A sleep staging method based on cardiac signal

Abstract

The present invention provides a sleep staging method, system and device based on cardiac impact signal, the system includes: a data acquisition module, used to acquire physiological data and PSG data; a feature extraction module, used to perform window processing on the physiological data to obtain window physiological data; perform feature extraction on the window physiological data to obtain physiological feature data; a classification model module, which establishes and stores the classification model; and a classification module, which classifies the physiological data to be classified based on the classification model and corrects unreasonable classification results. This solution uses a non-contact monitoring device to monitor the physical sign data of the test subject during sleep, which will not cause any burden to the test subject, and this algorithm can be deployed on an embedded device to realize offline sleep staging, does not rely on server operations, and has good scalability.

Description

Technical Field

[0001] The present invention relates to the field of biological sign data processing and computers, and in particular to the field of sleep staging and sleep quality analysis, and more particularly to a method, system and device for performing sleep staging by analyzing cardiac ballistometry sign data during sleep. Background Art

[0002] The pressure of life in modern society is getting higher and higher, and people's sleep quality at night will be affected to a certain extent. Sleep quality is closely related to human health and quality of life, and more and more people are beginning to pay attention to the quality of sleep at night. At present, the main method of sleep staging in medicine at home and abroad is polysomnography (PSG). PSG contains EEG, EOG, jaw electromyography, ECG, lower limb electromyography, respiratory airflow parameters, respiratory movement parameters, blood oxygen saturation, snoring sensor, body position sensor, etc. Through a variety of sensors, the physiological data of the monitored person during sleep is fully monitored, so as to diagnose sleep-related diseases, such as sleep breathing disorders, suspected narcolepsy, insomnia, etc. According to the American Academy of Sleep Medicine, sleep at night can be divided into 5 periods: wakefulness (W); rapid eye movement (REM); 3 non-rapid eye movement (NREM). The NREM stage is divided into non-rapid eye movement stage I (N1), non-rapid eye movement stage II (N2) and non-rapid eye movement stage III (N3). Stages N1 and N2 are defined as light sleep (Light Sleep, LS), and stage N3 is defined as deep sleep (also known as slow wave sleep (slowwave sleep, SWS)). A healthy individual usually goes through 3 to 5 sleep cycles during a full night of sleep. Each cycle lasts about 90 minutes, mainly including light sleep, deep sleep and rapid eye movement sleep. The wakefulness period refers to the period of falling asleep or waking up. A healthy individual usually goes through 3 to 5 sleep cycles during a full night of sleep. Each cycle lasts about 90 minutes, mainly including light sleep, deep sleep and rapid eye movement sleep. The wakefulness period refers to the period of falling asleep or waking up.

[0003] However, the PSG equipment used in medicine is bulky and expensive, and requires going to a hospital or a professional sleep monitoring center and requiring the person being monitored to wear many sensors. Changes in the sleeping environment and the wearing of sensors will have a certain impact on sleep quality. Among the current methods based on vital sign data, one method is to use pressure point sensors to collect heartbeat, breathing and body movement information during sleep to perform sleep staging. For example: Patent CN106419869A uses piezoelectric sensors to collect heartbeat, breathing and body movement data during sleep, and then uses the hidden Markov model of heart rate and breathing and body movement parameters to perform sleep staging. This method can distinguish between wakefulness and sleep states through body movement data, and further reduce the impact of body movement on heart rate and breathing signals. However, this method lacks heart parameters such as heart rate variability, and cannot accurately and objectively reflect the state of sleep. Another method is to extract heart rate, breathing and pulse signals from vital sign data and characterize them to generate feature vectors and then import them into deep learning models. For example: Patent CN110742585A extracts heart rate, breathing and pulse signals through the BCG waveform generated by the sleep monitoring mattress. The frequency domain features of heart rate variability and cardiopulmonary coupling power spectrum are calculated based on the acquired heart rate, pulse and respiratory signals, and then the features are extracted to obtain feature vectors and imported into the deep learning model to calculate the staging results. However, this method requires a deep learning model as a basis, requires a lot of computing power, and is not suitable for edge deployment. The present invention provides a sleep staging method based on a micro sleep sign monitoring device. Summary of the invention

