Waveform generation identification method and program

The method addresses the challenge of accurately determining IED time and sensors by using data comparison and threshold processing with machine learning to create an IED probability map, improving dipole estimation for epilepsy diagnosis.

JP7585655B2Active Publication Date: 2024-11-19RICOH CO LTD
View PDF 13 Cites 0 Cited by

Patent Information

Application Number
JP2020138813
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-10-30
Filing Date
2020-08-19
Publication Date
2024-11-19
Estimated Expiration
2040-08-19

AI Technical Summary

Technical Problem

Existing methods struggle to accurately determine the time of characteristic waveform information (IED) and narrow down the sensors for the equivalent current dipole method in magnetoencephalography due to the vast amount of data, making manual extraction difficult.

Method used

A method involving waveform data comparison, threshold-based determination, and filtering to accurately identify the time and sensors of characteristic waveforms using a combination of machine learning and threshold processing to create an IED probability map.

Benefits of technology

This approach enables precise determination of IED time and sensor selection, enhancing the accuracy of dipole estimation for epilepsy diagnosis by automating the search for IEDs and their origins.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007585655000001
    Figure 0007585655000001
  • Figure 0007585655000002
    Figure 0007585655000002
  • Figure 0007585655000003
    Figure 0007585655000003
Patent Text Reader

Abstract

To more accurately determine a time when characteristic waveform information (IED) appears and extract a sensor.SOLUTION: A waveform generation specification method includes a comparison step for comparing individual waveform data acquired by a plurality of sensors with at least one piece or more of characteristic waveform information, a determination step for determining appearance accuracy of the characteristic waveform information in at least fixed section of the waveform data on the basis of a correlation rate between a peak section of the waveform data and the characteristic waveform information, and an identification step for identifying a time when a section coinciding with the characteristic waveform information and a corresponding sensor on the basis of the appearance accuracy.SELECTED DRAWING: Figure 5
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a waveform generation specifying method and a program. [Background technology]

[0002] Traditionally, it has been very important to analyze characteristic waveforms in magnetoencephalography and electroencephalography. However, with technological advances, the sampling frequency and number of sensors during measurement are increasing, and as a result, the time required to visually search for waveforms is also increasing.

[0003] For example, in clinical epilepsy diagnosis using magnetoencephalography, a technique called the equivalent current dipole method is used to evaluate the localization of epileptic lesions. In the equivalent current dipole method, the current source (dipole) that generates the magnetic field measured on the scalp is estimated. In order to estimate the dipole, it is necessary to narrow down the time (origin) when characteristic waveform information (IED: Interictal Epileptiform Discharge) occurs from the time series of multiple sensors and the sensors in which that waveform information appears.

[0004] Currently, doctors manually search for IEDs and determine their origins, but because the amount of magnetoencephalography data is enormous, it is difficult to manually and accurately extract the sensor and IED time for each IED.

[0005] Therefore, Patent Document 1 discloses a technology for detecting characteristic waveforms, including IEDs, by performing wavelet transformation on time-series data of brain electromagnetic waves from a single sensor or multiple sensors in order to detect characteristic waves with high accuracy. Summary of the Invention [Problem to be solved by the invention]

[0006] However, the technique disclosed in Patent Document 1 has a problem in that it is not possible to appropriately determine the time of an IED or narrow down the number of sensors to perform the equivalent current dipole method.

[0007] The present invention has been made in view of the above, and has an object to more accurately determine the time when characteristic waveform information (IED) appears and extract the sensor. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems and achieve the object, the waveform occurrence identification method of the present invention includes a comparison step of comparing each waveform data acquired by a plurality of sensors with at least one or more characteristic waveform information, a determination step of determining an occurrence probability of characteristic waveform information in at least a certain section of the waveform data based on a correlation ratio between the section of the waveform data and the characteristic waveform information, a narrowing down step of extracting a sensor area using a threshold value for the occurrence probability of the determined characteristic waveform information and applying a filter to the area to narrow down the time of the characteristic waveform information and the sensor area narrowed down in the narrowing down step, and region Using the above, the time when the section matching the characteristic waveform information appeared and Sensor Index and a step of identifying the target object. Effect of the Invention

