JP2022061146A - Psychoneurotic activity estimation device - Google Patents

Abstract

To provide a psychoneurotic activity estimation device that can estimate affection possibility and severity of a mental disorder to a patient or a subject in less computational complexity with high-speed and high accuracy.SOLUTION: A psychoneurotic activity estimation device calculates three scalar values of a pupil diameter average value, a pupil diameter causal relation and a pupil diameter complexity, from a pupil diameter data train of a subject or a patient, and inputs the values to a classifier or a regression device based on a learning algorithm. Thereby, the psychoneurotic activity estimation device can estimate the mental disorder affection possibility or the severity of the mental disorder of the subject or the patient extremely objectively.SELECTED DRAWING: Figure 3

Description

The present invention mainly relates to a neuropsychiatric activity estimation device capable of estimating mental illness with high accuracy and objectively.

At the time of filing the patent application, there are no biochemical or genetic biomarkers (Biomarkers: objectively indicating the existence or progression of a certain disease) of psychiatric disorders.
If I force you to mention it, it is a seizure wave on the brain wave in epilepsy, low-γ entropy, and tau-PET in dementia, but these are rather neurological diseases.

Physiologically, the relationship between each component of event-related potentials and higher-order functions has been elucidated to some extent, but none has been established as a biomarker for psychiatric disorders at the time of filing the patent application.
That is, until now, as a means for diagnosing a psychiatric disorder, there has been no completely mechanical and objective diagnostic method or device that does not involve the subjectivity of a psychiatrist at all. That is, whether a patient or subject has a given mental illness and how advanced the mental illness is depends on the observations of the psychiatrist performing the examination. It cannot be denied that human subjectivity is mixed.

The prior art documents of the present invention are shown in Patent Documents 1 to 3 and Non-Patent Documents 1 to 5.
Patent Document 1 discloses a technique relating to a method for diagnosing ADHD using a machine learning method based on past diagnostic data of developmental disorders.
Patent Document 2 discloses a technique relating to a method for diagnosing a psychiatric disorder based on a biological index based on eye movement.
Non-Patent Document 1 discloses a technique using a time series of pupil diameters for learning for a total of several tens to several hundred hours for learning a time series pattern of pupil diameters of ASD mice.
Non-Patent Document 2 discloses a research report on a time-series pattern of pupil diameters in the left and right eyes of ADHD patients.
Non-Patent Document 3 discloses a research report on the size of the pupil diameter of ADHD patients.
Non-Patent Document 4 discloses technical contents relating to sample entropy.
Non-Patent Document 5 discloses that in Alzheimer's disease, which is a kind of dementia, an abnormality appears in the locus coeruleus, which is the starting point of pupil diameter control, from the MCI stage before the onset.

Special Table 2018-533448 Gazette Japanese Patent Publication No. 2017-528991

Artoni, Pietro, et al. "Deep learning of spontaneous arousal fluctuations detects early cholinergic defects across neurodevelopmental mouse models and patients." Proceedings of the National Academy of Sciences (2019): 201820847, Internet <https://www.pnas.org / content / early / 2019/07/16/1820847116 ＞ William D. Poynter, "Pupil-size asymmetry is a physiologic trait related to gender, attentional function, and personality" Internet <https://doi.org/10.1080/1357650X.2016.1268147> G. Wainstein, D. Rojas-Libano, N. A. Crossley, X. Carrasco, F. Aboitiz & T. Ossandon, "Pupil Size Tracks Attentional Performance In Attention-Deficit / Hyperactivity Disorder" Internet <https://www.nature.com / articles / s41598-017-08246-w ＞ Joshua S. Richman, and J. Randall Moorman, "Physiological time-series analysis using approximate entropy and sample entropy" Internet <https://journals.physiology.org/doi/full/10.1152/ajpheart.2000.278.6.H2039> Kelly, S. C., He, B., Perez, S. E., Ginsberg, S. D., Mufson, E. J., & Counts, S. E. (2017). Locus coeruleus cellular and molecular pathology during the progression of Alzheimer's disease. Acta Neuropathologica Communications, 5 (1) , 1-14.

Non-Patent Document 1 discloses a completely mechanical and objective diagnostic method and a technique relating to the possibility of a diagnostic device as a means for diagnosing a psychiatric disorder without any intervention of the subjectivity of a psychiatrist. However, this technique requires an extremely heavy arithmetic process of inputting moving image data of the pupil into the neural network.

The present invention has been made in view of the above circumstances, and is a neuropsychiatric activity estimation device capable of estimating the susceptibility and severity of a psychiatric disorder or the like for a patient or a subject with a small amount of calculation and with high speed and high accuracy. The purpose is to provide.

In order to solve the above problem, the neuropsychiatric activity estimation device of the present invention calculates the average value of the pupil diameter from the left and right pupil diameter data obtained by continuously measuring the pupil diameter of the left and right eyes of the subject or the patient. The pupil diameter complexity calculation processing unit that calculates the regularity of the variation pattern of the pupil diameter from the average calculation processing unit and the left and right pupil diameter data, and the pupil diameter that calculates the causality of the left and right pupil diameters from the left and right pupil diameter data. Pupil diameter average value output by the causal calculation processing unit and the pupil diameter average calculation processing unit, pupil diameter causality output by the pupil diameter causal calculation processing unit, and pupil diameter output by the pupil diameter complexity calculation processing unit. It comprises a mental illness estimation processing unit that outputs the prevalence and / or severity of a subject or patient in multiple mental illnesses based on complexity.

According to the present invention, it is possible to provide a neuropsychiatric activity estimation device capable of estimating the susceptibility and severity of a psychiatric disorder or the like for a patient or a subject with a small amount of calculation and with high speed and high accuracy.
Issues, configurations and effects other than those described above will be clarified by the following description of the embodiments.

It is a schematic diagram which shows the whole structure of the mental illness estimation apparatus which concerns on embodiment of this invention. It is a block diagram which shows the hardware composition of the mental illness estimation apparatus. It is a block diagram which shows the whole software function in the estimation phase of the mental illness estimation apparatus which concerns on 1st Embodiment of this invention. It is a block diagram which shows the software function in the learning phase of the mental illness estimation device. It is a block diagram which shows the whole software function in the estimation phase of the mental illness estimation apparatus which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows the whole software function in the learning phase of the mental illness estimation apparatus which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows the detail of a feature vector calculation processing part. It is a block diagram which shows the detail of the sample entropy calculation processing part. It is a block diagram which shows the detail of the left-right movement entropy calculation processing part. It is a block diagram which shows the detail of the right-left movement entropy calculation processing part. It is a graph which shows the ROC curve (Receiver Operating Characteristic) of the mental illness estimation apparatus which concerns on 1st Embodiment of this invention.

