WO2017065241A1 - Automated diagnostic device - Google Patents

Abstract

Provided is an automated diagnostic device 1 for cranial nerve disorders. A movement measuring unit 10 includes sensors 12 which are attached to a patient 2, and measures movements, namely the posture and/or vibrations of the patient 2, on the basis of outputs from the sensors 12. A feature extracting unit 20 extracts feature quantities of the movements on the basis of measured data S2 from the movement measuring unit 10. On the basis of the feature quantities an interpreter 30 generates diagnostic data indicating the results of a cranial nerve disorder diagnosis.

Description

Automatic diagnostic equipment

The present invention relates to a pathological diagnosis apparatus for cranial nerve diseases such as Parkinson's disease.

Parkinson's disease (PD) is a progressive disease that exhibits extrapyramidal signs and is characterized by a lack of dopamine in the brain and a relative increase in acetylcholine. It is one of neurodegenerative diseases and is designated as an intractable disease (specific disease) in Japan. There are more than 150,000 PD patients in the country with only severe cases, and the number is increasing rapidly.

症状 Symptoms are broadly divided into motor and non-motor, but there is a major feature where significant movement disorders appear. This can be divided into resting tremors (trembling), postural reflex disturbances (postural abnormalities), and gait disturbances such as freezing legs, which are important clues for doctors to diagnose the stage.

JP 2011-177278 A

In addition to the above movement disorders, PD is diagnosed from multiple viewpoints such as CT findings and the effects of L-DOPA administration. However, only the findings of movement disorders depended on the visual evaluation of doctors, and a big problem remained in the quantitativeness and objectivity.

As an example, the Hoehn-Yahr classification is known as an index widely used for diagnosis of PD stage. If a system for automatically diagnosing the stage of PD can be constructed by simply measuring and evaluating such movement disorders in PD patients, it will be a gospel for doctors and patients involved in treatment. Such an automatic diagnosis system is useful not only for PD but also for various cranial nerve diseases with other movement disorders such as stroke and dementia.

The present invention has been made in such a situation, and one of the exemplary purposes of an embodiment thereof is an automatic diagnostic apparatus for a cranial nerve disease that can easily and quantitatively or objectively measure a stage, a degree of progression, and the like. Is in the provision of.

According to an aspect of the present invention, an apparatus for automatically diagnosing a cranial nerve disease is provided. This automatic diagnostic device focuses on posture, vibration, and walking among physical movements of patients to easily diagnose movement disorders that appear in at least one of them in order to diagnose the stage and progression of PD patients with cranial neurological disorders. And measure quantitatively, and diagnose the disease and progression based on it.

Specifically, features related to posture abnormalities as posture reflex disturbance from posture measurement, features related to tremor as resting tremor from vibration measurement, features related to freezing as a walking obstacle from walking measurement The amount may be evaluated. A feature vector including at least one feature quantity obtained in this way (that is, including one dimension) may be configured, and an automatic diagnosis apparatus may be constructed based on machine learning.

It should be noted that a combination of the above-described components arbitrarily or a conversion of the expression of the present invention between a method, an apparatus, etc. is also effective as an aspect of the present invention.

According to an aspect of the present invention, it is possible to easily and quantitatively or objectively measure the stage of disease and the degree of progression.

It is a block diagram of the automatic diagnostic apparatus which concerns on embodiment. It is a figure explaining Hoehn-Yahr classification. It is a block diagram of an automatic diagnostic apparatus. 4A to 4D are diagrams for explaining posture measurement. FIGS. 5A and 5B are diagrams showing the inclination φ in the left-right direction and the inclination θ in the front-rear direction actually obtained for each young healthy person and PD patient. FIG. 6A is a diagram illustrating another example of sensor arrangement, and FIG. 6B is a diagram illustrating a forearm and an upper arm. FIGS. 7A to 7D are diagrams in which mean, time, variance, and kurtosis are plotted on a dimension plane as feature vectors of gradients φ and θ. FIGS. 8A to 8D are diagrams showing the trajectories for 30 seconds of the horizontal inclination φ and the front-rear inclination θ obtained for each of the four groups. It is a figure which shows the result of a Kruskal-Wallis test. FIGS. 10 (a) to 10 (c) are graphs showing the values of the average and variance of the back and forth inclinations of the respective groups when sitting, standing and walking. FIG. 11A is a diagram illustrating an example of a classifier for young healthy individuals and PD patients, and FIG. 11B is a diagram illustrating an example of a mild and severe classifier for PD patients. 12A to 12E are diagrams for explaining vibration measurement. FIGS. 13A and 13B are diagrams showing the results of vibration measurement of PD patients and healthy individuals. FIGS. 14A and 14B are diagrams showing the power spectrum of a resting tremor and its time waveform measured for a healthy person and a PD patient, respectively. It is the figure which plotted the power of 4-6Hz of the seismic war at posture and the resting tremor on a two-dimensional plane. It is a time waveform figure of the acceleration norm of the right index finger measured about PD patient. FIGS. 17A and 17B are diagrams showing power spectra of a resting tremor measured for healthy subjects and PD patients. FIG. 18A is a diagram illustrating an example of a classifier for young healthy individuals and PD patients, and FIG. 18B is a diagram illustrating an example of a mild and severe classifier for PD patients. It is a figure which shows the angular velocity data of a Z-axis. It is a flowchart of orbit estimation. FIG. 21A is a diagram for explaining the correction of the X-axis angle, and FIG. 21B is a three-dimensional view of the ankle obtained from the acceleration and angular velocity data during walking by the estimation method according to the embodiment. It is a figure which shows an orbit (for 1 period). It is a figure which shows the track | orbit of an ankle when taking the front-back direction on a horizontal axis and taking the height direction on a vertical axis | shaft. It is a figure explaining the feature-value regarding a walking track. 24A and 24B are diagrams showing factor loadings of the first principal component and the second principal component in the principal component analysis. It is a figure which shows the feature space which plotted the walk state of the some participant. FIG. 26 (a) is a diagram showing the results of applying SVM to the mild PD group and the healthy elderly group, and FIG. 26 (b) is a diagram showing the results of performing SVM on the mild PD group and severe PD group. is there. It is a figure which shows an example of distribution on each feature vector of a young healthy person and Hoehn-Yahr classification once and twice. It is a figure which shows an example of the discriminator which classify | categorizes a young healthy person and PD patient constructed | assembled using SVM based on the vibration measurement data of a fingertip.

Hereinafter, the present invention will be described based on preferred embodiments with reference to the drawings. The same or equivalent components, members, and processes shown in the drawings are denoted by the same reference numerals, and repeated descriptions are omitted as appropriate. The embodiments do not limit the invention but are exemplifications, and all features and combinations thereof described in the embodiments are not necessarily essential to the invention.

