JP7653688B2 - Information processing device, determination method, and determination program - Google Patents

Description

The present invention relates to an information processing device etc. that determines whether a subject has a brain disease based on biological signals emitted from the subject's brain.

Research is being conducted into diagnosing whether a subject has a brain disease using the subject's biosignal data. For example, the following Patent Document 1 discloses a technology that extracts a specific feature from the subject's cerebral blood flow data while the subject is performing a cognitive task, and determines the degree of the subject's cognitive impairment from the feature.

International Publication No. 2012/165602 (published December 6, 2012)

However, the above conventional technology has a problem in that it is difficult to determine appropriate feature quantities in the case of a disease in which it is unclear what features will appear in the biosignal data. For example, in Patent Document 1, cerebral blood flow data measured by fNIRS (functional near-infrared spectroscopy) is filtered with various filters, and the maximum amplitude of the signal that has passed through the filters is used as a feature quantity to determine the degree of cognitive impairment. However, there is no guarantee that the above feature quantities are optimal in determining the degree of cognitive impairment, and there is a possibility that information correlated with the degree of cognitive impairment may be missing without being extracted as a feature quantity.

One aspect of the present invention aims to realize an information processing device, etc. that can determine whether a subject has a brain disease without performing processing such as feature point extraction on biosignal data.

In order to solve the above problem, an information processing device according to one embodiment of the present invention includes a data acquisition unit that acquires time series data of biosignals emitted from the brain of a subject, and a judgment unit that inputs the time series data acquired by the data acquisition unit into a trained model constructed by machine learning the relationship between a brain disease and the time series data of biosignals emitted from the brain of a patient with the brain disease, and makes a judgment regarding the brain disease of the subject based on the output of the trained model.

In addition, a determination method according to one aspect of the present invention is a determination method using an information processing device, and includes a data acquisition step of acquiring time series data of biosignals emitted from the brain of a subject, and a determination step of inputting the time series data acquired in the data acquisition step into a trained model constructed by machine learning the relationship between the time series data of the biosignals emitted from the brain and brain diseases, and making a determination regarding the brain disease of the subject based on the output of the trained model.

According to one aspect of the present invention, it is possible to determine whether a subject has a brain disease without performing processing such as feature point extraction on the biosignal data.

1 is a block diagram showing an example of a configuration of a main part of an information processing device according to a first embodiment of the present invention; FIG. 13 is a diagram showing an example of the configuration of a trained model constructed using electroencephalogram data as training data. FIG. 13 is a diagram showing the discrimination success rate in an accuracy demonstration experiment of the above-mentioned trained model. 4 is a flowchart illustrating an example of a process executed by the information processing device. FIG. 11 is a diagram showing an example of the configuration of a trained model constructed using magnetoencephalography data as training data in accordance with a second embodiment of the present invention. FIG. 13 is a diagram showing the discrimination success rate in an accuracy verification experiment of a trained model constructed using magnetoencephalography data as training data.

[Embodiment 1]
(Device configuration)
The configuration of an information processing device 1 according to an embodiment of the present invention will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the main configuration of the information processing device 1. The information processing device 1 is a device that performs a determination regarding a brain disease of a subject, as will be described in detail below. Note that, although an example in which the information processing device 1 is one device (computer) will be described below, an information processing system in which multiple devices are connected to be able to communicate with each other may also be constructed, and each device, which is a component of the information processing device, may have each function of the information processing device 1.

The information processing device 1 includes a control unit 10 that controls each unit of the information processing device 1, a storage unit 20 that stores various data used by the information processing device 1, an input unit 30 that accepts input to the information processing device 1, and an output unit 40 that causes the information processing device 1 to output information. The control unit 10 also includes a data acquisition unit 101, a normalization unit 102, a judgment unit 103, and an output control unit 104. The storage unit 20 stores brainwave data 201 and a trained model 202. The storage unit 20, the input unit 30, and the output unit 40 may be external devices connected to the information processing device 1 by wire or wirelessly.

