CN112220482B - Method for detecting and eliminating magnetoencephalogram eye movement artifact based on neural network and electronic device - Google Patents
Method for detecting and eliminating magnetoencephalogram eye movement artifact based on neural network and electronic device Download PDFInfo
- Publication number
- CN112220482B CN112220482B CN202011026445.9A CN202011026445A CN112220482B CN 112220482 B CN112220482 B CN 112220482B CN 202011026445 A CN202011026445 A CN 202011026445A CN 112220482 B CN112220482 B CN 112220482B
- Authority
- CN
- China
- Prior art keywords
- signal
- eye movement
- magnetoencephalogram
- segment
- movement artifact
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
- A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
- A61B5/0042—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part for the brain
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
- A61B5/7207—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal of noise induced by motion artifacts
Abstract
The invention provides a method for detecting and eliminating magnetoencephalography eye movement artifacts based on a neural network and an electronic device, wherein the method comprises the following steps: cutting the magnetoencephalogram signal to be detected, and drawing a plurality of magnetoencephalogram signal views according to the signal segments and the signal detection positions of the magnetoencephalogram sensor; extracting signal space distribution characteristics of a brain magnetic signal view, and classifying the signal space distribution characteristics; acquiring an eye movement signal interference segment according to a fixed ratio and an eye movement eye electrical signal segment of a corresponding time point of a magnetoencephalogram signal segment containing an eye movement artifact noise signal; subtracting the magnetoencephalogram signal segment containing the eye movement artifact noise signal from the eye movement signal interference segment, and reducing the obtained eye movement artifact noise signal-removed segment to the corresponding position. The invention does not need the eye electrical signal measured by the electrode during detection, and does not interfere the brain magnetic signal which is not influenced by the eye electrical signal during removal, so that the information in the original brain magnetic signal can be retained to the maximum extent.
Description
Technical Field
The invention relates to the field of magnetoencephalogram artifact signal removal in the field of biological feature recognition, in particular to a magnetoencephalogram eye movement artifact detection and removal method based on a neural network and an electronic device.
Background
The magnetoencephalogram is an applied brain function image detection technology which is obtained by utilizing an extremely sensitive magnetic sensor to realize the detection of weak magnetic signals emitted by the human brain and has the advantages of no invasion, no damage and the like. At present, the magnetoencephalogram technology has been used for the research of higher brain functions such as thinking, emotion and the like, is widely used for the surgical treatment of functional diseases such as neurosurgical preoperative brain function positioning, epileptic focus surgical positioning, Parkinson's disease, psychosis, drug addiction treatment and the like, and also has important clinical medical application in cerebrovascular diseases and fetal neurological diseases. Magnetoencephalography is therefore a leading medical diagnostic technique for brain diseases.
In the application of magnetoencephalography, one of the most important is the localization of the location of a disease in the brain using magnetoencephalography. Because the measurement of the magnetoencephalogram does not need craniotomy, and only the sensor is tightly attached to the brain of a human, the magnetoencephalogram is more friendly to patients and simpler and more convenient to operate compared with other brain disease diagnosis technologies. However, because the magnetic signal emitted by the human brain is extremely weak, various noises appearing in the magnetoencephalogram need to be eliminated before the magnetoencephalogram signal is used for positioning the position of the brain disease, so that the signal-to-noise ratio is improved to the maximum extent. And then tracing and positioning the characteristic signals emitted by the disease in the brain according to the denoised magnetoencephalogram signals. Since the measurement of magnetoencephalogram is generally performed in a magnetic shielding room, the noise sources are mainly 3 types: the first is power frequency noise, which is typically 50Hz or 60 Hz. The harmonics of which are typically 50Hz or integer multiples of 60 Hz. The power frequency noise and its harmonics are generally not in one frequency band with the signal we are interested in the magnetoencephalogram (magnetoencephalogram signals are generally 0-40 Hz). This type of noise can be removed by simple filtering. The second type of noise is experimental noise, which may be caused by improper operation during the experiment, such as relative displacement between the sensor and the subject's head, or damage to the experimental equipment. These problems all cause fluctuations in the magnetoencephalogram signal, which masks the signal we want, and this part of the fluctuations is experimental noise. The experimental noise is irregular and appears randomly, but the characteristics are obvious under general conditions, and a professional physician needs to check the magnetoencephalogram frame by frame, and the magnetoencephalogram is screened out and rejected manually. The last type of noise is called artifact noise and is a weak magnetic signal generated by the heartbeat, eye swipe, blink, twitching of the face muscles and neck muscles of the subject. These signals are also detected by our magnetoencephalogram detection sensors. Thereby interfering with the magnetic signals emitted by the brain of the subject. But in general the magnetic signal generated by the muscle is too weak and therefore does not need to be considered. In addition, because the heart is far away from the brain, the heartbeat magnetic signal detected by the magnetoencephalogram sensor is very weak and can not be considered. However, the eye is very close to the magnetoencephalogram sensor, and magnetic signals generated by the eye sweeping or blinking can generate very large interference on the detected magnetoencephalogram signals, so that special elimination of the eye artifact noise signals is needed.
