KR102691350B1 - Single electroencephalogram-based sleep stage classification method and sleep stage classification device - Google Patents

Abstract

A single electroencephalogram-based sleep stage classification method performed by a sleep stage classification device includes the steps of measuring only an electroencephalogram (EEG) signal for a certain period of time to obtain an original signal; Preprocessing the original signal to generate a plurality of extracted signals; Generating integrated data by integrating the original signal and the plurality of extracted signals; calculating a posterior probability based on the integrated data; It includes a single electroencephalogram-based sleep stage, including the step of classifying the sleep stage from the calculated posterior probability. This not only makes it possible to automatically classify sleep stages without the help of highly trained sleep experts, but also facilitates the measurement of biological signals using only a single electroencephalogram signal (EEG).

Description

Single electroencephalogram-based sleep stage classification method and sleep stage classification device}

The present invention relates to a sleep stage classification method and a sleep stage classification device based on a single electroencephalogram. It relates to a sleep stage classification method and a sleep stage classification device based on a single electroencephalogram that classifies sleep stages using only a single electroencephalogram signal without the help of a medical professional.

Today, modern people live under stress due to school work, household chores, human relationships, and excessive work. Because of this stress, modern people are unable to sleep well and suffer from insufficient sleep time. In addition, since the spread of computers and smartphones has increased, the quality of sleep has dropped drastically, and the number of patients visiting hospitals complaining of sleep disorders is rapidly increasing.

In particular, sleep is very important because it is directly related to the quality of life. If you do not get adequate sleep, not only will your blood pressure increase or your immune system weaken, but serious problems such as cognitive impairment may also occur.

Accordingly, many researchers classified sleep stages to better understand sleep, which made it possible to diagnose various sleep disorders such as sleep apnea, insomnia, and narcolepsy.

However, in order to distinguish the stages of sleep, a polysomnography is performed to measure various biological signals such as electroencephalogram, electrocardiogram, electrocardiogram, electromyogram, and pulse oximetry while sleeping, and then a trained sleep expert manually distinguishes the stages of sleep. This not only requires a lot of time and resources, but also causes problems with poor accessibility.

Korean Patent Publication No. 2014-0058441

The present invention was created to solve the above problems, and the purpose of the present invention is to automatically classify sleep stages as well as to facilitate the measurement of biological signals using only a single electroencephalogram signal (EEG). , to provide a single electroencephalogram-based sleep stage classification method and sleep stage classification device.

A single electroencephalogram-based sleep stage classification method performed by a sleep stage classification device according to an embodiment of the present invention to achieve the above object includes the steps of measuring only electroencephalogram signals (EEG, electroencephalogram) for a certain period of time to obtain the original signal. ; Preprocessing the original signal to generate a plurality of extracted signals; Generating integrated data by integrating the original signal and the plurality of extracted signals; calculating a posterior probability based on the integrated data; and classifying sleep stages from the calculated posterior probability.

And in the step of calculating the posterior probability, a spatiotemporal convolution deep neural network (Spectral-Temporal Convolution Neural Network), a self-attention deep neural network (Self-Attention or Multi-Head Attention), and a bidirectional-Long Short Term neural network (Bidirectional-Long Short Term) The posterior probability can be calculated through a sleep classification model that includes memory.

In addition, the sleep classification model can learn the correlation between electroencephalogram feature information extracted from the integrated data through the spatio-temporal convolutional deep neural network through the self-attention deep neural network.

In addition, the sleep classification model can learn the correlation data between the electroencephalogram feature information based on the bidirectional long-term short-term memory recurrent neural network.

Meanwhile, the plurality of extracted signals may be generated by extracting signals for each specific frequency band from the original signal based on frequency filtering.

It may further include extracting posterior probability bundles while moving the posterior probabilities calculated in the calculating step at regular time intervals, and calculating an average value of the posterior probability bundles.

Additionally, in the classification step, the sleep stage may be classified based on the section having the largest value among the average values calculated in the average value calculation step.

Meanwhile, a sleep stage classification device according to an embodiment of the present invention for achieving the above object includes an acquisition unit that measures only electroencephalogram signals (EEG, electroencephalogram) for a certain period of time to obtain the original signal; a pre-processing unit that preprocesses the original signal to generate a plurality of extracted signals, and integrates the original signal and the plurality of extracted signals to generate integrated data; a calculation unit that calculates a posterior probability based on the integrated data; and a classification unit that classifies sleep stages from the calculated posterior probability.

And the calculation unit includes a spatiotemporal convolution deep neural network (Spectral-Temporal Convolution Neural Network), self-attention deep neural network (Self-Attention or Multi-Head Attention), and bidirectional-long short term memory (Bidirectional-Long Short Term Memory). The posterior probability can be calculated through a sleep classification model.

Additionally, the sleep classification model may learn the correlation between electroencephalogram feature information extracted from the integrated data through the spatio-temporal convolutional deep neural network based on the self-attention deep neural network.

