Sleep microstructure determination method and device, computer equipment and storage medium

Abstract

The invention relates to a sleep microstructure determination method and device, computer equipment and a storage medium. At least one detection physiological signal of the detected object in the sleep process is obtained, and the starting moment and the duration time of the target microstructure of the detected object are determined according to the at least one detection physiological signal and a preset microstructure detection model. In the embodiment of the invention, the microstructure detection model is obtained by training according to the reference probability value and the reference duration, and the reference probability value and the reference duration provide a continuous sequence tag for the microstructure detection model; therefore, deep features in physiological signals can be learned and detected based on the microstructure detection model, and dependence on rich signal processing knowledge or artificial experience is not needed. Moreover, on the basis of accurately detecting whether the target microstructure appears or not, accurate positioning of the occurrence moment and the duration time of the target microstructure can be realized, and more accurate information is provided for clinical diagnosis and treatment.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular to a method, apparatus, computer equipment and storage medium for determining the microstructure of sleep. Background Art

[0002] The microstructure of sleep includes sleep micro-arousals, obstructive sleep apnea, sleep spindles and transitions between sleep stages. Testing the microstructure of sleep can effectively avoid cardiovascular dysfunction, obesity, depression, low blood pressure, irritability and memory loss. Take sleep micro-arousals as an example. During non-rapid eye movement sleep, there is a sudden change in the frequency of the electroencephalogram (EEG), accompanied by the appearance of alpha waves, theta waves or waveforms with a frequency exceeding 16Hz, and this change must last for at least 3 seconds. During REM sleep, in addition to EEG changes, there is also a requirement for at least 1 second of increased chin electromyogram (EMG). However, humans are usually unaware of this short awakening during sleep, but frequent occurrences can lead to a significant decrease in sleep quality.

[0003] Currently, deep learning methods are used to classify each sampling point in the physiological signal, but this method cannot locate the microstructure during sleep. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment and storage medium for determining the microstructure of sleep that can obtain the starting time and duration of the microstructure of sleep in order to solve the above technical problems.

[0005] In a first aspect, the present application provides a method for determining the microstructure of sleep, comprising:

[0006] Acquiring at least one detected physiological signal of the detected subject during sleep;

[0007] Based on at least one of the detected physiological signals and a preset microstructure detection model, the starting time and duration of the target microstructure of the detected object are determined; the microstructure detection model is trained based on the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and the reference duration are obtained by performing Gaussian processing on the sample physiological signal.

[0008] In one embodiment, determining the starting time and duration of the target microstructure of the detection object according to at least one of the detected physiological signals and a preset microstructure detection model includes:

[0009] Inputting at least one of the detected physiological signals into the microstructure detection model to obtain a predicted probability value and a predicted duration corresponding to each sampling moment in the detected physiological signal;

[0010] Determining the target center moment according to each of the predicted probability values ​​and a preset microstructure merging condition;

[0011] The starting time and duration of the target microstructure are determined according to the target center time and each predicted duration.

[0012] In one embodiment, determining the target center moment according to each of the predicted probability values ​​and a preset microstructure merging condition includes:

[0013] Determining a plurality of candidate central moments according to each of the predicted probability values ​​and a preset probability threshold;

[0014] Determine the target microstructure according to the microstructure merging condition and the plurality of candidate center moments; the microstructure merging condition includes that the time difference between two adjacent candidate center moments is less than a preset time difference threshold;

[0015] The sampling moment corresponding to the maximum predicted probability value in the target microstructure is taken as the target center moment.

[0016] In one embodiment, the process of acquiring the microstructure detection model includes:

[0017] Acquiring a sample physiological signal, and annotating the sample microstructure in the sample physiological signal to obtain a reference label array;

[0018] Performing Gaussian processing on the reference tag array to obtain each reference probability value and each reference duration;

[0019] The initial detection model is trained according to each of the reference probability values ​​and each of the reference durations to obtain the microstructure detection model.

[0020] In one embodiment, performing Gaussian processing on the reference tag array to obtain each reference probability value and each reference duration includes:

[0021] Determining a reference mean value of the sample microstructure according to a reference start time and a reference end time of the sample microstructure in the reference tag array;

[0022] Determining a reference standard deviation of the sample microstructure according to the reference mean and the reference starting time;

[0023] According to the reference mean and the reference standard deviation, Gaussian processing is performed on the reference label array to obtain each reference probability value;

[0024] Each of the reference durations is determined according to the reference start time, the reference end time, each of the reference probability values, and a sampling frequency of the sample physiological signal.

[0025] In one embodiment, determining each reference duration according to the reference start time, the reference end time, each reference probability value and the sampling frequency of the sample physiological signal includes:

[0026] Determining a candidate duration according to a difference between the reference start time and the reference end time;

[0027] The ratio of the candidate duration to the sampling frequency is used as the reference duration corresponding to the first reference probability value in the reference probability value; the first reference probability value is the reference probability value corresponding to the center moment of the sample microstructure;

[0028] The preset duration is used as a reference duration corresponding to a second reference probability value among the reference probability values ​​except the first reference probability value.

[0029] In one embodiment, the initial detection model is trained according to each of the reference probability values ​​and each of the reference durations to obtain the microstructure detection model, including:

[0030] Inputting the sample physiological signal into the initial detection model to obtain a probability value of each sample and a duration of each sample;

[0031] Acquire a first loss value between each of the reference probability values ​​and the corresponding sample probability value based on a first loss function;

[0032] Acquire a second loss value between each of the reference durations and the corresponding sample duration based on a second loss function; the first loss function is determined based on the second loss function;

[0033] The initial detection model is trained according to the first loss value and the second loss value to obtain the microstructure detection model.

