CN111000557A - Noninvasive electroencephalogram signal analysis system applied to decompression skull operation - Google Patents
Noninvasive electroencephalogram signal analysis system applied to decompression skull operation Download PDFInfo
- Publication number
- CN111000557A CN111000557A CN201911244571.9A CN201911244571A CN111000557A CN 111000557 A CN111000557 A CN 111000557A CN 201911244571 A CN201911244571 A CN 201911244571A CN 111000557 A CN111000557 A CN 111000557A
- Authority
- CN
- China
- Prior art keywords
- layer
- electroencephalogram
- eeg
- groups
- data
- 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.)
- Granted
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
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
-
- 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
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- 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
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
- A61B5/726—Details of waveform analysis characterised by using transforms using Wavelet transforms
-
- 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
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Psychiatry (AREA)
- Animal Behavior & Ethology (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Signal Processing (AREA)
- Psychology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The utility model provides a be applied to noninvasive EEG analysis of ware of decompression skull postoperative system, including the receiving terminal who is used for acquireing final result, still including being used for gathering the EEG data of the person of being measured and carrying out the EEG acquisition front end of handling to and receive the EEG analysis and processing unit of the EEG data after the processing of EEG acquisition front end output through wireless transmission network, EEG analysis and processing unit is time-frequency map with EEG data conversion, and the rethread improved GoogleLeNet neural network model carries out analysis and processing to EEG data, obtains EEG data classification result, and will EEG data classification result send into receiving terminal through wireless transmission network. The invention can more accurately and stably detect and analyze the electroencephalogram signals of the patient after the decompression skull operation, and the obtained data classification probability can be more convenient for doctors to know the postoperative brain function rehabilitation condition of the patient and provide data support for brain function rehabilitation.
Description
Technical Field
The invention relates to a non-invasive electroencephalogram signal analysis system. In particular to a noninvasive electroencephalogram signal analysis system applied to decompression craniotomy.
Background
Electroencephalogram signals are widely concerned by people all the time as important indexes for researching human health. The brain-computer interface is a novel interactive system, plays a role of a bridge between a human body signal and external equipment, and people can obtain an electroencephalogram signal by using a brain-computer interface technology and then process the electroencephalogram signal to achieve the cognitive degree of brain functions. The brain-computer interface processing method has the advantages that the brain-computer signal data processing plays an important role in the application of the brain-computer interface technology, and the accuracy of the brain-computer interface processing method determines the stability of the brain-computer interface in some aspect.
Skull injury is a serious traumatic disease, and the disease has high disability rate and fatality rate due to the severity of the disease. Skull damage repair surgery is an important surgical formula for treating skull damage, but the surgery can cause skull damage patients to generate side effects such as intracranial pressure abnormity, and the side effects can cause the nerve functions of the patients to be greatly influenced, so people are always searching for a more accurate method to detect and analyze the electroencephalogram data of the patients after the skull damage surgery, and the electroencephalogram data is used as auxiliary data for diagnosis of doctors. With the development of computer science and artificial intelligence, people begin to apply a deep learning method to medical science, and because the method has high accuracy in electroencephalogram image identification and judgment, more and more medical workers begin to analyze human physiological data in a neural network mode, so that the effect of data quantification on human physiological indexes is achieved.
At present, data analysis on electroencephalogram signals after decompression craniotomy is less, and especially, the method is less in research on noninvasive electroencephalogram signal analysis systems applied after decompression craniotomy. The combination of decompression skull surgery and noninvasive electroencephalogram analysis belongs to the emerging field.
Disclosure of Invention
The invention aims to solve the technical problem of providing a noninvasive electroencephalogram signal analysis system applied to decompression craniotomy, which can provide basic data support for subsequent diagnosis of the brain function rehabilitation condition of a patient.
The technical scheme adopted by the invention is as follows: the utility model provides a be applied to noninvasive EEG analysis of ware of decompression skull postoperative system, including the receiving terminal who is used for acquireing final result, still including being used for gathering the EEG data of the person of being measured and carrying out the EEG acquisition front end of handling to and receive the EEG analysis and processing unit of the EEG data after the processing of EEG acquisition front end output through wireless transmission network, EEG analysis and processing unit is time-frequency map with EEG data conversion, and the rethread improved GoogleLeNet neural network model carries out analysis and processing to EEG data, obtains EEG data classification result, and will EEG data classification result send into receiving terminal through wireless transmission network.
The electroencephalogram signal acquisition front end comprises an electroencephalogram cap used for acquiring electroencephalograms of a detected person, an analog front end connected with an electroencephalogram cap signal output end and used for amplifying and filtering the acquired electroencephalograms of the detected person, an AD converter connected with a signal output end of the analog front end and used for converting the processed electroencephalograms output by the analog front end into electroencephalograms, and a microprocessor connected with an output end of the AD converter and used for controlling the working state of the AD converter, receiving the electroencephalograms output by the AD converter and sending the electroencephalograms into an electroencephalogram signal analysis and processing unit through a wireless transmission network.
