JP7072771B2 - Epilepsy determination device, epilepsy determination system, epilepsy determination method and epilepsy determination program - Google Patents

Description

The present invention relates to an epilepsy determination device, an epilepsy determination system, an epilepsy determination method, and an epilepsy determination program.

Conventionally, electrodes are placed on the head of a patient to measure electroencephalogram (EEG), and an electroencephalogram pattern peculiar to an epileptic attack may be observed. Observations of such EEG patterns may be made by an epilepsy specialist.

The following Non-Patent Document 1 describes a technique for classifying whether or not an epileptic seizure has occurred by using a brain wave as an input, extracting a feature amount by a wavelet transform, and using a neural network.

E. Juarez-Guerra, V. Alarcon-Aquino, P. Gomez- Gil, "Epilepsy seizure detection in eeg signals using wavelet transforms and neural networks", 2015, Lecture Notes in Electrical Engineering, volume 312, p. 261-269

However, it is known that there are various types of epileptic seizures and that the EEG patterns during the seizures are also diverse. Therefore, it has been difficult to determine an epileptic seizure with sufficiently high accuracy by the conventional technique.

Therefore, the present invention provides an epilepsy determination device, an epilepsy determination system, an epilepsy determination method, and an epilepsy determination program capable of determining an epilepsy attack with higher accuracy.

The epilepsy determination device according to one aspect of the present invention includes an acquisition unit that acquires brain wave data of a subject, a generation unit that cuts out the brain wave data in a predetermined time width and generates a plurality of images representing the brain waves, and a plurality of generation units. An image is input to the trained model, and the trained model is provided with a determination unit for determining whether an electroencephalogram of an epileptic seizure appears in any of a plurality of images.

According to this aspect, the EEG data is cut out in a predetermined time width, a plurality of images representing the EEG are generated, and the trained model is made to determine whether the EEG of the epileptic seizure appears in any of the plurality of images. As a result, the determination can be made under the same conditions as when a specialist diagnoses an epilepsy attack, and the epilepsy attack can be determined with higher accuracy.

In the above embodiment, a plurality of test images representing brain waves are input to the trained model, and a predetermined time width is set based on the evaluation unit that evaluates the judgment accuracy of the trained model and the evaluation unit. A setting unit, a plurality of learning images representing brain waves cut out in a predetermined time width set by the setting unit, and information on whether or not the plurality of learning images represent brain waves of epilepsy. A learning unit that trains a learning model as training data and generates a new trained model may be further provided.

According to this aspect, a new trained model can be generated so that the determination accuracy is improved by adjusting the time width for cutting out the EEG data, and the epilepsy attack can be determined with higher accuracy.

In the above aspect, the evaluation unit may evaluate the determination accuracy of the trained model based on a plurality of indexes.

According to this aspect, the determination accuracy of the trained model can be evaluated from various aspects, and the time width for cutting out the electroencephalogram data can be adjusted more appropriately.

In the above embodiment, the plurality of indicators are an index showing whether an image showing an epilepsy attack can be correctly determined to have an epilepsy attack, and an index showing whether an image showing no epilepsy attack does not show an epilepsy attack. An index showing whether it can be judged correctly, an index showing whether an image showing an epilepsy attack is erroneously judged not to have an epilepsy attack, and an image showing an epilepsy attack showing an epilepsy attack. It may include an index indicating whether or not it is erroneously determined.

According to this aspect, it is possible to evaluate the judgment accuracy of the trained model from various aspects in the cases of true positive, true negative, false negative and false positive, and the time width for cutting out the EEG data is more appropriate. Can be adjusted.

In the above embodiment, the plurality of images may include waveforms shown in a plurality of colors corresponding to the plurality of electrodes for which the electroencephalogram data is measured.

According to this aspect, it is possible to distinguish a plurality of electrodes for which EEG data are measured and make a judgment under the same conditions as a specialist diagnoses an epilepsy attack, and it is possible to judge an epilepsy attack with higher accuracy. ..

In the above embodiment, the trained model may be a trained convolutional neural network.

According to this aspect, a trained model having a determination accuracy equal to or higher than that of a human can be used, and an epileptic seizure can be determined with higher accuracy.

