Method of analysing electroencephalogram
SUBSTANCE: invention relates to medicine and can be used for automatic analysis of human electroencephalograms (EEG). The method comprises steps for recording an EEG; spectral analysis by continuous wavelet transform is carried out in two steps, at the first of which there is primary analysis of the recorded digital signal by the "common" wavelet, the basis function of which is a wavelet similar on its characteristics with the elementary area of the EEG; a matrix of recommendations on selection of the class of physiologically significant features of the recorded EEG is formed; physiologically significant parts of the analysed signal are selected; at the second step there is analysis of the physiologically significant parts of the analysed signal by the synthesised wavelets, the synthesis criterion of which is minimisation of the sum of squared deviations of the wavelet from the reference signal; a matrix of analysis results is formed; a video image of the moment of the patient is obtained, which is compared with the matrix of analysis results and, if the results match the corresponding information on moment, the patient has a physiologically significant feature and its type is determined, after which at least one trained artificial neural network is used to form a clinical conclusion matrix, based on which a clinical conclusion is formed in text form, which can be output for display and/or transmission to a remote reception unit; whether the patient has a disease is determined from the clinical conclusion.
EFFECT: automation of the process of analysing an EEG, high accuracy of analysis.
2 cl, 11 dwg
The invention relates to the field of medicine and is designed for automated research electroencephalograms (EEG) of a person, for example, with elements of epilepsy and using the wavelet spectrograms obtained by the method of continuous wavelet transform.
Known methods of registration of EEG with the surface of the scalp or directly from the brain with subsequent processing of EEG various methods.
In one of these ways (see Lresolv. Clinical electroencephalography, with elements of epilepsy, ed.: Medpress-inform, 2004, str-181) provide registration of EEG, record it in memory and then converting into digital form, processed by the method of Fourier transform. When this exercise selection physiologically significant areas EEG to localize the source of pathological activity using Fourier transform and implement the display of this information.
The display of this information (selection of physiologically significant areas EEG) carried out synchronously with the image of the patient obtained from a video camera.
This technical solution is implemented in a computerized analyzer EEG alpha-DEB-T-16-01 system digital video surveillance.
However, used in this way the Fourier transform is:
- prevents restores the signal with sufficient accuracy, what, in particular, is related to the Gibbs effect, which is that at the point of discontinuity convergence of Fourier series is not uniform and is of a special nature in terms of the appearance of pulsations near the point of discontinuity of the first kind, the maximum of which is left and right is about 9% of the amplitude-frequency characteristics (AFC). Therefore, the development of real systems that use this method with direct and inverse Fourier transforms, we have to limit the number of harmonics, and that, even if the exact (ideal) system is operating, it is not possible to retrieve the original signal without distortion;
- is ineffective in the study of signals with characteristics that change rapidly with time. To resolve this problem, you must limit the implementation of Fourier transform on some intervals of finite length, for example, using windowed transform. However, this leads to further complications, one of which is the choice of the window size. Increase time resolution leads to the reduction of resolution in the frequency domain, which is mainly localized basis function of the Fourier transform. Since the EEG signal is difficult to predict its change in the next moment of time is almost impossible, and after vetelino, using the Fourier transform does not allow for automated EEG analysis with a sufficient degree of accuracy.
There is a method of EEG analysis of humans and animals (see RF patent for the invention №2332160, M CL AU 5/04, publ. 27.08.2008,), which carry out registration of EEG and subsequent spectral analysis of the recorded EEG using the continuous wavelet transform by the formula:
where ψ(t) is the wavelet function, thea- determines the size of the wavelet, b - sets a shift along the time axis, so thata, b ∈ R,aa≠0.
The wavelet transform is a time base range and allows you to get more localized in time compared to the Fourier-transformation of the energy information.
After it on the basis of the wavelet coefficients is calculated salagrama on a discrete set of values of the argumentaiand bj, i=0,..., Na-1; j=0,..., Nb-1:
Based on scalogram calculated salagrama:
where N is the number of points that is used for averaging.
Next on salagrama produce physiologically relevant frequency bands on the basis of distances between neighboring local minima on the curve scalogram. The building with alahram. at short intervals allows you to track the temporal dynamics of the research process. Next, determine the changes wavelet power density in time, the change in frequency bands in time and values of specific wavelet power density over time, which reflects the dynamics of change in activity of the various generators of the EEG for short periods of time.
For registration of EEG uses electrodes installed on the surface of the scalp or removing signals directly from the brain.
This analog adopted for the prototype.
Used in the prototype wavelets constructed on the basis of mathematical functions that are not associated with EEG features, so their use in the prototype are not adapted to the analysis of EEG signals, which does not allow for the required accuracy of the analysis.
In addition, the output must be further subjected to manual analysis to obtain a clinical assessment of the different EEG features (artifacts, pathological activity and others). I.e. not achieved full automation of the process of EEG analysis.
Achievable technical result of the invention is to provide automation of the entire process of analysis of EEG while increasing the accuracy of the analysis.
This technical result is achieved in the proposed CSP is both analysis of electroencephalograms (EEG), at which carry out registration of EEG, spectral analysis of the recorded EEG using the continuous wavelet transform, the selection of physiologically significant sections of the analyzed signal, wherein the spectral analysis using the continuous wavelet transform is carried out in two stages, the first of which carry out an initial analysis of the registered digital signal "General" wavelet as the basis function which is used as a wavelet, similar in its characteristics with the basic plot of the EEG, for example, wavelet called "Mexican hat" (mexh), and form the matrix of recommendations on the selection of the class of physiologically significant features of the recorded EEG, based on the matrix recommendations secrete physiologically relevant sections of the analyzed signal, the second phase of the continuous wavelet transform, analyze physiologically relevant sections of the analyzed signal synthesized by the wavelets, the criterion of synthesis of which is chosen to minimize the sum of squared deviations of the wavelet from the reference signal, form a matrix of the results of the analysis, and get a video of the movement of the patient, which is correlated with the matrix of the analysis results and if the results matrix information about the appearance of physiologically significant the Oh features coincides with relevant information about the movement, confirm the patient is physiologically significant features and its type, and then use at least one trained artificial neural network to generate a matrix of clinical judgements on the basis of which form the clinical conclusion in the form of text that can be displayed to display or to a remote receiving point, on the basis of clinical findings, the conclusion about the presence of the disease.