The frequency of the eye movement artifact signal is typically 0-3Hz, overlapping with the brain magnetic signal we are interested in the frequency band. It cannot be removed using simple filtering. At present, people mainly use a signal space projection or an independent component analysis method to separate an eye movement artifact signal from a brain magnetic signal and then remove the eye movement artifact signal. In fact, however, these signal denoising algorithms were originally proposed for electroencephalogram signals and then extended to the field of electroencephalogram signals. The eye movement artifact noise signals are removed in the brain magnetic signals by using signal space projection or independent component analysis, and the brain magnetic signals which are not interfered by the eye movement artifact noise signals are influenced. This is determined by the nature of these two algorithms. They essentially look for a specific pattern of eye movement artifact noise signals and then remove from the signal pattern all components of the detected brain magnetic signal where this pattern is present. General signal space projection or independent component analysis requires electrodes to measure an eye electrical signal emitted by the eyes of a human subject, and then a fixed mode of the influence of the eye electrical signal on a brain magnetic signal is obtained through the eye electrical signal. However, since the brain magnetic signals are very complex, we generally consider the brain magnetic signals to be a superposition of signals from 15000 signal sources on the cerebral cortex. The pattern of the eye movement artifact noise signal cannot be found perfectly by these two algorithms in the first place. In addition, even if the method is found, due to the complexity of the brain magnetic signal source and the complexity of signal superposition, the eye movement artifact signal noise is difficult to be perfectly separated from the brain magnetic signal concerned by people. Therefore, the two algorithms are implemented to influence the brain magnetic signals which are not interfered by the eye movement artifact noise signals.
Disclosure of Invention
Aiming at the defects of the traditional method for removing the eye movement artifact noise signals in the magnetoencephalogram, the invention provides a method for detecting and removing the eye movement artifact of the magnetoencephalogram based on a neural network and an electronic device, which can effectively and accurately identify the eye movement artifact noise signals in the magnetoencephalogram and remove the eye movement artifact noise signals.
The technical scheme of the invention is as follows:
a magnetoencephalography eye movement artifact detection method based on a neural network comprises the following steps:
1) cutting the magnetoencephalogram signal to be detected to obtain a plurality of signal segments;
2) drawing a plurality of brain magnetic signal views according to the signal segments and the detection signal positions of the brain magnetic sensors;
3) and extracting the signal space distribution characteristics of the magnetoencephalography signal view, and classifying the signal space distribution characteristics to obtain whether the unknown magnetoencephalography signal contains eye movement artifact noise signals.
Further, the form of the signal segment includes: and M-N two-dimensional matrix data, wherein M is the number of the brain magnetic sensors, and N is a set time length.
Further, the brain magnetic signal view is a 2D view; obtaining the brain magnetic signal view by the following steps:
1) calculating the average power of the detection signals of each brain magnetic sensor in each signal segment;
2) and coding the average power by using colors, and drawing the average power to a spatial position corresponding to the brain magnetic sensor to obtain a brain magnetic signal view.
Further, extracting signal space distribution characteristics of the brain magnetic signal view through a trained two-dimensional deep convolution neural network, and classifying the signal space distribution characteristics; wherein a two-dimensional deep convolutional neural network is trained by:
a) collecting sample magnetoencephalogram signal fragments, and setting a label according to whether an eye movement artifact noise signal is contained;
b) drawing a plurality of sample magnetoencephalography signal views according to the sample magnetoencephalography signal fragments and the corresponding magnetoencephalography sensor detection signal positions;
c) and training the two-dimensional deep convolution neural network through a plurality of sample brain magnetic signal views to obtain the trained two-dimensional deep convolution neural network.
Further, the two-dimensional deep neural network comprises a GoogLeNet network, an AlexNet network or a two-dimensional deep convolutional neural network with node number and weight symbol requirements.
