CN109009102B - Electroencephalogram deep learning-based auxiliary diagnosis method and system - Google Patents
Electroencephalogram deep learning-based auxiliary diagnosis method and system Download PDFInfo
- Publication number
- CN109009102B CN109009102B CN201810909558.XA CN201810909558A CN109009102B CN 109009102 B CN109009102 B CN 109009102B CN 201810909558 A CN201810909558 A CN 201810909558A CN 109009102 B CN109009102 B CN 109009102B
- Authority
- CN
- China
- Prior art keywords
- electroencephalogram
- word embedding
- data
- preset
- deep learning
- 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
- A61B5/24—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
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
-
- 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
-
- 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
- A61B5/7271—Specific aspects of physiological measurement analysis
-
- 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
- A61B5/7271—Specific aspects of physiological measurement analysis
- A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
Abstract
The invention provides an electroencephalogram deep learning-based auxiliary diagnosis method and system, which are used for solving the problem of low accuracy of epilepsy diagnosis and comprise the following steps: s10: acquiring collected electroencephalogram sample data, and integrating the electroencephalogram sample data into a preset standardized model to obtain standardized electroencephalogram integer data; s20: converting the normalized electroencephalogram integer data into word embedding vectors according to a preset word embedding model; s30: extracting the characteristics of the word embedding vectors according to a preset deep learning model, and carrying out time marking and identification diagnosis on the extracted characteristics; s40: and outputting the disease attack probability according to the time mark and the identification diagnosis, and distinguishing electroencephalogram sample data with the disease attack probability exceeding a preset probability. The electroencephalogram of the patient is automatically diagnosed through the training model, the time region of the epileptic seizure in the electroencephalogram is automatically identified and marked, meanwhile, the ill probability is given, the working efficiency of a clinician is reduced, and the diagnosis efficiency is improved.
Description
Technical Field
The invention relates to the technical field of disease diagnosis, in particular to an auxiliary diagnosis method and system based on electroencephalogram deep learning.
Background
Epilepsy is transient dysfunction of the brain caused by paroxysmal abnormal over-discharge of cerebral neurons, the annual incidence is high, the detection and identification of electroencephalogram signals are the most important means for diagnosing epilepsy, electroencephalogram is a graph obtained by collecting weak bioelectricity generated by the human brain at the scalp through an electroencephalogram scanner of a medical instrument and amplifying and recording, and the graph is the main means and basis for a clinician to diagnose various related diseases in the brain, particularly epilepsy or epileptic seizure.
The recorded electric signal artifact which is not originated from the brain can be composed of a plurality of factors such as eye muscle activity, electroencephalogram electrode poor contact, swallowing action, head displacement and the like, because the electroencephalogram data always has artifact which is similar to electroencephalogram characteristics in epileptic seizure, the artifact and epileptic seizure characteristics are distinguished, and the method is a great difficulty for detecting the epileptic.
The patent with publication number CN106874694A discloses an intelligent diagnosis system for epilepsia electroencephalogram signal identification, which is characterized in that: the system comprises a background electroencephalogram digital signal processing program and a visual client operation interface; the characteristic extraction module of the background electroencephalogram digital signal processing program utilizes a discrete wavelet transform and a statistical method to extract characteristic quantities in original electroencephalogram signal segments, the model training module utilizes a radial basis function neural network and a minimum maximum probability mechanism to build a reliable epilepsy diagnosis model, and the trained diagnosis module utilizes the radial basis function neural network and a classification decision tree to diagnose newly input electroencephalogram signals; in the visual client operation interface, a data reading module is responsible for reading original electroencephalogram signal fragment data, a data communication module is responsible for initiating a request to a background and sending data information to wait for receiving the response and the data information of the background, and a data presentation module is responsible for displaying the original electroencephalogram signal fragments and the extracted electroencephalogram signal characteristic quantity in a chart mode and finally displaying a diagnosis result returned by the background. The system can automatically detect and diagnose the epileptic disease and improve the diagnosis efficiency of doctors, but the system cannot give accurate identification results and the probability of the epileptic disease and has poor diagnosis accuracy.
