Information processing method, program, information processing device, and brain function determination assistance device

Abstract

Proposed is a novel method capable of presenting information relating to brain functions.　The information processing method of an information processing device includes: calculating, on the basis of a time-frequency analysis, time-frequency information in a set time range from brain wave signals acquired from a plurality of electrodes; and outputting information relating to brain functions based on the time-frequency information.

Description

Information processing method, program, information processing device, and brain function assessment assistance device

[0001] The present invention relates to an information processing method and the like.

[0002] Measuring and evaluating electroencephalograms (EEG) is an essential element in assessing the normality of brain function. A typical EEG test requires 19 channels of electrodes to be attached to the scalp based on the International 10-20 system. Attaching the electrodes requires maintaining low contact impedance between the electrodes and the scalp, requiring skill and experience. Furthermore, EEG tests for epilepsy and other conditions require continuous electrode attachment for extended periods of time to obtain EEG data during seizures, which can place significant physical and psychological strain on the subject. Furthermore, interpreting measured EEG data typically requires several years of specialized training. Therefore, even if EEG data can be measured, the number of EEG interpreters who can utilize the data for diagnosing diseases is limited, and measurement results may not be fully utilized. For example, Non-Patent Document 1 discloses an artificial intelligence model for automated interpretation of clinical EEG data, which is capable of distinguishing between abnormal and normal EEG recordings based on the International 10-20 system and classifying abnormal EEG recordings into categories relevant to clinical decision-making in epilepsy diagnosis. Furthermore, Non-Patent Document 2 discloses a seizure detection device that automatically detects absence seizures based on a one-channel electroencephalogram measured using a wearable headband device.

[0003] Jesper Tveit et al. “Automated Interpretation of Clinical Electroencephalograms Using Artificial Intelligence”, JAMA Neurol. 2023;80(8):805-812. Giorgi Japaridze1 et al. “Automated detection of absence sequences using a wearable electroencephalographic device: a phase 3 validation “study and feasibility of automated behavioral testing”, Epilepsia. 2023;64:S40-S46.

[0004] The artificial intelligence model described in Non-Patent Document 1 may be able to distinguish between abnormal EEG recordings in epilepsy, an example of a disease, and normal EEG recordings. However, the artificial intelligence model described in Non-Patent Document 1 requires EEGs acquired from 19 channels of electrodes. The seizure detection device described in Patent Document 2 may be able to detect absence seizures, which are a pathological condition of epilepsy. However, absence seizures are not the only pathological condition of epilepsy, and the model is not sufficient for use in clinical practice, for example, to determine whether a subject is suffering from epilepsy, an example of a brain disease.

[0005] Although epilepsy has been used as an example of a disease here, there are various other diseases besides epilepsy. Cognitive dysfunction is one of these diseases. Sleep-related disorders are also attracting attention. In this case, being able to grasp and distinguish between sleep states, such as sleep stages, is important for solving sleep-related problems.

[0006] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to propose a new method that makes it possible to present information related to brain function.

[0007] According to a first aspect of the present invention, an information processing method for an information processing device includes calculating time-frequency information for a set time range from EEG signals acquired from a plurality of electrodes based on time-frequency analysis, and outputting information on brain function based on the time-frequency information. According to a second aspect of the present invention, a program for causing a computer to execute the following causes the computer to calculate time-frequency information for a set time range from EEG signals acquired from a plurality of electrodes based on time-frequency analysis, and output information on brain function based on the time-frequency information. According to a third aspect of the present invention, an information processing device includes a processing unit that calculates time-frequency information for a set time range from EEG signals acquired from a plurality of electrodes based on time-frequency analysis, and outputs information on brain function based on the time-frequency information. According to a fourth aspect of the present invention, a brain function assessment assistance device includes an information processing device and a display unit that displays information on brain function.

[0008] According to the present invention, information on brain function can be output based on EEG signals acquired from multiple electrodes. Furthermore, by using a small number of electrodes, information on brain function can be output based on EEG signals acquired from a small number of channels.

[0009] 1 is a block diagram showing an example of the functional configuration of an information processing device. A diagram showing an example of the placement of an electrode group for acquiring electroencephalograms. A flowchart showing an example of the flow of information processing. A diagram showing a specific example of the processing result in density spectrum array conversion processing. A diagram showing a specific example of the processing result in DSA time window extraction processing. A diagram showing an example of an extracted spectrum used for learning of the brain function assessment unit. A table showing an example of benchmark evaluation. A graph showing an example of an ROC curve. A table showing another example of benchmark evaluation. A diagram showing an example of extracted spectra for frontal lobe epilepsy and non-REM parasomnia. A table showing an example of benchmark evaluation. A diagram showing an example of extracted spectra for each sleep stage. A table showing an example of benchmark evaluation. A diagram showing an example of an experimental result of brain function assessment using wavelet analysis. A flowchart showing another example of the flow of information processing. A block diagram showing an example of the functional configuration of a brain function assessment assistance device.

[0010] An example of an embodiment of the present invention will be described below with reference to the drawings. In the description of the drawings, the same elements are designated by the same reference numerals, and duplicate descriptions may be omitted. Furthermore, the components described in this embodiment are merely examples, and are not intended to limit the scope of the present invention.

[0011] [Embodiment] Hereinafter, an example of an embodiment for realizing the information processing technology of the present invention will be described.