Further, the method for classifying the signal space distribution features comprises the following steps: a Softmax classifier was used.
A magnetoencephalography eye movement artifact removing method based on a neural network comprises the following steps:
1) obtaining a magnetoencephalogram signal segment containing an eye movement artifact noise signal in the magnetoencephalogram signal to be detected by the method;
2) acquiring a corresponding eye movement signal interference segment according to a fixed ratio and the eye movement electric signal segment of the corresponding time point of the magnetoencephalogram signal segment containing the eye movement artifact noise signal;
3) subtracting the magnetoencephalogram signal segment containing the eye movement artifact noise signal from the corresponding eye movement signal interference segment to obtain an eye movement artifact noise removal signal segment, and reducing the eye movement artifact noise removal signal segment to the corresponding position of the magnetoencephalogram signal segment containing the eye movement artifact noise signal.
Further, the fixed ratio is obtained by:
1) calculating the ratio of the sample magnetoencephalogram signal segment to the corresponding sample eye movement electric signal segment according to a plurality of sample magnetoencephalogram signal segments containing eye movement artifact noise signals and the sample eye movement electric signal segments at corresponding time points;
2) and calculating the average value of the ratios to obtain the fixed ratio.
A storage medium having a computer program stored therein, wherein the computer program is arranged to perform the above-mentioned method when executed.
An electronic device comprising a memory having a computer program stored therein and a processor arranged to run the computer to perform the method as described above.
Compared with the prior art, the invention has the following advantages:
1) eye movement artifact noise signals in unknown brain magnetic signals can be automatically identified, eye electric signals measured by electrodes are not needed, and hardware is saved;
2) based on the recognition result of the deep convolutional network, according to the influence rule of the eye electrical signals on the brain magnetic signals, the interference of the eye electrical signals on the brain magnetic image data can be eliminated, compared with the traditional algorithm, the brain magnetic signals which are not influenced by the eye electrical signals cannot be interfered, and the information in the original brain magnetic signals can be reserved to the maximum extent.
Drawings
Fig. 1 is a flowchart of an automatic detection algorithm of an eye movement artifact noise signal according to an embodiment of the present invention.
Fig. 2 is a flowchart of an automatic eye movement artifact noise signal removal algorithm according to an embodiment of the present invention.
FIG. 3 is a diagram of a 360 second, 600Hz sample rate of a known magnetoencephalography signal used in an embodiment of the present invention.
FIG. 4 is a diagram illustrating an eye movement artifact noise signal in a known brain magnetic signal according to an embodiment of the present invention.
Fig. 5 is a 2D view of an embodiment of the invention using eye movement artifact noise signal segments.
FIG. 6 is a schematic diagram of an eye movement electrical signal measured simultaneously with a known brain magnetic signal according to an embodiment of the present invention.
FIG. 7 is a diagram of a 360 second, 600Hz sample rate of known magnetoencephalography signals used in an embodiment of the present invention.
FIG. 8 is a schematic diagram of an eye movement electrical signal measured simultaneously with an unknown brain magnetic signal according to an embodiment of the present invention.
Fig. 9 is a schematic diagram of the detection result of google lenet on eye movement artifact noise signals in known brain magnetic signals and unknown brain magnetic signals adopted in the embodiment of the present invention.
Fig. 10 is a diagram illustrating a result of removing eye movement artifact noise signals in one of the unknown brain magnetic signals according to the embodiment of the present invention.
FIG. 11A is a schematic diagram of eye movement artifact noise signals in raw magnetoencephalogram data using a signal space projection method.
Fig. 11B is a signal diagram obtained after removing the eye movement artifact noise signal using the signal space projection method.
Fig. 11C is a schematic diagram of the difference between the signals in fig. 11A and 11B.
Detailed Description
In order to better explain the technical scheme of the invention, the invention is further described in detail with reference to the accompanying drawings and specific embodiments.
The invention discloses an automatic detection method of eye movement artifact noise signals in a magnetoencephalogram, which comprises the following steps as shown in figure 1:
1) and (3) carrying out target signal interception on the known magnetoencephalogram data, wherein the target signal is an eye movement artifact noise signal segment. Obtaining eye movement artifact noise signal segments therein: each eye movement artifact noise signal segment is a data set in a two-dimensional matrix form with the size of M x N, and M is the number of the brain magnetic sensors; and N is a set time length.