Disclosure of Invention
The invention aims to provide an electroencephalogram deep learning-based auxiliary diagnosis method and system, which are used for solving the problems of low accuracy and low diagnosis efficiency of automatic diagnosis of epileptic diseases.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
an assistant diagnosis method based on electroencephalogram deep learning comprises the following steps:
s10: acquiring collected electroencephalogram sample data, and integrating the electroencephalogram sample data into a preset standardized model to obtain standardized electroencephalogram integer data;
s20: converting the normalized electroencephalogram integer data into word embedding vectors according to a preset word embedding model;
s30: extracting the characteristics of the word embedding vector according to a preset deep learning model, and carrying out time marking and identification diagnosis on the extracted characteristics;
s40: and outputting the disease attack probability according to the time mark and the identification diagnosis, and distinguishing electroencephalogram sample data of which the disease attack probability exceeds a preset probability.
Further, the step S10 specifically includes the steps of:
s101: converting the electroencephalogram sample data into a two-dimensional matrix according to the number of the collected electrodes and the recording duration;
s102: performing pooling treatment on the two-dimensional matrix to obtain a pooled matrix after pooling;
s103: and carrying out u-law nonlinear compression transformation on the pooled matrix to obtain electroencephalogram integer data.
Further, the step S20 specifically includes the steps of:
s201: according to the recording time sequence, taking the electroencephalogram integer data as a corpus, and performing word prediction supervised learning through a word bag model and a jump model to form a word embedding matrix;
s202: and converting the electroencephalogram integer data to obtain a three-dimensional data matrix, and multiplying the three-dimensional data matrix by the word embedding matrix to obtain the word embedding vector.
Further, the step S30 specifically includes the steps of:
s301: performing multilayer superposition on the three-dimensional convolution layer to serve as a feature extraction layer;
s302: performing feature extraction on the word embedding vector through the feature extraction layer to obtain electroencephalogram data features;
s303: overlaying the attention layer and the recurrent neural network and time-stamping and identifying the electroencephalogram data features.
Further, the method also comprises the following steps:
s11: and carrying out supervision training on the preset word embedding module and the preset deep learning module through a preset disease attack database, wherein the preset disease attack database comprises a absence attack database, a tonic attack database, a generalized nonspecific attack database, a clonic attack database, a focus nonspecific attack database, a tonic clonic attack database, a simple partial attack database, a dystonic attack database, a complex partial attack database and a myoclonic attack database.
An aided diagnosis system based on electroencephalogram deep learning, comprising:
an input normalization module: the electroencephalogram sampling module is used for acquiring acquired electroencephalogram sample data, and integrating the electroencephalogram sample data into a preset standardized model to obtain standardized electroencephalogram integer data;
a word embedding module: the electroencephalogram normalization module is used for converting the normalized electroencephalogram integer data into word embedding vectors according to a preset word embedding model;
a deep learning module: the system is used for extracting the features of the word embedding vectors according to a preset deep learning model, and carrying out time marking and identification diagnosis on the extracted features;
an output module: the electroencephalogram sample data is used for outputting the disease attack probability according to the time mark and the identification diagnosis, and distinguishing the electroencephalogram sample data with the disease attack probability exceeding a preset probability.
Further, the system specifically includes:
a two-dimensional matrix conversion unit: the electroencephalogram sample data are converted into a two-dimensional matrix according to the collected electrode number and the recording duration;
a pooling treatment unit: the system is used for pooling the two-dimensional matrix to obtain a pooled matrix after pooling;
a compression transformation unit: and the method is used for carrying out u-law nonlinear compression transformation on the pooled matrix to obtain electroencephalogram integer data.
