Mild cognitive impairment screening method, computer program accompanied by the method, and screening device

Abstract

To provide a mild cognitive impairment screening method, a computer program accompanied by the method, and a screening device capable of executing highly accurate MCI screening easily and in a short time without using a large-sized complicated and expensive device and without needing specialized knowledge.SOLUTION: A mild cognitive impairment screening method includes: a measurement step S1 for measuring a finger-tapping motion which is an opening and closing motion of two fingers; a calculation step S3 for calculating a predetermined feature amount based on measurement data obtained by the measurement step within a predetermined time at the time of finger-tapping with only the right hand and / or at the time of finger-tapping with only the left hand and / or at the time of finger-tapping with both hands simultaneously and / or at the time of finger-tapping with both hands alternately; and a determination step S5 for making a determination regarding mild cognitive impairment based on the feature amount calculated in the calculation step S3. In the determination step S5, a discriminant model is generated by machine-learning including artificial intelligence (step S4), and a determination regarding mild cognitive impairment is made using the generated discriminant model.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a method for screening for mild cognitive impairment, and in particular to a method for measuring finger tapping movements and screening for mild cognitive impairment using the measurement results, as well as a computer program and screening device that accompany the method. [Background technology]

[0002] As the aging society progresses, the number of patients with Alzheimer's disease is increasing year by year, and if it is detected early, the progression of the disease can be slowed with medication. However, it can be difficult to distinguish between symptoms associated with aging, such as forgetfulness, and the disease itself, and in many cases, patients only visit a hospital once the condition has progressed to a severe stage.

[0003] In this situation, conventional screening tests for the early detection of Alzheimer's disease have included blood tests, olfactory tests, and tests that replicate a doctor's interview on a tablet device. However, these tests have presented problems such as the pain of blood sampling and the length of the test time, which places a heavy burden on the subject. Meanwhile, cognitive function assessments based on button presses or measuring finger movements of one hand using a tablet device have also been conducted as less burdensome tests, but these have the drawback of not achieving sufficient test accuracy. A simple screening test that is highly accurate and less burdensome for the subject could lead to the early detection of Alzheimer's disease, improve patients' quality of life, and contribute to reducing medical and nursing care costs.

[0004] In recent years, it has become clear that the opening and closing movements of two fingers (finger tapping movements) using the thumb and index finger of both hands can reveal movement patterns specific to Alzheimer's disease. High correlations have been confirmed between finger movement measurements and dementia tests based on general interviews. These findings are believed to be the result of finger tapping measurements capturing the decline in rhythmic motor function of both hands due to brain atrophy in Alzheimer's disease. Furthermore, the hands are said to be the second brain, and many brain regions are involved in finger function. Finger movements are also said to be related not only to Alzheimer's disease, but also to dementias such as vascular dementia and Lewy body dementia, Parkinson's disease, and developmental coordination disorder (e.g., difficulty skipping or jumping rope). In other words, finger tapping can reveal brain function. Furthermore, finger tapping can be used as a "ruler" for brain health, allowing for the quantification of fine motor function of the hands and fingers, which can be used in a variety of fields, including healthcare, rehabilitation, and lifestyle assistance.

[0005] As a method for accurately measuring and evaluating finger tapping movements, for example, Patent Document 1 discloses a motor function evaluation system that includes a motor function measurement device that calculates motor data based on the relative distance between a pair of a transmitter coil and a receiver coil attached to a moving part of a living body, and an evaluation device that evaluates the motor function of the living body based on the motor data received from the motor function measurement device. That is, Patent Document 1 shows that a magnetic sensor attached to the fingertips converts changes in magnetic force that fluctuate during tapping movements with two fingers into electrical signals, and measures and quantifies the movements to capture feature quantities that indicate the characteristics of the finger movements, thereby revealing the state of brain function. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-49123 Summary of the Invention [Problem to be solved by the invention]

[0007] As mentioned above, early detection of dementia is extremely important, and to do so, it is also necessary to distinguish between mild cognitive impairment (MCI), a precursor to dementia that signifies a decline in subjective cognitive function.

[0008] Such screening for MCI has traditionally been carried out through a diagnosis by a dementia specialist, a comprehensive geriatric assessment (CGA) by a clinical psychologist, psychological tests (FAB, RCPM, ADAS, etc.), magnetic resonance imaging (MRI), single-photon emission computed tomography, or electrocardiogram and blood tests, and ultimately a comprehensive judgment by a dementia specialist based on the results of these tests.