2) And (2) drawing a brain magnetic signal 2D view by using the eye movement artifact noise signal segment obtained in the step (1) and the position information of the detection signal of each sensor. The spatial distribution of the eye movement artifact noise signal segments is obtained. The spatial distribution of the noise signal segments is very obvious for the eye movement artifact, and is similar to the spatial distribution of the magnetic dipoles. All the resulting 2D views constitute the training set 1 and are labeled as eye movement artifact signals.
3) Randomly intercepting the known magnetoencephalogram data to obtain random magnetoencephalogram signal fragments: each magnetoencephalogram signal segment is also a data set in a two-dimensional matrix form with the size of M x N, and M is the number of magnetoencephalogram sensors; and N is a set time length.
4) And 3) drawing a 2D view of the magnetoencephalography signal by using the random magnetoencephalography signal fragments obtained in the step 3) and the position information of the detection signal of each sensor. The spatial distribution of random magnetoencephalogram signal segments is obtained. Due to the random truncation, where eye movement artifact noise signal segments may be present, their 2D view is labeled eye movement artifact signal and is included in the training set 1. For non-eye movement artifact noise signal segments, they are labeled as normal magnetoencephalogram signals and are included in training set 2.
5) Training an eye movement artifact noise signal detection model by using the training set 1 and the training set 2 obtained in the steps 2) and 4) as training sets; the eye movement artifact noise signal detection model comprises a feature learning module and a classification output module; the feature learning module comprises a two-dimensional deep convolutional neural network and is used for learning the spatial distribution features of the eye movement artifact noise signals presented in the training set, namely the 2D view, extracting the spatial distribution features and constructing a corresponding deep convolutional neural network model. Obtaining feature data of a corresponding channel, namely local feature data of the magnetoencephalogram data segment; and then, the classification output module divides the training set data into eye movement artifact noise signal segments and normal magnetoencephalogram signal segments according to the eye movement artifact noise signal space distribution characteristics extracted from the characteristic learning module.
6) And carrying out cutting operation on the unknown magnetoencephalogram signal, and dividing the unknown magnetoencephalogram signal into signal segments with the time length of N.
7) And 3) drawing 2D views reflecting signal space distribution by using the signal segments obtained in the step 6), and then respectively inputting the 2D views into the trained eye movement artifact noise signal detection model to obtain the eye movement artifact noise signal detection result corresponding to the magnetoencephalogram data segments.
8) Determining which data segments in the unknown magnetoencephalogram signal are influenced by the eye movement artifact noise signal according to the eye movement artifact noise signal detection result obtained in the step 7).
Further, the M depends on the number of channels for detecting the brain magnetic signals, and the minimum N may be a data length of one sampling, and the maximum N may be 0.4 second. Since the eye movement artifact noise signal length typically does not exceed 0.4 seconds.
Furthermore, the 2D view is drawn according to the average power of the detection signal of each sensor in the signal segment, and the average power is color-coded and drawn to the spatial position corresponding to the sensor, so as to obtain the 2D view of the spatial distribution of the signal power.
Further, the two-dimensional deep convolutional neural network in the feature learning module may adopt a pre-training network such as google lenet or AlexNet, and may also design a two-dimensional deep convolutional neural network whose parameters such as node number and weight meet requirements.
Further, the classification output module can be realized by adopting a full connection layer, and finally, the probability of different classes is calculated by using a Softmax function.
A method for automatically eliminating noise signals of eye movement artifacts in a magnetoencephalogram, as shown in FIG. 2, comprises the steps of:
1) and (3) carrying out target signal interception on the known magnetoencephalogram data, wherein the target signal is an eye movement artifact noise signal segment. Obtaining eye movement artifact noise signal segments therein: each eye movement artifact noise signal segment is a data set in a two-dimensional matrix form with the size of M x N, and M is the number of the brain magnetic sensors; and N is a set time length. Suppose the number of the intercepted eye movement artifact noise signal segments is n. The eye movement artifact noise signal fraction in known magnetoencephalogram data can be formulated as:
MEG＝{data1,data2,...,datan}
2) and carrying out target signal interception on the known electro-oculogram data, wherein the target signal is an eye movement electro-oculogram signal segment. Obtaining an eye movement eye electrical signal segment within the time corresponding to the eye movement artifact noise signal segment in the step 1): each eye movement electric signal segment is a data set with the size of l × N matrix, l is the number of sensors for measuring the eye movement artifact noise signal segment, and N is a set time length, wherein l is 1 in the embodiment. The intercepted eye movement electric signal segment is also n. The eye movement electric signal segment corresponding to the known magnetoencephalography data can be expressed by the following formula:
EOG＝{edata1,edata2,...,edatan}
edatai＝{b1,b2,...,bN}i
3) dividing the signals of the M channels in the eye movement artifact noise signal segment obtained in the step 1) by the eye movement signals obtained in the step 2). M × n fixed ratios are obtained. This ratio can be formulated as:
4) averaging the ratio obtained in the step 3) according to the number n of eye movement artifact noise signal segments to obtain M fixed ratios riFixed ratio R of composition, where RiCorresponding to M channels, respectively. R can be formulated as:
R＝{r1,r2,...,rM}T
5) and carrying out cutting operation on the unknown magnetoencephalogram signal, and dividing the unknown magnetoencephalogram signal into signal segments with the time length of N.