FIG. 1 is a block diagram of an automatic diagnosis apparatus according to an embodiment. This automatic diagnosis apparatus 1 diagnoses a patient with a cranial nerve disease accompanied by a movement disorder, and generates diagnostic data S4 indicating the diagnosis result. The diagnosis data S4 may be an index indicating a stage, a degree of progression, a sign, and the like. In the present embodiment, for ease of understanding, the automatic diagnosis apparatus 1 for PD will be described, and the diagnosis data S4 indicates the Hoehn-Yahr classification (1 to 6) related to the PD disease. Shall.

FIG. 2 is a diagram for explaining the Hoehn-Yahr classification. In the revised Hoehn-Yahr classification, 2.5-4 degrees with posture reflex disorder may be classified as severe, and 1-2 degrees without attitude may be classified as mild. Alternatively, diagnostic data S4 indicating mildness may be generated.

Return to Figure 1. The automatic diagnosis apparatus 1 includes a motion measurement unit 10, a feature extraction unit 20, and an interpreter 30. The motion measurement unit 10 measures the motion S1 of the patient 2. Specifically, the motion measurement unit 10 measures at least one of (i) posture and (ii) vibration (tremor) among (i) posture, (ii) vibration (tremor), and (iii) walking of the patient 2. To do. The movement S1 can be measured, for example, by one or more sensors 12 attached to the patient 2. The motion measurement unit 10 generates measurement data S2 indicating the motion S1 by converting the output of the sensor 12 into digital data. Various sensors such as an acceleration sensor, a speed sensor, a gyro sensor, and a geomagnetic sensor can be used as the sensor. The measurement data S2 shows a time waveform of the motion S1. The sensor 12 may be wireless or wired. Alternatively, in place of or in addition to the sensor 12, the marker 14 may be attached to one or more specific parts of the patient 2 and observed using a video camera, and the motion S1 may be measured based on the movement of the marker 14. Good.

The feature extraction unit 20 extracts a feature value S3 based on the measurement data S2. Examples of the feature amount S3 include an average, a variance, a skewness, a kurtosis, and a spectrum with respect to a time waveform of certain measurement data. Or when several measurement data S2 are obtained, it is good also considering those difference, the sum, a product, correlation, etc. as feature-value S3. The type of exercise and the feature amount may be determined according to the disease to be diagnosed. This will be described later.

When a plurality of feature amounts S3 are measured, a combination of them can be understood as a vector (hereinafter referred to as a feature vector). The single feature quantity S3 can also be interpreted as a one-dimensional feature vector. Therefore, hereinafter, it is also referred to as a feature vector S3 regardless of the number of feature quantities S3.

The interpreter (semantic understanding unit) 30 generates diagnostic data S4 by comparing the calculated feature vector S3 with the database 32. The database 32 can be generated by prior machine learning. That is, measurement data (also referred to as a learning sample) S2 of a young healthy person, a healthy elderly person, and an affected person with a different Hoehn-Yahr classification frequency are collected in advance. The interpreter 30 finds the correlation between the feature vector S3 obtained from the learning sample and the frequency of the Hoehn-Yahr classification. Specifically, in the feature vector space, a hyperplane serving as a boundary of the frequency of Hoehn-Yahr classification is learned, and a discriminator is constructed. The interpreter 30 can determine the frequency of Hoehn-Yahr classification by referring to the database 32 using a pattern classifier such as a support vector machine or principal component analysis.

The database 32 may be generated for each patient 2 property, that is, attribute or characteristic. For example, even if young people and elderly people have the same Hoehn-Yahr classification frequency, the feature vector S3 may show different tendencies. In this case, a separate database 32 may be generated for each age group. In addition to age, groups may be grouped by sex and physique, and an individual database 32 may be constructed.

Further, when the automatic diagnostic apparatus 1 is operated, the database 32 may be updated every time in order to reflect the newly measured feature vector S3 in the database 32. The database 32 may be stored in a part of the automatic diagnosis apparatus 1 or a hard disk of a computer associated therewith. Alternatively, the database 32 may be stored on a server connected to the automatic diagnosis apparatus 1 via a network. By storing the database 32 on the server, it is possible to share information with many medical institutions, thereby increasing the number of subjects, so that the database 32 can be enriched and the accuracy of diagnosis can be improved. . In machine learning, it is meaningful to divide the database 32 because the amount of computation increases explosively as the number of learning samples increases.

Further, the automatic diagnosis apparatus 1 may be mounted using a cloud computing architecture. For example, part or all of the processing of the feature extraction unit 20 and the interpreter 30 may be executed by a server on the cloud.

The feature extraction unit 20 and the interpreter 30 can be configured by a computer, that is, they can be a combination of hardware such as CPU and memory and software.

The above is the basic configuration of the automatic diagnosis apparatus 1. Next, the automatic diagnostic apparatus 1 will be specifically described. FIG. 3 is a block diagram of the automatic diagnosis apparatus 1. In the block diagram of FIG. 3, the patient 2 is shown inside the motion measurement unit 10, but this is for convenience of explanation, and it goes without saying that the patient 2 is not a component of the automatic diagnostic apparatus 1.

In the present embodiment, the motion measurement unit 10 measures (i) posture, (ii) vibration, and (iii) walking as the motion of the patient 2, and generates measurement data S2A to S2C indicating each of them. The measurement data S2 is measured by the wearable sensor 12. As the wearable sensor 12, a small 6-axis sensor (acceleration 3 axes and angular velocity 3 axes) that can be worn on the body of the patient 2 can be used. It is desirable that the time resolution is 100Hz or more, the acceleration range is ± 2g / 16bit or more, and the angular velocity range is ± 250dps / 16bit or more. The body translational motion and the inclination with respect to the direction of gravitational acceleration are evaluated from the acceleration information, and the rotational motion is evaluated from the angular velocity information. The measurement data S2 obtained in the motion measurement unit 10 is input to a computer (that is, the feature extraction unit 20 and the interpreter 30) by wire or wireless, and data analysis is performed online or offline. Note that the sensor mounting position and the feature value extraction method differ between posture measurement, vibration measurement, and walking measurement, and will be described in detail later.

The feature extraction unit 20 generates feature amounts S3A to S3C for each of the measurement data S2A to S2C. Specifically, the feature extraction unit 20 extracts feature amounts S3A to S3C from motions such as posture, vibration, and walking estimated by an acceleration integration system, an angular velocity integration system, a correction algorithm, and the like. These feature amounts S3A to S3C are input to the interpreter 30 as a feature vector S3.

The interpreter 30 refers to the database 32, determines a hyperplane (frequency threshold) in the vector space based on machine learning, and determines the stage based on the hyperplane and the feature vector S3 from the feature extraction unit 20. Classify.

Hereinafter, each type of motion measured by the automatic diagnosis apparatus 1 will be described in detail.

1. Posture Evaluation Posture evaluation focuses on postural reflex disorder in PD patients. This is a symptom that the vertical body axis is tilted back and forth or left and right when standing, and mainly uses angle information about the axis of gravity acceleration obtained from a group of acceleration sensors mounted in the body axis direction. It is possible to use the inclination (gradient) in the front-rear direction, the inclination in the left-right direction, their spatial correlation, temporal variation, and the like as feature quantities.