The data acquisition unit 101 acquires time series data of biosignals emitted from the brain of the subject of the judgment by the information processing device 1. Specifically, the data acquisition unit 101 acquires the electroencephalogram data of the subject measured by an electroencephalograph as the above-mentioned time series data. Note that the data acquired by the data acquisition unit 101 may be time series data of biosignals emitted from the brain of the subject, and is not limited to electroencephalogram data. For example, the data acquisition unit 101 may acquire magnetoencephalogram data described in embodiment 2, or may acquire data measured by fMRI (functional magnetic resonance imaging) or fNIRS. However, from the viewpoint of ease of measurement, it is preferable to acquire electroencephalogram data or magnetoencephalogram data. Note that, although details will be described later, the data acquisition unit 101 acquires data that can be used as input data for the trained model 202 (data of the same type as the teacher data used for training).

The normalization unit 102 normalizes the time series data of the biosignals of multiple channels. Specifically, when the data acquisition unit 101 acquires electroencephalogram data of multiple channels, the normalization unit 102 normalizes the electroencephalogram data. Details of normalization will be explained in the "Processing flow" section below.

The judgment unit 103 makes a judgment regarding the subject's brain disease. Specifically, the judgment unit 103 inputs the EEG data acquired by the data acquisition unit 101 to the trained model 202, and makes the above judgment based on the output of the trained model 202. Details of the judgment will also be explained in the "Processing flow" section below.

The output control unit 104 causes the output unit 40 to output the judgment result of the judgment unit 103. For example, if the output unit 40 is a display device that outputs an image, the output control unit 104 causes the output unit 40 to display an image showing the judgment result. The manner in which the judgment result is output is not particularly limited, and may be, for example, audio output or printed output. The output control unit 104 may also upload the judgment result to a webpage or the like to allow the subject to view the judgment result, or may notify the subject of the judgment result by email or the like.

The electroencephalogram data 201 is the electroencephalogram data of the subject measured by an electroencephalograph, and indicates fluctuations in the electrical potential generated by the subject's brain. Specifically, the electroencephalogram data 201 is waveform data measured over a predetermined period of time by a predetermined number of detection electrodes attached to the subject's head.

The trained model 202 is a judgment model for making a judgment regarding a subject's brain disease, constructed by machine learning the relationship between a brain disease and time series data of biosignals emitted from the brain of a patient with the brain disease. Specifically, the trained model 202 is constructed by machine learning the relationship between the type of dementia and the electroencephalogram data of various types of dementia patients. Details of the trained model 202 will be explained in the "Trained Model" section below.

As described above, the information processing device 1 includes a data acquisition unit 101 that acquires electroencephalogram data as time-series data of biological signals emitted from the subject's brain. The information processing device 1 also includes a judgment unit 103 that makes a judgment regarding the subject's brain disease. More specifically, the judgment unit 103 inputs the electroencephalogram data acquired by the data acquisition unit 101 into a trained model 202 that has been constructed by machine learning the relationship between the brain disease and the electroencephalogram data of patients with the brain disease, and makes the above judgment based on the output of the trained model 202.

In this way, the information processing device 1 can input EEG data that has not been subjected to processing such as feature point extraction into the trained model 202, thereby making it possible to make a judgment regarding the subject's brain disease. Furthermore, in order to obtain the above judgment result, the subject only needs to measure his or her EEG. In other words, the information processing device 1 also has the advantage of imposing less physical and economic burden on the subject compared to dementia diagnosis using amyloid PET (positron emission tomography) or the like. Furthermore, as will be described in detail later, the information processing device 1 can also distinguish the type of dementia, which was difficult to do with conventional technology.

(Pre-trained model)
Fig. 2 is a diagram showing an example of the configuration of the trained model 202. The trained model 202 shown in Fig. 2 is a convolutional neural network (CNN) constructed to receive electroencephalogram data of a subject of judgment as input and output the probability that the subject belongs to a predetermined cluster.

The clusters to be judged by the trained model 202 in FIG. 2 are four: healthy, AD (Alzheimer's disease), DLB (dementia with Lewy bodies), and iNPH (idiopathic normal pressure hydrocephalus). Of course, the clusters may be set appropriately so that the desired judgment item can be judged, and are not limited to this example. For example, if it is only necessary to judge whether or not it is AD, only two clusters, AD and healthy, may be used. Also, for example, if it is necessary to judge the degree of dementia, a cluster may be set for each degree to be judged. Furthermore, as described in embodiment 2, a cluster may be set for cognitive dysfunction other than dementia.