6) Respectively inputting the magnetoencephalogram data segments obtained in the step 5) into the trained eye movement artifact noise signal detection model to obtain eye movement artifact noise signal classification results of unknown magnetoencephalogram data segments.
7) Intercepting all eye movement artifact noise signal segments in the position signal according to the classification result obtained in the step 6). Each eye movement artifact noise signal segment is a data set in a two-dimensional matrix form with the size of M x N, and M is the number of the brain magnetic sensors; and N is a set time length. Assuming m segments of eye movement artifact noise signal segments, it can be formulated as:
unMEG＝{undata1,undata2,...,undatam}
8) and intercepting the eye movement electric signal within the known corresponding time according to the time point of the eye movement artifact noise signal segment intercepted in the step 7) to obtain the eye movement electric signal segment within the corresponding time. Where l × N is a data set in the form of a matrix, l is the number of sensors measuring the eye movement artifact noise signal segment, and N is a set time length, where l is set to 1 in this embodiment. The truncated eye movement eye electrical signal segment is also m. Known eye movement electrical signal segments can be formulated as:
unEOG＝{unedata1,unedata2,...,unedatam}
unedatai＝{b1,b2,...,bN}i
9) multiplying the eye movement eye electrical signal segments obtained in the step 8) by the fixed ratio R obtained in the step 4) to obtain m eye movement signal interference segments. Each eye movement signal interference segment is a data set in a two-dimensional matrix form with the size of M x N, and M is the number of sensors for measuring the eye movement artifact noise signal segment; n is the length of time the eye movement artifact noise signal segment lasts. The eye movement signal interference segment may be formulated as:
Interference＝{I1,I2,...,IM}T＝unEOG×R
10) subtracting the eye movement signal interference segment obtained in the step 9) from the eye movement artifact noise signal segment obtained in the step 7) to obtain a brain magnetic signal segment without the influence of the eye movement artifact noise signal. Can be formulated as:
correctMEG＝unMEG-Interference
11) restoring the magnetoencephalography signal segment which is obtained in the step 10) and is removed of the influence of the eye movement artifact noise signal back to the original position of the unknown magnetoencephalography signal in the step 5). The automatic elimination of the eye movement artifact noise signals of the unknown brain magnetic signals is realized.
The following is a specific embodiment of the present invention, which specifically includes the following steps:
firstly, for known brain magnetic signals, the eye movement artifact noise signal segments are intercepted. The signals used here are brain magnetic signals measured by SQUID-MEG equipment of CTF, canada, which has 274 sensors, so 274 sets of signals are measured. The length of each intercepted signal is the sampling time, the sampling rate is 600Hz, and therefore the interception time is 1/600 seconds. The truncated signal segment is a vector of 274 x 1. As in fig. 3, for the 360 seconds of known magnetoencephalogram signal used, where the spikes are eye movement artifact noise signal fragments, the signals detected by the 274 sensors are placed in the same coordinate system.
Since there are 19 eye movement artifact noise signals in the data used, one eye movement artifact noise signal is about 0.4 seconds long. We cut 60 groups of data out of an eye movement artifact noise signal, i.e. 0.1 seconds of data (sample rate 600 Hz). We cut a total of 1140 sets of data from the 0.4 x 19 second long data. With 1140 sets of data as a training set 1, which is color coded according to the position of the probe, 1140 2D views representing the spatial distribution of the signal can be obtained. As in fig. 4, one of the eye movement artifact noise signal segments, which is about 0.4 seconds in duration, has 0.1 seconds of data, i.e., 60 sets of data, truncated therein. The 19 eye movement artifact noise signals may result in 1140 groups of data. As in fig. 5, a 2D view of the spatial distribution of the eye movement artifact noise signal can be displayed, plotted against the vector data of 274 x 1 in combination with its probe position.