[0004] Based on the problems existing in the prior art, the present invention proposes a sleep staging method based on cardiac impact sign data and machine learning, which mainly consists of: data acquisition, feature extraction, semi-supervised classification, and classification staging. In the data acquisition stage, a cardiac impact sign monitoring sensor is used to collect changes in the heart rate, breathing, and body movement of the test subject, and invalid data is eliminated; in the feature extraction stage, all collected data are windowed and the required feature vectors are extracted; in the semi-supervised classification stage, the windowed feature vectors obtained in the previous stage are manually marked, and the data that can be clearly marked is marked. After the manual marking can clarify the staging data, it is imported into the label propagation algorithm for semi-supervised learning marking; in the classification staging stage, the samples of the previous stage are used to build an effective sleep staging model.

[0005] Specifically, the present invention provides the following technical solutions:

[0006] In one aspect, the present invention provides a sleep staging method based on a heartbeat signal, the method comprising:

[0007] Step 1: Obtain physiological data;

[0008] Step 2: performing window processing on the physiological data to obtain window physiological data; performing feature extraction on the window physiological data to obtain physiological feature data;

[0009] Step 3, performing unsupervised clustering on the physiological characteristic data and marking the category number; marking the window physiological data into sleep stages; statistically mapping the clustered category number with the sleep stage; if the statistical mapping deviation does not meet the preset requirements, adjusting the clustering parameters and performing unsupervised clustering again until the mapping deviation meets the preset requirements, and determining the final mapping relationship;

[0010] Step 4: classify and mark the sample data based on the clustering parameters corresponding to the final mapping relationship, and perform classification training based on the classified and marked sample data to establish a classification model;

[0011] Step 5: Classify the physiological data to be classified based on the classification model, and correct unreasonable classification results.

[0012] Preferably, the physiological characteristic data includes one or any combination of the following characteristics: heart rate mean, heart rate standard deviation, respiration mean, respiration standard deviation, body momentum, body momentum mean, body momentum standard deviation, expiratory heart rate ratio, respiratory frequency band peak, heart rate band peak, respiratory frequency band centroid, heart rate band peak, etc.

[0013] Preferably, in step 3, the clustering parameters include a cluster number K value and an initialization random number.

[0014] Preferably, in step 3, the preset requirement is: in the statistical mapping relationship, the proportion of abnormal points is less than 20%.

[0015] Preferably, in step 3, the specific method of marking the window physiological data into sleep stages is:

[0016] Marking sleep stages for some window physiological data with significant features. These physiological data features have extremely high confidence not only from the reference PSG data, but also from the annotation results of sleep experts and the principle of state changes in the sleep cycle.

[0017] Based on the window physiological data after sleep stage marking, classification reasoning is performed on the window physiological data that has not been marked with sleep stages, and sleep stage marking is performed.

[0018] Preferably, classification inference of window physiological data that are not marked for sleep stages can be achieved through label propagation: some of the data sets are marked and some are unmarked. If a data point is unmarked and M of its N nearest neighbors are marked, then these M neighbors can be voted through a simple majority vote, plus voting weights (for example, neighbors with closer distances have greater weights, etc.), and the classification result generated after voting can be used as the label of the unmarked data point.

[0019] Preferably, in step 5, the method of correcting the unreasonable classification result is:

[0020] Reasonable reasoning is performed based on the rules of sleep stages, and unreasonable results are reclassified into sleep stages to achieve correction.

[0021] Preferably, the rules referred to in the rational reasoning include but are not limited to: the user's sleep stages will repeat, each cycle lasts about 90 minutes, and the standard stage switching is as follows: wakefulness->REM->N1->N2->N3->N2->N1->REM->wakefulness. This stage switching rule is likely to be followed by the general public, and the sleep classification result is re-corrected according to this rule.