The collecting electrode of the brain electric cap is 75 leads, and a connection mode of a single-stage lead method is selected, wherein,
the brain leads used are Fp1, Fp2, Fpz, AF3, AF4, AF7, AF8, AFz, F1, F2, F3, F4, F5, F6, F7, F8, F9, F10, Fz, FC1, FC2, FC3, FC4, FC5, FC6, FCz, FT7, FT8, FT9, FT10, C3, C4, C5, C6, T7, T8, T9, T10, A1, A2, CP3, CP4, CP5, CP6, C7, 8, 9, TP10, TP 46z;
the electroencephalogram cap takes the Cz lead as a reference point to obtain electroencephalogram signals at the prefrontal lobe and bilateral temporal lobes.
The analog front end comprises a front end amplifying circuit, a high-pass filter circuit with the cutoff frequency of 0.48Hz and a low-pass filter circuit with the cutoff frequency of 41Hz which are sequentially connected.
The electroencephalogram signal analysis and processing unit comprises an electroencephalogram signal processing unit, a training unit, a prediction unit, a time-frequency diagram conversion unit and a receiving terminal, wherein the electroencephalogram signal processing unit is used for carrying out electroencephalogram signal preprocessing on an improved GoogleNet network model for training, the training unit is used for receiving an electroencephalogram signal processed by the electroencephalogram signal processing unit so as to train the improved GoogleNet network model, the prediction unit is used for receiving the improved GoogleNet network model trained by the training unit and carrying out prediction and classification on an electroencephalogram digital signal of a testee, the time-frequency diagram conversion unit is used for receiving the electroencephalogram digital signal output by the microprocessor in the electroencephalogram signal acquisition front end and converting the electroencephalogram digital signal into an electroencephalogram signal time-frequency diagram to be sent to the prediction unit, and the electroencephalogram signal data classification result output by the.
The receiving terminal is a mobile terminal or a computer.
The improved GoogLeNet neural network model is characterized in that a residual block network is added on the basis of the GoogLeNet neural network model, and specifically comprises the following steps:
from the input layer to the first three groups of parallel structures, and then from the three groups of parallel structures to the first channel merging layer, the first three groups of parallel structures are respectively: a connected 1X1 convolutional layer and pooling layer; a 3X3 convolutional layer and a pooling layer connected; connecting the 5X5 convolutional layer and the pooling layer; the number of convolution kernels of the convolution layers is 32;
and uniformly outputting the data from the first channel merging layer to the second three groups of parallel structures and then from the second three groups of parallel structures to the second channel merging layer, wherein the second three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 32;
and the third three groups of parallel structures are uniformly output to the third channel merging layer from the second channel merging layer to the third three groups of parallel structures, wherein the third three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 64;
and uniformly outputting the three groups of parallel structures from the third channel merging layer to the fourth three groups of parallel structures and then outputting the four groups of parallel structures to the fourth channel merging layer, wherein the fourth three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 128;
and the third three groups of parallel structures are uniformly output to the fifth channel merging layer from the fourth channel merging layer to the fifth three groups of parallel structures, and the fifth three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 256;
and finally, outputting the data to the average pooling layer and the output layer in sequence.
The working process of the electroencephalogram signal analyzing and processing unit comprises the following steps:
1) acquiring abnormal electroencephalogram signals of a patient after clinical decompression craniotomy and normal electroencephalogram signals of a patient after the clinical decompression craniotomy from electroencephalogram signal data set, wherein the abnormal electroencephalogram signals and the normal electroencephalogram signals are both frontal lobe electroencephalogram signals and bilateral temporal lobe electroencephalogram signals;
2) filtering the abnormal electroencephalogram signal and the normal electroencephalogram signal by a three-order Butterworth band-pass filter at 0.48Hz-41 Hz; in order to ensure that the improved GoogleLeNet neural network model acquires the same data volume in the same time, resampling the abnormal electroencephalogram signal and the normal electroencephalogram signal which are obtained after filtering;
3) converting the signals into electroencephalogram time-frequency graphs by methods such as wavelet transformation and the like, and randomly dividing all the acquired electroencephalogram time-frequency graphs into a training set, a cross validation set and a test set;
4) designing an improved GoogLeNet neural network model, wherein the output of a residual error unit involved in the improved GoogLeNet neural network model is cascaded by a plurality of convolution layers, so that the output and input elements are added, and the output and input elements of the convolution layers are ensured to have the same dimension;
5) training an improved GoogLeNet neural network model by using a training set, a cross validation set and a test set, inputting a sample of the training set into an improved GoogLeNet neural network input layer for training, and performing forward propagation; transforming layer by layer and transmitting to an output layer of a residual error neural network; defining labels and rehabilitation categories by using a cross entropy loss function, updating the weight and deviation of each layer of the network through an Adam algorithm and back propagation, verifying and adjusting parameters of the improved GoogleLeNet neural network on a cross verification set, comparing a predicted result with the labels, if the predicted result is the same as the label, keeping the parameters unchanged, training after iteration to obtain optimal parameters, and immediately stopping training until the accuracy rate is converged; if not, adjusting the parameters of the model through a back propagation algorithm to obtain a trained improved GoogLeNet neural network model;
6) the electroencephalogram signal data output by the electroencephalogram signal acquisition front end are converted into a time-frequency diagram, the time-frequency diagram is sent into a trained improved GoogLeNet neural network model, electroencephalogram signal data classification results are obtained through layer-by-layer transformation, and the electroencephalogram signal data classification results are sent into a receiving terminal through a wireless transmission network.