The EEG data input to the trained model 202 consists of 19 waveform data measured over a specified period of time by 19 detection electrodes attached to the subject's head. For example, if EEG measurements are performed for about 5 to 10 minutes, the specified period may be 2 seconds of that time, and the data may be sampled at 200 Hz. In this case, the input layer includes 19 (channels) x 400 (nodes). Of course, the EEG measurement site, the number of waveform data (number of channels), and the specified period may be changed as appropriate within the range in which the required judgment accuracy can be maintained.

The trained model 202 is designed to first perform convolution on the above-mentioned EEG data input to the input layer. More specifically, the trained model 202 in FIG. 2 is configured to perform convolution using a filter of size 19×4. In this way, when the EEG data includes biosignals of multiple channels measured simultaneously, it is preferable that the trained model 202 includes a layer that convolves the biosignals of multiple channels at the same time. This makes it possible to aggregate the features of the EEG data of multiple channels.

This configuration is based on the design concept of giving the trained model 202 a function similar to feature point extraction by frequency analysis. In other words, the trained model 202 can be said to include an extraction unit that extracts feature points for each frequency component of the EEG data. However, the feature extraction by the trained model 202 is optimized by machine learning and is different in nature from the uniform feature extraction performed by conventional technology.

After the above convolution, the network structure may be the same as that of a general CNN. For example, after the convolution and pooling are repeated a predetermined number of times, the network structure may be such that the output layer is connected via a predetermined number of fully connected networks. In addition, for example, max pooling may be performed in the pooling. As described above, the data obtained by the first convolution is an aggregate of data from multiple channels. For this reason, it is also possible to apply the network structure of a CNN with a one-channel input layer (for example, a CNN for speech recognition) as it is as the network structure after the first convolution.

(Machine Learning)
As described above, the trained model 202 includes multiple intermediate layers between the input layer and the output layer, and is constructed by optimizing the weights between the layers through deep learning. The training data used for training is electroencephalogram data obtained by measuring the electroencephalograms of multiple subjects who are known in advance to belong to one of the four clusters described above.

For example, if the trained model 202 is constructed by machine learning using electroencephalogram data emitted from the brains of patients with different types of dementia as training data, the determination unit 103 will determine the type of dementia of the subject based on the output of the trained model 202. In this way, the information processing device 1 can determine types of dementia that were difficult to determine using conventional technology.

As described in the above-mentioned Patent Document 1, in conventional dementia diagnosis, the brain activity state of a subject performing a cognitive task is measured, but the above-mentioned EEG measurement may be performed when the subject is at rest. "Resting" refers to a state in which the eyes are closed and no work is being performed. In this case, the trained model 202 is constructed by machine learning using EEG data measured when the subject is at rest. In this case, the data acquisition unit 101 acquires the EEG data measured when the subject is at rest as time-series data. This makes it possible to determine whether the subject has a brain disease without making the subject perform complicated tasks such as performing cognitive tasks. Another advantage is that it is possible to determine whether the subject has a brain disease when the subject has difficulty performing cognitive tasks.

Note that, up until now, there has been no knowledge that resting EEG data contains waveform patterns specific to dementia, and therefore resting EEG data has not been used to diagnose dementia. At this technical level, the inventors of the present application have discovered that a trained model 202 constructed using resting EEG data can determine not only whether a person has dementia or is healthy, but also the type of dementia. The accuracy of the determination will be described below with reference to FIG. 3.

(Experimental Results)
An experiment was conducted by inputting EEG data from a healthy subject, an AD patient, a DLB patient, and an iNPH patient into the trained model 202 to confirm the discrimination accuracy of the trained model 202. The results are shown in Fig. 3. Fig. 3 is a diagram showing the discrimination success rate of the trained model 202 in the above experiment.

