Epilepsy prediction method, computer equipment and computer readable storage medium

Abstract

The invention discloses an epilepsy prediction method. The epilepsy prediction method comprises the following steps: acquiring various physiological signal data; training a corresponding auto-encoder by using each kind of physiological signal data; acquiring physiological signal sequences in the same sequence form according to various physiological signal data and the trained auto-encoders; and training an epilepsy prediction model by using the physiological signal sequence and the epilepsy behavior event tag. According to the epilepsy prediction method, a multi-modal data fusion framework is provided, unified feature coding can be carried out on data of different modals from multiple sources, and the epilepsy prediction model is trained by using a coded sequence, so that the anomaly detection capability of the epilepsy prediction model is stronger, and the epilepsy prediction accuracy is improved. Some unobvious potential abnormal attacks are further detected, and different behavior phenotypes of epilepsy objects can be classified.

Description

Technical Field

[0001] The present invention relates to the technical fields of bio-signal processing and intelligent medical treatment, and particularly relates to an epilepsy prediction method, a computer device, and a computer-readable storage medium. Background Art

[0002] The accurate detection of epilepsy has always been a technical field that doctors and researchers have paid close attention to. The existing technologies mainly use the detection of abnormal discharges in electroencephalogram as the gold standard for judging epileptic seizures. They mainly perform time-frequency conversion on the electroencephalogram signals recorded by EEG, and use time-domain features, frequency-domain features, or information entropy as the classification basis to determine whether there are obvious seizures in patients or epileptic model animals. In addition, there are also technologies that use cameras to record the videos of patients' seizures and further judge and confirm the electroencephalogram detection results from the behavior. There are also some technologies that use deep learning methods to use the spatio-temporal features of electroencephalogram as the training set of neural network models, so as to realize the analysis of epilepsy.

[0003] However, the existing technical methods all extract features from single-modal data and use a variety of algorithms for analysis and detection. However, some actual epileptic seizures are latent, and it is difficult to accurately find the suspicious seizure time period from the data of a single source. For example, some seizures do not have significant features in the time-frequency domain of electroencephalogram, but there are obvious abnormalities in behavior; there are also some absence seizures, and there are no particularly obvious abnormal features in both electroencephalogram and behavior. Therefore, relying on single-modal data for seizure discrimination is not accurate and reliable enough, and it is necessary to conduct comprehensive judgment and analysis from the data of other modalities. Summary of the Invention

[0004] In order to solve the above technical problems existing in the prior art, the present invention provides an epilepsy prediction method, a computer device, and a computer-readable storage medium.

[0005] According to one aspect of the present invention, an epilepsy prediction method is provided, which includes: obtaining a variety of physiological signal data; training a corresponding autoencoder using each physiological signal data; obtaining physiological signal sequences in the same sequence form according to the variety of physiological signal data and the trained multiple autoencoders; training an epilepsy prediction model using the physiological signal sequences and epilepsy behavior event labels.

[0006] In an example of the epilepsy prediction method provided in the above aspect, the epilepsy prediction method further includes: real-time collecting a variety of physiological signal data; respectively inputting the real-time collected variety of physiological signal data into the corresponding trained autoencoders to obtain real-time physiological signal sequences in the same sequence form; inputting the real-time physiological signal sequences into the trained epilepsy prediction model for classification and detection.

[0007] In an example of the epilepsy prediction method provided in the above-mentioned aspect, the multiple physiological signal data includes: behavioral activity data, electroencephalogram signal data, electromyogram signal data, and heart rate signal data.

[0008] In an example of the epilepsy prediction method provided in the above-mentioned aspect, the multiple autoencoders include: a behavioral autoencoder, an electroencephalogram autoencoder, an electromyogram autoencoder, and a heart rate autoencoder; wherein, the behavioral autoencoder is trained using the behavioral activity data, the electroencephalogram autoencoder is trained using the electroencephalogram signal data, the electromyogram autoencoder is trained using the electromyogram signal data, and the heart rate autoencoder is trained using the heart rate signal data.

[0009] In an example of the epilepsy prediction method provided in the above-mentioned aspect, the sequences output by the behavioral autoencoder, the electroencephalogram autoencoder, the electromyogram autoencoder, and the heart rate autoencoder have the same form.