Step 6), unifying the format of the picture data of the time-frequency graph as follows: 224, and then feed the picture data into a modified google net neural network for prediction.
According to the non-invasive electroencephalogram signal analysis system applied to decompression cranioplasty, after electroencephalogram signals are collected and processed through the portable electroencephalogram signal collection system, the electroencephalogram signals of a patient after decompression cranioplasty are detected and analyzed more accurately and stably by using an improved Google-ResNet neural network algorithm, and therefore the obtained data classification probability can be more convenient for a doctor to know the postoperative brain function rehabilitation condition of the patient and provide data support for brain function rehabilitation. The method has the following advantages:
(1) on the premise of ensuring the safety of the human body, the electroencephalogram signals of the patient can be more conveniently and accurately acquired, and basic data support is provided for the subsequent diagnosis of the brain function rehabilitation condition of the patient;
(2) after the neural network is applied to the technology of brain function detection after decompression of non-invasive electroencephalograms, the accuracy of classification of non-invasive electroencephalograms is effectively improved.
Drawings
FIG. 1 is a block diagram of a noninvasive EEG signal analysis system applied after decompression craniotomy according to the invention;
FIG. 2 is a channel profile employed by the brain cap of the present invention;
FIG. 3 is a flow chart of a non-invasive EEG signal analysis system applied after decompression craniotomy of the present invention;
FIG. 4 is a schematic diagram of a residual block network in the present invention;
fig. 5 is a schematic structural diagram of an improved google lenet neural network model in the invention.
Detailed Description
The following describes a non-invasive electroencephalogram signal analysis system applied after decompression craniotomy in detail with reference to the embodiments and the accompanying drawings.
According to the non-invasive electroencephalogram signal analysis system applied to decompression craniotomy, electroencephalogram signals of a patient after decompression craniotomy are collected and processed through the portable electroencephalogram signal collection system, and then the improved GoogleLeNet neural network algorithm is utilized, so that more accurate and stable data analysis is achieved, and the obtained data classification probability can be more convenient for a doctor to analyze the postoperative brain function rehabilitation condition of the patient.
As shown in fig. 1 and fig. 3, the non-invasive electroencephalogram analysis system applied after decompression craniotomy comprises a receiving terminal 3 for obtaining a final result, an electroencephalogram signal acquisition front end 1 for acquiring and processing electroencephalogram signal data of a tested person, and an electroencephalogram signal analysis processing unit 2 for receiving the processed electroencephalogram signal data output by the electroencephalogram signal acquisition front end 1 through a wireless transmission network, wherein the electroencephalogram signal analysis processing unit 2 converts the electroencephalogram signal data into a time-frequency diagram, the electroencephalogram signal data is analyzed and processed through an improved google neural network model to obtain an electroencephalogram signal data classification result, and the electroencephalogram signal data classification result is sent to the receiving terminal 3 through the wireless transmission network, and the receiving terminal 3 is a mobile terminal or a computer.
The electroencephalogram signal acquisition front end 1 comprises an electroencephalogram cap 1.1 used for acquiring electroencephalograms of a tested person, an analog front end 1.2 connected with a signal output end of the electroencephalogram cap 1.1 and used for amplifying and filtering the acquired electroencephalograms of the tested person, an AD converter 1.3 connected with a signal output end of the analog front end 1.2 and used for converting the processed electroencephalograms analog signals output by the analog front end 1.2 into electroencephalograms digital signals, and a microprocessor 1.4 connected with an output end of the AD converter 1.3 and used for controlling the working state of the AD converter 1.3, receiving the electroencephalograms digital signals output by the AD converter 1.3 and sending the electroencephalograms digital signals into an electroencephalogram signal analysis processing unit 2 through a wireless transmission network.
As shown in fig. 2, the collecting electrode of the electroencephalogram cap 1.1 is 75 leads, and a connection mode of a single-stage lead connection method is selected, wherein,
the brain leads used are Fp1, Fp2, Fpz, AF3, AF4, AF7, AF8, AFz, F1, F2, F3, F4, F5, F6, F7, F8, F9, F10, Fz, FC1, FC2, FC3, FC4, FC5, FC6, FCz, FT7, FT8, FT9, FT10, C3, C4, C5, C6, T7, T8, T9, T10, A1, A2, CP3, CP4, CP5, CP6, C7, 8, 9, TP10, TP 46z;
the electroencephalogram cap 1.1 takes the Cz lead as a reference point to obtain electroencephalogram signals at the forehead and bilateral temporal lobes.
The analog front end 1.2 comprises a front end amplifying circuit, a high-pass filter circuit with cut-off frequency of 0.48Hz and a low-pass filter circuit with cut-off frequency of 41Hz which are connected in sequence.
The AD converter 1.3 has higher signal-to-noise ratio and data acquisition rate. The wireless transmission network has low power consumption, small volume and low cost.