The discrimination success rate was calculated by K-fold cross-validation. Specifically, K=10, and the discrimination success rate was calculated as the average value of 10 validation results ±1 SD (standard deviation). In addition, resting EEG data was used to construct the trained model 202, and the EEG data input to the trained model 202 in the experiment was also measured at rest.

As shown in the figure, the trained model 202 was able to discriminate between healthy subjects and AD patients with a high accuracy of 88.3±15.0(%). It was also able to discriminate between healthy subjects and DLB patients, healthy subjects and iNPH patients, AD patients and DLB patients, and AD patients and iNPH patients with 100±0(%). It also had a sufficient discrimination accuracy of 82.4±13.1(%) for DLB patients and iNPH patients, and a high discrimination accuracy of 83.0±10.4(%) for the four types (healthy subjects, AD patients, DLB patients, and iNPH patients).

As described above, the information processing device 1 was able to determine with high accuracy whether the subject was a healthy individual, an AD patient, a DLB patient, or an iNPH patient, using the trained model 202 constructed from resting EEG data and the subject's resting EEG data. The ability to make such highly accurate discrimination using resting EEG data was something that could not have been predicted at the state of the art at the time of filing this application.

(Processing flow)
The flow of the process (determination method) executed by the information processing device 1 will be described with reference to Fig. 4. Fig. 4 is a flowchart showing an example of the process executed by the information processing device 1.

In S1 (data acquisition step), the data acquisition unit 101 acquires brainwave data 201. The brainwave data 201 is data obtained by measuring the subject's brainwaves, and is input in advance to the information processing device 1 via the input unit 30 and stored in the memory unit 20. Note that in S1, the data acquisition unit 101 acquires brainwave data 201 measured in the same manner as the brainwave data used as teacher data when constructing the trained model 202. For example, when resting brainwave data is used as teacher data, the data acquisition unit 101 acquires brainwave data 201 measured at rest.

In S2, the normalization unit 102 normalizes the EEG data 201 acquired in S1. Since the target of normalization is the EEG data 201, which is time-series data of multiple channels, the normalization method can be performed by performing normalization in the time direction for each channel or by performing normalization in the channel direction for each time, but the former method is preferable. In other words, it is preferable that the normalization unit 102 normalizes each of the EEG data of multiple channels in the time direction. This is because the inventors of the present application have found that the accuracy of judgment is higher when the EEG data of multiple channels is normalized in the time direction through experiments. When normalization is performed, the judgment unit 103 inputs the normalized EEG data into the trained model 202 to make a judgment regarding the brain disease of the subject.

In addition, when the sampling rate of the EEG data 201 is different from the sampling rate of the EEG data used as the teacher data, the normalization unit 102 may perform resampling from the EEG data 201 before or after normalization. By performing resampling, it is possible to obtain EEG data having the same sampling rate as the sampling rate of the EEG data used as the teacher data.

In S3 (determination step), the determination unit 103 inputs the EEG data normalized in S2 into the trained model 202 and calculates the probability that the subject belongs to each cluster (e.g., healthy, AD, DLB, and iNPH). Then, the determination unit 103 determines which cluster the subject belongs to based on the calculated probability value. For example, the determination unit 103 may determine that the subject belongs to the cluster with the highest calculated probability value.

In S4, the output control unit 104 outputs the judgment result of S4 to the output unit 40. For example, if the output unit 40 is a display device, the output control unit 104 may cause the output unit 40 to display an image showing the judgment result of S4. In this case, the probability values output by the trained model 202 and clusters with second or subsequent probability values may also be displayed. The flow in FIG. 4 ends when the processing of S5 ends.

As described above, the determination method using the information processing device 1 includes a data acquisition step (S1) of acquiring electroencephalogram data, which is time-series data of biological signals emitted from the subject's brain, and a determination step (S3) of determining whether the subject has a brain disease. In addition, in S3, the electroencephalogram data acquired in S1 is input to a trained model 202 constructed by machine learning the relationship between the electroencephalogram data and brain diseases, and a determination of the subject's brain disease is made based on the output of the trained model 202.