[0034] In a second aspect, the present application also provides a device for determining the microstructure of sleep, comprising:

[0035] An acquisition module, used to acquire at least one detected physiological signal of the detected object during sleep;

[0036] A determination module is used to determine the starting time and duration of the target microstructure of the detection object according to at least one of the detected physiological signals and a preset microstructure detection model; the microstructure detection model is trained according to the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and the reference duration are obtained by performing Gaussian processing on the sample physiological signal.

[0037] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method steps of the first aspect when executing the computer program.

[0038] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method steps of the first aspect when executed by a processor.

[0039] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which implements the method steps of the first aspect when executed by a processor.

[0040] The above-mentioned sleep microstructure determination method, device, computer equipment and storage medium obtain at least one detection physiological signal of the detection object during sleep, and determine the starting time and duration of the target microstructure of the detection object according to at least one detection physiological signal and a preset microstructure detection model; the microstructure detection model is trained according to the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and the reference duration are obtained by Gaussian processing of the sample physiological signal. In the embodiment of the present application, since the microstructure detection model is trained according to the reference probability value and the reference duration, the reference probability value and the reference duration provide a continuous sequence label for the microstructure detection model, so that the deep features in the detection physiological signal can be learned based on the microstructure detection model, without relying on rich signal processing knowledge or manual experience, which greatly improves the universality and ease of use of the detection. Moreover, the starting time and duration of the target microstructure can be obtained on the basis of accurately detecting whether the target microstructure appears, so as to accurately locate the occurrence time and duration of the target microstructure, and provide more accurate information for clinical diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A diagram of an application environment of a method for determining a microstructure of sleep in one embodiment;

[0043] Figure 2 A schematic flow chart of a method for determining the microstructure of sleep in one embodiment;

[0044] Figure 3 A schematic diagram of a microstructure detection model provided by an embodiment;

[0045] Figure 4 A schematic flow chart of a method for determining a starting time and a duration in one embodiment;

[0046] Figure 5 A schematic flow chart of a method for determining a starting time and a duration in another embodiment;

[0047] Figure 6 A schematic diagram of prediction results in one embodiment;

[0048] Figure 7 A schematic diagram of a process for obtaining a microstructure detection model in one embodiment;

[0049] Figure 8 A schematic diagram of a method for acquiring a label array of a sample physiological signal in one embodiment;

[0050] Fig. 9 A schematic diagram of a label of a sample physiological signal in one embodiment;

[0051] Fig.10 A schematic flow chart of a method for determining a reference probability value and a reference duration in one embodiment;

[0052] Fig.11 A schematic diagram of a flow chart of a microstructure detection model in another embodiment;

[0053] Fig.12 A schematic flow chart of a method for determining the microstructure of sleep in another embodiment;

[0054] Fig.13 A structural block diagram of a device for determining a microstructure of sleep in one embodiment. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] In traditional methods, the detection of microstructures mainly relies on manual labeling by experienced clinicians, and the labeling rate is low, which cannot meet the needs of large-scale data processing; at the same time, due to the subjective differences of doctors, the consistency of the labeling results is often difficult to ensure, resulting in the reliability of the detection results of the microstructure being affected. In addition, although some existing machine learning methods can reduce the manual burden to a certain extent, they usually require rich signal processing knowledge or manual experience, which limits the universality and ease of use of these methods. Therefore, it is particularly important to develop a sleep micro-awakening detection algorithm based on deep learning. In addition, the existing algorithms can only determine whether there is a microstructure in the detection of microstructures, but cannot accurately locate the specific time when the microstructure occurs and its duration. This coarse-grained detection result cannot meet the accuracy requirements of clinical diagnosis and treatment. Therefore, the present application proposes a microstructure determination method, device, computer equipment and storage medium that can solve the above technical problems.

[0057] The sleep microstructure determination method provided in the embodiment of the present application can be applied to Figure 1 The application environment includes a computer device, which may be a server, and its internal structure diagram may be as shown in FIG. Figure 1 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store relevant data for determining the microstructure during sleep. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for determining the microstructure of sleep is implemented. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0058] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0059] In an exemplary embodiment, Figure 2 As shown, a method for determining the microstructure of sleep is provided, and the method is applied to Figure 1 The computer device in the example is used for explanation, including the following S201 to S202. Among them:

[0060] S201, obtaining at least one detected physiological signal of a detected subject during sleep.

[0061] Among them, when detecting different microstructures, the detected physiological signals may be different; for the same microstructure, the detected physiological signals may be of one type or of multiple different types. For example, when the microstructure is sleep microarousal, the detected physiological signals may include at least one type of electroencephalogram, electrooculogram (EOG), electromyogram and electrocardiogram (ECG). When multiple types of physiological signals such as EEG, EMG, EOG and ECG are included at the same time, it may also be called polysomnography (PSG); when the microstructure is obstructive sleep apnea, the detected physiological signals may include one or more types of respiratory signals such as abdominal effort (ABD), thoracic effort (THO), cannula flow (Canflow), and oxygen saturation (SpO2).

[0062] In an embodiment of the present application, the detection subject wears a wearable device, uses the wearable device to collect detection physiological signals during sleep, and sends the detection physiological signals to a computer device, which acquires them in real time.

[0063] In a possible implementation, the detected physiological signal may also be stored in a storage medium of the computer device, and after receiving an instruction from the user, the detected physiological signal is obtained from the storage medium.

[0064] S202, determining the starting time and duration of the target microstructure of the detection object based on at least one detected physiological signal and a preset microstructure detection model; the microstructure detection model is trained based on the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and reference duration are obtained by performing Gaussian processing on the sample physiological signal.

