby improving the robustness of feature extraction. Large convolution kernels have a wider field of view on the time axis, capturing relevant conditions in low-frequency signals over a longer period of time. For blood oxygen saturation signals, large convolution kernels effectively smooth short-term fluctuations and highlight the overall trend of signal change. For example, during apnea events, blood oxygen saturation typically exhibits a slowly decreasing trend. While a small convolution kernel may be affected by short-term fluctuations, resulting in inaccurate feature extraction, a large convolution kernel can accurately identify such changes.

[0037] S22: extracting high-frequency modal data in the discrete physiological data using a deep network with a small convolution kernel to obtain second recognition physiological data, wherein the high-frequency modal data includes electroencephalogram, eye movement graph, mandibular electromyogram and snoring data.

[0038] In the embodiment of the present invention, a 6-layer deep network is used, and a small convolution kernel is used for feature extraction.

[0039] Through multiple layers of nonlinear transformations, deep networks can gradually extract high-level signal features and fit the complex changing patterns of high-frequency signals. For EEG and snoring signals, the receptive field of deep networks can range from a few seconds to over ten seconds, covering the entire cycle of an apnea event. This design not only captures short-term features but also identifies patterns of signal change over longer timeframes.

[0040] Small convolution kernels have higher resolving power on the time axis, allowing them to accurately capture subtle changes in the waveform of high-frequency signals. In contrast, large convolution kernels may blur details within a short period of time, thereby losing important features. For example, in EEG signals, spikes are short-lived, and a large convolution kernel may smooth them out, reducing the accuracy of feature extraction. Therefore, small convolution kernels can effectively identify these transient features.

[0041] S23: The first recognition physiological data and the second recognition physiological data are combined to obtain recognition physiological data.

[0042] Specifically, step S3 includes: S31: Align the first recognized physiological data and the second recognized physiological data to obtain unified input data.

[0043] Multimodal data often has significant differences in sampling frequency and time series length. For example, EEG signals are high-frequency, while blood oxygen saturation is low-frequency. This discrepancy can cause time series misalignment between modalities, thus impacting subsequent fusion analysis. Therefore, we employ a time series alignment preprocessing method to unify the features of each modality into a fixed 100 time steps through adaptive average pooling. This eliminates sampling gaps between high and low frequencies, thereby preprocessing the features of different modalities.

[0044] Adaptive mean pooling can dynamically adjust the size of the pooling window based on the temporal length of the input features, thereby mapping features of different modalities over the same number of time steps. For example, for high-frequency data such as EEG signals, embodiments of the present invention use a larger pooling window to reduce the number of time steps and lower computational complexity; while for low-frequency data such as blood oxygen saturation signals, a smaller pooling window is used to retain more temporal detail information. This adaptive pooling strategy enables temporal alignment of multimodal data without losing important information.

[0045] After time series alignment, the features of each modality are concatenated along the channel dimension to form a unified input. This unified input not only preserves the unique characteristics of each modality but also provides a unified input for subsequent time series modeling.

[0046] S32: Input the unified input data into a bidirectional two-layer LSTM to perform time series modeling to obtain an initial warning model.

[0047] To capture temporal dependencies and inter-modal correlations in multimodal data, this embodiment of the present invention employs a bidirectional LSTM for time series modeling. Using two independent LSTM probes, one scanning from the start point to the end point and the other from the end point to the start point, the bidirectional LSTM can simultaneously capture the predictive relationship between the current state and the future, as well as the impact of the historical state on the current state.

[0048] In the bidirectional scanning mechanism, the forward LSTM scans from the start of the time series to the end, capturing the relationship between the current state and the future. For example, a decreasing trend in blood oxygen saturation may indicate an impending apnea event. The reverse LSTM scans from the end of the time series to the start, capturing the impact of historical states on the current state. For example, a gradually increasing snoring volume may indicate the onset of apnea.

[0049] This embodiment of the present invention uses a two-layer stacked LSTM structure to learn the local temporal interactions between modalities and the global state evolution across modalities. The first LSTM layer is used for local temporal interactive learning of intermodal interactions; the second LSTM layer is used to study the cross-modal state of global state evolution.

[0050] Through bidirectional LSTM time series modeling, high-quality feature representation is provided for subsequent event location and classification, fully exploring the temporal dependencies and inter-modal correlations in multimodal data.

