JP2011177278A - Gait disorder automatic analysis system - Google Patents

Abstract

The present invention provides an automatic analysis system for gait disturbance that can be easily measured and quantitatively evaluate the degree of gait disturbance in a clinical setting.
A walking obstacle automatic analysis system according to the present invention includes a walking motion measuring unit 3 that measures a walking motion of a subject 2 by motion sensors 31 and 32, and a waist trajectory from the walking motion measured by the walking motion measuring unit 3. The desired walking motion tracking unit 4, the walking obstacle feature extracting unit 5 that extracts the walking obstacle feature amount as the feature amount related to the walking disorder from the waist trajectory obtained by the walking motion tracking unit 4, and the walking obstacle feature extracting unit 5 Based on the extracted walking disorder feature amount, the walking disorder degree determination unit 6 that determines the walking disorder degree of the subject 2 and the walking disorder degree of the subject 2 whose walking disorder degree is known are associated with the walking disorder feature amount. A walking obstacle learning unit 7 that causes the walking obstacle degree determination unit 6 to learn as learning data, and the walking obstacle degree determination unit 6 makes a determination with reference to the learning data.
[Selection] Figure 1

Description

  The present invention relates to a gait disorder automatic analysis system. More specifically, the present invention relates to a walking disorder automatic analysis system that extracts a walking disorder feature amount from a waist trajectory of a subject and automatically analyzes a walking disorder degree based on the extracted walking disorder feature quantity.

  With the progress of an aging society, the number of physically disabled people is steadily increasing. Psychosomatic dysfunction due to hemiplegia or the like hinders activities of daily living (Activities of Daily Living: ADL), resulting in emotional problems and social disadvantage of the patient. Therefore, supporting the patient's independence through rehabilitation (hereinafter referred to as rehabilitation) improves the quality of life (QOL) of the patient and the surrounding people and realizes a healthy social life. It is an essential factor above.

  Walking rehabilitation is carried out in accordance with walking evaluation by doctors and physical therapists and rehabilitation programs based on it. However, gait evaluation in the clinical field is often performed by subjective evaluation by the examiner's visual observation, and there is a problem that it depends on the examiner's knowledge and experience and lacks reproducibility and consistency of evaluation. Therefore, methods using an electric angle meter, a floor reaction force meter, a three-dimensional motion analysis device, an electromyogram, etc. have been proposed as a method for objectively and quantitatively evaluating walking. However, due to limitations such as cost, time, and difficulty in using measuring instruments, the current situation is that they are not so popular in clinical practice. In the clinical field, there is a need for a system that can not only quantitatively evaluate gait but also easily measure.

  The inventors have developed a method to calculate the amount of displacement of the waist during walking using information obtained from a three-dimensional acceleration sensor attached to the waist so that measurement can be easily performed, and track the time evolution. Attempts have been made to apply the visualized waist trajectory to walking evaluation. The advantage of such a measurement method is that it is possible to measure kinematic information during walking with relatively high accuracy while being compact and lightweight and simple in measurement compared to conventional measurement devices. Another advantage is that there is no need to limit the measurement location. (For example, see Patent Document 1)

JP 2006-102156 A (paragraphs 0031 to 0112, FIGS. 1 to 15)

  However, the evaluation method regarding the degree of gait disturbance in the waist track has not been sufficiently studied, and its evaluation index and evaluation method have not been established. If the evaluation index and evaluation method are established, there is a possibility that it can be applied to clinical introduction and eventually to rehabilitation at home.

  It is an object of the present invention to provide an automatic gait disorder analysis system that can easily measure and quantitatively evaluate the degree of gait disturbance in a clinical setting.

  In order to solve the above-described problem, the walking disorder automatic analysis system according to the first aspect of the present invention is a walking that measures the walking motion of the subject 2 using motion sensors 31 and 32 as shown in FIGS. 1 and 2, for example. The motion measuring unit 3, the walking motion tracking unit 4 for obtaining the waist trajectory from the walking motion measured by the walking motion measuring unit 3, and the walking as the feature amount related to the walking disorder from the waist trajectory obtained by the walking motion tracking unit 4 A walking disorder feature extraction unit 5 that extracts a disorder feature, a walking disorder degree determination unit 6 that determines the degree of walking disorder of the subject 2 based on the walking disorder feature amount extracted by the walking disorder feature extraction unit 5, and a walking A gait obstacle learning unit 7 that causes the gait obstacle degree determination unit 6 to learn the walking obstacle degree of the subject 2 whose degree of obstacle is known as learning data in association with the gait obstacle feature amount, See training data A determination is made as Te.

  Here, the motion sensor includes an acceleration sensor for detecting the position of the waist and its movement, a position sensor (which can be seen by changing the position with time), a foot sensor for detecting landing and takeoff of the leg. . Further, according to the feature amount to be acquired, for example, both an acceleration sensor and a foot sensor may be used, or one of them may be used. In addition, the extracted gait disturbance feature is a vector that characterizes gait disturbance and can be either a geometric feature or a temporal feature as long as it can be extracted from the waist trajectory. A high value is desirable. In addition, the number of extracted gait disturbance feature quantities changes according to the type of obstacle, the correlation between the degree of gait disturbance and the gait disorder feature quantity. That is, one is sufficient when the correlation is large, and it is desirable to use a plurality when the correlation is small. Moreover, when using two or more, it is preferable that mutual correlation is small and independence is high. In addition, it is desirable to use an objective standard established as the degree of gait disturbance, but usually, if hemiplegia, Brunnstrom Stage (BS), which is an indicator of the degree of gait disturbance, is Parkinson's disease. The Yahr index is used. The gait obstacle learning unit 7 has a function as a database for the gait obstacle degree determination unit 6 to determine the gait obstacle degree. The gait obstacle degree determination unit 6 uses, for example, a pattern such as a support vector machine. This can be determined by referring to a database using a classifier or principal component analysis.

  When configured as in this aspect, an acceleration sensor is used as the motion sensor, and the system is put into clinical practice by using the computing function of a personal computer for tracking of the lower back trajectory, extraction of gait disturbance features, and determination of the degree of gait disturbance. It can be easily brought in and can be measured easily. In addition, since a determination value having a high correlation with the degree of gait disturbance can be obtained, the degree of gait disturbance can be quantitatively evaluated at the clinical site. Therefore, it is possible to provide an automatic gait disorder analysis system that can easily measure and quantitatively evaluate the degree of gait disturbance in the clinical field.

Moreover, the gait disturbance automatic analysis system 1 according to the second aspect of the present invention is the gait disturbance degree determination unit 6 in the first aspect, for example, as shown in FIG. The walking disorder degree of the subject 2 is determined using a pattern classifier to be classified.
Here, multidimensional means two or more dimensions. In addition, it is preferable to use a support vector machine as a pattern classifier because high classification accuracy is obtained, but a linear classifier, quadratic classifier, k approximation method, boosting, decision tree, neural network, Bayesian network, hidden Markov model Etc. may be used. If comprised like this aspect, since the walking disorder degree determination part 6 classify | categorizes a pattern statistically using a pattern classifier, it can determine a walking disorder degree with high precision.

In addition, in the gait disorder automatic analysis system according to the third aspect of the present invention, in the first aspect, for example, as shown in FIG. Determine the degree.
If comprised in this way, it will become possible to detect the time-dependent change of a walking disorder degree with respect to the test subject of the same BS by selecting a walking disorder feature-value appropriately and using a principal component analysis.

In addition, in the gait disturbance automatic analysis system according to the fourth aspect of the present invention, as shown in FIGS. 7 to 11, for example, as shown in FIGS. Geometric features related to left-right asymmetry are used.
With this configuration, the geometric feature amount related to the left-right asymmetry of the hip movement has a high correlation with, for example, the hemiplegic walking disorder degree. A determination can be obtained.

In addition, in any one of the first to third aspects, the gait disturbance automatic analysis system according to the fifth aspect of the present invention is characterized by the temporal characteristics of the waist movement as a gait disorder feature amount, for example, as shown in FIG. Use quantity.
With this configuration, for example, the walking period gradient of the temporal feature amount of hip movement has a high correlation with the degree of walking disorder of Parkinson's disease. Obtainable.

Further, in the gait disorder automatic analysis system according to the sixth aspect of the present invention, in any of the first to fifth aspects, for example, as shown in FIG. 2, the motion sensor is a three-dimensional mounted near the lumbar vertebra. This is an acceleration sensor 32.
If comprised in this way, a three-dimensional acceleration sensor is small and lightweight, can perform a measurement easily, and can provide the gait disorder automatic analysis system which can be easily brought into a clinical field.

  According to the present invention, it is possible to provide an automatic gait disorder analysis system that can easily measure and quantitatively evaluate the degree of gait disturbance in the clinical field.

It is a figure which shows the structural example of the gait disorder automatic analysis system in Example 1. FIG. It is a figure which shows the structural example of a walking movement measuring part. It is a figure which shows the movement amount calculation result of the left-right direction. It is a figure for demonstrating waist track data. It is a figure which shows the example of a healthy person's waist | orbit track data. It is a figure which shows the example of the waist track data of a hemiplegic patient. It is a figure which compares and shows the example of the waist track | orbit when a front face is observed from back about a healthy person and a hemiplegic patient. It is a figure for demonstrating the up-and-down difference of a load response period. It is a figure for demonstrating the asymmetry of the raising width of a waist. It is a figure for demonstrating the asymmetry of the raising width of a waist. It is a figure for demonstrating the asymmetry of the left-right amplitude. It is a figure for demonstrating the amplitude of the left-right direction. It is a figure which illustrates the relationship between each gait feature amount and BS. FIG. 3 is a diagram illustrating an overview of an SVM algorithm according to the first embodiment. It is a figure which shows the example of a processing flow of the walking disorder automatic analysis in Example 1. FIG. It is a figure which illustrates the time-dependent change for every gait disorder feature-value. It is a figure which illustrates the relationship between a 1st main component and BS. It is the figure which plotted the time-dependent change of the 1st main component for every patient. It is a figure for explaining relaxation of acceleration walking.

  Embodiments of the present invention will be described below with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals, and redundant description is omitted.

  In the first embodiment, an example will be described in which five walking obstacle feature amounts are extracted from the waist trajectory and the walking obstacle degree of the subject is determined using a support vector machine.

[Device configuration]
In FIG. 1, the structural example of the gait disorder automatic analysis system 1 by Example 1 is shown. In FIG. 1, the walking disorder automatic analysis system 1 includes a walking movement measuring unit 3, a walking movement tracking unit 4, a walking disorder feature extracting unit 5, a walking disorder degree determining unit 6, and a control unit 8. The walking motion measuring unit 3 measures the walking motion of the subject 2 using a motion sensor. The walking motion tracking unit 4 obtains the waist trajectory from the walking motion measured by the walking motion measuring unit 3. The walking obstacle feature extraction unit 5 extracts a walking obstacle feature amount from the waist trajectory obtained by the walking movement tracking unit 4. The walking disorder degree determination unit 6 determines the walking disorder degree of the subject 2 based on the walking disorder feature amount extracted by the walking disorder feature extraction unit 5. The gait obstacle learning unit 7 causes the gait obstacle degree determination unit 6 to learn as learning data by associating the gait obstacle degree of the subject 2 whose gait obstacle degree is known with the gait disorder feature amount. The walking disorder degree determination unit 6 determines the walking disorder degree with reference to the learning data. The control unit 8 controls the gait disorder automatic analysis system 1 and the flow of signals and data of each part thereof, so that the function of the gait disorder automatic analysis system 1 can be realized. The walking movement tracking unit 4, the walking obstacle feature extraction unit 5, the walking obstacle degree determination unit 6, and the control unit 8 can be realized by a personal computer (PC) 10 and are configured in the PC 10.

[Walking motion measurement unit]
The walking motion measuring unit 3 measures the walking motion of the subject 2 using a motion sensor. Walking measurement is performed by a foot contact timing detection device 31 that detects the ground contact and takeoff timing, and a waist trajectory measurement device 32 that measures the motion of the waist. Analysis from the side is possible. The two measuring devices are synchronized with each other.

FIG. 2 shows a configuration example of the walking motion measuring unit 3. The foot contact timing detection device 31 uses a foot sensor to detect the timing of ground contact and takeoff during walking. The foot sensor incorporates a tape-shaped pressure sensor. When a certain pressure (for example, 220 g / cm ² ) is applied to a part of the bottom of the foot, it is grounded, and the timing when the pressure is lowered to the certain pressure is taken off. Detect as. The waist trajectory measuring device 32 measures the spatial displacement of the waist using a three-dimensional acceleration sensor. The waist trajectory measuring device 32 measures the acceleration change of the waist by fixing the three-dimensional acceleration sensor to the waist using a band. In order to accurately capture the displacement of the lumbar region, it was worn near the lumbar vertebra, where there was little influence of body rotation. The sampling frequency is 100 Hz, for example. The measured acceleration information and leg contact information are transmitted to the walking movement tracking unit 4 in the personal computer 10 by wireless communication. Since the walking movement measuring unit 3 is small and light and uses wireless communication, the walking place is not limited, and continuous measurement over a long distance is also possible.

[Walking movement tracking unit]
The walking motion tracking unit 4 obtains the waist trajectory from the walking motion measured by the walking motion measuring unit 3. The acceleration information and the leg contact information measured by the walking motion measuring unit 3 are received, the walking motion is analyzed, and the waist trajectory is obtained. By integrating the acceleration information twice, the spatial displacement (vertical, horizontal, longitudinal) of the waist during walking is calculated.

  First, a walking trajectory calculation method will be described (see Patent Document 1). As the spatial coordinate system of the sensor attached to the waist, the acceleration in the three-dimensional direction is integrated with X in the left-right direction (positive on the left side), Y in the vertical direction, and Z in the traveling direction to obtain velocity and position information. However, in obtaining the velocity V from the acceleration A, an offset error due to grounding accumulates with a simple integration such as (Equation 1). Therefore, as a proposed method, regarding the walking motion in the vertical and left-right directions, it is assumed that the swing motion is periodic by exchanging both legs, and the baseline that is the center of the trajectory is calculated from the walking cycle, and the displacement from the measured value is calculated. Find the trajectory by integrating.

  Vertical direction: When integrating acceleration information, factors that cause errors include the effect of gravitational acceleration and impact when the legs touch the ground, and this particularly affects the error in the vertical direction. However, since the walking motion here is limited to the horizontal plane, when attention is paid to the vertical direction, the walking motion should always have the same waist height at the leg ground contact. Therefore, as shown in (Equation 2), the average value of the velocity Vy is obtained from a short-term time scale of 1 second before and after walking at a certain time point, and the displacement from the average value is used as a baseline serving as a zero point of walking motion. The position Y from which the error is eliminated is obtained by integrating the minute as the velocity V′y. The same calculation is performed for position integration to obtain Y '.

Left and right direction: There is a deviation of the walking trajectory with respect to the set course in the left and right direction with respect to the vertical direction where the waist height is always equal when the legs touch the ground. Therefore, it is necessary to separately evaluate the change in walking trajectory for each short-term step and the deviation from the course in the left-right direction when viewed in the long term in the walking analysis in the left-right direction. The movement amount X is calculated by integrating the speed as in (Expression 3), and the speed is calculated in the same manner as in (Expression 2). The walking trajectory centered at each step uses the average X _{1 sec} of the short-term time scale (1 second before and after) (Equation 4), and the deviation from the course is the average X of the long-term time scale (5 seconds before and after). _It is calculated from (Equation 5) using 5 _sec .

FIG. 3 shows the calculation result of the movement amount in the left-right direction. The vertical axis indicates the amount of horizontal movement, and the horizontal axis indicates time. Movement locus (solid line), the average _{X 1 sec} short-term time scale (about one second) (dashed line), the average _{X 5 sec} long-term time scale (longitudinal 5 seconds) (dashed line) is shown.

  Advancing direction: Regarding the advancing direction, first, as in the vertical and left-right directions, the velocity V′z from which the offset is removed is obtained using the average of the velocity Vz. However, as a difference from the vertical and left-right directions, the traveling direction is determined as the movement distance Z. Therefore, in the method of (equation 2) that removes the constant velocity component assuming that the movement is periodic, V′z is during walking. Only the velocity amplitude component of Therefore, the following walking speed estimation method is used.

  Since the speed amplitude and walking speed can be approximated by a regression line, the average walking speed on a short-time scale is obtained by multiplying the amplitude of V′z (V′zmax) by a constant α as shown in (Equation 6). Is added to the velocity amplitude component V′z to obtain the walking speed V ″ z. The movement distance Z is obtained by integrating V ″ z again. The relationship between amplitude and speed varies from individual to individual, and the value of α needs to be calculated and set for each individual from the actual walking distance and time.

In this way, a walking trajectory in three dimensions can be obtained by using a calculation method corresponding to each movement in the three-dimensional direction.

FIG. 4 is a diagram for explaining waist trajectory data. FIG. 4A shows a waist trajectory when the frontal surface of a healthy person is observed from the rear, with the vertical axis representing the vertical direction and the horizontal axis representing the horizontal displacement. FIG. 4B shows a method of dividing one walking cycle into four phases, focusing on the difference in the role of the lower limbs in walking motion. FIG. 4C shows a three-dimensional relationship between the walking motion and the waist locus. The markers in FIGS. 4A to 4C correspond to the following information, respectively. FIG. 4 (d) shows the reference numerals.
● Left foot contact (LIC)
□ Right Initial Swing (RISw)
■ Right foot contact (RIC)
○ Left foot swing (LISW)

The walking cycle is divided using these four ground contact and takeoff information. Specifically, there are the following four phases.
Phase 1: Load Response Period from LIC to RISW (Loading Response : LRL)
Phase 2: Left single leg support period from RISw to RIC (Left Single Limb Support: LSLS)
Phase 2: The right leg at the time of LSLS is the swing phase (Swing: the time when the foot is away from the ground and the leg is carried forward by swinging, RSw)
Phase 3: Load response period (LRR) from RIC to LISW
Phase 4: Right single leg support period from LISW to LIC (Right Single Limb Support: RSLS)
Phase 4: Left leg during RSLS is the swing phase (LSw)

  FIG. 5 shows an example of waist trajectory data of a healthy person. FIG. 5A is a graph in which the vertical axis is the displacement in the vertical direction and the time variation is plotted. FIG. 5B is a graph in which the vertical axis is the displacement in the left-right direction and the time variation is plotted. Positive is the displacement in the right direction of the body, and negative is the displacement in the left direction. FIG. 5C shows a waist trajectory when the frontal plane is observed from the rear with the vertical axis as the vertical direction and the horizontal axis as the horizontal displacement.

  The third lumbar vertebra of a healthy person who is walking draws a locus of a sine curve twice in the vertical direction and once in the left-right direction during one walking cycle. Also, in the vertical direction, the highest point (the first stage of FIG. 5 (a)) is the middle point of stance (Mid Stance: MSt, included in the single leg support period, LMSt is the middle stage of the left leg, and RMSt is the middle stage of the right leg). , 3 and 5), and passes through the lowest point (the 2nd and 4th circles in FIG. 5A) in the load response period LR (LRL and LRR). The amount of change is said to be an average of 4.5 cm. In the left-right direction, the maximum movement in each direction (first, third, fifth ● in FIG. 5B) occurs in the MSt of each of the left and right legs. It is said that the average value of the amplitude is 4.5 cm. Moreover, according to FIG.5 (c), the waist track pattern of the healthy person seen from back is bilaterally symmetrical, and has shown V shape or infinity shape. In addition, the triple ring of FIG.5 (b) represents the waist circumference from the outer side, the center axial part of a body, and a sole.

  FIG. 6 shows an example of waist trajectory data of a hemiplegic patient. FIG. 6A is a graph in which the vertical axis is the displacement in the vertical direction and the time variation is plotted. FIG. 6B is a graph in which the vertical axis is the displacement in the left-right direction and the time variation is plotted. Positive is the displacement in the right direction of the body, and negative is the displacement in the left direction. FIG. 6C shows a waist trajectory when the frontal plane is observed from the rear with the vertical axis as the vertical direction and the horizontal axis as the horizontal direction displacement. Hemiplegia is a case where a movement disorder appears in either the left or right body due to a cerebrovascular disorder or a musculoskeletal disorder. 6 (a) to 6 (c), peculiarities due to hemiplegia are seen in the movement of the waist. For example, the stride is asymmetric (the stride on the healthy support leg is long and the stride on the paralysis support leg is short). In addition, the vertical positions of LRL and LRR are different, so that the locus of the sine curve in the vertical direction and the horizontal direction is broken. Further, the waist trajectory pattern viewed from the rear is asymmetrical and shows an elliptical shape.

  Table 1 shows the attributes of the subjects. Test subject 2 was 32 hemiplegic patients and 10 healthy subjects. Subjects were classified in association with BS. Based on clinical observations of the hemiplegic recovery process, the BS moves from a relaxed state (a state in which the body cannot be moved at all) to a joint movement (for example, trying to move only the arm moves the elbow and moves the arm independently). The recovery process until the separation movement (for example, only the arm can be moved independently) occurs. I to BS. This is an index indicating the degree of gait disorder ordered in six stages up to VI, and is one of the evaluation indexes often used in clinical practice. In addition, BS is evaluated separately for the upper limbs, fingers, and lower limbs, but here, mainly refers to the lower limb BS. BS. I is the most severe and BS. BS., Which is evaluated as the lightest VI and enables joint exercise. Walking movement is often possible from about III. The subject was BS. III is 3, BS. IV is 9 people, BS. V is 10 people, BS. There were 10 VIs, and all subjects were able to walk alone without assistance. In addition, (f) in a table | surface shows a woman. The subjects were given an outline of the walking experiment in advance, and the experiment was conducted after obtaining consent. The subject was challenged to walk at a free speed on a straight 30 m walkway with the measuring device attached. However, for a patient who is difficult to walk for 30 m, the distance was set to 20 m. In addition, the use of walking sticks was restricted during walking. In the experiment, 30m walk was performed twice, the first trial was the training trial, the second trial was the main trial, and the data obtained in this trial was the subject of analysis.

  The gait analysis is performed on a steady walk excluding the transition period at the beginning and end of walking. 3 cycles at the beginning of walking and 4 cycles at the end of walking are excluded as transition periods, and the analysis range is 10 walking cycles based on LIC, and the range in which the coefficient of variation of walking cycles for 10 consecutive cycles is minimized It was.

  FIG. 7 shows a comparison of examples of waist trajectories when the front face is observed from the rear for healthy subjects and hemiplegic patients. FIGS. 7 (a) and 7 (b) show examples of hip trajectory patterns of healthy persons, and FIGS. 7 (c) to 7 (h) show examples of hip trajectory patterns of hemiplegic patients. A healthy person often shows an ∞-shaped or V-shaped shape as shown in (a), and it is said that the displacement in the vertical and horizontal directions falls within a range of about 2 cm to 2.5 cm. Further, the displacement in each of the up, down, left and right directions is substantially symmetrical. When comparing the pattern of a healthy person with the pattern of a hemiplegic patient, (h) is relatively similar to the pattern of a healthy person, but the other patterns clearly deviate from the pattern of the healthy person. From this, it is predicted that the peculiarity of the waist trajectory as a geometric feature is related to the asymmetry of the gait, which is a feature of hemiplegic walking.

[Walking disorder feature extraction unit]
The walking obstacle feature extraction unit 5 extracts a walking obstacle feature amount from the waist trajectory obtained by the walking movement tracking unit 4. Pay attention to five feature values (three feature values in the vertical direction and two feature values in the horizontal direction) to quantitatively indicate the peculiarities of the waist trajectory from the hip trajectory pattern of hemiplegic walking. Defined. Next, the walking obstacle feature amount will be described.

[(1) Vertical difference in load response period (LRdif)]
FIG. 8 is a diagram for explaining the vertical difference in the load response period. Fig. 8 (a) shows the waist trajectory captured from the dorsal frontal face of a healthy person and a left hemiplegic patient, and Figs. 8 (b) and 8 (c) show the time fluctuations in the vertical direction of the healthy person and left hemiplegic patient. FIG. 8 (d) shows the definition formula of the vertical difference (LRdif) in the load response period, and FIG. 8 (e) shows the evaluation method. Here, attention is focused on the lowest points of LRL and LRR in FIGS. 8B and 8C. LR is a phase in which the waist passes through the lowest position during walking movement. In a healthy person, as shown in FIG. 8B, LRL and LRR pass through almost the same height, so the difference between them is almost zero. However, in the case of a left hemiplegic patient, as shown in FIG. 8C, the LRR passes through a position higher than the LRL, and a vertical difference is generated. This is predicted to be related to the asymmetry of the stride that is characteristic of hemiplegic walking. It can be intuitively understood that if the stride becomes wider, the lowest point through which the center of gravity passes is lowered, and conversely, if the stride becomes narrower, it becomes higher. Therefore, it is considered that a difference occurs in the heights of LRR and LRL because the stride becomes asymmetric due to hemiplegia. The cause of this asymmetry in stride is that the muscle strength on the paralyzed side, the lowering of supportability due to spasticity, or the excessive extension of the knee, which is a compensatory exercise to compensate for it, occurs normally during support on the paralyzed side. This is because the side cannot be swung out sufficiently. In walking, the force of dropping the center of gravity is converted into propulsive force by interlocking the ankle, knee, and hip joints with the accompanying muscles. It is thought that.

Therefore, the vertical difference in the LR is regarded as an index characterizing hemiplegic walking, and the vertical difference (LRdif) in the load response period is defined as shown in (Expression 7) (see FIG. 8D).

LRdif = min (LRL) −min (LRR) (Expression 7)

Note that min () in (Expression 7) indicates the lowest point (minimum displacement amount) in LRL or LRR. LRdif is evaluated to be closer to a normal pattern as the value is closer to zero.

[(2) Waist lift width asymmetry (VUAsym)]
FIG. 9 is a diagram for explaining the asymmetry of the lifting width of the waist. Fig. 9 (a) shows the waist trajectory captured from the dorsal frontal face of the healthy subject and the right hemiplegic patient, and Figs. 9 (b) and 9 (c) show the time variations in the vertical direction of the healthy subject and the right hemiplegic patient. FIG. 9D shows the definition of the asymmetry (VUasym) of the waist lifting width, and FIG. 9E shows the evaluation method. Here, attention is paid to the lifting width of the waist indicated by the upward arrows in FIGS. The left leg lift width VUL is the width (difference) from the lowest point (minimum displacement amount) to the highest point (maximum displacement amount) included after LIC to LMSt, and the right leg lift width VUR is , The width from the lowest point to the highest point included from RMS to RMSt. In FIGS. 9B and 9C, VUL and VUR are approximately the same width in healthy subjects, but in the case of a right hemiplegic patient, VUL is larger than VUR and is asymmetric to the waist lifting width. Sex has arisen. Walking is equipped with a mechanism that keeps the movement of the center of gravity to the minimum necessary in order to advance the body efficiently.In normal walking, the ankle joint, knee joint, and hip joint are dorsiflexed at the appropriate timing. By bending, the vertical fluctuation that can be up to 9.5 cm is suppressed to 2 cm to 2: 5 cm. However, in hemiplegic walking, it is predicted that excessive vertical movement will occur during the support period on the paralyzed side due to such a decrease in the adjustment function. In hemiplegic walking, when the ankle joint on the paralyzed side is excessively bent or when the knee joint cannot be bent sufficiently, the user may walk while dragging the paralyzed side. In order to avoid such dragging, hemiplegic patients may lift their hips excessively during the support phase on the healthy side. This is a kind of compensatory movement seen in hemiplegic walking such as lifting and pelvic elevation. The right hemiplegic patient shown in FIG. 9 (c) shows the latter tendency, and it is considered that the hip lifting width on the healthy side is very large.