[0010] In an example of the epilepsy prediction method provided in the above-mentioned aspect, the physiological signal sequence includes: a behavioral sequence, an electroencephalogram sequence, an electromyogram sequence, and a heart rate sequence, and the sequences of the behavioral sequence, the electroencephalogram sequence, the electromyogram sequence, and the heart rate sequence have the same form;

[0011] wherein, the behavioral sequence is obtained according to the behavioral activity data and the trained behavioral autoencoder, the electroencephalogram sequence is obtained according to the electroencephalogram signal data and the trained electroencephalogram autoencoder, the electromyogram sequence is obtained according to the electromyogram signal data and the trained electromyogram autoencoder, and the heart rate sequence is obtained according to the heart rate signal data and the trained heart rate autoencoder.

[0012] In an example of the epilepsy prediction method provided in the above-mentioned aspect, the behavioral autoencoder, the electroencephalogram autoencoder, the electromyogram autoencoder, and the heart rate autoencoder are all convolutional neural networks.

[0013] In an example of the epilepsy prediction method provided in the above-mentioned aspect, the epilepsy prediction model is a multi-layer perceptron neural network.

[0014] According to another aspect of the present invention, there is provided a computer device, which includes a processor and a memory connected to the processor, wherein the memory stores program instructions; the processor is used to execute the program instructions stored in the memory to implement the epilepsy prediction method as described above.

[0015] According to still another aspect of the present invention, there is provided a computer-readable storage medium, which stores program instructions, and when the program instructions are executed, the epilepsy prediction method as described above is implemented.

[0016] Beneficial effects: The epilepsy prediction method of the present invention proposes a multi-modal data fusion framework, which can perform unified feature encoding on different modalities of data from multiple sources, and use the encoded sequences to train an epilepsy prediction model, making the abnormal detection ability of the epilepsy prediction model stronger, further detecting some unobvious potential abnormal seizures, and being able to classify different behavioral phenotypes of epilepsy subjects. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. In the drawings:

[0018] Figure 1 is a flowchart of the epilepsy prediction method according to an embodiment of the present invention;

[0019] Figure 2 is a schematic diagram of mouse body key point recognition used in an embodiment of the present invention;

[0020] Figure 3 is a schematic diagram of the geometric posture of mouse body key points used in an embodiment of the present invention;

[0021] Figure 4 is a schematic diagram of the sequence of the distance between the nose and the root of the tail of a mouse extracted according to an embodiment of the present invention;

[0022] Figure 5 is a schematic diagram of the sequence of the moving speed of a mouse extracted according to an embodiment of the present invention;

[0023] Figure 6 is a schematic diagram of epilepsy prediction for a mouse using respective encoders and a multi-layer perceptron neural network according to an embodiment of the present invention;

[0024] Figure 7 is a schematic diagram of the result of behavior classification for a mouse using respective encoders and a multi-layer perceptron neural network according to an embodiment of the present invention;

[0025] Figure 8 is a framework diagram of a computer device according to an embodiment of the present invention;

[0026] Figure 9 is a schematic diagram of a computer storage medium according to an embodiment of the present invention. Detailed Embodiments

[0027] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0028] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data used in appropriate cases can be interchanged so that the embodiments of the present application described here, for example, can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0029] Figure 1 is a flowchart of an epilepsy prediction method according to an embodiment of the present invention.

[0030] Referring to Figure 1 , in step S110, a variety of physiological signal data is acquired. The physiological signal data here is mainly used to train the model, so it does not have to be collected in real time. As an example, the variety of physiological signal data includes: behavioral activity data, electroencephalogram signal data, electromyogram signal data, and heart rate signal data, but the present invention is not limited thereto. The following will detail how to acquire this data.

[0031] A) Construction of an epilepsy model mouse

[0032] The epilepsy mouse models used when collecting physiological signal data are temporal lobe epilepsy mice and Scn1a gene mutation mice. The construction of the temporal lobe epilepsy mice is to inject 450 nanoliters of kainic acid into the dorsal CA3 region of the hippocampus in the mouse brain. After waiting for 2 months, the neurons of the modeled mouse brain slices are stained, and the stained slices show that the distribution of neurons in the CA1\CA3 and DG regions of the hippocampus is divergent, which can cause the mice to have spontaneous temporal lobe epilepsy.