At the same time, as recorded from multiple electrodes EEG is multi-channel, i.e. represents after digitization when registering EEG matrix, the columns of which correspond to samples (characterize the signal change in time), at the high sampling rate, the size of the matrix becomes very large.
Therefore, the computational threads (digital signal) can be divided into channels, and further processing can be performed for each channel or side channels (depending on need and from clinical cases).
However, to reduce the load on the computer it is preferable to carry out subsequent separation channel signals on the parts of fixed length, which gives the ability to zoom in or move to the analysis in real-time. In this case, processing (wavelet transform "General" wavelet) Khujand is realized over parts of fixed length channel signals.
The proposed method differs from the prototype in that the results of the continuous wavelet transform shall not be further processed in order of receipt of scalogram and scalogram and analyzed directly, i.e. they are the main source of information for identifying the characteristics of the digital signal. This approach allows us to estimate the shape of the EEG signal, its frequency, amplitude, temporal characteristics. Wavelet spectrogram, which can be obtained based on the continuous wavelet transform is more informative than scalogram and salagrama. The main purpose of the transition to scalograms in the prototype is the possibility of avoiding the dependency ratio and the amplitude of the wavelet spectrogram. Thus scalogram allows us to describe the distribution of power on the scale. But because of this, you lose the information about the shape of the signal changes in the negative half-plane. Salagrama allows to estimate the global energy spectrum. But because it is missing crucial information about the localization of singularities, there is a fundamental problem with choosing the length of the analyzed area, which should change depending on changes physiologically meaningful features, which, in turn, requires a priori knowledge of the type of physiologically C is achimas features and its precise location. This makes it impossible to fully automated EEG analysis method described in the prototype to obtain the necessary accuracy of the analysis.
In addition, in the proposed method, unlike the prototype is an initial analysis of each section "General" wavelet common to all plots, but as the basic functions of such a wavelet is proposed to use wavelet, similar in its characteristics with the basic plot of the EEG. This wavelet was chosen wavelet called "Mexican hat" (mexh).
This processing allows to estimate the EEG from the point of view of the presence of any physiologically significant features.
According to the results of the initial evaluation is the matrix of recommendations.
This allows you to more accurately select physiologically relevant sections of the analyzed signal, those sites that contain any feature. As a guide when selecting plot from a matrix of recommendations selects a number other than zero, the location of which corresponds to the original matrix, and the value corresponds to the conditional number of the group alleged physiologically meaningful features.
Unlike the prototype of the proposed system to improve the accuracy of EEG analysis is achieved by analysis of physiologically relevant sections of the analyzed signal is and the synthesized wavelets. The criterion of synthesis is proposed to use the minimization of the sum of squared deviations of the wavelet from the reference signal. Used or algebraic polynomial, or orthogonal polynomials (Chebyshev, Lagrange or other), or other method, the mathematical representation of the sample. In this case, synthesized wavelets must satisfy the following requirements:
1) received a basic function must be compact storage.
2) synthesized wavelets must have a zero integral value.
The best result of the continuous wavelet transform synthetic wavelets will be received at the description and display the features of the EEG, on the basis of the sample which it was derived (synthesized). A sample is a portion of the EEG, with appropriate physiological feature, serves as the basis for the synthesis wavelet.
Processing of the results of the analysis of the synthesized wavelet is performed using the algorithm, the block diagram of which is shown in figure 2.
Based on this processing automatically compiled the results matrix.
For analysis of the synthesized wavelet and additional processing was used synthesized base of wavelets.
As an additional tool to improve the accuracy of EEG analysis to identify psychologists who Eski significant features was applied mapping video contains information about the movement of the patient, with a matrix of results.
In that case, if the matrix results information about the appearance of physiologically significant features coincides with relevant information about the movement, it is possible with high probability to speak about the identity of one or other physiological characteristics (for additional indication of the artifacts associated with the movement of patients).
The result for fully automated EEG analysis is to obtain clinical opinion, which is formed using at least one training artificial neural network. The conclusion appears when the implementation of the method on the monitor in the form of text or can be transmitted to a remote receiving point.
When additional channel separation digital signals and their subsequent division into plots of a fixed length may be used when forming the clinical findings first trainable artificial neural network-channel processing, and then the learned artificial neural network processing the set of events that helps automate the process of analysis of EEG in real-time.
Thus, unlike the prototype, provides full automation of the process of analysis of EEG with simultaneous behavior of the solution accuracy of its analysis.
An example implementation of the proposed method is illustrated by drawings.
Figure 1 shows the block diagram of the auxiliary algorithm initial processing results of wavelet transformations (transformation General wavelet).
Figure 2 shows the block diagram of the processing algorithm of the results of the analysis of synthetic wavelets.
Figure 3 shows an example functional diagram of a system that implements the proposed method.
On figa, b, C, d, e, f, g, h is a block-diagram of the algorithm of the proposed method.
According to figure 3, the system for implementing the method comprises a serial connected device 1 registration of EEG with recorder EEG amplifier and ADC, with the corresponding inputs of the device 1 registration of EEG is connected to the electrodes mounted on the patient's head, block 2 spectral analysis using the wavelet transform ("General" wavelet), with the processing results of the wavelet transformation unit 3 forming the matrix of recommendations, unit 4 selection of physiologically significant sections of the analyzed signal, block 5 analysis of physiologically relevant sections of the analyzed signal synthesized by the wavelet, with processing results of the analysis of synthetic wavelet and block 6 forming the matrix, when this corresponding to the input unit 4 selection physiologically C is acimah plots of the analyzed signal and the first corresponding to the input unit 5 analysis of physiologically relevant sections of the analyzed signal synthesized by the wavelet is connected to the digital signal output device 1 registration of EEG and other relevant input unit 5 analysis of physiologically relevant sections of the analyzed signal synthesized by the wavelet is connected to the output unit 7 storage synthesized base of wavelets, the input connected to the output unit 8 synthesis adapted wavelets. The output of block 6 forming the matrix is connected to the input of block 9 of the comparison, the input/output bus associated with the unit 10 to the recording, storage and processing of video information on the movement of the patient, one input of which is connected to the output device 11 registration video. The device 11 registration video image is connected to the input/output bus unit 12 of the backlight, and the output to the input of the detector 13 motion, the output of which is connected to the second input unit 10 to the recording, storage and processing of video information on the movement of the patient. The output of block 9 of comparison is connected to the input unit 14 trainees artificial neural networks, the output of which is connected to the block 15 of the formation of the clinical findings, the output of which is connected to the appropriate input device 16 to output information, other relevant input connected to the output 1 of EEG registration. The device 16 to output information may include, for example, a printer and other devices. For remote transmission of clinical conclusion is to be placed in remote reception points in the system to implement the proposed method can be provided by the device 17 data corresponding inputs connected to the output unit 15 of the formation of the clinical findings. Data transfer can be performed using a local network, the Internet, the device read-write disks and other
The system for implementing the proposed method can also contain additional channel separation digital signals and their subsequent division into sections of fixed length connected in series block 18 channel separation digital signal and the block 19 of the separation channel signals on the parts of fixed length. In this case, the digital signal output device 1 of EEG registration can be connected to the input unit 18 channel separation digital signal, and the output unit 19 of the separation channel signals on the parts of fixed length connected to the input of block 2 spectral analysis using the wavelet transform ("General" wavelet) with the processing results.