[Summary of the invention: Objective estimation of psychiatric disorders based on research on the relationship between time-series changes in pupil diameter and psychiatric disorders]
It is known that many psychiatric disorders are expected to have multiple neurotransmitter abnormalities. For example, ADHD is caused by abnormalities of dopamine and noradrenaline, schizophrenia is caused by abnormalities of dopamine, glutamate and noradrenaline, and depression is caused by abnormalities of serotonin and noradrenaline.
For example, as illustrated in Non-Patent Documents 1, 2 and 3, several papers have been published so far that a significant association is found between mental illness and time-series changes in pupil diameter. That is, the argument is that the above-mentioned abnormal function of the neurotransmitter appears in the time-series change of the pupil diameter.

In the field of psychiatry, it is difficult to obtain a large number of data obtained from a subject or a patient. Therefore, the estimation method using a neural network as in Non-Patent Document 1 is not realistic.
Therefore, based on several past papers, the inventors calculated the scalar values of the pupil diameter over time using three indexes, and estimated from these three scalar values based on the learning algorithm, which was extremely good. I succeeded in getting the result.

The first scalar value is the average value of the left and right pupil diameters.
The second scalar value is the complexity of the left and right pupil diameters. The complexity of the pupil diameter is to calculate the regularity of the fluctuation pattern of the pupil diameter stochastically or deterministically.
The third scalar value is the causality of the left and right pupil diameters. The causal nature of the pupil diameter is how much the variation pattern of the pupil diameter of the right eye depends on the variation pattern of the pupil diameter of the left eye, or how much the variation pattern of the pupil diameter of the left eye depends on the variation pattern of the pupil diameter of the right eye. The degree of dependence of the left and right pupil diameters, that is, causality, whether or not it depends on the fluctuation pattern of the above, is calculated probabilistically or deterministically.

[Mental illness estimation device 101: Overall configuration]
FIG. 1 is a schematic view showing the overall configuration of the mental illness estimation device 101 according to the embodiment of the present invention.
The subject or patient 102 receives a pupil diameter measurement with an eye movement measuring device 103 (for example, TX300 manufactured by Tobi).
The eye movement measuring device 103 measures, for example, the pupil diameters of the left and right eyes at a sampling frequency of 300 Hz, and outputs the left and right pupil diameter data.
In order to accurately photograph the pupil diameter of the subject or the patient 102, the head of the subject or the patient 102 is fixed by the chin rest 104 when acquiring the left and right pupil diameter data.

In order to acquire the left and right pupil diameter data in a relaxed state as much as possible for the subject or the patient 102, the environment for acquiring the left and right pupil diameter data is in a living room such as a general office where the subject or the patient 102 is not exposed to direct sunlight. good.
The left and right pupil diameter data may be either a file or a data stream, but it is more convenient to finally create a file.

The left and right pupil diameter data generated by the eye movement measuring device 103 is transmitted to the mental illness estimation device 101.
The mental illness estimation device 101 including a computer such as a personal computer reads the left and right pupil diameter data, performs predetermined arithmetic processing, and outputs the possibility of morbidity for a plurality of mental illnesses.

[Mental illness estimation device 101: hardware configuration]
FIG. 2 is a block diagram showing a hardware configuration of the mental illness estimation device 101.
The mental illness estimation device 101 includes a CPU 202, a ROM 203, a RAM 204, a display unit 205, an operation unit 206, a non-volatile storage 207, and a NIC (Network Interface Card) 208 connected to the bus 201.

In FIG. 2, the mental illness estimation device 101 receives the left and right pupil diameter data from the eye movement measuring device 103 via the network, but may also be via a serial port such as USB. Moreover, you may go through a well-known USB memory.
The non-volatile storage 207 stores an application program for operating the personal computer as the mental illness estimation device 101.

[First embodiment: Mental illness estimation device 101: Software function of estimation phase]
FIG. 3 is a block diagram showing the entire software function in the estimation phase of the mental illness estimation device 101 according to the first embodiment of the present invention.
The left and right pupil diameter data file 301 is input to the sample number regulation processing unit 302.
The sample number regulation processing unit 302 extracts a sufficient number of data samples necessary for estimating the mental illness from the left and right pupil diameter data stored in the left and right pupil diameter data file 301. At this time, the sample number defining processing unit 302 divides the left and right pupil diameter data into epoch, which is a predetermined number of data samples, and linearly interpolates the missing value section of the data caused by the subject's blinking or the like. When the ratio of missing values in the epoch exceeds a predetermined ratio, for example, 10%, the epoch is not used and is discarded. Then, collect the required number of epoches.

In the first embodiment of the present invention and the second embodiment described later, as an example, the sampling frequency of the eye movement measuring device 103 is 300 Hz, and the echo is set to 2 seconds, that is, 600 samples. The left and right pupil diameter data is 140 seconds, that is, 70 epoch and 42000 samples.

A noise component is removed from the left and right pupil diameter data output from the sample number regulation processing unit 302 by a low-pass filter (hereinafter, “LPF”) 303.
The cutoff frequency of the LPF 303 is preferably half or less of the sampling frequency of the eye movement measuring device 103.
In the first embodiment of the present invention and the second embodiment described later, the cutoff frequency of the LPF 303 is set to 50 Hz.

After the number of data samples is specified by the sample number specification processing unit 302, the left and right pupil diameter data from which the noise component is removed by the LPF 303 are the pupil diameter average calculation processing unit 304, the pupil diameter causal calculation processing unit 305, and the pupil diameter. Each is input to the complexity calculation processing unit 306.
The pupil diameter average calculation processing unit 304 reads the left and right pupil diameter data and outputs the average value of the left and right pupil diameters.
The pupil diameter causal calculation processing unit 305 reads the left and right pupil diameter data, and outputs the average value of the mutual transfer entropy in the left and right pupil diameters. That is, mobile entropy is one of the methods for calculating the causality of the pupil diameter.

The pupil diameter complexity calculation processing unit 306 reads the left and right pupil diameter data, and outputs the average value of the time-series sample entropy (Sample Entropy) in the left and right pupil diameters. That is, sample entropy is one of the methods for calculating the complexity of the pupil diameter.
The LPF 303, the pupil diameter average calculation processing unit 304, the pupil diameter causality calculation processing unit 305, and the pupil diameter complexity calculation processing unit 306 constitute a feature vector calculation processing unit 307.

The pupil diameter average value output by the pupil diameter average calculation processing unit 304, the pupil diameter causality output by the pupil diameter causality calculation processing unit 305, and the pupil diameter complexity output by the pupil diameter complexity calculation processing unit 306 are It is input to the classifier 308.
The classifier 308 uses these data input from the feature vector calculation processing unit 307 as a feature vector, executes an estimation calculation process based on a well-known learning algorithm, and outputs the morbidity rate in a plurality of mental illnesses. The classifier 308 includes an approximate function parameter 309, which will be described later in FIG. 4.

