Sleep staging detection method and sleep staging detection device

Abstract

The invention provides a sleep staging detection method and a sleep staging detection device.The method comprises the steps that a polysomnogram of a target object within a preset time period is obtained; inputting the plurality of physiological signals into a first model to obtain frequency signals under different frequency characteristics; the multiple frequency signals are input into a second model to obtain two intermediate layer signals, the second model comprises a character string operation module and two first channels, the character string operation module is used for integrating the multiple frequency signals into a target frequency signal, and the two first channels are used for receiving the target frequency signal; the input of each first channel is a target frequency signal received from the character string operation module, and the output of each first channel is a corresponding first intermediate layer signal; and inputting the two first intermediate layer signals into a third model to obtain a sleep staging result of the target object. According to the sleep staging method and device, the physiological signals in the polysleep map are processed and analyzed, and the accuracy of the sleep staging result is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a sleep stage detection method and a sleep stage detection device. Background Art

[0002] One-third of a person's life is spent sleeping, and sleep quality plays a vital role in human health. With the development of medicine, people pay more and more attention to sleep, and also pay attention to sleep-related diseases. If you want to explore the basis of these sleep diseases, you can use sleep staging. Sleep staging refers to the various changes in the EEG during sleep, and these changes vary with the depth of sleep. According to the different characteristics of the EEG, sleep is divided into two states: non-rapid eye movement sleep (which can be divided into three sleep stages: N1, N2, and N3) and rapid eye movement sleep. The two are distinguished by whether there are paroxysmal rapid eye movements and different brain wave characteristics.

[0003] The current sleep staging methods have problems such as complex design and the need to extract signal features in advance. In addition, the final accuracy of sleep staging is often not high enough. These problems make the existing algorithms unsuitable for sleep EEG staging and cannot accurately reflect the brain's activity during sleep. Summary of the invention

[0004] In view of this, the purpose of the present application is to provide a sleep stage detection method and a sleep stage detection device to overcome at least one of the above-mentioned defects.

[0005] In a first aspect, an embodiment of the present application provides a sleep staging detection method, the method comprising: obtaining a polysomnogram of a target object within a preset time period, the polysomnogram comprising multiple physiological signals; inputting the multiple physiological signals into a first model to obtain frequency signals under different frequency characteristics, wherein the first model comprises multiple CNN sequences, the input of each CNN sequence is the multiple physiological signals, and the output of each CNN sequence is a frequency signal corresponding to a frequency characteristic; inputting the multiple frequency signals into a second model to obtain two intermediate layer signals, the second model comprises a string operation module and two first channels, the string operation module is used to integrate the multiple frequency signals into a target frequency signal, the input of each first channel is the target frequency signal received from the string operation module, and the output of each first channel is a corresponding first intermediate layer signal; inputting the two first intermediate layer signals into a third model to obtain the sleep staging result of the target object. In an optional embodiment of the present application, the method also includes: for the physiological signals under each channel of the polysomnogram, dividing the multiple physiological signals under the channel according to a preset time to obtain multiple sleep stages, each sleep stage including the multiple physiological signals under the multiple channels within the preset time; inputting each sleep stage into a sleep stage model in turn to obtain the sleep stage result of the target object, the sleep stage model including the first model, the second model and the third model.

[0006] In an optional embodiment of the present application, each CNN sequence is a 1-dimensional CNN sequence, and the parameter ratios between different CNN sequences are different, wherein the multiple physiological signals are input into the first model to obtain frequency signals under different frequency characteristics, including: for each CNN sequence, frequency feature extraction of the multiple physiological signals according to the parameter ratio corresponding to the CNN sequence to obtain the frequency signal under the frequency feature corresponding to the CNN sequence.

[0007] In an optional embodiment of the present application, the two first channels corresponding to the second model are both 1-dimensional CNN sequences, wherein the multiple frequency signals are input into the second model to obtain two first intermediate layer signals, including: inputting the multiple frequency signals into a string operation module to output a target frequency signal; and inputting the target frequency signals into the first channels respectively to obtain two first intermediate layer signals.

[0008] In an optional embodiment of the present application, the third model includes a staging result determination module and two second channels, the input of each second channel is the first intermediate layer signal output from the first channel, and the output of each second channel is the corresponding second intermediate layer signal, and the staging result determination module is used to map the second intermediate layer signal to a sleep staging result, wherein the inputting the two first intermediate layer signals into the third model to obtain the sleep staging result of the target object includes: inputting the two first intermediate layer signals into the two second channels to obtain the second intermediate layer signal corresponding to each second channel; inputting the two second intermediate layer signals into the staging result determination module to obtain the sleep staging result of the target object.