Consider the implementation of the proposed method in the system shown in figure 3 and with regard to figure 1, figure 2 and figa, b, C, d, e, f, g, h.
When a clinical study is being recorded electroencephalogram of a patient. The signal from the surface of the head (brain) is removed by means of electrodes. Further from the electrodes, the signal enters the device 1 of EEG registration, where the signal is amplified and is recording under memory. For example, you can use the computer complex EEG studies, then the entry is made in the computer memory. Registered signal in the device 1 of EEG registration is digitized and output of this block, the signal is digital. The recording format of the digital signal may be different. Consider the case when the signal is a matrix where the number of columns corresponds to the number of channels and the number of rows corresponds to the number of samples of the signal, and indicates the length of the EEG recording. The number of samples of the signal depends on the recording time and the sampling frequency. For example, if the recording is continued for 15 minutes (900 s), with a sampling rate of 250 Hz, this signal will contain 225,000 times.
I.e. at the output of the device 1 registration of EEG signal are of the view matrix samples.
For convenience of further work, as well as to reduce requirements to the computer, you can divide the entire multi-channel (containing many derivations) signal (matrix) on separate channels (vectors). This will allow you to analyze the signal in the future through the channels, and also allows you to enter the procedure division of computational threads. In this procedure all the computational steps are performed componentwise, and synchronize with the main computing thread. The procedure of separation of the spacecraft is Aly is performed in block 18 channel separation digital signal.
Hereinafter, for convenience of further analysis, the demand for computing, as well as for the approximation of the operation of the device to the analysis mode (work) in real time, the signal is divided into sections of fixed length. This procedure is performed in block 19 of the separation channel signals on the parts of fixed length. The sections can be of different length, which depends on the tasks and possibilities of computer technology. For example, it is possible to separate the signals from each channel into sections based on the signal size that will fit on a standard screen of the computer monitor EEG complex. Conventionally, select 18 with that with a sampling rate of 250 Hz corresponds to 4500 samples.
In General, it can not produce channel digital signal separation and separation channel signals on parts of fixed length, then the digital EEG signal output device 1 registration of EEG (electroencephalograph) immediately subjected to the continuous wavelet transform (block 2 spectral analysis using the wavelet transform ("General" wavelet) processing results). I.e. the continuous wavelet transform "shared" by all wavelet multi-channel EEG signal.
In the case of separation of the analyzed signal channels and sections of fixed length, each part is OK subjected to wavelet transformation. This procedure is performed in unit 2 spectral analysis using the wavelet transform ("General wavelet") with the processing results. The wavelet transform "General wavelet" is the wavelet transform, which for all the sites, with the continuous wavelet transform is used the same wavelet. Studies have shown that of all the traditional families of wavelets is the best way to do this is analogous to the wavelet Mexican hat" (but can be used and different wavelet). This is because the basic function of a wavelet Mexican hat" is similar in its characteristics to the basic plot of the EEG signal. In view of this, the results of wavelet analysis, for example, the wavelet spectrogram, have a brighter intensity features of the EEG signal. However, in contrast to several families of wavelets, the use of this wavelet network-to-one correspondence of the maximum (minimum) of the original signal with the response of the wavelet spectrogram.
In addition, the wavelet Mexican hat" has a representation in the form of formulas that for most of wavelets is an exception.
The output of block 2 spectral analysis using the wavelet transform ("General wavelet") with the processing results obtained wavelet spectrogram. Wavelet spectrogram is a graphical display in which information about the signal. Is a three-dimensional graph (axis: axis - time (can be measured in ticks), the ordinate represents the change of scale and describes the scaling coefficients, applicat the results of wavelet analysis of W(a, b) (see formula 1)). Each of the scaling factor corresponds to a certain frequency value of the signal. I.e. if you choose a set of wavelet coefficients corresponding to the specific number (value) of the scaling factor on the basis of these data we can build a graph (two-dimensional), which is called the line of wavelet coefficients. This graph will describe the processes that occur in the monitored signal level of the selected frequency band. Has axis: axis - time (can be measured in ticks), y - the results of wavelet analysis of W(a, b) (see formula 1) for the selected value.
Next, the wavelet spectrogram processed results of the wavelet transform in accordance with the algorithm (figure 1).
It is known that different EEG features peculiar to their characteristics (frequency, amplitude, shape and so on). Choosing, building and analyzing lines of wavelet coefficients for frequency bands specific features (such frequency bands are more pronounced and make a great contribution to component this dia is Altanbulag), analyzing for all sites and channels, you can get a preliminary assessment (fairly accurate) about the composition of the whole record.
Thus, the matrix of wavelet coefficients of the wavelet spectrogram is subjected to decomposition by the vectors, resulting in the formation of the line wavelet coefficients. This procedure is performed for a certain numbers of wavelet coefficients, which correspond to a certain scale (the frequency of the original signal corresponding to the particular physiological characteristics). Further, the graphic lines of the wavelet coefficients (vectors of numbers), constructed for each of the characteristics (frequency range) examines one of the possible methods, for example:
- evaluation of amplitude values, checking for compliance with data values, the values of the amplitude characteristic features for the selected frequency range;
- check the characteristics of the lines of the wavelet coefficients (LAC) by the substitution in the template of acceptable values. The template is built with an indication of the dispersion characteristics of the signal, with a certain step in the selected window. The pattern is formed in advance and stored in the device memory;
- other methods aimed at extracting information about the availability of features in the EEG recordings.
Most preferred is to check the characteristics of the lines of wavelet-coefficients is of antov (LAC) by the substitution in the template of acceptable values, which is used in this example implementation.