As the learning algorithm used for the classifier 308, these well-known learning algorithms such as SVM (Support Vector Machine), Naive Bayes, Gaussian process, decision tree (classification tree), random forest, Convolutional Neural Network, etc. can be applied. Is.
In the verification described later in FIG. 11, SVM was adopted. In the medical field where it is difficult to secure a large number of samples of training data, it seems preferable to adopt an SVM that is resistant to overfitting.

The morbidity rates of a plurality of types of mental illness output from the classifier 308 are input to the input / output control unit 310.
The input / output control unit 310 compares the morbidity rate of a plurality of mental illnesses with a predetermined threshold, and determines what kind of psychiatric disorder the subject or patient 102 is suffering from, or whether the subject or patient 102 is mentally healthy. Estimate and display the information on the display unit 205. At the same time, the input / output control unit 310 files a plurality of psychiatric morbidity rates in the non-volatile storage 207.

[First Embodiment: Mental Illness Estimator 101: Software Function in Learning Phase]
FIG. 4 is a block diagram showing the entire software function in the learning phase of the mental illness estimation device 101 according to the first embodiment of the present invention.
The mental illness estimation device 101 acquires a left and right pupil diameter data file 402 for learning, which is different from the left and right pupil diameter data file 301 for estimation, from a subject or a patient 102 that provides learning data to the learner 401.

In addition, a group of flag information indicating whether or not the subject or patient 102 has a predetermined mental illness, which is obtained in advance by a predetermined diagnosis, is prepared as teacher data 403. It is assumed that the left and right pupil diameter data files 402 and the teacher data 403 are associated with each other by the ID information that uniquely identifies the subject or the patient 102.

The feature vector calculation processing unit 307 performs the same calculation processing as in FIG. 3, and outputs the pupil diameter mean value, the pupil diameter causality, and the pupil diameter complexity from the left and right pupil diameter data files 402.
The learner 401 uses the pupil diameter average value, the pupil diameter causality, and the pupil diameter complexity input from the feature vector calculation processing unit 307 as feature vectors, and whether or not the subject or the patient 102 suffers from a predetermined mental illness. The learning process is performed using the group of flag information indicating the above as the teacher data 403. Then, the approximation function parameter 309 is generated and updated.
By combining the estimation calculator that implements the learning algorithm and the approximation function parameter 309, the classifier 308 of FIG. 3 is realized.

[Second embodiment: Mental illness estimation device 501: Software function of estimation phase]
FIG. 5 is a block diagram showing the entire software function in the estimation phase of the mental illness estimation device 501 according to the second embodiment of the present invention.
The difference between the mental illness estimation device 501 shown in FIG. 5 and the mental illness estimation device 101 shown in FIG. 3 is that the mental illness estimation device 501 of FIG. This is the point where 502 is used.

The feature vector calculation processing unit 307 performs the same calculation processing as in FIG. 3, and outputs the pupil diameter mean value, the pupil diameter causality, and the pupil diameter complexity from the left and right pupil diameter data files 402.
The pupil diameter average value output by the pupil diameter average calculation processing unit 304, the pupil diameter causality output by the pupil diameter causality calculation processing unit 305, and the pupil diameter complexity output by the pupil diameter complexity calculation processing unit 306 are It is input to the retriever 502.

The regressor 502 uses these data input from the feature vector calculation processing unit 307 as a feature vector, executes an estimation calculation process based on a well-known learning algorithm, and outputs estimated values of severity in a plurality of mental illnesses. The regressor 502 includes an approximate function parameter 503, which will be described later in FIG.

Similar to the classifier 308 in the first embodiment described above, the learning algorithm used for the regressionr 502 also includes an SVM (Support Vector Machine), a Gaussian process, a decision tree (regression tree), a random forest, and a convolutional neural. These well-known learning algorithms such as Network can be applied.

A plurality of types of mental illness severity estimates are input to the input / output control unit 310.
The input / output control unit 310 displays a plurality of mental illness severity estimates on the display unit 205. At the same time, the input / output control unit 310 files a plurality of mental illness severity estimates in the non-volatile storage 207.

[Second embodiment: Mental illness estimation device 501: Software function of learning phase]
FIG. 6 is a block diagram showing the entire software function in the learning phase of the mental illness estimation device 501 according to the second embodiment of the present invention.
The learner 601 uses the pupil diameter average value, the pupil diameter causality, and the pupil diameter complexity input from the feature vector calculation processing unit 307 as feature vectors, and is severe in a predetermined mental illness affecting the subject or the patient 102. The learning process is performed using the group of information as the teacher data 602. Then, the approximation function parameter 503 is generated and updated.
By combining the estimation calculator that implements the learning algorithm and the approximation function parameter 503, the regression device 502 of FIG. 5 is realized.

[Mental illness estimation device 101: Feature vector calculation processing unit 307]
The feature vector calculation processing unit 307 shown in the dotted frame including the LPF 303, the pupil diameter average calculation processing unit 304, the pupil diameter causality calculation processing unit 305, and the pupil diameter complexity calculation processing unit 306 according to FIGS. 3 to 6 is , The contents of processing are common. Hereinafter, the details of the feature vector calculation processing unit 307 will be described.

FIG. 7 is a block diagram showing details of the feature vector calculation processing unit 307.
The right pupil diameter data 701 and the left pupil diameter data 702 from which the noise component is removed by the LPF 303 after the number of data samples is specified by the sample number specifying processing unit 302 are subdivided into the first average value calculation processing unit 703. It is supplied to the chemical processing unit 704.
The first average value calculation processing unit 703 calculates and outputs the pupil diameter average value, which is the average value of the entire left and right pupil diameter data. The first average value calculation processing unit 703 is an entity of the pupil diameter average calculation processing unit 304.

The data subdivision processing unit 704 outputs the right echo 705 from the right pupil diameter data 701 and the left echo 706 from the left pupil diameter data 702 based on the control of the input / output control unit 310.
The input / output control unit 310 manages various loop processing and sequence control, that is, management of the order of execution for each functional block.

The right echo 705 and the left echo 706 are input to the sample entropy calculation processing unit 708 via the changeover switch 707.
The sample entropy value output by the sample entropy calculation processing unit 708 is input to the second average value calculation processing unit 709.

The second average value calculation processing unit 709 calculates the average of the sample entropy values of all epoches and outputs the pupil diameter complexity.
The changeover switch 707, the sample entropy calculation processing unit 708, and the second mean value calculation processing unit 709 are the entities of the pupil diameter complexity calculation processing unit 306.

The right echo 705 and the left echo 706 are input to the left / right (left to right) moving entropy calculation processing unit 710 and the right / left (right to left) moving entropy calculation processing unit 711, respectively.
The left-right movement entropy calculation processing unit 710 calculates the left-right pupil diameter causality and outputs it to the third average value calculation processing unit 712.
The right-left movement entropy calculation processing unit 711 calculates the right-left pupil diameter causality and outputs it to the third average value calculation processing unit 712.