[0009] According to the present invention, it is possible to more accurately determine the time when characteristic waveform information (IED) appears and extract the sensor. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a schematic diagram showing the configuration of a biosignal measuring system according to an embodiment. [Diagram 2] FIG. 2 is a block diagram illustrating a schematic functional configuration of the server. [Diagram 3] FIG. 3 is a diagram illustrating an example of a hardware configuration of an information processing device. [Figure 4] FIG. 4 is a diagram illustrating an example of a functional block configuration of the information processing device. [Diagram 5]FIG. 5 is a flowchart showing the flow of the dipole estimation process. [Figure 6] FIG. 6 is a flowchart showing the flow of the learning process when generating an IED probability map using machine learning. [Figure 7] FIG. 7 is a diagram illustrating an example of an IED probability map. [Figure 8] FIG. 8 is a diagram illustrating an example of an IED probability map after threshold processing. [Figure 9] FIG. 9 illustrates an example of an IED probability map after post-processing. [Figure 10] FIG. 10 is a diagram showing an example of a UI using an IED probability map. [Figure 11] FIG. 11 is a diagram showing an example of calculating an IED probability map using a method other than machine learning. [Figure 12] FIG. 12 is a conceptual diagram showing an example of a method for expanding the number of sensors. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] The following describes in detail the embodiments of the waveform generation specification method and program with reference to the accompanying drawings. The present invention is not limited to the following embodiments, and the components in the following embodiments include those that a person skilled in the art can easily come up with, those that are substantially the same, and those that are in the so-called equivalent range. Furthermore, various omissions, substitutions, modifications, and combinations of the components can be made without departing from the spirit of the following embodiments.

[0012] This embodiment has the following features when calculating the probability map of an IED (Interictal Epileptiform Discharge). In short, the feature is that by using the probability map of an IED, it becomes possible to detect the onset time of an IED and select a sensor, which are necessary for the equivalent current dipole method.

[0013] When a doctor performs an analysis, it is ideal to analyze the origin of the IED using the equivalent current dipole method, but since the origin is known to have a low S / N ratio, the time of the IED is actually determined between the origin and the peak of the IED. In this embodiment, the analysis is not limited to the origin.

[0014] (Outline of biosignal measurement system) 1 is a schematic diagram showing the configuration of a biosignal measurement system according to an embodiment of the present invention, and a biosignal measurement system 1 according to the present embodiment will be outlined with reference to FIG.

[0015] The biosignal measurement system 1 is a system that measures and displays multiple types of biosignals of a subject from a specific source (biological part) (e.g., magneto-encephalography (MEG) signals and electro-encephalography (EEG) signals). Note that in the present invention, the biosignals to be measured are not limited to magneto-encephalography (MEG) signals and electro-encephalography (EEG) signals, but may be, for example, electrical signals generated in response to cardiac activity (electrical signals that can be expressed as an electrocardiogram).

[0016] As shown in Fig. 1, the biosignal measurement system 1 includes a measurement device 3 that measures one or more biosignals of a subject, a server 40 that records one or more types of biosignals measured by the measurement device 3, and an information processing device 50 that is a biosignal display device that analyzes one or more types of biosignals recorded in the server 40. The measurement device 3 is, for example, a magnetoencephalograph that collects measurement values ​​such as brain magnetic fields and timing of application of stimuli. Note that, although the server 40 and the information processing device 50 are depicted separately in Fig. 1, for example, at least a part of the functions of the server 40 may be incorporated into the information processing device 50.

[0017] In the example of FIG. 1, the subject (subject) lies on his / her back on the measurement table 4 with electrodes (or sensors) for measuring electroencephalograms attached to his / her head, and places his / her head in the recess 32 of the Dewar 31 of the measurement device 3. The Dewar 31 is a holding container in an extremely low temperature environment using liquid helium, and a number of magnetic sensors for measuring magnetoencephalograms are arranged inside the recess 32 of the Dewar 31. The measurement device 3 collects electroencephalogram signals from the electrodes and magnetoencephalogram signals from the magnetic sensors, and outputs data including the collected electroencephalogram signals and magnetoencephalogram signals (hereinafter, sometimes referred to as "measurement data") to the server 40. The measurement data output to the server 40 is read out and displayed by the information processing device 50, and is analyzed. In general, the Dewar 31 and the measurement table 4 incorporating the magnetic sensors are arranged in a magnetically shielded room, but for convenience, the magnetically shielded room is not shown in FIG. 1.

[0018] The information processing device 50 is a device that displays waveform data of MEG signals from multiple magnetic sensors and waveform data of EEG signals from multiple electrodes in synchronization on the same time axis. An EEG signal is a signal that represents the electrical activity of nerve cells (the flow of ionic charges that occurs in the dendrites of neurons during synaptic transmission) as a voltage value between electrodes. A MEG signal is a signal that represents minute electric field fluctuations caused by electrical activity in the brain. The brain magnetic field is detected by a highly sensitive superconducting quantum interference device (SQUID) sensor. These EEG signals and MEG signals are examples of "biological signals".