[0033] The Scn1a gene mutation mouse model is to use the CRISPR / Cas9 technology to introduce a sequence into the gene fragment encoding the sodium ion voltage-gated channel 1a in normal mice, causing a toxic exon to be generated in this fragment. This exon causes premature termination of transcription, and the transcription level of Scn1a drops to half of that of wild-type mice, resulting in strong spontaneous epilepsy symptoms.

[0034] B) Behavioral video recording

[0035] The camera for mouse behavior recording is an ordinary industrial camera with a frame rate of 30 Hz. The mouse is placed in a square acrylic box, and the video recording duration is 24 hours. While video recording, the electroencephalogram (EEG) and neck muscle electromyogram (EMG) signals of the mouse are synchronously recorded using a multi-channel EEG recording device, and the heart rate change of the mouse is synchronously recorded at the neck of the mouse.

[0036] C) EEG signal recording

[0037] The EEG signal is recorded using a Medusa EEG recording device. Two-channel electrodes are implanted subdurally in the mouse to record the cortical EEG of the mouse. Two-channel electrodes are implanted in the neck muscles to record the neck muscle EMG signals. The sampling frequency is 500 Hz for both.

[0038] D) Heart rate signal recording

[0039] The heart rate of the mouse is measured using a mouse physiological signal recording device. The hair on the neck of the mouse is completely removed using a depilatory to expose the skin, and an infrared sensor is clipped on its neck to perform real-time sampling of the mouse's heart rate, with a sampling frequency of 15 Hz.

[0040] In step S120, the corresponding autoencoders are trained using each type of physiological signal data. In one example, a behavior autoencoder is trained using behavioral activity data, an EEG autoencoder is trained using EEG signal data, an EMG autoencoder is trained using EMG signal data, and a heart rate autoencoder is trained using heart rate signal data.

[0041] One of the objectives of the present invention is to encode different modal data forms into a unified feature sequence. Here, taking the behavior autoencoder as an example, a detailed description of how to train the encoder is given. It should be understood that the training methods of other autoencoders (EEG autoencoder, EMG autoencoder, and heart rate autoencoder) are the same as that of the behavior autoencoder, and will not be elaborated here.

[0042] The open-source tool DeepLabCut is used to identify the body key points of the mouse and extract the morphological and motion signals of the mouse, such as Figure 2 and Figure 3 as shown. Among them, Figure 2 is a schematic diagram of the mouse body key point identification used in the embodiments of the present invention. Figure 3 is a schematic diagram of the geometric posture of the mouse body key points used in the embodiments of the present invention.

[0043] Geometric calculations and extractions are performed on the morphology of the mouse in each frame of the video, and the action sequences of the mouse are respectively extracted, including: the distance between the nose and the root of the tail, the distance between the bilateral hip bones, the geometric area, the body direction, the body twist degree, the moving speed, the moving angular velocity, and the turning rate. Figure 4 It is a schematic diagram of the sequence of the distance between the nose and the root of the tail of the mouse extracted according to an embodiment of the present invention. Figure 5 It is a schematic diagram of the sequence of the moving speed of the mouse extracted according to an embodiment of the present invention.

[0044] For the electroencephalogram (EEG) and electromyogram (EMG) of the mouse, it is not necessary to extract features. The original data is processed for outliers (including the IQR quartile method and the Z-score method, and corresponding empirical parameters are set according to different experimental conditions), and then band-pass filtering is performed. For the EEG, the data in the range of 0.5 - 100 Hz is retained, and for the EMG, the data in the range of 40 - 250 Hz is retained.

[0045] Data of different modalities have their unique features and laws, such as large differences in sampling rates, different amounts of information contained, and different degrees of variation. Therefore, it is necessary to convert data of different modalities into the same sequence form to facilitate subsequent action and epilepsy detection tasks. Here, an autoencoder based on a convolutional neural network (CNN) is provided to process mouse behavior activity data. The behavior autoencoder has 6 one-dimensional convolutional layers. After each convolutional layer, the ReLU function (i.e., the activation function) is used for activation. The input of the convolutional layer is the dimension of the mouse behavior activity data, the hidden layer is 256, the convolutional kernel size is 3, the stride is 2, the padding is 1, and the output of the behavior autoencoder is 128.