[0065] Among them, the microstructure detection model is trained based on the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and reference duration are obtained by Gaussian processing of the sample physiological signal. Based on each reference probability value, the reference center moment of the sample microstructure can be obtained, and based on each reference duration, the duration of the sample microstructure at the reference center moment can be obtained. A continuous sequence label that can be used for microstructure detection model training is successfully constructed, avoiding the current situation of discrete label learning in traditional technology.

[0066] Among them, the microstructure detection model can be a model that integrates multiple time series signal detection algorithms. For example, the microstructure detection model includes but is not limited to Deepsleep2.0, Recurrent Neural Network (RNN) and Deep Residual Networks (ResNet) and other models that can process continuous time series data. Each of the above networks has its own unique advantages, and the above networks can also be combined to obtain a microstructure detection model, which can more comprehensively extract the key features in the detection of physiological signals, thereby improving the accuracy of microstructure detection.

[0067] Figure 3 A schematic diagram of a microstructure detection model provided in an embodiment of the present application, such as Figure 3As shown, the microstructure detection model includes an encoding structure, a decoding structure and an output structure. The encoding structure includes 5 encoder blocks, namely encoder block E1, encoder block E2, encoder block E3, encoder block E4 and encoder block E5, each of which includes two convolutional layers and uses ReLU as the activation function. A normalization layer is included after each convolutional layer to improve the stability and convergence speed of the microstructure detection model. A maximum pooling layer is set after two convolutional layers and two normalization layers. The decoding structure includes 4 decoder blocks, namely decoder block D4, decoder block D3, decoder block D2 and decoder block D1, each decoder block consists of 4 double convolutional layers and 4 upsampling layers, that is, an upsampling layer is set after a double convolutional layer. The output structure includes two output heads, the first output head is used for the predicted probability value of each sampling moment, and the second output head is used to output the predicted duration of each sampling moment. Among them, the first output head includes two convolutional layers, each of which is followed by a ReLU activation function. After the two convolution layers, the predicted probability value of each sampling moment is output through the sigmoid function. The second output head includes two convolution layers, and the predicted duration of each sampling moment is directly output after the convolution layers.

[0068] For example, the convolution layer in the above encoder block can be a convolution layer with a convolution kernel size of 7*7 and a stride of 1; the first convolution layer in the first output head can be a convolution layer with a convolution kernel size of 3*3, a stride of 1 and padding, and the second convolution layer in the first output head can be a convolution layer with a convolution kernel size of 1*1 and a stride of 1; the two convolution layers in the second output head can be consistent with the two convolution layers in the first output head.

[0069] In the embodiment of the present application, Figure 3 For example, the detected physiological signal is a PSG signal including EEG, EMG, EOG, and ECG multi-channels, and the length of each detected physiological signal is 1*10000. At least one detected physiological signal is input into the microstructure detection model, and the encoding structure in the microstructure detection model is used to obtain the characteristics of the detected physiological signal at different scales, which is helpful for detecting microstructures of different time lengths. The upsampling layer in the decoding structure restores the feature map output by the encoding structure, and combines the feature map of the encoder block corresponding to the encoding structure. This operation is called skip connection, which effectively retains more spatial information, so that the microstructure detection model can locate the microstructure more accurately. The feature map output by the decoding structure is input into the two output heads respectively to obtain the predicted probability value and predicted duration corresponding to each sampling moment.

[0070] In a possible implementation, the target center moment of the target microstructure can be determined based on each predicted probability value and a preset microstructure merging condition, and the start moment and duration of the target microstructure can be determined based on the target center moment and each predicted duration.

[0071] In another possible implementation, the maximum predicted probability value can be determined from the predicted probability values, and the starting time and duration of the target microstructure can be determined based on the sampling time corresponding to the maximum predicted probability value and the duration of the sampling time corresponding to the maximum predicted probability value.

[0072] In the above-mentioned method for determining the microstructure of sleep, at least one detected physiological signal of the detected object during sleep is obtained, and the starting time and duration of the target microstructure of the detected object are determined according to the at least one detected physiological signal and the preset microstructure detection model; the microstructure detection model is obtained by training according to the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and the reference duration are obtained by Gaussian processing of the sample physiological signal. In the embodiment of the present application, since the microstructure detection model is obtained by training according to the reference probability value and the reference duration, the reference probability value and the reference duration provide a continuous sequence label for the microstructure detection model, so that the deep features in the physiological signal can be learned based on the microstructure detection model, without relying on rich signal processing knowledge or manual experience, which greatly improves the universality and ease of use of the detection. Moreover, the starting time and duration of the target microstructure can be obtained on the basis of accurately detecting whether the target microstructure appears, so as to accurately locate the occurrence time and duration of the target microstructure, and provide more accurate information for clinical diagnosis and treatment.

[0073] Figure 4 FIG. 1 is a flow chart of a method for determining a starting time and a duration in one embodiment. Figure 4 As shown, the embodiment of the present application relates to a possible implementation method of how to determine the starting time and duration of the target microstructure of the detection object according to at least one detected physiological signal and a preset microstructure detection model, including the following steps:

[0074] S401, inputting at least one detected physiological signal into a microstructure detection model to obtain a predicted probability value and a predicted duration corresponding to each sampling moment in the detected physiological signal.

[0075] In the embodiments of the present application, Figure 5As shown, at least one detected physiological signal is input into the microstructure detection model, and the microstructure detection model is used to obtain the characteristics of the detected physiological signal at different scales, and the predicted probability value and predicted duration corresponding to each sampling moment are output based on the obtained characteristic information. Assuming that the length of the detected physiological signal is 10000, that is, it includes 10000 sampling moments, 10000 predicted probability values ​​and 10000 predicted durations are obtained.

[0076] S402, determining the target center moment according to each predicted probability value and a preset microstructure merging condition.