[0019] 2 is a block diagram showing a schematic functional configuration of the server 40. As shown in FIG. 2, the server 40 includes a data acquisition unit 41 and a data storage unit .

[0020] The data acquiring unit 41 periodically acquires measurement data from the measuring device 3. Here, the measurement data is individual waveform data measured by a plurality of magnetic sensors in the Dewar 31 of the measuring device 3.

[0021] The data storage unit 42 stores the measurement data acquired from the measurement device 3 .

[0022] (Hardware configuration of information processing device) 3 is a diagram showing an example of a hardware configuration of the information processing device 50. The hardware configuration of the information processing device 50 will be described with reference to FIG.

[0023] As shown in FIG. 3, the information processing device 50 has a CPU (Central Processing Unit) 101, a RAM (Random Access Memory) 102, a ROM (Read Only Memory) 103, an auxiliary storage device 104, a network I / F 105, an input device 106, and a display device 107, which are interconnected by a bus 108.

[0024] The CPU 101 is a computing device that controls the overall operation of the information processing device 50 and performs various information processing. The CPU 101 executes an information display program stored in the ROM 103 or the auxiliary storage device 104, and controls the display operation of the measurement collection screen and the analysis screen.

[0025] The RAM 102 is a volatile storage device used as a work area for the CPU 101 and stores main control parameters and information. The ROM 103 is a non-volatile storage device that stores basic input / output programs and the like. For example, the above-mentioned information display program may be stored in the ROM 103.

[0026] The auxiliary storage device 104 is a storage device such as a hard disk drive (HDD) or a solid state drive (SSD). The auxiliary storage device 104 stores, for example, a control program that controls the operation of the information processing device 50, and various data and files necessary for the operation of the information processing device 50.

[0027] The network I / F 105 is a communication interface for communicating with devices on a network, such as the server 40. The network I / F 105 is realized by, for example, a network interface card (NIC) conforming to the Transmission Control Protocol (TCP) / Internet Protocol (IP).

[0028] The input device 106 is a user interface such as a touch panel input function, a keyboard, a mouse, and operation buttons. The display device 107 is a display device that displays various information. The display device 107 is realized by, for example, a touch panel display function, a liquid crystal display (LCD: Liquid Crystal Display) or an organic EL (Electro-Luminescence) display. The display device 107 displays a measurement collection screen and an analysis screen, and the screens are updated in response to input / output operations via the input device 106.

[0029] The hardware configuration of the information processing device 50 shown in Fig. 3 is an example, and other devices may be included. The information processing device 50 shown in Fig. 3 has a hardware configuration assuming, for example, a PC (Personal Computer), but is not limited thereto, and may be a mobile terminal such as a tablet. In this case, the network I / F 105 may be a communication interface having a wireless communication function.

[0030] (Functional block configuration of information processing device) 4 is a diagram showing an example of a functional block configuration of the information processing device 50. The functional block configuration of the information processing device 50 will be described with reference to FIG.

[0031] As shown in FIG. 4, the information processing device 50 includes a pre-processing unit 501, an IED probability map calculation unit 502, a threshold processing unit 503, a post-processing unit 504, and a dipole estimation unit 505.

[0032] The preprocessing unit 501 performs preprocessing such as extraction and expansion of sensors, downsampling and application of frequency filters, removal of artifacts, poor channel processing, extraction of time windows, and standardization of magnetic field data.

[0033] In addition to using all sensors to calculate the IED probability map described below, sensor extraction and expansion can also be performed using only a certain number of sensors that have been grouped in advance. Sensor grouping can be based on anatomical criteria such as the temporal lobe or frontal lobe, or simply by grouping an arbitrary number of nearby sensors. In addition, if the number of sensors used in the calculation is less than the number of sensors used in the learning process described below, it is also possible to expand the number of fictitious sensors.

[0034] Downsampling is applied to match the sampling frequency used during training. The same frequency filter is also applied, and the same filtering applied during training is applied. Commonly used filters include a 35Hz low-pass filter and a 3Hz-35Hz band-pass filter.

[0035] For the purpose of removing artifacts due to heartbeats and artifacts due to blinking and body movements, ICA (see Non-Patent Document 1) and DSSP (see Non-Patent Document 2) are used.

[0036] Bad channel processing refers to the process of excluding sensors that observe magnetic field changes exceeding a preset threshold, or of interpolating using surrounding sensor values.

[0037] There are several ways to extract the time window, including no overlap where the time window is shifted by the length of the time window, overlapping half the length of the time window, overlapping 1 / 4 the length, etc. When overlapping, the overlapping amount is averaged when calculating the IED probability map described below.

