CN110013249B - Portable adjustable head-mounted epilepsy monitor - Google Patents

Abstract

The invention discloses a portable adjustable head-mounted epilepsy monitor, which comprises: the electroencephalogram cap is used for collecting electroencephalogram signals of a patient; the debugging module is used for processing according to electroencephalogram signals of a patient during epileptic seizure, in an early seizure stage and in a seizure stage, acquired by all electrodes on the electroencephalogram cap, so as to determine the most appropriate electrode; the training module is used for carrying out model training according to the electroencephalogram signals collected by the most appropriate electrodes; the monitoring module is used for receiving the electroencephalogram signals and detecting by using the model trained by the training module so as to judge whether the patient is in the early stage of epileptic seizure currently; and the early warning module is used for carrying out early warning when the monitoring module monitors that the patient is in the early stage of the epileptic seizure. The invention can realize portable monitoring and instant early warning, and effectively prevent the occurrence of secondary injury when an epileptic is ill.

Description

Portable adjustable head-mounted epilepsy monitor

Technical Field

The invention relates to the field of brain-computer interfaces and epilepsy disease monitoring, in particular to a portable and adjustable head-mounted epilepsy monitor which utilizes an OpenBCI electroencephalogram cap, utilizes a data transmission system to acquire electroencephalogram signals and utilizes an epilepsy disease detection algorithm to monitor epilepsy diseases.

Background

The brain-computer interface is a technology for establishing communication between the brain and an external machine so that the brain can directly control the external machine, and is a special human-computer interaction technology. The brain-computer interface can be applied to many fields, and can help a patient with dyskinesia to communicate with the outside medically and assist the postoperative patient to carry out rehabilitation training; the brain can be used for directly controlling the game in entertainment, so that the game experience is enhanced; the identity authentication aspect can carry out accurate identity identification; and the monitoring and early warning aspect can carry out early warning reminding on a driver fatigue patient and a sudden disease patient, reduce accident damage and the like.

The epileptic disease monitoring is to carry out real-time monitoring to the patient who suffers from epileptic disease, sends the suggestion early warning to patient and guardian when the morbidity, avoids the patient to suffer secondary physical injury. There are two main methods for monitoring epilepsy: one is video monitoring, and the other is brain-computer interface monitoring. And video monitoring is not easy to be portable, and the accuracy of the video monitoring is easily influenced by similar actions of morbidity and external complex environment. The brain-computer interface monitoring takes the electroencephalogram signals of a patient as an analysis object, and distinguishes epileptic seizure from normal state through efficient and accurate algorithm analysis. The existing brain-computer interface technology generally comprises the following three modules: signal acquisition, signal analysis, corresponding controller and feedback link. The signal acquisition mainly comprises the steps of acquiring physiological electrical signals reflecting human brain intention from the brain, acquiring scalp electroencephalograms, and rarely meeting the functions of portability and attractiveness of the conventional electroencephalogram acquisition equipment; secondly, in order to efficiently analyze and process the acquired electroencephalogram signals, a more accurate and faster electroencephalogram signal processing algorithm is required.

Disclosure of Invention

The invention aims to provide a portable adjustable head-mounted epilepsy monitor, which is characterized in that electroencephalogram signals of a wearer are collected through a head-mounted electroencephalogram cap, the collected electroencephalogram signals are transmitted to a computing module for analysis and processing, if epileptic seizure occurs, detection results are transmitted to a mobile phone APP end of the wearer for reminding, and if the mobile phone APP of the wearer is reminded, the APP automatically sends information to a guardian, so that the guardian can know the condition of a patient in time.

In order to realize the task, the invention adopts the following technical scheme:

a portable adjustable wear epilepsy monitor, includes:

the electroencephalogram cap is used for collecting electroencephalogram signals of a patient;

the debugging module is used for carrying out filtering processing and data division on electroencephalogram signals according to electroencephalogram signals of a patient during epileptic seizure, pre-seizure and seizure, collected by all electrodes on an electroencephalogram cap, classifying the electroencephalogram signals according to channels, constructing a data set for each channel, carrying out ensemble empirical mode decomposition on each data set, dividing samples through a sliding time window, constructing a feature vector by using the samples in the time window, constructing a feature set for each channel, classifying each feature set to obtain the classification accuracy of each channel, and selecting the electrodes corresponding to the channels with the highest classification accuracy as the most appropriate electrodes;