In the above embodiment, the generation unit may generate a plurality of images by applying at least one of a low-pass filter, a high-pass filter, and a notch filter to the electroencephalogram data, and then cutting out the electroencephalogram data in a predetermined time width. good.

According to this aspect, it is possible to generate a plurality of images after removing noise that may be included in the electroencephalogram data, and it is possible to determine an epileptic seizure with higher accuracy.

The epilepsy determination system according to another aspect of the present invention is an epilepsy determination system including a measuring device for measuring brain wave data of a subject and an epilepsy determination device for determining epilepsy based on the brain wave data. The determination device cuts out the electroencephalogram data in a predetermined time width to generate a plurality of images representing the electroencephalogram, and inputs a plurality of images to the trained model, and depending on the trained model, one of the plurality of images. A determination unit that determines whether an epilepsy brain wave is appearing, and a notification unit that notifies a predetermined terminal when it is determined by the determination unit that an epilepsy brain wave is appearing in any of a plurality of images. And have.

According to this aspect, the EEG data is cut out in a predetermined time width, a plurality of images representing the EEG are generated, and the trained model is made to determine whether the EEG of the epileptic seizure appears in any of the plurality of images. As a result, the determination can be made under the same conditions as when a specialist diagnoses an epilepsy attack, and the epilepsy attack can be determined with higher accuracy. In addition, when it is determined that an electroencephalogram of an epileptic seizure is appearing, the burden of checking the electroencephalogram can be reduced by notifying a predetermined terminal.

The epilepsy determination method according to another aspect of the present invention is to acquire the electroencephalogram data of the subject, to cut out the electroencephalogram data in a predetermined time width to generate a plurality of images representing the electroencephalogram, and to generate a plurality of images. Is input to the trained model, and the trained model is used to determine whether the brain wave of the epileptic seizure appears in any of a plurality of images.

According to this aspect, the EEG data is cut out in a predetermined time width, a plurality of images representing the EEG are generated, and the trained model is made to determine whether the EEG of the epileptic seizure appears in any of the plurality of images. As a result, the determination can be made under the same conditions as when a specialist diagnoses an epilepsy attack, and the epilepsy attack can be determined with higher accuracy.

In the epilepsy determination program according to another aspect of the present invention, a computer provided in the epilepsy determination device is used as an acquisition unit for acquiring electroencephalogram data of a subject, and a plurality of electroencephalogram data are cut out in a predetermined time width to represent an electroencephalogram. It functions as a generation unit that generates an image and a determination unit that inputs a plurality of images to a trained model and determines whether an electroencephalogram of an epileptic seizure appears in any of the plurality of images by the trained model.

According to this aspect, the EEG data is cut out in a predetermined time width, a plurality of images representing the EEG are generated, and the trained model is made to determine whether the EEG of the epileptic seizure appears in any of the plurality of images. As a result, the determination can be made under the same conditions as when a specialist diagnoses an epilepsy attack, and the epilepsy attack can be determined with higher accuracy.

According to the present invention, it is possible to provide an epilepsy determination device, an epilepsy determination system, an epilepsy determination method, and an epilepsy determination program capable of determining an epilepsy attack with higher accuracy.

It is a figure which shows the functional block of the epilepsy determination system which concerns on embodiment of this invention. It is a figure which shows the physical structure of the epilepsy determination apparatus which concerns on this embodiment. This is an example of an image input to the trained model of the epilepsy determination device according to the present embodiment. It is a 1st graph which shows the determination accuracy of the epilepsy determination apparatus which concerns on this embodiment. It is a 2nd graph which shows the determination accuracy of the epilepsy determination apparatus which concerns on this embodiment. It is a flowchart of the epilepsy determination process executed by the epilepsy determination system which concerns on this embodiment. It is a flowchart of the learning process executed by the epilepsy determination system which concerns on this embodiment.

An embodiment of the present invention will be described with reference to the accompanying drawings. In each figure, those with the same reference numerals have the same or similar configurations.

FIG. 1 is a diagram showing a functional block of the epilepsy determination system 1 according to the embodiment of the present invention. The epilepsy determination system 1 includes an epilepsy determination device 10 and a measuring device 20. The measuring device 20 is a device that measures the brain waves of the subject, and measures the electrical activity of the subject's brain by a plurality of electrodes attached to the subject's head. The measuring device 20 may have, for example, 21 electrodes arranged according to the international 10-20 method, but the number and arrangement of the electrodes are arbitrary.