In addition, the normal brain magnetic signal segments with the same length are randomly intercepted, and here, the normal brain magnetic signal segments of 2000 groups are intercepted, and are vectors of 274 x 1. Depending on the position of the probe, it is color coded to obtain 2000 2D views representing the spatial distribution of normal brain magnetic signals as the training set 2.
In addition, it is necessary to intercept the eye movement eye electrical signal segment corresponding to the time point, and the eye movement eye electrical signal segment is measured by placing the electrode beside the eye of the subject. The eye movement artifact noise signal of 19 × 274 × 1 is divided by the eye movement signal at the corresponding time point to obtain a ratio of 19 × 274 × 1, and this is averaged by the number 19 of eye movement eye electrical signal segments to obtain a ratio R of 274 × 1. As in fig. 6, is the eye movement electrical signal measured simultaneously with the known brain magnetic signal.
And combining the training set 1 and the training set 2 into a training set total, and training the pre-training network GoogLeNet. The structure of the pre-training network GoogLeNet is adjusted correspondingly, so that the recognition of eye movement artifact noise signal segments can be completed. The main adjustment modifies the final full-link filter number to equal the number of classifications (i.e., eye movement artifact noise signal versus normal brain magnetic signal). The learning rate factor of the full connection layer is increased, so that the data can be converged more quickly; furthermore we modify the number of output classes of a given network in the classification layer. The classification layer is replaced with a new classification layer without class labels. When training the network, it will automatically set the output class of the layer; meanwhile, the last dropout layer is modified, and the dropout layer is mainly used for randomly setting parameters transmitted from the previous layer to be 0 so as to prevent overfitting of the network. Here we adjust the probability of setting the element to 0 for this layer from 0.5 to 0.6. The network is then trained.
The trained network can be used for detecting the eye movement artifact noise signals of the original training data, and can also be used for detecting the eye movement artifact noise signals existing in the unknown brain magnetic signals to obtain the positions of the eye movement artifact noise signals in the unknown brain magnetic signals. In an embodiment, the success rate of detection of eye movement artifact noise signal fragments in known brain magnetic data by our network is 100%. The success rate of detecting the eye movement artifact noise signal segments in the unknown brain magnetic data is 85%. There were 3 eye movement artifact noise signal segments misjudged by the network. As in fig. 7, is the unknown brain magnetic signal used. As in fig. 8, is the eye movement electrical signal measured simultaneously with the unknown brain magnetic signal. As shown in fig. 9, the detection results of eye movement artifact noise signals in known and unknown brain magnetic signals are obtained by using the modified google net. Wherein the dashed lines represent misjudged eye movement artifact noise signal segments.
According to the detection result, the segment of the unknown brain magnetic signal in which the eye movement artifact noise signal appears is intercepted, and the fact that 20 eye movement artifact noise signals exist in the unknown brain magnetic signal can be known. And multiplying R by the eye movement eye electric signal at the corresponding time point according to the calculated ratio R to obtain the eye movement signal interference segment. The method can remove the interference of the eye electrical signals by subtracting the corresponding eye movement signal interference fragment from the eye movement artifact noise signal fragment in the unknown magnetoencephalogram to obtain a clean magnetoencephalogram signal, and cannot influence the undisturbed part in the original magnetoencephalogram signal. As shown in fig. 10, the result of removing the eye movement artifact noise signal in one of the unknown brain magnetic signals is shown.
And the same is to process a segment of eye movement artifact noise signal in the data 1, and fig. 11A to fig. 11C are diagrams illustrating the effect of the signal space projection method on removing the eye movement artifact noise signal. Fig. 11A is a schematic diagram of an eye movement artifact noise signal in original magnetoencephalogram data by using a signal space projection method, where the eye movement artifact noise signal lasts for approximately 0.4 second, fig. 11B is a schematic diagram of a signal obtained after the eye movement artifact noise signal is removed by using the signal space projection method, it can be seen that the eye movement artifact noise signal is completely removed, but the algorithm also affects other signal segments without the eye movement artifact noise signal, and fig. 11C is a schematic diagram of a signal difference between fig. 11A and fig. 11B. The visible signal space projection algorithm also has a large effect on the original brain magnetic signal that is not affected by the eye movement artifact noise signal, which is determined by the intrinsic nature of the signal space projection method.