the training module is used for constructing a characteristic vector according to the electroencephalogram signals collected by the most suitable electrodes in the same method as the debugging module, constructing the characteristic vector into a characteristic matrix, training the characteristic matrix through an H-ELM learning machine, and storing the trained model;

the monitoring module is used for receiving the electroencephalogram signals acquired by the most suitable electrodes, intercepting the electroencephalogram signals by a sliding time window, and detecting the electroencephalogram signals by using a model trained by the training module after the intercepted fragments construct a characteristic matrix so as to judge whether the patient is in the early stage of epileptic seizure currently;

and the early warning module is used for carrying out early warning when the monitoring module monitors that the patient is in the early stage of the epileptic seizure.

Further, the filtering and data dividing of the electroencephalogram signal includes:

carrying out filtering pretreatment of 0.5HZ-60HZ on the electroencephalogram signals, and removing the influence of the electro-oculogram signals and the electromyogram signals;

and (3) carrying out data division, wherein the electroencephalogram signal during epileptic seizure is defined as a seizure period, the electroencephalogram signal 10min before seizure is defined as a seizure prophase, and the rest are defined as seizure intervals.

Further, the performing ensemble empirical mode decomposition on each data set, then dividing samples by sliding a time window, and constructing feature vectors by using the samples in the time window includes:

and performing ensemble empirical mode decomposition on each data set to obtain an intrinsic mode function, adding a sliding time window of 6s to the intrinsic mode function to divide samples, and calculating power energy of sample signals in each time window to construct a feature vector.

Further, the ensemble empirical mode decomposition is performed on each data set to obtain an intrinsic mode function, wherein the signal decomposition satisfies the following formula:

wherein r is_nIs a data set d_i(t) residual errors after extraction of n eigenmode functions, c_jThe eigenmode function IMF obtained for signal decomposition.

Further, the calculation formula for dividing the sample is as follows:

S_in＝d_i[t,t+6]

in the above formula, S_inN-th sample representing the i-th channel, d_iIs a single-channel electroencephalogram signal data set, t is the time at the left window side of a sliding time window, [ t, t +6 ]]Is the length of the sliding time window.

Further, the calculation formula for calculating the power energy is as follows:

wherein x_iAre signal data signal points within a time window.

Further, the detector also comprises:

and the mobile phone APP is used for receiving the early warning information sent by the early warning module and carrying out early warning reminding.

Further, the early warning of the early warning module comprises:

if the early-stage signal of the outbreak is detected, the early warning module sends notification information to a mobile phone APP of the patient through Bluetooth; if the other two signals are detected, no notification information is sent;

the information is read by a mobile phone APP of a wearer, if the information is information before epileptic seizure, early warning information is displayed on the mobile phone, and the wearer is reminded through vibration and ringing;

the mobile phone of the wearer sends a confirmation message to the wearer again after early warning for 10min, and if the wearer cancels the confirmation message manually, the early warning is false alarm; if the epilepsy is not cancelled manually, the epilepsy is shown to be in attack, the mobile phone APP of the wearer automatically sends an alarm to the mobile phone APP of the guardian for help, and the positioning information of the mobile phone APP of the wearer is sent to the mobile phone APP of the guardian.

The invention has the following technical characteristics:

1. in the aspect of an epilepsia detection algorithm, compared with other epilepsia detection algorithms, the method provided by the invention mainly considers the position information among channels and the loss of signal quantity during feature extraction, constructs an electroencephalogram signal feature matrix through signal decomposition, and then carries out deeper feature coding through an autoencoder, so that epilepsia and non-epilepsia have stronger distinguishability.

2. In the aspect of epilepsy monitoring, the epilepsy monitoring device has the functions of portable wearing and carrying, timely epilepsy attack reminding and the like. The combination of the daily electroencephalogram cap, the microcomputer and the smart phone facilitates carrying of a wearer, and real-time monitoring is achieved by embedding of the algorithm.

Drawings

FIG. 1 is a flow chart of a process for performing the most appropriate electrode selection;

FIG. 2 is a single channel raw brain electrical signal (abscissa: time; ordinate: amplitude);

FIG. 3 is a multi-channel raw brain electrical signal (abscissa: time; ordinate: amplitude);

FIG. 4 is a schematic flow chart of the training module during model training;

FIG. 5 is a schematic diagram of sliding window partitioning of electroencephalogram signals;

FIG. 6 is a feature matrix construction block diagram;

FIG. 7 is a block diagram of the architecture of the present invention;

FIG. 8 is a wearer's cell phone APP display normal interface;

FIG. 9 is a wearer's mobile phone APP display pre-warning interface;

FIG. 10 is a flow chart of a monitor monitoring process.