Further, the system specifically includes:
word embedding matrix unit: the electroencephalogram integer data are used as a corpus according to the recording time sequence, and word prediction supervised learning is carried out through a word bag model and a jump model to form a word embedding matrix;
word embedding vector unit: and the three-dimensional data matrix is obtained by converting the electroencephalogram integer data and is multiplied by the word embedding matrix to obtain the word embedding vector.
Further, the system specifically includes:
a superimposing unit: the three-dimensional convolution layer is used for carrying out multilayer superposition to be used as a characteristic extraction layer;
a feature extraction unit: the electroencephalogram data extraction layer is used for extracting the features of the word embedding vectors to obtain electroencephalogram data features;
a mark recognition unit: the electroencephalograph data processing system is used for overlaying an attention layer and a recurrent neural network, time marking the electroencephalograph data characteristics and identifying diagnosis.
Further, the method also comprises the following steps:
a supervision training module: the system comprises a preset word embedding module and a preset deep learning module, and is used for carrying out supervision training on the preset word embedding module and the preset deep learning module through a preset disease outbreak database, wherein the preset disease outbreak database comprises a absence outbreak database, a tonic outbreak database, a generalized nonspecific outbreak database, a clonic outbreak database, a focus nonspecific outbreak database, a tonic clonic outbreak database, a simple partial outbreak database, a hypotonia outbreak database, a complex partial outbreak database and a myoclonic outbreak database.
By adopting the method and the system, the preset seizure database is utilized to supervise and train each model, so that the epileptic seizures possibly existing in the electroencephalogram can be marked and the potential illness probability can be output, thereby quickly providing diagnosis basis and diagnosis suggestions for clinicians, improving the diagnosis efficiency of the epileptic diseases, and ensuring the diagnosis accuracy through supervised training.
Drawings
Fig. 1 is a flowchart of an assistant diagnosis method based on electroencephalogram deep learning according to an embodiment of the present invention;
FIG. 2 is a flowchart of an assistant diagnosis method based on electroencephalogram deep learning according to an embodiment of the present invention;
FIG. 3 is a flowchart of an assistant diagnosis method based on electroencephalogram deep learning according to an embodiment of the present invention;
FIG. 4 is a flowchart of an assistant diagnosis method based on electroencephalogram deep learning according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating an auxiliary diagnosis system based on deep electroencephalogram learning according to an embodiment of the present invention;
fig. 6 is a flowchart illustrating an assistant diagnosis method based on electroencephalogram deep learning according to another embodiment of the present invention.
Detailed Description
The following are specific embodiments of the present invention and are further described with reference to the drawings, but the present invention is not limited to these embodiments.
Example one
Referring to fig. 1, the present embodiment provides an aided diagnosis method based on electroencephalogram deep learning, including:
s10: acquiring collected electroencephalogram sample data, and integrating the electroencephalogram sample data into a preset standardized model to obtain standardized electroencephalogram integer data;
s20: converting the normalized electroencephalogram integer data into word embedding vectors according to a preset word embedding model;
s30: extracting the characteristics of the word embedding vector according to a preset deep learning model, and carrying out time marking and identification diagnosis on the extracted characteristics;
s40: and outputting the disease attack probability according to the time mark and the identification diagnosis, and distinguishing electroencephalogram sample data of which the disease attack probability exceeds a preset probability.
In this application, diagnose the disease personnel through corresponding medical equipment, be about to corresponding diagnosis step and system set up in medical equipment, carry out automatic diagnosis to the disease personnel through medical equipment, wherein, the disease in this application is epilepsy.
In step S10, the electroencephalogram sample image is an electroencephalogram of a scalp of a patient with epilepsy diagnosis, wherein the scalp electroencephalogram can be obtained by collecting weak bioelectricity generated by the brain of a human body at the scalp through an electroencephalogram scanner of a medical instrument, and the collected weak bioelectricity is enlarged and recorded to form a graph as an electroencephalogram of the scalp.
The preset normalized model is used for integrating electroencephalogram data obtained by scanning of the scanning instrument into a normalized input matrix format of the model according to the format and the numerical value.