The epilepsy determination device 10 includes an acquisition unit 11, a generation unit 12, a determination unit 13, an evaluation unit 14, a setting unit 15, a learning unit 16, a notification unit 17, and a storage unit 18. The acquisition unit 11 acquires the brain wave data of the subject. The acquisition unit 11 may acquire the electroencephalogram data of the subject measured by the measuring device 20, but may acquire the electroencephalogram data of the subject previously stored in the storage unit 18 or another storage device.

The generation unit 12 cuts out the electroencephalogram data in a predetermined time width to generate a plurality of images representing the electroencephalogram. Here, the predetermined time width can be arbitrarily set, and may be, for example, 0.5 seconds, 1 second, 2 seconds, 5 seconds, 10 seconds, or the like. For example, when the predetermined time width is 10 seconds and the recording time of the brain wave data is 1 hour (3600 seconds), the generation unit 12 cuts out the brain wave data in the range of 0 to 10 seconds and represents the brain wave. An image is generated, brain wave data in the range of 1 to 11 seconds is cut out to generate a second image representing an electroencephalogram, this processing is continued, and finally, brain wave data of 3590 to 3600 seconds is cut out to represent an electroencephalogram 3591. You may generate an image. The generation unit 12 may shift the cutout range at equal intervals so that the range of the brain wave data cut out by the plurality of images overlaps, but the range of the brain wave data cut out by the plurality of images may not overlap. good.

The generation unit 12 may generate a plurality of images by applying at least one of a low-pass filter, a high-pass filter, and a notch filter to the electroencephalogram data, and then cutting out the electroencephalogram data in a predetermined time width. Here, the low-pass filter may be a filter that passes data having a frequency of, for example, 60 Hz or less, the high-pass filter may be a filter that passes data having a frequency of 0.5 Hz or more, for example, and the notch filter may be a filter that passes data having a frequency of 0.5 Hz or higher. It may be a filter that blocks the data of the frequency of. As a result, it is possible to generate a plurality of images after removing noise that may be included in the electroencephalogram data, and it is possible to determine an epileptic seizure with higher accuracy.

The determination unit 13 inputs a plurality of images to the trained model 13a, and causes the trained model 13a to determine whether the brain wave of the epileptic seizure appears in any of the plurality of images. In the epilepsy determination device 10 according to the present embodiment, the trained model 13a is a trained convolutional neural network (CNN). By using the trained convolutional neural network as the trained model 13a, it is possible to achieve a judgment accuracy equal to or higher than that of a human being, and it is possible to judge an epileptic seizure with higher accuracy.

The evaluation unit 14 inputs a plurality of test images representing brain waves into the trained model 13a, and evaluates the determination accuracy of the trained model 13a. Here, the evaluation unit 14 may evaluate the determination accuracy of the trained model 13a based on a plurality of indexes. As a result, the determination accuracy of the trained model 13a can be evaluated from various aspects, and the time width for cutting out the electroencephalogram data can be adjusted more appropriately.

More specifically, the multiple indicators show epileptic seizures for images that show epileptic seizures, and for images that do not show epileptic seizures. An index showing whether or not an epilepsy attack can be correctly determined, an index indicating whether or not an image showing an epilepsy attack is mistakenly judged as not showing an epilepsy attack, and an image showing no epilepsy attack show an epilepsy attack. It may include an index indicating whether or not it is erroneously determined. In this way, by evaluating the determination accuracy of the trained model 13a from multiple aspects in the cases of true positive, true negative, false negative, and false positive, the determination accuracy can be evaluated from multiple aspects. The time width for cutting out the EEG data can be adjusted more appropriately.

The setting unit 15 sets a predetermined time width based on the evaluation by the evaluation unit 14. The setting unit 15 may set a predetermined time width so that the evaluation by the evaluation unit 14 is improved.