[0077] In the embodiment of the present application, multiple candidate center moments can be determined according to each predicted probability value and a preset probability threshold, and the target microstructure can be determined according to the preset microstructure merging condition and the multiple candidate center moments, and the sampling moment corresponding to the maximum predicted probability value in the target microstructure is used as the target center moment. The microstructure merging condition at this time includes that the time difference between two adjacent candidate center moments is less than the preset time difference threshold.

[0078] In another possible implementation, the initial microstructure can be directly determined according to each predicted probability value, and the initial microstructure can be adjusted based on the microstructure merging condition to obtain the target microstructure, and the sampling time corresponding to the maximum predicted probability value in the target microstructure is used as the target center time. The microstructure merging condition at this time includes that the time difference between the end time of the previous initial microstructure and the start time of the current initial microstructure is less than a preset time difference threshold.

[0079] S403, determining the start time and duration of the target microstructure according to the target center time and each predicted duration.

[0080] In the embodiment of the present application, the predicted duration corresponding to the target center moment is obtained from each predicted duration corresponding to each sampling moment, and the predicted duration corresponding to the target center moment is used as the duration of the target microstructure. According to the duration of the target microstructure and the target center moment of the target microstructure, the starting moment of the target microstructure can be obtained. For example, if the target center moment is the 15th second, and the predicted duration corresponding to the 15th second is 6 seconds, then the duration of the target microstructure is 6 seconds, and the starting moment of the target microstructure is the 12th second. Figure 6As shown, the prediction result diagrams of two (two frames) physiological detection signals are respectively shown, wherein the dotted line is the predicted probability value and predicted duration based on the microstructure detection model, the start time and end time of the target microstructure obtained, and the solid line detects the actual start time and actual end time of the target microstructure corresponding to the physiological signal. It can be seen that the start time and end time predicted in the first frame of the physiological detection signal are basically overlapped with the actual start time and actual end time, and there are differences between the start time and end time predicted in the second frame of the physiological detection signal and the actual start time and actual end time, but the differences are very small.

[0081] In an embodiment of the present application, at least one detected physiological signal is input into a microstructure detection model to obtain predicted probability values ​​and predicted durations corresponding to each sampling moment in the detected physiological signal, and the target center moment is determined according to each predicted probability value and a preset microstructure merging condition, and the start moment and duration of the target microstructure are determined according to the target center moment and each predicted duration. The embodiment of the present application uses a microstructure detection model to quickly analyze detected physiological signals and automatically determine the occurrence time and duration of the microstructure, which not only improves the accuracy and precision of microstructure detection, but also reduces the running time, providing strong support for clinical diagnosis and treatment.

[0082] In one embodiment, the target center moment is determined according to each predicted probability value and a preset microstructure merging condition, including: determining multiple candidate center moments according to each predicted probability value and a preset probability threshold; determining the target microstructure according to the microstructure merging condition and the multiple candidate center moments; the microstructure merging condition includes that the time difference between two adjacent candidate center moments is less than a preset time difference threshold; and taking the sampling moment corresponding to the maximum predicted probability value in the target microstructure as the target center moment.

[0083] Continue to combine Figure 5 As shown, due to the influence of signal noise, there may be a predicted probability value greater than 0 at the sampling moment when the target microstructure does not occur. In order to further improve the determination of the target microstructure. In the embodiment of the present application, each predicted probability value is compared with a preset probability threshold, and the sampling moment corresponding to the predicted probability value greater than the preset probability threshold is used as a plurality of candidate center moments.

[0084] Since it is impossible for multiple microstructures to appear in a short time, the time difference between two adjacent candidate center moments among multiple candidate center moments is determined. If the time difference is less than the preset time difference threshold, it is determined that they belong to the same microstructure, that is, the target microstructure. The sampling moment corresponding to the maximum predicted probability value in the target microstructure is taken as the target center moment of the target microstructure.

[0085] The time difference between two adjacent candidate center moments being less than the preset time difference threshold may include that the adjacent candidate center moments are continuous, or that the two adjacent candidate center moments are discontinuous, but the time difference is less than the preset time difference threshold. For example, the preset time difference threshold is 0.5 seconds, and when the sampling frequency is 100, the difference between the corresponding sampling points of the detected physiological signal is 50 (i.e. Figure 5 The smallest dimensioning unit in .

[0086] Multiple candidate center moments that meet the above microstructure merging conditions are merged into the same target microstructure, and the sampling time corresponding to the maximum probability value of the target microstructure is used as the target center moment. If multiple candidate center moments do not meet the microstructure merging conditions, each candidate center moment corresponds to a target microstructure, and the corresponding candidate center moment is the target center moment of each target microstructure.

[0087] Among them, the probability threshold and the preset time difference threshold can be automatically adjusted according to the sleep characteristics and microstructure patterns of different individuals. This method can better adapt to the detection needs of microstructures in different situations and improve the sensitivity and specificity of detection.

[0088] In the embodiment of the present application, multiple candidate center moments are determined according to each predicted probability value and a preset probability threshold; the target microstructure is determined according to the microstructure merging condition and the multiple candidate center moments; the microstructure merging condition includes that the time difference between two adjacent candidate center moments is less than the preset time difference threshold; the sampling moment corresponding to the maximum predicted probability value in the target microstructure is used as the target center moment. In the embodiment of the present application, the target center moment is determined by the preset probability threshold and the microstructure merging condition, which improves the accuracy of the target center moment determination.

[0089] Figure 7 FIG. 1 is a flow chart of a method for obtaining a microstructure detection model in an embodiment. Figure 7 As shown, the following steps are included:

[0090] S701, acquiring a sample physiological signal, and annotating the sample microstructure in the sample physiological signal to obtain a reference label array.

[0091] The sample microstructure and the target microstructure are of the same type of microstructure.