On the basis of the received data (with a preliminary indication of the anticipated features are vectors on the basis of which into block 3, the formation of a matrix of recommendations is a matrix (table) recommendations. The zeros in the table correspond to the EEG without features. Each specific feature relates to the appropriate group (class) (e.g., ocular artifact or appearance of pathological activity), it is assigned a group number, which is stored in the table of recommendations in the appropriate place (where was discovered the feature). (For example, 1 - eye artifact, 2 - pathological activity etc). Thus, the table of recommendations is a matrix consisting of numbers, with different values. For EEG norms these numbers (if not selected special ways of further analysis) is equal to zero.
Next, the values of the table of recommendations received on the unit 4 selection of physiologically significant sections of the analyzed signal.
Block 4 is the analysis matrix (table) recommendations: there are elements different from zero. Presumably each element corresponds to the matrix element of the EEG signal, presumably containing the feature (the class features are already pre-defined). On the basis of full color is s recommendations formed "physiologically important areas". This is done as follows: in accordance with the line number and column non-null reference matrix of recommendations determined appropriate reference (signal) in the matrix EEG records received from the output device 1 of EEG registration. Relative to this reference is delayed a certain number of samples before and after (on the channel). Thus, a vector of a certain length. The length of this vector can be different (depending on the length of the "alleged" according to the table of recommendations features), and can be fixed and correspond to, for example, the minimum length of the area needed to display features (based on the conducted experiments, approximately one-third of a second). For example, let's take a length of 100 samples (i.e. 49 times before and 50 after the signal), with a sampling rate of 250 Hz is approximately the minimum length that is required. The signal readout is also part of the plot. Of course, in this case, the inevitable is some redundancy, i.e. if the table of recommendations sequentially (channel vector) are indicators features, the processing unit 4, you will get several almost identical plots (difference in one reference). To resolve this problem, you can perform decimation, i.e. zeroing through nectariniidae groups of repeating elements of the matrix of recommendations. However, to preserve the accuracy of the appropriate decimation is not performed.
Then, the obtained plots are sent to the input unit 5 analysis of physiologically relevant sections of the analyzed signal synthesized by the wavelets with the processing results.
The procedure of synthesis wavelet for the continuous wavelet transform is performed in advance, using the block 8 synthesis adapted wavelets, and is as follows.
1. The choice of the area of EEG signal containing the feature, the so-called sample.
2. The mathematical representation of this plot. This involves the approximation by the method of least squares is used or algebraic polynomial or orthogonal polynomial (or other method of mathematical representation of the sample).
Thus, a basis function of the wavelet. It performs the basic requirements to wavelets:
- compact media;
- zero integral value.
The resulting wavelet almost (or completely) coincides with the overlap with the original sample. Thus, the synthesized wavelet is adapted to the particular characteristics (based on plot-sample which was synthesized). The resulting wavelets are stored in block 7 base storage synthesized Valleta is. If necessary, they can be called from this block and used in block 5 analysis of physiologically relevant sections of the analyzed signal synthesized by the wavelets with the processing results.
In the same block 5, the obtained wavelet spectrograms are subjected to a comprehensive analysis of the evaluation of the wavelet spectrogram and is the comparison matrix of the original signal from the output device 1 registration of EEG and synthesized wavelet-specific features, in accordance with the algorithm (figure 2).
When the plot of EEG containing feature is analyzed using synthesized wavelet obtained from a sample of similar features (for example, a parcel containing ocular artifact is analyzed synthesized (adapted) the wavelet obtained from a sample of eye artifact), then the resulting wavelet spectrogram this feature will find vivid display. This comes at a time when synthesized wavelet acquires a value of the scale at which almost completely (or exactly) the same feature in the analyzed signal. Analyzing the "similarity" of the characteristics of part of the site EEG and synthesized wavelet, it is possible with a certain probability to judge about the nature of features (its class). The similarity frequent the area signal and the synthesized (adapted) wavelet can be expressed as a percentage. When this value reaches (or exceeds) the threshold values (for example, above 90%), with a certain confidence (90%) to say that it is exactly this feature. When you have different probability to assume that several different types of features, in the case of a large (significant) difference in value (for example, 82% and 36%) is the higher value (with the possibility of entry into the program log several options for possible additional testing, if needed).
When the difference is small (for example, 82% and 80%), it is possible to conduct more research in this part of the doctor in case of serious (critical) study.
According to the results of the analysis in block 5 analysis of physiologically significant sections of the analyzed signal synthesized by the wavelets with the processing results obtained vectors results, from which then a matrix (table) results in block 6 forming the matrix.
After receipt of a matrix (table) of the results it arrives on block 9 of the comparison. This block is used as an additional validation subsystem. During the procedure, EEG studies, simultaneously with the recording of EEG signal synchronous recording of video images of the patient. For this study alzueta device 11 registration image (video, camera). Because EEG study takes place in low light conditions (pritenennoe) room, you must use the block 12 illumination (for example, IR-floodlight or other).
The very video that can serve as a source of additional information about the artifacts caused by patient movement, and to automate the procedure of video display artifacts can be used detector 13 movement. The detector 13 motion allows to detect motion in a pre-selected regions of the zone of vision of the cameras. I.e. when installing and configuring cameras you can choose the area in which you will be detecting movement, the so-called active zones, and areas that will be analyzed will not be. This is necessary to avoid triggering by changing the position of the chest of a person during respiration, but also to be able to select multiple zones (e.g., face, eyes, hands etc) when the separation of the various artifacts.
Information from the device 11 registration video (cameras, video cameras) and the sensor 13 movement is recorded in the unit 10 to the recording, storage and processing of video information on the movement of the patient. In block 9 of the comparison is a comparison of the information matrix and information from the detector 13 movement. If information about the artifact coincides with information about stopping the Institute - this is an additional confirmation of the results of the automated analysis.
The results come on the block 14 trained artificial neural networks, from block 9 comparison with output 1 or output 2 (depending on the case) is augmented (modified) results matrix.
In block 14 of trained artificial neural networks the results matrix is divided into vectors (lines). The number of samples in the vector equal to (matches) the number of lead wires (channels). Further, each vector is analyzed for the presence of at least one nonzero element.
I) In case I, if at least one element in the vector is non-zero, the vector received at the input of the trained artificial neural network-channel processing. As the neural network can act as a multilayer perceptron. The choice of a neural network is determined by the number of inputs is determined by the number of possible features), capability analysis (processing) such amount of input information, education and health.
