RU2751137C1 - Method for determining sleep phase in long-term eeg recording - Google Patents

Abstract

FIELD: computing technology.

SUBSTANCE: proposed is a computer-implemented method for determining the sleep phase in a long-term EEG recording, containing a preparatory stage whereat: at least one EEG signal from the database preprocessed, at least one channel is filtered, at least one epoch is isolated from at least one EEG signal and channel-by-channel normalisation is performed; the first neural network is trained, at the input whereof at least one epoch of at least one EEG signal is supplied, wherein at least two channels common to all subjects are used, wherein features from at least one epoch of at least one EEG signal are automatically extracted by applying a one-dimensional convolutional layer with a large filter size and applying at least one convolutional block providing a greater abstraction level; the obtained features are combined into input sequences, wherein a sequence consists of the current epoch and at least one previous epoch of the first neural network; each epoch is divided into equal parts by the length of the segment, and for each segment, in each channel the power of the signal in the alpha range is calculated, the obtained features are added to the input sequence; the second neural network is trained, to the input whereof the input sequence obtained at the previous stage is supplied to determine the temporal dynamics of the correlation between the epochs. The method also comprises a working stage, whereat: the first and second neural networks trained at the previous stage are initialised, at least one epoch of at least one EEG signal is supplied to the input of the first neural network, the input sequence obtained from the first neural network and containing features calculated based on the power of the signal in the alpha range is then supplied to the input of the second neural network; classification of at least one epoch of at least one EEG signal is performed to determine the probability of attribution of at least one epoch to at least one sleep phase; the results of the classification are output.

EFFECT: invention provides improved accuracy of classification of an EEG signal according to the sleep phases.

4 cl, 3 dwg

Description

FIELD OF TECHNOLOGY

This technical solution relates to the fields of medicine and computing, in particular, to a method for automatically determining the phase of sleep in a long-term recording of an electroencephalogram (hereinafter - EEG) using machine learning.

LEVEL OF TECHNOLOGY

From the prior art, a source of information KR 101395197 B1, 05/15/2014 disclosing a method for determining the sleep state of a patient for a certain period of time is known, containing the stages at which EEG data for a patient is obtained for a specified period of time, and the EEG data contains at least one first frequency band having a lower power in the frequency spectrum, a second frequency band in the frequency spectrum, a receiving step showing a dynamic range. Brain wave data are segmented into one or more epochs. The frequency powers of one or more epochs are weighted over time to normalize the brainwave data to generate normalized data, the weighting step including: the step of weighting the power of the frequency spectrum, increasing the dynamic range of the power within the first frequency range by comparing the dynamic range of the power with the second frequency range; and a step of generating one or more frequency-weighted epochs. A second weighting step of low power weighting to re-normalize the normalized data according to a second measure to generate twice normalized data. The patient's sleep state is classified based on one or more frequency-weighted epochs.

A source of information RU 2672684 C2, 11/19/2018 is known from the prior art, which discloses a method for detecting sleep stages of a patient using a sleep stage detection system. This provides a stimulus to the patient using the stimulus source. Using the sensors, output signals are generated that transmit information related to one or more physiological parameters of the patient related to EEG, EMG, EOG and patient movement. Analyzed by a response module based on one or more physiological stimulus response parameters. The sleep stage module generates an estimate of the patient's sleep stage based on one or more physiological parameters, and the likelihood of an accuracy of the sleep stage estimate is determined. It is determined using the comparison module whether the probability of the estimation accuracy exceeds a predetermined threshold. The stimulus control module is provided, in response to the probability of the estimation accuracy not exceeding a predetermined threshold, a reevaluation stimulus for the patient. Providing a reevaluation incentive to the patient is repeated with increasing intensity until the probability of the accuracy of the estimate exceeds a predetermined threshold. The response module is analyzed by a response module to respond to a reevaluation stimulus based on one or more physiological parameters. The sleep stage module reevaluates the sleep stage score and / or the likelihood of the accuracy of the score based on the analyzed response to the reevaluation stimulus. EFFECT: increased accuracy of estimation of sleep stages determination and increased probability of accurate estimation of sleep stages determination.

The proposed solution differs from the solutions known from the prior art in that when training a neural network, signs of the EEG signal power in the alpha range, which are associated with sleep processes, are taken into account, so alpha waves decrease with open eyes, drowsiness and sleep, occipital alpha waves in periods of closed eyes are the strongest EEG signals from the brain. These parameters will help to accurately correlate the EEG signal to sleep phases.

SUMMARY OF THE INVENTION

The technical problem to be solved by the claimed technical solution is the creation of a computer-implemented method for automatically determining the sleep phase in a long-term EEG recording, which is characterized in an independent claim. Additional embodiments of the present invention are presented in the dependent claims.

