EP3946003A1

EP3946003A1 - Method for classifying a polysomnography recording into defined sleep stages

Info

Publication number: EP3946003A1
Application number: EP20711544.5A
Authority: EP
Inventors: Muthuraman MUTHURAMAN; Haralampos GOUVERIS; Philipp Tjarko BOEKSTEGERS
Original assignee: Johannes Gutenberg Universitaet Mainz; Universitaetsmedizin der Johannes Gutenberg-Universitaet Mainz
Current assignee: Johannes Gutenberg Universitaet Mainz; Universitaetsmedizin der Johannes Gutenberg-Universitaet Mainz
Priority date: 2019-03-26
Filing date: 2020-03-10
Publication date: 2022-02-09
Also published as: CA3134205A1; US20220183619A1; WO2020193116A1; JP2022526516A; DE102019107666A1

Abstract

The invention relates to a method for classifying a polysomnography recording into defined sleep stages, the method comprising essentially the following steps: dividing the sleep of a person into different identifiable sleep stages; capturing a plurality of pieces of information regarding bodily functions over a specified time period in the form of data, the data comprising at least one data set of a data type of a first nature, by means of which data set the sleep stages can be identified; splitting the captured data into time-dependent data blocks; manually selecting a limited number of training data blocks from the data blocks and associating them with a sleep stage; evaluating the data set of the first nature of each training data block by means of a data-processing method; creating training objects, each training object comprising the evaluated data sets of the first nature of a training data block and the association of the training data block with a sleep stage; transmitting the training objects to a support vector machine for creating a classification; transmitting at least some of the data blocks that were not selected as training data blocks to the support vector machine.

Description

"Method for the classification of a polysomnography recording into defined sleep stages" Description:

Technical area:

The present invention relates to a method for classifying sleep stages on the basis of a polysomnography recording. In particular, the present invention relates to a

Method for classifying or dividing a cardiorespiratory polysomnography recording into defined sleep stages.

State of the art:

There are a large number of people who suffer from insomnia. The sleep disorders are sometimes very different and can therefore have a variety of different causes.

It is known that polysomnography records can provide clues as to the causes of the sleep disturbance. A polysomnography involves a variety of

Body function data of a patient recorded during sleep. In particular from the course of the brain waves in certain areas of the brain, the heart activity and the breathing intensity and frequency during sleep, causes of

Recognize sleep disorders. Therefore, during a polysomnography, the brain waves in different parts of the brain are recorded using electroencephalography (EEG) and the heart's activity is recorded using electrocardiography (EKG). Beyond that

Respiratory parameters and, if necessary, snoring noises measured with the help of a microphone or the electrical muscle activity measured by means of electromyography (EMG).

Usually, the polysomnography is carried out in a specially equipped sleep laboratory.

It is also known that sleep can be divided into different sleep stages.

Usually the sleep is divided into five different stages, namely the stage N1, the stage N2 and the stage N3 (as part of the non-REM sleep), the REM stage and the waking stage, which corresponds to the epochs or the period during of sleep when the person is awake. The physical activity or the physical function data differ in these stages. This does Noticeable, for example, that the brain waves, which are recorded by means of electroencephalography (EEG), are different in the individual stages. Among other things, both the frequency and the intensity of the brain waves differ. Heart activity, especially heart rate, also changes from one sleep stage to another.

In a healthy person, the stages of sleep run in a more or less regular pattern. In patients with insomnia, this pattern may differ from that of a healthy person. In addition, various bodily functions, depending on the absolute or percentage sleep stage division during sleep, differ from those of a healthy person.

In order to find the cause of sleep disorders, it is therefore helpful to recognize the individual sleep stages of a patient and to assign body functions to specific sleep stages. On the basis of deviations found in the timing of the

Sleep stages and on the basis of deviations in individual body functions in the various sleep stages compared to a healthy person, the causes of a

Recognize or better narrow down sleep disorders.