The third average value calculation processing unit 712 calculates and outputs the pupil diameter causality as the average of the left and right pupil diameter causality and the right and left pupil diameter causality of all epoches.
The left-right movement entropy calculation processing unit 710, the right-left movement entropy calculation processing unit 711, and the third average value calculation processing unit 712 are the entities of the pupil diameter causal calculation processing unit 305.

The human eye has a "dominant eye" similar to the limbs. Therefore, the above-mentioned average pupil diameter value, pupil diameter causality, and pupil diameter complexity are biased in the values output for each of the left and right eyes. If the value in which the left-right bias appears is input to the estimation calculation process of the learning algorithm as it is, the bias of the dominant eye of the person who provides the learning data is reflected in the estimation result. In order to cancel this bias, the feature vector calculation processing unit 307 calculates the average value of the values output for each of the left and right eyes. Therefore, the feature vector calculation processing unit 307 is provided with a first average value calculation processing unit 703, a second average value calculation processing unit 709, and a third average value calculation processing unit 712.

[Mental illness estimation device 101: sample entropy calculation processing unit 708]
FIG. 8 is a block diagram showing details of the sample entropy calculation processing unit 708.
The sample entropy is calculated by the following formula. The sample entropy calculation processing unit 708 executes the calculation of the following equation (1) disclosed in Non-Patent Document 4.

The right echo 705 or the left echo 706 output via the changeover switch 707 is input to the Z-scoring processing unit 801.
The Z scoring processing unit 801 performs a well-known Z scoring process (Z score) on the input right echo 705 or left echo 706. Specifically, the value obtained by subtracting the average from the input variable is divided by the standard deviation of the input variable. By this arithmetic processing, the input epoch is normalized to have an average value of 0 and a standard deviation of 1, and becomes a normalized epoch 802.
The normalized epoch 802 is input to the m-dimensional vector generation processing unit 803 and the m + 1-dimensional vector generation processing unit 804, respectively.

The m-dimensional vector generation processing unit 803 creates two m-dimensional vectors x ^m _i and x ^m _j , which are composed of m adjacent elements in the time series in the normalized epoch 802.
From this point onward, when the term x ^m _i is used, the superscript indicates the dimension of the vector, and the subscript indicates the vector number. x ^m _i means "the i-th vector of m dimension in x". Also, when written as x _i , the subscript shall indicate the number of the scalar value. x _i means "the i-th scalar value in x".

Now, when one of the vectors generated by the m-dimensional vector generation processing unit 803 x ^m _i = {x _i , x _{i + 1} ,…, x _{i + m-1} }, the other vector x ^m _j = {x _j , x _{j + 1} ,…, x _{j + m-1} }. The following conditions are imposed on i and j.
(i! = j, i = 1, m + 1, m + 2, m + 3,… Nm, j = m + 1, m + 2, m + 3,…, N-m + 1)

The two vectors output from the m-dimensional vector generation processing unit 803 are input to the norm calculation processing unit 805a.
The norm calculation processing unit 805a outputs the absolute value of the norm of the composite vector obtained by subtracting the vectors x ^m _i and x ^m _j .
The absolute value of the norm output by the norm calculation processing unit 805a is compared with the threshold value r807 by the comparator 806a, and if it is less than the threshold value r807, the comparator 806a outputs the "true" of the logic.
The probability calculation processing unit 808a counts the logical output of the comparator 806a for each set of vectors output by the norm calculation processing unit 805a, and outputs the probability C _m (r) that the absolute value of the norm is less than the threshold value r807.

The arithmetic processing of the m-dimensional vector generation processing unit 803 will be described with reference to an actual example.
In the embodiment of the present invention, the echo is 2 seconds and the sampling frequency is 300 Hz, so that the number of samples N is 600. m is 2.
The data string X on the right echo705 is X = {x ₁ , x ₂ ,…, x _t ,…, x _N } = {x ₁ , x ₂ ,…, x _t ,…, x ₆₀₀ }.
The data string Y on the left echo 706 is Y = {y ₁ , y ₂ ,…, y _t ,…, y _N } = {y ₁ , y ₂ ,…, y _t ,…, y ₆₀₀ }
Since the vector x ^m _i = {x _i , x _{i + 1} }, i = 1, m + 1, m + 2, m + 3… Nm, i = 1,3,4,… 598. The vector x ^m _i is x ² ₁ = {x ₁ , x ₂ }, x ² ₃ = {x ₃ , x ₄ }, x ² ₄ = {x ₄ , x ₅ },…, x ² ₅₉₈ = {x It becomes ₅₉₈ , x ₅₉₉ }.

On the other hand, the vector x ^m _j = {x _j , x _{j + 1} }, j = m + 1, m + 2, m + 3,… N-m + 1, so j = 3,4,5 , ... 599. Therefore, the vector x ^m _j is x ² ₃ = {x ₃ , x ₄ }, x ² ₄ = {x ₄ , x ₅ }, x ² ₅ = {x ₅ , x ₆ },…, x ² ₅₉₉ = It becomes {x ₅₉₉ , x ₆₀₀ }.
First, the m-dimensional vector generation processing unit 803 sets j = 3,4,5,… 599 for the vector x ² ₁ = {x ₁ , x ₂ } of i = 1, and the vector x ² ₃ = {x. ₃ , x ₄ }, x ² ₄ = {x ₄ , x ₅ }, x ² ₅ = {x ₅ , x ₆ },…, x ² ₅₉₉ = {x ₅₉₉ , x ₆₀₀ }, Norm arithmetic processing unit 805a Output to. The norm calculation processing unit 805a is a composite vector obtained by subtracting the vector x ² ₃ = {x ₃ , x ₄ } from the vector x ² ₁ = {x ₁ , x ₂ } input from the m-dimensional vector generation processing unit 803. Calculate the absolute value of the norm.

The m + 1 dimensional vector generation processing unit 804 creates two (m + 1) dimensional vectors composed of (m + 1) adjacent elements in the time series in the normalized epoch 802.
Hereinafter, the norm calculation processing unit 805b performs the same calculation processing as the norm calculation processing unit 805a. The comparator 806b makes a comparison with the threshold value r807, which is the same as the comparator 806a, and outputs a logical value.
The probability calculation processing unit 808b counts the logical output of the comparator 806b for each set of vectors output by the norm calculation processing unit 805b, and outputs the probability C _{m + 1} (r) that the absolute value of the norm is less than the threshold value r807. do.

The divider 809 divides the probability C _{m + 1} (r) as the divisor and the probability C _m (r) as the divisor.
The logarithm calculation processing unit 810 calculates and outputs the natural logarithm of the output of the divider 809.
The multiplier 811 inverts the sign of the output value of the logarithmic arithmetic processing unit 810 by multiplying the output value of the logarithmic arithmetic processing unit 810 by the negative value 812 which is "-1", and the sample entropy h (r, m). Is output.