[0046] The decoder uses a transposed convolutional layer symmetric to the behavior autoencoder, and all parameters are the same. The output dimension is the dimension of the mouse behavior sequence (the decoded behavior activity data). After extracting the behavior sequence of the mouse (the behavior activity data before encoding) from the video, the behavior sequence is input into the autoencoder. The behavior sequence is encoded by the autoencoder and then restored by the decoder. The original behavior sequence and the behavior sequence decoded by the decoder are compared to calculate the loss. The loss function is the mean absolute error (MAE). After repeated training, a behavior autoencoder of the mouse is finally obtained. This behavior autoencoder can encode the mouse behavior sequence to reduce the sequence dimension, while the decoder raises the dimension to reconstruct the behavior sequence.

[0047] The behavior autoencoder compresses the input behavior activity data into a behavior sequence with a shape of 1 * 128. That is to say, for any length of behavior activity data (or the original behavior sequence) input into the behavior autoencoder, it will ultimately be compressed into a behavior sequence of the same length, which is conducive to subsequent behavior classification and epilepsy detection.

[0048] In addition, the electroencephalogram (EEG) sequence (i.e., EEG signal data), electromyogram (EMG) sequence (i.e., EMG signal data), and heart rate sequence (i.e., heart rate signal data) of the mice are all trained using autoencoders with the same structure, and then their respective autoencoders are obtained. The original sequences are encoded using their respective autoencoders and uniformly compressed into sequences of 1*128.

[0049] In step S130, physiological signal sequences in the same sequence form are obtained based on various physiological signal data and multiple trained autoencoders. In one example, a behavior sequence is obtained based on behavior activity data and a trained behavior autoencoder, an EEG sequence is obtained based on EEG signal data and a trained EEG autoencoder, an EMG sequence is obtained based on EMG signal data and a trained EMG autoencoder, and a heart rate sequence is obtained based on heart rate signal data and a trained heart rate autoencoder. Here, the behavior sequence, EEG sequence, EMG sequence, and heart rate sequence have the same sequence form.

[0050] In step S140, an epilepsy prediction model is trained using the physiological signal sequences and epilepsy behavior event labels. Specifically, an epilepsy prediction model is trained using the behavior sequence, EEG sequence, EMG sequence, and heart rate sequence obtained in step S130 and the epilepsy behavior event labels.

[0051] A multi-layer perceptron neural network (MLP) is trained using the uniformly encoded behavior sequence, EEG sequence, EMG sequence, heart rate sequence of the mice and the behavior event labels of the mice with epilepsy as an epilepsy prediction model. This multi-layer perceptron neural network can accurately classify and detect the encoded unified sequence, not only find epilepsy seizure events, but also classify the behavior of the mice. Figure 6 It is a schematic diagram of epilepsy prediction for mice using respective encoders and a multi-layer perceptron neural network according to an embodiment of the present invention.

[0052] Refer to Figure 6 , first, various physiological signal data are collected in real time. For example, behavior activity data, EEG signal data, EMG signal data, and heart rate signal data are collected in real time.

[0053] Next, the behavior activity data, EEG signal data, EMG signal data, and heart rate signal data collected in real time are respectively input into their corresponding behavior autoencoder, EEG autoencoder, EMG autoencoder, and heart rate autoencoder to output behavior sequences, EEG sequences, EMG sequences, and heart rate sequences with the same sequence form.

[0054] Finally, the behavior sequences, EEG sequences, EMG sequences, and heart rate sequences with the same sequence form are input into the multi-layer perceptron neural network for behavior classification and epilepsy detection.

[0055] Figure 7 It is a schematic diagram of the result of classifying the behaviors of mice by using respective encoders and a multi-layer perceptron neural network according to an embodiment of the present invention. Refer to Figure 7 , the multi-layer perceptron neural network according to the embodiment of the present invention can detect most of the behavioral phenotypes and epileptic seizures, and has a high detection efficiency for low-level minor seizures.

[0056] In summary, the epileptic seizure prediction method according to the embodiment of the present invention proposes a multi-modal data fusion framework, which can perform unified feature encoding on different-modal data from multiple sources, and use the encoded sequences to train an epileptic seizure prediction model, making the abnormal detection ability of the epileptic seizure prediction model stronger, further detecting some unobvious potential abnormal seizures, and being able to classify different behavioral phenotypes of epileptic subjects.