[0009] However, MCI screening using such methods is accompanied by various problems, such as the tests being invasive, taking a long time to complete, requiring specialized knowledge, and requiring the use of complex, large, and expensive equipment.

[0010] The present invention has been made in consideration of the above circumstances, and aims to provide a mild cognitive impairment screening method, a computer program associated with the method, and a screening device that enable highly accurate screening of MCI to be performed easily and in a short time without using complex, large, and expensive equipment and without requiring specialized knowledge. [Means for solving the problem]

[0011] In order to solve the above problems, the mild cognitive impairment screening method of the present invention includes: a measuring step of measuring finger tapping movement, which is an opening and closing movement of two fingers; a calculation step of calculating a predetermined feature amount based on measurement data obtained by the measurement step within a predetermined time when finger tapping is performed with only the right hand and / or only the left hand and / or both hands simultaneously and / or both hands alternately; a determination step of determining mild cognitive impairment based on the feature amount calculated in the calculation step; Including, The determination step is characterized by generating a discriminant model by machine learning including artificial intelligence, and making a determination regarding mild cognitive impairment using the generated discriminant model.

[0012] The present inventors have conducted clinical research focusing on finger tapping movements and have succeeded in extracting movement patterns specific to individuals with MCI from tapping movements involving the thumb and index finger of both hands. The inventors have then utilized technology that uses magnetic sensors to quantify dexterous finger movements, and have discovered that highly accurate screening for mild cognitive impairment can be performed by processing the various features of the tapping movement of repeatedly opening and closing two fingers using machine learning including artificial intelligence. Based on these results, the present invention provides a testing method for supporting the early detection of individuals with MCI, i.e., a mild cognitive impairment screening method.

[0013] Specifically, in machine learning, data on healthy individuals and data on individuals with MCI are provided as information regarding finger tapping movements of the right hand only and / or finger tapping movements of the left hand only and / or finger tapping movements of both hands simultaneously and / or finger tapping movements of both hands alternately. Based on this information, a discriminant model for distinguishing between healthy individuals and individuals with MCI (determining whether a person has MCI) is generated by machine learning according to a predetermined algorithm, including artificial intelligence. Then, a diagnosis of MCI is made using the generated discriminant model. More specifically, in this machine learning, a previously prepared healthy group and MCI group are input into a machine learning algorithm capable of distinguishing between the two groups, which will be described later, to generate a discriminant model. Then, a diagnosis result is output when new subject data is input.

[0014] According to the above-mentioned configuration of the present invention, a huge amount of data, which is a wide variety of feature quantities, is processed by machine learning including artificial intelligence, so that MCI screening can be performed easily and quickly with high accuracy without requiring specialized knowledge. Moreover, the finger tapping measurement itself can be performed non-invasively, easily, and using a simple and compact device, so measurement and evaluation can be performed in a short measurement time with little burden on the subject.

[0015] In the above configuration, the measurement step of measuring finger tapping movements, which are the opening and closing movements of two fingers, may use any measurement method as long as it can measure finger tapping movements. For example, it may be a measurement method using magnetic sensors attached to two fingers, a measurement method that processes image data obtained by photographing finger movements, or a measurement method based on detecting finger movements tapping on a touch panel. Furthermore, in the above configuration, "determination of mild cognitive impairment" includes determining whether or not a subject has MCI, determining whether the likelihood of MCI is high or low, and presenting (determining) the probability of MCI and the degree of MCI.

[0016] Furthermore, in the above configuration, it is preferable that, among the many feature amounts obtained by various finger tapping movements, MCI is determined based on feature amounts that show particularly significant differences between healthy individuals and individuals with MCI.