A polysomnography scan usually takes seven to eight hours, as this is how long a person usually sleeps. Since some sleep disorders can only last a few seconds, the data are displayed at very short intervals, i.e. almost continuously, recorded.

Due to the large amount of data, it goes without saying that the evaluation of such

Polysomnography recording is very time consuming. Alone for the classification of the

Sleep stages of a full night's polysomnography recording require one

Specialist around one to two hours, during which sleep is divided into 30-second units, so-called epochs, with each epoch being assigned to a sleep stage. Furthermore, the quality of the classification depends on the experience of the specialist.

Attempts have already been made to automatically carry out the classification of a polysomnography recording. So far, however, no satisfactory method has been found for automatically classifying a polysomnography recording into different sleep stages. Presentation of the invention:

It is the object of the present invention to provide a method for classifying sleep stages of a polysomnography recording, which can be carried out in a time-saving and cost-effective manner.

According to the invention, the object is achieved by a method for classifying a

A polysomnography record according to claim 1, which comprises the following steps:

First, the sleep of a person is divided into different sleep stages, the sleep stages being identifiable on the basis of at least one data type of the first type. A large number of items of information on body functions are then recorded over a predetermined period of time in the form of data, the data comprising at least one data record of the data type of the first type. The recorded data is divided into time-dependent data blocks. This can be done manually, i.e. by a person, or automatically by a computer or the like. A limited number of training data blocks are then manually selected from the data blocks and assigned sleep stages, the training data blocks being selected such that the data contained in the training block can each be clearly assigned to a defined sleep stage. This selection is preferably made by a trained person or a specialist. Each data record of the first type of each training data block is evaluated by means of a data processing method. Training objects are created from the evaluated data of each training data block, each training object comprising the data sets of the first type of training data block evaluated by means of the data processing method and the assignment of the training data block to a sleep stage. The training objects are then transmitted to a support vector machine for creating a classification in the support vector machine. Thereafter, at least some of the data blocks, preferably all data blocks that are not as

Training data blocks have been selected, transmitted to the support vector machine, and these data blocks are automatically divided into the known sleep stages based on the data of the data type of the first type of each data block.

Under the process step evaluation of the data record of the first type of each

Training data blocks by means of a data processing method are understood in the context of the invention to include processes such as the processing and / or analysis of data records. With the aid of the method described, it is possible to automatically carry out most of the classification of a polysomnography recording in sleep stages.

If the process is also a partially automated process for

Classification of polysomnography recordings is, however, only a very small proportion of the time required by a specialist or a trained person in the

Carrying out the classification. The classification can thus be carried out much more cost-effectively than before.

Surprisingly, it was found that a classification carried out by means of the method described has a comparatively high accuracy if the data set of the first type has data for the following body functions: brain waves, cardiac activity, air flow of breathing, breathing noises, in particular snoring noises, eye movement patterns, electrical muscle activity in the chin area and on the Lower leg (tibialis anterior muscle).

Preferably, in the described method according to claim 1, at least one of the following measuring methods or measuring devices is used to determine the data record of the first type: electroencephalography, electrocardiography, microphone, air flow meter.

The invention is based on the knowledge that the brain waves generated by the

Electroencephalography are measured, allow conclusions to be drawn particularly well about the sleep stage present in a data block. In particular, the C3 / C4 data of the

Electroencephalograms are comparatively easy to determine and, due to their symmetrical arrangement on a person's head, also enable the measurement results to be compared with one another. Therefore, according to a preferred embodiment of the method described, the data record of the first type of data is provided

Having electroencephalography, in particular C3 / C4 data of an electroencephalography.

In preferred embodiments of the method described, the following data processing methods could be used to classify the polysomnography recording well into

Sleep stages can be achieved: cross-frequency coupling method, entropy method,

Power spectrum analysis and heart rate variability determination if the data set of the first type includes data on cardiac function.