[0022] In addition, the present invention also provides a sleep staging system based on the heartbeat signal, the system comprising:

[0023] A data acquisition module, used for acquiring heart shock signal data and multiple PSG data;

[0024] The feature extraction module performs bandpass filtering on the heartbeat signal data to extract the respiratory signal and the heart rate signal; performs sliding window processing on the respiratory signal and the heart rate signal to obtain window physiological data; and performs feature extraction on the window physiological data to obtain physiological feature data; the physiological feature data includes time domain features and frequency domain features;

[0025] Classification model module, builds and stores classification models;

[0026] The classification module classifies the physiological data to be classified based on the classification model and corrects unreasonable classification results.

[0027] Preferably, the system further comprises:

[0028] A mapping module performs unsupervised clustering on the physiological characteristic data and marks the cluster numbers; marks the window physiological data into sleep stages with reference to the staging results of the multi-PSG data; statistically maps the cluster numbers with the sleep staging marks; if the statistical mapping deviation does not meet the preset requirements, adjusts the clustering parameters and performs unsupervised clustering again until the mapping deviation meets the preset requirements, and determines the final mapping relationship.

[0029] Preferably, when the classification model is established, the classification model module is also used to: cluster and mark subsequent sample data based on the unsupervised clustering model corresponding to the final mapping relationship, and form a new feature data set based on the marked sample data; perform classification training based on the new feature data set to establish a classification model.

[0030] Preferably, the physiological characteristic data includes one or any combination of the following characteristics: heart rate mean, heart rate standard deviation, respiration mean, respiration standard deviation, body momentum, body momentum mean, body momentum standard deviation, expiratory heart rate ratio, respiratory frequency band peak, heart rate band peak, respiratory frequency band centroid, heart rate band peak, etc.

[0031] Preferably, the clustering parameters include a clustering number K value and an initialization random number.

[0032] Preferably, the preset requirement is that in the statistical mapping relationship, the proportion of abnormal points is less than 20%.

[0033] Preferably, the specific method of marking the sleep stages of the window physiological data is:

[0034] Some window physiological data with significant features are marked as sleep stages. These physiological data features have extremely high confidence not only from the reference PSG data, but also from the annotation results of sleep experts and the principle of state changes in the sleep cycle. Based on the window physiological data after sleep stage marking, the window physiological data that has not been marked with sleep stages are classified and inferred, and sleep stages are marked.

[0035] Preferably, the method of correcting the unreasonable classification results is:

[0036] Reasonable reasoning is performed based on the rules of sleep stages, and unreasonable results are reclassified into sleep stages to achieve correction.

[0037] On the other hand, the present invention also provides a sleep staging device based on cardiac impact signals, which includes a processor, a memory and a bus, wherein the memory stores instructions readable by the processor; the processor is used to call the instructions in the memory to execute the sleep staging method based on cardiac impact signals as described above; the bus connects the functional components to transmit information.

[0038] Compared with the prior art, the technical solution of the present invention uses a non-contact monitoring device to monitor the vital signs data of the test subject during sleep, which will not cause any burden to the test subject, and this algorithm can be deployed on embedded devices to realize offline sleep staging, and does not rely on server operations. It can realize effective sleep staging only by monitoring the mattress, and can be well applied in the fields of daily household items. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 paying creative work.

[0040] Figure 1 Generate a flow chart for the training set of an embodiment of the present invention;

[0041] Figure 2 A flow chart of a sleep staging method according to an embodiment of the present invention;

[0042] Figure 3 is a system structure diagram of an embodiment of the present invention;

[0043] Figure 4 Schematic diagram of a method for establishing a mapping between cluster numbers and sleep stages according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. It should be clear that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Those skilled in the art should know that the following specific embodiments or specific implementations are a series of optimized settings listed by the present invention to further explain the specific content of the invention, and these settings can be combined or used in association with each other, unless the present invention clearly states that some or a specific embodiment or implementation cannot be associated or used together with other embodiments or implementations. At the same time, the following specific embodiments or implementations are only used as the most optimized settings, and are not to be understood as limiting the scope of protection of the present invention.