[Mental illness estimation device 101: left-right movement entropy calculation processing unit 710]
FIG. 9 is a block diagram showing details of the left-right movement entropy calculation processing unit 710.
The mobile entropy is calculated by the following formula. The left-right movement entropy calculation processing unit 710 and the right-left movement entropy calculation processing unit 711 described with reference to FIG. 10 below execute the calculation of the following equation (2).

P (y _t | y _t-τ ^dy , x _t-τ ^dx ) and p (y _t | y _t-τ ^dy ) in the equation (2) are conditional probabilities, respectively, and the following equations (3) and It is calculated by the formula (4).

The right echo 705 and the left echo 706 are input to the bin width adjusting processing unit 901.
The bin width adjustment processing unit 901 rounds the value input by dividing the range width of the variables of the right echo 705 and the left echo 706 by a predetermined number.
Thus, the right epoch 705 and the left epoch 706 become the normalized right epoch 902 and the normalized left epoch 903.

The normalized right echo 902 and the normalized left echo 903 are input to the delay vector generation processing units 904a and 904b, respectively.
The delay vector generation processing units 904a and 904b have a right delay vector 906 (x _t-τ ^dx ) and a left delay vector 907 (y _t-τ ^dy ) according to a delay vector setting value 905 composed of a sampling interval τ, a number of dimensions dx and dy. ) Is generated.

The right delay vector 906 (x _t-τ ^dx ) and the left delay vector 907 (y _t-τ ^dy ) are input to the joint probability calculation processing unit 908a.
The simultaneous probability calculation processing unit 908a has a right delay vector 906 (x _t-τ ^dx ) and a left delay vector 907 (y _t-τ ) among all the delay vector sets included in the normalized right echo 902 and the normalized left echo 903. Calculate the probability p (y _t-τ ^dy , x _t-τ ^dx ) that a set of the same values as ^dy ) exists.

The arithmetic processing of the joint probability arithmetic processing unit 908a will be described with reference to an actual example.
In the embodiment of the present invention, the echo is 2 seconds and the sampling frequency is 300 Hz, so that the number of samples N is 600. τ is 10, dx and dy are 5.
The data string X on the right echo705 is X = {x ₁ , x ₂ ,…, x _t ,…, x _N } = {x ₁ , x ₂ ,…, x _t ,…, x ₆₀₀ }.
The data string Y on the left echo 706 is Y = {y ₁ , y ₂ ,…, y _t ,…, y _N } = {y ₁ , y ₂ ,…, y _t ,…, y ₆₀₀ }
Since the delay vector x _t ^dx is x _t ^dx = {x _t , x _t-τ , x _{t-2 τ} ,…, x _{t- (dx-1) τ} }, so x ₆₀₀ ⁵ = {x ₆₀₀ , x ₅₉₀ , x ₅₈₀ , x ₅₇₀ , x ₅₆₀ }, x ₅₉₉ ⁵ = {x ₅₉₉ , x ₅₈₉ , x ₅₇₉ , x ₅₆₉ , x ₅₅₉ },…, x ₄₁ ⁵ = {x ₄₁ , x ₃₁ , x ₂₁ , x ₁₁ , x It becomes ₁ }.

As described above, the delay vector x _t ⁵ extracts elements from x _t , which is the final element of the data string X, at every sampling interval τ = 10, and creates a vector having dx = 5 dimensions. The subscript " ₄₁ " of the last vector x 4 1 ⁵ is obtained by back-calculating t from t- (dx-1) τ = 1.
The simultaneous probability p (y _t-τ ^dy , x _t-τ ^dx ) is obtained by changing the time t from 600 to 41 + 10 = 51 (the number of subscripts in the last vector plus the sampling interval τ). It refers to the probability that a set of delay vectors having the same value as the delay vector at a desired time t exists in the group of delay vectors to be obtained. For example, when t = 600, the set of delay vectors is
{y ₅₉₀ ⁵ , x ₅₉₀ ⁵ } = {y ₅₉₀ , y ₅₈₀ , y ₅₇₀ , y ₅₆₀ , y ₅₅₀ , x ₅₉₀ , x ₅₈₀ , x ₅₇₀ , x ₅₆₀ , x ₅₅₀ }
Is. Similarly, when t = 51, the set of delay vectors is
{y ₄₁ ⁵ , x ₄₁ ⁵ } = {y ₄₁ , y ₃₁ , y ₂₁ , y ₁₁ , y ₁ , x ₄₁ , x ₃₁ , x ₂₁ , x ₁₁ , x ₁ }
Is.

The joint probability calculation processing unit 908a calculates the probability that a set of delay vectors having the same value as this set of delay vectors exists. For example, assuming that the data column X and the data column Y take any of the six values of 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, the set of delay vectors is
{y ₅₉₀ ⁵ , x ₅₉₀ ⁵ } = {0.3,0.4,0.5,0.4,0.3,0.2,0.3,0.4,0.3,0.2}
If the set of values is, the joint probability p (y ₅₉₀ ⁵ , x ₅₉₀ ⁵ ) is the set of delay vectors consisting of the same set of values as {y ₅₉₀ ⁵ , x ₅₉₀ ⁵ } among all the sets of delay vectors. It is obtained by counting the number of pairs and dividing that number by the number of pairs of all delay vectors.

The right delay vector 906 (x _t-τ ^dx ), the left delay vector 907 (y _t-τ ^dy ), and the scalar value y _t taken out from the normalized left echo 903 are input to the addition joint probability calculation processing unit 909a.
The additional simultaneous probability calculation processing unit 909a has a right delay vector 906 (x _t-τ ^dx ) and a left delay vector 907 among all the delay vector sets included in the normalized right echo 902, the normalized left echo 903, and y _t . Calculate the probability p (y _t ∩ y t _-τ ^dy , x _t-τ ^dx ) that a set of the same values as (y _t-τ ^dy ) and y _t exists.

The arithmetic processing of the additional joint probability arithmetic processing unit 909a will be described with reference to an actual example. Similar to the joint probability calculation processing unit 908a described above, N = 600, τ = 10, and dx = dy = 5.
The additional simultaneous probability p (y _t ∩ y t _-τ ^dy , x _t-τ ^dx ) is the addition of the scalar value y _t to the group of delay vectors obtained by changing the time t from 600 to 41 + 10 = 51. It refers to the probability that a set of delay vector and scalar value y _t having the same value as the delay vector and scalar value y _t at a desired time t exists in the group. For example, when t = 600, the set of delay vector and scalar value y _t is
{y ₆₀₀ , y ₅₉₀ ⁵ , x ₅₉₀ ⁵ } = {y ₆₀₀ , y ₅₉₀ , y ₅₈₀ , y ₅₇₀ , y ₅₆₀ , y ₅₅₀ , x ₅₉₀ , x ₅₈₀ , x ₅₇₀ , x ₅₆₀ , x ₅₅₀ }
Is. Similarly, when t = 51, the set of delay vector and scalar value y _t is
{y ₅₁ , y ₄₁ ⁵ , x ₄₁ ⁵ } = {y ₅₁ , y ₄₁ , y ₃₁ , y ₂₁ , y ₁₁ , y ₁ , x ₄₁ , x ₃₁ , x ₂₁ , x ₁₁ , x ₁ }
Is. The additional joint probability calculation processing unit 909a calculates the probability that the set of the delay vector and the set of the delay vector and the scalar value y _t having the same value as the scalar value y _t exist.