Furthermore, the invention is based on the knowledge that the means of a

Electroencephalogram recorded data result from a superposition of several oscillating signals. The electroencephalogram thus records various

Frequency components that interact with each other. Classic analyzes of the Power frequency based, for example, on the (almost) Fourier transformation (FFT) or various transformations of time (e.g. Flilbert transformation)

Modulations of the amplitudes within a defined frequency per time represent. However, you cannot identify the relationships between different frequencies or frequency components. With the help of the cross-frequency coupling method it is possible to synthesize coupling frequencies. Using the cross-frequency coupling method, it is possible for a support vector machine to correctly classify comparable data with a high degree of certainty. Among the various cross-frequency coupling methods, a phase-to-amplitude method has proven particularly useful. According to a preferred development, the step of evaluating the data record of the first type of each training data block by means of a data processing method thus includes a cross-frequency coupling method that includes a phase-to-amplitude method. With this phase-to-amplitude method, the data of an electroencephalogram can be classified particularly accurately in sleep stages.

With regard to an evaluation of a polysomnography recording, it is advantageous that the recorded data are subdivided into a predefined time interval, the time interval in particular being in the range from 15 seconds to 5 minutes and, in particular with regard to electroencephalographic signals, preferably 30 seconds (so-called. 30-second epoch).

In a preferred embodiment of the method, defined for each

Sleep stage two to six, preferably four training data blocks selected.

In order to precisely evaluate the large number of available data blocks in a short period of time, it is advantageous that the support vector machine has an algorithm which uses a non-linear basic kernel function.

In a first embodiment of the method, the data on the body functions can be recorded in a sleep laboratory, with the data on the body functions preferably being recorded on the second night in the sleep laboratory.

Alternatively, the data on the body functions can be recorded in the home environment.

A very high match rate or a very high hit rate in the classification of sleep stages according to the method described could be achieved if the data record of the data type of the first type consists of the data from an electroencephalography, in particular from C3 / C4 data from an electroencephalography, and if the Evaluation of the Data set of the first type of each training data block using a

Cross-frequency coupling method is carried out with a phase-to-amplitude method.

Even with a method in which the data record of the data type of the first type consists of the data from an electroencephalography, in particular C3 / C4 data from an electroencephalography and the evaluation of the data record of the first type of each training data block is carried out by means of a power spectrum analysis, it was possible to achieve high agreement rates in the

Classification of sleep stages can be achieved.

This also applies to methods in which the data record of the data type of the first type consists of at least one of the following data types: data from an electroencephalography, in particular from C3 / C4 data from an electroencephalography, respiratory flow, snoring noises and the evaluation of the data record of the first type of each training data block using an entropy process takes place.

Finally, a method has also proven itself in which the data record of the data type of the first type consists of the data from an electrocardiography and the data processing method comprises a method for establishing the heart rate variability.

Brief description of the drawings:

Preferred embodiments are explained in more detail with reference to the accompanying drawings, in which:

1 shows a schematic representation of the sequence of a method for classifying a polysomnography recording into defined sleep stages on the basis of

Electroencephalography (EEG) data associated with a

Coupling frequency method;

Fig. 2 shows an overview of the usable data of the first type and corresponding

Data processing procedures;

3 shows a schematic representation of the sequence of a method for classifying a polysomnography recording into defined sleep stages on the basis of

Electroencephalography (EEG) data associated with power spectral analysis;

4 shows a schematic illustration of the sequence of a method for classifying a polysomnography recording into defined sleep stages on the basis of

Electroencephalography (EEG) data associated with an entropy procedure; 5 shows a schematic representation of the sequence of a method for classifying a

Polysomnography recording in defined sleep stages using electrocardiography (EKG) data in connection with the determination of heart rate variability.

Best way to carry out the invention:

1 shows a schematic representation of the sequence of a method for classifying a polysomnography recording into defined sleep stages on the basis of electroencephalography (EEG) data in connection with a coupling frequency method.