In this embodiment, step S10 further includes the following steps:
s101: converting the electroencephalogram sample data into a two-dimensional matrix according to the number of the collected electrodes and the recording duration;
s102: performing pooling treatment on the two-dimensional matrix to obtain a pooled matrix after pooling;
s103: and carrying out u-law nonlinear compression transformation on the pooled matrix to obtain electroencephalogram integer data.
In step S101, when acquiring an electroencephalogram, a plurality of electrodes are mounted on the scalp to acquire brain wave signals of different parts of the brain scalp, so as to ensure accuracy of acquired data, and recording duration is the acquisition duration corresponding to each electrode in the process of acquiring brain waves. Among them, the brain waves of epilepsy and epileptic seizure are mainly manifested as acute waves.
And converting the original electroencephalogram data transmitted by the scanner into a two-dimensional matrix according to the number of electrodes collected on the scalp and the recording duration through a preset normalized model.
In step S102, in the convolutional neural network, a pooling operation is used to reduce the feature vectors output by the convolutional layer and improve the output result, and the general pooling operation includes random pooling, average pooling and maximum pooling. After the two-dimensional matrix is subjected to the pooling operation, the two-dimensional matrix is converted into a 16 × N matrix, namely the pooling matrix, wherein N is the time recording data length of 100 HZ.
In step S103, u-law nonlinear compression transformation is performed on the pooled matrix so that each data point is converted into 0-256 shaped data, i.e., electroencephalogram shaping data.
In step S20, a natural language processing algorithm is used when a word embedding model is preset, and the word embedding model is obtained by supervised learning training using normalized database data as a dictionary vector, and is used to transform electroencephalogram database data into a word embedding vector with features at a low latitude.
In this embodiment, step S20 specifically includes:
s201: according to the recording time sequence, taking the electroencephalogram integer data as a corpus, and performing word prediction supervised learning through a word bag model and a jump model to form a word embedding matrix;
s202: and converting the electroencephalogram integer data to obtain a three-dimensional data matrix, and multiplying the three-dimensional data matrix by the word embedding matrix to obtain the word embedding vector.
In step S201, the time sequence is recorded for recording brain waves for each electrode, for example, a plurality of electrodes are provided on the scalp, and the recording start time and the recording time length of different electrodes are different from each other.
And taking 0-256 electroencephalogram integer data as a corpus according to the recording time sequence, performing word prediction supervised learning through a word bag model and a jump model, and learning to obtain a 16-dimensional word embedding vector to form a 257 x 16 word embedding matrix.
In step S202, the 0-256 electroencephalogram integer data points in step S103 are converted into 257-dimensional one-hot vectors according to size, and are multiplied by the word embedding matrix in step S201, so as to obtain electroencephalogram feature data after the output word is embedded, that is, word embedding vectors.
The one-hot is a common sample marking method, for samples with a plurality of different categories, one category or a combination of a plurality of categories can be represented by an n-dimensional vector, and when the vector represents a plurality of categories, which combination of categories the vector represents can be determined according to the positions of elements in the vector.
In step S30, the preset deep learning model accepts a preset word embedding model, which is obtained by performing artificial labeling supervised learning training on electroencephalogram data converted into word embedding vectors, and is used for further feature extraction of the word embedding vectors, and performing time labeling and recognition diagnosis according to a recurrent neural network.
Wherein, step S30 specifically includes the steps of:
s301: performing multilayer superposition on the three-dimensional convolution layer to serve as a feature extraction layer;
s302: performing feature extraction on the word embedding vector through the feature extraction layer to obtain electroencephalogram data features;
s303: overlaying the attention layer and the recurrent neural network and time-stamping and identifying the electroencephalogram data features.
In step S301, the three-dimensional convolution layer is stacked into a plurality of layers, wherein the plurality of layers is at least three layers.