The first divider 910 uses the probability p (y _t ∩ y t _-τ ^dy , x _t-τ ^dx ) output by the additional simultaneous probability calculation processing unit 909 as a divisor, and the probability p output by the simultaneous probability calculation processing unit 908a. Divide by using (y _t-τ ^dy , x _t-τ ^dx ) as a divisor. The output data of the first divider 910 is a conditional probability p (y _t | y _t-τ ^dy , x _t-τ ^dx ).

The left delay vector 907 (y _t-τ ^dy ) is input to the joint probability calculation processing unit 908b.
The joint probability calculation processing unit 908b has a probability p (y _t-τ ) that a set having the same value as the left delay vector 907 (y _t-τ ^dy ) exists among all the sets of delay vectors included in the normalized left echo 903. ^dy ) is calculated.

Similar to the above, to give an example of the joint probability calculation processing unit 908b, the joint probability p (y _t-τ ^dy ) is a group of delay vectors obtained by changing the time t from 600 to 41 + 10 = 51. Of these, it refers to the probability that a set of delay vectors having the same value as the delay vector at a desired time t exists. For example, when t = 600, the set of delay vectors is
{y ₅₉₀ ⁵ } = {y ₅₉₀ , y ₅₈₀ , y ₅₇₀ , y ₅₆₀ , y ₅₅₀ }
Is. The joint probability calculation processing unit 908b calculates the probability that a set of delay vectors having the same value as this set of delay vectors exists.

The left delay vector 907 (y _t-τ ^dy ) and the scalar value y _t taken out from the normalized left echo 903 are input to the addition joint probability calculation processing unit 909b.
The additional simultaneous probability calculation processing unit 909b includes the left delay vector 907 (y _t-τ ^dy ) and the scalar value in the set of all the delay vectors included in the normalized right echo 902, the normalized left echo 903, and the scalar value y _t . Calculate the probability p (y _t ∩ y _t-τ ^dy ) that a set of the same values as y _t exists.

Similar to the above, to give an example of the additional simultaneous probability calculation processing unit 909b, the additional simultaneous probability p (y _t ∩ y _t-τ ^dy ) can be obtained by changing the time t from 600 to 41 + 10 = 51. It refers to the probability that a set of a delay vector having the same value as the delay vector and yt exists at a desired time t among the group of delay vectors and yt. For example, when t = 600, the set of delay vector and yt is
{y ₆₀₀ , y ₅₉₀ ⁵ } = {y ₆₀₀ , y ₅₉₀ , y ₅₈₀ , y ₅₇₀ , y ₅₆₀ , y ₅₅₀ }
Is. The additional joint probability calculation processing unit 909b calculates the probability that this set of delay vectors and the set of delay vectors and yt having the same value as yt exist.

The second divider 911 divides with the probability p (y _t ∩ y t _- τ ^dy ) as the divisor and the probability p (y _t-τ ^dy ) as the divisor. The output of the second divider 911 is a conditional probability p (y _t | y _t-τ ^dy ).

The third divider 912 divides the conditional probability p (y _t | y _t-τ ^dy , x _t-τ ^dx ) as the divisor and the conditional probability p (y _t | y _t-τ ^dy ) as the divisor. .. The output of the third divider 912 is input to the logarithmic operation processing unit 913.
The logarithm calculation processing unit 913 calculates the natural logarithm of the value input from the third divider 912.
The output data of the logarithmic operation processing unit 913 is multiplied by the conditional probability p (y _t | y _t-τ ^dy , x _t-τ ^dx ) output by the first divider 910 by the multiplier 914.
The evaluation value output by the multiplier 914 is input to the total sum calculation processing unit 915 under the control of the input / output control unit 310, the sum of the evaluation values in all the delay vectors is calculated, and the left-right movement entropy is output.

[Mental illness estimation device 101: Right-left moving entropy calculation processing unit 711]
FIG. 10 is a block diagram showing details of the right-left moving entropy calculation processing unit 711.
All functional blocks are the same as in FIG. 9, except that the arrangements of the normalized right echo 902 and the normalized left echo 903 are exchanged, and the arrangements of the right delay vector 906 and the left delay vector 907 are also exchanged, so the details are omitted. ..
The evaluation value output by the multiplier 914 is input to the total sum calculation processing unit 915 under the control of the input / output control unit 310, the sum of the evaluation values in all the delay vectors is calculated, and the right-left movement entropy is output.

[Mental illness estimation device 101: Performance by simulation calculation]
FIG. 11 is a graph showing a ROC curve (Receiver Operating Characteristic) of the mental illness estimation device 101 according to the first embodiment of the present invention.
In FIG. 11, "Baseline" is the result of only the average value, "TranEn" is the result of only the mobile entropy, "SampEn" is the result of only the sample entropy, and "All" is the average value, the mobile entropy, and the sample entropy. The result using is shown.
Since the AUC (Area Under the Curve) outputs a result of 0.9, which is extremely close to 1, the mental illness estimation device 101 according to the first embodiment of the present invention has high estimation accuracy for mental illness. It can be said that it is prepared.

The psychiatric disorders that can be estimated by the psychiatric disorder estimator according to the first embodiment and the second embodiment of the present invention include ADHD, ASD, schizophrenia, depression, and dementia. There seems to be one. In particular, in Alzheimer's disease, which is a type of dementia, it has been reported in Non-Patent Document 5 that abnormalities appear in the locus coeruleus, which is the starting point of pupil diameter control, from the MCI stage before the onset. It is possible that the pupillary reaction can be used as an indicator of such early locus coeruleus dysfunction.

In the embodiment described above, the following modification can be adopted.
(1) The m-dimensional vector generation processing unit 803 and the m + 1-dimensional vector generation processing unit 804 described with reference to FIG. 8 can also perform the processing procedure shown below.
The m-dimensional vector generation processing unit 803 creates two m-dimensional vectors composed of m adjacent elements in the time series in the normalized epoch 802.
Now, when x ^m _i = {x _i , x _{i + 1} ,…, x _{i + m-1} }, then x ^m _j = {x _j , x _{j + 1} ,… x _{j + m-1} } be. The following conditions are imposed on i and j.
(i! = j, i = 1,2,3,…, Nm-1, j = i + 1, i + 2, i + 3,…, Nm)

The two vectors output from the m-dimensional vector generation processing unit 803 are input to the norm calculation processing unit 805a.
The norm calculation processing unit 805a outputs the absolute value of the norm of the composite vector obtained by subtracting the vectors x ^m _i and x ^m _j .
The absolute value of the norm is compared to the threshold value r807 by the comparator 806a, and if it is less than the threshold value r807, the comparator 806a outputs the "true" of logic.
The probability calculation processing unit 808a counts the logical output of the comparator 806a for each set of vectors output by the norm calculation processing unit 805a, and outputs the probability Cm (r) that the absolute value of the norm is less than the threshold value r807.

The arithmetic processing of the m-dimensional vector generation processing unit 803 will be described with reference to an actual example.
In the embodiment of the present invention, the echo is 2 seconds and the sampling frequency is 300 Hz, so that the number of samples N is 600. m is 2.
The data string X on the right echo705 is X = {x ₁ , x ₂ ,…, x _t ,…, x _N } = {x ₁ , x ₂ ,…, x _t ,…, x ₆₀₀ }.
The data string Y on the left echo 706 is Y = {y ₁ , y ₂ ,…, y _t ,…, y _N } = {y ₁ , y ₂ ,…, y _t ,…, y ₆₀₀ }
Since the vector x ^m _i = {xi, xi + 1}, i = 1,2,3,… Nm-1, i = 1,2,3,…, 597. The vector x ^m _i is x ² ₁ = {x ₁ , x ₂ }, x ² ₃ = {x ₃ , x ₄ }, x ² ₄ = {x ₄ , x ₅ },…, x ² ₅₉₇ = {x It becomes ₅₉₇ , x ₅₉₈ }.
On the other hand, the vector x ^m _j = {x _j , x _{j + 1} }, j = m + 1, m + 2, m + 3,… N-m + 1, so j = 3,4,5 , ... 599. The vector x ^m _j is x ² ₃ = {x ₃ , x ₄ }, x ² ₄ = {x ₄ , x ₅ }, x ² ₅ = {x ₅ , x ₆ },…, x ² ₅₉₉ = {x It becomes ₅₉₉ , x ₆₀₀ }.

First, the m-dimensional vector generation processing unit 803 sets j = 2,3,4, ... 599 for the vector x ² ₁ = {x ₁ , x ₂ } of i = 1, and the vector x ² ₂ = {x. ₂ , x ₃ }, x ² ₃ = {x ₃ , x ₄ }, x ² ₄ = {x ₄ , x ₅ },…, x ² ₅₉₉ = {x ₅₉₉ , x ₆₀₀ }, sequentially, norm arithmetic processing Output to unit 805a.
The norm calculation processing unit 805a is a composite vector obtained by subtracting the vector x ² ₂ = {x ₂ , x ₃ } from the vector x ² ₁ = {x ₁ , x ₂ } input from the m-dimensional vector generation processing unit 803. Calculate the absolute value of the norm.

Next, the norm calculation processing unit 805a subtracts the vector x ² ₃ = {x ₃ , x ₄ } from the vector x ² ₁ = {x ₁ , x ₂ } sequentially input from the m-dimensional vector generation processing unit 803. Calculate the absolute value of the norm of the composite vector, calculate the absolute value of the norm of the composite vector by subtracting the vector x ² ₄ = {x ₄ , x ₅ } from the vector x ² ₁ = {x ₁ , x ₂ }, ..., Calculate the absolute norm of the composite vector by subtracting the vector x ² ₅₉₉ = {x ₅₉₉ , x ₆₀₀ } from the vector x ² ₁ = {x ₁ , x ₂ }.

Next, the m-dimensional vector generation processing unit 803 increments i by one. Then, for the vector x ² ₂ = {x ₂ , x ₃ } of i = 2, the vector x ² ₃ = {x ₃ , x ₄ }, x ² ₄ = where j = 3,4,5,… 599 {x ₄ , x ₅ }, x ² ₅ = {x ₅ , x ₆ },…, x ² ₅₉₉ = {x ₅₉₉ , x ₆₀₀ } are sequentially output to the norm calculation processing unit 805a.
The norm calculation processing unit 805a is a composite vector obtained by subtracting the vector x ² ₃ = {x ₃ , x ₄ } from the vector x ² ₂ = {x ₂ , x ₃ } input from the m-dimensional vector generation processing unit 803. Calculates the absolute value of the norm of. Similarly, the norm calculation processing unit 805a calculates the absolute value of the norm of the difference between the vectors sequentially input from the m-dimensional vector generation processing unit 803.

The m + 1 dimensional vector generation processing unit 804 creates two (m + 1) dimensional vectors composed of (m + 1) adjacent elements in the time series in the normalized epoch 802.
Hereinafter, the m + 1 dimensional vector generation processing unit 804 performs the same arithmetic processing as the m dimensional vector generation processing unit 803.

(2) Pupil diameter complexity calculation processing unit 306 calculated the sample entropy in order to quantify the complexity of the pupil diameter data sequence. The method of calculating the index of the complexity of the pupil diameter data sequence is not limited to the sample entropy. In addition to the sample entropy, for example, Higuchi's fractal dimension or Omega-complexity may be used as a method for calculating an index of the complexity of the pupil diameter data sequence.

As an example of a feasible index of complexity other than sample entropy, the correlation dimension will be described below. Correlation dimension, unlike stochastic sample entropy, realizes deterministic complexity.
x ^m _i = {x _i , x _{i + 1} ,…, x _{i + m-1} } x ^m _j = {x _j , x _{j + 1} ,…, x _{j +} generated by the m-dimensional vector generation processing unit 803 For _m-1 }, the probability of entering a sphere with a radius ε centered on x ^m _i is given by the following equation (5).

Other indicators of deterministic complexity include the maximum Lyapunov exponent.

(3) The pupil diameter causality calculation processing unit 305 calculated the moving entropy in order to quantify the left and right causality of the pupil diameter data string. As another quantity related to causality, the method of calculating the left-right reciprocity measure of the pupil diameter data sequence is not limited to mobile entropy. In addition to the mobile entropy, it may be evaluated by the mutual correlation coefficient and the mutual information amount of the left and right pupil diameter time series. Unlike mobile entropy, when mutual correlation coefficient and mutual information amount are used, the direction of left to right and right to left disappears, but as shown in FIG. 7, the direction of left to right and right to left. Since the average value is calculated for the mobile entropy having the property, there is no particular problem even if the mutual correlation coefficient or the mutual information amount is adopted.

In addition to mobile entropy, Granger causality is an index for observing causality between time series.
Also, instead of causality, the degree of synchronization between the x _i and y _i time series may be evaluated. In that case, the index includes the amount of mutual information and the correlation coefficient. Specifically, the correlation coefficient R is described by the following equation (6).