[0038] Standardization of the magnetic field data is performed so that the average is 0 and the variance is 1 within the extracted time window. In addition to standardization, a method of normalizing the magnetic field within a preset range to be between -1 and 1 may be used.

[0039] The IED probability map calculation unit 502 calculates a probability map of a characteristic waveform (IED: Interictal Epileptiform Discharge).

[0040] IED refers to waveforms such as spike wave, spike and wave, poly spike and wave, and sharp wave that are characteristic of epilepsy lesions. In this embodiment, the IED probability map calculation unit 502 calculates the probability map using a machine learning model that has been trained in advance. A method for creating the machine learning model will be described later.

[0041] The threshold processing unit 503 uses a threshold to narrow down the time and sensor area with a high IED probability from the IED probability map obtained in the IED probability map calculation unit 502.

[0042] The post-processing unit 504 performs post-processing to extract sampling time points and sensors to be used for dipole estimation from the IED probability map thresholded by the threshold processing unit 503. In addition, the post-processing unit 504 performs post-processing to separate the maps when the peak detection method or the map has multiple peaks (such as when there are two lesions).

[0043] The dipole estimation unit 505 performs dipole estimation (see Non-Patent Document 3) using the sensor obtained in the processing before the dipole estimation unit 505 and the sampling time point of the IED's origin.

[0044] Next, the flow of the dipole estimation process will be described.

[0045] 5 is a flowchart showing the flow of the dipole estimation process. Prior to the process, the measurement device 3 performs magnetoencephalography measurement and outputs each waveform data measured by the multiple magnetic sensors of the Dewar 31 to the server 40. The server 40 stores each waveform data measured by the multiple magnetic sensors of the Dewar 31 of the measurement device 3 in the data storage unit 42.

[0046] Then, as shown in FIG. 5, first, the preprocessing unit 501 acquires each piece of waveform data measured by the multiple magnetic sensors of the Dewar 31 of the measuring device 3 from the data storage unit 42 of the server 40 (step S1).

[0047] Next, the preprocessing unit 501 performs preprocessing on the acquired data, such as extracting and expanding sensors, downsampling and applying frequency filters, removing artifacts, processing bad channels, extracting time windows, and standardizing magnetic field data (step S2).

[0048] Next, the IED probability map calculation unit 502 calculates a probability map of the characteristic waveform (IED) (step S3).

[0049] The IED probability map may be calculated using a conventional spike position detection algorithm (see Non-Patent Document 4) in addition to a method of applying a model calculated using machine learning. When machine learning is used, the time of IED manually detected by a doctor is learned as the correct answer, so compared to conventional spike position detection methods, it has the effect of obtaining results similar to those obtained when a doctor performs analysis. Conventional methods are developed with the aim of detecting the origin or peak of an IED, but in actual analysis, doctors perform analysis while adjusting the time at which dipoles are estimated from the waveform, findings, etc., between the origin and the peak.

[0050] Here, a case where the IED probability map calculation unit 502 uses machine learning to generate the IED probability map will be described.

[0051] Fig. 6 is a flowchart showing the flow of the learning process when generating an IED probability map using machine learning. As shown in Fig. 6, when the IED probability map calculation unit 502 acquires preprocessed magnetoencephalography data preprocessed by the preprocessing unit 501 (step S11), it performs learning for outputting an IED probability map for the preprocessed magnetoencephalography data (step S12). Note that, as the correct answer data used in learning, the sampling time recorded when dipole estimation was performed and the information of the sensor used at that time are used.

[0052] Specifically, this is realized by applying a network that is often used in a task called Semantic Segmentation, such as U-Net described in Non-Patent Document 5. Semantic segmentation is a task that labels each element of all input sequences, and is used in medical images to automatically estimate brain tumors and cancer regions. Here, the mask data created from the sensor selected when the doctor performed dipole estimation and the time of estimation is used as the correct answer, and learning is performed so that the input magnetoencephalography data becomes mask data. At this time, in order to improve generalization performance, a data augmentation method such as randomly rearranging the order of sensors or setting a certain range of values ​​of a specific sensor to 0 may be applied.

[0053] This can also be achieved with networks used in tasks other than Semantic Segmentation, such as Object Detection and Instance Segmentation. Object Detection is a method for estimating where a target object is located in an input sequence using rectangles, etc. When using this method, it is not possible to calculate an IED probability map, but it is possible to directly obtain the time (e.g., the midpoint of the rectangle) and sensor for dipole estimation from the detected rectangle. Instance Segmentation is a task that combines Semantic Segmentation and Object Detection, and is capable of calculating the classification and number of detected objects as well as the area of ​​the object. Instance Segmentation can calculate an IED probability map, so it can be applied as is.