The electroencephalogram signal analyzing and processing unit 2 comprises an electroencephalogram signal processing unit 2.2 used for carrying out electroencephalogram signal preprocessing on an improved GoogleNet network model for training and adopting, a training unit 2.3 used for receiving an electroencephalogram signal processed by the electroencephalogram signal processing unit 2.2 and training the improved GoogleNet network model, a prediction unit 2.4 used for receiving the improved GoogleNet network model trained by the training unit 2.3 and carrying out prediction and classification on an electroencephalogram digital signal of a testee, a time-frequency diagram conversion unit 2.1 used for receiving the electroencephalogram digital signal output by a microprocessor 1.4 in the electroencephalogram signal acquisition front end 1, converting the electroencephalogram digital signal into an electroencephalogram signal and sending the electroencephalogram signal to the prediction unit 2.4, and a time-frequency diagram conversion unit 2.1 used for receiving an electroencephalogram signal data classification result 2.5 output by the prediction unit 2.4 and sending the electroencephalogram data classification result to a receiving terminal 3 through a wireless transmission network.
The improved GoogleLeNet neural network model is formed by adding a residual block network shown in figure 4 on the basis of the GoogleLeNet neural network model. As shown in fig. 5, the improved google lenet neural network model specifically includes:
from the input layer to the first three groups of parallel structures, and then from the three groups of parallel structures to the first channel merging layer, the first three groups of parallel structures are respectively: a connected 1X1 convolutional layer and pooling layer; a 3X3 convolutional layer and a pooling layer connected; connecting the 5X5 convolutional layer and the pooling layer; the number of convolution kernels of the convolution layers is 32;
and uniformly outputting the data from the first channel merging layer to the second three groups of parallel structures and then from the second three groups of parallel structures to the second channel merging layer, wherein the second three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 32;
and the third three groups of parallel structures are uniformly output to the third channel merging layer from the second channel merging layer to the third three groups of parallel structures, wherein the third three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 64;
and uniformly outputting the three groups of parallel structures from the third channel merging layer to the fourth three groups of parallel structures and then outputting the four groups of parallel structures to the fourth channel merging layer, wherein the fourth three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 128;
and the third three groups of parallel structures are uniformly output to the fifth channel merging layer from the fourth channel merging layer to the fifth three groups of parallel structures, and the fifth three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 256;
and finally, outputting the data to the average pooling layer and the output layer in sequence.
The working process of the electroencephalogram signal analyzing and processing unit 2 comprises the following steps:
1) acquiring abnormal electroencephalogram signals of a patient after clinical decompression craniotomy and normal electroencephalogram signals of a patient after the clinical decompression craniotomy from electroencephalogram signal data set, wherein the abnormal electroencephalogram signals and the normal electroencephalogram signals are both frontal lobe electroencephalogram signals and bilateral temporal lobe electroencephalogram signals;
2) filtering the abnormal electroencephalogram signal and the normal electroencephalogram signal by a three-order Butterworth band-pass filter at 0.48Hz-41 Hz; in order to ensure that the improved GoogleLeNet neural network model acquires the same data volume in the same time, resampling the abnormal electroencephalogram signal and the normal electroencephalogram signal which are obtained after filtering;
3) converting the signals into electroencephalogram time-frequency graphs by methods such as wavelet transformation and the like, and randomly dividing all the acquired electroencephalogram time-frequency graphs into a training set, a cross validation set and a test set;
4) an improved google lenet neural network model is designed that is different from conventional convolutional neural networks. In order to facilitate the modification of the neural network, the improved GoogLeNet neural network adopts a modular structure. The improved GoogLeNet neural network does not increase the calculation cost while expanding the depth and the width, and fully utilizes the internal resources of a computer; the output of a residual error unit related in the improved GoogLeNet neural network model is cascaded by a plurality of convolution layers, so that the output and input elements are added, and the output and input elements of the convolution layers are ensured to have the same dimension;
5) training an improved GoogLeNet neural network model by using a training set, a cross validation set and a test set, inputting a sample of the training set into an improved GoogLeNet neural network input layer for training, and performing forward propagation; the data are transformed layer by layer and then transmitted to an output layer of a residual block network; defining labels and rehabilitation categories by using a cross entropy loss function, updating the weight and deviation of each layer of the improved GoogleLeNet neural network through an Adam algorithm and back propagation, verifying and adjusting parameters of the improved GoogleLeNet neural network on a cross verification set, comparing a predicted result with the labels, if the predicted result is the same as the label, keeping the parameters unchanged, training after iteration to obtain optimal parameters, and immediately stopping training until the accuracy rate is converged; if not, adjusting parameters of the improved GoogLeNet neural network through a back propagation algorithm to obtain a trained improved GoogLeNet neural network model;
(1) in this example, the cross entropy loss function used is:
L＝-[ylogyl+(1-y)log(1-yl)]
wherein y is a real tag, ylIs a predicted value.
(2) The Adam algorithm in this embodiment is as follows:
s←ρ1s+(1-ρ1)g
γ←ρ2γ+(1-ρ2)g⊙g
where ρ is1、ρ2Are all constants (default value is ρ)1＝0.9、ρ20.999), g represents the first order gradient of the loss function, s represents the biased first order moment estimate, and γ represents the biased second order moment estimate.