The technical result consists in increasing the accuracy of the classification of the EEG signal by sleep phases.

The claimed result is achieved by implementing a computer-implemented method for automatically determining the phase of sleep in a long-term EEG recording, which contains:

The preparatory stage, at which:

preprocessing at least one EEG signal from the database, filtering at least one channel, extracting at least one epoch from at least one EEG signal and performing channel-by-channel normalization;

train the first neural network, at the input of which at least one epoch of at least one EEG signal is supplied, in which at least two channels are used, which are common for all subjects, while

perform automatic feature extraction from at least one epoch of at least one EEG signal by applying a one-dimensional convolutional layer with a large filter size and applying at least one convolutional block that provides a higher level of abstraction;

the obtained features are combined into input sequences, and the sequence consists of the current epoch and at least one previous epoch of the first neural network;

each epoch is divided into equal parts along the length of the segment and for each segment, in each channel, the signal power in the alpha range is calculated, the obtained features are added to the input sequence;

the second neural network is trained, at the input of which the input sequence obtained at the previous stage is fed, to determine the temporal dynamics of the relationship between eras;

The working stage at which:

the first and second neural networks trained at the previous stage are launched, at least one epoch of at least one EEG signal is fed to the input of the first neural network, then the input sequence obtained from the first neural network and containing the features calculated based on signal strength in the alpha range;

carry out the classification of at least one epoch of at least one EEG signal to determine the probability of belonging at least one epoch to at least one sleep phase;

displays the classification results.

In a private embodiment of the proposed method, sleep phases are divided into:

● W (wakefulness) - wakefulness;

● N1 - drowsiness;

● N2 - light sleep;

● N3 - deep / slow sleep;

● REM (rapid eye movement) - sleep phase, accompanied by rapid eye movements and dreams.

In another particular embodiment of the claimed method, the channels that are common to all subjects are: A1, A2, C3, C4, Cz, F3, F4, F7, F8, Fp1, Fp2, Fz, O1, O2, P3, P4, Pz, T3, T4, T5, T6.

In another particular embodiment of the claimed method, if at least one EEG signal is longer than 30 seconds, then the signal is divided into 30-second epochs.

DESCRIPTION OF DRAWINGS

The implementation of the invention will be described in the following in accordance with the accompanying drawings, which are presented to clarify the essence of the invention and in no way limit the scope of the invention. The following drawings are attached to the application:

1 illustrates a general view of the claimed computer-implemented method for determining the phase of sleep in a long-term EEG recording.

2 illustrates an example of a convolutional recurrent neural network architecture for classifying sleep into phases.

FIG. 3 illustrates an example of a general arrangement of a computing device.

DETAILED DESCRIPTION OF THE INVENTION

In the following detailed description of an implementation of the invention, numerous implementation details are set forth in order to provide a thorough understanding of the present invention. However, it will be obvious to those skilled in the art how the present invention can be used, with or without these implementation details. In other instances, well-known techniques, procedures, and components have not been described in detail so as not to obscure the details of the present invention.

In addition, it will be clear from the above description that the invention is not limited to the above implementation. Numerous possible modifications, alterations, variations and substitutions, while retaining the spirit and form of the present invention, will be apparent to those skilled in the art.

Sleep quality can be associated with human health problems related to cardiovascular disease, diabetes, and psychiatric disorders also affect sleep. Dividing sleep into phases can help diagnose diseases that can be detected by differences in sleep stage patterns from a healthy person.

The present invention is directed to the implementation of a computer-implemented method for automatically determining the phase of sleep in a long-term EEG recording.

Figure 1 illustrates the proposed computer-implemented method for automatically determining the phase of sleep in a long-term EEG recording (100), performed on a computing device (300), containing two stages of work: preparatory (110) (learning stage) and working (120) (stage testing).

At the preparatory stage (110), at least one EEG signal is received from the database, which has already been marked by a specialist in sleep phases.

The current American Academy of Sleep Medicine (AASM) standard distinguishes 5 main stages of sleep:

W (wakefulness) - wakefulness;

N1 - drowsiness;

N2 - light sleep;

N3 - deep / slow sleep;

REM (rapid eye movement) - sleep phase, accompanied by rapid eye movements and dreams.

Normal sleep consists of 3-5 characteristic cycles. One cycle lasts approximately 90 minutes and consists of N2, N3 and REM phases. Phase N1 occurs mainly during falling asleep or waking up during sleep.

At step 111, the received signal from the database is preprocessing.

If the EEG signal is longer than 30 seconds, then such a signal is divided into several 30-second epochs.