The preset deep learning model adopts an encoder-decoder architecture, and outputs the epileptic seizure marking probability of 0-1 for each input time data node.
In this embodiment, the method further includes the steps of:
s11: and carrying out supervision training on the preset word embedding module and the preset deep learning module through a preset disease attack database, wherein the preset disease attack database comprises a absence attack database, a tonic attack database, a generalized nonspecific attack database, a clonic attack database, a focus nonspecific attack database, a tonic clonic attack database, a simple partial attack database, a dystonic attack database, a complex partial attack database and a myoclonic attack database.
In this embodiment, the preset word embedding model and the preset deep learning model are supervised and trained through preset seizure data, wherein the preset seizure database in this embodiment includes a plurality of different seizure databases, so that each model can be accurately supervised and trained, and it is ensured that epilepsy or seizures can be accurately identified or diagnosed.
When the deep learning model is trained, a focus loss function is adopted to solve the problem of data nonuniformity, wherein seizure marks in the epileptic seizure data set occupy less data.
The convolutional neural network adopts an exponential linear activation function, adopts Adam optimization and random inactivation and L2 regularization to prevent overfitting, adopts specificity as a first metric index, sensitivity as a second metric index and F1 score as an evaluation index, is used for reducing error diagnosis or marking, and improves sensitivity on the premise of higher specificity.
In this embodiment, the predetermined probability in step S40 is fifty percent, i.e., 0.5.
Counting epileptic seizure marks of time nodes output by a coder-decoder of a preset deep learning model, marking a part with a mark probability larger than 0.5 as 1 on the basis of the input of an original electroencephalogram, and distinguishing the part with different colors, wherein the embodiment adopts red, and simultaneously, according to the output value of each recursion node, a fully-connected network outputs a 0-1 sickness probability through a sigmoid activation function.
By carrying out supervision training on each model by utilizing a preset seizure database, the epileptic seizure possibly existing in an electroencephalogram can be marked and potential illness probability can be output, so that diagnosis basis and diagnosis advice can be quickly and accurately provided for a clinician, and the diagnosis efficiency of the epilepsia is improved.
Example two
Referring to fig. 2, the present embodiment provides an aided diagnosis system based on electroencephalogram deep learning, including:
input normalization module 21: the electroencephalogram sampling module is used for acquiring acquired electroencephalogram sample data, and integrating the electroencephalogram sample data into a preset standardized model to obtain standardized electroencephalogram integer data;
the word embedding module 22: the electroencephalogram normalization module is used for converting the normalized electroencephalogram integer data into word embedding vectors according to a preset word embedding model;
the deep learning module 23: the system is used for extracting the features of the word embedding vectors according to a preset deep learning model, and carrying out time marking and identification diagnosis on the extracted features;
the output module 24: the electroencephalogram sample data is used for outputting the disease attack probability according to the time mark and the identification diagnosis, and distinguishing the electroencephalogram sample data with the disease attack probability exceeding a preset probability.
In this embodiment, the word embedding module 22 includes a preset word embedding model conversion layer frame, a stored word embedding matrix, and a port for data input and output of the word embedding module 22, and the deep learning module 23 includes a preset deep learning network structure frame, a stored network weight matrix of each layer, and a port for data input and output of the deep learning module 23.
In this embodiment, the method further includes:
the supervised training module 25: the system comprises a preset word embedding module and a preset deep learning module, and is used for carrying out supervision training on the preset word embedding module and the preset deep learning module through a preset disease outbreak database, wherein the preset disease outbreak database comprises a absence outbreak database, a tonic outbreak database, a generalized nonspecific outbreak database, a clonic outbreak database, a focus nonspecific outbreak database, a tonic clonic outbreak database, a simple partial outbreak database, a hypotonia outbreak database, a complex partial outbreak database and a myoclonic outbreak database.