[0054] Next, when the learning in step S2 is completed, the IED probability map calculation unit 502 creates a trained model (step S13). The created trained model is used for machine learning inference, that is, when detecting IEDs in unknown data.

[0055] Here, the IED probability map calculated by the IED probability map calculation unit 502 will be described.

[0056] Fig. 7 is a diagram showing an example of an IED probability map. In Fig. 7, the horizontal axis represents time, and the vertical axis represents the sensor index. In the example shown in Fig. 7, it is estimated that an IED appeared when the time index was 100 and the sensor index was around 40 to 80. Note that in Fig. 7, the IED probability is expressed in shades of black and white, but a color display in which the color changes according to the IED probability may also be used.

[0057] 5, next, the threshold processing unit 503 performs threshold processing on the IED probability map obtained by the IED probability map calculation unit 502 (step S4). Specifically, the threshold processing unit 503 narrows down the time and sensor area with a high IED probability from the IED probability map obtained by the IED probability map calculation unit 502 using a threshold.

[0058] In the threshold processing unit 503, the higher the threshold is set, the more likely the waveforms are to be selected as those that are IED-like, but the number of IEDs for which dipole estimation is performed decreases. On the other hand, in the threshold processing unit 503, when the threshold is set low, the number of false positives increases, but more IEDs can be detected. Furthermore, since magnetoencephalography data is strongly influenced by individual differences, there is a possibility that IEDs will hardly be detected unless the threshold is lowered.

[0059] Therefore, the threshold processing unit 503 may be configured to interactively control the number of IEDs to be detected in cooperation with a UI (User Interface).

[0060] The threshold processing unit 503 basically uses a preset value such as "0.8" as the threshold.

[0061] Here, Fig. 8 is a diagram showing an example of an IED probability map after threshold processing. In the example shown in Fig. 8, the threshold processing unit 503 uses "0.5" as the threshold. In the IED probability map 605, which is a two-dimensional map shown in Fig. 8, this corresponds to extracting locations where an IED exists with a probability of 50% or more.

[0062] Returning to FIG. 5, next, the post-processing unit 504 executes post-processing to extract sampling time points and sensors to be used for dipole estimation from the IED probability map 605 thresholded by the threshold processing unit 503 (step S5).

[0063] Specifically, when the threshold processing unit 503 is applied, the time and the sensor area having a value equal to or greater than a predetermined probability are extracted. In order to clarify the peak detection or the detection time of the IED, a filter such as a Gaussian filter may be applied to the extracted sensor area before the peak detection. The time of the IED is determined by applying a filter such as a Gaussian filter to the extracted sensor area before the peak detection and then performing the peak detection. When performing the peak detection, the detection may be performed on a one-dimensional probability map obtained by averaging the probability map including the extracted sensor area in the sensor direction. If the averaging is not performed, the peak detection may be performed for each sensor, and the time of the IED may be determined by averaging the time points.

[0064] The threshold processing unit 503 extracts sensors having a value equal to or greater than a predetermined probability at the time of the detected peak as sensors to be used for dipole estimation. At this time, it is known that the stability of the dipole estimation solution decreases when the number of sensors used for dipole estimation is reduced. Therefore, it is possible to prevent dipole estimation from being performed at the IED when the number of selected sensors is less than a specified number.

[0065] As another method of extracting sensors, the stability of the dipole estimation solution can be improved by defining a group of sensors in advance and expanding the extracted sensor to all sensors in the group to which the sensor belongs. Here, the group is basically set based on anatomical criteria such as the vicinity of the temporal lobe or the vicinity of the frontal lobe, but a neighboring area containing any number of sensors may simply be set as one group.

[0066] Here, FIG. 12 is a conceptual diagram showing an example of a sensor expansion method. FIG. 12(a) shows a subject's head 200, magnetoencephalogram sensors 210 arranged to cover the head 200, and a group 220 of the sensors 210 set in advance. In FIG. 12(b), the sensors 210a shown by black circles represent sensors extracted at the threshold processing stage. It is possible to perform dipole estimation using only the sensors 210a extracted at the threshold processing stage, but as described above, it is also possible to perform sensor expansion in order to increase the stability of the dipole estimation solution. FIG. 12(c) shows the result of expanding the number of sensors. That is, as shown in FIG. 12(c), the sensor 210a extracted at the threshold processing stage is expanded to all sensors 210 in the group 220 to which the sensor 210a belongs. This can increase the stability of the dipole estimation solution.