In the sequence shown in FIG. 1, the first step is to divide a person's sleep into different sleep stages. Usually, sleep is divided into the five known stages, namely stage N1, stage N2, stage N3, REM stage and waking stage.

Each of these known stages can be identified using at least one data type of the first type. In the specific case, provision is made for the individual stages to be automatically identified and assigned using the brain waves recorded by means of electroencephalography

classify.

In the next step, a large amount of information about body functions about the duration of sleep of a person is recorded in the form of a known polysomnography recording in a sleep laboratory. A polysomnography scan usually takes seven to eight hours.

The recorded data is divided into time-dependent data blocks with a duration of 30 seconds. This can be done manually, i.e. by a person, or automatically by a computer or the like.

A trained person or a specialist selects a limited number of training data blocks from the data blocks and assigns these selected training data blocks each to a sleep stage, the person or the specialist selecting the training data blocks in such a way that the data contained in the training block are each clearly assigned to a defined sleep stage can be. Ideally, the person or professional selects the same number of training data blocks for each sleep stage. It has been shown that the selection of four training data blocks per sleep stage is sufficient. It understands However, that in the context of the described method also more or less

Training data blocks can be selected.

The polysomnography recording and thus the data blocks contain, among other things, the brain waves recorded by means of electroencephalography. The brain waves were recorded in different parts of the brain. For the further method for classifying a polysomnography recording into sleep stages, the data that were acquired at positions C3 and C4 on the head of a patient by means of electroencephalography are used (see FIG. 1 in FIG. 1). Positions C3, C4 are the positions that are usually referred to as C3, C4 in electroencephalography.

The data of each training data block obtained at the C3 / C4 positions of an electroencephalography are evaluated by means of a data processing method.

It is known that the frequency and the amplitude of the brain waves change in the individual sleep stages. Each sleep stage is characterized by the presence or the intensity or amplitude of different known frequency groups. The data that the electroencephalogram shows at one position in the brain are thus a superposition of various signals that the brain sends out in the form of brain waves. A simple frequency analysis of the recorded data, for example in the form of an (almost) Fourier transformation, does not provide any frequency sequences that can be clearly assigned to a sleep stage due to the superimposed signals.

For this reason, the data obtained at the C3 / C4 positions of the electroencephalography are processed by means of a cross-frequency coupling method (see Fig. 2 in Fig. 1). It has surprisingly been found here that a cross-frequency coupling method with a phase-to-amplitude method is particularly suitable for converting the data of a

Assign sleep stages to the electroencephalogram.

From the data recorded in the course of electroencephalography, two frequency groups are identified at the C3 / C4 positions, the course and intensity of which can be precisely described using a phase-to-amplitude method. In a phase-to-amplitude method, the dependency between the amplitude of a higher-frequency signal and the phase of a low-frequency signal is shown. The characteristic course of the frequency groups processed by means of the phase-to-amplitude method can be clearly assigned to a sleep stage. The data of a data block obtained with the aid of the cross-frequency coupling method, in particular with the aid of the phase-to-amplitude method, are compared with that of a

A person skilled in the art correlates a certain sleep stage and thus forms a training object.

The training objects obtained from the selected data blocks are transmitted to a support vector machine to create a classification in the support vector machine (see Fig. 3 in Fig. 1).

An algorithm contained in the Support Vector Machine marks each data element as a point in an n-dimensional space, where n represents the number of features. The algorithm must calculate the best mean value from these between different separating straight lines in order to determine the best common separating plane for all points, i.e. in this case a line with the maximum possible distance to all data points. The classification is carried out by determining the so-called optimal hyperplane. As a next step, the algorithm looks for the hyperplane on which the data points are located with the smallest distance to the optimal hyperplane, the so-called support vectors. This distance is called margin. The optimal separating hyperplane now maximizes the margin in order to obtain clearly separated classification groups. The Support Vector Machine thus divides the training data blocks into the specified sleep stages.