Where ρ is1、ρ2Are all constants (default value is ρ)1＝0.9、ρ20.999), s stands for biased first order moment estimate, γ stands for biased second order moment estimate, slRepresenting deviations of the corrected first-order matrix, gammalRepresenting the deviation of the modified second moment.
Wherein, both epsilon and delta are constants (default values are epsilon 0.001 and delta 10)-8) And gamma represents the biased second moment estimate, slRepresenting the deviation of the modified first order matrix and delta theta representing the variation of the parameter theta.
6) The electroencephalogram signal data output by the electroencephalogram signal acquisition front end are converted into a time-frequency diagram, and the image data of the time-frequency diagram are uniformly formatted as follows: 224 and 224, sending the picture data into a trained improved GoogLeNet neural network model, obtaining an electroencephalogram data classification result through layer-by-layer transformation, and sending the classification result into a receiving terminal through a wireless transmission network.
Claims (9)
1. The utility model provides a be applied to noninvasive EEG analysis of ware of decompression skull postoperative system, including receiving terminal (3) that are used for obtaining final result, its characterized in that still including being used for gathering testee's EEG data and carrying out the EEG collection front end (1) of handling to and EEG analysis and processing unit (2) of the EEG data after the processing of receiving EEG collection front end (1) output through wireless transmission network, EEG analysis and processing unit (2) convert EEG data into time-frequency diagram, and the analytic processing is carried out to EEG data to the neural network model of rethread modified GoogleNet, obtains EEG data classification result, and will EEG data classification result send into receiving terminal (3) through wireless transmission network.
2. The noninvasive brain electrical signal analysis system for decompression craniotomy according to claim 1, it is characterized in that the electroencephalogram signal acquisition front end (1) comprises an electroencephalogram cap (1.1) for acquiring the electroencephalogram signal of a tested person, an analog front end (1.2) which is connected with the signal output end of the electroencephalogram cap (1.1) and is used for amplifying and filtering the acquired electroencephalogram signals of the tested person, an AD converter (1.3) which is connected with the signal output end of the analog front end (1.2) and is used for converting the processed electroencephalogram analog signal output by the analog front end (1.2) into an electroencephalogram digital signal, and the microprocessor (1.4) is connected with the output end of the AD converter (1.3) and is used for controlling the working state of the AD converter (1.3), receiving the electroencephalogram digital signals output by the AD converter (1.3) and sending the electroencephalogram digital signals to the electroencephalogram signal analysis processing unit (2) through a wireless transmission network.
3. The system for noninvasive brain electrical signal analysis after decompression craniotomy according to claim 2, characterized in that the collecting electrode of the brain electrical cap (1.1) is 75 lead, and the connection mode of single-stage lead method is selected, wherein,
the brain leads used are Fp1, Fp2, Fpz, AF3, AF4, AF7, AF8, AFz, F1, F2, F3, F4, F5, F6, F7, F8, F9, F10, Fz, FC1, FC2, FC3, FC4, FC5, FC6, FCz, FT7, FT8, FT9, FT10, C3, C4, C5, C6, T7, T8, T9, T10, A1, A2, CP3, CP4, CP5, CP6, C7, 8, 9, TP10, TP 46z;
the electroencephalogram cap (1.1) takes the Cz lead as a reference point to obtain the electroencephalogram signals at the forehead and bilateral temporal lobes.
4. The system for noninvasive electroencephalogram signal analysis after decompression craniotomy according to claim 2, wherein the analog front end (1.2) comprises a front end amplifying circuit, a high-pass filter circuit with a cut-off frequency of 0.48Hz and a low-pass filter circuit with a cut-off frequency of 41Hz which are connected in sequence.
5. The system of claim 1, wherein the EEG analysis processing unit (2) comprises an EEG signal processing unit (2.2) for preprocessing EEG signals used for training the improved GoogleNet network model, a training unit (2.3) for receiving the EEG signals processed by the EEG signal processing unit (2.2) and training the improved GoogleNet network model, a prediction unit (2.4) for receiving the improved GoogleNet network model trained by the training unit (2.3) and predicting and classifying the EEG digital signals of the testee, a time-frequency diagram conversion unit (2.1) for receiving the EEG digital signals output by the microprocessor (1.4) in the EEG signal acquisition front end (1) and converting the EEG digital signals into time-frequency diagrams of the EEG signals and sending the time-frequency diagrams into the prediction unit (2.4), and receiving the classification result (2.5) of the electroencephalogram signal data output by the prediction unit (2.4) and sending the classification result to a receiving terminal (3) through a wireless transmission network.
6. The system for noninvasive brain electrical signal analysis after decompression craniotomy according to claim 1 or 5, characterized in that the receiving terminal (3) is a mobile terminal or a computer.