(4) In the first embodiment of the present invention, a psychiatric disorder estimation device 101 for estimating the morbidity rate of psychiatric disorders of a subject or a patient 102 using a classifier 308 is disclosed.
In the second embodiment of the present invention, a mental illness estimation device 501 for estimating the severity of a mental illness of a subject or a patient 102 using a regression device 502 is disclosed.
All mental illness estimators have in common that they use learning algorithms to estimate psychiatric parameters (morbidity or severity).

The classifier 308 and the retriever 502 have a pupil diameter average value output by the pupil diameter average calculation processing unit 304, a pupil diameter causality output by the pupil diameter causality calculation processing unit 305, and a pupil diameter complexity calculation processing unit 306. Based on the pupil diameter complexity output by, it can be summarized by the superordinate concept of a mental illness estimation processing unit that outputs the morbidity or severity of a plurality of mental illnesses of a subject or patient 102.

(5) The psychiatric disorder estimation device according to the present invention utilizes the fact that the behavior of the sympathetic nervous system and the parasympathetic nervous system affects the pupil diameter. Phenomena that can be estimated from the behavior of the sympathetic nervous system and the parasympathetic nervous system are not limited to mental illness.
Pupil diameter is thought to reflect human attention and alertness. Inadvertent accidents are a problem when driving a car or the like. Therefore, from the pupil diameter index, it may be possible to determine a state in which the driver is distracted or a state in which he / she is not able to concentrate on driving due to a decrease in arousal.

Therefore, an eye movement measuring device 103 is installed on the ceiling of the driver's seat of the automobile and on the peripheral portion of the sun visor to estimate a state in which the driver's concentration is reduced and a state in which arousal is reduced. The 101 is connected to the eye movement measuring device 103. By providing the mental illness estimation device 101 in the automobile, it is possible to sound an alarm informing the driver of the danger at an early stage from the situation before the driver is drowsy.

That is, the phenomenon estimated by the mental illness estimation device according to the present invention is not limited to the mental illness. Therefore, the mental illness estimation device according to the present invention can estimate not only the mental illness but also the state in which the subject or the patient 102 has decreased concentration and the state in which the arousal is decreased. It can be paraphrased by the superordinate concept of a device.

In the embodiment of the present invention, the mental illness estimation device is disclosed.
The mental illness estimation device 101 and the mental illness estimation device 501 calculate three scalar values of pupil diameter average value, pupil diameter causality, and pupil diameter complexity from the pupil diameter data sequence of the subject or patient 102, and a learning algorithm. It is input to the classifier 308 or the regressor 502 based on the above. This makes it possible for the mental illness estimation device 101 and the mental illness estimation device 501 to extremely objectively estimate the susceptibility to mental illness or the severity of mental illness of the subject or patient 102.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and other modifications and applications are provided as long as they do not deviate from the gist of the present invention described in the claims. including.

101 ... mental disease estimation device, 102 ... patient, 103 ... eye movement measuring device, 104 ... jaw stand, 201 ... bus, 202 ... CPU, 203 ... ROM, 204 ... RAM, 205 ... display unit, 206 ... operation unit, 207 ... Non-volatile storage, 208 ... NIC, 301 ... Left and right pupil diameter data file, 302 ... Sample number regulation processing unit, 303 ... LPF, 304 ... Puppy diameter average calculation processing unit, 305 ... Puppy diameter causal calculation processing unit, 306 ... Puppy diameter complexity arithmetic processing unit, 307 ... feature vector arithmetic processing unit, 308 ... classifier, 309 ... approximate function parameter, 310 ... input / output control unit, 401 ... learner, 402 ... left and right pupil diameter data file, 403 ... teacher Data, 501 ... Psychiatric disease estimator, 502 ... Retractor, 503 ... Approximate function parameter, 601 ... Learner, 602 ... Teacher data, 701 ... Right pupil diameter data, 702 ... Left pupil diameter data, 703 ... First average value Arithmetic processing unit, 704 ... Data subdivision processing unit, 705 ... Right epoch, 706 ... Left epoch, 707 ... Switch, 708 ... Sample entropy calculation processing unit, 709 ... Second average value calculation processing unit, 710 ... Left / right movement entropy calculation Processing unit, 711 ... right / left movement entropy calculation processing unit, 712 ... third average value calculation processing unit, 801 ... Z scoring processing unit, 802 ... normalized epoch, 803 ... m-dimensional vector generation processing unit, 804 ... m + 1-dimensional vector Generation processing unit, 805a, 805b ... Norm calculation processing unit, 806a, 806b ... Comparator, 807 ... Threshold r, 808a, 808b ... Probability calculation processing unit, 809 ... Divider, 810 ... Logarithmic processing unit, 811 ... Multiplier, 812 ... Negative value, 901 ... Bin width adjustment processing unit, 902 ... Normalized right echo, 903 ... Normalized left echo, 904a, 904b ... Delay vector generation processing unit, 905 ... Delay vector setting value, 906 ... Right delay vector, 907 ... Left delay vector, 908a, 908b ... Simultaneous probability calculation processing unit, 909a, 909b ... Additional simultaneous probability calculation processing unit, 910 ... First divider, 911 ... Second divider, 912 ... Third divider, 913 ... Logarithmic arithmetic processing unit, 914 ... Multiplier, 915 ... Total arithmetic processing unit

Claims (5)

A pupil diameter averaging processing unit that calculates the average value of the pupil diameter from the left and right pupil diameter data that continuously measures the pupil diameter of the left and right eyes of the subject or patient.
From the left and right pupil diameter data, a pupil diameter complexity calculation processing unit that calculates the regularity of the variation pattern of the pupil diameter, and
A pupil diameter causality calculation processing unit that calculates the causality of the left and right pupil diameters from the left and right pupil diameter data,
Based on the pupil diameter average value output by the pupil diameter average calculation processing unit, the pupil diameter causality output by the pupil diameter causal calculation processing unit, and the pupil diameter complexity output by the pupil diameter complexity calculation processing unit. A psychiatric activity estimation device comprising a psychiatric disorder estimation processing unit that outputs the morbidity rate and / or severity of the subject or patient in a plurality of psychiatric disorders. The mental illness estimation processing unit is a classifier that outputs the morbidity rate of the subject or patient in a plurality of mental illnesses.
The neuropsychiatric activity estimation device according to claim 1. The psychiatric disorder estimation processing unit is a regressor that outputs the severity of the subject or patient in a plurality of psychiatric disorders.
The neuropsychiatric activity estimation device according to claim 1. The pupil diameter complexity calculation processing unit calculates the average value of the sample entropy of the left and right pupil diameter data.
The neuropsychiatric activity estimation device according to any one of claims 1 to 3. The pupil diameter causality calculation processing unit calculates the average value of the moving entropy of the left and right pupil diameter data.
The neuropsychiatric activity estimation device according to any one of claims 1 to 3.

Images