Channels are selected from at least one epoch of at least one EEG signal. To obtain an accurate result, 21 channels were selected, which are common for all subjects in the sample, namely: A1, A2, C3, C4, Cz, F3, F4, F7, F8, Fp1, Fp2, Fz, O1, O2, P3 , P4, Pz, T3, T4, T5, T6. However, the proposed solution does not limit the use of more channels.

Further, each channel is filtered with a two-sided Butterworth filter with cutoff frequencies of 0.3 and 35.0 Hertz in order to remove interference that is not related to brain activity: interference from the network, muscle activity, and others.

Since at the working stage of the algorithm, data can come from potentially different sources of EEG recording with different sampling rates, while the algorithm must be tuned to one specific frequency, all channels were resampled to a frequency of 100 Hertz.

The last step of signal preprocessing is channel-by-channel normalization. For each channel, the average value and standard deviation of the entire signal for a given record is calculated, after which the channel is normalized - that is, the average value is subtracted and the result is divided by the standard deviation.

At step 112, the first convolutional neural network is trained. An example of a neural network architecture is shown in Fig. 2 At least one preprocessed EEG signal is supplied to the input of a convolutional neural network, namely at least one 30-second epoch of a preprocessed EEG signal, which includes 21 channels that are common to all subjects.

In this case, the automatic extraction of features from at least one epoch of the EEG signal is carried out, where a one-dimensional convolutional layer with a large filter size is applied to the input signal (50). This filter is used to get into the window low-frequency oscillations, at which there are significant signs to highlight the N3 stage - deep sleep, where there are characteristic oscillations in the delta range of 0.5-4 Hz.

Further, the obtained features are fed to at least one convolutional block to isolate features with a high level of abstraction. Three convolutional blocks were used to obtain the most accurate result, however, more blocks can be used. Each block includes:

1) One-dimensional convolutional layer with a filter of size 5, for feature extraction (steps and number of filters vary);

2) The max-pooling layer, which consists in performing a passage by a window of a fixed size over the image and aggregating the values of the intensities of the window pixels to reduce the computational complexity, regularization, and translational invariance.

Each convolutional layer is followed by a sequence of BatchNormalization and Dropout layers to speed up convergence and prevent overfitting.

Next comes the classification part of fully connected layers, which ends with a softmax layer that produces a normalized probability distribution of the epoch belonging to at least one sleep phase.

The convolutional neural network does not take into account the temporal dynamics of the relationship between epochs, which is very important in determining the sleep phases, therefore, it is proposed to additionally use a recurrent neural network. It consists of two Long Short-Term Memory (LSTM) layers using a recurrent kind of Dropout, and one softmax classification layer at the end.

In order to generate the input for the recurrent model, all training data (at least one epoch of at least one EEG signal) is passed through the convolutional network up to the first fully connected layer. Further, the output of the first fully connected layer is taken as a new feature description of the original object. Since the convolutional part of the model has already been trained, these features will characterize the original signal well.

Then, at step 113, these features are combined into input sequences of 6 epoch length. Thus, the input of the recurrent network is a sequence of the current and 5 previous epochs in the feature description of the convolutional network.

To eliminate the blurring of the boundaries between the phases W, N1, N2 and improve the accuracy of the classification of sleep phases, at step 114, the power of the EEG signal in the alpha range is calculated:

1) at least one epoch is divided into several segments equal in length;

2) for each segment in each channel, the signal power in the alpha range is considered. To do this, perform a discrete Fourier transform, then calculate the DFT sampling frequency and select frequencies in the range from 8 Hz to 13 Hz. Calculate the sum of the squares of the spectral coefficients that correspond to the selected frequencies;

3) the obtained features are added to the features at the stage of forming the input sequence for the recurrent network;

For example, an epoch is divided into 3 segments of equal length, 21 channels are allocated in one epoch. For each segment, in each channel, the signal power in the alpha range is counted, 63 signs of the signal power in the alpha range are obtained from one epoch of the EEG signal. The obtained features are added to the features that form the input sequence.

At step 115, a recurrent neural network is trained, to which the input sequence obtained at the previous step is fed, to determine the temporal dynamics of the relationship between epochs. At this stage, the preparatory stage ends.

At the working step 120, the first and second neural networks trained in the previous step are launched (step 121), at least one epoch of at least one EEG signal is fed to the input of the first neural network, then the input sequence obtained from the first neural network and containing features calculated based on the signal strength in the alpha range.

At least one epoch of at least one EEG signal is classified in step 122 to determine the likelihood that at least one epoch belongs to at least one sleep phase. The obtained classification results are displayed on the working screen, at step 123.

FIG. 3, a general diagram of a computing device (300) will be presented below, which provides data processing necessary for the implementation of the claimed solution.