The remaining part of the data blocks that were not selected as training data blocks are then transmitted to the support vector machine and these data blocks are automatically divided into the known sleep stages based on the C3 / C4 data

Electroencephalography.

In a test phase, with the method described, a correct allocation of the data blocks in sleep stages and thus a hit rate of more than 93% could be achieved (see Fig. 4 in Fig. 1).

A particularly precise classification of the data blocks not selected as training data blocks is achieved in that a non-linear basic kernel function is used in the algorithm of the support vector machine.

2 shows an overview of the data of the first type which can be used in the context of the present method and which are suitable for performing a classification, and corresponding data processing methods for evaluating the data of the first type. Instead of the electroencephalography data used in connection with FIG. 1, the data on the respiratory flow, the snoring and the data from an electrocardiography are also suitable as data of the first type for carrying out the method described if suitable data processing methods are used.

As an alternative to the cross-frequency method used in connection with FIG. 1, the data of the first type can also be evaluated by means of a power spectral analysis or an entropy method.

In power spectrum analysis, the frequency-related power of a signal is specified in a frequency band. The power spectral analysis is suitable for example for data from an electroencephalography (see Fig. 3). The multi-tap process is particularly suitable for this.

As already mentioned above, classical frequency transformations such as the Fourier transformation cannot, or only inadequately, be applied to data from electroencephalography. In a Fourier transformation, for example, there is no temporal reference to the respective frequencies. The multi-taper process is generated by multiplying the

Frequency domain such a time-frequency representation.

The entropy method is a non-linear dynamic analysis. The main principle of the entropy method is the quantification of information about a signal as well as the probability of the occurrence of certain patterns within a finite number of patterns and within a time series of the signal. The more information that is conveyed within a signal, the higher the entropy of the signal. There are different types of entropy method, which differ in connection with the classification of

The sample entropy method, which is a modification of the approximate entropy method, is particularly suitable for sleep stages.

In the approximate entropy method, time series are examined for similar epochs, with more frequent and more similar epochs leading to lower values of the approximate entropy. Thus, lower values of the approximate entropy mean a high level of regularity of the signal and conversely, high values of the approximate entropy mean an irregular signal.

The approximate entropy method, however, depends on the data record length. To avoid the results being dependent on the data record length, a Entropy method used, in which self-matching sequences are not counted and which works regardless of the data record length. This entropy method is the aforementioned sample entropy method, which is a modification of the approximate entropy method. The sample entropy method also has the advantage that it can be carried out more quickly.

The sample entropy method can, as shown in FIG. 4, particularly advantageously in

In connection with electroencephalography data. For this procedure, data on the respiratory flow and snoring can also be used.

In the case in which the data are taken from an electrocardiography, a data processing method that determines the heart rate variability from the recorded data is also suitable (see FIG. 5).

The distance between two heartbeats is usually defined as the time between the start of two contractions of the heart chambers. This beginning of the ventricular contraction is shown in the electrocardiogram as an R-wave, the distance between two R-waves being called the RR interval. As a rule, the RR intervals are not of the same length, but are subject to fluctuations. The quantification of these fluctuations is known as heart rate variability (HRV).

The heart rate variability of a selected training data block determined from the data of an electrocardiography, together with the assignment to the sleep stage, already serves as a training object that can be transmitted to the support vector machine. The data blocks not selected as training data blocks can be classified into sleep stages using the support vector machine based on the heart rate variability.

Even if the best hit rate or best assignment of the data blocks to sleep stages with the C3 / C4 data using a

Cross-frequency method were achieved, so was the methods that use the

Use entropy methods or power spectral analysis, a satisfactory one

Hit rate achieved. Except for the procedure in which the snoring noises were used as data of the first type, the hit rates were generally over 50%, sometimes well over 50%.