[0051] Specifically, the calculation formula is: in, For the moment The forward LSTM output of is the forward LSTM activation function, is the input layer forward propagation weight matrix, For the moment The unified input, is the weight matrix of the forward propagation itself, For the moment The forward LSTM output of time The backward LSTM output of is the backward LSTM activation function, is the input layer backpropagation weight matrix, is the weight matrix of the back propagation itself, For the moment The backward LSTM output of For the moment The predicted value of the final output gate, is the activation function of the forward propagation and backward propagation splicing, is the weight matrix for forward propagation to the output layer, is the weight matrix for back propagation to the output layer; Specifically, step S4 includes: S41: Design a localization head, which is used to enable the initial warning model to focus more on capturing changes in event boundaries, thereby improving localization accuracy by biasing gradient weights to channels related to waveform mutations; The present invention extracts signal features through a multi-scale convolutional network (convolutional layer design) and aligns different features by time step. Each node corresponds to a high-dimensional feature vector to design a localization head. The embodiment of the present invention designs a classification convolution layer. Based on the feature extraction, a 1×1 convolution layer (equivalent to a fully connected layer) is used to generate a category probability distribution prediction for each default event, that is, the output dimension is Dimension, among which is the total number of real event categories, plus one default event category.

[0052] Output probability distribution normalization, for each default event Apply point-wise Softmax activation to the dimensional output to generate normalized probabilities: in is the normalized probability of the i-th default event, For the i-th default event The raw score of the occurrence of each category, in, The corresponding 0th category is the default event.

[0053] S42: Design a classification head that enables the initial warning model to more accurately identify event types, thereby improving classification performance, by shifting gradient weights to channels related to the blood oxygen decline trend.

[0054] The classification head of this embodiment is designed as follows: first, a definition is given: a positive sample is a sample whose overlap ratio in the positioning head is higher than the set IoT threshold, and a negative sample is a sample whose overlap ratio is lower than the set IoT threshold.

[0055] Since a large amount of time during sleep is not in the process of developing obstructive sleep apnea, a large number of negative samples will appear in the default samples. Direct training will cause the model to fall into default events and thus reduce the prediction accuracy. The embodiment of the present invention adopts an innovative method to improve the prediction accuracy of the classification head - dynamic negative sample mining.

[0056] The basic steps of the dynamic negative sample mining algorithm are: basic screening: retain all positive samples (usually 1-5%); probability sorting: sort negative samples in ascending order of background class probability (that is, negative samples that are most difficult for the model to distinguish); dynamic sampling: select several negative samples with the highest probability so that the positive-to-negative sample ratio remains at 1:3; loss weighting: calculate the classification loss only for the screened samples.

[0057] This negative sample mining method can reduce false positives by forcing the model to focus on easily confused negative samples, forcing the model to learn more refined feature differentiation capabilities. In short, the embodiments of the present invention dynamically select the negative samples that are most valuable for model training, balancing the ratio of positive and negative samples and preventing the model from biased prediction of background classes.

[0058] S43: Designing an optimization function based on the positioning head and the classification head : in, is the first hyperparameter, is the positioning head loss function, is the second hyperparameter, is the classification head loss function, when the optimization function The initial warning model corresponding to the minimum is the final warning model.

[0059] Figure 3 Graph of the loss function for training with the localization head and the classification head. Figure 4 This is a model index diagram of the final model corresponding to the method provided by the present invention. Figure 3 As shown in (a), after adding the positioning head, the loss function converges more quickly; according to Figure 3 As shown in (b), after adding the classification head, the loss function converges more quickly. Figure 4 As shown in Figure 2, as the number of training rounds increases, the precision, recall rate, and F1 score of the model continue to rise, and the accuracy of the model is getting higher and higher. The initial warning model corresponding to the minimum is the final warning model.

[0060] Figure 2 An example of a physical structure diagram of an electronic device is shown below. Figure 2 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. The processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may invoke logic instructions in the memory 830 to execute an apnea warning system based on blood oxygen and snoring. This system measures a patient's physiological data, inputs the data into a final warning model, and generates a warning result.