In this embodiment, the input normalization module 21 further includes:
the two-dimensional matrix conversion unit 211: the electroencephalogram sample data are converted into a two-dimensional matrix according to the collected electrode number and the recording duration;
the pooling processing unit 212: the system is used for pooling the two-dimensional matrix to obtain a pooled matrix after pooling;
compression conversion unit 213: and the method is used for carrying out u-law nonlinear compression transformation on the pooled matrix to obtain electroencephalogram integer data.
In this embodiment, the word embedding module 22 includes:
word embedding matrix unit 221: the electroencephalogram integer data are used as a corpus according to the recording time sequence, and word prediction supervised learning is carried out through a word bag model and a jump model to form a word embedding matrix;
word embedding vector unit 222: and the three-dimensional data matrix is obtained by converting the electroencephalogram integer data and is multiplied by the word embedding matrix to obtain the word embedding vector.
In this embodiment, the deep learning module 23 further includes:
the superimposing unit 231: the three-dimensional convolution layer is used for carrying out multilayer superposition to be used as a characteristic extraction layer;
the feature extraction unit 232: the electroencephalogram data extraction layer is used for extracting the features of the word embedding vectors to obtain electroencephalogram data features;
the mark recognition unit 233: the electroencephalograph data processing system is used for overlaying an attention layer and a recurrent neural network, time marking the electroencephalograph data characteristics and identifying diagnosis.
Through mutual cooperation among all modules, the time region of epileptic seizure in the electroencephalogram can be identified through rapid analysis of the input electroencephalogram data, and meanwhile, the potential illness probability is given, so that the diagnosis efficiency is improved.
EXAMPLE III
Referring to fig. 3, the present embodiment provides a flowchart of an aided diagnosis method based on electroencephalogram deep learning, in which a patient is scanned by a scanner to obtain an electroencephalogram sample image, and the format and the numerical value of data input by the scanner are integrated into a normalized model by an input normalization module to obtain an input matrix format; converting and compressing electroencephalogram database data into word embedding vectors with characteristics at low latitudes through a word embedding model; marking, identifying and diagnosing the word embedding vectors through a deep learning model; finally, a seizure mark of 0 or 1 and a disease probability are given by an output module, wherein 0 represents no seizure and 1 represents seizure.
Taking this embodiment as an example, 118 patients were collected with training set seizure electroencephalogram, 146 patients with normal background electroencephalogram, and 330.08 hours of total recording time, 38 patients were collected with testing set seizure electroencephalogram, 12 patients with normal background electroencephalogram, and 171.42 hours of total time. The Mini-batch size is set to 512, 120 epochs are trained, and the test set after training is represented as:
the model test ROC curve AUC is 0.73;
the model test specificity was 95.90%, the test sensitivity was 53.68%, and the data set was compared with the performance index of the conventional model method, which is shown in the following table.
As can be seen from the above table, the method of the present embodiment has improved performance compared to the previous model method, and particularly improves sensitivity, i.e. attack detection rate on the basis of ensuring specificity and not generating misjudgment, and the performance is kept equivalent in the case of a conventional electroencephalogram in a shorter time. Can improve more accurate diagnosis basis support and auxiliary diagnosis opinions for the diagnosis of the clinician.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.
Claims (4)
1. An aided diagnosis system based on electroencephalogram deep learning, comprising:
an input normalization module: the electroencephalogram sampling module is used for acquiring acquired electroencephalogram sample data, and integrating the electroencephalogram sample data into a preset standardized model to obtain standardized electroencephalogram integer data;
a word embedding module: the electroencephalogram normalization module is used for converting the normalized electroencephalogram integer data into word embedding vectors according to a preset word embedding model;
the word embedding module specifically includes:
word embedding matrix unit: the electroencephalogram integer data are used as a corpus according to the recording time sequence, and word prediction supervised learning is carried out through a word bag model and a jump model to form a word embedding matrix;
word embedding vector unit: the electroencephalogram integer data are converted to obtain a three-dimensional data matrix, and the three-dimensional data matrix is multiplied by the word embedding matrix to obtain the word embedding vector;
a deep learning module: the system is used for extracting the features of the word embedding vectors according to a preset deep learning model, and carrying out time marking and identification diagnosis on the extracted features;
an output module: the electroencephalogram sample data is used for outputting the disease attack probability according to the time mark and the identification diagnosis, and distinguishing the electroencephalogram sample data with the disease attack probability exceeding a preset probability.