Detailed Description

The invention discloses a portable adjustable head-mounted epilepsy monitor, which comprises:

1. electroencephalogram cap

The electroencephalogram cap is used for collecting electroencephalogram signals of a patient, and in the embodiment, the electroencephalogram cap is an OpenBCI electroencephalogram cap. When electroencephalogram signals are collected, the electrodes on the electroencephalogram cap are attached to the scalp of a patient, and electroencephalogram analog signals are collected, for example, fig. 2 shows a single-channel electroencephalogram signal, and fig. 3 shows a multi-channel electroencephalogram signal.

The collected electroencephalogram signals pass through an amplifier and an ADS1299 analog-to-digital converter; the amplifier amplifies weak electroencephalogram signals, so that subsequent processing is facilitated, and the analog-to-digital converter converts electroencephalogram analog signals into digital signals, so that subsequent transmission is facilitated. The electroencephalogram signal converted into the digital signal is transmitted through the 802.11b protocol, that is, the WIFI protocol.

2. Debugging module

The electroencephalogram cap classification method comprises the steps of performing filtering processing and data division on electroencephalogram signals according to electroencephalogram signals of a patient in epileptic seizure periods, pre-seizure periods and seizure periods, collected by all electrodes on the electroencephalogram cap, classifying according to channels, constructing a data set for each channel, performing ensemble empirical mode decomposition on each data set, dividing samples through a sliding time window, constructing feature vectors by using the samples in the time window, constructing a feature set for each channel, classifying each feature set to obtain the classification accuracy of each channel, and selecting the electrodes corresponding to the channels with the highest classification accuracy as the most appropriate electrodes; the specific number of the most suitable electrodes can be selected according to actual needs; as shown in fig. 1, the specific steps are as follows:

2.1 acquisition of initialization data

Firstly, sticking all electrodes of an electroencephalogram cap on the scalp of a patient, and collecting electroencephalogram signals of the patient during epileptic seizure, at the prophase of seizure and at the seizure stage according to all the electrodes; in this embodiment, a total of 10h of signals are collected.

2.2 Filter processing

And (3) carrying out filtering pretreatment of 0.5HZ-60HZ on the electroencephalogram signals, and removing the influence of the electro-oculogram signals and the electromyogram signals.

2.3 data partitioning

Defining the EEG signal of epileptic seizure as seizure period, defining the EEG signal 10min before seizure as seizure prophase, and defining the rest as seizure interval.

2.4 data Classification

Classifying the divided data, constructing a data set for each channel, wherein the data set comprises the electroencephalogram signals of the attack period, the early attack period and the interval of attack at the same time, and the classification formula of the data channels is as follows:

wherein D (t) is an electroencephalogram signal data set of all channels, d_j(t) is a single-channel electroencephalogram signal data set, and n is the number of all channels.

2.5 data decomposition

Each data set d_j(t) carrying out Ensemble Empirical Mode Decomposition (EEMD) to obtain an Intrinsic Mode Function (IMF), wherein the signal decomposition satisfies the following formula:

wherein r is_nIs a data set d_i(t) residual errors after extraction of n eigenmode functions, c_jThe eigenmode function IMF obtained for signal decomposition. The EEMD decomposition can process nonlinear and non-stationary electroencephalogram signals into linear and stable waveforms, and meanwhile, influences of impurity signals are further filtered.

2.6 sample partitioning

Dividing the sample by a sliding time window of IMFs plus 6s, wherein the calculation formula of the sample division is as follows:

S_in＝d_i[t,t+6]，t＝t+6

in the above formula, S_inN-th sample representing the i-th channel, d_iIs a single-channel electroencephalogram signal data set, and t is the left side of a sliding time windowTime of window side, [ t, t +6 ]]Is the length of the sliding time window.

2.7 feature vectors

And (3) solving power energy of the sample signals in each time window and constructing a feature vector, wherein the calculation formula of the power energy is as follows:

wherein x_iAre signal data signal points within a time window.