7. The system of claim 1, wherein the improved google lenet neural network model is obtained by adding a residual block network on the basis of the google lenet neural network model, and specifically comprises:
from the input layer to the first three groups of parallel structures, and then from the three groups of parallel structures to the first channel merging layer, the first three groups of parallel structures are respectively: a connected 1X1 convolutional layer and pooling layer; a 3X3 convolutional layer and a pooling layer connected; connecting the 5X5 convolutional layer and the pooling layer; the number of convolution kernels of the convolution layers is 32;
and uniformly outputting the data from the first channel merging layer to the second three groups of parallel structures and then from the second three groups of parallel structures to the second channel merging layer, wherein the second three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 32;
and the third three groups of parallel structures are uniformly output to the third channel merging layer from the second channel merging layer to the third three groups of parallel structures, wherein the third three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 64;
and uniformly outputting the three groups of parallel structures from the third channel merging layer to the fourth three groups of parallel structures and then outputting the four groups of parallel structures to the fourth channel merging layer, wherein the fourth three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 128;
and the third three groups of parallel structures are uniformly output to the fifth channel merging layer from the fourth channel merging layer to the fifth three groups of parallel structures, and the fifth three groups of parallel structures are respectively as follows: a connected 1X1 convolutional layer and pooling layer; four 3X3 convolutional layers and a pooling layer which are connected in sequence; four 5X5 convolutional layers and a pooling layer which are connected in sequence; the number of convolution kernels of the convolution layer is also 256;
and finally, outputting the data to the average pooling layer and the output layer in sequence.
8. The system for analyzing the brain electrical signal without the wound after the decompression craniotomy as claimed in claim 5, wherein the working process of the brain electrical signal analyzing and processing unit (2) comprises the following steps:
1) acquiring abnormal electroencephalogram signals of a patient after clinical decompression craniotomy and normal electroencephalogram signals of a patient after the clinical decompression craniotomy from electroencephalogram signal data set, wherein the abnormal electroencephalogram signals and the normal electroencephalogram signals are both frontal lobe electroencephalogram signals and bilateral temporal lobe electroencephalogram signals;
2) filtering the abnormal electroencephalogram signal and the normal electroencephalogram signal by a three-order Butterworth band-pass filter at 0.48Hz-41 Hz; in order to ensure that the improved GoogleLeNet neural network model acquires the same data volume in the same time, resampling the abnormal electroencephalogram signal and the normal electroencephalogram signal which are obtained after filtering;
3) converting the signals into electroencephalogram time-frequency graphs by methods such as wavelet transformation and the like, and randomly dividing all the acquired electroencephalogram time-frequency graphs into a training set, a cross validation set and a test set;
4) designing an improved GoogLeNet neural network model, wherein the output of a residual error unit involved in the improved GoogLeNet neural network model is cascaded by a plurality of convolution layers, so that the output and input elements are added, and the output and input elements of the convolution layers are ensured to have the same dimension;
5) training an improved GoogLeNet neural network model by using a training set, a cross validation set and a test set, inputting a sample of the training set into an improved GoogLeNet neural network input layer for training, and performing forward propagation; transforming layer by layer and transmitting to an output layer of a residual error neural network; defining labels and rehabilitation categories by using a cross entropy loss function, updating the weight and deviation of each layer of the network through an Adam algorithm and back propagation, verifying and adjusting parameters of the improved GoogleLeNet neural network on a cross verification set, comparing a predicted result with the labels, if the predicted result is the same as the label, keeping the parameters unchanged, training after iteration to obtain optimal parameters, and immediately stopping training until the accuracy rate is converged; if not, adjusting the parameters of the model through a back propagation algorithm to obtain a trained improved GoogLeNet neural network model;
6) the electroencephalogram signal data output by the electroencephalogram signal acquisition front end are converted into a time-frequency diagram, the time-frequency diagram is sent into a trained improved GoogLeNet neural network model, electroencephalogram signal data classification results are obtained through layer-by-layer transformation, and the electroencephalogram signal data classification results are sent into a receiving terminal through a wireless transmission network.