According to one aspect of the present invention described above, by providing a sleep stage classification method and a sleep stage classification device based on a single electroencephalogram, it is possible to automatically classify sleep stages without the help of a highly trained sleep expert, as well as a single electroencephalogram signal (EEG). ) can be used to easily measure biosignals.

1 is a schematic diagram of classifying sleep stages in a sleep stage classification device according to an embodiment of the present invention;
Figure 2 is a diagram for explaining the configuration of a sleep stage classification device according to an embodiment of the present invention;
Figure 3 is a diagram for explaining the pre-processing process in the sleep stage classification device according to an embodiment of the present invention;
4 is a diagram for explaining a learning unit according to an embodiment of the present invention;
5 is a flowchart illustrating a sleep stage classification method according to an embodiment of the present invention;
Figure 6 is a diagram for explaining the detailed configuration of a sleep classification model according to an embodiment of the present invention;
7 is a schematic diagram of the structure of a spatiotemporal convolutional deep neural network according to an embodiment of the present invention;
8 is a schematic diagram of a prediction algorithm based on posterior probability according to an embodiment of the present invention;
Figure 9 is a diagram for comparing before and after applying the prediction algorithm based on the posterior probability of Figure 8, and
Figure 10 is a diagram for comparing sleep stage prediction results according to the sleep stage classification method of the present invention and sleep stages judged by a sleep expert.

The detailed description of the present invention described below refers to the accompanying drawings, which show by way of example specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the present invention are different from one another but are not necessarily mutually exclusive. For example, specific shapes, structures and characteristics described herein may be implemented in one embodiment without departing from the spirit and scope of the invention. Additionally, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the detailed description that follows is not intended to be taken in a limiting sense, and the scope of the invention is limited only by the appended claims, together with all equivalents to what those claims assert, if properly described. Similar reference numbers in the drawings refer to identical or similar functions across various aspects.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

Figure 1 is a schematic diagram of classifying sleep stages in the sleep stage classification device 100 according to an embodiment of the present invention, and Figure 2 illustrates the configuration of the sleep stage classification device 100 according to an embodiment of the present invention. FIG. 3 is a diagram for explaining the pre-processing process in the sleep stage classification device 100 according to an embodiment of the present invention, and FIG. 4 is a diagram for explaining the learning unit 140 according to an embodiment of the present invention. FIG. 5 is a diagram for explaining the detailed configuration of a sleep classification model according to an embodiment of the present invention, and FIG. 6 is a schematic diagram of the structure of a spatiotemporal convolutional deep neural network according to an embodiment of the present invention.

Unlike the prior art, the sleep stage classification device 100 of the present invention uses only a single electroencephalogram (EEG) signal and uses an advanced deep neural network algorithm to automatically classify sleep stages without the help of a sleep expert. do.

As shown in FIG. 1, the sleep stage classification device 100 first acquires only the electroencephalogram signals measured from the patient for a certain period of time, preprocesses the obtained electroencephalogram signals, and finally uses a deep learning algorithm to classify the preprocessed data into sleep stages. can be classified. Through this, objective standards can be presented according to sleep stage classification, judgment errors between judges can be reduced, and analysis of polysomnography results can be automated, providing the effect of minimizing the time required for polysomnography. It becomes possible.

For this purpose, the sleep stage classification device 100 may be provided including an acquisition unit 110, a storage unit 120, a control unit 130, and a learning unit 140.

The acquisition unit 110 may be provided to acquire original signals collected by measuring only electroencephalogram signals (EEG) from the patient for a certain period of time. This acquisition unit 110 is connected wired or wirelessly to an electroencephalogram sensor attached to the patient's body and can acquire the original signal directly or indirectly in real time.

In addition, the original signal acquired by the acquisition unit 110 may be an electroencephalogram signal stored in the storage unit 120 rather than in real time, and the original signal acquired in real time by the acquisition unit 110 is stored in the storage unit 120. It can be transmitted as much as possible, or it can be transmitted to the control unit 130 for sleep stage classification.

Meanwhile, the storage unit 120 may store the electroencephalogram signal acquired by the acquisition unit 110 as the original signal, or store the electroencephalogram signal recorded in the polysomnography database by a previously performed polysomnography test as the original signal.

And in the storage unit 120, a program for performing a single electroencephalogram-based sleep stage classification method is recorded. Additionally, the data processed by the control unit 130 is stored temporarily or permanently, and may include, but is not limited to, volatile storage media or non-volatile storage media.

Additionally, the storage unit 120 may store data accumulated while performing a sleep stage classification method. For example, learning data used in a sleep stage classification model, a sleep stage classification model, and embedding vectors may be stored.

The control unit 130 is provided to control the entire process of providing a single electroencephalogram-based sleep stage classification method, and can be controlled by software (application) for performing the single electroencephalogram-based sleep stage classification method.

And the control unit 130 controls the acquisition unit 110 and the storage unit 120, and uses the sleep stage classification model generated through the learning unit 140 based on the single electroencephalogram signal acquired by the acquisition unit 110. Sleep stages can be classified. As shown, the control unit 130 may include a preprocessing unit 131, a calculation unit 133, and a classification unit 135.

The pre-processing unit 131 is provided to generate a plurality of extracted signals by pre-processing the original signal received from the acquisition unit 110 and to generate integrated data by integrating the original signal and the plurality of extracted signals.

, 4~8Hz), 알파(α, 8~12Hz), 스핀들(spindle, 12~15Hz), 베타(, 15~30Hz)의 주파수 대역별 신호를 추출할 수 있다. Specifically, since the frequency bands activated are different depending on the stage of sleep, a process of extracting signals in a specific frequency band from the original signal is necessary. Accordingly, the preprocessor 131 uses a Band-Pass filter as shown in FIG. 3 to filter delta (δ, 0.5 to 4 Hz) and theta (

, 4~8Hz), alpha (α, 8~12Hz), spindle (12~15Hz), beta ( , 15~30Hz) signals can be extracted for each frequency band.

Then, the preprocessor 131 extracts signals for each frequency band from the original signal to generate a plurality of extracted signals, then integrates the five extracted signals and the original signal to produce 20 pieces of data at 30-second intervals as integrated data. can be created. Thereafter, the pre-processing unit 131 may transmit the integrated data generated in this way to the calculation unit 133.

Meanwhile, the calculation unit 133 is provided to receive the integrated data generated in the pre-processing unit 131 and input the received integrated data into a sleep classification model to calculate a posterior probability.

At this time, the sleep classification model includes Spectral-Temporal Convolution Neural Network, Self-Attention or Multi-Head Attention, and Bidirectional-Long Short Term Memory. can do. The sleep classification model is an artificial intelligence model that uses time series data in epoch units and allows sleep stages to be classified by referring to the AASM (American Academy of Sleep Medicine) standard or the R&K (Rechtschaffen and Kales) standard.

Accordingly, the calculation unit 133 can take the integrated data as input and go through a total of three deep neural networks above to finally calculate the posterior probability. This sleep classification model can be generated by the learning unit 140, and the specific learning process for generating the sleep classification model will be described later in the learning unit 140.

Additionally, the calculation unit 133 according to this embodiment can more accurately distinguish sleep stages using a learned sleep classification model.

To this end, the calculation unit 133 inputs integrated data, which is 20 signals at 30-second intervals, into the learned sleep classification model and extracts a bundle of posterior probabilities by moving the calculated posterior probabilities at regular time intervals, that is, at 30-second intervals. there is. Additionally, the calculation unit 133 can calculate the average value of the posterior probability bundles extracted in this way.

Meanwhile, the classification unit 135 can classify the place with the largest value among the average values of the posterior probability bundles calculated in this way as a sleep stage. Specifically, it can be classified as shown in Table 1 below.

Event Frequency/Duration Related Stage Alpha rhythm 8-13 Hz W Low amplitude, mixed frequency activity 4-7 Hz N1 Vertex shape waves <0.5s N1 K complex 1.6-4 Hz >= 0.5 s N2 Sleep spindle 11-16 Hz >= 0.5 s N2, N3 Slow wave activity 0.5-2 Hz N3 Sawtooth waves 2-6 Hz R.E.M.

Meanwhile, the learning unit 140 may be provided to generate a sleep stage classification model for sleep stage classification, and the artificial intelligence-based sleep stage classification model receives epoch-level time series data with processed characteristics as input and performs deep learning. Sleep stages can be classified by analysis using technology. In order to implement a sleep stage classification model for optimal performance, hyperparameters, which are optimized variables, exist and can be defined for each layer.

As described above, the sleep stage classification model of this embodiment is a spatiotemporal convolution deep neural network (Spectral-Temporal Convolution Neural Network), a self-attention deep neural network (Self-Attention or Multi-Head Attention), and a bidirectional-long memory recurrent neural network. It is designed in the form of a deep learning model, including Short Term Memory.

For this purpose, the learning unit 140 may be provided including a space-time convolution module 141, a self-attention module 143, and a bidirectional LSTM module 145.

As shown in FIG. 5, the spatio-temporal convolution module 141 is a total of 20 electroencephalogram signals measured in 30 second units (epochs) in the acquisition unit 110, which are generated by preprocessing the preprocessor 131. Data will be input.

Then, the space-time convolution module 141 can automatically extract the features of the EEG signal by applying a space-time convolution deep neural network, focus on important parts of the feature vector, extract the final feature vector, and generate an embedding vector.

More specifically, the spatiotemporal convolutional deep neural network extracts neurophysiological features of the EEG signal using several parameters from the integrated data, as shown in Figure 6, and at this time, adjusts the kernel size to process can be performed. And as shown in Figure 6, batch normalization and ELU (Exponential linear unit) activation functions for the two convolutional layers can be applied to the feature maps generated from the two convolutional layers, and bias is applied to reduce the number of parameters. It can be removed.

More specifically, in a spatiotemporal convolutional deep neural network, two convolutional layers can extract temporal features from EEG signals using a kernel size of (1, k). The kernel size of the first convolutional layer is set to (1, 50), which is half of the 100Hz sampling rate, and for the second convolutional layer, the kernel size is set to (1, 100) to capture frequency information above 0.5Hz in the EEG signal. It can be set to . The two convolutional layers can be padded with 25 and 50 to maintain the dimensionality of the input.

The spatiotemporal convolutional deep neural network may include a depthwise convolution layer and a pointwise convolution layer, as shown in FIG. 6. At this time, the depth-specific convolution layer learns spatial information by frequency using a kernel size of (1, 6) and captures information between frequencies at each sleep stage. Meanwhile, the point-wise convolution layer optimally mixes each feature map using a kernel size of (1, 1). This spatiotemporal convolutional deep neural network does not fully apply pooling to all previous feature maps, thus reducing the number of trainable parameters. Two stacked pooling layers (fully connected layers) with 1000 output neuron units and 5 output neurons can be used to classify sleep stages using the final feature maps of the convolution process.

At this time, the feature map is flattened into a single vector and input into two stacked pooling layers, which allows scores for the five sleep stages to be calculated based on the five output vectors. Then the next SoftMax layer is applied.

More specific layers and hyperparameters of the spatiotemporal convolutional deep neural network may be as shown in Table 2 below.

The spatiotemporal convolution module (141) can perform transfer learning using a pre-trained spatiotemporal convolutional deep neural network to generate an embedding vector including the characteristics of brain wave signals related to sleep. Transfer learning is used in various machine learning, which can improve performance and shorten training time by transferring knowledge of related source tasks to a target task. As shown in Fig. 6, the final feature map extracted from the pre-trained spatiotemporal convolutional deep neural network can be linearly projected to a pooling layer including 2,000 output neurons to generate an embedding vector.

The embedding vector generated in this way is input to the self-attention module 143 as shown in FIG. 5 and can be set to a non-static state so that it can be fine-tuned to fit the sleep classification model.

Meanwhile, the self-attention module 143 can learn the correlation between 20 electroencephalogram feature information extracted from the spatio-temporal convolution module 141 by applying a self-attention deep neural network (Self-Attention or Multi-Head Attention). Through this, it is possible to check whether the integrated data has similar characteristics or not.

Self-attentive deep neural networks can capture correlations and dependencies between multiple epochs with different timestamps.

Twenty consecutive epochs from the space-time convolution module 141 are converted into embedding vectors and then input into the self-attention deep neural network. Since the self-attention mechanism does not reflect sequence information, position information is added to the embedding matrix. Position encoding can be applied, and this position encoding can be implemented by the following equation 1.

[Equation 1]

)

Here, i is the index of the dimension, pos is 20, which means the position index of the fixed-length partial sequence for learning dependencies between epochs, and d _model can be defined as 2,000, which is the same size as the embedding vector.

And the self-attentive deep neural network may include a query, key, and value with the same size as the input vector, and the attention weight can be calculated through three steps.

First, to derive similarity between keys and queries, the dot product of queries with all keys is calculated and divided by the square of the vector dimension. Second, the attention weights of different values can be obtained using SoftMax activation, and finally, the attention value can be calculated by adding the weights of all values generated in the second step.

The attention weight is calculated by packing the query, key, and value into Q, K, and V, respectively, and the output matrix can be calculated through Equation 2 below.

[Equation 2]

where Q, K and V are matrices formed by the input embedding vectors, is the dimension of the vector, and T may mean the transpose matrix.

And the self-attention mechanism can extract features from various perspectives by performing h linear projection on Q, K, and V, where h can be the number of heads, and in this case, Equation 3 below can be used.

[Equation 3]

From above may mean a learnable parameter.

Afterwards, the self-attention module 143 can connect the output of another head and then perform linear projection again. The self-attention module 143 according to this embodiment projects an embedding vector with a dimension of 2,000 (i.e., 10 heads) using a fully connected layer with 100 output neurons, and the final output vector is deep It can be added to the input vector through a skip connection to improve the convergence of learning, overcome performance degradation issues, and improve information flow.

And the self-attentive deep neural network is equally applied to the position-wise feed-forward network consisting of two pooling layers with the GELU activation function. The position-wise feed-forward network is implemented as shown in Equation 4 below. It can be.

[Equation 4]

here class are respectively and It is a learnable parameter of the form and f is 100, and may mean bias terms.

By applying the above self-attention deep neural network, it is possible to capture correlations between different signals in the self-attention module 143, and the self-attention module 143 converts the output vector output through the self-attention deep neural network into a two-way LSTM module (145). ) can be entered.

Meanwhile, the bidirectional LSTM module 145 can be prepared to learn time-series characteristics of the electroencephalogram that change with time by applying a bidirectional-long short term memory (Bidirectional-Long Short Term Memory). In the case of long-term memory recurrent neural networks, they are mainly used when learning time series data sequentially, and prevent past information from being lost as learning progresses. By learning this long-term memory recurrent neural network in both directions, it becomes possible to extract and learn various information from the signal.

More specifically, although it is possible to capture correlations between different signals in the self-attention module 143, there are limitations in reflecting changes in sleep stages over time, and to address these issues, deep self-attention The output vector extracted through the neural network can be input to the bidirectional LSTM module 145 that considers the correlation between different epochs and changes in sleep stages.

The bidirectional LSTM module 145 according to this embodiment applies a bidirectional long-term short-term memory recurrent neural network including two LSTMs in the forward and backward directions to increase the amount of information considered to learn changes in sleep stages over time. It is possible to solve the long-term dependency problem of RNN.

The Long Short Term Memory (LSTM) of the bidirectional long short term memory recurrent neural network applied to this bidirectional LSTM module 145 can be implemented according to Equation 5 below.

[Equation 5]

From above is a sigmoid function, and w and b represent the weight matrix and bias, respectively. I, f, c, and o refer to the input gate, forget gate, cell vector, and output gate, and may have the same size as the hidden vector h.

The bidirectional long-short-term memory recurrent neural network according to this embodiment may include 200 hidden units and 2,000 input neurons per layer. In order to classify the five sleep stages, the output vectors of the self-attention deep neural network and the bidirectional long-term memory recurrent neural network can be connected (the dimension of the output vector is 2,400) and then sequentially input to the pooling layer and SoftMax layer.

Through the above process, the learning unit 140 can complete learning using three deep neural networks to create a sleep classification model. And through the sleep classification model created in this way, the control unit 130 can classify the final sleep stage.

In addition, in order to successfully learn a sleep stage classification model, the class imbalance problem must be solved. Generally, the duration of each sleep stage varies, so there is a difference in the number of instances of each class. For example, the 'N2' stage accounts for about 45-55% of total sleep, while the 'N1' stage only accounts for about 2-5%.

Most machine learning algorithms learn by assuming that the misclassification cost is the same for all classes, but cost-sensitive learning minimizes the total cost by appropriately assigning different misclassification costs to each class. It's a training method. This method has the advantage of being able to learn all classes equally without increasing training time or computational cost, and can alleviate class imbalance problems compared to other data sampling methods.

Accordingly, in the case of the Sleep-EDF (Expanded) dataset, which is the learning target of this embodiment, there may be a problem of data imbalance in which the number of classes in each sleep stage is uneven.

Accordingly, when learning is performed with the corresponding dataset, learning is performed with more interest in multiple classes, so in order to perform learning evenly for all classes, the learning unit 140 in this embodiment uses focus loss. The misclassification cost can be minimized by using functions. The focus loss function can be implemented based on Equation 6 below.

[Equation 6]

At this time may refer to the probability associated with a specific sleep stage occurring based on the SoftMax activation function, is a weighting factor, a focusing parameter. is a balance modulation factor, which was introduced to solve the class imbalance problem.

In addition, since the performance of the 'N1' class is relatively poor compared to other classes, the learning unit 140 in this embodiment frequently trains samples with the 'N1' class by applying weighted random sampling to the 'N2' class. ' We sought to improve the performance of the class.

At this time, the learning unit 140 may group multiple outputs into three groups to apply weighted random sampling. Group 1 is when all labels are in the 'w' stage, Group 2 is when 'N1' is not included in the label and does not belong to group 1, and Group 3 is when 'N1' is included in the label and does not belong to group 1. This may not be the case.

The sampling probability was set at a ratio of 1:5:5 for each group, and this procedure effectively rescaled the training set so that the majority of classes focused on sleep stages other than 'W' (particularly N1).

Figure 7 is a schematic diagram of a prediction algorithm based on posterior probability according to an embodiment of the present invention.

As described above, the calculation unit 133 can extract the posterior probabilities output through the sleep classification model generated after completing learning in the learning unit 140 at intervals of 30 seconds. This sleep stage classification device 100 can classify sleep stages using a prediction algorithm based on posterior probability. As shown in the figure, the input sequence consists of 20 epochs and is offset by 1 epoch. .

The calculation unit 133 can add up the posterior probabilities for each epoch and calculate the average value for each class, and the time index t of the final posterior probability can be calculated according to Equation 7 below.

[Equation 7]

here is the nth prediction result, is the ith data, means the posterior probability.

Meanwhile, Figure 8 is a flowchart for explaining a sleep stage classification method according to an embodiment of the present invention. The single electroencephalogram-based sleep stage classification method according to this embodiment can be performed by the sleep stage classification device 100 described above. there is.

The sleep stage classification method according to this embodiment includes the step of acquiring the original signal (S100), the step of generating a plurality of extracted signals (S200), the step of generating integrated data (S300), and the step of calculating the posterior probability. (400) and a step of classifying sleep stages (S500).

The step of acquiring the original signal (S100) may be a step of acquiring the electroencephalogram signal (EEG) as the original signal by measuring only the electroencephalogram signal (EEG) for a certain period of time. The step (S100) of acquiring these original signals may be collected directly by the EEG sensor provided in the sleep classification device 100, or may be transmitted through wired or wireless communication with a separate EEG sensor.

, 4~8Hz), 알파(α, 8~12Hz), 스핀들(spindle, 12~15Hz), 베타(, 15~30Hz)의 주파수 대역별 신호를 추출하는 단계일 수 있다. Meanwhile, the step of generating a plurality of extracted signals (S200) is a step of generating a plurality of extracted signals by pre-processing the original signal obtained from the acquisition unit 110 in the pre-processing unit 131. The original signal is generated using a Band-Pass filter. From the signal, delta (δ, 0.5~4Hz), theta (

, 4~8Hz), alpha (α, 8~12Hz), spindle (12~15Hz), beta ( , 15 to 30 Hz), this may be a step of extracting signals for each frequency band.

After a plurality of extracted signals are generated in this way, step 300 of generating integrated data can be performed. In the step 300 of generating integrated data, the preprocessor 131 may generate integrated data by integrating the original signal and the five extracted signals.

Then, step 400 of calculating the posterior probability based on the integrated data can be performed. The step of calculating the posterior probability (S400) is a step of calculating the posterior probability by inputting the generated integrated data into a sleep classification model. At this time, the sleep classification model is a spatiotemporal convolution deep neural network, It may include deep attention neural networks (Self-Attention or Multi-Head Attention) and Bidirectional-Long Short Term Memory. The sleep classification model is an artificial intelligence model that uses time series data in epoch units and allows sleep stages to be classified by referring to the AASM (American Academy of Sleep Medicine) standard or the R&K (Rechtschaffen and Kales) standard.

Accordingly, in the step of calculating the posterior probability (S400), the posterior probability can be finally calculated through a sleep classification model including a total of three stages of deep neural network generated by the learning unit 140 described above. At this time, the correlation between electroencephalogram feature information extracted from integrated data through a spatio-temporal convolutional deep neural network can be extracted through a self-attention deep neural network. And based on the bidirectional long-term memory recurrent neural network, it is possible to extract correlation data between EEG feature information and time-series features of the EEG that change with time in epoch units.

In addition, the single electroencephalogram-based sleep stage classification method (S500) according to this embodiment may further include a step (not shown) of calculating the average value of the posterior probability bundles after the step of calculating the posterior probability (S400).

In the step of calculating the average value of these posterior probability bundles (not shown), a sleep classification model including a three-stage deep neural network detailed on 20 signals at 30-second intervals generated through the step of generating integrated data (S300). The posterior probabilities obtained by inputting can be extracted by moving at 30 second intervals. Then, the average values of the posterior probability bundles extracted in this way are obtained.

Then, a step (S500) of classifying the sleep stage can be performed. In the step of classifying sleep stages (S500), the sleep stages can be classified based on the position with the largest value among the average values calculated in the step of calculating the average value of the posterior probability bundles (S500).

Meanwhile, Figure 9 is a diagram for comparing before and after applying the prediction algorithm based on the posterior probability of this embodiment, and Figure 10 is a diagram for comparing the sleep stage prediction results according to the sleep stage classification method of the present invention with the sleep stage judged by a sleep expert. As a drawing for this purpose, the following will describe the results of an experiment evaluating the performance of the sleep stage classification model according to this embodiment.

In this experiment, we used the Sleep-EDF (Expanded) database (version 2018), and the data set consists of two subsets: sleep cassette subset and sleep telemetry subset. The sleep cassette (SC) subset consisted of polysomnography (PSG) from 153 subjects aged 25–101 years, measured to study the relationship between sleep and age.

The sleep telemetry subset consisted of polysomnograms from 44 subjects aged 18–79 years who were measured to study the relationship between sleep and the prescription sleep medication temazepam. And the sleep cassette subset was adopted as the target domain in this experiment.

Polysomnography consisted of two bipolar EEG channels (Fpz-Cz and Pz-OZ), a horizontal EOG channel and a submental EMG channel. EEG and EOG signals were measured at 100 Hz, and EMG signals were sampled at 1 Hz. In each 30-second epoch of recording, the sleep expert selects one of eight classes according to R&K standards: 'W', 'N1', 'N2', 'N3', 'N4', 'REM', 'M', '?' One is labeled, and 'M' and '?' mean movement and unknown, respectively. According to the AASM standard, 'N3' and 'N4' were merged into 'N3', 'M' and '?' were excluded, and among all physiological signals, the Fpz-Cz EEG signal was used only for learning according to this embodiment.

In order to compare the performance of the sleep stage classification model according to this embodiment, two existing machine learning methods and eight latest methods using deep learning were implemented.

Two existing methods are support vector machine and random forest, which learn power spectral density from EEG signals, and published code to implement the state-of-the-art method (Phan, et al. 2019; Mousavi, et al. 2019; Supratak and Guo 2021).

The sleep stage classification model of the present invention was implemented in PyTorch v1.3.1, a library designed to enable easy and fast research on machine learning models, and training and evaluation were performed using E5-2680 2.40GHz CPU, 256GB RAM, and NVIDIA TITAN XP 12GH GPU. It was performed on a computer equipped with .

For learning, the spatiotemporal convolutional deep neural network was first pre-trained using a minibatch of size 400, and after the spatiotemporal convolutional deep neural network learning was completed, the remaining parameters except the two pretrained pooling layers were initialized. It was trained using mini-batches of size 50, and an early stopping criterion based on the validation loss for 10 epochs was used to stop learning when no improvement was detected during the learning process.

To optimize the parameters of the sleep stage classification model, the AdamW (Loshchilov and Hutter 2017) optimization program was adopted. The parameters learning rate, beta1, beta2, and weight decay parameters were set to 1e-3, 0.9, 0.999, and 1e-2, respectively, in the AdamW optimizer.

Additionally, to evaluate the sleep stage classification model, 10-ford cross-validation was performed on the Sleep-EDFX data set, and a total of 153 polysomnography subjects were divided into a training set for model learning and a test set for evaluation, and the training set Four topics were subdivided for verification, and this process was repeated 10 times to test all topics, and more specifically, as shown in Table 3 below.

, 4~8Hz), 알파파(α, 8~12Hz), 스핀들(spindle, 12~15Hz), 베타파(, 15~30Hz), 감마파(γ, 30~50Hz)로 분류된다. 또한 알파파를 기준으로 해서 8Hz 미만을 서파(Slow Wave), 13Hz이상을 속파(Fast Wave)라고 구분한다. 뇌파에 의해 연구되어 온 자발뇌파는 일반적 생리현상에서 감각 등 뇌 활동으로 나타나며, 유발뇌파는 뇌 활동 상태를 알아보기 위해 인위적으로 뇌 활동을 유도하여 관찰할 수 있다.In general, brain waves are divided into delta waves (δ, 0.5~4Hz) and theta waves (depending on the frequency band).

, 4~8Hz), alpha wave (α, 8~12Hz), spindle (12~15Hz), beta wave ( , 15~30Hz), and gamma waves (γ, 30~50Hz). Additionally, based on alpha waves, those below 8Hz are classified as slow waves, and those above 13Hz are classified as fast waves. Spontaneous EEG, which has been studied through EEG, appears as brain activity such as sensation in general physiological phenomena, and evoked EEG can be observed by artificially inducing brain activity to determine the state of brain activity.

In order to evaluate the sleep stage classification performance of the sleep stage classification model according to this embodiment, overall accuracy (ACC), macro-F1 score (MF1), and Cohen's Kappa coefficient (k) were calculated. Three metrics were used. ACC and MF1 are commonly used evaluation indicators, and in particular, MF1 is a useful indicator when data classes are unbalanced, and this evaluation indicator can be calculated through Equation 8 below.

[Equation 8]

here and N are the number of true positives and total epochs, respectively, and K may mean the F1 score of the ith sleep stage and the total number of all sleep stages, respectively.

And Cohen's Kappa coefficient (k) is a method of measuring the degree of agreement between measurement category values between two observers and can be calculated through Equation 9 below.

[Equation 9]

At this time represents accuracy, represents the change rate, and this formula indicates how well the classification result matches the actual value.

In addition, as shown in Table 4 below, as a result of comparing the performance of the sleep stage classification model according to this embodiment proposed based on various metrics with previous studies, the performance of the sleep stage classification model according to this embodiment was different. It can be seen that the overall performance is the best compared to other algorithms.

In addition, as shown in Figure 9, the result of classifying sleep stages using the sleep stage classification model of the present invention together with posterior probability calculation (right), and the sleep stage classification using only the sleep stage classification model without calculating posterior probability. Comparing the results (left), it can be seen that the sleep stage classification method according to this embodiment shows excellent prediction performance for all classes.

And, as shown in Figure 10, when comparing the results of classifying sleep stages through the sleep stage classification method according to this embodiment (below) and the sleep stage classification results manually determined by a sleep expert (above), the results of sleep stage classification are very You can see the similarity.

In other words, the single electroencephalogram-based sleep stage classification method according to this embodiment not only can automatically classify sleep stages without the help of a highly trained sleep expert, but also improves time and resource efficiency by using only a single electroencephalogram signal (EEG). can be raised.