[0009] In an optional embodiment of the present application, the multiple physiological signals are determined in the following manner: a filtering range of each physiological signal is determined; for each physiological signal, the physiological signal in the polysomnogram is filtered according to the corresponding filtering range to obtain multiple physiological signals of the physiological signal within the corresponding filtering range.

[0010] In an optional embodiment of the present application, the sleep staging results include a first result, a second result, a third result, a fourth result and a fifth result, wherein the sleep staging results of the target object are determined in the following manner: each sleep stage is input into the sleep staging model in turn to determine the result corresponding to the sleep stage.

[0011] In a second aspect, an embodiment of the present application further provides a sleep staging detection device, the device comprising: a polysomnogram acquisition module, used to acquire a polysomnogram of a target object within a preset time period, the polysomnogram comprising a plurality of physiological signals; a frequency signal acquisition module, used to input the plurality of physiological signals into a first model to obtain frequency signals under different frequency characteristics, wherein the first model comprises a plurality of CNN sequences, the input of each CNN sequence is the plurality of physiological signals, and the output of each CNN sequence is a frequency signal under a corresponding frequency characteristic; a first intermediate layer signal acquisition module, used to input the plurality of frequency signals into a second model to obtain two first intermediate layer signals, the second model comprising a string operation module and two first channels, the string operation module being used to integrate the plurality of frequency signals into a target frequency signal, the input of each first channel being the target frequency signal received from the string operation module, and the output of each first channel being the corresponding first intermediate layer signal; a sleep staging result acquisition module, used to input the two first intermediate layer signals into a third model to obtain the sleep staging result of the target object. In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the method described above are performed.

[0012] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are executed.

[0013] The sleep staging detection method and sleep staging detection device provided by the embodiment of the present application obtain a polysomnogram of a target object within a preset time period, wherein the polysomnogram includes multiple physiological signals; the multiple physiological signals are input into a first model to obtain frequency signals under different frequency characteristics, wherein the first model includes multiple CNN sequences, each CNN sequence has multiple physiological signals as input, and each CNN sequence has a frequency signal under a corresponding frequency characteristic as output; the multiple frequency signals are input into a second model to obtain two intermediate layer signals, wherein the second model includes a string operation module and two first channels, the string operation module is used to integrate multiple frequency signals into a target frequency signal, the input of each first channel is a target frequency signal received from the string operation module, and the output of each first channel is a corresponding first intermediate layer signal; the two first intermediate layer signals are input into a third model to obtain the sleep staging result of the target object. Through the present application, multiple physiological signals in the polysomnogram are processed and analyzed, and the accuracy of the sleep staging result is improved.

[0014] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flowchart of the sleep stage detection method provided in an embodiment of the present application; Figure 2 A flowchart for determining multiple physiological signals provided in an embodiment of the present application; Figure 3 A schematic diagram of a sleep stage model provided in an embodiment of the present application; Figure 4 A schematic diagram of a first model provided in an embodiment of the present application; Figure 5 A schematic diagram of a second model provided in an embodiment of the present application; Figure 6 A flowchart of obtaining two first intermediate layer signals provided in an embodiment of the present application; Figure 7 A schematic diagram of a third model provided in an embodiment of the present application; Figure 8 A schematic diagram of the structure of a sleep stage detection device provided in an embodiment of the present application; Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0018] First, the application scenarios to which the present application is applicable are introduced. The present application can be applied in the field of data processing.

[0019] Research has found that the existing technology in sleep staging has problems such as complex design, limited generalization ability and poor real-time performance, resulting in the final accuracy of sleep staging being often not high enough. These problems make the existing algorithms unsuitable for sleep EEG staging and cannot accurately reflect the brain's activity during sleep.

[0020] Based on this, the embodiment of the present application provides a sleep staging detection method and a sleep staging detection device, which obtains a polysomnogram of a target object within a preset time period; inputs multiple physiological signals into a first model to obtain frequency signals with different frequency characteristics; inputs multiple frequency signals into a second model to obtain two intermediate layer signals; and inputs two first intermediate layer signals into a third model to obtain the sleep staging results of the target object. Through the present application, multiple physiological signals in a polysomnogram are processed and analyzed, thereby improving the accuracy of the sleep staging results.

[0021] See also Figure 1 , Figure 1 This is a flow chart of the sleep stage detection method provided in the embodiment of the present application. Figure 1 As shown in , the sleep stage detection method provided in the embodiment of the present application includes: S101, obtaining a polysomnogram of a target subject within a preset time period.

[0022] Among them, polysomnography (PSG) includes multiple physiological signals.