The learning unit 16 includes a plurality of learning images representing brain waves cut out in a predetermined time width set by the setting unit 15, and information on whether or not the plurality of learning images represent epileptic seizure brain waves. , Is used as training data to train a learning model, and a new trained model is generated. Here, the plurality of learning images representing the brain waves may be images that have been measured by the measuring device 20 in the past and are associated with information regarding whether or not they represent an epileptic seizure. When the learning model is a neural network such as a convolutional neural network, the learning unit 16 may perform learning processing of the weighting coefficient of the neural network by an error back propagation method. In this way, the setting unit 15 can adjust the time width for cutting out the brain wave data, and the learning unit 16 can generate a new trained model so that the determination accuracy is improved, and the new trained model is determined. By mounting it on the unit 13, the epilepsy determination device 10 can determine an epilepsy attack with higher accuracy.

When the determination unit 13 determines that the brain wave of the epileptic seizure appears in any of the plurality of images, the notification unit 17 notifies a predetermined terminal. Here, the predetermined terminal may be an arbitrary information processing terminal, and may be, for example, a terminal used by a specialist, a terminal used by a person who cares for the subject, or a terminal used by the subject himself / herself. You may do it. When it is determined that an electroencephalogram of an epileptic seizure is appearing, the burden of checking the electroencephalogram can be reduced by notifying a predetermined terminal.

The storage unit 18 may store the learning data used by the learning unit 16. However, the learning unit 16 and the storage unit 18 may not be provided in the epilepsy determination device 10, and the learning process of the learning model may be executed by another device accessible via the communication network. In that case, the epilepsy determination device 10 may transmit the conditions of the learning process to another device performing the learning process and receive the information for constructing the trained model. The information for constructing the trained model is the information that identifies the training model, and in the case of a neural network, the number of layers, the type of layer, the connection of nodes between layers, and the weighting coefficient (including the threshold value) of the connection of nodes. ) And the type of activation function and the like.

FIG. 2 is a diagram showing a physical configuration of the epilepsy determination device 10 according to the present embodiment. The epilepsy determination device 10 includes a CPU (Central Processing Unit) 10a corresponding to a calculation unit, a RAM (Random Access Memory) 10b corresponding to a storage unit, a ROM (Read only Memory) 10c corresponding to a storage unit, and a communication unit. It has a 10d, an input unit 10e, and a display unit 10f. Each of these configurations is connected to each other via a bus so that data can be transmitted and received. In this example, the case where the epilepsy determination device 10 is composed of one computer will be described, but the epilepsy determination device 10 may be realized by combining a plurality of computers. Further, the configuration shown in FIG. 2 is an example, and the epilepsy determination device 10 may have a configuration other than these, or may not have a part of these configurations.

The CPU 10a is a control unit that controls execution of a program stored in the RAM 10b or ROM 10c, calculates data, and processes data. The CPU 10a is a calculation unit that executes a program (epilepsy determination program) for determining epilepsy based on an image representing an electroencephalogram. The CPU 10a receives various data from the input unit 10e and the communication unit 10d, displays the calculation result of the data on the display unit 10f, and stores the data in the RAM 10b and the ROM 10c.

The RAM 10b is a storage unit capable of rewriting data, and may be composed of, for example, a semiconductor storage element. The RAM 10b may store an epilepsy determination program executed by the CPU 10a. It should be noted that these are examples, and data other than these may be stored in the RAM 10b, or a part of these may not be stored.

The ROM 10c is a storage unit capable of reading data, and may be composed of, for example, a semiconductor storage element. The ROM 10c may store, for example, a search program or data that is not rewritten.

The communication unit 10d is an interface for connecting the epilepsy determination device 10 to another device. The communication unit 10d may be connected to a communication network such as the Internet or a LAN (Local Area Network).

The input unit 10e receives data input from the user, and may include, for example, a keyboard and a touch panel.

The display unit 10f visually displays the calculation result by the CPU 10a, and may be configured by, for example, an LCD (Liquid Crystal Display). The display unit 10f may display, for example, an image representing an electroencephalogram or a determination result.

The epilepsy determination program may be stored in a storage medium readable by a computer such as RAM 10b or ROM 10c and provided, or may be provided via a communication network connected by the communication unit 10d. In the epilepsy determination device 10, the CPU 10a executes the epilepsy determination program to realize various operations described with reference to FIG. 1. It should be noted that these physical configurations are examples and do not necessarily have to be independent configurations. For example, the epilepsy determination device 10 may include an LSI (Large-Scale Integration) in which the CPU 10a and the RAM 10b or ROM 10c are integrated.