[0012] 1 is a block diagram showing an example of the functional configuration of an information processing device 1 according to one aspect of the present embodiment. The information processing device 1 may be regarded as an electroencephalogram (EEG) diagnosis support device (EEG diagnosis support device), a brain disease assessment support device (brain disease assessment support device), or the like. The information processing device 1 includes, for example, an EEG signal acquisition unit 110, a density modulated spectral array (DSA) conversion unit 120, a time window extraction unit 130, and a brain function assessment unit 140. These may be functional units (functional blocks) included in, for example, a processing unit (processing device) or a control unit (control device) of the information processing device 1 (not shown), and are configured with processing circuits such as a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), or a field programmable gate array (FPGA).

[0013] In the following, epilepsy will be taken as an example of a brain disease caused by brain dysfunction. Note that the brain functions that can be determined by this device are not limited to brain functions related to epilepsy. For example, the brain functions that can be determined may be memory functions related to dementia or sleep functions related to sleep-related disorders.

[0014] The electroencephalogram signal acquiring section 110 receives, as input, an electroencephalogram signal acquired by an electroencephalogram acquiring electrode group 5 external to the information processing device 1, for example.

[0015] The EEG acquisition electrode group 5 includes two or more electrodes 5x placed on the subject's head, earlobes, etc. For example, if the number of electrodes in the EEG acquisition electrode group 5 is "N" (where "N" is a natural number equal to or greater than two), the EEG acquisition electrode group 5 includes N electrodes 5A, 5B, 5C, ..., 5N. FIG. 2 shows an example of a method for placing the EEG acquisition electrode group 5. In this figure, electrode placement positions based on the International 10-20 System are indicated by white circles. For example, if the number of electrodes in the EEG acquisition electrode group 5 is "3," electrode 5A may be placed at "Cz," electrode 5B at "A1," and electrode 5C at "A2."

[0016] In this example, an example of acquiring one-channel electroencephalograms using "A1" and "A2" as reference electrodes is illustrated, but the reference electrodes may be placed on the mastoid processes "M1" and "M2", for example. In other words, the number of reference electrodes may be any number as long as it is one or more. The reference electrodes may also be called standard electrodes.

[0017] Furthermore, the number and positions of the head electrodes are not limited to the above example. For example, electrodes may be placed at "C3" and "C4." That is, the number of head electrodes may be any number as long as it is one or more. The head electrodes may also be called search electrodes.

[0018] The reference electrode may be placed on the subject's head (e.g., "Cz"), and the head electrodes may be placed on the earlobes (e.g., "A1" and "A2").

[0019] Hereinafter, the time-series data of potential (EEG signal) acquired by electrode 5A will be referred to as the "channel 1 signal," the EEG signal acquired by electrode 5B as the "channel 2 signal," ..., and the EEG signal acquired by electrode 5N as the "channel N signal." Note that "channel" may be abbreviated as "Ch." Furthermore, the electrode that acquires the "channel x signal" may be referred to as "electrode Ch.x." Furthermore, the potential of the "channel x signal" at time "t" may be referred to as "Ch.x(t)."

[0020] The EEG signal acquisition unit 110 has the function of, for example, receiving an EEG signal as input from the EEG acquisition electrode group 5, buffering the EEG signal for a predetermined period (for example, "180 minutes"), and outputting the EEG signal within the predetermined period to the density spectrum array conversion unit 120.

[0021] The density spectrum array conversion unit 120 has a function of, for example, receiving as input an EEG signal acquired by the EEG signal acquisition unit 110, performing time-frequency analysis on the EEG signal based on the potential difference between the reference electrode and the head-mounted electrode, converting it into a power spectrum (DSA), and outputting the result to the time window extraction unit 130. The DSA may be considered a signal (tensor data) that expresses the amplitude (power) of the EEG signal into multiple channels (e.g., (R, G, B)) through time-frequency analysis. The DSA may also be considered a type of color image data.

[0022] The time window extraction unit 130 has a function of generating an extracted spectrum by extracting the EEG signal from the DSA-processed EEG signal over a time range (time window) of a predetermined width from a predetermined reference point (e.g., the time point at which an abnormality in brain function is detected), and outputting the extracted spectrum to the brain function assessment unit 140. The extracted spectrum may be considered a multi-channel signal (tensor data) that expresses the EEG signal in the predetermined time window in the frequency domain. The extracted spectrum may also be a type of color image data.

[0023] The brain function assessment unit 140 is configured, for example, with a convolutional neural network (CNN) model such as ResNet or a neural network model such as Vision Transformer (ViT). For example, upon receiving a cut-out spectrum as input, the brain function assessment unit 140 has a function of outputting a probability (e.g., "0" to "1") that serves as a guide for a pre-trained brain function assessment (e.g., the presence or absence of a brain disease, a sleep state, etc.) in accordance with a pre-trained neural network model. The brain function assessment unit 140 may also be configured to determine and output a brain function state (e.g., "0" or "1") based on the probability of the brain function assessment based on the cut-out spectrum.

[0024] For example, when the brain function assessment is the presence or absence of a brain disease (e.g., epilepsy), the probability of the brain function assessment may be referred to as the probability of the presence or absence of a brain disease. In this case, the brain functional state may be referred to as the presence or absence of a brain disease. Furthermore, for example, when the brain function assessment is the assessment of a sleep stage, the probability of the brain function assessment may be referred to as the probability of belonging to each sleep stage estimated from the measured DSA. In this case, the brain functional state may be referred to as a sleep stage.

[0025] The output layer of the brain function assessment unit 140 may have multiple output nodes to enable multi-class classification, or may have one output node to enable two-class classification (e.g., presence or absence of disease).

[0026] Here, the "output" of the judgment result may include not only the display of the judgment result on the device itself (display output), but also, for example, the output of the judgment result to another functional unit on the device itself (internal output), or the output (external output) or transmission (external transmission) of the judgment result to a device other than the device itself (external device).