2.8 determining the most suitable electrode

And constructing feature sets by using the feature vectors, constructing one feature set for each channel, classifying the feature sets of each channel by using a kernel Support Vector Machine (SVM) respectively to obtain the classification accuracy of each channel, and selecting the electrodes corresponding to the 5 channels with the best classification accuracy as the most appropriate electrodes.

According to the most suitable electrode position determined by initialization, wearing the electroencephalogram cap on the head of a patient, and placing the most suitable electrode and a reference electrode; and turning on the power supply of the electroencephalogram cap, and matching the electroencephalogram cap with the training module and the monitoring module by using the radio module.

3. Training module

The device is used for constructing a characteristic vector according to the electroencephalogram signals collected by the most appropriate electrode and the same method as the debugging module, then constructing the characteristic vector into a characteristic matrix, training through an H-ELM learning machine, and storing the trained model; the method comprises the following steps:

3.1 for the electroencephalogram signals collected by the most suitable electrodes, constructing the feature vectors by using the same method of 2.2 to 2.7, as shown in FIG. 4;

and fusing the eigenvectors obtained from each channel among the channels to construct an eigenvector matrix. The characteristic matrix is considered from the perspective of overall data, channel information among data is fused, information loss in the signal characteristic extraction process is reduced, and data integrity can be embodied compared with characteristic vectors. Fig. 5 is a schematic diagram of sliding window division, and fig. 6 is a schematic diagram of a feature matrix configuration.

And 3.2, sending the constructed feature matrix into an H-ELM learning machine for feature coding and training, and storing the trained model.

4. Monitoring module

The electroencephalogram signal acquisition module is used for receiving an electroencephalogram signal acquired by the most appropriate electrode, intercepting the electroencephalogram signal by a sliding time window, and detecting the electroencephalogram signal by using a model trained by the training module after the intercepted segment constructs a characteristic matrix so as to judge whether the patient is in the early stage of epileptic seizure currently; the method specifically comprises the following steps:

the monitoring module receives the electroencephalogram signals collected by the most appropriate electrodes, the data sent in time are segmented and intercepted according to a sliding time window of 6s, the power energy of the intercepted segments is calculated according to the method in the step 2.7, a characteristic matrix is constructed, and the characteristic matrix is sent into a trained model to obtain a detection label, wherein the label is either in the early stage of onset, or in the period of onset, or in the interval of onset.

5. Early warning module

The early warning is carried out when the monitoring module monitors that the patient is in the early stage of the epileptic seizure.

After detection of the monitoring module, if a signal at the early stage of the attack is detected, the early warning module sends notification information to a mobile phone APP of a patient through Bluetooth; if the other two signals are detected, no notification information is sent. Fig. 7 is a schematic diagram of a communication module.

The information is read by a mobile phone APP of a wearer (patient) through a scoket interface, if the information is information before epileptic seizure, early warning information is displayed on the mobile phone, and the wearer is reminded through vibration and ringing; the display of the APP terminal early warning interface of the mobile phone of the wearer is shown in fig. 9.

The mobile phone of the wearer sends a confirmation message to the wearer again after early warning for 10min, and if the wearer cancels the confirmation message manually, the early warning is false alarm; if the epilepsy is not cancelled manually, the epilepsy is shown to be seized, the mobile phone APP of the wearer automatically sends an alarm to the mobile phone APP of the guardian for help, and the positioning information of the mobile phone APP of the wearer is sent to the mobile phone APP of the guardian.

In this embodiment, the debugging module, the training module, the monitoring module and the early warning module are all integrated in a computer, and a raspberry-type microcomputer is adopted in the embodiment.

In order to verify the effectiveness of the epilepsy detection algorithm, the invention selects a public epilepsy data set for experimental verification:

the experiment selects CHB-MIT epilepsy public data set, which is obtained from Boston children hospital and contains cerebral cortex electroencephalogram data of 11 children. The data contains 645 electroencephalographic recordings, of which 136 recordings contain epileptic segments and 509 recordings do not contain epileptic segments. For experimental needs, we chose out a subset of the data to perform the experiment, which satisfies the following condition:

(a) each record contains only one seizure event;

(b) in each recording, the epileptic seizure event is preceded by at least 30min of electroencephalogram data.

To prove the validity of the results, we used 5-fold cross-validation to partition the training set and the test set, and mean values were used to represent the effect of the model. Through experiments, the epilepsy detection algorithm can achieve 99.98% of sensitivity and 91.67 of specificity, and has a high epilepsy recognition effect.