[0023] In addition to the electroencephalogram, current regular PSG monitoring should include physiological signals from channels such as the electrocardiogram, electromyogram, oculogram, chest and abdominal respiratory tension graph, nasal and oral ventilation, body position and movement, blood oxygen saturation, and cavernous muscle volume.

[0024] PSG examinations are performed in a specially set up sleep laboratory. An ideal sleep laboratory should include an instrument operation room and a sleep monitoring room for the subject. The layout of the sleep monitoring room should be simple and quiet, with a suitable temperature and good air circulation, and should be as close to the home environment as possible, so that the subject can relax and feel comfortable.

[0025] According to monitoring needs, electrodes and sensors are placed on the subject's scalp, face, chest, abdomen, limbs and other parts. The placement of electrodes and sensors should follow internationally accepted standards to ensure the accuracy and comparability of monitoring data. The electrodes and sensors are connected to the PSG monitoring equipment to ensure stable and reliable signal transmission.

[0026] According to the monitoring needs, the sampling rate, filter and other parameters of the equipment are adjusted to obtain the best monitoring effect. After the subject falls asleep, the PSG monitoring device will begin to continuously and synchronously record physiological signals such as electroencephalogram (EEG), electromyogram (EMG), electrooculogram (EOG), electrocardiogram (ECG), and respiratory recording device. If needed, it can also simultaneously monitor physiological indicators reflecting cardiovascular function such as blood pressure and pulse.

[0027] The PSG monitoring device will automatically collect physiological signal data during sleep throughout the night, namely a polysomnogram, and store it inside the device or in an external storage device.

[0028] The collected data are preprocessed, including filtering, signal calibration and other steps to improve the accuracy and reliability of the data. Professional sleep analysis software is used to conduct detailed analysis of the preprocessed data, including evaluation of sleep structure, process, abnormal EEG, respiratory function, cardiovascular function, etc. Based on the analysis results and combined with clinical information, the sleep disorders of the subjects are diagnosed, classified and differentially diagnosed.

[0029] Here, multiple physiological signals can be determined in the following ways: First, ensure that the equipment used to record polysomnography (such as electroencephalograph, electrooculogram, electromyograph, etc.) is in normal working condition and has been correctly connected to the target object. According to research or diagnostic needs, set the parameters for recording polysomnography, such as sampling rate, recording time, number of channels, etc. Start the equipment and record the polysomnography of the target object within the preset time period. Ensure that the equipment is stable and the data is complete during the recording process. Store the collected polysomnography data in a safe and reliable storage medium for subsequent analysis and processing.

[0030] See also Figure 2 , Figure 2 This is a flow chart for determining multiple physiological signals provided by an embodiment of the present application. Figure 2 As shown in , the sleep stage detection method provided in the embodiment of the present application includes: S201, determining a filtering range for each physiological signal; First, a preliminary analysis is performed on each physiological signal to understand its frequency characteristics and noise distribution. Then, based on the signal analysis results, the filter range of each physiological signal is set. The filter range should be able to retain the useful information in the signal while removing noise and interference.

[0031] Finally, according to the filtering range, select the appropriate filter type (such as low-pass filter, high-pass filter, band-pass filter, etc.) and parameters.

[0032] During specific preprocessing, the desired channel needs to be filtered to a reasonable range.

[0033] Specifically: EEG filtering of 0.3~35HZ, EOG filtering of 0.3~35HZ, ECG filtering of 0.3~70HZ and EMG filtering of 10~100HZ.

[0034] In addition, the 50HZ of the relevant channel can be notched to remove power frequency interference.

[0035] S202 . For each physiological signal, filter the physiological signal in the polysomnogram according to a corresponding filtering range to obtain a plurality of physiological signals of the physiological signal within the corresponding filtering range.

[0036] Preprocess the original physiological signal, such as removing DC components and trend items, to improve the filtering effect, and input the preprocessed physiological signal into the corresponding filter for filtering. Ensure the integrity and continuity of the signal during the filtering process, and post-process the filtered physiological signal, such as removing edge effects and smoothing, to improve the signal quality.

[0037] Specifically, for the physiological signals under each channel of the polysomnogram, multiple physiological signals under the channel are divided according to preset time to obtain multiple sleep stages.

[0038] Each sleep stage includes multiple physiological signals under multiple channels within the preset time; Here, the physiological signals under each channel are divided according to a preset time (such as 30 seconds, 1 minute, etc., which can be specifically limited according to actual conditions) to obtain multiple sleep stages, ensuring that the signals in each sleep stage are representative, and organizing multiple physiological signals under multiple channels in each sleep stage to form a unified data format.