[0017] In addition to the aforementioned mild cognitive impairment screening method (for example, by processing by a computer according to a predetermined algorithm, etc.), the present invention also provides a computer program that causes a computer to execute the method, and a mild cognitive impairment screening device that can execute the method. [Effects of the Invention]

[0018] The mild cognitive impairment screening method, computer program, and device of the present invention enable highly accurate screening for MCI to be performed easily and quickly without using complex, large, and expensive equipment and without requiring specialized knowledge. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a block diagram showing a schematic configuration of a mild cognitive impairment screening device according to one embodiment of the present invention. [Figure 2] FIG. 1 is a schematic diagram showing both hands of a subject with tapping sensors attached to the thumbs and index fingers. [Figure 3] 2 is a flowchart showing an example of a mild cognitive impairment screening method according to one embodiment of the present invention that can be executed by the mild cognitive impairment screening device of FIG. 1. [Figure 4] 1 is a table showing an example of feature amounts required for determining mild cognitive impairment screening. [Figure 5] 1 is a table showing an example of a machine learning algorithm (machine learning model) for discriminating between a healthy group and an MCI group. [Figure 6] 10 is a table showing the discrimination accuracy (F1 score is in the top 5%) between the MCI group and the healthy group for individual feature amounts. [Figure 7] 1 is a table showing the results of discrimination between a healthy group and an MCI group using each machine learning model; DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In this embodiment, by providing the following technology, highly advanced technology contributes to the development of medical care and the realization of a healthy society. The realization of this mild cognitive impairment screening method (device and computer program) contributes to "9. Build resilient infrastructure, promote inclusive and sustainable industrialization, promote innovation and build resilient infrastructure" of the Sustainable Development Goals (SDGs) advocated by the United Nations.

[0021] In addition, in the following embodiments, a mild cognitive impairment screening method (apparatus) will be described, but the present invention may also be configured as a computer program that enables a computer to perform the processing executed by the mild cognitive impairment screening method (apparatus).

[0022] 1 shows a schematic configuration of a mild cognitive impairment screening device 1 according to one embodiment of the present invention. As shown in the figure, this mild cognitive impairment screening device 1 includes a measurement unit 10 having a tapping sensor 2 that magnetically detects finger tapping movement, which is the opening and closing movement of two fingers, and a processor 30 that processes measurement data measured by the measurement unit 10.

[0023] The measurement unit 10 measures finger movement data based on the relative distance between a pair of a transmitter coil and a receiver coil attached to a living finger (or other moving part), and, for example, detects information on the subject's finger movement in a time series, and can acquire the subject's movement information regarding at least one of distance, speed, acceleration, and jerk (acceleration differentiated with respect to time) as time series data (waveform data).

[0024] The measurement unit 10 includes a tapping sensor 2, first and second switching circuits 4 and 5, an AC generator 6 for generating AC, an amplifier / filter circuit 7, an A / D converter 8, a detector 9, a downsampler 10 for downsampling, and a controller 11 for controlling the operation of these components.

[0025] The tapping sensor 2 is composed of a pair of a transmitter coil 2A (2A') and a receiver coil 2B (2B') (or multiple rows of coil pairs), and is attached to the fingers (e.g., nails) of the subject's hand 100 (100A, 100A') using, for example, double-sided tape or a fixing band, as shown in FIG. 2. Specifically, in FIG. 2, a pair of a transmitter coil 2A and a receiver coil 2B is attached to the thumb 100a and index finger 100b of the subject's right hand 100A, and a pair of a transmitter coil 2A' and a receiver coil 2B' is attached to the thumb 100a and index finger 100b of the subject's left hand 100A' (the fingers on which the coils are attached may be reversed, or other fingers may be attached). In this case, the transmitter coil 2A (2A') emits a magnetic field, and the receiver coil 2B (2B') receives (detects) the magnetic field emitted by the transmitter coil 2A (2A').

[0026] One AC generator 6 is connected to the transmitter coil 2A (2A') via a first switching circuit 4. Through the switching operation of the first switching circuit 4, AC current (e.g., a 20 kHz current) from the AC generator 6 flows sequentially through the transmitter coil 2A (2A'), causing the transmitter coil 2A (2A') to generate an AC magnetic field through which the AC current flows. The AC generator 6 generates an AC current of a predetermined frequency, and the timing at which the current flows is controlled by a controller 11. The signal generated by the AC generator 6 is used as a reference signal for the detection operation of the detector 9.

[0027] The controller 11 generates a synchronization signal for controlling the first and second switching circuits 4 and 5. This synchronization signal enables the first switching circuit 4 and the second switching circuit 5 to be switched simultaneously, and they operate sequentially for each pair of the transmitting coil 2A (2A') and the receiving coil 2B (2B').

[0028] The receiving coil 2B (2B') is also connected to an amplifier / filter circuit 7 via a second switching circuit 5, and the output signal from the amplifier / filter circuit 7 is converted into a digital signal by an A / D converter 8, which is then transmitted to a detector 9. The conversion of analog data into digital data by the A / D converter 8 facilitates subsequent processing (downsampling, etc.). The detector 9 also performs processing to remove a predetermined period of the AC magnetic field waveform (noise portion) detected by the receiving coil 2B (2B') immediately after switching by the second switching circuit 5.

[0029] In addition, the time of the deletion process in the AC magnetic field waveform of each receiving coil 2B (2B') is precisely controlled by the controller 11. After this deletion process, the detector 9 performs full-wave rectification and filtering (mainly processing using a low-pass filter (LPF)) using the reference signal described above. Finally, the digital signal processed by the detector 9 is converted (downsampled) by the downsampler 10 into coarse data with a sampling frequency (e.g., 200 Hz) that is approximately 1 / 1000 (a predetermined ratio) of the sampling frequency (e.g., 200 kHz) of the A / D converter 8. This makes it possible to reduce the overall data volume. Therefore, the output signal can be transmitted at high speed as data from multiple receiving coils even with limited communication capacity. In other words, because the amount of data received by the communication interface 12 of the measurement unit 10 from the downsampler 10 is small, finger movement data related to multiple receiving coils can be transferred to the processor 30 (via the communication interface 31 of the processor 30) wirelessly or via a wired connection at once.

[0030] The processor 30, which processes the measurement data measured by the measurement unit 10, includes an arithmetic circuit 33 that calculates a predetermined number of feature quantities within a predetermined time based on detection information detected by the tapping sensor 2, specifically, based on measurement data output from the measurement unit 10 that measures finger tapping movements of only the right hand and / or only the left hand and / or simultaneous finger tapping movements of both hands and / or alternate finger tapping movements of both hands (for each group of healthy subjects and MCI subjects) using the tapping sensor 2 (the selection of the calculated feature quantities or their number may be performed automatically based on a predetermined algorithm or may be performed based on a command input through an operation input interface 38, which will be described later), and a judgment circuit 34 that makes a judgment regarding mild cognitive impairment based on the feature quantities calculated by the arithmetic circuit 33. Here, in this embodiment, the judgment circuit 34 generates a discriminant model by machine learning including artificial intelligence, and makes a judgment regarding mild cognitive impairment using the generated discriminant model. The processor 30 further includes an adjustment circuit 40 that adjusts optimal values ​​of hyperparameters in machine learning performed by the determination circuit 34 so that the F1 score (F1 value), which is an evaluation index in a machine learning model, is 0.7 or greater.

[0031] An example of features used to determine MCI is shown in Figure 4 and listed below. 1. Maximum distance amplitude between two fingers 2. Total travelling distance of two fingers 3. Average of local maximum distance between two fingers 4. Standard deviation of the local maximum distance between two fingers (SD of local maximum distance) 5. Slope of approximate line of local maximum point of distance between two fingers (an approximate curve is a regression line when time is plotted on the horizontal axis and the maximum value on the vertical axis). 6. Coefficient of variation of the local maximum distance between two fingers (CV of local maximum distance) (The coefficient of variation is the standard deviation divided by the mean.) 7. SD of local maximum distance in three adjacent taps, i.e., the standard deviation of the maximum distance between the two fingers in three adjacent taps. 8. Maximum velocity amplitude, i.e., the maximum value of finger velocity during two-finger opening and closing movement 9. Average of local maximum velocity of fingers during two-finger open movement 10. Average of local maximum velocity of fingers during closed finger movement 11. Standard deviation of local maximum velocity of fingers during two-finger open movement (SD of local maximum velocity) 12. Standard deviation of the local minimum velocity of the fingers during closed finger movements (SD of local minimum velocity) 13. Energy balance, i.e., the ratio of the sum of squares of the velocities during the open finger movement to the sum of squares of the velocities during the closed finger movement 14. Total energy, i.e., the sum of the squares of the velocities during the entire measurement time 15. Coefficient of variation of local maximum velocity of finger during two-finger open movement (CV of local maximum velocity) 16. Coefficient of variation of local minimum velocity of fingers during closed finger movement (CV of local minimum velocity) 17. Number of freezings calculated from velocity, i.e., the number of times the velocity waveform changes positive and negative minus the number of large open and close finger taps. 18. Average distance rate of peak velocity in extending movement (Average of distance rate of peak velocity in extending movement), i.e., the average value of the ratio of the distance at the maximum velocity during the extending movement of two fingers, when the amplitude of the finger tap is set to 1.0 19. Average distance rate of peak velocity in flexing movement, i.e., the average ratio of the distance at the minimum velocity during flexing movement to the amplitude of finger tapping set to 1.0. 20. Ratio of 18 to 19 (ratio of distance rates of velocity peak in extending and flexing movements) 21. Standard deviation of distance rate of velocity peak in extending movement 22. Standard deviation of distance rate of peak velocity in flexing movement 23. Maximum amplitude of finger acceleration during two-finger closing movement 24. Average of local maximum acceleration in extending movement 25. Average of local minimum acceleration in extending movement 26.Average of local maximum acceleration in flexing movement 27.Average of local minimum acceleration in flexing movement 28.Average contact duration between two fingers during tapping 29. Standard deviation of contact duration between two fingers during tapping (SD of contact duration) 30. Coefficient of variation of contact duration between two fingers during tapping (CV of contact duration) 31. Number of zero crossover points of acceleration during tapping, i.e., the average number of times acceleration changes positive and negative during one cycle of finger tapping. 32. Number of freezing calculated from acceleration, i.e., the number of times the acceleration changes positive and negative during one finger tap cycle minus the number of large open / close finger taps 33. Number of taps 34. Average of intervals, which is the time interval between each tap. 35. Average tapping interval frequency (Frequency of taps) 36. Standard deviation of inter-tapping interval (SD of inter-tapping interval) 37. Coefficient of variation of inter-tapping interval (CV of inter-tapping interval) 38. Inter-tapping interval variability, i.e., the integrated value of the frequency range of 0.2 to 2.0 Hz when the tap interval is spectrally analyzed. 39. Skewness of inter-tapping interval distribution, i.e., the skewness in the frequency distribution of tap intervals, and the degree to which the frequency distribution is skewed compared to a normal distribution. 40. Local standard deviation of inter-tapping interval (SD of inter-tapping interval in three adjacent taps), i.e., the standard deviation of the inter-tapping interval in three adjacent taps 41. Average of phase difference between the left and right tapping 42. SD of phase difference between the left and right finger tapping 43. Similarity of the waveforms of both hands, that is, the value that represents the correlation when the time lag is 0 when applying the cross-correlation function to the waveforms of the left and right hands 44. Time lag of similarity of hands, i.e., the time lag at which the correlation calculated in the previous section is maximized.

[0032] Of the 44 features, items 1 to 40 are obtained by finger tapping with either the right or left hand, and items 41 to 44 are obtained by simultaneous or alternating tapping with both hands. Forty features (items 1 to 40) are obtained for each hand, resulting in a total of 40 × 2 = 80 features for both hands. For simultaneous and alternating tapping with both hands, 80 features (40 × 2 = 80 for items 1 to 40 and 4 for items 41 to 44) are obtained for each hand, resulting in a total of 84 features (84 × 2 = 168 for simultaneous and alternating tapping). As a result, a total of 80 + 168 = 248 features are obtained per subject.

[0033] Although a diagnosis of mild cognitive impairment may be made based on all of these features, in this embodiment in particular, a diagnosis of mild cognitive impairment is made based on the following features out of these 248 features, which show a significant difference between healthy individuals and MCI individuals (high correlation between healthy individuals and MCI individuals):

[0034] That is, in this embodiment, the feature amounts of items 4, 12, 15, 16, 21, 25, 27, 28, 29, 32, 34, 36, 38, and 40 are used for the finger tapping movement of only the left hand, the feature amounts of items 16, 25, 27, 28, 29, 34, and 38 are used for the finger tapping movement of only the right hand, and the feature amounts of items 12, 15, 18, 25, 27, 28, 29, 34, and 38 are used for the left hand and items 18, 29, 34, and 38 are used for the finger tapping movement of both hands simultaneously. The features of items 0, 25, 28, 29, 33, 34, 35, and 38, as well as the features of items 17, 18, 20, 21, 22, 25, 28, 29, 30, 31, 32, 34, 36, 37, 38, and 40 on the left hand and items 3, 20, 25, 27, 28, 29, 31, 32, 34, 36, 37, 38, and 40 on the right hand (hereinafter referred to as selected features) are used to determine mild cognitive impairment.

[0035] In addition, the mild cognitive impairment screening device 1 further includes a display 37 that displays various data (such as the calculation results calculated by the calculation circuit 33) including the judgment results by the judgment circuit 34 of the processor 30, a memory 36 that stores the various data, and an operation input interface 38 that can input necessary data and commands to the processor 30 by operation.

[0036] In the above configuration, the processor 30 is composed of a CPU, etc., and executes programs such as an operating system (OS) and various operation control applications stored in the memory 36, thereby performing operation control processing for the various circuits 33, 34, 40 described above and controlling the startup operations of various applications.

[0037] Memory 36 is configured with a flash memory or the like, and stores programs such as an operating system and applications for controlling the operation of various processes such as images, audio, documents, displays, measurements, etc. Memory 36 also stores information data such as base data required for basic operations by the operating system and file data used by various applications.

[0038] The processing by the processor 30 may be stored as one application, and the measurement of finger movements and the calculation and analysis of various feature quantities may be performed by activating the application. Alternatively, an external server device with high computing performance and large capacity may receive the measurement results from the information processing terminal and calculate and analyze the feature quantities.

[0039] The operation input interface 38 generally uses input means such as a keyboard, key buttons, touch keys, etc., but may also use, for example, gesture operation or voice input, and is used to set and input the information that the subject should enter.

[0040] The communication interface 31 may not only receive measurement results from the measurement unit 10 but also wirelessly communicate with a server device or the like located in a different location via short-range wireless communication, wireless LAN, or base station communication. In this case, measurement data and analyzed and calculated features may be transmitted and received via the transmitting / receiving antenna 39 to and from the server device or the like. While short-range wireless communication may be performed using, for example, an electronic tag, this is not limiting. Wireless LANs such as Bluetooth (registered trademark), IrDA (Infrared Data Association, registered trademark), Zigbee (registered trademark), HomeRF (Home Radio Frequency, registered trademark), or Wi-Fi (registered trademark) may also be used, as long as they are capable of wireless communication when located near other information terminals. Long-range wireless communication, such as W-CDMA (Wideband Code Division Multiple Access) or GSM (Global System for Mobile communications, registered trademark), may be used for base station communication. It is also possible to detect the positional relationship and orientation between terminals using an ultra-wideband (UWB) wireless system. Although not shown, the communication interface 31 may use other methods such as optical communication or acoustic communication as a means of wireless communication. In that case, instead of the transmitting and receiving antenna 39, a light emitting / receiving unit and an acoustic wave output / input interface are used, respectively.

[0041] In this embodiment, the measurement unit 10 and the processor 30 have each of the aforementioned components individually, but they may also be provided with a functional unit that integrates at least some or all of these components.In short, any configuration is acceptable as long as the functions of each of these components are ensured.

[0042] The mild cognitive impairment screening device 1 configured as described above is non-invasive and has a compact design that is small, lightweight, and highly biosafe, allowing for measurement and evaluation in a short measurement time with minimal burden on the subject.

[0043] Next, with reference to the flowchart of FIG. 3, an example of the operation of the mild cognitive impairment screening device 1 configured as described above (mild cognitive impairment screening method) will be described in more detail.

[0044] FIG. 3 shows an example of processing steps (steps S1 to S8 of a mild cognitive impairment screening method) executed by the mild cognitive impairment screening device 1 (steps S2 and onward show an example of processing steps executed by the processor 30). As shown in the figure, in the mild cognitive impairment screening device 1 (mild cognitive impairment screening method) of this embodiment, first, the measurement unit 10 measures the finger tapping movement performed by the subject (measurement step S1). Specifically, the measurement unit 10 magnetically measures (detects) the finger tapping movement of only the right hand and / or the finger tapping movement of only the left hand and / or the finger tapping movement of both hands simultaneously and / or the finger tapping movement of both hands alternately using the tapping sensor 2. During this measurement process, the processor 30 acquires detection data (measurement data) from the tapping sensor 2 (step S2).

[0045] In this way, the subject's finger tapping movement is measured by the measurement unit 10, and the measurement data is received by the processor 30. Subsequently, the arithmetic circuit 33 of the processor 30 calculates the above-mentioned feature amounts based on the received measurement data (calculation step S3). Then, once such feature amounts have been calculated by the arithmetic circuit 33, the judgment circuit 34 of the processor 30 then makes a judgment regarding MCI based on a predetermined threshold value for the calculated feature amounts.

[0046] Specifically, in this embodiment, the judgment circuit 34 generates a discriminant model by machine learning including artificial intelligence (model generation step S4), and performs a judgment regarding mild cognitive impairment using the generated discriminant model (judgment step S5). Specifically, a previously prepared healthy group and MCI group are input into a machine learning algorithm capable of distinguishing between two groups, which will be described later, to generate a discriminant model. Then, a judgment result is output when new subject data is input. At this time, if the F1 score, which is an evaluation index in the discriminant model (machine learning model), is not 0.7 or more (YES in step S6), an adjustment step S7 is performed to adjust the optimal values ​​of hyperparameters in the machine learning so that the F1 score becomes 0.7 or more.

[0047] When an MCI assessment is performed with high accuracy, such that the F1 score is 0.7 or more, in this manner, the assessment result is displayed on the display 37 (step S8). In addition, by inputting a predetermined command, for example, via the operation input interface 38, data corresponding to the command, such as the calculated data calculated by the arithmetic circuit 33 (or the measured data output from the measurement unit 10), is also displayed on the display 37.

[0048] FIG. 5 is a table showing examples of machine learning algorithms (machine learning models; discrimination models) for discriminating between a healthy group and an MCI group. Examples of machine learning models include, but are not limited to, decision trees, random forests, XGBoost, LightGBM, and support vector machines (SVMs). FIG. 5 also shows hyperparameters and their search ranges for each machine learning model. The accuracy of discriminating between a healthy group and an MCI group varies by changing the specific hyperparameters of each learning model. To find optimal values ​​for the hyperparameters, a machine learning model may be determined based on the relationship between input feature data and the hyperparameters. Regarding machine learning evaluation, nested cross validation (nCV), which performs cross validation in a double-nested structure, may be used to simultaneously adjust hyperparameters and compare accuracy.

[0049] The discrimination accuracy between the MCI group and the healthy control group was calculated for the 248 features described above, and the features with the top 5% F1 values ​​were selected and shown in Figure 6 . The feature numbers (Char. No.) in the figure correspond to the item numbers of the 44 features described above. As shown in the figure, of the four tasks (i.e., right-hand only finger tapping, left-hand only finger tapping, both-hand simultaneous finger tapping, and both-hand alternate finger tapping), only the feature for both-hand alternate tapping (related to the left hand in the figure) was selected, while the other three tasks were not selected. Furthermore, among the features, item 29 relates to the duration of two fingers together, item 31 to the number of tremors, and items 34, 36, and 40 to the rhythm of tapping, while the feature related to the amplitude of movement was not selected. In this embodiment, feature items with F1 values ​​exceeding 0.5 were selected and used to determine mild cognitive impairment. These are the selected features described above. In Figure 6, Recall (precision rate) is the proportion of those predicted as positive (patients) that were actually correctly predicted, Precision (recall rate) is the proportion of those actually predicted as positive (also called sensitivity) among all positives, and F1 is the harmonic mean of Recall and Precision.

[0050] Figure 7 shows the results of hyperparameter tuning and classification accuracy evaluation using nCV for the five machine learning models already shown in Figure 5. The results show that Random Forest has the highest classification accuracy (F1 value). Figure 7 also shows the optimal values ​​of the hyperparameters, along with the hyperparameters, F1 value, precision, and recall.

[0051] As described above, according to the configuration of this embodiment, a huge amount of data, which is a wide variety of feature quantities, is processed by machine learning including artificial intelligence, so that MCI screening can be performed easily and quickly with high accuracy without requiring specialized knowledge. Moreover, the finger tapping measurement itself can be performed non-invasively, easily, and using a simple and compact device, so measurement and evaluation can be performed in a short measurement time with little burden on the subject.