[0067] Post-processing unit 504 can also perform post-processing, described below, to separate the IED probability map if it has multiple peaks (such as may occur when there are two lesions).

[0068] Another example of setting the neighborhood region is to construct a convex hull from the selected sensor coordinates, then perform a morphological transformation on the convex hull with a predetermined threshold value to expand the space, and group the sensors included in the expanded space into one group.

[0069] Alternatively, the sensor data may be acquired in order from the nearest point based on the midpoint of the spatial coordinates of the sensor (source) having the maximum magnetic field value and the sensor (sink) having the minimum magnetic field value.

[0070] In addition, a spatial isomagnetic field map may be reconstructed at a predetermined magnetic field threshold from the magnetic field information obtained from the sensor data, and from the magnetic field information of the selected sensor, a sensor that is in the vicinity of the magnetic field information of the already selected sensor and belongs to the same isomagnetic field line may be selected.

[0071] Sensor groups may also be configured from a spatial and anatomical perspective by configuring groups based on a combination of a determination means using the spatial distribution of the magnetic field in the vicinity area and anatomical criteria.

[0072] Finally, the dipole estimation unit 505 performs dipole estimation (see Non-Patent Document 3) using the sensor obtained by the processing before the dipole estimation unit 505 and the sampling time point of the IED's origin (step S6). In this case, in addition to dipole estimation, it is possible to use a current source analysis method using a spatial filter method such as the minimum norm method (see Non-Patent Document 6) or the LCMV Beamformer method (see Non-Patent Document 7). The spatial filter method is a method in which tens of thousands of dipoles are placed in the brain in advance and the change in current over time in each dipole can be obtained.

[0073] Here, Fig. 9 is a diagram showing an example of an IED probability map after post-processing. The post-processing unit 504 executes peak detection and threshold processing on the IED probability map after threshold processing obtained in Fig. 8, thereby extracting the sampling time point of dipole estimation and the sensor. In the example shown in Fig. 9, the post-processing unit 504 executes peak detection and threshold processing to obtain the designated sampling time point 511 of dipole estimation and the channel 512 from which the sensor is extracted.

[0074] The dipole estimation unit 505 performs dipole estimation using a sampling time point 511 for dipole estimation and a channel 512 from which the sensor is extracted.

[0075] Note that the example shown in FIG. 9 simulates the case where there is only one spike, but if there are two or more spikes, the post-processing unit 504 can apply a process to separate the maps using a watershed algorithm (see Non-Patent Document 8) or a random walk partitioning algorithm (see Non-Patent Document 9), making it possible to perform dipole estimation targeting only a single IED.

[0076] Here, Fig. 10 is a diagram showing an example of a UI using an IED probability map. As shown in Fig. 10, by superimposing the IED probability map, which is a two-dimensional map calculated as described above, on the original waveform, it is possible to get an overview of which part of the waveform the trained network focused on. This makes it possible to eliminate data that adversely affects the estimation results, such as false positives that adversely affect the diagnosis of the localization of the estimated epilepsy lesion. When displaying, all sensors may be displayed as is, but it is also possible to display only manually selected sensors or to display in units of preset sensor groups.

[0077] By displaying the data in this manner, it is possible to compare the estimated results with the original waveform, enabling more accurate selection of the IED time and sensor, thereby enabling more accurate localization diagnosis of epileptic lesions.

[0078] In the example shown in Fig. 10, the image is drawn so that the shade changes according to the estimated probability, but it is also possible to set a color map according to the probability. For example, a color map may be applied that changes the color linearly so that a probability of 0 is black, and as the probability increases, the color changes from blue to red, and a probability of 1 is white.

[0079] In particular, the IED probability map calculation unit 502 learns the ideal IED time point and the narrowed down sensors as correct answer data in the equivalent current dipole method during machine learning, thereby improving the accuracy of dipole estimation. The example shown in Fig. 10 is an example in which the origin of an epileptic IED and the sensor where the IED appears are extracted using machine learning. The dark areas in Fig. 10 represent the areas estimated by machine learning.

[0080] Although the above description has been given using an example of a magnetoencephalograph, the present invention is not limited to this and can also be applied to an electroencephalograph.

[0081] As described above, according to this embodiment, by using the IED probability map, it becomes possible to detect the IED onset time and select the sensor required for the equivalent current dipole method, making it possible to more accurately determine the time when characteristic waveform information (IED) appears and extract the sensor, thereby enabling localized diagnosis of epileptic lesions using the equivalent current dipole method.

[0082] In addition, by using an automatic IED detection method that uses machine learning, it is possible to automate the detection of IEDs and the determination of their origin.