The above embodiments are only intended to illustrate the technical solution of the present invention and not to limit the same, and a person skilled in the art can modify the technical solution of the present invention or substitute the same without departing from the spirit and scope of the present invention, and the scope of the present invention should be determined by the claims.
Claims (9)
1. A brain magnetic pattern eye movement artifact detection and elimination method based on a neural network comprises the following steps:
1) cutting the magnetoencephalogram signal to be detected to obtain a plurality of signal segments;
2) drawing a plurality of brain magnetic signal views according to the signal segments and the detection signal positions of the brain magnetic sensors;
3) extracting signal space distribution characteristics of a brain magnetic signal view, and classifying the signal space distribution characteristics to obtain whether each signal segment contains an eye movement artifact noise signal;
4) for the signal segment containing the eye movement artifact noise signal, acquiring a corresponding eye movement signal interference segment according to a fixed ratio and the eye movement electrical signal segment of the magnetoencephalogram signal segment containing the eye movement artifact noise signal at the corresponding time point;
5) subtracting the magnetoencephalogram signal segment containing the eye movement artifact noise signal from the corresponding eye movement signal interference segment to obtain an eye movement artifact noise removal signal segment, and reducing the eye movement artifact noise removal signal segment to the corresponding position of the magnetoencephalogram signal segment containing the eye movement artifact noise signal.
2. The method of claim 1, wherein the form of the signal segment comprises: and M-N two-dimensional matrix data, wherein M is the number of the brain magnetic sensors, and N is a set time length.
3. The method of claim 1, wherein the brain magnetic signal view is a 2D view; obtaining the brain magnetic signal view by the following steps:
1) calculating the average power of the detection signal of each sensor in each signal segment;
2) and coding the average power by using colors, and drawing the average power to a spatial position corresponding to the sensor to obtain a brain magnetic signal view.
4. The method as claimed in claim 1, characterized in that, through a trained two-dimensional deep convolutional neural network, the signal space distribution characteristics of the brain magnetic signal view are extracted and classified; wherein a two-dimensional deep convolutional neural network is trained by:
a) collecting sample magnetoencephalogram signal fragments, and setting a label according to whether an eye movement artifact noise signal is contained;
b) drawing a plurality of sample magnetoencephalography signal views according to the sample magnetoencephalography signal fragments and the corresponding magnetoencephalography sensor detection signal positions;
c) and training the two-dimensional deep convolution neural network through a plurality of sample brain magnetic signal views to obtain the trained two-dimensional deep convolution neural network.
5. The method of claim 4, wherein the two-dimensional deep convolutional neural network comprises a GooglLeNet network or an AlexNet network.
6. The method of claim 1, wherein the method of classifying the signal spatial distribution features comprises: a Softmax classifier was used.
7. The method of claim 1, wherein the fixed ratio is obtained by:
1) calculating the ratio of the sample magnetoencephalogram signal segment to the corresponding sample eye movement electric signal segment according to a plurality of sample magnetoencephalogram signal segments containing eye movement artifact noise signals and the sample eye movement electric signal segments at corresponding time points;
2) and calculating the average value of the ratios to obtain the fixed ratio.
8. A storage medium having a computer program stored thereon, wherein the computer program is arranged to, when run, perform the method of any of claims 1-7.