[0061] Furthermore, the logic instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0062] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute an apnea warning model based on blood oxygen and snoring provided by the above methods. The system is used to measure the patient's physiological data, input the patient's physiological data into a final warning model, and obtain a warning result.

[0063] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the above-mentioned apnea warning model based on blood oxygen and snoring. The system is used to measure the patient's physiological data, input the patient's physiological data into the final warning model, and obtain a warning result.

[0064] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0065] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

[0067] It should be noted that the embodiments of the present disclosure can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system such as a microprocessor or dedicated hardware. Those skilled in the art will understand that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such code is provided on a programmable memory or a data carrier such as an optical or electronic signal carrier.

[0068] In addition, although the operations of the method of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that these operations must be performed in this particular order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flow chart can change the order of execution. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps. It should also be noted that the features and functions of two or more devices according to the present disclosure can be embodied in one device. Conversely, the features and functions of a device described above can be further divided into being embodied by multiple devices.

[0069] Although the present disclosure has been described with reference to several specific embodiments, it should be understood that the present disclosure is not limited to the specific embodiments disclosed. The present disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A sleep apnea warning model based on blood oxygen and snoring, characterized by: The model training method is: S1: Acquire physiological data, and preprocess the physiological data to obtain discrete physiological data; S2: extracting heterogeneous features from the discrete physiological data to obtain recognized physiological data; S3: performing spatiotemporal fusion on the identified physiological data to obtain an initial warning model; S4: Jointly optimize the initial warning model to obtain a final warning model.

2. The apnea warning model based on blood oxygen and snoring according to claim 1, characterized in that: The physiological data includes electroencephalogram, oculogram, mandibular electromyogram, nasal and oral airflow, blood oxygen saturation, respiratory rate, electrocardiogram and snoring data.

3. The apnea warning model based on blood oxygen and snoring according to claim 1, characterized in that: Step S1 includes: S11: Perform standard deviation screening on the physiological data to determine the values exceeding the mean. The data points in the range are marked as outliers, and the outliers are eliminated to obtain the first physiological data; S12: performing singular value decomposition on the first physiological data to obtain second physiological data; S13: Perform dynamic sample adjustment on the second physiological data to obtain discrete physiological data.

4. The apnea warning model based on blood oxygen and snoring according to claim 3, characterized in that: Step S13 includes: S131: Design default event time; S132: Design sample length; S133: discretizing the second physiological data according to the default event time, and splicing the discretized second physiological data according to the sample length to obtain discrete physiological data.

5. The apnea warning model based on blood oxygen and snoring according to claim 2, characterized in that: Step S2 includes: S21: extracting low-frequency modal data from the discrete physiological data using a shallow network with a large convolution kernel to obtain first recognized physiological data, wherein the low-frequency modal data includes nasal and oral airflow, blood oxygen saturation, respiratory rate, and electrocardiogram; S22: extracting high-frequency modal data from the discrete physiological data using a deep network with a small convolution kernel to obtain second recognition physiological data, wherein the high-frequency modal data includes electroencephalogram, eye movement graph, mandibular electromyogram, and snoring data; S23: The first recognition physiological data and the second recognition physiological data are combined to obtain recognition physiological data.

6. The apnea warning model based on blood oxygen and snoring according to claim 5, characterized in that: Step S3 includes: S31: Aligning the first recognized physiological data with the second recognized physiological data to obtain unified input data; S32: Input the unified input data into a bidirectional two-layer LSTM to perform time series modeling to obtain an initial warning model.

7. The apnea warning model based on blood oxygen and snoring according to claim 6, characterized in that: Step S4 includes: S41: Design a localization head, which is used to make the initial warning model more focused on capturing changes in event boundaries, thereby improving localization accuracy by biasing gradient weights to channels related to waveform mutations; S42: Design a classification head that enables the initial warning model to more accurately identify event types, thereby improving classification performance by shifting gradient weights to channels related to the blood oxygen decline trend; S43: Designing an optimization function based on the positioning head and the classification head : in, is the first hyperparameter, is the positioning head loss function, is the second hyperparameter, is the classification head loss function, when the optimization function The initial warning model corresponding to the minimum is the final warning model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the apnea warning system based on blood oxygen and snoring as claimed in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the apnea warning system based on blood oxygen and snoring as claimed in any one of claims 1 to 7 is run.

Images