[0027] [Information Processing Procedure] Fig. 3 is a flowchart showing an example of an information processing procedure in this embodiment. The processing in the flowchart in Fig. 3 is realized, for example, by the processing unit of the information processing device 1 reading out the code of a program stored in a storage unit (not shown) into a RAM (Random Access Memory) (not shown) and executing it.

[0028] Each symbol S in the flowchart in Fig. 3 represents a step. Note that the flowchart described below merely shows an example of the information processing procedure in this embodiment, and other steps may be added or some steps may be deleted. Also, some of the steps in the flowchart may be interchanged and executed.

[0029] First, the EEG signal acquiring unit 110 executes an EEG signal acquiring process (S110). In the EEG signal acquiring process, for example, the EEG signal acquiring unit 110 acquires EEG signals from the EEG acquisition electrode group 5 and accumulates and stores the EEG signals in a buffer (not shown).

[0030] Then, for example, when the EEG signal acquiring section 110 acquires EEG signals for a predetermined period (for example, "180 minutes"), for example, the density spectrum array converting section 120 executes density spectrum array converting processing (S120).

[0031] In the density spectrum array conversion process, the density spectrum array conversion unit 120 calculates a unipolar EEG signal for one channel based on, for example, the potential at the reference electrode and the potential at the head-mounted electrode. For example, when the electrode positions are as shown in FIG. 2, the unipolar EEG signal Vd(t) at time "t" may be calculated using the following formula: Vd(t) = abs(Ch.1(t) - (Ch.2(t) + Ch.3(t)) ÷ 2), where "abs(x)" represents the absolute value of "x."

[0032] The unipolar EEG signal may be considered as time series data of the differential voltage between the average potential of the head electrode and the average potential of the reference electrode. The unipolar EEG signal may be considered as an EEG signal based on unipolar induction. The unipolar EEG signal may be considered as a signal for observing activity at the head electrode location relative to the reference electrode location.

[0033] The density spectrum array transform unit 120 then applies a fast Fourier transform to the monopolar EEG signal at predetermined time intervals (e.g., 5.12 seconds) to calculate the power in each frequency domain at each time interval. The unit then calculates a density spectrum array (DSA) expressed as a color spectrum in which the maximum power domain is (R, G, B) = (255, 0, 0) and the minimum power domain is (R, G, B) = (0, 0, 255). The sampling time interval may be set to overlap at a predetermined interval (e.g., 2.56 seconds).

[0034] FIG. 4 shows an example of a unipolar EEG signal calculated from an EEG signal acquired by the EEG signal acquisition unit 110 and a DSA calculated from the unipolar EEG signal. In this figure, the color spectrum of the DSA (primary colors: "red" indicates maximum power, "blue" indicates minimum power, and "yellow" indicates intermediate power) is converted to a grayscale, where "(R,G,B)=(255,0,0)" is black (0), "(R,G,B)=(0,255,255)" is white (0), and "(R,G,B)=(0,0,255)" is gray (128). That is, in each region of the DSA, the closer the color is to black (across gray and white), the greater the amplitude (power) of the unipolar EEG signal in that time period and frequency band. For example, the frequency band for the DSA is set to 0 Hz to 20 Hz.

[0035] 3 , for example, after the density spectrum array conversion process is performed, the time window extracting unit 130 executes a brain function analysis start point detection process (S130). In the brain function analysis start point detection process, the time window extracting unit 130 analyzes a movement sequence of the subject captured by an imaging unit (not shown) in synchronization with the electroencephalogram signal time series during electroencephalogram acquisition, and detects, for example, a seizure onset (clinical onset).

[0036] Body tremors are one of the characteristics of epileptic seizures, but analysis of EEG signals is essential to distinguish them from syncope and psychogenic non-epileptic seizures, which are conditions that also cause body tremors. Therefore, the onset of a seizure may be taken as the start of body tremors, for example.

[0037] The time window extracting unit 130 may detect the seizure onset based on, for example, an input from the subject via an input unit (e.g., a push button, not shown). The time window extracting unit 130 may also detect the seizure onset based on, for example, spikes or sharp waves detected using an EEG signal or DSA. The time window extracting unit 130 may also detect the seizure onset based on, for example, a time series analysis of the subject's weight shift using a pressure-sensitive sensor (not shown).

[0038] The time window extractor 130 may also receive time information indicating the start point of a seizure via an input unit (not shown), for example.

[0039] In addition, when the brain function to be assessed is a cognitive function, for example, the time point at which a cognitive dysfunction (e.g., memory impairment, disorientation, or executive dysfunction) is observed may be analyzed from the subject's conversation sequence acquired by a sound input unit (not shown) and treated as the onset of a seizure. Furthermore, in epilepsy, for example, abnormal waveforms may occur in areas showing hyperexcitability during periods when a seizure is not occurring (so-called interictal periods). Therefore, for example, the time point at which an abnormal waveform is observed in an electroencephalogram during sleep may be analyzed and treated as the onset of a predetermined symptom.

[0040] For example, when the brain function analysis starting point detection process is executed, the time window cutout unit 130 executes a DSA time window cutout process (S140). In the DSA time window cutout process, the time window cutout unit 130 cuts out the DSA in a predetermined time window (e.g., ±1 minute or ±3 minutes) based on the seizure onset point, and generates a cutout spectrum with normalized magnitude.