[0092] Combination Figure 8As shown in the figure, taking the sample microstructure of sleep microarousal as an example, sleep microarousal is a short-term awakening state that briefly invades the sleep process. Its characteristics are that in the N1, N2 and N3 (non-rapid eye movement) sleep periods, the EEG frequency changes suddenly, accompanied by alpha waves, theta waves or waveforms with a frequency exceeding 16Hz (but not including spindle waves), and this change must last for at least 3 seconds. Prior to this, there was at least 10 seconds of stable sleep state. In the rapid eye movement (R) sleep period, in addition to EEG changes, the determination of microarousal requires at least 1 second of chin electromyography (EMG) increase. Based on the changes in sleep microarousal in EEG, the reference start time and reference end time of sleep microarousal in the sample physiological signal can be obtained, and the sampling point of the sampling time when sleep microarousal occurs in the sample physiological signal is marked as 1, and the sampling points of the other sampling times are marked as 0. A binary reference label array can be obtained. The reference label array is as follows Fig. 9 The first one is shown in .

[0093] In one possible implementation, according to the American Academy of Sleep Medicine (AASM) scoring manual, when experts mark sleep microarousals, they will clearly mark the reference start time and reference end time of the microarousals, and use computer equipment to directly obtain sample physiological signals with the reference start time and reference end time of the microarousals marked in the AASM scoring manual, and mark the sampling points at the sampling times when sleep microarousals occur in the sample physiological signals as 1, and mark the sampling points at the remaining sampling times as 0. This can obtain a binary reference label array, which can make full use of the existing label information and improve the efficiency of sample microstructure labeling.

[0094] Optionally, 380 multi-lead PSG signals are acquired in the embodiment of the present application, covering a total of 32 channels of rich information, aiming to meet the screening and diagnosis needs of various sleep-related information. Among them, the sampling frequency of the sample physiological signal is as high as 200Hz, which ensures the accuracy of signal acquisition. The 6 EEG channels (F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1) most closely related to the sleep state, 2 EOG channels (LOC-A2, ROC-A2) and 1 EMG channel are specially selected.

[0095] In order to balance the efficiency and accuracy of data processing, the sampling frequency of the sample physiological signal is resampled to 100Hz. The multi-lead PSG signal with a sampling frequency of 100Hz is preprocessed, including noise removal and signal standardization, to improve the data quality of the sample physiological signal, reduce the interference of subsequent analysis, and ensure the accuracy and reliability of the analysis results. Furthermore, in order to achieve accurate time positioning of micro-awakening, the pre-processed PSG signal is finely segmented in units of preset short-time windows, such as 100s for the short-time window, so as to obtain multiple frames of sleep windows (sample physiological signals), that is, each frame of sleep window includes 10,000 sampling points, that is, the pre-processed PSG signal is segmented into multiple frames of sleep windows using the patching technology, and the reference start time and reference end time of the micro-awakening are clearly marked in each frame. In each frame, not only is the presence of micro-awakenings accurately marked, but the sampling rate of the sample window is also intelligently adjusted according to the presence or absence of micro-awakenings. For frames without micro-awakenings, the sampling rate is lowered to reduce the burden of data processing; while for frames with micro-awakening events, the sampling rate is increased to capture more detailed information about the micro-awakenings. This method of dynamically adjusting the sampling rate not only ensures the efficiency of data processing, but also improves the accuracy of micro-awakening event analysis.

[0096] S702, performing Gaussian processing on the reference tag array to obtain reference probability values ​​and reference durations.

[0097] In an embodiment of the present application, each sample microstructure is traversed, and a reference mean of the sample microstructure is determined based on a reference start time and a reference end time of the sample microstructure in a reference label array, and a reference standard deviation of the sample microstructure is determined based on the reference mean and the reference start time. Gaussian processing is performed on the reference label array based on the reference mean and the reference standard deviation to obtain reference probability values ​​and reference durations, that is, to obtain a label array of reference probability values ​​and a label array of reference durations.

[0098] In a possible implementation, the reference tag array may also be input into a preset Gaussian process regression model to obtain reference probability values ​​and reference durations.

[0099] S703, training the initial detection model according to each reference probability value and each reference duration to obtain a microstructure detection model.

[0100] In an embodiment of the present application, the sample physiological signal can be input into the initial detection model to obtain the sample probability value of each sampling point and the sample duration. The loss function is used to calculate the first loss value between each sample probability value and the corresponding reference probability value. The loss function is then used to calculate the second loss value between each sample duration and the corresponding reference duration. The initial detection model is trained using the first loss value and the second loss value to obtain a microstructure detection model.

[0101] In a possible implementation, different loss functions can be used to calculate the first loss value between each sample probability value and the corresponding reference probability value, and the second loss value between each sample duration and the corresponding reference duration. The first loss value and the second loss value are used to train the initial detection model to obtain a microstructure detection model.

[0102] By using different types of sample physiological signals to train the initial detection model and adjusting the parameters and structure of the initial detection model, it can be adapted to the characteristics and patterns of different microstructures, thereby achieving comprehensive detection of the microstructures of multiple sleep.

[0103] In an embodiment of the present application, a sample physiological signal is obtained, and the sample microstructure in the sample physiological signal is annotated to obtain a reference label array, and the reference label array is Gaussian processed to obtain each reference probability value and each reference duration, and the initial detection model is trained according to each reference probability value and each reference duration to obtain a microstructure detection model. In an embodiment of the present application, the reference label array is Gaussian processed to obtain a continuous sequence label, and through the combination of the initial detection model and the continuous sequence label, the training efficiency of the initial detection model is improved, and the prediction ability of the obtained microstructure detection is improved, so that it can show better stability and robustness when processing complex and changeable detection physiological signals, so as to achieve precise time positioning and high-performance detection of microstructures.