[0083] Here, a case will be described in which the IED probability map calculation unit 502 uses a method other than machine learning to generate the IED probability map.

[0084] Fig. 11 is a diagram showing an example of calculating the probability map of an IED using a method other than machine learning. In the example shown in Fig. 11, the IED probability map calculation unit 502 uses template matching. The IED probability map calculation unit 502 detects an IED using a template that generalizes the shape of an IED created from a document (see Non-Patent Document 10) that investigates the morphological characteristics of IEDs in magnetoencephalograms.

[0085] 11, the IED probability map calculation unit 502 scans templates such as templates 601 to 603 for the time series of MEG data from one sensor in MEG data 604, and creates a two-dimensional map, IED probability map 605, based on the similarity. Although the map created here is not strictly a probability map, it can be used as a map for the purpose, and is therefore called a probability map for convenience.

[0086] Similarity can be calculated using one-dimensional methods of calculating similarity in template matching in images, such as normalized cross-correlation (NCC) or zero-mean normalized cross-correlation (ZNCC), or a simple normalized convolution with a filter. In addition to template matching, several methods can be applied to detect IEDs, such as a method that derives the IED by detecting a steep transition from the baseline.

[0087] In addition, if machine learning is not used in the IED probability map calculation unit 502, dipole estimation at the peak position may produce undesirable results, so post-processing is required, such as performing the estimation at a time point slightly earlier than the sampling time.

[0088] By using an automatic IED detection method that uses an algorithm in this way, it is possible to automate the detection of IEDs and the determination of their origin.

[0089] Furthermore, in the above-described embodiment, when at least any of the functional units of the biosignal measurement system 1 is realized by execution of a program, the program is provided by being pre-installed in a ROM, etc. Furthermore, the program executed by the biosignal measurement system 1 according to the above-described embodiment may be provided by being recorded in a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R (Compact Disc Recordable), or a DVD (Digital Versatile Disc) in the form of an installable or executable file.

[0090] Furthermore, the programs executed by the biosignal measuring system 1 of the above-described embodiment may be configured to be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network.

[0091] The program executed in the biosignal measurement system 1 of the above embodiment may be configured to be provided or distributed via a network such as the Internet. The program executed in the biosignal measurement system 1 of the above embodiment has a modular configuration including at least any one of the above-mentioned functional units, and as actual hardware, the CPU reads out the program from a ROM or the like and executes it, so that the above-mentioned functional units are loaded and generated on a main storage device. [Explanation of symbols]

[0092] 605 2D Map [Prior art documents] [Patent documents]

[0093] [Patent Document 1] Patent No. 3107020 [Non-patent literature]

[0094] [Non-Patent Document 1] E. Javier, H. Roberto, A. Daniel, F. Alberto, and LC Miguel, “Artifact removal in magnetoencephalogram background activity with independent component analysis,” IEEE Trans Biomed Eng, vol. 54, no. 11, pp. 1965-1973, 2007. [Non-Patent Document 2] K. Sekihara, Y. Kawabata, S. Ushio, S. Sumiya, S. Kawabata, Y. Adachi, and S. S. Nagarajan, “Dual signal subspace projection (DSSP): a novel algorithm for removing large interference in biomagnetic measurements,” Journal of Neural Engineering, vol. 13, no. 3, p. 036007, 2016. [Non-Patent Document 3] M. Scherg, “Fundamentals of dipole source potential analysis,” in Auditory Evoked Magnetic Fields and Potentials, M. Hoke, F. Grandori, and G. L. Romani, Eds. Basel, Switzerland: Karger, 1989, vol. 6 [Non-Patent Document 4] A. Ossadtchi, S. Baillet, J. Mosher, D. Thyerlei, W. Sutherling, and R. Leahy, “Automated interictal spike detection and source localization in magnetoencephalography using independent components analysis and spatio-temporal clustering,” Clinical Neurophysiology, vol. 115, no. 3, pp. 508-522, 2004. [Non-Patent Document 5] O. Ronneberger, P. Fischer, and T. Brox, "U-net: Convolutional networks for biomedical image segmentation," Proc. Int. Conf. Medical Image Comput. Comput.-Assisted Intervention, pp. 234 - 241, 2015. [Non-Patent Document 6] K. Sekihara, M. Sahani, and S. S. Nagarajan, “Localization bias and spatial resolution of adaptive and non-adaptive spatial filters for MEG source reconstruction,” NeuroImage, vol. 25, no. 4, pp. 1056 - 1067, 2005. [Non-Patent Document 7] B. V. Veen, W. V. Drongelen, M. Yuchtman, and A. Suzuki, “Localization of brain electrical activity via linearly constrained minimum variance spatial filtering,” IEEE Transactions on Biomedical Engineering, vol. 44, no. 9, pp. 867 - 880, 1997. [Non-Patent Document 8] J. Cousty, G. Bertrand, L. Najman, and M. Couprie, “Watershed Cuts: Minimum Spanning Forests and the Drop of Water Principle,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 8, pp. 1362 - 1374, 2009. [Non-Patent Document 9] L. Grady, “Random Walks for Image Segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 11, pp. 1768 - 1783, 2006.