The components according to the present invention are components defined by functional division rather than physical division, and can be defined by the functions each performs. Each component may be implemented as hardware or program code and processing units that perform each function, and the functions of two or more components may be included and implemented in one component. Therefore, the names given to the components in the following embodiments are not intended to physically distinguish each component, but are given to suggest the representative function performed by each component, and the names of the components are used to describe the present invention. It should be noted that the technical idea is not limited.

The single electroencephalogram-based sleep stage classification method of the present invention can be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc., singly or in combination.

The program instructions recorded on the computer-readable recording medium may be those specifically designed and configured for the present invention, as well as those known and usable by those skilled in the computer software field.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specifically configured to store and perform program instructions, such as ROM, RAM, flash memory, etc.

Examples of program instructions include not only machine language code such as that created by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform processing according to the invention and vice versa.

Although various embodiments of the present invention have been shown and described above, the present invention is not limited to the specific embodiments described above, and may be used in the technical field to which the invention pertains without departing from the gist of the invention as claimed in the claims. Of course, various modifications can be made by those skilled in the art, and these modifications should not be understood individually from the technical idea or perspective of the present invention.

100: Sleep stage classification device 110: Acquisition unit
120: storage unit 130: control unit
131: preprocessing unit 133: calculation unit
135: classification unit 140: learning unit
141: Space-time convolution module 143: Self-attention module
145: Bidirectional LSTM module

Claims (10)

In a single electroencephalogram-based sleep stage classification method performed by a sleep stage classification device,
Obtaining the original signal by measuring only the electroencephalogram (EEG) signal for a certain period of time;
Preprocessing the original signal to generate a plurality of extracted signals;
Generating integrated data by integrating the original signal and the plurality of extracted signals;
calculating a posterior probability based on the integrated data;
Including the step of classifying sleep stages from the calculated posterior probability,
The posterior probability is,
The probability that the integrated data is output by inputting it into a pre-trained sleep classification model is the probability that the integrated data corresponds to each class corresponding to the sleep stage,
The sleep classification model is,
A single electroencephalogram-based sleep stage classification method that uses a focus loss function that uses a focusing parameter, which is a balance modulation factor, to solve the problem of imbalance in the number of classes corresponding to the sleep stage. According to paragraph 1,
The sleep classification model is,
A single neural network characterized by a spatio-temporal convolution deep neural network (Spectral-Temporal Convolution Neural Network), a self-attention deep neural network (Self-Attention or Multi-Head Attention), and a Bidirectional-Long Short Term Memory (Bidirectional-Long Short Term Memory). Sleep stage classification method based on electroencephalogram. According to paragraph 2,
The sleep classification model is,
A single electroencephalogram-based sleep stage classification method, characterized in that the correlation between electroencephalogram feature information extracted from the integrated data through the spatio-temporal convolution deep neural network is learned through the self-attention deep neural network. According to paragraph 3,
The sleep classification model is,
A single electroencephalogram-based sleep stage classification method, characterized in that the correlation data between the electroencephalogram feature information is learned based on the bidirectional long-term short-term memory recurrent neural network for time-series characteristics of the electroencephalogram that change with time in epoch units. According to paragraph 1,
The plurality of extracted signals are,
A single electroencephalogram-based sleep stage classification method, characterized in that it is generated by extracting signals for each specific frequency band from the original signal based on frequency filtering. According to paragraph 1,
A sleep stage classification method based on a single electroencephalogram, further comprising extracting posterior probability bundles while moving the posterior probabilities calculated in the calculating step at regular time intervals, and calculating an average value of the posterior probability bundles. According to clause 6,
In the classification step,
A single electroencephalogram-based sleep stage classification method, characterized in that the sleep stage is classified based on the section with the largest value among the average values calculated in the step of calculating the average value. An acquisition unit that acquires the original signal by measuring only the electroencephalogram (EEG) signal for a certain period of time;
a pre-processing unit that preprocesses the original signal to generate a plurality of extracted signals, and integrates the original signal and the plurality of extracted signals to generate integrated data;
a calculation unit that calculates a posterior probability based on the integrated data; and
It includes a classification unit that classifies sleep stages from the calculated posterior probability,
The posterior probability is,
The probability that the integrated data is output by inputting it into a pre-trained sleep classification model is the probability that the integrated data corresponds to each class corresponding to the sleep stage,
The sleep classification model is,
A sleep stage classification device that uses a focus loss function that uses a focusing parameter, which is a balance modulation factor, to solve the problem of imbalance in the number of classes corresponding to the sleep stages. According to clause 8,
The sleep classification model is,
Sleep characterized by including a spectral-temporal convolution neural network, self-attention or multi-head attention, and bidirectional-long short term memory. Step classification device. According to clause 9,
The sleep classification model is,
A sleep stage classification device, characterized in that the correlation between electroencephalogram feature information extracted from the integrated data through the spatio-temporal convolution deep neural network is learned based on the self-attention deep neural network.