[0041] FIG. 5 shows examples of generated cut-out spectra for various time windows. The DSA and cut-out spectra in this figure are shown in grayscale, similar to FIG. 4 . For example, when (A) the time window is ±1 minute and the reference point is (A2) the seizure onset (referred to as "middle 1 / 3" in the evaluation experiment), when (A1) the reference point is -2 minutes before the seizure onset (referred to as "early 1 / 3" in the evaluation experiment), and when (A3) the reference point is +2 minutes before the seizure onset (referred to as "late 1 / 3" in the evaluation experiment), the cut-out spectra obtained from the same DSA are significantly different. Furthermore, when (B) the time window is ±3 minutes and the reference point is the seizure onset (referred to as "full" in the evaluation experiment), the cut-out spectrum appears similar to (A) due to the wide time window. It appears that the amount of information is greater than when the time window is ±1 minute. Note that the extracted spectrum may be extracted using a time window and reference point other than those described above.

[0042] Returning to FIG. 3 , for example, when the DSA time window extraction process is performed, the brain function assessment unit 140 executes the brain function assessment process (S150). In the brain function assessment process, for example, the brain function assessment unit 140 receives the generated extracted spectrum as input and executes an inference process according to pre-trained weights. Then, the brain function assessment unit 140 calculates, for example, the probability of brain function assessment (for example, a value between "0" corresponding to the absence of a brain disease and "1" corresponding to the presence of a brain disease). Note that the probability of brain function assessment may also be referred to as the "probability of a disease" or "probability of illness." Furthermore, the probability of brain function assessment may also be referred to as the probability that the brain is in a predetermined state (for example, REM sleep state).

[0043] The brain function assessment unit 140 may output the brain function assessment as a binary value such as "0" or "1" by, for example, performing threshold processing on the calculated probability of brain function assessment. For example, the brain function assessment unit 140 may output "1" if the probability of brain function assessment is equal to or greater than a preset threshold, and may output "0" if not. In other words, the user may be able to change the assessment criteria for brain function assessment by setting a threshold.

[0044] Then, the brain function assessment unit 140 outputs, for example, the calculated probability of the brain function assessment or a brain function assessment result related to the brain function assessment (S160). Note that the brain function assessment result may include the unipolar EEG signal or DSA used in the assessment, information related to the seizure onset and time window, the generated extracted spectrum, etc. Furthermore, when the brain function assessment is a multi-class assessment, the result may include the probability of belonging to each class, or the multi-class classification result corresponding to the output node that gives the maximum value.

[0045] For example, when the brain function assessment result output process is executed, the information processing device 1 determines whether to terminate the process (S170). For example, if it is determined to terminate the process based on a user operation (an example of a user input) on an operation unit (an example of an input unit) not shown (S170: YES), the information processing device 1 terminates the process. If it is determined not to terminate the process (S170: NO), the information processing device 1 returns the process to, for example, the EEG signal acquisition process.

[0046] In addition, for example, if a seizure starting point is not detected during the time when the EEG signal is acquired in the brain function analysis starting point detection process, the information processing device 1 may, for example, return the process to the EEG signal acquisition process.

[0047] Furthermore, for example, when multiple seizure onset points are detected in the brain function analysis onset point detection process, the time window extraction unit 130 may, for example, set reference points based on each seizure onset point and generate multiple extracted spectra.The brain function assessment unit 140 may then output the probability of brain function assessment or brain function assessment for each of the generated extracted spectra based on the multiple extracted spectra.The brain function assessment unit 140 may, for example, output the average value of the brain function assessment probabilities calculated for the multiple extracted spectra as the final brain function assessment probability.The brain function assessment unit 140 may, for example, output the voting result of the brain function assessment calculated for the multiple extracted spectra as the final brain function assessment.

[0048] While Figure 3 shows the inference processing in the brain function assessment unit 140, the learning processing in the brain function assessment unit 140 can be realized, for example, by labeling the extracted spectra generated in accordance with S110 to S140 and using them as training data.

[0049] Figure 6 shows examples of extracted spectra from epileptic and non-epileptic seizures used as training data. The extracted spectra in this figure are shown in grayscale, as in Figure 4. Epileptic and non-epileptic seizures, which are difficult to distinguish at first glance from body movements (tremors), have significantly different characteristics when comparing the extracted spectra, which are the results of time-frequency analysis of EEG signals.

[0050] In the learning process of the brain function assessment unit 140, transfer learning may be performed using a pre-trained network that has been trained in advance using another image database, etc. In addition, in the learning process of the brain function assessment unit 140, learning may be performed using only the extracted spectra that are training data.

[0051] [Evaluation Experiment Results] Figure 7 shows a table of the evaluation results of brain function assessment based on various head-mounted electrodes and time windows. Note that "A1" and "A2" were used as reference electrodes in all experiments. Regarding the head-mounted electrodes, "C3-C4" indicates that measurements were taken using electrodes placed at positions "C3" and "C4" based on the International 10-20 System. The same applies to other electrodes such as "Fp1-Fp2" and "O1-O2." In the evaluation experiments, EEG signal data for temporal lobe epilepsy and psychogenic non-epileptic seizures from patients hospitalized in the epilepsy department of the hospital to which the inventor belongs between 2014 and 2022 and undergoing video EEG monitoring were collected. Training data included 40 patients with temporal lobe epilepsy (91 seizures) and 40 patients with psychogenic non-epileptic seizures (82 seizures) from 2014 to 2020, while test data included 8 patients with temporal lobe epilepsy (15 seizures) and 11 patients with psychogenic non-epileptic seizures (18 seizures) from 2021 to 2022. The brain function assessment unit 140 was configured using ResNet34, a type of CNN. In addition, during the DSA time window extraction process, the extracted spectrum was normalized to 224 x 224 pixels to match the input to ResNet34. In other words, the extracted spectrum can be considered a "224 x 224 x 3" dimensional spatiotemporal feature based on the EEG signal. The threshold for the probability of brain function determination for determining the presence or absence of disease was set to "0.5."