Measured postures include three types of postures: standing posture, sitting posture and walking posture. The standing position is a posture in which the hand is vertically lowered on the body side, and the sitting position is a posture when sitting without putting a back on the backrest. In addition, walking means walking for about one minute on a straight line for a total of one round trip.

1.1 Posture measurement When posture measurement is performed, the attachment site of the sensor 12 and the posture of the patient at that time are important. Various positions are conceivable as the attachment site. 4A to 4D are diagrams for explaining posture measurement. 4A and 4B show an example of sensor arrangement. In this example, sensors 12_1 and 12_2 are mounted 10 cm below C7 from the vertebra and L4, respectively. C7 is the portion that protrudes most behind the neck when the head is lowered, and mainly detects the state of the back. L4 is the intersection of the Jacobi line and the lumbar vertebra that connects the upper ends of the hip bones, and mainly detects the state of the waist. In addition to the feature that the posture useful for staging can be accurately measured, these combinations also have an advantage that it is easy to determine the attachment position when the sensor 12 is attached to the patient 2.

As shown in FIGS. 4C and 4D, the posture can be evaluated by a combination of the front-rear direction inclination θ and the left-right direction inclination φ. When the gravitational acceleration is a, the forward / backward inclination θ is expressed by the following equation. The X-axis represents the right hand direction of the patient, the Y-axis represents the forward direction, the Z-axis represents the vertical direction, and the subscripts x, y, and z represent components in the respective directions.
θ ＝ arctan (a _z / a _x )… a _x > 0
θ = arctan (a _z / a _x ) + π… a _x <0 and a _z > 0
θ ＝ arctan (a _z / a _x ) −π… a _x <0 and a _z <0
θ = π / 2… a _x = 0 and a _z > 0
θ = −π / 2… a _x = 0 and a _z <0
θ: No definition… a _x = 0 and a _z = 0

Moreover, the inclination φ in the left-right direction is expressed by the following equation.
φ = arctan (a _z / a _y )… a _y > 0
φ = arctan (a _z / a _y ) + π… a _y <0 and a _z > 0
φ ＝ arctan (a _z / a _y ) −π… a _y <0 and a _z <0
φ = π / 2… a _y = 0 and a _z > 0
φ = −π / 2… a _y = 0 and a _z <0
φ: Not defined… a _y = 0 and a _z = 0

FIGS. 5 (a) and 5 (b) are diagrams showing the left-right inclination φ and the front-rear inclination θ actually obtained for young healthy individuals and PD patients, respectively. This result is obtained by measuring the gradient of the body axis for 1 minute with the acceleration sensor 12_1 mounted at a position 10 cm below C7 in the standing posture shown in FIG. 4B. FIG. 5 (a) shows the results for seven healthy subjects, and FIG. 5 (b) shows the results for seven PD patients.

Clearly, it can be seen that the healthy person is near the origin, that is, the tilt angle is close to 0 degrees in the front-rear direction and the left-right direction. On the other hand, the PD patient is located in a region far away from the origin, and it can be confirmed that the body axis is greatly inclined in the front-rear direction and the left-right direction. Further, it can be seen that the fluctuations of the gradients φ and θ are very small and stable in healthy subjects, but are greatly fluctuating in PD patients.

According to the sensor arrangement in FIG. 4B, posture evaluation can be performed simply, but detailed posture evaluation may be performed by increasing the number of sensors. FIG. 6A is a diagram illustrating another example of sensor arrangement. In this example, it is possible to measure the posture together with the skeleton model of the body by attaching a plurality of sensors to locations where the rotation matrix of each joint can be obtained. In this case, at least 11 sensors are required for the upper body only, and at least 17 sensors are required for the lower body.

An example of an algorithm for obtaining the rotation angle of the joint will be described. FIG. 6B shows the forearm and the upper arm. As shown in the figure, consider a left-handed coordinate system in which the X axis is taken from the shoulder to the elbow, the elbow to the hand, and the Y axis is taken from the elbow head as shown in the figure. At this time, rolls, pitches, and yaws with rotation angles in the counterclockwise directions of the upper arms X, Y, and Z are φ _u , θ _u , and ψ _u , respectively, and φ _f , θ _f , and ψ _f are the front arms, respectively. .

In determining θ _f at this time, steady components of acceleration with respect to the forearm X axis and Y axis are a _x and a _y , the angular velocity in the pitch direction is ω _θ , and measurement errors with _respect to each are _ex , e _y , and e _θ . Then, the pitch angle θ _fgyr obtained from the angular velocity is expressed by Expression (1).
θ _fgyr = θ _f0 + ∫ _{0 to t} ω _θ (t) dt + ∫ _{0 to t} e _θ (t) dt (1)

Pitch angle theta _facc determined from the acceleration is expressed by equation (2).
a _x = g _x0 cos φ _u + e _x
a _y = g _y0 cos φ _u + e _{y
θ facc = arctan (a x /} a y) ... (2)

θ _fgyr accumulates accumulated errors as the measurement time increases, and θ _facc uses an acceleration sensor with a high S / N ratio to increase the angular velocity estimation accuracy.

Here, when | φ _u | <ε or | φ _u |> π−ε, the measurement error of pitch angle θ _{fgyr obtained} from acceleration (that is, the third term on the right side of equation (1)) is sufficiently small. Therefore, θ _{ft ′} = θf _acc is calculated at time t ′, and thereafter, θ _f = θ _fgyr can be calculated according to equation (3).
θ _fgyr = θ _{ft ′} + ∫ _{t ′ to t} ω _θ (t) dt (3)

When the measurement error component of _{θf acc} obtained from the measurement value of the acceleration sensor is large, θ _f = θ _fgyr and θ _f−u is the angle between the forearm and the upper arm is θ _f−u = θ _f −θ _u and the rotation matrix R for converting from the upper arm coordinate system to the forearm coordinate system can be obtained as follows.

¡By applying this method to multi-degree-of-freedom joints, the rotation matrix of each joint can be calculated, and the posture of the upper body can be estimated based on the skeleton model.

1.2 Posture Feature Amount Next, the posture feature amount will be described. Various features can be considered, but as an example here, as noted above, when focusing on the body axis gradient θ in the front-rear direction and the body axis gradient φ in the left-right direction, the first order statistics (Average), second order statistics (variance), third order statistics (distortion), and fourth order statistics (kurtosis) can be used as feature quantities. 7A to 7D are diagrams in which mean, time, variance, and kurtosis are plotted on a two-dimensional plane as feature vectors of gradients φ and θ.

In each figure, healthy subjects are indicated by circles and PD patients are indicated by triangles. As is clear from this result, significant differences between the healthy subjects and PD patients are observed in terms of the mean, variance, and kurtosis of the gradient, and it can be seen that this is an effective feature amount for separating both groups. This is an important finding for automatic diagnosis of the stage of PD patients.