9. The system for analyzing noninvasive brain electrical signals applied after decompression craniotomy according to claim 8, wherein the image data of the time-frequency diagram in step 6) is uniformly formatted as follows: 224, and then feed the picture data into a modified google net neural network for prediction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911244571.9A CN111000557B (en) | 2019-12-06 | 2019-12-06 | Noninvasive electroencephalogram signal analysis system applied to decompression skull operation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911244571.9A CN111000557B (en) | 2019-12-06 | 2019-12-06 | Noninvasive electroencephalogram signal analysis system applied to decompression skull operation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111000557A true CN111000557A (en) | 2020-04-14 |
| CN111000557B CN111000557B (en) | 2022-04-15 |
Family
ID=70113904
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911244571.9A Active CN111000557B (en) | 2019-12-06 | 2019-12-06 | Noninvasive electroencephalogram signal analysis system applied to decompression skull operation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111000557B (en) |
Citations (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070213629A1 (en) * | 2006-03-08 | 2007-09-13 | David Greene | Implantable seizure monitor |
| CN102715903A (en) * | 2012-07-09 | 2012-10-10 | 天津市人民医院 | Method for extracting electroencephalogram characteristic based on quantitative electroencephalogram |
| CN102824171A (en) * | 2012-07-16 | 2012-12-19 | 天津大学 | Method for extracting electroencephalogram features of patients suffering from PSD (post-stroke depression) |
| CN104887224A (en) * | 2015-05-29 | 2015-09-09 | 北京航空航天大学 | Epileptic feature extraction and automatic identification method based on electroencephalogram signal |
| CN105595961A (en) * | 2015-12-21 | 2016-05-25 | 天津大学 | Alzheimer's disease detecting system and method based on electroencephalogram signals |
| CN106909784A (en) * | 2017-02-24 | 2017-06-30 | 天津大学 | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks |
| CN106963373A (en) * | 2017-04-12 | 2017-07-21 | 博睿康科技（常州）股份有限公司 | A kind of electric adaptive filter method of brain |
| US20180085573A1 (en) * | 2016-09-27 | 2018-03-29 | National Guard Health Affairs | Skull implanted electrode assembly for brain stimulation |
| US20180146916A1 (en) * | 2015-10-23 | 2018-05-31 | Hrl Laboratories, Llc | Neural model-based controller |
| CN108433722A (en) * | 2018-02-28 | 2018-08-24 | 天津大学 | Portable brain electric collecting device and its application in SSVEP and Mental imagery |
| WO2018182827A1 (en) * | 2017-03-28 | 2018-10-04 | Hrl Laboratories, Llc | A neural model-based controller |
| US20180289311A1 (en) * | 2017-04-06 | 2018-10-11 | EPIC Neuro, Inc. | Eeg recording device |
| CN109711383A (en) * | 2019-01-07 | 2019-05-03 | 重庆邮电大学 | Convolutional neural networks Mental imagery EEG signal identification method based on time-frequency domain |
| CN110058691A (en) * | 2019-04-18 | 2019-07-26 | 西安交通大学 | Based on Embedded wearable wireless dry electrode brain wave acquisition processing system and method |
| CN110236536A (en) * | 2019-06-04 | 2019-09-17 | 电子科技大学 | A kind of brain electricity high-frequency oscillation signal detection system based on convolutional neural networks |
| CN110236533A (en) * | 2019-05-10 | 2019-09-17 | 杭州电子科技大学 | Epileptic seizure prediction method based on the study of more deep neural network migration features |
| CN110302462A (en) * | 2019-08-06 | 2019-10-08 | 张铎 | A kind of neural feedback training system based on EEG signals |
| CN110321856A (en) * | 2019-07-08 | 2019-10-11 | 山东大学 | A kind of brain-machine interface method and device of the multiple dimensioned divergence CSP of time-frequency |
| CN110353673A (en) * | 2019-07-16 | 2019-10-22 | 西安邮电大学 | A kind of brain electric channel selection method based on standard mutual information |
-
2019
- 2019-12-06 CN CN201911244571.9A patent/CN111000557B/en active Active
Patent Citations (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070213629A1 (en) * | 2006-03-08 | 2007-09-13 | David Greene | Implantable seizure monitor |
| CN102715903A (en) * | 2012-07-09 | 2012-10-10 | 天津市人民医院 | Method for extracting electroencephalogram characteristic based on quantitative electroencephalogram |
| CN102824171A (en) * | 2012-07-16 | 2012-12-19 | 天津大学 | Method for extracting electroencephalogram features of patients suffering from PSD (post-stroke depression) |
| CN104887224A (en) * | 2015-05-29 | 2015-09-09 | 北京航空航天大学 | Epileptic feature extraction and automatic identification method based on electroencephalogram signal |
| US20180146916A1 (en) * | 2015-10-23 | 2018-05-31 | Hrl Laboratories, Llc | Neural model-based controller |
| CN105595961A (en) * | 2015-12-21 | 2016-05-25 | 天津大学 | Alzheimer's disease detecting system and method based on electroencephalogram signals |
| US20180085573A1 (en) * | 2016-09-27 | 2018-03-29 | National Guard Health Affairs | Skull implanted electrode assembly for brain stimulation |
| CN106909784A (en) * | 2017-02-24 | 2017-06-30 | 天津大学 | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks |
| WO2018182827A1 (en) * | 2017-03-28 | 2018-10-04 | Hrl Laboratories, Llc | A neural model-based controller |
| US20180289311A1 (en) * | 2017-04-06 | 2018-10-11 | EPIC Neuro, Inc. | Eeg recording device |
| CN106963373A (en) * | 2017-04-12 | 2017-07-21 | 博睿康科技（常州）股份有限公司 | A kind of electric adaptive filter method of brain |
| CN108433722A (en) * | 2018-02-28 | 2018-08-24 | 天津大学 | Portable brain electric collecting device and its application in SSVEP and Mental imagery |