[0052] In the evaluation experiment, the AUC (Area Under the Curve), accuracy, sensitivity, and specificity were calculated based on the results obtained under each condition and used for evaluation. From this table, it can be seen that the highest accuracy of 87.9% was achieved with the head electrode "Cz" and the time window "middle 1 / 3." The accuracy of a non-trained reader is estimated to be around 50%. Therefore, by using the results of this information processing device as diagnostic auxiliary information, it is expected to provide a simple measurement method for differentiation using emergency EEG and outpatient EEG.

[0053] Figure 8 shows the ROC curve for the head electrode "Cz" and time window "middle 1 / 3" that achieved the highest accuracy. This ROC curve shows that the increase in the true positive rate is significantly greater than the increase in the false positive rate. In other words, the ROC curve demonstrates the high usefulness of this method.

[0054] Figure 9 shows a table of the evaluation results of brain function assessment using EEG signal data for generalized epilepsy and psychogenic non-epileptic seizures. The table of evaluation results can be interpreted in the same way as in Figure 7 . From this table, it can be seen that the highest accuracy of "79.9%" was achieved with the head electrode "Cz" and the time window "middle 1 / 3." It should be noted that, although not shown, it was confirmed that similarly good results were obtained for frontal lobe epilepsy and other conditions. In other words, it can be seen that this method works effectively regardless of the type of epilepsy.

[0055] [Actions and Effects of the Embodiment] The information processing device 1 calculates time-frequency information (e.g., a cut-out spectrum) within a set time range from EEG signals acquired from multiple electrodes based on time-frequency analysis, and outputs information about brain function based on the time-frequency information (e.g., a probability of the presence or absence of a brain disease, or a binary value indicating the presence or absence of the disease). The time-frequency information within a set time range is calculated from EEG signals acquired from multiple electrodes based on time-frequency analysis. Then, information about brain function is output based on the calculated time-frequency information. This makes it possible to present information about brain function. Furthermore, by using a small number of electrodes, it becomes possible to present information about brain function based on EEG signals acquired from a small number of channels.

[0056] Furthermore, the information on brain function may include information on the probability of developing a brain disease, in which case it becomes possible to present the probability of developing a brain disease, etc.

[0057] Furthermore, the brain disease may be a disease accompanied by a predetermined symptom (for example, seizures if the brain disease is epilepsy, or cognitive impairment if the brain disease is dementia), and the set time range may be a range that uses the start point of the predetermined symptom as a reference point. Time-frequency information is calculated within the range that uses the start point of the predetermined symptom as a reference point, and based on the calculated time-frequency information within the range that uses the start point of the predetermined symptom as a reference point, it is possible to calculate a highly accurate value related to the presence of the brain disease.

[0058] Furthermore, the EEG signals may be acquired from one or more head electrodes placed on the head (for example, one or more electrodes placed on the Cz or other parts of the head) and one or more reference electrodes (for example, one or more reference electrodes placed on the earlobes, mastoid processes, etc.). This makes it possible to calculate time-frequency information within a set time range based on time-frequency analysis from the EEG signals acquired from the one or more head electrodes placed on the head and one or more reference electrodes.

[0059] Alternatively, the head electrode may be a single electrode located at Cz in the International 10-20 system. This allows time-frequency information within a set time range to be calculated based on time-frequency analysis from EEG signals acquired from a single head electrode located at Cz in the International 10-20 system and one or more reference electrodes. In this case, only one electrode, Cz, is required to be located on the head, allowing time-frequency information to be appropriately calculated with a small number of electrodes.

[0060] The time-frequency analysis may be density spectrum array analysis, which makes it possible to appropriately calculate time-frequency information in a set time range based on the density spectrum array analysis.

[0061] The set time range may be a predetermined range before and after the reference point. By using the time-frequency information in the predetermined range before and after the reference point, highly accurate information regarding brain function can be output.

[0062] Furthermore, the time-frequency information may be normalized based on the frequency band resolution of the density spectrum array analysis, and the information on brain function may be calculated based on the normalized time-frequency information. By using the time-frequency information normalized based on the frequency band resolution of the density spectrum array analysis, it is possible to calculate a highly accurate value related to the incidence of brain disease.

[0063] Furthermore, the output of information regarding brain function may be performed using a neural network model. This makes it possible to easily calculate values ​​related to the incidence of brain diseases using the neural network model. The neural network model may be a ResNet model.

[0064] The brain disease may be epilepsy (various types of epilepsy including temporal lobe epilepsy, frontal lobe epilepsy, generalized epilepsy, etc.). This makes it possible to present information about epilepsy. Furthermore, by using a small number of electrodes, it becomes possible to present information about epilepsy based on EEG signals acquired through a small number of channels.

[0065] [First Modification of the Embodiment] In the above embodiment, an example in which the brain function assessment unit 140 is configured using a neural network model has been described, but this is not limiting. For example, the brain function assessment unit 140 may be configured using a statistical machine learning model such as an SVM (support-vector machine). Furthermore, for example, when the brain function assessment unit 140 receives a cut-out spectrum as input, it may convert the spectrum into a lower-dimensional feature vector using a pre-trained ViT or the like, and then discriminate the converted feature vector using a machine learning model such as an SVM. In other words, the brain function assessment unit 140 may use any mathematical model that can calculate the probability of brain function assessment or binary information on brain function assessment when it receives a cut-out spectrum as input.

[0066] [Second Modification of the Embodiment] In the above embodiment, the determination of an epileptic seizure and a psychogenic non-epileptic seizure is illustrated as an example of the determination of a brain function abnormality, but the present invention is not limited to this. For example, the brain function determination unit 140 may determine a frontal lobe epilepsy (FLE) and a non-REM parasomnia (NREMP).