The heart rate variability is also suitable for the classification of sleep stages using the method described. Here, too, the hit rates are over 50%.

Claims

1. Procedure for classifying a polysomnography recording into defined ones

Sleep stages comprising the following steps:

• Division of a person's sleep into different sleep stages, the sleep stages being identifiable on the basis of at least one data type of the first type;

• Gathering a lot of information about body functions through one

predetermined time period in the form of data, the data comprising at least one data record of the data type of the first type;

• Subdivision of the recorded data into time-dependent data blocks;

• manually selecting a limited number of training data blocks from the data blocks and assigning them to a sleep stage, the training data blocks being selected so that the data contained in the training block can each be clearly assigned to a defined sleep stage;

• Evaluation of the data record of the first type of each training data block by means of a data processing method;

• Creation of training objects, each training object comprising the data sets of the first type of training data block evaluated by means of the data processing method and the assignment of the training data block to a sleep stage;

• Transmission of the training objects to a support vector machine for creating a classification in the support vector machine;

• Transmission of at least some of the data blocks that are not as

Training data blocks were selected to the support vector machine and automatic division of these data blocks into the known sleep stages based on the data of the data type of the first type of data blocks.

2. The method according to claim 1, characterized in that the data set of the first type has data from the following body functions: brain waves, cardiac activity, air flow of breathing, breathing sounds, in particular snoring noises, eye movement patterns, electrical muscle activity in the chin area and on the lower leg.

3. The method according to claim 1 or 2, characterized in that at least one of the following measuring methods or measuring devices for determining the data record of the first type is used: electroencephalography, electrocardiography, microphone,

Air flow meter.

4. The method according to any one of the preceding claims, characterized in that the data set of the first type has data from electroencephalography, in particular C3 / C4 data from electroencephalography.

5. The method according to any one of the preceding claims, characterized in that the data preparation method comprises at least one of the following methods: cross-frequency coupling method, entropy method, power spectral analysis and determination of heart rate variability if the data set of the first type includes data on cardiac function.

6. The method according to any one of the preceding claims, characterized in that the cross-frequency coupling method comprises a phase-to-amplitude method.

7. The method according to any one of the preceding claims, characterized in that the recorded data are divided into a predefined time interval, wherein

in particular the time interval is in the range from 15 seconds to 5 minutes, preferably 30 seconds.

8. The method according to any one of the preceding claims, characterized in that two to six, preferably four, training data blocks are selected for each defined sleep stage.

9. The method according to any one of the preceding claims, characterized in that the support vector machine has an algorithm which uses a non-linear basic kernel function.

10. The method according to any one of the preceding claims, characterized in that the data on the body functions are recorded in a sleep laboratory, the data on the body functions preferably being recorded on the second night in the sleep laboratory.

1 1. The method according to any one of claims 1 to 9, characterized in that the data on the body functions are recorded in a domestic environment.

12. The method according to any one of the preceding steps, characterized in that the data record of the data type of the first type from the data of an electroencephalography, consists in particular of C3 / C4 data from an electroencephalography and that the evaluation of the data record of the first type of each training data block is carried out by means of a cross-frequency coupling method with a phase-to-amplitude method.

13. The method according to any one of claims 1 to 1 1, characterized in that the

Dataset of the data type of the first type consists of the data of an electroencephalography, in particular of C3 / C4 data of an electroencephalography and that the evaluation of the dataset of the first type of each training data block takes place by means of a power spectral analysis.

14. The method according to any one of claims 1 to 1 1, characterized in that the

Dataset of the data type of the first type consists of at least one of the following data types: data from an electroencephalography in particular from C3 / C4 data from an electroencephalography, respiratory flow, snoring noises and that the evaluation of the dataset of the first type of each training data block is carried out using an entropy method.

15. The method according to any one of claims 1 to 1 1, characterized in that the

Data record of the data type of the first type consists of the data of an electrocardiography and the data processing method comprises a method to determine the heart rate variability.