In general, the device (300) contains such components as: one or more processors (301), at least one memory (302), data storage means (303), input / output interfaces (304), I / O means ( 305), networking tools (306).

The processor (301) of the device performs the basic computational operations necessary for the operation of the device (300) or the functionality of one or more of its components. The processor (301) executes the necessary computer readable instructions contained in the main memory (302).

Memory (302), as a rule, is made in the form of RAM and contains the necessary program logic that provides the required functionality.

The data storage medium (303) can be performed in the form of HDD, SSD disks, raid array, network storage, flash memory, optical information storage devices (CD, DVD, MD, Blue-Ray disks), etc. The means (303) allows performing long-term storage of various types of information, for example, the aforementioned files with user data sets, a database containing records of time intervals measured for each user, user identifiers, etc.

Interfaces (304) are standard means for connecting and working with the server side, for example, USB, RS232, RJ45, LPT, COM, HDMI, PS / 2, Lightning, FireWire, etc.

The choice of interfaces (304) depends on the specific implementation of the device (300), which can be a personal computer, mainframe, server cluster, thin client, smartphone, laptop, etc.

As means of I / O data (305) in any embodiment of the system that implements the described method, a keyboard should be used. The hardware design of the keyboard can be any known: it can be either a built-in keyboard used on a laptop or netbook, or a stand-alone device connected to a desktop computer, server or other computer device. In this case, the connection can be either wired, in which the connecting cable of the keyboard is connected to the PS / 2 or USB port located on the system unit of the desktop computer, or wireless, in which the keyboard exchanges data via a wireless communication channel, for example, a radio channel, with base station, which, in turn, is directly connected to the system unit, for example, to one of the USB ports. In addition to the keyboard, I / O data can also include: joystick, display (touch screen), projector, touchpad, mouse, trackball, light pen, speakers, microphone, etc.

Networking means (306) are selected from a device that provides network reception and transmission of data, for example, Ethernet card, WLAN / Wi-Fi module, Bluetooth module, BLE module, NFC module, IrDa, RFID module, GSM modem, etc. The means (305) provide the organization of data exchange via a wired or wireless data transmission channel, for example, WAN, PAN, LAN, Intranet, Internet, WLAN, WMAN or GSM.

The components of the device (300) are interconnected via a common data bus (310).

In the present application materials, the preferred disclosure of the implementation of the claimed technical solution has been presented, which should not be used as limiting other, particular embodiments of its implementation, which do not go beyond the scope of the claimed scope of legal protection and are obvious to specialists in the relevant field of technology.

Claims (21)

1. A computer-implemented method for automatically determining the sleep phase in a long-term EEG recording, containing a processor and a memory storing instructions executed by the processor, the method comprising two stages: preparatory stage, at which: perform preprocessing of at least one EEG signal from the database, filtering at least one channel, separating at least one epoch from at least one EEG signal and performing channel-by-channel normalization; train the first neural network, the input of which is fed at least one epoch of at least one EEG signal, in which at least two channels are used, which are common to all subjects, while performing automatic feature extraction from at least one epoch of at least one EEG signal by applying a one-dimensional convolutional layer with a large filter size and applying at least one convolutional block that provides a higher level of abstraction; the obtained features are combined into input sequences, the sequence consisting of the current epoch and at least one previous epoch of the first neural network; each epoch is divided into equal parts along the length of the segment and for each segment, in each channel, the signal power in the alpha range is calculated, the obtained features are added to the input sequence; the second neural network is trained, at the input of which the input sequence obtained at the previous stage is fed, to determine the temporal dynamics of the relationship between epochs; working stage at which: the first and second neural networks trained at the previous stage are launched, at least one epoch of at least one EEG signal is fed to the input of the first neural network, then the input sequence obtained from the first neural network and containing features calculated on based on the signal strength in the alpha range; carry out the classification of at least one epoch of at least one EEG signal to determine the probability of belonging at least one epoch to at least one sleep phase; the results of the classification of determining the probability of belonging at least one epoch to at least one sleep phase are displayed on the working screen. 2. The method according to claim 1, characterized in that the sleep phases are divided into: - W (wakefulness) - wakefulness; - N1 - drowsiness; - N2 - light sleep; - N3 - deep / slow sleep; - REM (rapid eye movement) - sleep phase, accompanied by rapid eye movements and dreams. 3. The method according to claim 1, characterized in that the channels common to all subjects are: A1, A2, C3, C4, Cz, F3, F4, F7, F8, Fp1, Fp2, Fz, O1, O2, P3, P4, Pz, T3, T4, T5, T6. 4. The method according to claim 1, characterized in that if at least one EEG signal is longer than 30 seconds, then the signal is divided into 30-second epochs.

Images