[0067] Frontal lobe epilepsy and non-REM parasomnia are known to cause seizures or locomotion during sleep. Differential diagnosis of these two conditions is important, but the number of cases of non-REM parasomnia is small and the frequency of symptoms is low. Therefore, it has been difficult to differentiate between frontal lobe epilepsy and non-REM parasomnia during hospitalization for tests, for example.

[0068] FIG. 10 shows an example of excised spectra used as training data during a frontal lobe epileptic seizure and during the occurrence of abnormal behavior due to non-REM parasomnia. The excised spectra in this figure are shown in grayscale, as in FIG. 4 . In the brain function analysis start point detection process, the time point at which abnormal behavior occurred as the seizure start point for non-REM parasomnia may be detected, for example, based on video monitoring images acquired by an imaging unit (not shown). In addition, in the DSA time window excision process, the DSA was excised with a time window of ±1 minute based on the seizure start point. These excised spectra reveal that stronger signals are observed across a wider bandwidth in non-REM parasomnia than in frontal lobe epilepsy.

[0069] Figure 11 shows a table of the evaluation results of brain function assessment based on various head electrodes and time windows. In the evaluation experiment, extracted spectra were generated from 26 seizures in 10 FLE cases and 26 events in 5 NREMP cases. Of these, 8 FLE cases and 4 NREMP cases were used as training data for learning, and 2 FLE cases and 1 NREMP case were used as test data for evaluation using 5-fold cross-validation. The results show that the AUC exceeded 70% for head electrodes "Cz" and "C3-C4," suggesting the potential for this method to differentiate frontal lobe epilepsy from non-REM parasomnia.

[0070] In this embodiment, the brain disease (e.g., frontal lobe epilepsy or non-REM parasomnia) is a disease accompanied by a predetermined symptom (e.g., seizures or abnormal behavior), and the set time range may be a range having a reference point at the onset of the predetermined symptom. As a result, time-frequency information is calculated within the range having a reference point at the onset of the predetermined symptom, and based on the calculated time-frequency information within the range having a reference point at the onset of the predetermined symptom, it becomes possible to calculate a highly accurate value relating to the incidence of the brain disease.

[0071] Also shown is an example of a configuration in which the brain disease is epilepsy or non-REM parasomnia, which makes it possible to easily distinguish between epilepsy and non-REM parasomnia.

[0072] In the above embodiment, the brain function assessment is based on the assessment of brain function abnormalities caused by epilepsy, but the assessment target may be, for example, the sleep stage of the subject.

[0073] Sleep stages are divided into five stages: awake (AWAKE), rapid eye movement (REM) sleep (REM), non-REM sleep stage 1 (NREM1), non-REM sleep stage 2 (NREM2), and non-REM sleep stage 3 (NREM3). Sleep stages are generally distinguished by electroencephalography (EEG) or polysomnography. However, typical EEG and polysomnography require large-scale equipment and specialized technicians, and there is a need for a method that allows subjects to easily measure and confirm their sleep stages.

[0074] In this embodiment, the brain function determination unit 140 may be configured, for example, with ResNet34, which is a neural network model, and may be configured to learn five nodes associated with each sleep stage as output nodes.

[0075] FIG. 12 shows an example of a cut-out spectrum used as training data for each sleep stage. The cut-out spectrum in this figure is shown in grayscale, similar to FIG. 4 . In the brain function analysis start point detection process, the predetermined time point may be detected based on, for example, a video monitoring image acquired by an imaging unit (not shown). The predetermined time point may be detected based on, for example, a change in body position (e.g., turning over in sleep or remaining immobile for a certain period of time). The predetermined time point may be, for example, a sampling point determined at a predetermined interval from the start of EEG acquisition. The predetermined time point may also be any time point during sleep determined based on, for example, a user input to an input unit (not shown).

[0076] In the DSA time window extraction process, the DSA was extracted with a time window of "±1 minute" based on the seizure onset point.

[0077] Figure 13 shows a table of the evaluation results of brain function assessment based on various head-mounted electrodes and time windows. The evaluation experiment targeted Awake 78 records, NREM1 63 records, NREM2 78 records, NREM3 69 records, and REM 78 records extracted from 26 patients who underwent polysomnography in 2022. Of these, Awake 60 records, NREM1 48 records, NREM2 63 records, NREM3 54 records, and REM 63 records were used as training data, and Awake 18 records, NREM1 15 records, NREM2 15 records, NREM3 15 records, and REM 15 records were used as test data for evaluation using 5-fold cross-validation. These results demonstrate that the average AUC for sleep stage assessment was 85% or higher for all head-mounted electrodes, demonstrating excellent classification ability.

[0078] In this embodiment, the information about brain function may include information about sleep state, which makes it possible to present information about sleep state. Furthermore, by using a small number of electrodes, it becomes possible to present information about sleep state based on EEG signals acquired through a small number of channels.

[0079] Alternatively, the information on the sleep state may be a sleep stage, and the set time range may be a range with a predetermined time point as a reference point. Time-frequency information is calculated for a range with a start point of the predetermined state as a reference point, and based on the calculated time-frequency information for the range with the predetermined time point as a reference point, highly accurate information on the sleep state can be output.

[0080] [Fourth Modification of the Embodiment] In addition to the aforementioned distinction (differentiation) between epilepsy and psychogenic non-epileptic seizures, for example, it may be possible to distinguish between epilepsy and syncope. In this case, for example, cut-out spectra from epileptic seizures (e.g., epileptic loss of consciousness, convulsions) and syncope (e.g., syncope loss of consciousness, convulsions) may be used as training data. Although not shown in the figure, a comparison of these cut-out spectra reveals different characteristics, which can be used to distinguish between epilepsy and syncope.