Note that the feature quantities that can be used are not limited to these. Although the description here is the feature amount at the time of standing, it is possible to define further feature amounts at the time of different postures and exercises by performing measurement at the time of sitting and walking.

Also, by looking at the differences and ratios in each posture, it is possible to see the change in the inclination of the body axis according to the posture of the same individual, and it is possible to define further feature quantities. It can be considered that the inclination of the body axis caused by the distortion in the skeleton also occurs in a different posture. Therefore, by looking at the difference in the tilt angle of the body axis between postures, for example, when sitting and standing, it may be possible to infer whether the tilt during standing is due to skeletal abnormalities or other causes.

Also, attention may be paid to time series information. For example, by looking at the change in the inclination of the body axis during walking, the influence due to fatigue can be seen.

In summary, in posture evaluation, a feature vector may be formed using any combination of the average value of gradient φ, variance, kurtosis, average value of gradient θ, variance, and kurtosis, and machine learning may be performed.

In addition, each of the above measurements is an example of a statistical feature quantity based on gradient measurement at one point, but further feature quantities can be defined by extending to spatial correlation or temporal correlation. Is possible.

For example, as shown in FIG. 4 (b), spatial characteristics can be taken into account if sensors are attached to a plurality of locations and motion is measured. PD patients are known to exhibit unilateral parkinsonism at Hoehn-Yahr classification once and bilateral parkinsonism at twice. Therefore, it is possible to distinguish between the right half and the left half by spatially characterizing the difference.

Furthermore, if time is measured continuously, it is possible to take into account characteristics related to time variation. For example, when the body is tilted, feedback is applied so that the body is returned to its original position in a short time in order to correct it, so that a periodic movement can be found on the time axis. On the other hand, PD patients are poor in such a response, and it is known that the time correlation in the fluctuation of movement is low when it becomes severe. Therefore, evaluation of an autocorrelation function, cross-correlation function, or self-similarity is also effective as a feature quantity.

1.3 Experimental Results on Posture Below, we will explain some of the experiments on the posture and the knowledge obtained from them.

(Experiment 1)
Participants are divided into 4 groups.
・ 19 healthy young people (19 men)
・ Healthy Elderly 17 people (8 men, 9 women)
・ Mild PD (Mild PD) (mH & Y: 1-2) 19 (11 men, 8 women)
・ Severe PD (mH & Y: 2.5-4) 24 (9 men, 15 women)

The posture was measured under the following three conditions.
・ Sitting
・ Standing
・ Walking

The measurement data is the following four.
-The back inclination in the left-right direction obtained from the upper sensor 12_1 (Frontal Angle) φ _u (t)
The back-and-forth back inclination (Sagittal Angle) θ _u (t) obtained from the upper sensor 12_1
・ Frontal Angle φ _L (t) obtained from the lower sensor 12_2 in the horizontal direction
・ Longitudinal waist inclination (Sagittal Angle) θ _L (t) obtained from the lower sensor 12_2

From these, the left / right / front / rear ratio is also calculated.
-Ratio of upper sensor 12_1 φ _u (t) / θ _u (t)
・ Ratio of lower sensor 12_2 φ _L (t) / θ _L (t)

The following feature amounts were calculated for each of the gradients φ _u , θ _u , φ _L , and θ _L obtained under each condition.
・ Range
・ Average
・ Variance
・ Skewness
・ Kurtosis

With regard to each feature amount thus obtained, attention was also paid to the following amounts.
・ Difference between standing and sitting (Standing-Sitting)
・ Walking-Sitting

Therefore, in this experiment, five items related to measurement conditions (measurement conditions of sitting / standing / walking and the difference between standing and sitting, walking and sitting), two items related to measuring position (back and waist), and tilt direction By combining 3 items (front / rear / left / right and their ratio) and 5 items related to statistics (range, average, variance, skewness, kurtosis) in a matrix, a total of 150 feature values are comprehensively extracted. .

The Kruskal-Wallis test was performed between the healthy young group, healthy elderly group, mild PD group, and severe PD group, and the presence or absence of a significant difference in each feature amount was confirmed. At this time, multiple comparisons were made using the Steel-Dwass method. As a result, a feature amount that has confirmed a significant difference between groups can be an index for quantitative evaluation of posture abnormalities in PD.

8 (a) to 8 (d) are diagrams showing the trajectories for 30 seconds of the horizontal inclination φ and the forward / backward inclination θ obtained for each of the four groups. As shown in FIGS. 8A and 8B, it can be seen that a healthy person regardless of age has an inclination angle close to 0 degrees in the vicinity of the origin, that is, in the front-rear direction and the left-right direction. On the other hand, the mild PD patient in FIG. 8C is located in a region away from the origin, and it can be confirmed that the body axis is greatly inclined in the front-rear direction and the left-right direction. The tendency for PD patients is even stronger. Further, it can be seen that the fluctuations (range) of the gradients φ and θ are very small and stable in normal subjects, but are greatly fluctuating in PD patients.

FIG. 9 is a diagram showing the results of the Kruskal-Wallis test. The p-value is plotted with 4 ranks. As is apparent from FIG. 9, among the 150 feature values, 41 features such as dispersion of the back and forth and right and left tilt angles when sitting, dispersion of the back when standing, and dispersion of the back and forth of the back when walking A significant difference between the groups in the amount could be confirmed.

FIGS. 10 (a) to 10 (c) are graphs showing the values of the average and variance of the back and forth inclinations of the respective groups when sitting, standing and walking. FIG. 10A shows the average value and variance of the inclination θ _u (t) in the front-rear direction of the back when sitting. FIG. 10B shows the average value and variance of the inclination θ _u (t) in the front-rear direction of the back when standing. FIG. 10C shows the average value and the variance of the inclination θ _u (t) in the front-rear direction of the back during walking.

In the average before and after the back in the sitting position shown in FIG. 10 (a), no significant difference can be confirmed between the groups, but the dispersion before and after the sitting position shown in FIG. Significant differences can be confirmed between groups in the mean, variance, and mean and variance before and after walking shown in FIG.

In addition, regarding the average during standing, the variance, and the average during walking, it was confirmed that the feature value tends to increase as it becomes more severe, but the value of the severe PD group becomes smaller with respect to variance during walking. A trend was confirmed.

As can be seen from FIGS. 10B and 10C, when standing or walking, the average of the forward and backward inclinations θ _u (t) of PD patients is larger than that of healthy individuals. This is thought to be because Camptocormia (back fracture) often seen in PD patients can be quantitatively evaluated. With regard to hip folding, it appears when standing and walking, but not when sitting.

The fact that the variance increases as the severity of PD increases can be attributed to a decrease in posture retention due to muscle weakness, but the variance in the PD group increases compared to the difference between the healthy young group and the healthy elderly group. Therefore, there is a possibility that an influence such as dyskinesia (involuntary movement) can be detected.