Claims (6)

1. The utility model provides a portable adjustable wear epilepsy monitor, its characterized in that includes:

the electroencephalogram cap is used for collecting electroencephalogram signals of a patient;

the debugging module is used for carrying out filtering processing and data division on electroencephalogram signals according to electroencephalogram signals of a patient during epileptic seizure, pre-seizure and seizure, collected by all electrodes on an electroencephalogram cap, classifying the electroencephalogram signals according to channels, constructing a data set for each channel, carrying out ensemble empirical mode decomposition on each data set, dividing samples through a sliding time window, constructing a feature vector by using the samples in the time window, constructing a feature set for each channel, classifying each feature set to obtain the classification accuracy of each channel, and selecting the electrodes corresponding to the channels with the highest classification accuracy as the most appropriate electrodes;

the training module is used for constructing a characteristic vector according to the electroencephalogram signals collected by the most suitable electrodes in the same method as the debugging module, constructing the characteristic vector into a characteristic matrix, training the characteristic matrix through an H-ELM learning machine, and storing the trained model;

the monitoring module is used for receiving the electroencephalogram signals acquired by the most suitable electrodes, intercepting the electroencephalogram signals by a sliding time window, and detecting the electroencephalogram signals by using a model trained by the training module after the intercepted fragments construct a characteristic matrix so as to judge whether the patient is in the early stage of epileptic seizure currently;

the early warning module is used for carrying out early warning when the monitoring module monitors that the patient is in the early stage of the epileptic seizure;

the early warning of the early warning module comprises:

if the early-stage signal of the outbreak is detected, the early warning module sends notification information to a mobile phone APP of the patient through Bluetooth; if the other two signals are detected, no notification information is sent;

the information is read by a mobile phone APP of a wearer, if the information is information before epileptic seizure, early warning information is displayed on the mobile phone, and the wearer is reminded through vibration and ringing;

the mobile phone of the wearer sends a confirmation message to the wearer again after early warning for 10min, and if the wearer cancels the confirmation message manually, the early warning is false alarm; if the epilepsy is not cancelled manually, the epilepsy is shown to be seized, the mobile phone APP of the wearer automatically sends an alarm to the mobile phone APP of the guardian for help, and the positioning information of the mobile phone APP of the wearer is sent to the mobile phone APP of the guardian;

the EEG signal is subjected to filtering processing and data division, and the method comprises the following steps:

carrying out filtering pretreatment of 0.5HZ-60HZ on the electroencephalogram signals, and removing the influence of the electro-oculogram signals and the electromyogram signals;

and (3) carrying out data division, wherein the electroencephalogram signal during epileptic seizure is defined as a seizure period, the electroencephalogram signal 10min before seizure is defined as a seizure prophase, and the rest are defined as seizure intervals.

2. The portable adjustable head-mounted epilepsy monitor of claim 1, wherein said performing an ensemble empirical mode decomposition of each data set, then dividing the samples by sliding the time window, and constructing feature vectors using the samples in the time window comprises:

and performing ensemble empirical mode decomposition on each data set to obtain an intrinsic mode function, adding a sliding time window of 6s to the intrinsic mode function to divide samples, and calculating power energy of sample signals in each time window to construct a feature vector.

3. The portable adjustable head-mounted epilepsy monitor of claim 1, wherein said ensemble empirical mode decomposition of each data set yields an eigenmode function, wherein the signal decomposition satisfies the following equation:

wherein r is_nIs a data set d_i(t) residual errors after extraction of n eigenmode functions, c_jThe eigenmode function IMF obtained for signal decomposition.

4. The portable adjustable head-mounted epilepsy monitor of claim 1, wherein said sample-dividing calculation formula is:

S_in＝d_i[t，t+6]

in the above formula, S_inN-th sample representing the i-th channel, d_iIs a single-channel electroencephalogram signal data set, t is the time at the left window side of a sliding time window, [ t, t +6 ]]Is the length of the sliding time window.

5. The portable adjustable head-mounted epilepsy monitor of claim 2, wherein said power energy is calculated by the formula:

wherein x_iAre signal data signal points within a time window.

6. The portable adjustable head-mounted epilepsy monitor of claim 2, wherein said monitor further comprises:

and the mobile phone APP is used for receiving the early warning information sent by the early warning module and carrying out early warning reminding.

Images