[0046] In a specific embodiment, the technical solution of the present invention, its core method mainly includes: data acquisition, feature extraction, semi-supervised classification, classification and staging. In the data acquisition stage, use equipment such as a cardiac shock sign monitoring sensor to collect changes in the heart rate, breathing, body movement, etc. of the test subject, and eliminate invalid data; in the feature extraction stage, all collected data are windowed and the required feature vectors are extracted; in the semi-supervised classification stage, the windowed feature vectors obtained in the previous stage are marked, and the data that can be clearly divided into stages are marked. After marking the data that can be clearly divided into stages, it is imported into the label propagation algorithm for semi-supervised learning marking; in the classification and staging stage, the samples of the previous stage are used to build an effective sleep staging model.

[0047] In a more preferred embodiment, each specific step of the method can be carried out as follows:

[0048] 1. Data acquisition

[0049] First of all, the purpose of the present invention is to enable contactless sleep monitoring and staging of users' sleep at night, and to detect relevant physiological data of the tested person through the intelligent heartbeat detection device. The detection device will start to collect parameters such as the heart rate, respiratory rate, body movement size, etc. of the monitored person after lying quietly in bed. While using the intelligent heartbeat sleep monitoring bed belt, a polysomnography device (PSG) is also used to synchronously collect data on parameters such as ECG, respiratory rate, blood oxygen saturation, and EEG. In a more preferred embodiment, this method only requires the sleep staging results of the PSG analysis method as a control.

[0050] 2. Feature extraction

[0051] Feature extraction is mainly to obtain hidden information from the collected data, so as to effectively explore the relationship between the vital sign data and the sleep stage during sleep. In this embodiment, the collected heart impact signal data is preferably band-pass filtered to extract the breathing signal and the heart rate signal. The frequency band range for extracting breathing can be set to the range of 0.1 to 2 Hz, and the frequency band range for extracting heart rate can be set to the range of 1 to 30 Hz.

[0052] In this embodiment, since the sleep state cannot be determined based on a single data frame, each sleep stage is divided according to a time interval, so the data is windowed to extract feature data within a period of time. In a specific implementation, the size of each window can be set to, for example, 30 seconds, and the sliding step of each window can be set to, for example, 15 seconds. The time range of the window can be adjusted, but it should be at least greater than 4 seconds, and the upper limit of the window is preferably 30 seconds.

[0053] After the window is divided, the mean, standard deviation, and frequency of occurrence of each parameter in the window are calculated, and the correlation parameters between some parameters are calculated, such as the mean standard deviation of heart rate, the mean standard deviation of respiratory rate, the mean standard deviation of body momentum, and the expiratory heart rate ratio are used as features. These acquired physiological feature data include time domain features and frequency domain features. In a more preferred embodiment, the acquired features are listed as follows:

[0054] Heart rate average: All heartbeat intervals that occurred in the past 30 seconds, after removing the maximum and minimum values, calculate the average value and then convert it into heart beats per minute.

[0055] Heart rate standard deviation: All heartbeat intervals that occurred in the past 30 seconds are converted to heart beats per minute, and the standard deviation is calculated after removing the maximum and minimum values.

[0056] Respiration mean: All breathing intervals that occurred in the past 30 seconds, after removing the maximum and minimum values, calculate the average value and then convert it into the number of breaths per minute.

[0057] Respiration standard deviation: All breathing intervals that occurred in the past 30 seconds are converted to the number of breaths per minute, and the standard deviation is calculated after removing the maximum and minimum values.

[0058] Body momentum: The standard deviation of the vibration amplitude of the heartbeat waveform every 4 seconds is used as the calculation method of body momentum. In a preferred embodiment, the calculation window of body momentum can be configured to range from 1 second to 10 seconds. Different configuration parameters will affect the sensitivity of body momentum, thereby affecting the parameters of subsequent clustering and classification methods. The sliding window for obtaining body momentum can be different from the calculation window. In specific implementation, half of the calculation window can be used as the sliding window for body motion calculation.

[0059] Mean of body momentum: Directly calculate the standard deviation of all body momentum that occurred in the past 30 seconds.

[0060] Body momentum standard deviation: Directly calculate the standard deviation of all body momentum that occurred in the past 30 seconds.

[0061] PHR: The ratio of the number of heartbeats in the past 30 seconds divided by the number of respirations in the past 30 seconds.

[0062] In addition, characteristic data such as respiratory frequency band peak value, heart rate band peak value, respiratory frequency band centroid, heart rate band peak value, etc. can also be added.

[0063] 3. Semi-supervised classification

[0064] The mean and standard deviation of the above-mentioned respiration, heart rate, body momentum, and the features of the heart rate and breathing rate are used as training vectors, and clustering is performed in an unsupervised manner. The clustering method can be implemented using common clustering algorithms such as kmeans, kmeans+, dbscan, etc. The number of clusters K in the algorithm implementation ranges from 5 to 20. Clustering can be performed using different K value parameters and initialization random numbers, and then the optimal K value and initial parameters of the clustering algorithm can be determined under the PSG control data. Figure 4 As shown in the F3 section.

[0065] After clustering, the training set is divided into multiple clusters and marked with corresponding cluster numbers. The windowed data in the previous stage is visualized and manually marked with data with obvious features according to the PSG staging results. Then, the data is classified and inferred by label propagation for unlabeled data, so as to preliminarily judge the current staging results through the algorithm. It should be noted that only the data with significant features are marked, and the remaining unlabeled data are marked by the label propagation algorithm. The label propagation algorithm guarantees the optimal marking result based on the mathematical nearest neighbor majority principle. After marking, the cluster number and the sleep staging stage form a mapping effect.

[0066] In a preferred embodiment, the label propagation mode can be implemented specifically by the following method:

[0067] Some of the data sets are labeled and some are unlabeled. If a data point is unlabeled, but M of its N nearest neighbors are labeled, then these M neighbors can vote by simple majority voting, plus voting weights (for example, neighbors with closer distances have greater weights, etc.), and the classification result produced after voting can be used as the label of the unlabeled data point. For example, for a data point that is not labeled with a stage, 13 of its 20 nearest neighbors are labeled with a stage, of which 9 are labeled as stage N1, 3 are labeled as stage N2, and 1 is labeled as stage N3. Then, according to the majority voting principle, this data point should be labeled as N1. In a more preferred solution, the distance between neighbors will be taken into account, and the larger the distance, the smaller the weight. For example, the inverse of the distance can be used as the weight.

[0068] When performing mapping, in a preferred embodiment, the mapping method is exemplified as follows: C1=>REM, C2=>REM, C2=>N2, etc., where C1 and C2 are example category numbers, and REM and N2 are different sleep stages. After counting the mapping relationships of enough samples, a statistical mapping relationship between the cluster number and the sleep stage is established. If the deviation of the mapping relationship is too large, it is necessary to return to the clustering stage to select different K values ​​and initialize random numbers for re-clustering. These two steps are iterated multiple times until the mapping relationship between the cluster number and the sleep stage is statistically robust enough. In a preferred embodiment, the criterion for judging the robustness is that the proportion of abnormal points is less than 20%. Here, the judgment of abnormal points in the mapping relationship is illustrated by example. For example, in Figure 4 As shown in S302, 17 C2 cluster points are mapped to N1 category, and 163 C2 cluster points are mapped to N2 category. According to the majority principle, C2 cluster should be mapped to N2 category. However, because 17 C2 are mapped to N1 category, there is an error in this mapping, and the calculation of this error is (17) / (17+163)=9.4%. The maximum error of the entire mapping is determined by the maximum cluster->classification mapping error.

[0069] When the mapping relationship of the above clustering results is stable enough, more samples are marked and classified with the model parameters of this clustering. When all samples are marked, all samples can be used for subsequent classification training.

[0070] Semi-supervised classification can display the classification results and the feature data collected by PSG synchronously through a visual interface. The semi-supervised classification results are marked as awake period, REM period, and N1-N3 by referring to the classification results of PSG. At the same time, each semi-supervised classification result and the possible reasons for this staging are carefully analyzed. And some obviously wrong or unreasonable staging results are corrected and re-staged by manual methods by referring to PSG data, such as rapid transitions in sleep stages, transitions across stages, and rapid fluctuations in staging stages, so as to generate a training set that can be used for learning and training. The training set generation process can be referred to Figure 1 shown.

[0071] 4. Classification and staging

[0072] In a preferred embodiment, in this step, based on the unsupervised clustering model corresponding to the final mapping relationship, the subsequent sample data are clustered and labeled, and a new feature data set is formed based on the labeled sample data. Specifically:

[0073] Step 4.1: The new feature data set described in step 4 includes the physiological feature data obtained after window processing and the clustering labels in this step. These two parts are also associated through the above-mentioned unsupervised clustering. Figure 4 As shown in F6, the new dataset no longer considers the labeling results of PSG, but directly classifies the features of the data points into cluster numbers.

[0074] Step 4.2: Perform classification training on the data set obtained in step 4.1 to establish a classification model. Common machine learning classification algorithms or deep learning classification algorithms can be used here, and the selection is based on the local computing power of the device.

[0075] A sleep staging classification model based on supervised learning is constructed through the classification samples in the previous stage. Decision trees, random forests, Xgboost and other machine learning classification algorithms can be used for specific implementation. The classified data set marked in the previous step is trained through the above classification algorithm and the cluster number is output, which is then mapped to the five-classification model of the sleep staging results, namely the awake period, REM rapid eye movement period, N1, N2, and N3 period. After the model completes the sleep classification, it will use a reasonable reasoning method to re-divide some unreasonable results in the classification results into stages. Finally, the model is processed and deployed on home or medical devices such as smart mattresses to complete offline sleep staging that does not rely on the back-end server. The specific process is as follows: Figure 2 shown.

[0076] In a more preferred embodiment, in the rationality reasoning method, the reference reasoning judgment rules include but are not limited to: the user's sleep stages will repeat, each cycle lasts about 90 minutes, and the standard stage switching is as follows: wakefulness->REM->N1->N2->N3->N2->N1->REM->wakefulness. This stage switching rule is likely to be followed by the general public, and the sleep classification results are re-corrected according to this rule.

[0077] In another implementation mode, the present solution can also be implemented in a system manner, and the system may include:

[0078] A data acquisition module, used for acquiring heart shock signal data and multiple PSG data;

[0079] The feature extraction module performs bandpass filtering on the heartbeat signal data to extract the respiratory signal and the heart rate signal; performs sliding window processing on the respiratory signal and the heart rate signal to obtain window physiological data; and performs feature extraction on the window physiological data to obtain physiological feature data; the physiological feature data includes time domain features and frequency domain features;

[0080] Classification model module, builds and stores classification models;

[0081] The classification module classifies the physiological data to be classified based on the classification model and corrects unreasonable classification results.

[0082] Preferably, at the initial stage of system establishment, it is necessary to establish a classification model. At the same time, the establishment of the classification model requires the formation of training sample data with reasonable sleep staging labels. At this time, it is necessary to first label the physiological characteristic data with reasonable sleep staging labels. To achieve this process, the system can also be provided with a mapping module. The mapping module performs unsupervised clustering on the physiological characteristic data and marks the cluster number; marks the window physiological data with sleep staging with reference to the staging results of the multi-PSG data; performs statistical mapping between the cluster number and the sleep staging label; if the statistical mapping deviation does not meet the preset requirements, the clustering parameters are adjusted to perform unsupervised clustering again until the mapping deviation meets the preset requirements, and the final mapping relationship is determined.

[0083] At this time, when the classification model is established, the classification model module is also used to: cluster and mark subsequent sample data based on the unsupervised clustering model corresponding to the final mapping relationship, and form a new feature data set based on the marked sample data; perform classification training based on the new feature data set to establish a classification model.

[0084] In specific implementation, the classification model can use machine learning classification algorithms such as decision tree, random forest, Xgboost, etc. The labeled classification data set is trained through the above classification algorithm and a five-classification model of the sleep stage results is output, namely, wakefulness, REM, N1, N2, and N3.

[0085] Preferably, the physiological characteristic data includes one or any combination of the following characteristics: heart rate mean, heart rate standard deviation, breathing mean, breathing standard deviation, body momentum, body momentum mean, body momentum standard deviation, expiratory heart rate ratio, etc., which will not be repeated here.

[0086] Preferably, the clustering parameters include a clustering number K value and an initialization random number.

[0087] Preferably, the preset requirement is that in the statistical mapping relationship, the proportion of abnormal points is less than 20%.

[0088] More preferably, the specific method of marking the sleep stages of the window physiological data is:

[0089] Marking sleep stages for some window physiological data with significant features. These physiological data features have extremely high confidence not only from the reference PSG data, but also from the annotation results of sleep experts and the principle of state changes in the sleep cycle.

[0090] Based on the window physiological data after sleep stage marking, classification reasoning is performed on the window physiological data that has not been marked with sleep stages, and sleep stage marking is performed.

[0091] Preferably, classification inference of window physiological data that are not marked for sleep stages can be achieved through label propagation: some of the data sets are marked and some are unmarked. If a data point is unmarked and M of its N nearest neighbors are marked, then these M neighbors can be voted through a simple majority vote, plus voting weights (for example, neighbors with closer distances have greater weights, etc.), and the classification result generated after voting can be used as the label of the unmarked data point.

[0092] Preferably, the method of correcting the unreasonable classification results is:

[0093] Reasonable reasoning is performed based on the rules of sleep stages, and unreasonable results are reclassified into sleep stages to achieve correction.

[0094] In another embodiment, the present solution can be implemented by means of a device, which may include a corresponding module for performing each or several steps in the above-mentioned embodiments. Therefore, each step or several steps of the above-mentioned embodiments can be performed by a corresponding module, and the electronic device may include one or more of these modules. The module may be one or more hardware modules specially configured to perform the corresponding steps, or implemented by a processor configured to perform the corresponding steps, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination.

[0095] The device can be implemented using a bus architecture. The bus architecture can include any number of interconnecting buses and bridges, depending on the specific application and overall design constraints of the hardware. The bus connects various circuits including one or more processors, memories, and / or hardware modules together. The bus can also connect various other circuits such as peripherals, voltage regulators, power management circuits, external antennas, etc.

[0096] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one connecting line is used in this figure, but it does not mean that there is only one bus or one type of bus.

[0097] Any process or method description in the flowchart or otherwise described herein can be understood as a module, fragment or portion of a code representing one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiment of the present solution includes other implementations, in which the functions may not be performed in the order shown or discussed, including performing the functions in a substantially simultaneous manner or in a reverse order according to the functions involved, which should be understood by a person skilled in the art of the present solution. The processor performs the various methods and processes described above. For example, the method implementation in the present solution can be implemented as a software program, which is tangibly contained in a machine-readable medium, such as a memory. In some embodiments, part or all of the software program can be loaded and / or installed via a memory and / or a communication interface. When the software program is loaded into the memory and executed by the processor, one or more steps in the method described above can be performed. Alternatively, in other embodiments, the processor can be configured to perform one of the above methods in any other appropriate manner (e.g., by means of firmware).

[0098] The logic and / or steps represented in the flowchart or otherwise described herein may be embodied in any readable storage medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from and execute instructions on an instruction execution system, apparatus, or device), or in combination with such instruction execution systems, apparatuses, or devices.

[0099] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0100] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A sleep staging method based on cardiac signal, characterized in that: The method comprises: Step 1, obtaining heartbeat signal data and collecting multiple PSG data; Step 2: bandpass filter the heartbeat signal data to extract the respiratory signal and the heart rate signal; perform sliding window processing on the respiratory signal and the heart rate signal to obtain window physiological data; and perform feature extraction on the window physiological data to obtain physiological characteristic data; the physiological characteristic data includes time domain features and frequency domain features; Step 3, performing unsupervised clustering on the physiological characteristic data and marking the cluster number; marking the window physiological data by sleep stage with reference to the staging results of the multi-PSG data; statistically mapping the cluster number and the sleep stage mark; if the statistical mapping deviation does not meet the preset requirements, adjusting the clustering parameters and performing unsupervised clustering again until the mapping deviation meets the preset requirements, and determining the final mapping relationship; Step 4: Based on the unsupervised clustering model corresponding to the final mapping relationship, cluster and label the subsequent sample data, and form a new feature data set based on the labeled sample data; perform classification training based on the new feature data set to establish a classification model; Step 5: classify the physiological data to be classified based on the classification model, and correct unreasonable classification results; In step 3, the clustering parameters include the clustering number K value and the initialization random number; In step 3, the preset requirement is: in the statistical mapping relationship, the proportion of abnormal points is less than 20%.

2. The method according to claim 1, characterized in that The physiological characteristic data includes one or any combination of the following characteristics: heart rate mean, heart rate standard deviation, respiration mean, respiration standard deviation, body momentum, body momentum mean, body momentum standard deviation, expiratory heart rate ratio, respiratory frequency band peak, heart rate frequency band peak, and respiratory frequency band centroid.

3. The method according to claim 1, characterized in that In step 1, the specific method of marking the window physiological data into sleep stages is: Some window physiological data with significant features are marked as sleep stages. Based on the window physiological data after sleep stage marking, the window physiological data that has not been marked as sleep stages are classified and inferred to mark the sleep stages.

4. The method according to claim 1, characterized in that: In step 5, the method for correcting unreasonable classification results is: Rational reasoning is performed based on the rules of sleep stages, and unreasonable results are reclassified into sleep stages to achieve correction.

5. A sleep staging system based on cardiac signal, characterized in that: The system comprises: A data acquisition module, used for acquiring heart shock signal data and multiple PSG data; The feature extraction module performs bandpass filtering on the heartbeat signal data to extract the respiratory signal and the heart rate signal; performs sliding window processing on the respiratory signal and the heart rate signal to obtain window physiological data; and performs feature extraction on the window physiological data to obtain physiological feature data; the physiological feature data includes time domain features and frequency domain features; Classification model module, builds and stores classification models; A classification module, which classifies the physiological data to be classified based on the classification model and corrects unreasonable classification results; The system further comprises: A mapping module performs unsupervised clustering on the physiological characteristic data and marks the cluster number; marks the window physiological data into sleep stages with reference to the staging results of the multi-PSG data; performs statistical mapping between the cluster number and the sleep stage mark; if the statistical mapping deviation does not meet the preset requirements, adjust the clustering parameters and perform unsupervised clustering again until the mapping deviation meets the preset requirements, and determine the final mapping relationship; When the classification model is established, the classification model module is also used to: cluster and mark subsequent sample data based on the unsupervised clustering model corresponding to the final mapping relationship, and form a new feature data set based on the marked sample data; perform classification training based on the new feature data set to establish a classification model.

6. A sleep staging device based on cardiac signal, characterized in that: The device includes a processor, a memory and a bus, wherein the memory stores instructions readable by the processor; the processor is used to call the instructions in the memory to execute the sleep staging method based on cardiac impulse signals as described in any one of claims 1 to 4; and the bus connects the functional components to transmit information.

Images