Also, from FIG. 10 (c), the variance during walking of severe PD patients is lower than that of the other groups. This can be related to the fact that you are walking without moving your back greatly due to the influence of Bradykinesia.

Thus, the above experimental results support that the automatic diagnosis apparatus 1 according to the embodiment is effective in quantitative evaluation of posture abnormalities in PD. In this experiment, the configuration of the classifier that distinguishes the severe PD group from the mild PD group is limited due to the number of samples, but by increasing the number of samples and selecting the feature vector appropriately, the Hoehn-Yahr classification Diagnosing 1-5 degrees is realistic enough.

In reality, since the patient's age is already known, there is no need for automatic diagnosis. Therefore, if a database is constructed for each age category, it is possible to more easily and accurately determine the severity of a PD patient based on a database constructed for the same age category.

(Experiment 2)
Experiment 2 has the same conditions as Experiment 1, but in Experiment 2, participants are classified into three groups.
・ Severe PD patients (13)
・ Mild PD patients (15)
・ Young healthy subjects (7)

FIG. 11A is a diagram showing an example of a classifier for young healthy individuals and PD patients, and FIG. 11B is a diagram showing an example of a mild and severe classifier for PD patients. FIG. 11A is a graph plotting the relationship between the dispersion of the tilt of the back and forth of the back when standing and the difference between the average values of the tilts of the back and left and right obtained when standing and sitting as feature vectors. It is. According to the classifier constructed by SVM, it is possible to diagnose young healthy persons and PD patients with a certainty of 72.2%.

FIG. 11B is a graph plotting the relationship between the difference between the average values of the back and front inclinations obtained during standing and sitting and the dispersion ratio of the waist inclination when standing as a feature vector. It is. According to the classifier constructed by SVM, it is possible to diagnose severe PD patients and mild PD patients with a certainty of 71.4%. The certainty of diagnosis can be further increased by increasing the dimension of the feature vector.

2. Vibration evaluation In vibration evaluation, we focus on the resting tremor of PD patients. This is a symptom in which hands and fingers spontaneously tremble when resting without voluntary movement or the like, and vibration information obtained from a plurality of acceleration sensors that are attached to the hand, fingers, and parts that easily vibrate may be mainly used. In addition to the vibrations that have been observed in the 4-6 Hz band, which has been attracting attention in the past, it is possible to pay attention to the lower and higher frequency bands, as well as their spatial and temporal correlations. .

For example, paying attention to the norm of fingertip acceleration, the time variation may be measured. Alternatively, the vibration may be spectrally analyzed and separated for each frequency band, and the respective characteristics may be examined.

2.1 Vibration measurement As a method for measuring seismic warfare, the location of the sensor and the posture of the patient at that time are important.
12A to 12E are diagrams for explaining vibration measurement. FIGS. 12A and 12B show an example of the arrangement of the sensor 12B. Although various positions are conceivable as the mounting site of the sensor 12B, it may be fixed to the upper surface of the distal phalanx of the index finger as shown in FIG. Or you may fix to the back of a hand, as shown in FIG.12 (b).

Various postures can be considered when measuring earthquakes. FIGS. 12C to 12E illustrate postures at the time of measuring the earthquake. FIG. 12 (c) shows a measurement of a resting tremor, FIG. 12 (d) shows a posture seismic measurement, and FIG. 12 (e) shows an intentional tremor measurement. In a resting seismic battle, the elbow and back of the hand are placed on the armrest, and the measurement is performed in a natural shape with the palm up. In the postural seismic battle, measure with the palm down and the arm in front horizontally. In an intention tremor, the tremor during the finger-nose test will be measured.

FIGS. 13A and 13B are diagrams showing the results of vibration measurement of PD patients and healthy individuals. As shown in FIG. 7 (a), the sensor 12B is attached to the distal phalanx of the index finger, and the time variation of the acceleration norm is shown for 4 seconds. In each figure, the light gray is the index finger on the right hand, and the black line is the index finger on the left hand. Obviously, significant fingertip shaking is observed in PD patients. Moreover, it is a typical symptom of unilateral parkinsonism that occurs only in the right hand and is likely to occur in Hoehn-Yahr classification once. On the other hand, such vibration does not occur in healthy persons. In other words, it was confirmed that vibration measurement is very useful for frequency classification of PD patients.

2.2 Features of vibration Next, features of vibration will be described.

Here, it is divided and evaluated for each frequency band. First of all, it is effective to pay attention to power as a characteristic for vibrations having a characteristic peak in the 4 to 6 Hz band, which has been attracting attention as a seismic battle. FIGS. 14A and 14B are diagrams showing the power spectrum of a resting tremor and its time waveform measured for a healthy person and a PD patient, respectively. As shown in FIG. 14B, it can be seen that the vibration power of the PD patient is strong in the 4 to 6 Hz band. Therefore, in this example, power of 4 to 6 Hz is used as a vibration feature.

If this feature quantity is measured by the seismic motion at posture and the seismic motion at rest, a two-dimensional feature vector can be formed. FIG. 15 is a diagram in which the 4 to 6 Hz powers of the post-posture and resting tremors are plotted on a two-dimensional plane. The 14 PD patients have increased power of resting and postural tremors, but 7 healthy individuals are distributed near the origin. This means that the vibration power in the 4 to 6 Hz band is effective in discriminating both groups.

FIG. 16 is a time waveform diagram of the acceleration norm of the right index finger measured for a PD patient. It is also clear that a burst phenomenon is specifically observed in the resting tremor of PD patients on the lower frequency side (period 1-5 seconds, that is, 0.2-1 Hz) than the previous 4-6 Hz. It was. Specifically, as shown in FIG. 16, a burst phenomenon having a period of several seconds is observed. This phenomenon is regarded as a periodic variation of the amplitude of vibration and is one of important feature quantities.

17 (a) and 17 (b) are diagrams showing the power spectrum of a resting tremor measured for healthy subjects and PD patients. Healthy individuals and PD patients have different degrees of power attenuation with respect to frequencies in the high-frequency region, and it is observed that the power spectrum decreases linearly in the 10-40 Hz band. It will be an important feature in distinguishing In other words, healthy individuals have high fractal characteristics, whereas PD patients have low fractal characteristics. Further, as shown on the right side of FIG. 17B, there is a case where a peak is shown in the range of 30 to 40 Hz, and it is also useful to use the power in this band as the feature amount.

There are some other features that can be considered. All of the above measurements are examples of feature values calculated from vibration measurements at one point of the fingertip. However, it is possible to define additional feature values by extending to spatial correlation and temporal correlation. become.

For example, spatial characteristics can be taken into account by measuring finger vibrations at multiple points. PD patients are known to exhibit unilateral parkinsonism at Hoehn-Yahr classification 1 and bilateral parkinsonism at 2 degrees. Therefore, it is possible to distinguish between the right half and the left half by spatially characterizing the difference.

Furthermore, since the vibration of the seismic war is a dynamic motion that changes with time, it is possible to take into account the characteristics related to time fluctuation. It is known that when PD patients become severe, the temporal correlation in movement fluctuation is low. Therefore, evaluation of autocorrelation function, cross-correlation function, or self-similarity is also effective.

It is possible to define new feature values not only by looking at differences in measurement conditions but also differences in left and right vibrations and ratios. For example, stationary tremor is known as a characteristic symptom of PD. For this reason, it is possible to determine whether the vibration is generated only when stationary or whether the vibration is generated regardless of the posture by looking at the ratio of the vibration power when the posture is stationary.

2.3 Results of experiments related to earthquakes The following describes the experiments related to earthquakes and the knowledge obtained from them.
(Experiment 3)
In this experiment, the same 28 PD patients and 7 young healthy individuals as in Experiment 2 described in 1.3 were used as participants.

Fig. 18 (a) is a diagram showing an example of a classifier for young healthy individuals and PD patients, and Fig. 18 (b) is a diagram showing an example of a mild and severe classifier for PD patients.

Fig. 18 (a) is a plot of the 4-6Hz band power and the 30-45Hz band power of the stationary tremor as feature vectors. According to the classifier constructed by SVM, it is possible to diagnose young healthy individuals and severe PD patients with a probability of 75.0%.

Note that the frequency bands (4 to 6 Hz, 30 to 45 Hz) are merely examples, and it can be seen that a discriminator can be configured by using powers of a plurality of different frequency bands as feature vectors.

Fig. 18 (b) is a plot of the feature vectors of the ratio of stationary and postural tremors in the 4-6 Hz band, and the ratio of stationary seismic and postseismic tremors in the entire area. is there. According to the classifier constructed by SVM, it is possible to diagnose young healthy individuals and mild PD patients with a certainty of 73.3%.

The frequency band (4 to 6 Hz) is merely an example, and it can be seen that the discriminator can be configured by using the power ratio of a predetermined frequency band and the power ratio of the entire band as feature vectors.

3. Gait evaluation In gait evaluation, attention is paid to freezing and accelerated walking of PD patients. This is a symptom in which the first step at the start of walking is difficult to occur and a symptom in which walking gradually accelerates after the start of walking.

3.1 Walking measurement In walking measurement, ankle trajectory information obtained from a group of acceleration and angular velocity sensors attached to the ankle, knee, waist, etc. is mainly used. We pay attention to the kinematic characteristics of the trajectory, the stride and the walking cycle, their left-right asymmetry and temporal variation.

The method of estimating the trajectory can be divided into two stages: a stage in which continuous walking data is divided for each period, and a stage in which the trajectory is estimated in each period. Details of each are shown below.

FIG. 19 is a diagram showing the angular velocity data of the Z axis. Since walking is a periodic motion, the acceleration and angular velocity data acquired from the sensor group shows a periodic pattern. Therefore, the measured data is divided into one period. It is desirable that the dividing point be in a stable state where the foot is grounded and the angular velocity is close to zero. This is because it is easy to assume an initial value at the time of integration. By dividing the walking for each cycle, it is possible to reduce an accumulated error when integrating acceleration and angular velocity. Furthermore, the feature quantity of each cycle can be extracted efficiently.

The trajectory is estimated for each of the divided periods. FIG. 20 is a flowchart of trajectory estimation. _First , an initial value θ _z0 of the angle of each axis is determined (S100). For example, the initial value θ _z0 can be determined based on the equation (4).
θ _z0 = tan ⁻¹ (^ a _y / − ^ a _x ) (4)

^ Indicates moving average. The moving average section is a plurality of points before and after the start point of each cycle, and can be, for example, 5 before and after, for a total of 11 points. Thereby, the steady component, that is, the gravity component can be taken out, and the initial angles of the Y axis and the Z axis can be estimated. For the X-axis, the initial angle is assumed to be zero and will be corrected later.

To find the trajectory, it is not enough to double-integrate acceleration on each axis. This is because the legs are accompanied by a rotational movement and the posture is always changed. Therefore, first, the attitude of the sensor is estimated by integrating the angular velocities ω (i) of the respective axes as in equation (5) (S102). At this time, the initial value of the angle at the time of integration is obtained by the equation (4).
θ _i = θ _i−1 + ω (i) × Δt (5)

Subsequently, the sensor posture T is estimated based on the angle of each axis (S104). The posture T is represented by a three-column matrix with the x-axis, y-axis, and z-axis as columns. Next, the acceleration a at each time i is decomposed into a traveling direction α ₁ , a vertical direction α ₂ , and a side surface direction α ₃ by matrix calculation using the estimated sensor posture T (S 106).

Then, using the equations (6) and (7), the position is obtained by double integration with respect to time in each direction (S108).
v _i = v _i−1 + αi × Δt (6)
p _i = p _i−1 + v _i × Δt (7)

It is necessary to set the initial value prior to integration. Therefore, the initial value of the velocity in each direction is assumed to be zero, and the starting point of each cycle is set as the origin for the position. Here, in the stable state at the time of ground contact, not only in the vertical and horizontal directions but also in the longitudinal direction, it is sufficiently smaller than the swinging motion of the foot, so it is approximated to zero. Based on this, double integration of equations (6) and (7) is performed for each direction to estimate the ankle trajectory during walking.

(Here, the accumulation of noise due to integration must be considered.) Therefore, when integrating the angle, velocity, and position, the weight corresponding to the distance from the start point / end point is calculated for the two waveforms obtained by integrating from both the start point and end point of each period (8) , (9), and a weighted average is taken according to equation (10). However, i indicates the time in each period (that is, what sampling point), and indicates the total number of samples in each period (that is, how many times Δt is the period).

In the present embodiment, the parameter m is preferably about 0.1. Also, the initial value must be set when integrating from the reverse direction (direction to return the time axis). Therefore, the velocity and position are similarly set to 0, and the initial value of the angle is the same as the angle at the start point of the next cycle.
w ₁ = 1−w ₂ (8)
w ₂ = 1 / {1 + exp {−m (i−T / 2)}} (9)
_{V = w1 × V fwrd + w2} × V back ... (10)

That is, using the fact that the error at the boundary of each cycle, that is, the start point and the end point is small, the integration in the direction in which the time is advanced from the start point of the cycle and the integration in the direction in which the time is returned from the end point of the cycle is the coefficient w _1. , W ₂ and adding them, the influence of errors can be reduced.

Finally, the error assuming that the initial value of the X-axis angle is 0 must be corrected. FIG. 21A is a diagram for explaining correction of the X-axis angle. If φ is inclined in the initial posture, a straight line connecting the origin and the end point is inclined from the traveling direction as shown in FIG. Accordingly, the position in the side surface direction is obtained without taking the above-described cumulative error countermeasure, and the trajectory is rotated and corrected so that the end point coincides with the traveling direction in the side surface-traveling direction plane (S110). The rotation of the trajectory can be performed by matrix calculation. FIG. 21B is a diagram showing an ankle three-dimensional trajectory (for one cycle) obtained from acceleration and angular velocity data during walking by the estimation method according to the embodiment.

3.2 Features of walking Next, features of walking will be described. Various features can be considered, but here, as an example, attention is paid to the feature amounts in the front-rear direction, the left-right direction, and the up-down direction of the ankle trajectory estimated by the above method. Specifically, the front-rear direction is the stride, the left-right direction is the swing width, and the up-down direction is the foot lift amount. Paying attention to these mean values (primary statistics) and variance (secondary statistics), PD patients have the characteristics that the mean value is small in both stride and lift, and the variance (trajectory fluctuation) is large. It was done. In normal subjects, the opposite was true, and the average value was large and the variance was small for both stride and lift. This means that the feature amount is effective in separating both groups. This is an important finding for automatic diagnosis of the stage of PD patients.

There are some other features that can be considered. Although the above measurement is an example of a statistical feature amount related to an ankle trajectory, further feature amounts can be defined by extending to a spatial correlation or a temporal correlation.

For example, spatial characteristics can be taken into account by measuring at multiple points instead of just one ankle. PD patients are known to exhibit unilateral parkinsonism at Hoehn-Yahr classification once and bilateral parkinsonism at twice. Therefore, it is possible to distinguish between the right half and the left half by spatially characterizing the difference. Alternatively, in addition to or instead of the ankle, the hip and knee trajectories may be measured. It is particularly useful to combine an ankle track and a waist track. Furthermore, it is possible to take into account the characteristics related to the temporal variation of the orbit. It is known that when PD patients become severe, the temporal correlation in movement fluctuation is low. Therefore, evaluation of autocorrelation function, cross-correlation function, or self-similarity is also effective.

FIG. 22 is a diagram illustrating an ankle trajectory when the horizontal axis indicates the front-rear direction and the vertical axis indicates the height direction. Regarding the lift amount of the foot, not only the maximum point P _MAX but also the height P _LOW immediately before the foot surrounded by the frame touches the ground can be defined as the feature amount. In addition, if the height of the feature amount is low, the possibility of hitting the foot during walking increases. Therefore, it is an intuitive feature amount to present to the measurement subject.

3.3 Results of experiments related to walking The following describes experiments related to walking and the knowledge obtained from them.
(Experiment 4)
In Experiment 4, participants are classified into 4 groups.
・ 27 severe PD patients (12 men, 15 women)
・ 30 patients with mild PD (14 men and 16 women)
・ 24 healthy elderly people (12 men and 12 women)
・ 25 healthy young people (24 men, 1 woman)
Mild corresponds to 1.0 to 2.0 degrees in the modified Hoehn-Yahr classification, and severe corresponds to 2.5 to 4.0 degrees in the modified Hoehn-Yahr classification.

In Experiment 4, six feature values were extracted. FIG. 23 is a diagram for explaining a feature amount related to a walking trajectory. The horizontal axis represents the front-rear direction, the vertical axis represents the height direction, and one cycle of walking is shown. Split points (Split Points) 1, 2 and 3 indicate the heel takeoff, the vertical maximum point, and the foot swing start point, respectively.

The six feature quantities are as follows.
Feature quantity 1: Advancing direction displacement at division point 1 Feature quantity 2: Advancing direction displacement at division point 2 Feature quantity 3: Advancing direction displacement at dividing point 3 Feature quantity 4: Advancing direction displacement at dividing point 4 Feature quantity 5: Displacement in the traveling direction at the dividing point 5 Feature 6: Displacement in the traveling direction at the dividing point 6

主 成分 Reduced to a two-dimensional feature space by applying principal component analysis to selected features. Principal component analysis is performed on the walking data of all participants, and the obtained first principal component and second principal component are defined as feature vectors.

In this feature space, a classifier was constructed using machine learning. In this experiment, two classifiers that classify mild PD patients and healthy elderly people, and mild PD patients and severe PD patients were constructed by SVM. Finally, 10-fold cross-validation was used as the classifier evaluation.

24 (a) and 24 (b) are diagrams showing factor loadings of the first principal component and the second principal component in the principal component analysis. The contribution ratio of each main component was 48.8% for the first main component and 30.2% for the second main component. The cumulative contribution rate is 79%.

Referring to FIG. 24 (a), the first principal component tended to have a large factor loading of feature amounts 4-6. On the other hand, as can be seen from FIG. 24B, the second principal component is contributed by the feature quantities 1 to 3. It can be said that the first principal component is the amount of the traveling direction component of the walking trajectory, and the second principal component is the amount of the vertical direction component. In addition, since the cumulative contribution rate is 79.0%, it can be said that the six feature amounts obtained from the walking trajectory can be sufficiently reduced to the two-dimensional feature space.

FIG. 25 is a diagram showing a feature space in which the walking state of each participant is plotted. The horizontal axis indicates the first main component, and the vertical axis indicates the second main component.

FIG. 26 (a) shows the results of applying SVM to the mild PD group and the healthy elderly group. The black solid line is the classification boundary. The accuracy of each classifier by 10-fold cross validation was 92.6%. FIG. 26B is a diagram showing the results of performing SVM on the mild PD group and the severe PD group. The accuracy of each classifier by 10-fold cross validation was 76.8%.

Klucken et al. Classified Hoehn-Yahr severity (H & Y) I (1 degree) and healthy subjects, H & Y II (1 degree) and healthy subjects, and showed that the accuracy was 70% and 86%, respectively. (Klucken, et al. "Unbiased and mobile gait analysis detects motor impairment in Parkinson's disease," PloS One, vol.8, no.2, e56956 (2013)). Compared with this result, it can be said that the accuracy of this system is high.

The automatic diagnosis apparatus 1 according to the embodiment is a simple system in which a sensor is simply attached to an ankle, and is a simple method of measuring walking. The fact that the method was able to classify with high accuracy is evidence that the walking trajectory is effective for PD diagnosis support. The classification accuracy of the mild PD group and the severe PD group was 76.8%. This is because the defined feature space is not appropriate for capturing the PD posture reflex disturbance, and there is room for improvement.

Furthermore, by reducing the feature space to two dimensions, it becomes possible to visualize gait changes. The automatic diagnosis apparatus 1 according to the embodiment uses a small sensor to realize measurement that does not limit the environment. Therefore, it is expected to be used in everyday environments such as homes. At this time, it is assumed that the user himself / herself uses it in the absence of an expert, and the analysis result must be fed back in an easy-to-understand form. Therefore, it is considered that visual information such as figures is intuitive and easy to understand, not numerical values such as feature quantities and indices, so that the user can grasp his / her walking state.

So far, the motion measured by the motion measuring unit 10 and the feature quantities generated by the feature extracting unit 20 have been described in detail based on several examples. Next, diagnosis by the interpreter 30 based on these feature amounts (feature vectors) will be described.

In the interpreter 30, feature vectors are constructed based on several feature quantities.
In constructing a feature vector, generally, the more feature quantity, the higher the classification accuracy, but the calculation cost becomes high. It can be considered that the higher the correlation between different feature amounts, the greater the amount of information when combining them. Therefore, feature quantities are selected and reconfigured as necessary using principal component analysis. This makes it possible to reduce the calculation cost without greatly reducing the classification accuracy. In addition, it is possible to improve performance such as diagnostic accuracy by constructing an appropriate feature vector according to the target disease or application problem.

FIG. 27 is a diagram showing an example of distributions on the feature vectors of the young healthy person and the Hoehn-Yahr classification once and twice. Here, a combination of the variance of the inclination of the left and right direction of the back and the variance of the inclination of the back and forth direction of the back is selected as the feature vector. It can be seen that Hoehn-Yahr classification can be automatically diagnosed based on the feature vector by generating an appropriate hyperplane (indicated by a solid line in FIG. 27) for dividing each group by machine learning.

Regarding machine learning, the database 22 is constructed by collecting measurement data of young healthy persons, healthy elderly persons, and affected patients. Machine learning is performed based on the information. As a result, it is possible to construct a discriminator for separating the presence or absence and severity of a disease. By using this discriminator, it becomes possible to classify the measurement data of a target whose disease presence or severity is not clear, and automatic diagnosis can be realized. FIG. 28 is a diagram illustrating an example of a classifier that classifies young healthy individuals and PD patients constructed using SVM based on fingertip vibration measurement data.

早期 Early detection of PD symptoms is particularly important from the viewpoint of treatment. Since the automatic diagnosis apparatus 1 according to the embodiment can distinguish a healthy person and a mild PD patient with high accuracy, it greatly contributes to the treatment of a PD disease patient.

According to the automatic diagnosis apparatus 1 according to the embodiment, it is possible to realize an automatic diagnosis system of severity with PD disease as an example. This system can be applied not only to automatic diagnosis but also to other applications. One of them is drug efficacy evaluation using quantitative evaluation of symptoms. By measuring and analyzing using this system before and after taking medicine, it is possible to confirm how effective the medicine is for which symptom. Moreover, by using this system on a daily basis, it becomes possible to grasp whether the patient has sustained drug efficacy, and it is possible to suggest the timing of medication by this system.

“Walking disorders and tremors are symptoms that are likely to occur in the early stage (mild) of PD disease. Therefore, it is possible to separate healthy subjects from mild patients by gait analysis and seismic analysis, and severe patient analysis by posture analysis. Therefore, it can be said that walking, trembling, and posture are in a complementary relationship. Therefore, by combining them well, it is possible to more accurately determine the severity of PD disease.

In addition, although the diagnosis of PD disease has been described in the embodiment, the present invention is also effective for early diagnosis of a disease that progresses gradually such as a neurodegenerative disease, and is applied to the diagnosis of dementia as described below. Is also expected.
・ Alzheimer type dementia ・ Lewy body type dementia ・ Cerebrovascular type dementia ・ Normal pressure hydrocephalus type dementia

Furthermore, the present invention can also be used for applications that evaluate the degree of improvement in the rehabilitation process. Specifically, the following are exemplified.
Rehabilitation of movement disorders due to hemiplegia of stroke Rehabilitation of movement disorders due to orthopedic diseases such as osteoarthritis

Although the present invention has been described using specific terms based on the embodiments, the embodiments only illustrate the principles and applications of the present invention, and the embodiments are defined in the claims. Many variations and modifications of the arrangement are permitted without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 1 ... Automatic diagnostic apparatus, 2 ... Patient, 10 ... Movement measurement part, 12 ... Sensor, 14 ... Marker, 20 ... Feature extraction part, 30 ... Interpreter, 32 ... Database, S1 ... Movement, S2 ... Measurement data.

The present invention can be used for pathological diagnosis of cranial nerve diseases.

Claims (12)

An automatic diagnostic device for cranial nerve diseases,
A motion measurement unit that includes a first sensor attached to a patient, and that measures an ankle trajectory of the patient based on an output of the first sensor;
A feature extraction unit that extracts a feature amount of the ankle trajectory based on first measurement data relating to the ankle trajectory from the motion measurement unit;
An interpreter for generating diagnostic data indicating a diagnostic result of the cranial nerve disease based on a characteristic amount of the ankle trajectory;
An automatic diagnostic apparatus comprising: 2. The automatic diagnosis apparatus according to claim 1, wherein the characteristic amount of the ankle trajectory is based on at least one of a step length, a swing width, and a lift amount of the foot. The automatic diagnostic apparatus according to claim 1 or 2, wherein the interpreter generates the diagnostic data based on a feature vector that is a combination of a plurality of ankle trajectory feature quantities. The said interpreter refers to the database which accumulated the feature-value of the said ankle track | orbit obtained with respect to the some patient, The said diagnostic data is produced | generated based on the said database. The automatic diagnostic apparatus in any one. The motion measurement unit further includes a second sensor attached to the patient, and measures a motion that is at least one of posture or vibration of the patient based on an output of the second sensor,
The feature extraction unit extracts a feature quantity of the motion based on second measurement data related to the motion from the motion measurement unit,
5. The automatic diagnosis apparatus according to claim 1, wherein the interpreter generates the diagnosis data based on the feature amount of the motion in addition to the feature amount of the ankle trajectory. 6. The exercise includes the posture of the patient;
The automatic diagnosis apparatus according to claim 5, wherein the feature amount includes at least one of an average value, variance, skewness, and kurtosis of the patient's inclination. The movement includes vibration of the patient;
The automatic diagnosis apparatus according to claim 5, wherein the feature amount is power in a specific frequency band of the vibration. An automatic diagnostic device for cranial nerve diseases,
A motion measuring unit including a sensor attached to a patient, and measuring a motion that is at least one of posture and vibration of the patient based on an output of the sensor;
Based on the measurement data from the motion measurement unit, a feature extraction unit that extracts the feature quantity of the motion;
An interpreter for generating diagnostic data indicating a diagnostic result of the cranial nerve disease based on the feature amount;
An automatic diagnostic apparatus comprising: The exercise includes the posture of the patient;
The automatic diagnosis apparatus according to claim 8, wherein the feature amount includes at least one of an average value, variance, skewness, and kurtosis of the patient's inclination. The movement includes vibration of the patient;
The automatic diagnosis apparatus according to claim 8 or 9, wherein the feature amount is power in a specific frequency band of the vibration. 11. The automatic diagnosis apparatus according to claim 8, wherein the exercise includes walking of the patient in addition to at least one of posture and vibration of the patient. The movement includes a plurality of movements measured at different sites or conditions with respect to the patient;
The automatic diagnosis apparatus according to claim 8, wherein the feature amount is information regarding a plurality of motion differences.

Images