FIG. 3 is an example of an image P input to the trained model of the epilepsy determination device 10 according to the present embodiment. The image P is an example of an image cut out from the brain wave data by the generation unit 12, and in this example, the brain wave data for 10 seconds is cut out.

Image P shows an electroencephalogram measured by 21 electrodes of the measuring device 20 and an electrocardiographic waveform. The electroencephalogram measured by the 21 electrodes is the waveform from the top to the 21st of the graph, and the electrocardiographic waveform is the waveform shown at the bottom of the graph.

The image P includes waveforms shown in a plurality of colors corresponding to the plurality of electrodes for which the electroencephalogram data was measured. In this example, the waveforms drawn next to each other on the top and bottom are shown in different colors, or the color of the waveform is changed to correspond to the arrangement of the electrodes, but the color selection is arbitrary. By color-coding in this way, it is possible to distinguish between multiple electrodes that have measured EEG data and make a judgment under the same conditions as a specialist diagnosing an epilepsy attack, and to judge an epilepsy attack with higher accuracy. Can be done.

FIG. 4 is a first graph showing the determination accuracy of the epilepsy determination device 10 according to the present embodiment. The first graph shows the results of measuring brain wave data for each of 24 subjects for several tens of hours and testing whether the epilepsy determination device 10 according to the present embodiment can correctly determine an epilepsy attack. The first graph shows the time width (Time Window) set by the setting unit 15 on the horizontal axis, and shows the determination accuracy of the trained model in which the learning process is performed in the time width set on the vertical axis. The determination accuracy is a true positive value evaluated by the evaluation unit 14. That is, the determination accuracy shown in the first graph is a determination accuracy evaluated based on an index indicating whether or not an image showing an epileptic seizure can be correctly determined to have an epileptic seizure.

According to the first graph, when the time width is 0.5 seconds, the median value represented by the boxplot is about 0, the upper quartile is slightly larger than 0.1, and the maximum value is 0.2. It is a degree. When the time width is 1 second, the minimum value represented by the boxplot is about 0, the lower quartile is about 0.5, the median is slightly larger than 0.7, and the upper side. The quartile is about 0.9 and the maximum value is about 1.0. When the time width is 2 seconds, the minimum value represented by the boxplot is about 0, the lower quartile is about 0.4, the median is about 0.7, and the upper side. The quartile is about 0.9 and the maximum value is about 1.0. When the time width is 5 seconds, the minimum value represented by the boxplot is about 0, the lower quartile is about 0.5, the median is slightly smaller than 0.7, and the upper side. The quartile is slightly smaller than 0.9 and the maximum value is slightly smaller than 1.0. When the time width is 10 seconds, the minimum value represented by the boxplot is about 0, the lower quartile is about 0.2, the median is about 0.6, and the upper side. The quartile is slightly smaller than 0.8, and the maximum value is about 1.0.

As described above, the time width for cutting out the electroencephalogram data affects whether or not the learning process of the learning model can be performed appropriately, and may be a factor that changes the determination accuracy. In the epilepsy determination device 10 according to the present embodiment, a new trained model can be generated so that the determination accuracy is improved by adjusting the time width for cutting out the electroencephalogram data, and the epilepsy attack is determined with higher accuracy. be able to.

FIG. 5 is a second graph showing the determination accuracy of the epilepsy determination device 10 according to the present embodiment. The second graph shows the results of measuring brain wave data for each of 24 subjects for several tens of hours and testing whether the normal state can be correctly determined by the epilepsy determination device 10 according to the present embodiment. The second graph shows the time width (Time Window) set by the setting unit 15 on the horizontal axis, and shows the determination accuracy of the trained model in which the learning process is performed in the time width set on the vertical axis. The determination accuracy is a true negative value evaluated by the evaluation unit 14. That is, the determination accuracy shown in the second graph is a determination accuracy evaluated based on an index indicating whether or not an image showing no epileptic seizure can be correctly determined as having no epileptic seizure.