[0104] Fig.10 FIG. 1 is a flow chart of a method for determining a reference probability value and a reference duration in one embodiment. Fig.10 As shown, the embodiment of the present application relates to a possible implementation method of how to perform Gaussian processing on a reference tag array to obtain each reference probability value and each reference duration, including the following steps:

[0105] S1001, determining a reference mean value of the sample microstructure according to a reference start time and a reference end time of the sample microstructure in a reference tag array.

[0106] In the embodiment of the present application, the reference starting time start, the reference ending time end, and the reference mean .

[0107] S1002, determining a reference standard deviation of the sample microstructure according to a reference mean and a reference starting time.

[0108] In the embodiment of the present application, according to the characteristics of Gaussian distribution, assuming that the width of the sample microstructure is 3 times the standard deviation, the reference standard deviation of the sample microstructure is . Among them, 3 times in the embodiment of the present application is just an example.

[0109] S1003, performing Gaussian processing on the reference label array according to the reference mean and the reference standard deviation to obtain each reference probability value.

[0110] In the embodiment of the present application, the reference label array is a binary label array that annotates the sample microstructure. The Gaussian value is calculated for each sampling point in the sample microstructure according to the reference mean and the reference standard deviation. , and obtain the reference probability value corresponding to each sampling moment in the sample microstructure, where x is the sampling moment corresponding to the sample microstructure. That is, the reference probability value of the sampling points other than the sample microstructure in the reference label array is 0, and the representation of each reference probability value is as follows Fig. 9 As shown in the second figure.

[0111] S1004, determining each reference duration according to the reference start time, the reference end time, each reference probability value and the sampling frequency of the sample physiological signal.

[0112] In an embodiment of the present application, a candidate duration can be determined based on the difference between a reference start time and a reference end time, and the ratio of the candidate duration to the sampling frequency can be used as a reference duration corresponding to a first reference probability value in the reference probability value, and the preset duration can be used as a reference duration corresponding to a second reference probability value in the reference probability value other than the first reference probability value; the first reference probability value is a reference probability value corresponding to the center moment of the sample microstructure.

[0113] In a possible implementation, a candidate duration may be determined according to the difference between the reference start time and the reference end time, and the ratio of the candidate duration to the sampling frequency may be used as the reference duration corresponding to the first reference probability value in the reference probability value. A time difference between a first sampling time corresponding to the first reference probability value and a second sampling time corresponding to the second reference probability value is determined, and a lower reference duration is set for a larger time difference.

[0114] In the embodiment of the present application, the reference mean of the sample microstructure is determined according to the reference start time and the reference end time of the sample microstructure in the reference label array, the reference standard deviation of the sample microstructure is determined according to the reference mean and the reference start time, the reference label array is Gaussian processed according to the reference mean and the reference standard deviation to obtain each reference probability value, and each reference duration is determined according to the reference start time, the reference end time, each reference probability value and the sampling frequency of the sample physiological signal. Each reference probability value emphasizes the central part of the sample microstructure, and each reference duration provides information about the length or strength of the sample microstructure. This label enhancement scheme not only simplifies the labeling process of the sample microstructure, but also makes the label information more continuous and smooth, which helps the initial detection model to better learn the characteristics of the sample microstructure.

[0115] In one embodiment, each reference duration is determined based on a reference start time, a reference end time, each reference probability value and a sampling frequency of a sample physiological signal, including: determining a candidate duration based on a difference between a reference start time and a reference end time; using a ratio of the candidate duration to the sampling frequency as a reference duration corresponding to a first reference probability value in the reference probability value; and using a preset duration as a reference duration corresponding to a second reference probability value in the reference probability value other than the first reference probability value.

[0116] The first reference probability value is a reference probability value corresponding to the center moment of the sample microstructure.

[0117] Optionally, the first reference probability value may be the maximum reference probability value among the reference probability values, and the second reference probability value may be a reference probability value other than the maximum reference probability value; the first reference probability value may be the maximum reference probability value and the second largest reference probability value among the reference probability values, and the second reference probability value may be a reference probability value other than the maximum reference probability value and the second largest reference probability value.

[0118] Optionally, the preset duration may be 0, or a value infinitely close to 0.

[0119] In the embodiments of the present application, Fig. 9 As shown in the third figure of , taking the first reference probability value as the maximum reference probability value and the preset duration as 0 as an example, the candidate duration is determined according to the difference between the reference start time and the reference end time, and the candidate duration is the duration obtained in seconds. The ratio of the candidate duration to the sampling frequency is used as the reference duration corresponding to the maximum reference probability value, and the reference durations corresponding to the sampling moments other than the maximum reference probability value are all set to 0.

[0120] In an embodiment of the present application, a candidate duration is determined based on the difference between a reference start time and a reference end time; the ratio of the candidate duration to the sampling frequency is used as the reference duration corresponding to the first reference probability value in the reference probability value, and the preset duration is used as the reference duration corresponding to the second reference probability value in the reference probability value other than the first reference probability value. This simplifies the annotation method of the reference duration, and the obtained label information is more continuous and smooth, which helps the initial detection model to better learn the characteristics of the sample microstructure.

[0121] Fig.11 FIG. 1 is a flow chart of a microstructure detection model in another embodiment. Fig.11 As shown, the embodiment of the present application relates to a possible implementation method of how to train the initial detection model according to each reference probability value and each reference duration to obtain a microstructure detection model, including the following steps:

[0122] S1101, inputting sample physiological signals into an initial detection model to obtain a probability value of each sample and a duration of each sample.

[0123] S1102: Obtain a first loss value between each reference probability value and the corresponding sample probability value based on a first loss function.

[0124] S1103, obtaining a second loss value between each reference duration and the corresponding sample duration based on a second loss function; the first loss function is determined based on the second loss function.