[0081] [Fifth Modification of the Embodiment] In the above embodiment, density spectrum array analysis is applied as the time-frequency analysis, but this is not limiting. For example, wavelet analysis may be applied as the time-frequency analysis. This method is also applicable to various brain function assessments.

[0082] Wavelet analysis is a technique for analyzing the frequency characteristics of waves by compressing and translating a basis function called a mother wavelet. Unlike Fourier transform, it does not separate the analysis interval by a window size, making it possible to extract frequency characteristics without losing time information. For example, a wavelet transform can be performed on measured electroencephalograms (EEG signals) using a mother wavelet called a Morlet, and an image can be generated by plotting the transition of the frequency characteristics of the waves over the measurement time on a time-scale plane.

[0083] FIG. 14 shows an example of experimental results in which brain function was assessed using wavelet analysis. As an example, the following shows the results of an experiment to classify generalized epileptic seizures (including absence seizures (A), tonic-clonic seizures (G), and myoclonic seizures (M)) from psychogenic non-epileptic seizures (PNES). In this experiment, the average of Cz and A1-A2, and data from "-3 to +3 minutes" (full) before clinical symptoms appeared were used. As a result, the AUC, Accuracy, Recall, and Precision were as follows, confirming favorable results.・AUC: (A) 91.5%, (G) 99.1%, (M) 90.7%, (PNES) 87.5% ・Accuracy: (A) 100%, (G) 100%, (M) 84.4%, (PNES) 70.5%・Recall: (A) 30%, (G) 100%, (M) 12.5%, (PNES) 95% ・Precision: (A) 90.2%, (G) 91.5%, (M) 88.8%, (PNES) 83.3%

[0084] Similarly, wavelet analysis may be applied as time-frequency analysis to various brain function assessments, including the above-mentioned differentiation between epilepsy and non-REM parasomnia, and sleep stage assessment.

[0085] FIG. 15 is a flowchart showing an example of the information processing procedure in this modified example. The diagram is viewed in the same way as in FIG. 3, and steps similar to those in FIG. 3 are assigned the same reference numerals, and a repeated explanation will be omitted. After S110, the processing unit of the information processing device 1 performs wavelet analysis processing (S220). Specifically, for example, a wavelet transform is performed on the monopolar EEG signal described above, and the transition of frequency characteristics within the measurement time is plotted on a time-scale plane to generate an image. The processing unit of the information processing device 1 then proceeds to S130.

[0086] For example, density spectrum array analysis and wavelet analysis may be applied as time-frequency analyses to perform the learning process of the brain function assessment unit 140. Specifically, for example, the learning process of the brain function assessment unit 140 may be performed using a cut-out spectrum generated by applying density spectrum array analysis (first trained model), and the learning process of the brain function assessment unit 140 may be performed using a cut-out spectrum generated by applying wavelet analysis (second trained model). Furthermore, during inference, for example, the cut-out spectrum generated by applying density spectrum array analysis and wavelet analysis may be input to the corresponding trained models, so that the output of each model can be referenced.

[0087] [Example] Next, an example of a brain function assessment assistance device 10 (brain function assessment assistance device) that applies or includes the above-described information processing device 1 will be described. This device may be configured as a terminal, electronic device (electronic device), or the like. Here, an example of a sleep stage assessment assistance device will be described as an example of brain function assessment. However, examples to which the present invention can be applied are not limited to this, and the present invention may also be applied to an assessment assistance device for brain diseases such as epilepsy and dementia.

[0088] 16 is a diagram showing an example of the functional configuration of the brain function assessment assistance device 10. The brain function assessment assistance device 10 includes, for example, a processing unit 100, a storage unit 200, an electroencephalogram acquisition electrode group 5, an operation unit 310, a display unit 320, a communication unit 330, and an imaging unit 340.

[0089] The processing unit 100 is a processing device that comprehensively controls each part of the brain function assessment assistance device 10 in accordance with various programs such as system programs stored in the memory unit 200 and performs various processes related to brain function assessment, and is configured with processing circuits such as a CPU, GPU, DSP, ASIC, FPGA, etc.

[0090] The processing unit 100 has, as its main functional units, for example, an electroencephalogram signal acquisition unit 110, a density spectrum array conversion unit 120, a time window extraction unit 130, a brain function assessment unit 140, and a display control unit 150. For example, the display control unit 150 has a function of, upon receiving the output (assessment result) of the brain function assessment unit 140, visualizing the assessment result and displaying it on the display unit 320. The other functional units correspond, for example, to the functional units included in the information processing device 1 in FIG. 1 .

[0091] The storage unit 200 is a storage device configured to include a memory circuit such as a ROM (Read Only Memory), a RAM, or a flash memory, a hard disk device, a magneto-optical disk device, or the like.

[0092] The storage unit 200 stores, for example, a brain function assessment assistance program 210 and an electroencephalogram signal temporary storage unit 220 .

[0093] The brain function assessment assistance program 210 is, for example, a program that is read by the processing unit 100 and executed as brain function assessment assistance application processing.

[0094] The EEG signal temporary storage unit 220 is, for example, a buffer that stores the EEG signals acquired by the EEG signal acquisition unit 110. Note that the EEG signal temporary storage unit 220 may store imaging information (video information) acquired by the imaging unit 340 while synchronizing the time with the EEG signals.

[0095] The operation unit 310 is configured to have an input device, such as an operation button or an operation switch, with which the user performs various operation inputs to the brain function assessment assistance device 10. The operation unit 310 may have, for example, a touch panel (not shown) configured integrally with the display unit 320, and this touch panel may function as an input interface between the user and the brain function assessment assistance device 10. For example, the operation unit 310 may output an operation signal in accordance with a user operation to the processing unit 100. Furthermore, an input device that accepts sound (including voice) input as user input may be configured.