[0039] Data segmentation is to facilitate data processing in deep learning models. At the same time, it is also to make the experimental data obtained more accurate and reliable, so as to lay a good foundation for further experiments and analysis, and provide favorable guarantees. The segmented data must also be baseline corrected. In the AASM (American Academy of Sleep Medicine) standard, each sleep stage is set to 30 seconds, so the data is cut into segments every 30 seconds.

[0040] Each sleep stage is sequentially input into the sleep stage model to obtain the sleep stage result of the target object. The sleep stage model includes a first model, a second model and a third model.

[0041] The sorted data are sequentially input into the sleep staging model, which includes a first model, a second model and a third model, each of which is responsible for different processing tasks.

[0042] See also Figure 3 , Figure 3 Schematic diagram of the sleep stage model provided in the embodiment of the present application. Figure 3 As shown in , the sleep staging model provided in the embodiment of the present application includes a first model (New MRCNN), a second model (DoubleFlow) and a third model (New SE Block).

[0043] S102: Input a plurality of physiological signals into a first model to obtain frequency signals with different frequency characteristics.

[0044] See also Figure 4 , Figure 4 A schematic diagram of the first model provided in an embodiment of the present application.

[0045] like Figure 4 As shown in , the first model provided in the embodiment of the present application includes multiple CNN sequences, the input of each CNN sequence is multiple physiological signals, and the output of each CNN sequence is a frequency signal corresponding to a frequency feature, Figure 4 The figure below is a CNN sequence.

[0046] like Figure 4 As shown, each block is a module inside the first model, and the specific contents are as follows: Conv1d: This stands for 1D Convolutional Layer. In deep learning, convolutional layers are used to extract features of input data. 1D convolutional layers are usually used to process time series data or 1D signals such as audio signals or text data. In this model, Conv1d layer is used to extract features of three different frequencies.

[0047] BatchNorm: This stands for Batch Normalization. Batch Normalization is a technique designed to speed up training, improve model stability and generalization. Its core function is to solve the problem of Internal Covariate Shift during training by normalizing the input distribution of the intermediate layers of the neural network. It stabilizes the training process by normalizing each mini-batch of data so that the input data of each layer has a similar distribution.

[0048] ELU: This stands for Exponential Linear Unit and is a type of activation function. Activation functions are very important in neural networks because they introduce nonlinearity, allowing the network to learn complex patterns. The ELU function outputs the input value when the input is positive and a negative value close to zero when the input is negative, which helps alleviate the vanishing gradient problem.

[0049] MaxPool: This stands for Maximum Pooling Layer. Pooling layers are often used to reduce the dimensionality of data (i.e. downsampling) while retaining important features. The maximum pooling layer works by selecting the maximum value in each pooling window, which helps reduce the amount of computation and extract the most salient features.

[0050] Dropout: This is a regularization technique used to prevent neural networks from overfitting. During training, Dropout randomly discards a portion of the neurons (and their connections) in the network, which prevents the network from being too dependent on specific neurons, thereby improving the generalization ability of the model.

[0051] Each CNN sequence is a 1-dimensional CNN sequence, and parameter ratios between different CNN sequences are different. Preferably, the first model may include three 1-dimensional CNN sequences.

[0052] For each CNN sequence, frequency features of multiple physiological signals are extracted according to the parameter ratio corresponding to the CNN sequence to obtain frequency signals under the frequency features corresponding to the CNN sequence.

[0053] Through three 1D CNN sequences, three frequency features are extracted according to different parameter ratios, and the three features are integrated and processed and transmitted to the Double Flow framework.

[0054] Here, the first model is a composite model (MRCNN) composed of multiple one-dimensional convolutional neural network (1D CNN) sequences. Each CNN sequence is specially designed to extract information under specific frequency characteristics from multiple input physiological signals. Different CNN sequences have different parameter ratios such as the number of network layers, convolution kernel size, step size, and padding method. The setting of these parameter ratios is based on the understanding of the prior knowledge of the frequency characteristics of physiological signals and experimental optimization. By adjusting these parameters, each CNN sequence can focus on different frequency components in the input signal.

[0055] The input layer of each CNN sequence receives multiple physiological signals as input. After appropriate preprocessing, these signals are organized into a data format suitable for convolution operations. In the convolution layer of the CNN sequence, convolution kernels with different parameter ratios are used to perform convolution operations on the input signals. These convolution kernels can capture local features in the signal and generate feature maps.