[0125] S1104, training the initial detection model according to the first loss value and the second loss value to obtain a microstructure detection model.

[0126] In an embodiment of the present application, the second loss function may be an L1 loss function, and the first loss function is a loss function after adjusting the L1 loss function. Optionally, the first loss function is a loss function that combines the characteristics of the L1 loss function and the L2 loss function. The first loss function performs well in dealing with regression problems and is used to emphasize the central part of the sample microstructure. The second loss function is used to regress the duration at the center of the sample microstructure. The performance of the initial detection model is jointly optimized by the first loss function and the second loss function during the training process to obtain a microstructure detection model.

[0127] The first loss function can be expressed as ,in, is a threshold parameter that can be reduced to emphasize the center of the sample microstructure, so that smaller errors near the center of the sample microstructure are squared and magnified, thereby increasing the weight of these errors in the total loss, which helps the microstructure detection model pay more attention to and accurately fit the sampling points in the center of the sample microstructure; is the reference probability value, and y is the sample probability value.

[0128] During the training process, these two loss functions will work together on the output layer of the initial detection model, and the parameters of the initial detection model will be updated through the back propagation algorithm. The internal weights of the initial detection model will be adjusted according to the first loss value and the second loss value to minimize the value of the loss function. In this way, the model can gradually learn how to regress the two label vectors more accurately, emphasize the central part of the event, and accurately identify the start and end time of the micro-arousal event.

[0129] In the embodiment of the present application, the sample physiological signal is input into the initial detection model to obtain each sample probability value and each sample duration, and the first loss value between each reference probability value and the corresponding sample probability value is obtained based on the first loss function, and the second loss value between each reference duration and the corresponding sample duration is obtained based on the second loss function. The initial detection model is trained according to the first loss value and the second loss value to obtain a microstructure detection model. The first loss function can generate a larger gradient in the central part of the sample microstructure, thereby guiding the microstructure detection model to pay more attention to the key areas in the sample microstructure, which helps the microstructure detection model to more accurately capture the central features of the sample microstructure and improve the accuracy of regression. Moreover, it lays the foundation for the subsequent start time and duration of the target microstructure based on the microstructure detection model, so as to automatically learn and detect the deep features of the physiological signal through the end-to-end microstructure detection model, realize accurate identification and time positioning of the microstructure, reduce subjective errors, and provide a powerful auxiliary tool for the clinical practice environment to reduce its workload.

[0130] Fig.12 FIG. 1 is a flow chart of a method for detecting and determining a microstructure in another embodiment. Fig.12 As shown, including:

[0131] S1201, obtaining at least one detected physiological signal of the detected subject during sleep;

[0132] S1202, inputting at least one detected physiological signal into a microstructure detection model to obtain a predicted probability value and a predicted duration corresponding to each sampling moment in the detected physiological signal;

[0133] S1203, determining multiple candidate central moments according to each predicted probability value and a preset probability threshold;

[0134] S1204, determining a target microstructure according to a microstructure merging condition and a plurality of candidate center moments; the microstructure merging condition includes that a time difference between two adjacent candidate center moments is less than a preset time difference threshold;

[0135] S1205, taking the sampling time corresponding to the maximum predicted probability value in the target microstructure as the target center time;

[0136] S1206, determining the start time and duration of the target microstructure according to the target center time and each predicted duration.

[0137] In the embodiment of the present application, since the microstructure detection model is obtained by training based on the reference probability value and the reference duration, the reference probability value and the reference duration provide a continuous sequence label for the microstructure detection model, so that the deep features in the physiological signal can be learned and detected based on the microstructure detection model, without relying on rich signal processing knowledge or manual experience, which greatly improves the universality and ease of use of the detection. Moreover, on the basis of accurately detecting whether the target microstructure appears, the starting time and duration of the target microstructure can be obtained, and the precise positioning of the occurrence time and duration of the target microstructure can be achieved, providing more accurate information for clinical diagnosis and treatment.

[0138] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0139] Based on the same inventive concept, the embodiment of the present application also provides a sleep microstructure determination device for implementing the sleep microstructure determination method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more sleep microstructure determination devices provided below can refer to the limitations of the sleep microstructure determination method above, and will not be repeated here.

[0140] In an exemplary embodiment, Fig.13 As shown, a device for determining the microstructure of sleep is provided, comprising: an acquisition module 131 and a determination module 132, wherein:

[0141] An acquisition module 131 is used to acquire at least one detected physiological signal of the detected subject during sleep;

[0142] The determination module 132 is used to determine the starting time and duration of the target microstructure of the detection object based on at least one detected physiological signal and a preset microstructure detection model; the microstructure detection model is trained based on the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and reference duration are obtained by performing Gaussian processing on the sample physiological signal.

[0143] In one embodiment, the determination module 132 is specifically used to input at least one detected physiological signal into the microstructure detection model to obtain the predicted probability value and predicted duration corresponding to each sampling moment in the detected physiological signal; determine the target center moment according to each predicted probability value and a preset microstructure merging condition; determine the start moment and duration of the target microstructure according to the target center moment and each predicted duration.

[0144] In one embodiment, the determination module 132 is specifically used to determine multiple candidate center moments based on each predicted probability value and a preset probability threshold; determine the target microstructure based on the microstructure merging condition and the multiple candidate center moments; the microstructure merging condition includes that the time difference between two adjacent candidate center moments is less than the preset time difference threshold; and use the sampling moment corresponding to the maximum predicted probability value in the target microstructure as the target center moment.

[0145] In one embodiment, the sleep microstructure determination device further comprises:

[0146] A labeling module is used to obtain sample physiological signals and label the sample microstructures in the sample physiological signals to obtain a reference label array;

[0147] A processing module, used for performing Gaussian processing on the reference tag array to obtain each reference probability value and each reference duration;

[0148] The training module is used to train the initial detection model according to each reference probability value and each reference duration to obtain a microstructure detection model.

[0149] In one embodiment, the processing module is specifically used to determine a reference mean of the sample microstructure based on a reference start time and a reference end time of the sample microstructure in a reference tag array; determine a reference standard deviation of the sample microstructure based on the reference mean and the reference start time; perform Gaussian processing on the reference tag array based on the reference mean and the reference standard deviation to obtain each reference probability value; determine each reference duration based on the reference start time, the reference end time, each reference probability value and the sampling frequency of the sample physiological signal.

[0150] In one embodiment, the processing module is specifically used to determine the candidate duration based on the difference between the reference start time and the reference end time; use the ratio of the candidate duration to the sampling frequency as the reference duration corresponding to the first reference probability value in the reference probability value; the first reference probability value is the reference probability value corresponding to the center moment of the sample microstructure; and use the preset duration as the reference duration corresponding to the second reference probability value in the reference probability value except the first reference probability value.

[0151] In one embodiment, a training module is used to input sample physiological signals into an initial detection model to obtain each sample probability value and each sample duration; obtain a first loss value between each reference probability value and the corresponding sample probability value based on a first loss function; obtain a second loss value between each reference duration and the corresponding sample duration based on a second loss function; the first loss function is determined based on the second loss function; and the initial detection model is trained according to the first loss value and the second loss value to obtain a microstructure detection model.

[0152] Each module in the above-mentioned sleep microstructure determination device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0153] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of any of the above method embodiments when executing the computer program.

[0154] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above method embodiments are implemented.

[0155] In one embodiment, a computer program product is provided, including a computer program, which implements the steps of any of the above method embodiments when executed by a processor.

[0156] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0157] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0158] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0159] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for determining the microstructure of sleep, characterized in that: The method comprises: Acquiring at least one detected physiological signal of the detected subject during sleep; Based on at least one of the detected physiological signals and a preset microstructure detection model, the starting time and duration of the target microstructure of the detected object are determined; the microstructure detection model is trained based on the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and the reference duration are obtained by performing Gaussian processing on the sample physiological signal.

2. The method according to claim 1, characterized in that The step of determining the starting time and duration of the target microstructure of the detection object according to at least one of the detected physiological signals and a preset microstructure detection model comprises: Inputting at least one of the detected physiological signals into the microstructure detection model to obtain a predicted probability value and a predicted duration corresponding to each sampling moment in the detected physiological signal; Determining the target center moment according to each of the predicted probability values ​​and a preset microstructure merging condition; The starting time and duration of the target microstructure are determined according to the target center time and each predicted duration.

3. The method according to claim 2, characterized in that Determining the target center moment according to each of the predicted probability values ​​and the preset microstructure merging condition includes: Determining a plurality of candidate central moments according to each of the predicted probability values ​​and a preset probability threshold; Determine the target microstructure according to the microstructure merging condition and the plurality of candidate center moments; the microstructure merging condition includes that the time difference between two adjacent candidate center moments is less than a preset time difference threshold; The sampling moment corresponding to the maximum predicted probability value in the target microstructure is taken as the target center moment.

4. The method according to claim 1, characterized in that: The process of acquiring the microstructure detection model includes: Acquiring a sample physiological signal, and annotating the sample microstructure in the sample physiological signal to obtain a reference label array; Performing Gaussian processing on the reference tag array to obtain each reference probability value and each reference duration; The initial detection model is trained according to each of the reference probability values ​​and each of the reference durations to obtain the microstructure detection model.

5. The method according to claim 4, characterized in that The performing Gaussian processing on the reference tag array to obtain each of the reference probability values ​​and each of the reference durations includes: Determining a reference mean value of the sample microstructure according to a reference start time and a reference end time of the sample microstructure in the reference tag array; Determining a reference standard deviation of the sample microstructure according to the reference mean and the reference starting time; According to the reference mean and the reference standard deviation, Gaussian processing is performed on the reference label array to obtain each reference probability value; Each of the reference durations is determined according to the reference start time, the reference end time, each of the reference probability values, and a sampling frequency of the sample physiological signal.

6. The method according to claim 5, characterized in that The determining each of the reference durations according to the reference start time, the reference end time, each of the reference probability values, and the sampling frequency of the sample physiological signal comprises: Determining a candidate duration according to a difference between the reference start time and the reference end time; The ratio of the candidate duration to the sampling frequency is used as the reference duration corresponding to the first reference probability value in the reference probability value; the first reference probability value is the reference probability value corresponding to the center moment of the sample microstructure; The preset duration is used as a reference duration corresponding to a second reference probability value among the reference probability values ​​except the first reference probability value.

7. The method according to claim 4, characterized in that The step of training the initial detection model according to each of the reference probability values ​​and each of the reference durations to obtain the microstructure detection model comprises: Inputting the sample physiological signal into the initial detection model to obtain a probability value of each sample and a duration of each sample; Acquire a first loss value between each of the reference probability values ​​and the corresponding sample probability value based on a first loss function; Acquire a second loss value between each of the reference durations and the corresponding sample duration based on a second loss function; the first loss function is determined based on the second loss function; The initial detection model is trained according to the first loss value and the second loss value to obtain the microstructure detection model.

8. A device for determining the microstructure of sleep, characterized in that: The device comprises: An acquisition module, used to acquire at least one detected physiological signal of the detected object during sleep; A determination module is used to determine the starting time and duration of the target microstructure of the detection object according to at least one of the detected physiological signals and a preset microstructure detection model; the microstructure detection model is trained according to the reference probability value and reference duration of the sample microstructure corresponding to each sampling moment in the sample physiological signal; the reference probability value and the reference duration are obtained by performing Gaussian processing on the sample physiological signal.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images