According to the second graph, when the time width is 0.5 seconds, 1 second, 2 seconds, 5 seconds and 10 seconds, the minimum value represented by the boxplot is about 0.97 to 0.99, which is lower. The side quadrant is about 0.98 to 0.99, the median is about 0.99, the upper quadrant is about 0.995, and the maximum value is about 1.0.

Comparing the first graph of FIG. 4 and the second graph of FIG. 5, it can be confirmed that the relationship between the time width for cutting out the electroencephalogram data and the determination accuracy differs depending on the index used for evaluating the determination accuracy. More specifically, in the first graph, the median determination accuracy is about 0 when the time width is 0.5 seconds, whereas in the second graph, it is close to 1.0. Further, in the first graph, when the time width is lengthened from 1 second to 10 seconds, the median judgment accuracy decreases, whereas in the second graph, the time width is increased from 1 second to 10 seconds. There is a difference that the median judgment accuracy increases as the length increases. The epilepsy determination device 10 according to the present embodiment evaluates the determination accuracy of the trained model based on a plurality of indexes, and collects the electroencephalogram data so as to improve the determination accuracy evaluated based on the plurality of indexes in a well-balanced manner. You may set the time width to cut out. For example, the time width for cutting out the electroencephalogram data may be set so as to maximize the sum of the determination accuracy evaluated based on a plurality of indexes.

FIG. 6 is a flowchart of the epilepsy determination process executed by the epilepsy determination system 1 according to the present embodiment. First, the electroencephalogram data of the subject is measured by the measuring device 20 (S10). Subsequent processing may be executed after the electroencephalogram data is accumulated, or may be sequentially executed while measuring the electroencephalogram data.

The generation unit 12 of the epilepsy determination device 10 filters the acquired electroencephalogram data and executes preprocessing (S11). After that, the generation unit 12 cuts out the electroencephalogram data in a predetermined time width to generate a plurality of images (S12).

The determination unit 13 of the epilepsy determination device 10 inputs a plurality of images into a trained model such as CNN, and determines whether or not an electroencephalogram of an epilepsy attack appears (S13). Here, when it is determined that there is an epilepsy attack (S14: YES), the notification unit 17 of the epilepsy determination device 10 notifies that the epilepsy attack has been detected at a predetermined terminal (S15). On the other hand, if it is not determined that there is an epilepsy attack (S14: NO), the epilepsy determination process ends.

According to the epilepsy determination device 10 according to the present embodiment, the electroencephalogram data is cut out in a predetermined time width to generate a plurality of images representing the brain waves, and whether the brain waves of the epilepsy attack appear in any of the plurality of images. By letting the trained model make a judgment, it is possible to make a judgment under the same conditions as a specialist diagnoses an epilepsy attack, and it is possible to judge an epilepsy attack with higher accuracy.

Further, according to the epilepsy determination system 1 according to the present embodiment, the electroencephalogram data is cut out in a predetermined time width to generate a plurality of images representing the brain waves, and the brain waves of the epilepsy attack appear in any of the plurality of images. By letting the trained model determine whether or not epilepsy is occurring, the determination can be made under the same conditions as when a specialist diagnoses an epilepsy attack, and the epilepsy attack can be determined with higher accuracy. In addition, when it is determined that an electroencephalogram of an epileptic seizure is appearing, the burden of checking the electroencephalogram can be reduced by notifying a predetermined terminal.

FIG. 7 is a flowchart of the learning process executed by the epilepsy determination system 1 according to the present embodiment. First, the evaluation unit 14 of the epilepsy determination device 10 inputs a test image into the trained model and evaluates the determination accuracy (S20). The setting unit 15 sets the time width for cutting out the electroencephalogram data based on the evaluation by the evaluation unit 14 (S21).

The learning unit 16 cuts out brain wave data within a set time width, executes a learning process of the learning model, and generates a new learned model (S22). Then, the epilepsy determination device 10 mounts a new trained model on the determination unit 13 (S23). With the above, the learning process is completed.

In this way, the time width for cutting out the EEG data can be adjusted to generate a new trained model so that the determination accuracy is improved, and the epilepsy attack can be determined with higher accuracy. Further, as the learning data is accumulated, it is possible to generate a trained model with higher judgment accuracy and update the trained model to be mounted on the judgment unit 13, so that the epilepsy judgment system 1 continues to be operated. , Will be able to determine epileptic seizures with higher accuracy.