As a training method can be applied a method of back-propagation errors, or other method that allows to obtain a sufficient accuracy of the neural network in solving the following tasks.
The output of the current training artificial neural network-channel processing which is is the vector of output values. The vector of output values contains information estimation neural network evaluation of the presence or absence of specific types of features (numerical estimation). The result is achieved by pre-training the neural network channel estimation by the operator on the set of training examples (containing certain types of features, combinations thereof, or the lack thereof).
Next, the vector of output values is analyzed for the violation of its elements thresholds. The threshold value is determined at the time of debugging device. If the threshold is exceeded, then this item is set to the value of the number corresponding to the output of the neural network for this element.
If the element value is less than the threshold value, it is assigned a null value.
The resulting vectors are combined sequentially into the output vector of the trained artificial neural network-channel processing.
This vector is fed to the input of the counter same sequence of events. With zero initial conditions (the counter value is zero), is determined by the first event (an element of the vector is not equal to zero). The counter is assigned a value that is one more, i.e. a unit. Next, analyze the next non-zero element. If the value of this element coincides with the previous one, taschetto is assigned a value that is one more than the previous one, i.e. for a specific example or two. This continues up until the next non-zero element will be a different value than the previous one, then, in this case, a vector containing two elements. The first element corresponds to the type of features (corresponding to element the expense of repetition is performed). The second element denotes the number of repetitions features. The counter is reset to zero and the procedure is repeated for other types of features.
Further, each such vector sequentially combined into a single vector, which is fed to the input of the trained artificial neural network processing set of events.
This neural network can also be implemented using a multilayer perceptron or a similar type of neural network that allows you to perform the below described functions (operations).
This neural network in the simplest case can contain a fixed number of inputs (multiple of two). Its implementation can also be done using a database of neural networks with different number of inputs (when the analysis is conducted with the neural network, the number of inputs which correspond to the number of elements in the vector (the vector length)).
As a training method can be applied a method of back-propagation errors, or other method will get the sufficient accuracy of the neural network in solving the following tasks.
In the case when the number of inputs is fixed, three possible scenarios.
1. The number of inputs corresponds to the number of elements in the vector (the vector length).
2. The number of elements in the vector is less than the number of inputs of the neural network. In this case, on the other inputs are zeros or other numeric values, taken into account when training the neural network.
3. The number of elements in the vector is greater than the number of inputs of the neural network. In this case, the vector is divided into parts, equal in length to the number of inputs of the neural network. At the same time is determined by the number of such vectors. Analysis procedure the training of the artificial neural network processing the set of events is repeated for each vector separately. Building a neural network should be organized in such a way that this case has arisen with a minimum frequency (neural network must have a large number of inputs, while maintaining performance).
At the output of the training of the artificial neural network processing the set of events generated vector number of elements in the vector (the number of outputs of the neural network processing set of events) corresponds to the number of features provided to identify the device. Each element of the assessment of the availability of certain features. When was isolated and analyzed several vectors that what was tupile to the input of the training of the artificial neural network processing the set of events, the results of the analysis are summarized (elementwise) and the total elements are multiplied by a weighting factor. The value of the weighting factor is determined at the time of debugging device that implements this method. On the basis of the received vector is a matrix of clinical findings. It contains two rows and number of columns equal to the number of potential features.
The first line contains the number of possible physiologically significant features. The second line contains the assessment of the presence or absence of this feature, obtained from outputs of the training of the artificial neural network processing the set of events and are listed in the matrix clinical findings in order to match the values of the first row. Thus, the matrix clinical opinion contains all types of features in the first row and the appropriate assessment of their presence in the EEG recording in the second. Each rating corresponds to a group of features located in the first line of your matrix column clinical findings.
II) In case II, if non-zero elements have been identified and the number of nonzero counts equal to the number of all counts, clinical formed the conclusion that the patient is healthy. Clinical conclusion contains a matrix of two vectors (rows): the first group number of all possible features, and the second is all zeros. Information output unit 1.
In case II, if the number of zero counts not equal to the number of all samples, it generates a system message about the availability of features in the signal.
Further, the output 1 or output 2 block 14 trainees artificial neural networks to the input of block 15 of the formation of the clinical findings.
In block 15 of the formation of clinical findings there is a rounding matrix clinical findings (up to the order required for the selected precision analysis).
Next, based on pre-established and stored in memory a table of text matches, in accordance with the values of the matrix clinical findings, with the substitution in the text template (prepared in advance and stored in memory), is formed text clinical conclusion in the form of a text message containing information about the presence or absence of a disease in a patient, and identified artifacts during the study.
If necessary, based on the matrix (table) results may be detected location (time) and the recording channel, where the detected feature (features). When applicable, this section can be deduced from the matrix analyzed (original) EEG signal output device 1 of EEG registration to view.
Text clinical conclusion, and when n is required, all other information is supplied to the device 16 to output information (monitor, the printer and so on) and / or device 17 data (recording device (optical or magnetic (floppy) drive, external storage devices: flash memory, external hard drives and others), network, Internet, LAN, and others).
When implementing the proposed method as the device 1 of registration of EEG can be used EEG, for example, a Telepath-D, which operates at a sampling rate of 2000 Hz, although it is considered the method makes it possible to get acceptable results even at a frequency of 250 Hz and below.
To create the necessary level of illumination as block 12 of the backlight can be used one or more PC-projectors. For example, you can use two IR illuminator PEAK 23 EAP "Plate", one of which must be installed at the head of the patient, and the other at the feet or at the waist.
As the device 11 registration video can be used camera, for example, Ai-WD75N. To connect to your computer video encoder which can be used AXIS Q7404, which connects via Ethernet to the network card of the computer. The sensor 13 can be used for motion detection available in the camera and video encoder.
The proposed method can be used:
- at high flow patients PR is different surveys with low probability of pathology;
- during preventive examinations at the medical commissions of officers of various organizations;
- for the Express analysis of the EEG;
- for pre-flight checking or before other activities aimed at monitoring the health condition of the employee before, during or after events associated with large psycho-emotional loads;
- when building simulators for training of medical personnel;
1. The method of analysis of electroencephalograms (EEG), which carry out registration of EEG, spectral analysis of the recorded EEG using the continuous wavelet transform, the selection of physiologically significant sections of the analyzed signal, wherein the spectral analysis using the continuous wavelet transform is carried out in two stages, the first of which carry out an initial analysis of the registered digital signal "General" wavelet as the basis function which is used as a wavelet, similar in its characteristics with the basic plot of the EEG, for example, wavelet called "Mexican hat" (mexh), and form the matrix of recommendations on the selection of the class of physiologically important features of the recorded EEG on the basis of which secrete physiologically relevant sections of the analyzed signal, to second the m stage of the continuous wavelet transform, analyze physiologically relevant sections of the analyzed signal synthesized by wavelets, the criterion of synthesis of which is chosen to minimize the sum of squared deviations of the wavelet from the reference signal, form a matrix of the results of the analysis, and get a video of the movement of the patient, which is correlated with the matrix of the analysis results and if the results matrix information about the appearance of physiologically significant features coincides with relevant information about the movement, confirm the patient is physiologically significant features and its type, and then use at least one trained artificial neural network to generate a matrix of clinical judgements on the basis of which form the clinical conclusion in the form of text that can be displayed to display and/or transfer to the remote receiving point, on the basis of clinical findings, the conclusion about the presence of the disease.
2. The method according to claim 1, characterized in that the first stage of the spectral analysis using the continuous wavelet transform ("General" wavelet) carry over of the registered digital signal is divided into channels, each channel signal is divided into sections of fixed length.
FIELD: information technology.
SUBSTANCE: digital filtering method enables to pick up a signal in conditions with powerful broadband interference and can be realised in a digital filter. The method involves successive transformation of the filtered signal such that in each filtration step, fourth-order transformation is carried out, having transformation members only during even steps for signal sampled values. As a result of such signal transformation, the number of operations is reduced and operating speed is increased during filtration. The digital filter has a simple design and can be made using standard components.
EFFECT: higher speed of operation during band-pass filtering.
3 cl, 7 dwg
FIELD: information technology.
SUBSTANCE: described are methods for efficiently performing full and scaled transformations with data received through full and scaled interfaces, respectively. Full transformation is transformation which realises full mathematical description of transformation. Full transformation operates with or provides full transformation coefficients. Scaled transformation is transformation which operates with or provides scaled transformation coefficients which are scaled versions of full transformation coefficients. Scaled transformation can have less computational complexity, while full transformation can be easier to use through versions of applications. Full and scaled transformations can be for 2D IDCT, which can be realised by a separate method with 1D IDCT transformations. Full and scaled transformations can also be for 2D DCT, which can be realised by a separate method with 1D DCT transformations. 1D IDCT transformations and 1D DCT transformations can be realised by a computationally efficient method.
EFFECT: reduced complexity of coding equipment.
26 cl, 16 dwg
FIELD: information technology.
SUBSTANCE: in one version, the device has an input module for receiving an audio signal and generating time-domain input values representing the audio signal; a transform module for transforming the input values into spectral coefficients using a modified discrete cosine transform (MDCT) which is recursively split into at least a discrete cosine transform type IV (DCT-IV), a discrete cosine transform type II (DCT-II), or both the DCT-IV and DCT-II, where each such transform is of smaller dimension than the MDCT, wherein at least some multiplication operations of the MDCT are merged with a prior windowing operation applied to the input values.
EFFECT: reduction in complexity of the device and less memory usage.
35 cl, 15 dwg
FIELD: information technologies.
SUBSTANCE: vector of coefficients of reverse discrete cosine conversion is calculated using a sequence of butterfly-type structure operations above fixed-point numbers. Values of centre shift and a value of additional displacement are added to a coefficient of a matrix of scaled coefficients. Reverse discrete cosine conversion is applied to a resulting matrix of scaled coefficients. Values in a resulting matrix are shifted to the right for produce a matrix of pixel component values.
EFFECT: reduction of errors in process of compression and recovery of images and video information.
57 cl, 12 dwg
FIELD: information technologies.
SUBSTANCE: device comprises a control device, an analogue-digital converter, an assembly of cells with homogeneous structure with size J×K, where J - maximum quantity of scales in time, K - maximum quantity of time shifts of discretised continuous wavelet transform, at the same time each cell comprises an addressing unit, an integrator, a multiplier, a unit of shift registers.
EFFECT: simplification of device design with preservation of its efficiency for discretised continuous wavelet transform of data.
FIELD: information technology.
SUBSTANCE: invention is intended for use in digital high-speed devices whose operation employs discrete cosine transform for encoding video and audio data. The device for calculating discrete cosine transform consists of a demultiplexer, a first scalar multiplier, a second scalar multiplier, a real coefficient generator, a first fast Fourier transform unit, a second fast Fourier transform unit, an imaginary coefficient generator, a first complex multiplier, a second complex multiplier, a multiplexer and a comparator.
EFFECT: few calculation operations, high-speed operation owing to paralleling of calculation operations and possibility of calculating variable-length discrete cosine transform.
FIELD: radio engineering.
SUBSTANCE: method involves synchronous accumulation, spectrum analysis and decision taking procedure. Range of divergence of radio frequencies is separated in several sub-ranges. Number of sub-ranges is chosen on the basis of the required evaluation accuracy. Synchronous signal accumulation at the inlet of signal conversion device (SCD) is performed parallel for medium frequencies of each sub-range. After accumulation there are coefficients of direct orthogonal conversion for each sub-range. Ratio of maximum coefficient of this sub-range to maximum coefficient of decomposition in spectrum of carrying oscillation of SCD is defined for each sub-range. The decision on the sub-range in which the carrying frequency of SCD oscillation is present is taken as per maximum value of obtained ratios for all sub-ranges.
EFFECT: improving evaluation accuracy of carrying oscillation of SCD.
FIELD: information technology.
SUBSTANCE: in one version, the device for performing data transformation has a first logic circuit for multiplying a first group from at least one data value by a first group, at least from one rational binary constant which approximates a first group from at least one irrational constant scaled by a first common factor, wherein each rational binary constant is a rational number with a binary denominator; and a second logic circuit for multiplying a second group from at least one data value by a second group from at least one rational binary constant which approximates a second group from at least one irrational constant scaled by a second common factor, wherein the first and second groups from at least one data value have different sizes.
EFFECT: reduced complexity and high accuracy of transformation.
43 cl, 9 dwg, 6 tbl
FIELD: information technology.
SUBSTANCE: method comprises the following steps: matrix of offset coefficients is generated by adding one or more offset values to the coefficient of the matrix of initial coefficients. A matrix of transformed coefficients is generated using fixed-point arithmetic for repeated application of one-dimensional transformation to coefficients in the matrix of offset coefficients. A matrix of output coefficients is generated via right-side shifting of coefficients in the matrix of transformed coefficients, where coefficients in the matrix of output coefficients are approximations of values which could have been obtained by transforming the matrix of initial coefficients using an ideal inverse discrete cosine transform. The media presentation unit is instructed to output audible or visible signals based on the matrix of output values.
EFFECT: reduced round-off errors when calculating discrete cosine transform using fixed-point calculations.
83 cl, 12 dwg
FIELD: information technology.
SUBSTANCE: method comprises steps on which each coefficient in an 8×8 matrix of encoded coefficients is scaled with one multiplier for creating a matrix of scaled coefficients, scaled one-dimensional fixed-point transforms are repeatedly performed to transform the matrix of scaled coefficients into a matrix of transformed coefficients, the transformed coefficients are shifted to the right to create a matrix of corrected coefficients, wherein each corrected coefficient in the matrix of corrected coefficients approximates the corresponding value in the matrix of values, which can be created with application of an ideal two-dimensional inverse discrete cosine transform to the matrix of encoded coefficients, an 8×8 block of pixels is displayed, wherein each pixel in the said block includes a pixel component value taking into account the corrected coefficient.
EFFECT: eliminating approximation errors for calculating an inverse discrete cosine transform using fixed-point calculations.
66 cl, 12 dwg, 3 tbl
SUBSTANCE: individual is exposed to rhythmic coherent light, audio and vibrotactile signals. The session is preceded by rest and functional test electroencephalogram recording. The individual alpha activity values are determined: alpha peak frequency, alpha-band width, desynchronisation depth. The session is three-staged. At the first stage, a base frequency is specified as alpha peak frequency, while at the second stage the base frequency is maintained within the range of an individual flexibility resource described by the relations: Rmin=4.5-0.2*(F-10)-0.2*(ΔF-6) and Rmax=22+0.8*(F-10)+0.9*(ΔF-6) wherein F is an individual alpha peak frequency, ΔF is an individual alpha-band width. At the third stage, the base frequency is reduced to a value related to the required target functional state. Total time of the stimulation session is described by formula T=30+0.3*(D/10-10)2 wherein D is an individual desynchronisation frequency, %.
EFFECT: method provides higher effectiveness of correction that is ensured by regarding the individual characteristics of the patient.
1 dwg, 3 ex
SUBSTANCE: invention relates to field of medicine, namely to neurology, immunology and occupational pathology. Rheoencephalography with hypercapnic test are performed, visual evoked potentials (VEP), encephalography (EEG) are registered, level of antibodies to protein S100, level of immunoglobulin G in blood serum are determined. Canonical value is calculated by formula: Cv=3.18-0.38×a1-1.61×a2-0.71×a3+0.93×a4+1.19×a5-0.81×a6, where Cv is canonical value; -0.38; -1.61; -0.71; 0.93; 1.19; -0.81 are discriminant coefficients; 3.18 is constant; a1 is intensity of blood supply in frontomastoid pool during hypercapnic test in ohm; a2 is amplitude of peak N1 of VEP in right occipital lead in mcV; a3 is level of antibodies to protein S100 in conv. units; a4 is level of immunoglobulin G in g/l; a5 is presence or absence of nidus of pathological activity on EEG: 0 - no, 1 - present; a6 is degree of expression of diffuse changes by EEG; 0 - no, 1 - mild, 2 - moderate, 3 - expressed, 4 - severe. If Cv is lower than the constant, early manifestations of chronic mercury intoxication are diagnosed, if Cv is higher or equals the constant, chronic mercury intoxication of the first.
EFFECT: method extends arsenal of means for diagnostics of early manifestations of chronic mercury intoxication.
1 tbl, 3 ex
SUBSTANCE: invention relates to medicine, namely to neurology and psychiatry. Electroencephalogram (EEG) is registered at the background of hyperventilation on orthostatic table first in horizontal position, then, after turning the table on 60-70 degrees and bringing the examined person in orthostatic position. Registration of EEG is carried out in patient within 1 hour after sleepless night. Turning of orthostatic table is performed during 3-5 seconds. Examined person is brought into orthostatic position for 10 minutes and hyperventilation in orthostatic position is performed for 3 minutes. If signs of orthostatic blood circulation disorder are absent and epileptoform patterns are present, epileptoform activity is diagnosed.
EFFECT: method extends arsenal of means for detection of epileptiform activity.
SUBSTANCE: invention is applied in field of medicine, namely in neurology. Electric encephalogram is registered for patient of young age. If detected are: in occipital region alpha-rhythm with amplitude in range from 79 mcV to 98 mcV, reduction of alpha-rhythm index from 71% to 61%; in occipital, central parietal and central- frontal-anterior temporal regions of brain patterns of teta-activity with amplitude in range from 42 mcV to 48 mcV, teta-activity index from 11% to 40%, chonic ischemia of brain tissue is diagnosed.
EFFECT: method extends arsenal of means for diagnostics of chronic disorders of cerebral circulation, caused by arterial hypertension.
SUBSTANCE: invention relates to medicine and can be used for correction of person's functional state. Registration of electric encephalogram (EEG) is performed, EEG signals are transposed into sound range and patient is exposed to influence of sound signals. Spectral analysis of EEG is carried out, local extremums of obtained frequency spectrum of EEG are detected, frequencies of detected extremums of maximums and minimums are determined. From obtained values of frequencies of extremums selected are frequencies multiple to minimal one, with multiplication factor 2n where n is an integer number, and impact with sound signals with frequency which is multiple to, simultaneously multiple, frequencies of extremums of maximums and extremums of minimums of EEG frequency spectrum.
EFFECT: method extends arsenal of means for treatment of mental disorders.
3 dwg, 1 ex
SUBSTANCE: electroencephalography is performed; β2 - and δ - rhymes are recorded in F4 and P3 lead; a canonical variate is calculated by formula: Cv=1.6+1.6xa1-1.2xa2+0.7xa3-0.4xa4-) wherein Cv is the canonical value 1.6; -1.2; 0.7; -0.4 are discriminant coefficients; 1,6 is a constant; a1,2,3,4 are numerical values of research results of bioelectric cerebral activity: a1 is a percentage β2 of P3-lead rhythm, a2 is a percentage δ of P3-lead rhythm, a3 is a percentage δ of F4-lead rhythm, a4 is a percentage β2 of F4-lead rhythm. If the Cv value is more than 1.6, the absence of chronic action of mercury on the brain is stated; if the Cv is 1.6 or less, toxic encephalopathy is diagnosed in the animal.
EFFECT: method extends the range of the agents for diagnosing of toxic encephalopathy in small laboratory animals in chronic action of metallic mercury vapour.
1 tbl, 3 ex
SUBSTANCE: invention relates to medicine and can be used in neurology in diagnostics of frontal dysfunction. Registration of EEG is performed, spectral power of delta range with analysis of absolute values of power of the δ1-st rhythm and δ2-nd rhythm in frontopolar leads of both cerebral hemispheres is carried out. Absolute values of spectral power of the δ1-st rhythm and δ2-nd rhythm are determined. If obtained values of power of the δ1-st rhythm are from 182 mcV2 to 222 mcV2, the δ2-nd rhythm - from 42 mcV2 to 72 mcV2, dysfunction of frontal lobes is diagnosed.
EFFECT: method extends arsenal of means for diagnosing dysfunction of frontal lobes in patients with vibration disease.
7 dwg, 2 ex
SUBSTANCE: invention relates to medicine, namely to professional pathology. EEG registration is performed and analysis of spectral power of delta-range before and after treatment is carried out. Absolute values of power of δ1-st rhythm and δ2-nd rhythm in frontal leads of both cerebral hemispheres are analysed. If absolute values of power of δ1-st rhythm in right leads decrease more than by 37%, and in left leads - by more than 44% and of δ2-nd rhythm in left leads by more than 70% in comparison with said indices before treatment, performed therapy is assessed as efficient.
EFFECT: method increases reliability of assessment of efficiency of performed therapy of cognitive disorders in patients with vibration disease.
SUBSTANCE: invention refers to medicine, namely to anaesthesiology, and may be used as an anaesthesia care of surgical correction of severe spinal scoliosis with a high risk of developing neurological complications. For this purpose, 30 minutes prior to the operation, intramuscular pre-medication with Dormicum 0.1 mg/kg and Dimedrol 0.4 mg/kg is required. The anaesthesia is induced by Phentanyl 0.002 mg/kg, Propofol 2.5 mg/kg. The introduction of Nimbex 0.1 mg/kg is followed by the trachea intubation. After the trachea intubation and transition to artificial pulmonary ventilation, loading doses of Clopheline 0.004 mcg/kg and Ketamine 0.25 mg/kg are introduced. Sevorane in the concentration of 4 vol. % immediately follows the trachea intubation and transition to artificial pulmonary ventilation assisted by the air and oxygen flow rate of 4-5 l/min to reach the breath-out sevorane concentration min. 2.6 vol. % (1.04 minimum alveolar concentration). Then the air and oxygen flow rate is decreased to 1 l/min. Artificial pulmonary ventilation is enabled by an anaesthesia apparatus for sevorane delivery in forced pulmonary ventilation mode with the low fresh gas flow rate 1 l/min with minute tidal volume to ensure the breath-out concentration of carbon dioxide within 32-37 mm Hg, the concentration of oxygen in the mixture 40%. The mandatory safety monitoring involves blood pressure, heart rate, electrocardiogram, arterial blood oxygen saturation, mixture oxygen concentration, breath-out carbon dioxide concentration, air and oxygen sevorane concentration, breath-out sevorane concentration, breath-in air and oxygen carbon dioxide and bispectral electroencephalogram and electromyography index recordings. The anaesthesia is maintained by sevorane inhalations 3-1.5 vol. %. (1.2-0.6 minimum alveolar concentration), bolus introductions of Fentanyl 0.004±0.001 mg/kg/h, continuous infusion of Clopheline 0.004 mcg/kg/h, Ketamine 0.25 mg/kg/h and supporting Nimbex 0.05-0.03 mg/kg/h. 30 Minutes before the patient wakes up, sevorane dose is maintained at 1.0-0.8 vol. %, 20 minures before, the Nimbex introduction is completed, 15 minutes before, sevorane delivery is completed, 30 minutes before the waking up, the Fentanyl introduction is completed, while Clopheline and Ketamine are kept to be introduced. The spinal function monitoring is controlled by electroencephalogram activity and nervomuscular conduction as shown by electromyography. Patient contact is considered to be allowed if observing the bispectral electroencephalogram index min. 75-78% and the degree of residual neuromuscular blocks max. 20%. After obtaining the spinal function monitoring data, the bolus introduction of Fentanyl 0.002 mg/kg, Nimbex 0.1 mg/kg, while sevorane is started to be introduced in the concentration of 4 vol. %. Then concentration of Sevorane is reduced to 3-1.5 vol. % (1.2-0.6 minimum alveolar concentration), Clopheline and Ketamine are kept to be infused in the previous dosages.
EFFECT: method enables high control of the anaesthesia and an effective level of antinociceptive protection while the patient wakes up that is ensured by multidirectional action of the presented components of the anaesthesia.
SUBSTANCE: invention relates to field of medicine, in particular, to neurology, electrophysiology, functional diagnostics, psychiatry and physiology. Value of correlation coefficient is determined by data of electroencephalographic (EEG) examination. Correlation coefficients are calculated in leads between temporal, frontal, central-parietal and occipital areas of the brain. If correlation coefficient increases between temporal areas and its simultaneous decrease in leads between the remaining areas of the brain, higher readiness for convulsions is determined in the person.
EFFECT: method extends arsenal of means for detection of person's readiness for convulsions.
FIELD: medicine, neurology, psychopathology, neurosurgery, neurophysiology, experimental neurobiology.
SUBSTANCE: one should simultaneously register electroencephalogram (EEG) to detect the level of constant potential (LCP). At LCP negativization and increased EEG power one should detect depolarizational activation of neurons and enhanced metabolism. At LCP negativization and decreased EEG power - depolarized inhibition of neurons and metabolism suppression. At LCP positivation and increased EEG power - either repolarized or hyperpolarized activation of neurons and enhanced metabolism. At LCP positivation and decreased EEG power - hyperpolarized suppression of neurons and decreased metabolism of nervous tissue. The method enables to correctly detect therapeutic tactics due to simultaneous LCP and EEG registration that enables to differentiate transition from one functional and metabolic state into another.
EFFECT: higher accuracy of diagnostics.
5 dwg, 1 ex, 1 tbl