[0057] To implement the epileptic seizure prediction method in the above embodiment, the present application also provides another computer device 300. Specifically, please refer to Figure 8 , the computer device 300 in the embodiment of the present application includes a processor 31, a memory 32, an input / output device 33, and a bus 34.

[0058] The processor 31, the memory 32, and the input / output device 33 are respectively connected to the bus 34. The memory 32 stores program data, and the processor 31 is configured to execute the program data to implement the epileptic seizure prediction method described in the above embodiment.

[0059] In the embodiment of the present application, the processor 31 may also be referred to as a CPU (Central Processing Unit, central processing unit). The processor 31 may be an integrated voltage control system chip and has the ability to process signals. The processor 31 may also be a general-purpose processor, a digital signal processor (DSP, Digital Signal Process), an application-specific integrated voltage control system (ASIC, Application Specific Integrated Circuit), a field programmable gate array (FPGA, FieldProgrammable GateArray), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor 31 may also be any conventional processor, etc.

[0060] The present application also provides a computer storage medium. Please continue to refer to Figure 9 , Figure 9 It is a schematic diagram of the structure of an embodiment of the computer storage medium provided by the present application. The computer storage medium 400 stores program data 41, and when the program data 41 is executed by a processor, it is used to implement the epileptic seizure prediction method in the above embodiment.

[0061] When the embodiments of the present application are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0062] The above are only the embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A method for predicting epilepsy, characterized in that: The epilepsy prediction method comprises: Acquire multiple physiological signal data; Use each physiological signal data to train the corresponding autoencoder; Acquire physiological signal sequences in the same sequence form according to a variety of physiological signal data and a plurality of trained autoencoders; An epilepsy prediction model is trained using the physiological signal sequence and epilepsy behavior event labels.

2. The epilepsy prediction method according to claim 1, characterized in that: The epilepsy prediction method further comprises: Collect multiple physiological signal data in real time; Inputting the multiple physiological signal data collected in real time into the corresponding trained autoencoders respectively to obtain real-time physiological signal sequences with the same sequence form; The real-time physiological signal sequence is input into the trained epilepsy prediction model for classification and detection.

3. The epilepsy prediction method according to claim 1 or 2, characterized in that: The multiple physiological signal data include: behavioral activity data, electroencephalogram signal data, electromyography signal data and heart rate signal data.

4. The epilepsy prediction method according to claim 3, characterized in that: The multiple autoencoders include: a behavior autoencoder, an EEG autoencoder, an EMG autoencoder, and a heart rate autoencoder; Among them, the behavioral activity data is used to train the behavioral autoencoder, the EEG signal data is used to train the EEG autoencoder, the EMG signal data is used to train the EMG autoencoder, and the heart rate signal data is used to train the heart rate autoencoder.

5. The epilepsy prediction method according to claim 4, characterized in that: The sequence forms of the outputs of the behavior autoencoder, the EEG autoencoder, the EMG autoencoder and the heart rate autoencoder are all the same.

6. The epilepsy prediction method according to claim 4, characterized in that: The physiological signal sequence includes: a behavior sequence, an EEG sequence, an EMG sequence and a heart rate sequence, and the behavior sequence, the EEG sequence, the EMG sequence and the heart rate sequence are all in the same sequence form; Among them, the behavior sequence is obtained according to the behavior activity data and the trained behavior autoencoder, the EEG sequence is obtained according to the EEG signal data and the trained EEG autoencoder, the EMG sequence is obtained according to the EMG signal data and the trained EMG autoencoder, and the heart rate sequence is obtained according to the heart rate signal data and the trained heart rate autoencoder.

7. The epilepsy prediction method according to claim 4, characterized in that: The behavior autoencoder, EEG autoencoder, myoelectric autoencoder and heart rate autoencoder are all convolutional neural networks.

8. The epilepsy prediction method according to claim 1 or 2, characterized in that: The epilepsy prediction model is a multi-layer perceptron neural network.

9. A computer device, characterized in that: The computer device comprises a processor and a memory connected to the processor, wherein: The memory stores program instructions; The processor is used to execute the program instructions stored in the memory to implement the epilepsy prediction method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The storage medium stores program instructions, and when the program instructions are executed, the epilepsy prediction method according to any one of claims 1 to 8 is implemented.

Images