Non - Patent Document 10

Claims

1. A comparison step of comparing each of the waveform data acquired by the plurality of sensors with at least one characteristic waveform information; a determining step of determining an appearance probability of characteristic waveform information in at least a certain section of the waveform data based on a correlation ratio between the section of the waveform data and the characteristic waveform information; a narrowing step of extracting a sensor area using a threshold value for the determined appearance probability of the characteristic waveform information, and narrowing down the time of the characteristic waveform information and the sensor area by applying a filter to the area; a step of identifying a time and an index of a sensor at which a section matching characteristic waveform information appears, using the time and sensor area narrowed down in the narrowing down step; 11. A method for identifying waveform occurrences comprising:

2. The determining step displays a two-dimensional map representing the likelihood of occurrence of characteristic waveform information in at least a certain section of the waveform data.

2. The method of claim 1, wherein the waveform generation is determined by a step of:

3. The determining step displays the two-dimensional map in a superimposed manner on each of the waveform data acquired by the plurality of sensors.

3. The method of claim 2, wherein the waveform generation is determined by the first step.

4. The determining step calculates a probability map of characteristic waveform information using a pre-trained machine learning model.

2. The method of claim 1, wherein the waveform generation is determined by a step of:

5. In the determining step, a time point of ideal characteristic waveform information in the equivalent current dipole method and an index of the narrowed down sensor are learned as correct answer data during machine learning.

5. The method of claim 4, wherein the waveform generation is determined by the first step.

6. The identification step includes: defining a group of sensors in advance; and narrowing down the number of sensors by expanding a sensor having a value equal to or greater than a predetermined probability in the probability map to all sensors in the group to which the sensor belongs.

5. The method of claim 4, wherein the waveform generation is determined by the first step.

7. The determining step displays the two-dimensional map in a color manner such that the color changes according to the probability of characteristic waveform information.

3. The method of claim 2, wherein the waveform generation is determined by the first step.

8. A comparison step of comparing each of the waveform data acquired by the plurality of sensors with at least one characteristic waveform information; a determining step of determining an appearance probability of characteristic waveform information in at least a certain section of the waveform data based on a correlation ratio between the section of the waveform data and the characteristic waveform information; a narrowing step of extracting a sensor area using a threshold value for the determined appearance probability of the characteristic waveform information, and narrowing down the time of the characteristic waveform information and the sensor area by applying a filter to the area; a step of identifying a time and an index of a sensor at which a section matching characteristic waveform information appears, using the time and sensor area narrowed down in the narrowing down step; A program for causing a computer to execute the following.

9. The determining step displays a two-dimensional map representing the likelihood of occurrence of characteristic waveform information in at least a certain section of the waveform data.

9. The program according to claim 8 .

10. The determining step displays the two-dimensional map in a superimposed manner on each of the waveform data acquired by the plurality of sensors.

10. The program according to claim 9 .

11. The determining step calculates a probability map of characteristic waveform information using a pre-trained machine learning model.

9. The program according to claim 8 .

12. In the determining step, a time point of ideal characteristic waveform information in the equivalent current dipole method and an index of the narrowed down sensor are learned as correct answer data during machine learning.

12. The program according to claim 11 .

13. The identification step includes: defining a group of sensors in advance; and narrowing down the number of sensors by expanding a sensor having a value equal to or greater than a predetermined probability in the probability map to all sensors in the group to which the sensor belongs.

12. The program according to claim 11 .

14. The determining step displays the two-dimensional map in a color manner such that the color changes according to the probability of characteristic waveform information.

10. The program according to claim 9 .

Citation Information

Patent Citations

  • Local current measuring device for organism

    JP1990082941A

  • Brain and heart primary current tomography system and method

    JP2001514031A

  • Apparatus and method for measuring, estimating, and displaying RMS current density diagrams

    JP2002500909A

  • A system for predicting, early detection, warning, preventing, or controlling changes in brain activity.

    JP2002502270A

  • Method of fetal and maternal ECG identification across multiple epochs

    JP2009160410A