According to the above-mentioned determination method, a determination regarding a subject's brain disease can be made simply by inputting EEG data that has not been subjected to processing such as feature point extraction into the trained model 202. Moreover, in order to obtain the above-mentioned determination result, the subject only needs to measure his/her EEG, which places a small physical and economic burden on the subject. Furthermore, according to the above-mentioned determination method, it is also possible to distinguish the type of dementia, which was difficult with conventional technology.

[Embodiment 2]
Other embodiments of the present invention will be described below. For ease of explanation, the same reference numerals are given to members having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

(Device configuration and processing flow)
In this embodiment, an example will be described in which magnetoencephalography data is used as time-series data of biosignals emitted from the brain. Magnetoencephalography data is waveform data obtained by measuring magnetic fields generated by electrical activity of the brain using a measuring device (e.g., a superconducting quantum interference meter).

The information processing device 1 of this embodiment differs from the information processing device 1 of embodiment 1 (see FIG. 1) in that it uses a trained model 200A constructed using magnetoencephalography data as training data instead of the trained model 200. The flow of the process (determination method) executed by the information processing device 1 of this embodiment is similar to the example of FIG. 4, except that the acquired time series data is changed from electroencephalography data to magnetoencephalography data.

(Pre-trained models and machine learning)
5 is a diagram showing an example of the configuration of a trained model 200A constructed using magnetoencephalography data as training data. When magnetoencephalography data is used, it is desirable to use time-series data of multiple channels, as in the case of using electroencephalography data. The trained model 202A is a CNN constructed to input magnetoencephalography data of a subject to be judged and output the probability that the subject belongs to a predetermined cluster. The clusters in the example of FIG. 5 are five: healthy, iNPH, epilepsy, Parkinson's disease, and spinal cord injury.

The magnetoencephalogram data input to the trained model 202A is 160-channel time series data (waveform data) generated by measuring the magnetic field at 160 locations on the head for a specified period of time. For this reason, the input layer of the trained model 202 is configured to include nodes equal to 160 (channels) x the number of samplings in the specified period. For example, if the specified period is 800 ms and the sampling rate is 1 kHz, the input layer will include 160 (channels) x 800 (nodes). Of course, the magnetoencephalogram measurement site, the number of waveform data (number of channels), and the measurement time can be changed as appropriate within the range in which the required judgment accuracy can be maintained.

Trained model 202A, like trained model 202 in embodiment 1, is designed to perform convolution using a filter across multiple channels on the data input to the input layer. This makes it possible to aggregate the features of magnetoencephalography data from multiple channels. The network structure after the above convolution is the same as that of trained model 202. However, when there are five clusters to be judged as described above, the number of nodes in the output layer is five.

Machine learning in constructing the trained model 202A can be performed in the same manner as the trained model 202. The teacher data used for learning may be magnetoencephalogram data obtained by measuring magnetoencephalograms of multiple subjects who are known in advance to belong to any of the five clusters described above. As with the EEG measurement in embodiment 1, the magnetoencephalogram may be measured while the subject is at rest. In this case, the trained model 202A is constructed by machine learning using magnetoencephalogram data measured while the subject is at rest, and the data acquisition unit 101 acquires the magnetoencephalogram measured while the subject is at rest as time-series data. This makes it possible to determine the subject's brain disease without having the subject perform complicated tasks such as performing cognitive tasks. There is also the advantage that it is possible to determine subjects who have difficulty performing cognitive tasks.

(Experimental Results)
Magnetoencephalogram data of a healthy subject, an iNPH patient, an epilepsy patient, a Parkinson's disease patient, and a spinal cord injury patient were input to the trained model 202A, and an experiment was conducted to confirm the accuracy of discrimination by the trained model 202A. The results are shown in the "Discrimination Success Rate (CNN)" section of FIG. 6. FIG. 6 is a diagram showing the discrimination success rate of the trained model 202A constructed using magnetoencephalogram data as training data. Note that magnetoencephalogram data at rest was used to construct the trained model 202A, and the magnetoencephalogram data input to the trained model 202A in the experiment was also measured at rest.