[0096] The display unit 320 is a display device configured to include, for example, an LCD (Liquid Crystal Display) or an OELD (Organic Electro-luminescence Display), and performs various displays based on display signals output from the display control unit 150.

[0097] The display unit 320 may display information based on values ​​related to the subject's brain function under the control of the display control unit 150. For example, various information may be displayed, such as the probability of classification into each sleep stage calculated by the brain function assessment unit 140, information (numerical information) on the values ​​of each sleep stage, and information (image information) on a time transition image (including graphs, etc.) of the sleep stages based on these values. Furthermore, the display unit 320 may display time-frequency information within a set time range under the control of the display control unit 150. For example, information such as a cut-out spectrum (e.g., the one shown in FIG. 5 ) generated by the time window cut-out unit 130 may be displayed.

[0098] The brain function assessment assistance device 10 may be configured to transmit various pieces of information to a display device (display device) provided separately from the brain function assessment assistance device 10, and the display device may display the information received from the brain function assessment assistance device 10. Alternatively, the information processing device 1 may be configured to transmit various pieces of information to a display device, and the display device may display the information received from the information processing device 1.

[0099] The communication unit 330 is a communication device for transmitting and receiving information used within the device with an external information processing device. The communication method of the communication unit 330 can be various methods, such as a wired connection via a cable conforming to a predetermined communication standard such as Ethernet or USB (Universal Serial Bus), a wireless connection using a wireless communication technology conforming to a predetermined communication standard such as Wi-Fi (registered trademark) or 5G (fifth generation mobile communication system), or a connection using short-range wireless communication such as Bluetooth (registered trademark). The communication unit 330 may transmit various information to an external device or receive various information from an external device under the control of the processing unit 100.

[0100] The imaging unit 340 is an imaging device configured with an imaging element (semiconductor element) such as a CCD (Charge Coupled Device) image sensor or a CMOS (Complementary MOS) image sensor. The imaging unit 340, for example, forms an image of light emitted from an object to be imaged on the light-receiving plane of the imaging element using a lens (not shown), and converts the brightness of the image into an electrical signal through photoelectric conversion. The converted electrical signal is converted into a digital signal by, for example, an A / D (Analog-Digital) converter (not shown), and output to the processing unit 100. Note that the imaging unit 340 is not necessary.

[0101] The processing unit 100 of the brain function assessment assistance device 10 performs brain function assessment processing, for example, in accordance with a brain function assessment assistance program 210 stored in the storage unit 200. The brain function assessment processing may be executed, for example, in accordance with the flowchart of FIG.

[0102] [Actions and Effects of the Example] The brain function assessment assistance device 10 of this example can achieve the same actions and effects as the above-described embodiment. Furthermore, the brain function assessment assistance device 10 may include a display unit 320 that displays information about brain function, thereby enabling the user to check the information about brain function. Furthermore, the display unit 320 may display time-frequency information within a set time range, thereby enabling the user to check the time-frequency information within the set time range.

[0103] REFERENCE SIGNS LIST 1 Information processing device 5 Electrode group for acquiring electroencephalogram 10 Brain function assessment assistance device 110 Electroencephalogram signal acquisition unit 120 Density spectrum array conversion unit 130 Time window extraction unit 140 Brain function assessment unit

Claims

1. An information processing method for an information processing device, comprising: calculating time-frequency information within a set time range based on time-frequency analysis from EEG signals acquired from a plurality of electrodes; and outputting information relating to brain function based on the time-frequency information.

2. The information processing method according to claim 1, wherein the information includes information relating to a sleep state.

3. The information processing method according to claim 2, wherein the information relating to the sleep state is a sleep stage, and the set time range is a range with a predetermined time point as a reference point.

4. The information processing method according to claim 1, wherein the information includes information relating to the occurrence of a brain disease.

5. The information processing method according to claim 4, wherein the brain disease is a disease accompanied by a predetermined symptom, and the set time range is a range having the onset of the predetermined symptom as a reference point.

6. The information processing method according to claim 4, wherein the brain disease is epilepsy or non-REM parasomnia.

7. An information processing method according to any one of claims 1 to 6, wherein the electroencephalogram signals are acquired from one or more head electrodes and one or more reference electrodes placed on the head.

8. The information processing method according to claim 7, wherein the head electrode is one, and is an electrode placed at Cz in the International 10-20 system.

9. The information processing method according to claim 3 or claim 5, wherein the time-frequency analysis is density spectrum array analysis.

10. The information processing method according to claim 9, wherein the set time range is a predetermined range before and after the reference point.

11. The information processing method according to claim 9, wherein the time-frequency information is normalized based on the frequency band resolution of the density spectrum array analysis, and the information is output based on the normalized time-frequency information.

12. The information processing method according to any one of claims 1 to 6, wherein the information is output using a neural network model.

13. The information processing method according to claim 12, wherein the neural network model is a ResNet model.

14. A program for causing a computer to perform the following steps: calculate time-frequency information within a set time range based on time-frequency analysis from EEG signals acquired from multiple electrodes; and output information related to brain function based on the time-frequency information.

15. An information processing device having a processing unit that calculates time-frequency information within a set time range based on time-frequency analysis from EEG signals acquired from multiple electrodes, and outputs information related to brain function based on the time-frequency information.

16. A brain function assessment assistance device comprising: an information processing device according to claim 15; and a display unit that displays the information.

17. The brain function assessment assistance device according to claim 16, wherein the display unit displays the time-frequency information.

Images