| CN109711383A (en) * | 2019-01-07 | 2019-05-03 | 重庆邮电大学 | Convolutional neural networks Mental imagery EEG signal identification method based on time-frequency domain |
| CN110058691A (en) * | 2019-04-18 | 2019-07-26 | 西安交通大学 | Based on Embedded wearable wireless dry electrode brain wave acquisition processing system and method |
| CN110236533A (en) * | 2019-05-10 | 2019-09-17 | 杭州电子科技大学 | Epileptic seizure prediction method based on the study of more deep neural network migration features |
| CN110236536A (en) * | 2019-06-04 | 2019-09-17 | 电子科技大学 | A kind of brain electricity high-frequency oscillation signal detection system based on convolutional neural networks |
| CN110321856A (en) * | 2019-07-08 | 2019-10-11 | 山东大学 | A kind of brain-machine interface method and device of the multiple dimensioned divergence CSP of time-frequency |
| CN110353673A (en) * | 2019-07-16 | 2019-10-22 | 西安邮电大学 | A kind of brain electric channel selection method based on standard mutual information |
| CN110302462A (en) * | 2019-08-06 | 2019-10-08 | 张铎 | A kind of neural feedback training system based on EEG signals |
Non-Patent Citations (8)
| Title |
|---|
| A. M. TAQI, F. AL-AZZO, M. MARIOFANNA AND J. M. AL-SAADI: "Classification and discrimination of focal and non-focal EEG signals based on deep neural network", 《2017 INTERNATIONAL CONFERENCE ON CURRENT RESEARCH IN COMPUTER SCIENCE AND INFORMATION TECHNOLOGY (ICCIT)》 * |
| CHUNFANG WANG,YUANYUAN CHEN,CHANGCHENG SUN,YING ZHANG,DONG MING,: "Electrophysiological changes in poststroke subjects with depressed mood: A quantitative EEG study", 《INTERNATIONAL JOURNAL OF GERIATRIC PSYCHIATRY》 * |
| S. ZHANG ET AL.: "Analysis and Classification for Single-Trial EEG Induced by Sequential Finger Movements", 《2019 41ST ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY (EMBC)》 * |
| X. WEI, W. WANG, L. CAI, M. LU, G. YI AND J. WANG: "Parametric Bispectral Estimation of EEG Signals in Disorders of Consciousness", 《2019 CHINESE CONTROL CONFERENCE (CCC)》 * |
| ZHANG, Y., WANG, C., SUN, C., ZHANG, X.,WANG, Y.,QI, H.,MING,D.: "Neural complexity in patients with poststroke depression: A resting EEG study", 《JOURNAL OF AFFECTIVE DISORDERS》 * |
| 张逵,郭俐,谭雪梅,张蓉: "定量脑电图对颅骨重建患者脑功能康复的评估研究", 《四川医学》 * |
| 赖春任: "颞叶癫痫患者的脑部MR影像多特征分析与自动识别", 《中国优秀硕士学位论文全文数据库》 * |
| 陈龙,王仲朋,顾斌,张磊,刘爽,明东: "基于脑电控制的功能性电刺激康复系统设计及应用", 《航天医学与医学工程》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111000557B (en) | 2022-04-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Panda et al. | Classification of EEG signal using wavelet transform and support vector machine for epileptic seizure diction | |
| CN104173043B (en) | Electrocardiogram (ECG) data analysis method suitable for mobile platform | |
| CN111616682B (en) | Epileptic seizure early warning system based on portable electroencephalogram acquisition equipment and application | |
| AU2012284246A1 (en) | Systems and methods for the physiological assessment of brain health and the remote quality control of EEG systems | |
| CN105595961A (en) | Alzheimer's disease detecting system and method based on electroencephalogram signals | |
| CN103989485A (en) | Human body fatigue evaluation method based on brain waves | |
| CN111616681B (en) | Anesthesia state monitoring system based on portable electroencephalogram acquisition equipment and deep learning | |
| CN110101384B (en) | Functional network analysis system and analysis method for complex network | |
| CN113274011A (en) | Brain function monitoring and management method integrating electroencephalogram and functional near infrared spectrum technology | |
| CN116269249B (en) | Cerebral apoplexy risk prediction method and system | |
| Djamal et al. | Significant variables extraction of post-stroke EEG signal using wavelet and SOM kohonen | |
| CN115349861A (en) | Mental stress detection system and method based on single-channel electroencephalogram signal | |
| Arif et al. | Design of EEG data acquisition system based on Raspberry Pi 3 for acute ischemic stroke identification | |
| CN112200221B (en) | Epilepsy prediction system and method based on electrical impedance imaging and electroencephalogram signals | |
| CN111000557B (en) | Noninvasive electroencephalogram signal analysis system applied to decompression skull operation | |
| CN111543983B (en) | Electroencephalogram signal channel selection method based on neural network | |
| CN110537907B (en) | Electrocardiosignal compression and identification method based on singular value decomposition | |
| CN111481193A (en) | Fall risk assessment and early warning method and system | |
| CN102813513A (en) | Non-invasive intracranial pressure measurement method and measurement instrument | |
| CN115486819B (en) | Method, system and device for multi-cascade detection and quantification of sensory-perceptual neural pathway | |
| CN112274145A (en) | Method and device for processing near-infrared brain function imaging data and storage medium | |
| Radhakrishnan et al. | Investigating EEG Signals of Autistic Individuals Using Detrended Fluctuation Analysis. | |
| Fathi et al. | An efficient compression technique for Foetal phonocardiogram signals in remote healthcare monitoring systems | |
| CN113080892A (en) | Detection data processing method and system for risk prediction of cardiovascular and cerebrovascular diseases | |
| CN206950156U (en) | A kind of detection means that disease risks are judged by human-body biological electromagnetic wave |
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 |