2. The electroencephalogram deep learning-based aided diagnosis system according to claim 1, specifically comprising:
a two-dimensional matrix conversion unit: the electroencephalogram sample data are converted into a two-dimensional matrix according to the collected electrode number and the recording duration;
a pooling treatment unit: the system is used for pooling the two-dimensional matrix to obtain a pooled matrix after pooling;
a compression transformation unit: and the method is used for carrying out u-law nonlinear compression transformation on the pooled matrix to obtain electroencephalogram integer data.
3. The electroencephalogram deep learning-based aided diagnosis system according to claim 1, specifically comprising:
a superimposing unit: the three-dimensional convolution layer is used for carrying out multilayer superposition to be used as a characteristic extraction layer;
a feature extraction unit: the electroencephalogram data extraction layer is used for extracting the features of the word embedding vectors to obtain electroencephalogram data features;
a mark recognition unit: the electroencephalograph data processing system is used for overlaying an attention layer and a recurrent neural network, time marking the electroencephalograph data characteristics and identifying diagnosis.
4. The aided diagnosis system based on electroencephalogram deep learning of claim 1, further comprising:
a supervision training module: the preset word embedding model and the preset deep learning module are supervised and trained through a preset disease attack database, wherein the preset disease attack database comprises a absence attack database, a tonic attack database, a generalized nonspecific attack database, a clonic attack database, a focus nonspecific attack database, a tonic clonic attack database, a simple partial attack database, a dystonia attack database, a complex partial attack database and a myoclonic attack database.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810909558.XA CN109009102B (en) | 2018-08-10 | 2018-08-10 | Electroencephalogram deep learning-based auxiliary diagnosis method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810909558.XA CN109009102B (en) | 2018-08-10 | 2018-08-10 | Electroencephalogram deep learning-based auxiliary diagnosis method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109009102A CN109009102A (en) | 2018-12-18 |
| CN109009102B true CN109009102B (en) | 2021-02-12 |
Family
ID=64632760
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810909558.XA Active CN109009102B (en) | 2018-08-10 | 2018-08-10 | Electroencephalogram deep learning-based auxiliary diagnosis method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109009102B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110957036B (en) * | 2019-10-24 | 2023-07-14 | 中国人民解放军总医院 | Disease prognosis risk assessment model method based on causal reasoning construction |
| CN110767314A (en) * | 2019-10-29 | 2020-02-07 | 湖南省脑科医院 | Schizophrenia state stability prediction model and construction method thereof |
| CN112842357B (en) * | 2019-11-26 | 2024-04-09 | 阿里健康信息技术有限公司 | Brain physiological data processing method, device and storage medium |
| CN111709446B (en) * | 2020-05-14 | 2022-07-26 | 天津大学 | X-ray chest radiography classification device based on improved dense connection network |
| CN111881665B (en) * | 2020-09-27 | 2021-01-05 | 华南师范大学 | Word embedding representation method, device and equipment |
| CN113197545B (en) * | 2021-04-08 | 2022-10-04 | 山东师范大学 | Epilepsy detection system based on graph attention residual error network and focus loss |
| CN114533084B (en) * | 2022-02-09 | 2023-07-14 | 北京师范大学 | Electroencephalogram feature extraction method and device, electronic equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105046111A (en) * | 2015-09-10 | 2015-11-11 | 济南市儿童医院 | Amplitude integrated electroencephalogram result automatic identifying system and method |
| CN107438398A (en) * | 2015-01-06 | 2017-12-05 | 大卫·伯顿 | Portable wearable monitoring system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8684921B2 (en) * | 2010-10-01 | 2014-04-01 | Flint Hills Scientific Llc | Detecting, assessing and managing epilepsy using a multi-variate, metric-based classification analysis |
-
2018
- 2018-08-10 CN CN201810909558.XA patent/CN109009102B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107438398A (en) * | 2015-01-06 | 2017-12-05 | 大卫·伯顿 | Portable wearable monitoring system |
| CN105046111A (en) * | 2015-09-10 | 2015-11-11 | 济南市儿童医院 | Amplitude integrated electroencephalogram result automatic identifying system and method |
Non-Patent Citations (3)
| Title |
|---|
| Automated EEG-Based Epileptic Seizure Detection Using Deep Neural Networks;J. Birjandtalab et al.;《2017 IEEE International Conference on Healthcare Informatics》;20171231;第552-555页 * |
| Rapid Annotation of Seizures and Interictal-ictal Continuum EEG Patterns;Jin Jing et al.;《Annual International Conference of the IEEE Engineering in Medicine and Biology Society》;20180731;第3394-3397页 * |
| 基于符号化表示学习的脑电信号情绪分类方法研究;王帅;《中国优秀硕士学位论文全文数据库 医药卫生科技辑》;20170215(第2期);第31-50页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109009102A (en) | 2018-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109009102B (en) | Electroencephalogram deep learning-based auxiliary diagnosis method and system | |
| JP7018133B2 (en) | ECG heartbeat automatic identification classification method based on artificial intelligence | |
| CN110876626B (en) | Depression detection system based on optimal lead selection of multi-lead electroencephalogram | |
| CN110840402B (en) | Atrial fibrillation signal identification method and system based on machine learning | |
| EP3803709B1 (en) | Detecting abnormalities in ecg signals | |
| CN109934089A (en) | Multistage epileptic EEG Signal automatic identifying method based on supervision gradient lifter | |
| CN108090509B (en) | Data length self-adaptive electrocardiogram classification method | |
| KR102141185B1 (en) | A system of detecting epileptic seizure waveform based on coefficient in multi-frequency bands from electroencephalogram signals, using feature extraction method with probabilistic model and machine learning | |
| CN104636580A (en) | Health monitoring mobile phone based on human face | |
| CN110731773A (en) | abnormal electrocardiogram screening method based on fusion of global and local depth features of electrocardiogram | |
| CN112733774A (en) | Light-weight ECG classification method based on combination of BiLSTM and serial-parallel multi-scale CNN | |
| CN113995419A (en) | Atrial fibrillation risk prediction system based on heartbeat rhythm signal and application thereof | |
| CN111370124A (en) | Health analysis system and method based on facial recognition and big data | |
| CN112641451A (en) | Multi-scale residual error network sleep staging method and system based on single-channel electroencephalogram signal | |
| WO2024040797A1 (en) | Electroencephalogram-based autism evaluation apparatus and method, terminal device, and medium | |
| CN115530788A (en) | Arrhythmia classification method based on self-attention mechanism | |
| CN108962379B (en) | Mobile phone auxiliary detection system for cranial nerve system diseases | |
| CN113080996A (en) | Electrocardiogram analysis method and device based on target detection | |
| Sane et al. | Detection of myocardial infarction from 12 lead ECG images | |
| Mitra et al. | Analyzing Clinical 12-Lead ECG Images Using Deep Learning Algorithms for Objective Detection of Cardiac Diseases | |
| CN114366116A (en) | Parameter acquisition method based on Mask R-CNN network and electrocardiogram | |
| Sharma et al. | A Smart Healthcare Framework for Accurate Detection of Schizophrenia using Multi-Channel EEG | |
| CN112545535B (en) | Sleep-wake cycle analysis method based on amplitude integrated electroencephalogram | |
| CN117457229B (en) | Anesthesia depth monitoring system and method based on artificial intelligence | |
| CN117292420A (en) | Depression detection and recognition system based on facial submicron features |
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 |