9. An electronic device comprising a memory having a computer program stored therein and a processor arranged to run the computer program to perform the method according to any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011026445.9A CN112220482B (en) | 2020-09-25 | 2020-09-25 | Method for detecting and eliminating magnetoencephalogram eye movement artifact based on neural network and electronic device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011026445.9A CN112220482B (en) | 2020-09-25 | 2020-09-25 | Method for detecting and eliminating magnetoencephalogram eye movement artifact based on neural network and electronic device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112220482A CN112220482A (en) | 2021-01-15 |
| CN112220482B true CN112220482B (en) | 2021-07-30 |
Family
ID=74108228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011026445.9A Active CN112220482B (en) | 2020-09-25 | 2020-09-25 | Method for detecting and eliminating magnetoencephalogram eye movement artifact based on neural network and electronic device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112220482B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113064490B (en) * | 2021-04-06 | 2022-07-29 | 上海金陵电子网络股份有限公司 | Eye movement track-based virtual enhancement equipment identification method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101474070A (en) * | 2009-01-21 | 2009-07-08 | 电子科技大学 | Method for removing ocular artifacts in brain-electrical signal |
| CN103717129A (en) * | 2011-05-24 | 2014-04-09 | 加利福尼亚大学董事会 | Magnetoencephalography source imaging |
| CN109009092A (en) * | 2018-06-15 | 2018-12-18 | 东华大学 | A method of removal EEG signals noise artefact |
| CN111340142A (en) * | 2020-05-14 | 2020-06-26 | 南京慧脑云计算有限公司 | Epilepsia magnetoencephalogram spike automatic detection method and tracing positioning system |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CU22179A1 (en) * | 1990-11-09 | 1994-01-31 | Neurociencias Centro | Method and system for evaluating abnormal electro-magnetic physiological activity of the heart and brain and plotting it in graph form. |
| US20080021340A1 (en) * | 2006-07-19 | 2008-01-24 | Mika Sarkela | Detection of focal epileptiform activity |
| CN110353672B (en) * | 2019-07-15 | 2022-02-22 | 西安邮电大学 | System and method for removing eye artifacts in electroencephalogram signals |
-
2020
- 2020-09-25 CN CN202011026445.9A patent/CN112220482B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101474070A (en) * | 2009-01-21 | 2009-07-08 | 电子科技大学 | Method for removing ocular artifacts in brain-electrical signal |
| CN103717129A (en) * | 2011-05-24 | 2014-04-09 | 加利福尼亚大学董事会 | Magnetoencephalography source imaging |
| CN109009092A (en) * | 2018-06-15 | 2018-12-18 | 东华大学 | A method of removal EEG signals noise artefact |
| CN111340142A (en) * | 2020-05-14 | 2020-06-26 | 南京慧脑云计算有限公司 | Epilepsia magnetoencephalogram spike automatic detection method and tracing positioning system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112220482A (en) | 2021-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110876626B (en) | Depression detection system based on optimal lead selection of multi-lead electroencephalogram | |
| Pizzagalli | Electroencephalography and high-density electrophysiological source localization | |
| Urigüen et al. | EEG artifact removal—state-of-the-art and guidelines | |
| Delorme et al. | Automatic artifact rejection for EEG data using high-order statistics and independent component analysis | |
| US9186083B2 (en) | Method and system for sleep stage determination | |
| Islam et al. | EEG mobility artifact removal for ambulatory epileptic seizure prediction applications | |
| Naghsh et al. | Spatial analysis of EEG signals for Parkinson’s disease stage detection | |
| Migliorelli et al. | Automated detection of epileptic ripples in MEG using beamformer-based virtual sensors | |
| Migliorelli et al. | SGM: a novel time-frequency algorithm based on unsupervised learning improves high-frequency oscillation detection in epilepsy | |
| Khalid et al. | Epileptic MEG spikes detection using amplitude thresholding and dynamic time warping | |
| CN107480635B (en) | Glance signal identification method and system based on bimodal classification model fusion | |
| CN112220482B (en) | Method for detecting and eliminating magnetoencephalogram eye movement artifact based on neural network and electronic device | |
| Bahador et al. | A correlation-driven mapping for deep learning application in detecting artifacts within the EEG | |
| Aarabi et al. | Detection of EEG transients in neonates and older children using a system based on dynamic time-warping template matching and spatial dipole clustering | |
| Zhao et al. | A multi-step blind source separation approach for the attenuation of artifacts in mobile high-density electroencephalography data | |
| CN111671421A (en) | Electroencephalogram-based children demand sensing method | |
| Yu et al. | A spatio-temporal filtering approach to denoising of single-trial ERP in rapid image triage | |
| CN112168167B (en) | Magnetoencephalogram single auditory evoked signal detection method and electronic device | |
| CN114861738A (en) | Electroencephalogram tracing and dipole selection-based motor imagery classification method | |
| Feng et al. | A new recognition method for the auditory evoked magnetic fields | |
| Hardani et al. | Analysis of emotional condition based on electrocardiogram signals | |
| Sanamdikar et al. | Classification of ECG Signal for Cardiac Arrhythmia Detection Using GAN Method | |
| Gao et al. | An adaptive joint CCA-ICA method for ocular artifact removal and its application to emotion classification | |
| Tian et al. | Modeling and reconstructing textile sensor noise: Implications for wearable technology | |
| Nancy et al. | A brain EEG classification system for the mild cognitive impairment analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |