CN116172517A - Seizure interval epileptiform discharge detection method and device based on double-view feature fusion framework - Google Patents
Seizure interval epileptiform discharge detection method and device based on double-view feature fusion framework Download PDFInfo
- Publication number
- CN116172517A CN116172517A CN202310155208.XA CN202310155208A CN116172517A CN 116172517 A CN116172517 A CN 116172517A CN 202310155208 A CN202310155208 A CN 202310155208A CN 116172517 A CN116172517 A CN 116172517A
- Authority
- CN
- China
- Prior art keywords
- ied
- spatial
- framework
- features
- seizure
- 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.)
- Pending
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
- A61B5/40—Detecting, measuring or recording for evaluating the nervous system
- A61B5/4076—Diagnosing or monitoring particular conditions of the nervous system
- A61B5/4094—Diagnosing or monitoring seizure diseases, e.g. epilepsy
-
- 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
- A61B5/369—Electroencephalography [EEG]
- A61B5/372—Analysis of electroencephalograms
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Pathology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Artificial Intelligence (AREA)
- Veterinary Medicine (AREA)
- Psychiatry (AREA)
- Neurology (AREA)
- Evolutionary Computation (AREA)
- Psychology (AREA)
- Neurosurgery (AREA)
- Physiology (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Fuzzy Systems (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a method and a device for detecting inter-seizure phase epileptiform discharge (IED) based on a double-view feature fusion framework, which select a multichannel electroencephalogram as a data basis for IED detection and design a discrimination detection model for fusing waveform features of electroencephalogram signals in a time domain and space features among multichannel signals. The model contains three main parts: firstly, directly treating an electroencephalogram as a multichannel time sequence, and explicitly learning deep morphology features by applying a neural network formed by multi-layer one-dimensional convolution; then, converting an electroencephalogram into a three-dimensional tensor arranged according to a channel topological structure, positioning a peak position based on the three-dimensional tensor, reserving a corresponding peak region in a fragment to be classified, sequentially using a three-dimensional convolutional neural network to extract spatial features of a peak region sampling frame, and capturing evolution of the spatial features by a long-term and short-term memory network; and finally, fusing the extracted features so as to judge whether the input signal segment is an IED event.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a seizure interval epileptiform discharge detection method and device based on a double-view feature fusion framework.
Background
Inter-seizure epileptic discharge (Interictal Epileptiform Discharge, IED) is a type of abnormal discharging activity of neurons that occurs frequently in seizure intervals of epileptic patients, and can be recorded by electroencephalogram (EEG) acquisition. IEDs are important biomarkers in the diagnosis of epileptic diseases, and recent studies further indicate that IEDs have important associations with neurodevelopmental abnormalities, cognitive impairment, etc. Considering the adverse factors of long time consumption of manual review of electroencephalogram report, high time cost of culturing doctors, inconsistent evaluation standards among different doctors and the like, the accurate and automatic IED detection tool has important significance in brain disease diagnosis and the like.
IEDs act as an electrophysiological activity whose appearance and diffusion process leaves traces in multiple brain electrical pathways throughout the brain. According to the standards of the clinical neurophysiologic international union (IFCN), typical IED signals should have both signal waveform characteristics and spatial correlation characteristics between multiple channels, so combining morphological characteristics with spatial characteristics is also considered as a key to distinguishing epileptic discharges from benign transients, artifacts. However, the existing IED detection methods are based on two ideas of traditional feature engineering and deep learning, and most of waveform feature recognition on a single multi-aspect channel cannot reflect the physiological characteristics of the IED; a few methods convert multi-channel electroencephalogram into two-dimensional electroencephalogram for identification, and give consideration to the characteristics of the IED in multiple channels, but the method changes the adjacent relation among channels, damages the original inter-channel association characteristics, and cannot meet the requirements of accurate identification of the IED.
Disclosure of Invention
The invention aims to provide a seizure interval epileptiform discharge detection method and device based on a double-view feature fusion frame, which are used for solving the technical problem of low detection accuracy in the prior art.
In order to solve the technical problems, the technical scheme of the invention is as follows:
a first aspect provides a method for inter-seizure epileptiform discharge detection based on a dual view feature fusion framework, comprising:
s1: respectively acquiring an original brain electrical signal of an epileptic patient at a seizure interval and an original brain electrical signal of a healthy subject;
s2: preprocessing the acquired original brain electrical signals, and marking;
s3: an IED detection framework is constructed, wherein the IED detection framework comprises a morphological feature learning module for extracting waveform features of depth signals, a spatial feature learning module for extracting related features of multichannel electroencephalogram signals and a feature fusion module for fusing the waveform features and the spatial features;
s4: acquiring training data from the preprocessed and marked data, and training the IED detection frame by using the training data;
s5: and detecting the data to be identified by using the trained IED detection framework.
In one embodiment, step S1 includes:
and acquiring the original electroencephalogram signals of the epileptic patient at the seizure interval and the original electroencephalogram signals of the healthy subject by using a 10-20 scalp electroencephalogram instrument.
In one embodiment, step S2 includes:
s2.1: resampling the acquired brain electrical data to 250Hz, and carrying out band-pass filtering of 1-70Hz and notch filtering of 50Hz on the resampled signals;
s2.2: labeling the signal obtained in the step S2.1 as positive and negative samples and cutting, wherein the positive sample consists of a seizure interval epileptiform discharge fragment from an epileptic patient, the negative sample consists of a background fragment from a healthy subject and a benign sharp transient fragment, and the length of each labeling fragment is 0.8S;
s2.3: resetting the lead combination for the signal segment obtained in the step S2.2, and storing the segments in the form of an average reference lead, a longitudinal bipolar lead and a transverse bipolar lead combination for each segment of signal.
In one implementation manner, in the IED detection framework constructed in step S3, the morphological feature learning module is sequentially composed of a one-dimensional convolution layer, three residual error connection layers, a one-dimensional convolution layer and an average pooling layer, where the convolution kernel size of the first one-dimensional convolution layer is 1, the number of input channels is 19, and the output channels are 128; the three residual error connecting layers have the same structure and sequentially comprise a batch normalization layer, two one-dimensional convolution layers with 3 convolution kernels and an activation layer, and are used for extracting the depth morphological characteristics of waveforms; the convolution kernel size of the subsequent one-dimensional convolution layer is 1, the number of input channels is 128, and the number of output channels is 1, so that characteristic dimension reduction is realized in the channel dimension; the average pooling layer step length is 2, and is used for realizing characteristic dimension reduction in the time domain.
In one implementation, in the IED detection framework constructed in step S3, the spatial feature learning module includes three parts including peak area positioning, electroencephalogram signal channel rearrangement and spatial feature extraction, where the peak area positioning measures the feature of phase inversion by calculating the variance between channels under the horizontal and vertical bipolar leads from sample frame to sample frame, and the center of the time window with the largest variance variation is used as the peak of the IED; rearranging the electroencephalogram signal channels, rearranging the signals under the average reference leads into three-dimensional tensors according to the channel physical topological structure, and supplementing the vacant parts in the three-dimensional tensors by using the average value of surrounding channels; the spatial feature extraction captures the spatial features of the sampling frames around each IED peak by using a three-dimensional convolution layer, and captures the evolution of the spatial features by using a long-term and short-term memory network, and the last frame output of the long-term and short-term memory network is used as a spatial feature vector.
In one embodiment, in the IED detection framework constructed in step S3, the feature fusion module splices the morphological feature vector and the spatial feature vector to form a fusion feature vector, and classifies the morphological feature vector, the spatial feature vector and the fusion feature vector by using three full-connection layer networks, so as to obtain classification results of three branches.
In one embodiment, during the training process of step S4, IED segments from epileptic patients and benign sharp transients or background segments from healthy subjects are input together to build an IED detection framework; and calculating the loss of the obtained classification results of the three branches by using cross entropy, taking the average value of the three cross entropy losses as the final classification loss and carrying out back propagation.
Based on the same inventive concept, a second aspect of the present invention provides an inter-seizure gap epileptiform discharge detection device based on a dual view feature fusion framework, comprising:
the data acquisition module is used for respectively acquiring the original brain electrical signals of the epileptic seizure interval and the original brain electrical signals of the healthy subjects;
the preprocessing module is used for preprocessing the acquired original electroencephalogram signals and labeling the acquired original electroencephalogram signals;
the detection framework construction module is used for constructing an IED detection framework, and the IED detection framework comprises a morphological feature learning module for extracting waveform features of depth signals, a spatial feature learning module for extracting related features of multichannel electroencephalogram signals and a feature fusion module for fusing the waveform features and the spatial features;
the training module is used for acquiring training data from the preprocessed and marked data and training the IED detection frame by using the training data;
and the detection module is used for detecting the data to be identified by using the trained IED detection framework.
Based on the same inventive concept, a third aspect of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed, implements the method of the first aspect.
Based on the same inventive concept, a fourth aspect of the present invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, said processor implementing the method according to the first aspect when executing said program.
Compared with the prior art, the invention has the following advantages and beneficial technical effects:
according to the high-precision inter-seizure interval epileptiform discharge detection method based on the dual-view feature fusion framework, which is provided by the invention, the IED detection framework is constructed, the waveform features of deep signals can be extracted through a morphological feature learning module, the associated features of multichannel electroencephalogram signals can be extracted through a spatial feature learning module, and the two features can be fused through a fusion module. Training the IED detection framework by using training data; the trained IED detection framework can be used for detecting the data to be identified, so that an accurate evaluation result is obtained.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of an IED detection framework constructed in the practice of the invention;
FIG. 2 is a schematic diagram of a morphological feature learning module constructed in the implementation of the present invention;
FIG. 3 is a schematic diagram showing the structure of signal channel rearrangement in the practice of the present invention
Fig. 4 is a schematic structural diagram of a spatial feature learning module constructed in the implementation of the present invention.
Detailed Description
The existing research work indicates that spatial features are key to distinguishing IEDs from benign sharp transients, but existing IED detection algorithms rely mainly on waveform features of signals, without paying attention to the spatial features exhibited by IED events between channels; and the partial IED detection algorithm maps the multichannel signals into two-dimensional pictures to extract multichannel information, so that the adjacent relation among channels is artificially given, and the real topological structure of the channels is broken. Existing algorithms are poor in performance in distinguishing between benign sharp transients similar to IEDs in waveform and IEDs, and are prone to misdiagnosis.
Based on the method, a detection algorithm taking waveform characteristics and spatial characteristics into comprehensive consideration is provided, and the morphological characteristics and the spatial characteristics of the fragments to be identified are respectively extracted, and the characteristics of the morphological characteristics and the spatial characteristics are fused for final classification, so that the high-precision detection of the IED is realized.
The main inventive concept of the present invention is as follows:
a multi-channel EEG signal of 10-20 systems is selected as a data base of IED detection, and a detection framework integrating EEG signal waveform characteristics and spatial characteristics is designed. The framework comprises three main parts, namely, extracting waveform characteristics and spatial characteristics of the brain electrical signals and fusing the waveform characteristics and the spatial characteristics. In the feature fusion module, respectively outputting a prediction result according to the waveform feature vector, the spatial feature vector and the fusion feature vector; during training, cross entropy loss is calculated for all three prediction results, and mean value back propagation is taken; in predicting unknown samples, the prediction result based on the fusion feature vector is used as the control.
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
The invention provides a seizure interval epileptiform discharge detection method based on a double-view feature fusion framework, which comprises the following steps:
s1: respectively acquiring an original brain electrical signal of an epileptic patient at a seizure interval and an original brain electrical signal of a healthy subject;
s2: preprocessing the acquired original brain electrical signals, and marking;
s3: an IED detection framework is constructed, wherein the IED detection framework comprises a morphological feature learning module for extracting waveform features of depth signals, a spatial feature learning module for extracting related features of multichannel electroencephalogram signals and a feature fusion module for fusing the waveform features and the spatial features;
s4: acquiring training data from the preprocessed and marked data, and training the IED detection frame by using the training data;
s5: and detecting the data to be identified by using the trained IED detection framework.
The inter-seizure epileptiform discharge (IED) is the specific discharge of epileptic patients in epileptic inter-seizure electroencephalogram, is an important biological indicator for diagnosing epileptic diseases, and has research to show that the IED has important connection with diseases such as nerve dysplasia, cognitive impairment and the like. The main objective of the IED detection task is to judge the IED fragment from the brain electrical signal of the subject, and considering the IED as an electrophysiological event, the IED has a diffusion characteristic in the whole brain domain, so that the comprehensive analysis of the whole brain multichannel brain electrical signal is more in accordance with physiological significance and has higher accuracy. Based on the detection framework, the invention relates to a detection framework based on double-view feature fusion, and direct detection of IED events is realized.
Please refer to fig. 1, which is a schematic diagram of an IED detection framework constructed in the implementation of the present invention.
In one embodiment, step S1 includes:
and acquiring the original electroencephalogram signals of the epileptic patient at the seizure interval and the original electroencephalogram signals of the healthy subject by using a 10-20 scalp electroencephalogram instrument.
In one embodiment, step S2 includes:
s2.1: resampling the acquired brain electrical data to 250Hz, and carrying out band-pass filtering of 1-70Hz and notch filtering of 50Hz on the resampled signals;
s2.2: labeling the signal obtained in the step S2.1 as positive and negative samples and cutting, wherein the positive sample consists of a seizure interval epileptiform discharge fragment from an epileptic patient, the negative sample consists of a background fragment from a healthy subject and a benign sharp transient fragment, and the length of each labeling fragment is 0.8S;
s2.3: resetting the lead combination for the signal segment obtained in the step S2.2, and storing the segments in the form of an average reference lead, a longitudinal bipolar lead and a transverse bipolar lead combination for each segment of signal.
In the specific implementation process, firstly, downsampling (resampling) and filtering are carried out on the acquired electroencephalogram signals, and then IED events are marked in the resting state electroencephalogram of an epileptic by the length of 0.8s; to ensure the equalization of training data, a certain amount of background segments or benign sharp transient segments are marked in the brain electrical of healthy subjects, making the number of segments from healthy subjects equal to the number of IED events. Wherein a polyphase filtering algorithm can be used to resample the acquired brain electrical data to 250Hz. In labeling, IED segments are labeled in the brain of epileptic patients and background segments are labeled in the brain of healthy subjects and benign sharp transient segments that are easily confused with IEDs are stored in an average reference lead format of length 0.8 s.
In order to further augment the data features to increase the robustness of the model, the invention uses two augmentation strategies of the electroencephalogram signals: 1) Shifting the marked fragments back and forth, and increasing the data characteristics of the time dimension; 2) And exchanging the channel signals of the fragments according to the left brain region and the right brain region, and increasing the data characteristics of the space dimension.
The dimension of the single sample finally obtained is [ _h, seg_length ]. n_ch represents the number of channels of electrophysiological data, where the number of channels of the 10-20 average reference leads is 19; seg_length represents the segment length, where the sample duration is 0.8s, the sampling rate is 250Hz and thus seg_length is 200. The final obtained sample data may be expressed as [ _, n_h, seg_length ], where n_samples represent the number of samples.
In one implementation manner, in the IED detection framework constructed in step S3, the morphological feature learning module is sequentially composed of a one-dimensional convolution layer, three residual error connection layers, a one-dimensional convolution layer and an average pooling layer, where the convolution kernel size of the first one-dimensional convolution layer is 1, the number of input channels is 19, and the output channels are 128; the three residual error connecting layers have the same structure and sequentially comprise a batch normalization layer, two one-dimensional convolution layers with 3 convolution kernels and an activation layer, and are used for extracting the depth morphological characteristics of waveforms; the convolution kernel size of the subsequent one-dimensional convolution layer is 1, the number of input channels is 128, and the number of output channels is 1, so that characteristic dimension reduction is realized in the channel dimension; the average pooling layer step length is 2, and is used for realizing characteristic dimension reduction in the time domain.
Specifically, the input of the morphological feature learning module is the segment to be classified under the average reference lead, and the output is the feature vector embedded with the morphological feature of the signal. The first layer one-dimensional convolution can realize automatic learning of weights among channels and enrichment of feature space under the condition of not damaging time domain correlation. The purpose of the residual connection layer is to extract the depth morphology features of the waveform.
In the implementation process, the input of the morphological feature extraction module is a sample signal [ _, n_h, seg_length ], and the output is a feature vector [ _, m_ ] corresponding to each sample, wherein the parameters of the input sample are consistent with the sample obtained in the step S2, and the length m_embedding of the morphological feature vector is 100.
In one implementation, in the IED detection framework constructed in step S3, the spatial feature learning module includes three parts including peak area positioning, electroencephalogram signal channel rearrangement and spatial feature extraction, where the peak area positioning measures the feature of phase inversion by calculating the variance between channels under the horizontal and vertical bipolar leads from sample frame to sample frame, and the center of the time window with the largest variance variation is used as the peak of the IED; rearranging the electroencephalogram signal channels, rearranging the signals under the average reference leads into three-dimensional tensors according to the channel physical topological structure, and supplementing the vacant parts in the three-dimensional tensors by using the average value of surrounding channels; the spatial feature extraction captures the spatial features of the sampling frames around each IED peak by using a three-dimensional convolution layer, and captures the evolution of the spatial features by using a long-term and short-term memory network, and the last frame output of the long-term and short-term memory network is used as a spatial feature vector.
Specifically, the spatial feature learning module comprises three parts, namely IED peak positioning based on phase inversion characteristics under bipolar leads, signal channel rearrangement based on a spatial topological structure and spatial feature extraction based on a neural network. The input of the method is three lead fragments of the same signal to be classified under the average reference lead and the transverse and longitudinal bipolar leads, wherein the fragments under the transverse and longitudinal bipolar leads are used for positioning the wave crest of the signal, the signal under the average reference lead is used for channel rearrangement and space feature extraction, and the output of the signal under the average reference lead is a feature vector embedded with the space feature of the signal.
Wherein the peak area localization algorithm aims to localize and preserve the peak area of the IED, thereby avoiding masking the spatial features of the relatively distinct peak area with a relatively smooth and lengthy background section. The peak region localization algorithm exploits the inverse abrupt nature of the phase of the IED under bipolar leads, with the peak of the IED centered in the time window where variance changes most rapidly (i.e., variance of variance is greatest) in the time domain.
In the specific implementation process, in order to improve the resolution of the bipolar leads, the invention converts the original sample signal into two bipolar lead modes of transverse and longitudinal modes and stacks the two bipolar lead modes together, and the single sample signal S= [ __ ch, seg_length after stacking]Where n_bi_h is the number of channels after bipolar lead stacking, and the value is 36, then the variance v=var (S, dim=0) of the sample in the time domain, dim represents the dimension, and the above formula represents the calculation of the variance of the single sample signal S in the 0 th dimension (time domain dimension). Measuring the variance variation degree in the time domain by using a sliding time window, and locating the peak
Where l is the signal length seg_length, l r 、l w The length of the reserved peak area and the length of the sliding time window are respectively, N is a natural number set, i is the coordinate (index) of the position, i is the i frame, i p Representing the coordinates of the peak, and p represents the peak. After the positioning is completed, retaining the two sides of the peak in the original average reference lead signal>
And the corresponding output of each sample is [ _h, l [) r ]In the present embodiment l r 50.
Signal channel rearrangement refers to rearranging the signals under the average reference leads into a three-dimensional tensor according to the physical topological structure of the channel, and supplementing the vacant part of the three-dimensional tensor by using the average value of surrounding channels. The mapping relation is shown in figure 3, and the three-dimensional tensor corresponding to each sample after rearrangement is [5, l ] r ]。
The spatial feature extraction module aims at capturing IED spatial features and the evolution process of the IED spatial features, and sequentially comprises three parts, namely a three-dimensional convolution layer, a downsampling layer and a long-term and short-term memory network. Firstly, three-dimensional convolution layers are used for extracting spatial characteristics of each sampling frame, wherein the sampling frames refer to voltage spatial distribution recorded by each electroencephalogram channel at a certain sampling moment, the step length of the three-dimensional convolution layers in the time domain is 1 so as to ensure that the characteristics of each sampling frame are independently extracted, and the spatial characteristics of each sampling frame after extraction form a spatial characteristic sequence which can be recorded as [ _, lr together]Where s_emmbedding is the feature dimension extracted for each sample frame, l r Is the number of sampling frames, i.e. the length of the aforementioned three-dimensional tensor in the time domain. And then, in order to capture microscopic physiological information and macroscopic state of the spatial feature change, performing 50% and 25% downsampling treatment on the extracted spatial feature sequence, wherein the spatial feature sequence after downsampling and the spatial feature sequence before downsampling form three independent spatial feature sequences in total. Three independent long-short-term memory network pairs are used for extracting evolution of spatial features, and the superposition sum of states of the last hidden layer of the three long-short-term memory networks is used as a spatial feature vector, so that fusion of spatial change features under different scales is realized. The final spatial feature vector has a size of [ _]In this embodiment, s_nesting is 32.
In one embodiment, in the IED detection framework constructed in step S3, the feature fusion module splices the morphological feature vector and the spatial feature vector to form a fusion feature vector, and classifies the morphological feature vector, the spatial feature vector and the fusion feature vector by using three full-connection layer networks, so as to obtain classification results of three branches.
The feature fusion module is used for fusing the morphological features and the spatial features and realizing final classification. However, the direct splicing and training of the morphological feature vector and the spatial feature vector may cause inconsistent convergence speeds of the morphological feature extraction module and the spatial feature extraction module, which deviates from the optimal performance. In order to solve the problem, the invention introduces a gradient mixing idea, namely three independent full-connection layers are used for respectively carrying out classification prediction on the morphological feature vector, the spatial feature vector and a one-dimensional fusion feature vector formed by splicing the morphological feature vector and the spatial feature vector, wherein the activation function of all the full-connection layers is Sigmoid. In the training stage, the classification results of the three full-connection layers respectively calculate losses, and the average value of the three losses is taken for carrying out loss feedback; in the prediction stage, only the output result of the full-connection layer corresponding to the fusion feature vector is used as a final prediction result.
In one embodiment, during the training process of step S4, IED segments from epileptic patients and benign sharp transients or background segments from healthy subjects are input together to build an IED detection framework; and calculating the loss of the obtained classification results of the three branches by using cross entropy, taking the average value of the three cross entropy losses as the final classification loss and carrying out back propagation.
In the detection process, a sample to be classified is input into a model to extract waveforms and spatial features, and confidence level output by a full connection layer corresponding to the fusion feature vector is used as a prediction result.
The method provided by the invention is illustrated by the following specific examples.
Step 1: data acquisition refers to the acquisition of brain electricity of a tested person in a resting state by using a standard 10-20 brain electricity analyzer. And downsampling the acquired electroencephalogram data to 250Hz based on a polyphase filtering algorithm, and then applying 1-70Hz band-pass filtering and 50Hz notch filtering to the downsampled electroencephalogram data.
Step 2: the data preprocessing refers to preprocessing an acquired data set and labeling IED fragments, background and benign sharp transient fragments. Since the duration of IED events typically does not exceed 0.2s, the present embodiment takes 0.8s as the length of the clip in order to be able to include the complete IED event signal; meanwhile, as the events of the IED are rare, in order to increase the diversity of the features, two data augmentation schemes are used in the embodiment: by moving the extracted segments back and forth in the time domain, the diversity of features in the time dimension is increased while the complete IED event is not destroyed; by opposing channel signals at symmetrical locations of left and right brain regions of the brain, the diversity of features in the spatial dimension is increased.
Step 3: constructing an IED event detection framework;
(3.1) in order to extract morphological features contained in the signal, the present embodiment designs a morphological feature extraction module composed of a one-dimensional convolutional network. As shown in fig. 2, the input of the morphological feature extraction module is a segment under the average reference lead of 0.8s, and the one-dimensional convolutional neural network can adaptively learn the weights between the channels, so that the morphological feature of the signal is extracted without artificially introducing the association information between the channels, and the output of the morphological feature extraction module is a feature vector with the length of 100.
(3.2) in order to extract the spatial features contained in the signal, the embodiment designs a morphological feature extraction module composed of a three-dimensional convolution network and a long-short-term memory network, which is used for extracting the spatial features of each sampling frame in the time domain and extracting the evolution of the spatial features in the time domain respectively. The method comprises the steps that firstly, a wave crest is positioned by a wave crest positioning algorithm based on a segment under a 0.8s bipolar reference lead in the initial stage of a spatial feature extraction module, and then 0.05s parts on two sides of the wave crest under an average reference lead are reserved; the segments are then arranged in a spatial topology as shown in fig. 3, and the portions of the tensor that lack the corresponding physical channels are filled with the mean of the surrounding channels. The structure of the subsequent spatial feature extraction module is shown in fig. 4, and the output of the spatial feature extraction module is a feature vector with length of 32.
(3.3) in order to achieve fusion of morphological features and spatial features, the present embodiment adopts the idea of gradient mixing, that is, three full-connection layers are used to process morphological and spatial feature vectors and fusion feature vectors respectively, so that the lengths of input vectors of the three are 100, 32 and 132 respectively, and the output is a confidence level between 0 and 1.
Step 4: training the IED event detection framework using the training data;
in order to realize parameter sharing and common training among all modules, the embodiment calculates cross entropy loss for the outputs of three full-connection layers respectively during training, and trains by taking the average value of the three as the final loss, thereby relieving performance loss caused by inconsistent convergence speeds of different modules.
Step 5: and detecting the data to be identified by using the trained IED event detection framework.
In order to confirm the effect of the present embodiment, the present embodiment uses data of the same data set that is not used for training, and performs the same data slicing and downsampling operations, resulting in a series of data to be identified. And inputting the data to be identified into a detection framework to obtain a final identification result.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention firstly tries to extract the form and waveform characteristics of the IED event respectively and combine the form and waveform characteristics, and shows good effect on the IED event detection task, which shows that the evolution of correctly capturing the spatial characteristics is helpful for the identification of the physiological event.
2. According to the invention, an IED wave crest positioning algorithm is provided by utilizing the discharge characteristics of the IED event, and the IED wave crest positioning is creatively applied to IED event detection, so that the detection effect is improved.
3. In the invention, during data preprocessing and labeling, the seizure interval epileptiform discharge fragments from an epileptic patient are labeled as positive samples, the background fragments and benign sharp transient fragments from a healthy subject are labeled as negative samples, and by fusing morphological characteristics and spatial characteristics of an electroencephalogram signal, the model has better performance in distinguishing IED (intelligent electronic device) from the background electroencephalogram signal and is distinguished from benign sharp transient (Benign Sharp Transient) with similar waveforms
There are also outstanding advantages.
Example two
Based on the same inventive concept, the invention discloses a seizure interval epileptiform discharge detection device based on a double-view feature fusion framework, which comprises:
the data acquisition module is used for respectively acquiring the original brain electrical signals of the epileptic seizure interval and the original brain electrical signals of the healthy subjects;
the preprocessing module is used for preprocessing the acquired original electroencephalogram signals and labeling the acquired original electroencephalogram signals;
the detection framework construction module is used for constructing an IED detection framework, and the IED detection framework comprises a morphological feature learning module for extracting waveform features of depth signals, a spatial feature learning module for extracting related features of multichannel electroencephalogram signals and a feature fusion module for fusing the waveform features and the spatial features;
the training module is used for acquiring training data from the preprocessed and marked data and training the IED detection frame by using the training data;
and the detection module is used for detecting the data to be identified by using the trained IED detection framework. Since the device described in the second embodiment of the present invention is a device for implementing the inter-seizure gap epileptiform discharge detection method based on the dual-view feature fusion framework in the first embodiment of the present invention, based on the method described in the first embodiment of the present invention, a person skilled in the art can understand the specific structure and deformation of the device, and therefore, the detailed description thereof is omitted herein. All devices used in the method of the first embodiment of the present invention are within the scope of the present invention.
Example III
Based on the same inventive concept, the present invention also provides a computer-readable storage medium having stored thereon a computer program which, when executed, implements the method as described in embodiment one.
Since the computer readable storage medium described in the third embodiment of the present invention is a computer readable storage medium used for implementing the inter-seizure gap epileptiform discharge detection method based on the dual-view feature fusion framework in the first embodiment of the present invention, based on the method described in the first embodiment of the present invention, a person skilled in the art can understand the specific structure and the modification of the computer readable storage medium, and therefore, the detailed description thereof is omitted herein. All computer readable storage media used in the method according to the first embodiment of the present invention are included in the scope of protection.
Example IV
Based on the same inventive concept, the present application also provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the method in the first embodiment when executing the program.
Since the computer device described in the fourth embodiment of the present invention is a computer device used for implementing the inter-seizure gap epileptiform discharge detection method based on the dual-view feature fusion framework in the first embodiment of the present invention, based on the method described in the first embodiment of the present invention, a person skilled in the art can understand the specific structure and deformation of the computer device, and therefore, the detailed description thereof is omitted herein. All computer devices used in the method of the first embodiment of the present invention are within the scope of the present invention.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention. It will be apparent to those skilled in the art that various modifications and variations can be made to the embodiments of the present invention without departing from the spirit or scope of the embodiments of the invention. Thus, if such modifications and variations of the embodiments of the present invention fall within the scope of the claims and the equivalents thereof, the present invention is also intended to include such modifications and variations.
Claims (10)
1. The method for detecting the inter-seizure epileptiform discharge based on the double-view feature fusion framework is characterized by comprising the following steps of:
s1: respectively acquiring an original brain electrical signal of an epileptic patient at a seizure interval and an original brain electrical signal of a healthy subject;
s2: preprocessing the acquired original brain electrical signals, and marking;
s3: an IED detection framework is constructed, wherein the IED detection framework comprises a morphological feature learning module for extracting waveform features of depth signals, a spatial feature learning module for extracting related features of multichannel electroencephalogram signals and a feature fusion module for fusing the waveform features and the spatial features;
s4: acquiring training data from the preprocessed and marked data, and training the IED detection frame by using the training data;
s5: and detecting the data to be identified by using the trained IED detection framework.
2. The method for detecting inter-seizure epileptiform discharge based on the dual view feature fusion framework of claim 1, wherein step S1 comprises:
and acquiring the original electroencephalogram signals of the epileptic patient at the seizure interval and the original electroencephalogram signals of the healthy subject by using a 10-20 scalp electroencephalogram instrument.
3. The method for detecting inter-seizure epileptiform discharge based on the dual view feature fusion framework of claim 1, wherein step S2 includes:
s2.1: resampling the acquired brain electrical data to 250Hz, and carrying out band-pass filtering of 1-70Hz and notch filtering of 50Hz on the resampled signals;
s2.2: labeling the signal obtained in the step S2.1 as positive and negative samples and cutting, wherein the positive sample consists of a seizure interval epileptiform discharge fragment from an epileptic patient, the negative sample consists of a background fragment from a healthy subject and a benign sharp transient fragment, and the length of each labeling fragment is 0.8S;
s2.3: resetting the lead combination for the signal segment obtained in the step S2.2, and storing the segments in the form of an average reference lead, a longitudinal bipolar lead and a transverse bipolar lead combination for each segment of signal.
4. The method for detecting the inter-seizure gap epileptiform discharge based on the double-view feature fusion framework as claimed in claim 1, wherein in the IED detection framework constructed in the step S3, the morphological feature learning module is sequentially composed of one-dimensional convolution layer, three residual error connection layers, one-dimensional convolution layer and an average pooling layer, wherein the convolution kernel of the first one-dimensional convolution layer is 1, the number of input channels is 19, and the number of output channels is 128; the three residual error connecting layers have the same structure and sequentially comprise a batch normalization layer, two one-dimensional convolution layers with 3 convolution kernels and an activation layer, and are used for extracting the depth morphological characteristics of waveforms; the convolution kernel size of the subsequent one-dimensional convolution layer is 1, the number of input channels is 128, and the number of output channels is 1, so that characteristic dimension reduction is realized in the channel dimension; the average pooling layer step length is 2, and is used for realizing characteristic dimension reduction in the time domain.
5. The method for detecting inter-seizure gap epileptiform discharge based on a dual view feature fusion framework as claimed in claim 1, wherein in the IED detection framework constructed in step S3, the spatial feature learning module includes three parts including peak region positioning, electroencephalogram signal channel rearrangement and spatial feature extraction, wherein the peak region positioning measures the feature of phase inversion by calculating variances between channels under transverse and longitudinal bipolar leads from sample frame to sample frame, and takes the center of a time window with the largest variance variation as a peak of the IED; rearranging the electroencephalogram signal channels, rearranging the signals under the average reference leads into three-dimensional tensors according to the channel physical topological structure, and supplementing the vacant parts in the three-dimensional tensors by using the average value of surrounding channels; the spatial feature extraction captures the spatial features of the sampling frames around each IED peak by using a three-dimensional convolution layer, and captures the evolution of the spatial features by using a long-term and short-term memory network, and the last frame output of the long-term and short-term memory network is used as a spatial feature vector.
6. The method for detecting the inter-seizure gap epileptiform discharge based on the dual-view feature fusion framework as claimed in claim 1, wherein in the IED detection framework constructed in the step S3, the feature fusion module splices the morphological feature vector and the spatial feature vector to form a fusion feature vector, and classifies the morphological feature vector, the spatial feature vector and the fusion feature vector by using three full-connection layer networks respectively to obtain classification results of three branches.
7. The method for inter-seizure gap epileptiform discharge detection based on dual view feature fusion framework of claim 6, wherein during the training process of step S4, IED segments from epileptic patients and benign sharp transients or background segments from healthy subjects are input together to construct an IED detection framework; and calculating the loss of the obtained classification results of the three branches by using cross entropy, taking the average value of the three cross entropy losses as the final classification loss and carrying out back propagation.
8. Interval seizure-like discharge detection device based on double-view feature fusion frame, which is characterized by comprising:
the data acquisition module is used for respectively acquiring the original brain electrical signals of the epileptic seizure interval and the original brain electrical signals of the healthy subjects;
the preprocessing module is used for preprocessing the acquired original electroencephalogram signals and labeling the acquired original electroencephalogram signals;
the detection framework construction module is used for constructing an IED detection framework, and the IED detection framework comprises a morphological feature learning module for extracting waveform features of depth signals, a spatial feature learning module for extracting related features of multichannel electroencephalogram signals and a feature fusion module for fusing the waveform features and the spatial features;
the training module is used for acquiring training data from the preprocessed and marked data and training the IED detection frame by using the training data;
and the detection module is used for detecting the data to be identified by using the trained IED detection framework.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when executed, implements the method of any one of claims 1 to 7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 7 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310155208.XA CN116172517A (en) | 2023-02-21 | 2023-02-21 | Seizure interval epileptiform discharge detection method and device based on double-view feature fusion framework |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310155208.XA CN116172517A (en) | 2023-02-21 | 2023-02-21 | Seizure interval epileptiform discharge detection method and device based on double-view feature fusion framework |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116172517A true CN116172517A (en) | 2023-05-30 |
Family
ID=86434226
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310155208.XA Pending CN116172517A (en) | 2023-02-21 | 2023-02-21 | Seizure interval epileptiform discharge detection method and device based on double-view feature fusion framework |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116172517A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116500335A (en) * | 2023-06-30 | 2023-07-28 | 国网山东省电力公司邹城市供电公司 | Smart power grid electricity larceny detection method and system based on one-dimensional features and two-dimensional features |
| CN116996697A (en) * | 2023-07-24 | 2023-11-03 | 南通大学 | HEVC (high efficiency video coding) frame-oriented video recovery method |
-
2023
- 2023-02-21 CN CN202310155208.XA patent/CN116172517A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116500335A (en) * | 2023-06-30 | 2023-07-28 | 国网山东省电力公司邹城市供电公司 | Smart power grid electricity larceny detection method and system based on one-dimensional features and two-dimensional features |
| CN116500335B (en) * | 2023-06-30 | 2023-10-13 | 国网山东省电力公司邹城市供电公司 | Smart power grid electricity larceny detection method and system based on one-dimensional features and two-dimensional features |
| CN116996697A (en) * | 2023-07-24 | 2023-11-03 | 南通大学 | HEVC (high efficiency video coding) frame-oriented video recovery method |
| CN116996697B (en) * | 2023-07-24 | 2024-02-23 | 南通大学 | HEVC (high efficiency video coding) frame-oriented video recovery method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116172517A (en) | Seizure interval epileptiform discharge detection method and device based on double-view feature fusion framework | |
| CN110353673B (en) | Electroencephalogram channel selection method based on standard mutual information | |
| CN109543526B (en) | True and false facial paralysis recognition system based on depth difference characteristics | |
| CN107837082A (en) | Electrocardiogram automatic analysis method and device based on artificial intelligence self study | |
| CN104473629B (en) | Automatic electrocardioelectrode placement error detection method based on kernel function classification algorithm | |
| CN109009102B (en) | Electroencephalogram deep learning-based auxiliary diagnosis method and system | |
| CN105342569B (en) | A kind of state of mind detecting system based on brain electricity analytical | |
| CN109276242A (en) | The method and apparatus of electrocardiosignal type identification | |
| CN113499086B (en) | HFO automatic check out system based on degree of depth study | |
| KR20190105180A (en) | Apparatus for Lesion Diagnosis Based on Convolutional Neural Network and Method thereof | |
| CN106725452A (en) | Based on the EEG signal identification method that emotion induces | |
| CN106580307A (en) | Quality judgement method and quality judgement system for electrocardiogram | |
| CN111956208B (en) | ECG signal classification method based on ultra-lightweight convolutional neural network | |
| CN114093501B (en) | Intelligent auxiliary analysis method for child movement epilepsy based on synchronous video and electroencephalogram | |
| CN111797901A (en) | Retinal artery and vein classification method and device based on topological structure estimation | |
| CN113509148B (en) | Schizophrenia detection system based on mixed high-order brain network | |
| CN115081486B (en) | System and method for positioning epileptic focus by using intracranial brain electrical network in early stage of epileptic seizure | |
| CN107411738A (en) | A kind of mood based on resting electroencephalogramidentification similitude is across individual discrimination method | |
| CN104095627A (en) | Electrocardiogram digitized signal quality soft-decision method and device | |
| CN110786849A (en) | Electrocardiosignal identity recognition method and system based on multi-view discriminant analysis | |
| CN110141245A (en) | A kind of Characteristics of electrocardiogram vector extracting method | |
| CN106096544A (en) | Non-contact blink and heart rate joint detection system and method based on second-order blind identification | |
| CN117084693A (en) | Encephalopathy detecting system based on multi-domain feature fusion and attention mechanism | |
| CN116035598B (en) | Sleep spindle wave intelligent recognition method and system | |
| CN110507299B (en) | Heart rate signal detection device and method |
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 |