[0056] In order to reduce the dimension of the feature map and reduce the amount of calculation, a pooling layer is usually added after the convolution layer. Pooling operations (such as maximum pooling or average pooling) can retain important feature information while removing redundant information. By stacking multiple convolutional layers and pooling layers, each CNN sequence can gradually extract higher-level features from the input signal. Finally, the output layer of each CNN sequence will generate a frequency signal corresponding to its specific frequency characteristics.

[0057] S103. Input multiple frequency signals into the second model to obtain two intermediate layer signals.

[0058] See also Figure 5 , Figure 5 A schematic diagram of the second model provided in an embodiment of the present application.

[0059] like Figure 5 As shown in , the second model provided in the embodiment of the present application includes a string operation module and two first channels. The string operation module is used to integrate multiple frequency signals into a target frequency signal. The input of each first channel is the target frequency signal received from the string operation module, and the output of each first channel is the corresponding first intermediate layer signal.

[0060] Among them, the two first channels corresponding to the second model are both 1-dimensional CNN sequences.

[0061] like Figure 5 As shown, each block is a module inside the second model, and the specific contents are as follows: Concat: The full name is Concatenate. The Concat operation can connect two or more feature vectors along a certain dimension (usually the channel dimension). In this way, the originally independent features are integrated into a larger feature space, providing richer information for subsequent convolution, pooling and other operations.

[0062] exist Figure 5 In the figure, the "Concat" layer is at the top, indicating that it is one of the starting points of the network processing flow. It receives two or more input feature vectors, concatenates them together, and then passes them to the subsequent "Conv1d" layer for further processing.

[0063] Conv1d: This stands for 1D convolutional layer. In deep learning, convolutional layers are used to extract features from input data. 1D convolutional layers are particularly useful for processing time series data or 1D signals such as audio signals, brain wave data, or text data. They extract local features by sliding one or more convolution kernels (also called filters).

[0064] BatchNorm: This stands for Batch Normalization layer. Batch normalization is a technique used to speed up deep neural network training and reduce model overfitting. It normalizes each channel of each mini-batch of data so that the input data of each layer has a similar distribution. This helps stabilize the training process and improves the convergence speed and performance of the model.

[0065] ReLU: This stands for Rectified Linear Unit and is a commonly used non-linear activation function. The ReLU function outputs the input value when the input is positive and outputs zero when the input is negative. It helps introduce non-linearity, enables the network to learn complex patterns, and alleviates the gradient vanishing problem.

[0066] Adaptive AvgPool: This stands for Adaptive Average Pooling Layer. Adaptive pooling layers are often used to reduce the spatial dimensionality of data while retaining important feature information. Adaptive average pooling layers work by calculating the average of the input data at each output position. It allows the network to handle input data of varying sizes in a more flexible way.

[0067] Here, the results of MRCNN are integrated and processed by Double Flow, and the outputs are input to the improved SE Block after passing through two 1D CNN sequences respectively.

[0068] The second model is a complex structure (Double Flow) designed to process frequency signals, which includes two key subcomponents: a string operation module and two first channels.

[0069] The main task of the string operation module is to integrate multiple input frequency signals into a single target frequency signal. This process may involve signal merging, superposition, weighting or other forms of signal processing technology, the purpose is to extract key information from the original signal and provide a clearer and more effective input for subsequent processing.

[0070] The two first channels are parallel processing paths, each receiving the target frequency signal from the string operation module as input. These channels are usually a sequence of 1D convolutional neural networks (1D CNNs), which can extract and transform the input signal to generate their own first intermediate layer signals. These intermediate layer signals are important inputs for subsequent processing steps, and they contain the signal features after preliminary processing.

[0071] See also Figure 6 , Figure 6 This is a flow chart of obtaining two first intermediate layer signals provided in an embodiment of the present application. Figure 6 As shown in , the sleep stage detection method provided in the embodiment of the present application includes: S601, input multiple frequency signals into a string operation module, and output a target frequency signal; In this step, the original multiple frequency signals are input into the string operation module, which integrates and processes these signals according to preset algorithms or rules, and finally outputs a single target frequency signal. This process may involve complex signal processing algorithms, such as Fourier transform, filtering, noise reduction, etc., to ensure the quality and accuracy of the output signal.

[0072] S602: Input the target frequency signals into the first channels respectively to obtain two first intermediate layer signals.

[0073] After the processing of the string operation module is completed, the obtained target frequency signal is input into two first channels respectively. The two channels work in parallel and process the input signal independently.

[0074] Since these channels are usually 1D CNN sequences, they are able to leverage the power of convolution operations to extract features from the input signal. These features may include the frequency components, amplitude changes, phase information, etc. of the signal, which are crucial for subsequent signal analysis or recognition tasks.

[0075] Through the processing of these two channels, two first intermediate layer signals are obtained. These signals contain signal information after preliminary feature extraction, providing important input for subsequent processing steps. These intermediate layer signals may be further processed, such as feature fusion, classification, regression and other tasks, to complete specific signal processing or recognition tasks.

[0076] S104. Input the two first intermediate layer signals into the third model to obtain the sleep staging result of the target object.

[0077] Here, the two results of Double FLow are input into the improved SE Block framework, and the output obtained is the final classification result.

[0078] The two first intermediate layer signals are input into the third model to accurately obtain the sleep staging results of the target object. This step is a key link in the sleep monitoring and analysis process. It relies on advanced algorithms and models to improve the accuracy and reliability of sleep staging.

[0079] The third model takes into account the complexity and diversity of sleep data. The model contains a staging result determination module and two parallel processing second channels. These two channels are independent of each other but have similar functions. They are designed to perform in-depth analysis and feature extraction of the input first intermediate layer signal from different angles or feature spaces.

[0080] See also Figure 7 , Figure 7 A schematic diagram of the third model provided in an embodiment of the present application.

[0081] like Figure 7 As shown in , the third model provided in the embodiment of the present application includes a staging result determination module and two second channels, the input of each second channel is the first intermediate layer signal output from the first channel, and the output of each second channel is the corresponding second intermediate layer signal, and the staging result determination module is used to map the second intermediate layer signal to a sleep staging result, wherein the two first intermediate layer signals are input into the two second channels to obtain the second intermediate layer signal corresponding to each second channel; the two second intermediate layer signals are input into the staging result determination module to obtain the sleep staging result of the target object.

[0082] like Figure 7 As shown, each block is a module inside the third model, and the specific contents are as follows: Linear: This stands for linear layer, also known as fully connected layer or dense layer. In a neural network, the linear layer is one of the most basic layers, which performs a linear transformation on the input data, that is, y=Wx+b, where W is the weight matrix, x is the input data, b is the bias term, and y is the output data. The linear layer can learn the linear relationship between the input data.

[0083] ELU: This stands for Exponential Linear Unit, which is a nonlinear activation function. Activation functions are very important in neural networks because they introduce nonlinearity that allows the network to learn complex patterns. The ELU function outputs the input value when the input is positive, and outputs a negative value close to zero when the input is negative (specifically a(exp(x)−1), where a is a hyperparameter that is usually set to 1). This helps alleviate the vanishing gradient problem because the derivative of ELU is non-zero when the input is negative.

[0084] Sigmoid: is a nonlinear activation function that compresses the input data between 0 and 1. The Sigmoid function is often used in the output layer of binary classification problems because it can map any value between 0 and 1 to represent probability.

[0085] Figure 7 The “+” position in indicates that the second intermediate layer signal corresponding to each second channel is the sum of the first intermediate layer signal and the target intermediate layer signal. Here, the target intermediate layer signal is the signal obtained after the first intermediate layer signal is processed by the module in the second channel.

[0086] Figure 7 The “×” position in represents that the second intermediate layer signals outputted by each second channel are multiplied to determine a target signal, and the target signal is inputted into the staging result determination module to obtain the sleep staging result of the target object.

[0087] In this step, the input of each second channel comes directly from the first intermediate layer signal output by the first channel. These signals have been preliminarily processed and contain basic information about the target object's sleep state, such as brain waves, eye movement signals, muscle activity, etc.

[0088] Inside each second channel, the input first intermediate layer signal undergoes a series of complex transformations and feature extraction processes. These processes may include deep learning techniques such as convolution operations, pooling operations, and fully connected layers, which aim to extract highly discriminative features for sleep stages from the original signal. Ultimately, each channel outputs a deeply processed second intermediate layer signal that is more focused on specific sleep features or stages.

[0089] After obtaining the two second intermediate layer signals, they will be simultaneously input into the staging result determination module. The module is designed with an efficient feature fusion mechanism that can comprehensively consider the information output by the two channels to achieve information complementarity and enhancement.

[0090] Based on the fused feature information, the staging result determination module will use advanced machine learning algorithms (such as support vector machines, random forests, deep learning networks, etc.) for final mapping and decision-making. This process maps complex feature vectors into specific sleep staging results, such as the various stages of non-rapid eye movement (NREM) (N1, N2, N3) and rapid eye movement (REM).

[0091] Preferably, the sleep staging result includes a first result, a second result, a third result, a fourth result and a fifth result. The sleep staging result of the target object can be determined by: Each sleep stage is input into the sleep stage model in turn to determine the result corresponding to the sleep stage.

[0092] Here, the first result, the second result, the third result, the fourth result and the fifth result are wake (awake period), NREM1 (non-rapid eye movement sleep stage 1), NREM2 (non-rapid eye movement sleep stage 2), NREM3 (non-rapid eye movement sleep stage 3) and REM (rapid eye movement sleep period), respectively.

[0093] Among them, wake is the preparation stage before falling asleep, during which people are conscious and can feel changes in the surrounding environment. EEG (electroencephalogram) signals are mainly expressed as beta waves and alpha waves.

[0094] NREM1 (non-rapid eye movement sleep 1), also known as light sleep, is the typical transition period from wakefulness to sleep. This stage is the lightest sleep period, easy to wake up, and there will be some muscle tremors and reduced sensory reactions.

[0095] NREM2 (non-rapid eye movement sleep stage 2), also known as light sleep or intermediate sleep stage, is a stage in which the human body further relaxes and breathing and heart rate gradually slow down. This stage usually accounts for 45% to 50% of the total sleep time of normal middle-aged adults, and is the stage in which most of the sleep time is spent.

[0096] NREM3 (non-rapid eye movement sleep stage 3), also known as deep sleep or slow-wave sleep, is the most relaxing stage of the human body, and the most active stage of body repair and recovery. This stage usually accounts for 10% to 20% of the total sleep time of young / middle-aged adults, and decreases with age. During deep sleep, the muscles of the whole body are relaxed, the sensitivity to external stimuli is reduced, the breathing is stable, and it is difficult to be awakened.

[0097] REM (rapid eye movement sleep) occurs after the NREM period, usually once every 90 to 120 minutes, with a lower proportion in the first half of the night and mainly concentrated in the second half of the night. During the REM period, the eye muscles are active, the eyeballs move rapidly, and dreams may occur. It is generally difficult to wake up, but if you wake up, you can usually remember your dreams and the content of your dreams. During this stage, the body's heart rate, blood pressure, breathing, etc. all fluctuate greatly, such as increased blood pressure, faster heart rate, fast and irregular breathing, etc., and a large amount of adrenaline is secreted, and other muscles except the eye muscles are relaxed.

[0098] By inputting a plurality of physiological signals within a sleep stage of every 30 seconds into the sleep stage model, a target result among five sleep stage results corresponding to the sleep stage can be determined.

[0099] The sleep staging detection method and sleep staging detection device provided in the embodiments of the present application obtain a polysomnogram of a target object within a preset time period; input multiple physiological signals into a first model to obtain frequency signals with different frequency characteristics; input multiple frequency signals into a second model to obtain two intermediate layer signals; input two first intermediate layer signals into a third model to obtain the sleep staging result of the target object. Through the present application, multiple physiological signals in a polysomnogram are processed and analyzed, thereby improving the accuracy of the sleep staging result.

[0100] The present application is capable of adaptively processing physiological signals of multiple channels, reducing the complex data processing operations of the original sleep staging algorithm. In addition, as shown in Table 1 below, good classification accuracy has been achieved on multiple public data sets (with different numbers of channels).

[0101] Table 1

[0102] Based on the same inventive concept, the embodiments of the present application also provide a sleep stage detection device corresponding to the sleep stage detection method. Since the principle of solving the problem by the device in the embodiments of the present application is similar to the above-mentioned sleep stage detection method in the embodiments of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0103] See also Figure 8 , Figure 8 This is a schematic diagram of the structure of the sleep stage detection device provided in the embodiment of the present application. Figure 8 As shown in , the sleep stage detection device 800 includes: A polysomnography acquisition module 801 is used to acquire a polysomnography of a target object within a preset time period, wherein the polysomnography includes a plurality of physiological signals; A frequency signal acquisition module 802 is used to input the multiple physiological signals into a first model to obtain frequency signals under different frequency characteristics, wherein the first model includes multiple CNN sequences, the input of each CNN sequence is the multiple physiological signals, and the output of each CNN sequence is a frequency signal under a corresponding frequency characteristic; A first intermediate layer signal acquisition module 803 is used to input the multiple frequency signals into a second model to obtain two first intermediate layer signals, the second model includes a string operation module and two first channels, the string operation module is used to integrate the multiple frequency signals into a target frequency signal, the input of each first channel is the target frequency signal received from the string operation module, and the output of each first channel is the corresponding first intermediate layer signal; The sleep staging result obtaining module 804 is used to input the two first intermediate layer signals into the third model to obtain the sleep staging result of the target object.

[0104] See also Fig. 9 , Fig. 9 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present application. Fig. 9 As shown in , the electronic device 300 includes a processor 310 , a memory 320 and a bus 330 .

[0105] The memory 320 stores machine-readable instructions executable by the processor 310. When the electronic device 300 is running, the processor 310 communicates with the memory 320 via the bus 330. When the machine-readable instructions are executed by the processor 310, the above-mentioned Figure 1 The steps of the sleep stage detection method in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0106] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 The steps of the sleep stage detection method in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0108] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0109] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0110] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0111] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0112] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A sleep stage detection method, characterized in that: include: Acquiring a polysomnogram of a target subject within a preset time period, wherein the polysomnogram includes a plurality of physiological signals; Input the multiple physiological signals into a first model to obtain frequency signals under different frequency characteristics, wherein the first model includes multiple CNN sequences, the input of each CNN sequence is the multiple physiological signals, and the output of each CNN sequence is a frequency signal under a corresponding frequency characteristic; Input the multiple frequency signals into a second model to obtain two intermediate layer signals, the second model comprising a string operation module and two first channels, the string operation module is used to integrate the multiple frequency signals into a target frequency signal, the input of each first channel is the target frequency signal received from the string operation module, and the output of each first channel is the corresponding first intermediate layer signal; The two first intermediate layer signals are input into a third model to obtain a sleep staging result of the target object.

2. The method according to claim 1, characterized in that The method further comprises: For the physiological signals under each channel of the polysomnogram, the multiple physiological signals under the channel are divided according to a preset time to obtain multiple sleep stages, each sleep stage including the multiple physiological signals under the multiple channels within the preset time; Each sleep stage is sequentially input into a sleep stage model to obtain a sleep stage result of the target object, wherein the sleep stage model includes the first model, the second model and the third model.

3. The method according to claim 2, characterized in that Each CNN sequence is a 1-dimensional CNN sequence, and the parameter ratios between different CNN sequences are different. The step of inputting the plurality of physiological signals into a first model to obtain frequency signals with different frequency characteristics includes: For each CNN sequence, frequency features of the multiple physiological signals are extracted according to the parameter ratio corresponding to the CNN sequence to obtain a frequency signal under the frequency feature corresponding to the CNN sequence.

4. The method according to claim 1, characterized in that The two first channels corresponding to the second model are both 1-dimensional CNN sequences. The step of inputting the multiple frequency signals into a second model to obtain two first intermediate layer signals includes: Input the multiple frequency signals into a string operation module, and output a target frequency signal; The target frequency signals are input into the first channels respectively to obtain two first intermediate layer signals.

5. The method according to claim 2, characterized in that: The third model includes a staging result determination module and two second channels, the input of each second channel is a first intermediate layer signal output from the first channel, the output of each second channel is a corresponding second intermediate layer signal, and the staging result determination module is used to map the second intermediate layer signal to a sleep staging result. The step of inputting the two first intermediate layer signals into a third model to obtain a sleep staging result of the target object includes: Inputting the two first intermediate layer signals into the two second channels to obtain a second intermediate layer signal corresponding to each second channel; The two second intermediate layer signals are input into the sleep staging result determination module to obtain the sleep staging result of the target object.

6. The method according to claim 1, characterized in that The plurality of physiological signals are determined by: Determine the filtering range for each physiological signal; For each physiological signal, filtering is performed on the physiological signal in the polysomnogram according to a corresponding filtering range to obtain a plurality of physiological signals of the physiological signal within the corresponding filtering range.

7. The method according to claim 5, characterized in that The sleep staging results include a first result, a second result, a third result, a fourth result and a fifth result. The sleep stage result of the target object is determined by: Each sleep stage is sequentially input into the sleep stage model to determine the result corresponding to the sleep stage.

8. A sleep stage detection device, characterized in that: include: A polysomnography acquisition module, used to acquire a polysomnography of a target object within a preset time period, wherein the polysomnography includes a plurality of physiological signals; A frequency signal acquisition module, used for inputting the multiple physiological signals into a first model to obtain frequency signals under different frequency characteristics, wherein the first model includes multiple CNN sequences, the input of each CNN sequence is the multiple physiological signals, and the output of each CNN sequence is a frequency signal under a corresponding frequency characteristic; a first intermediate layer signal acquisition module, used for inputting the multiple frequency signals into a second model to obtain two first intermediate layer signals, the second model comprising a string operation module and two first channels, the string operation module being used for integrating the multiple frequency signals into a target frequency signal, the input of each first channel being the target frequency signal received from the string operation module, and the output of each first channel being the corresponding first intermediate layer signal; The sleep staging result obtaining module is used to input the two first intermediate layer signals into the third model to obtain the sleep staging result of the target object.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate via the bus, and the processor executes the machine-readable instructions to perform the steps of any method as claimed in claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are executed.

Images