The embodiments described above are for facilitating the understanding of the present invention, and are not for limiting the interpretation of the present invention. Each element included in the embodiment and its arrangement, material, condition, shape, size, and the like are not limited to those exemplified, and can be appropriately changed. Further, it is possible to partially replace or combine the configurations shown in different embodiments.

1 ... Epilepsy determination system, 10 ... Epilepsy determination device, 10a ... CPU, 10b ... RAM, 10c ... ROM, 10d ... Communication unit, 10e ... Input unit, 10f ... Display unit, 11 ... Acquisition unit, 12 ... Generation unit, 13 ... Judgment unit, 13a ... Learned model, 14 ... Evaluation unit, 15 ... Setting unit, 16 ... Learning unit, 17 ... Notification unit, 18 ... Storage unit, 20 ... Measuring device

Claims (10)

The acquisition unit that acquires the brain wave data of the subject,
A generation unit that cuts out the brain wave data in a predetermined time width and generates a plurality of images representing the brain waves.
A determination unit that inputs the plurality of images to the trained model and determines whether or not the brain wave of the epileptic seizure appears in any of the plurality of images by the trained model.
An epilepsy determination device equipped with. An evaluation unit that inputs a plurality of test images representing brain waves into the trained model and evaluates the determination accuracy of the trained model.
Based on the evaluation by the evaluation unit, the setting unit that sets the predetermined time width and the setting unit.
Learning a plurality of learning images representing brain waves cut out in the predetermined time width set by the setting unit, and information on whether or not the plurality of learning images represent epileptic seizure brain waves. A learning unit that trains a learning model as data and generates a new trained model,
The epilepsy determination device according to claim 1. The evaluation unit evaluates the determination accuracy of the trained model based on a plurality of indexes.
The epilepsy determination device according to claim 2. The plurality of indicators are
An index showing whether an image showing an epileptic seizure can be correctly determined to have an epileptic seizure,
For images that do not show seizure, an index that indicates whether it can be correctly determined that seizure does not appear,
An index that indicates whether an image showing an epileptic seizure is erroneously determined not to have an epileptic seizure,
Includes an index that indicates whether an image that does not show seizure is erroneously determined to have seizure.
The epilepsy determination device according to claim 3. The plurality of images include waveforms shown in a plurality of colors corresponding to the plurality of electrodes for which the electroencephalogram data was measured.
The epilepsy determination device according to any one of claims 1 to 4. The trained model is a trained convolutional neural network.
The epilepsy determination device according to any one of claims 1 to 5. The generation unit applies at least one of a low-pass filter, a high-pass filter, and a notch filter to the electroencephalogram data, and then cuts out the electroencephalogram data in the predetermined time width to generate the plurality of images.
The epilepsy determination device according to any one of claims 1 to 6. An epilepsy determination system including a measuring device for measuring an electroencephalogram data of a subject and an epilepsy determination device for determining epilepsy based on the electroencephalogram data.
The epilepsy determination device is
A generation unit that cuts out the brain wave data in a predetermined time width and generates a plurality of images representing the brain waves.
A determination unit that inputs the plurality of images to the trained model and determines whether or not the brain wave of the epileptic seizure appears in any of the plurality of images by the trained model.
It has a notification unit for notifying a predetermined terminal when it is determined by the determination unit that an electroencephalogram of an epileptic seizure appears in any of the plurality of images.
Epilepsy judgment system. Acquiring the subject's EEG data and
By cutting out the brain wave data in a predetermined time width to generate a plurality of images representing the brain waves,
The plurality of images are input to the trained model, and the trained model is used to determine whether the brain wave of the epileptic seizure appears in any of the plurality of images.
Epilepsy determination method including. A computer equipped with an epilepsy determination device,
Acquisition department that acquires the brain wave data of the subject,
An electroencephalogram data is cut out in a predetermined time width to generate a plurality of images representing brain waves, and the plurality of images are input to a trained model, and one of the plurality of images is input by the trained model. Judgment unit that determines whether brain waves of epileptic attacks are appearing,
Epilepsy judgment program that functions as.

Images