[0052] Although the embodiments of the present invention have been described above with reference to the drawings, the present invention is not limited to the above-described embodiments and can include various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0053] Furthermore, the above-described configurations, functions, processing units (processors), processing means, etc. may be partially or entirely implemented in hardware, for example, by designing them as integrated circuits. The above-described configurations, functions, etc. may also be implemented in software, with a processor interpreting and executing a program that implements each function. Information such as the programs, tables, and files that implement each function may be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD, or may be stored in a device on a communication network.

[0054] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the product. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0055] 2 Tapping Sensor 10. Measurement section 30 processors 33 Arithmetic circuit 34 Judgment circuit 40 Adjustment circuit

Claims

1. a measuring step of measuring finger tapping movement, which is an opening and closing movement of two fingers; a calculation step of calculating a predetermined feature amount based on measurement data obtained by the measurement step within a predetermined time when finger tapping is performed with only the right hand and / or only the left hand and / or both hands simultaneously and / or both hands alternately; a determination step of determining mild cognitive impairment based on the feature amount calculated in the calculation step; Including, A mild cognitive impairment screening method characterized in that the judgment step generates a discriminant model using machine learning including artificial intelligence, and makes a judgment regarding mild cognitive impairment using the generated discriminant model.

2. The mild cognitive impairment screening method according to claim 1, further comprising an adjustment step of adjusting optimal values ​​of hyperparameters in machine learning so that the F1 score, which is an evaluation index in the machine learning model, is 0.7 or higher.

3. The mild cognitive impairment screening method according to claim 1, characterized in that the machine learning algorithm is a random forest.

4. 2. The mild cognitive impairment screening method according to claim 1, wherein the feature amount includes at least one of a standard deviation of contact time between two fingers during tapping, a difference in intersection point with zero finger acceleration during tapping, an average tapping interval which is the time interval between each tap, a standard deviation of tapping intervals, and a standard deviation of local tapping intervals.

5. A computer program for processing measurement data of finger tapping movement, which is an opening and closing movement of two fingers, a calculation step of calculating a predetermined feature amount based on measurement data obtained by the measurement step within a predetermined time when finger tapping is performed with only the right hand and / or only the left hand and / or both hands simultaneously and / or both hands alternately; a determination step of determining mild cognitive impairment based on the feature amount calculated in the calculation step; on the computer, A computer program characterized in that the judgment step generates a discriminant model using machine learning including artificial intelligence, and makes a judgment regarding mild cognitive impairment using the generated discriminant model.

6. 6. The computer program according to claim 5, further comprising an adjustment step of adjusting optimal values ​​of hyperparameters in machine learning so that an F1 score, which is an evaluation index in a machine learning model, is 0.7 or higher.

7. 6. The computer program of claim 5, wherein the machine learning algorithm is a random forest.

8. 6. The computer program according to claim 5, wherein the feature amount includes at least one of a standard deviation of contact time between two fingers during tapping, an intersection difference between finger acceleration and zero during tapping, an average tapping interval which is the time interval between each tap, a standard deviation of tapping intervals, and a local standard deviation of tapping intervals.

9. The device includes a measuring unit that measures finger tapping movement, which is the opening and closing movement of two fingers, and a processor that processes measurement data obtained by the measurement unit, The processor: an arithmetic circuit that calculates a predetermined feature amount based on measurement data obtained by the measuring step within a predetermined time when finger tapping is performed with only the right hand and / or only the left hand and / or both hands simultaneously and / or both hands alternately; a determination circuit that determines whether the subject has mild cognitive impairment based on the feature amount calculated in the calculation step; and The mild cognitive impairment screening device is characterized in that the judgment circuit generates a discriminant model through machine learning including artificial intelligence, and makes a judgment regarding mild cognitive impairment using the generated discriminant model.

10. The mild cognitive impairment screening device according to claim 9, further comprising an adjustment circuit that adjusts the optimal values ​​of hyperparameters in machine learning so that the F1 score, which is an evaluation index in the machine learning model, is 0.7 or higher.

11. The mild cognitive impairment screening device according to claim 8, characterized in that the machine learning algorithm is a random forest.

12. 10. The mild cognitive impairment screening device of claim 9, wherein the feature amount includes at least one of a standard deviation of contact time between two fingers during tapping, a difference in intersection point with zero finger acceleration during tapping, an average tapping interval which is the time interval between each tap, a standard deviation of tapping intervals, and a standard deviation of local tapping intervals.

Images