As a control experiment, discrimination was performed using a support vector machine (SVM) trained on spectrograms obtained by frequency analysis of magnetoencephalography data from healthy subjects and patients with the above four types of cognitive impairment. The results are shown in the "Discrimination success rate (SVM)" section of Figure 6. When the discrimination success rates of CNN and SVM were compared using Student's t-test, the accuracy of CNN was statistically significantly higher for some items, as shown in the "p-value" in Figure 6.

As shown in the figure, the trained model 202A was able to discriminate between iNPH patients and epilepsy patients with an extremely high accuracy of 96.8±3.9%. It also achieved high accuracy of 88.0±9.9% for discriminating between iNPH patients and healthy subjects. In this discrimination, many of the healthy subjects who were erroneously determined to be iNPH patients were young. Therefore, when the discrimination accuracy was calculated for only the healthy subjects aged 40 years or older, it was 95.2±6.7% (see the "Age correction" item in Figure 5). From these results, it can be said that it is more preferable to limit the subjects to those aged 40 years or older when discriminating between iNPH patients using magnetoencephalography data.

In addition, the discrimination accuracy was high for iNPH patients and Parkinson's disease patients, and for iNPH patients and spinal cord injury patients, at 89.3±12.7% and 88.2±11.9%, respectively. Thus, it was found that trained model 202A can discriminate between iNPH patients and not only healthy subjects, but also patients with other diseases, such as Parkinson's disease, that affect magnetoencephalography data.

In addition, the discrimination success rate was nearly 70% when discriminating between a group of patients with the above four types of diseases and healthy subjects. In all experimental cases, the discrimination success rate using trained model 202A exceeded the discrimination success rate using SVM. The fact that such highly accurate discrimination could be performed using magnetoencephalography data at rest was something that could not have been predicted at the state of the art at the time of filing this application.

As described above, the information processing device 1 of this embodiment can determine whether the subject is a healthy individual or has one of the four types of cognitive impairments described above, using the trained model 202A constructed from magnetoencephalography data at rest and the subject's magnetoencephalography data at rest. This configuration has the advantage of placing less strain on the subject compared to having the subject perform a cognitive task. It also has the advantage of being able to determine whether the subject has difficulty performing a cognitive task.

[About diagnosable diseases]
The information processing device 1 can determine whether or not a subject has a disease that affects electroencephalograms or magnetoencephalograms and is accompanied by abnormal cortical activity. When determining whether or not a subject has a disease other than those described in the above embodiments, a trained model corresponding to the disease to be determined may be constructed in advance. The method of constructing the trained model is the same as in the above embodiments, and machine learning may be performed using electroencephalogram or magnetoencephalogram data of a patient with the disease to be determined as teacher data.

[Software implementation example]
The control block of the information processing device 1 (particularly each unit included in the control unit 10) may be realized by a logic circuit (hardware) formed on an integrated circuit (IC chip) or the like, or may be realized by software.

In the latter case, the information processing device 1 includes a computer that executes the instructions of a program (determination program), which is software that realizes each function. This computer includes, for example, one or more processors, and a computer-readable recording medium that stores the program. The object of the present invention is achieved by the processor reading the program from the recording medium and executing it in the computer. The processor can be, for example, a CPU (Central Processing Unit). The recording medium can be a "non-transient tangible medium", such as a ROM (Read Only Memory), as well as a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, etc. The computer can also include a RAM (Random Access Memory) that expands the program. The program can be supplied to the computer via any transmission medium (such as a communication network or a broadcast wave) that can transmit the program. One aspect of the present invention can also be realized in the form of a data signal embedded in a carrier wave, in which the program is embodied by electronic transmission.

The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.
[Additional Notes]
An information processing device according to one embodiment of the present invention includes a data acquisition unit that acquires time series data of biosignals emitted from the brain of a subject, and a determination unit that inputs the time series data acquired by the data acquisition unit into a trained model constructed by machine learning the relationship between brain diseases and the time series data of biosignals emitted from the brains of patients with the brain disease, and makes a determination regarding the brain disease of the subject based on the output of the trained model.
In the information processing device according to one aspect of the present invention, the trained model may be constructed by machine learning using time series data of biological signals emitted from the brains of patients with different types of dementia as teacher data. In this case, the determination unit may determine the type of dementia of the subject based on an output of the trained model.
In an information processing device according to one aspect of the present invention, the time series data may include biosignals of multiple channels measured simultaneously, and the trained model may be a convolutional neural network including a layer that convolves the biosignals of the multiple channels at the same time.
An information processing device according to one embodiment of the present invention includes a normalization unit that performs time-direction normalization on each of the time series data of multiple channels, and the determination unit inputs the normalized time series data of the biological signal to the trained model and performs a determination regarding brain disease of the subject based on the output of the trained model.
In the information processing device according to an aspect of the present invention, the trained model may be constructed by machine learning using electroencephalogram data measured while the subject is at rest. In this case, the data acquisition unit may acquire the electroencephalogram data measured while the subject is at rest as the time-series data.
In the information processing device according to an aspect of the present invention, the trained model may be constructed by machine learning using magnetoencephalogram data measured while the subject is at rest. In this case, the data acquisition unit may acquire the magnetoencephalogram measured while the subject is at rest as the time-series data.
A determination method according to one embodiment of the present invention is a determination method using an information processing device, and includes a data acquisition step of acquiring time series data of biological signals emitted from the brain of a subject, and a determination step of inputting the time series data acquired in the data acquisition step into a trained model constructed by machine learning the relationship between the time series data of the biological signals emitted from the brain and brain diseases, and making a determination regarding the brain disease of the subject based on the output of the trained model.
A determination program according to one aspect of the present invention is a determination program for causing a computer to function as the information processing device, and causes the computer to function as the data acquisition unit and the determination unit.

1 Information processing device 101 Data acquisition unit 102 Normalization unit 103 Determination unit 201 Electroencephalogram data (time series data of biological signals)
202 Trained Model

Claims (5)

A data acquisition unit that acquires waveform data of multiple channels of electroencephalograms measured simultaneously while the subject is at rest, or waveform data of multiple channels of magnetoencephalograms measured simultaneously while the subject is at rest;
a determination unit that inputs only the electroencephalogram waveform data or magnetoencephalogram waveform data acquired by the data acquisition unit into a trained model that includes a layer for convolving waveform data of multiple channels at the same time, the trained model being constructed by machine learning using as teacher data the waveform data of multiple channels of electroencephalograms measured simultaneously while a patient with a brain disease is at rest, or the waveform data of multiple channels of magnetoencephalograms measured simultaneously while the patient is at rest; and determines whether the subject has a brain disease based on an output of the trained model ;
The output of the trained model indicates the probability that the subject falls into a predetermined cluster related to a brain disease.
23. An information processing apparatus comprising: The trained model was constructed using machine learning that uses waveform data of electroencephalograms or magnetoencephalograms of patients with different types of dementia as training data.
The information processing device according to claim 1, characterized in that the determination unit determines the type of dementia of the subject based on the output of the trained model. a normalization unit that normalizes the waveform data of the electroencephalogram of the multiple channels or the waveform data of the magnetoencephalogram of the multiple channels in a time direction,
The information processing device according to claim 1, characterized in that the judgment unit inputs the normalized waveform data into the trained model and makes a judgment regarding a brain disease of the subject based on the output of the trained model. A determination method using an information processing device including a processor,
A data acquisition step in which the processor acquires waveform data of multiple channels of electroencephalograms measured simultaneously while the subject is at rest, or waveform data of multiple channels of magnetoencephalograms measured simultaneously while the subject is at rest;
the processor includes a layer for convolving waveform data of multiple channels at the same time, and inputs only the EEG waveform data or the magnetoencephalogram waveform data acquired in the data acquisition step into a trained model constructed by machine learning using as teacher data waveform data of multiple channels of EEGs simultaneously measured while a patient with a brain disease is at rest, or waveform data of multiple channels of magnetoencephalograms simultaneously measured while the patient is at rest; and a determination step for making a determination regarding the brain disease of the subject based on an output of the trained model ;
The output of the trained model indicates the probability that the subject falls into a predetermined cluster related to a brain disease.
A determination method comprising: A determination program for causing a computer to function as the information processing device according to claim 1, the determination program causing the computer to function as the data acquisition unit and the determination unit.

Images