From these characteristics, the asymmetry of the hip lifting width is regarded as an important factor in evaluating hemiplegic walking, and the asymmetry (VUasym) of the hip lifting width is expressed as shown in (Equation 8). Define (see FIG. 9D).

VUasym = (VUL−VUR) / (VUL + VUR) (Equation 8)

VUasym indicates that the closer the value is to 0, the closer it is to a healthy person's walking pattern.

[(3) Asymmetry of lowering width of waist (VDasym)]
FIG. 10 is a diagram for explaining the asymmetry of the lowering width of the waist. Fig. 10 (a) shows the waist trajectory captured from the dorsal frontal face of the healthy subject and the left hemiplegic patient, and Figs. 10 (b) and 10 (c) show the time variations in the vertical direction of the healthy subject and the left hemiplegic patient. FIG. 10 (d) shows the definition of the asymmetry (VDasym) of the lowering width of the waist, and FIG. 10 (e) shows the evaluation method. Here, attention is focused on the lowering width of the waist indicated by the downward arrows in FIGS. Lifting width is the left leg lifting width VDL is the width (difference) from the highest point (maximum displacement amount) to the lowest point (minimum displacement amount) included in LMSt to LISw, right The leg lifting width VDR is a width from the highest point to the lowest point included from RMSt to RISw. In FIGS. 10 (b) and 10 (c), VDL and VDR are almost the same width in healthy subjects, but in the case of a left hemiplegic patient, VDR is larger than VDL. Asymmetry has occurred. Sufficiently lowering the waist is an indispensable factor for obtaining sufficient propulsive force, but the role of the ankle joint is particularly important for lifting. Appropriate dorsiflexion and flexion of the ankle joint is the starting point for converting the force of dropping the center of gravity into propulsive force, and even if the movement of the knee joint and hip joint is possible to some extent if the movement of the ankle joint is insufficient It is considered difficult to obtain sufficient driving force. In addition, recovery of motor paralysis is said to be slow as it goes to the end like a limb, and separation of the ankle joint is also BS. V and BS. It is thought to occur when paralysis such as VI is mild. From this, it is predicted that the decrease in the lifting width due to the ankle joint failure is a common factor in hemiplegic walking.

From the above points, the asymmetry of the width to lower the waist is considered to be an important factor in evaluating hemiplegic walking, and as shown in (Equation 9), the asymmetry of the lowering width of the waist (VDasym) Is defined (see FIG. 10D).

VDasym = (VDL−VDR) / (VDL + VDR) (Equation 9)

VDasym indicates that the closer the value is to 0, the closer it is to a healthy person's walking pattern.

[(4) Left-right amplitude asymmetry (Hasym)]
FIG. 11 is a diagram for explaining the left-right amplitude asymmetry. Fig. 11 (a) shows the waist trajectory captured from the dorsal frontal face of a healthy person and a right hemiplegic patient, and Figs. FIG. 11 (d) shows the definition of the left-right amplitude asymmetry (Hasym), and FIG. 11 (e) shows the evaluation method. Here, attention is paid to the amplitude in the left or right direction indicated by the arrows in FIGS. The amplitude HAL in the left direction indicates the maximum amplitude from the LIC to the left side, and the amplitude HAR in the right direction indicates the maximum amplitude from the RIC to the right side. In FIGS. 11 (b) and 11 (c), HAL and HAR are almost the same width in healthy subjects, but the HAL and HAR are asymmetric in the hip trajectory pattern of the right hemiplegic patient. It is predicted that the balance support ratio is largely biased toward the healthy side with the decline in sex. In conventional walking rehabilitation, it was inevitable that such a bias was inevitable, since walking training compensated for the function of the paralyzed side on the healthy side.

From such characteristics, the left-right amplitude asymmetry is regarded as an important factor in evaluating hemiplegic walking, and the left-right amplitude asymmetry (Hasym) is defined as shown in (Equation 10) (FIG. 11). (See (d)).

Hasym = (HAL−HAR) / (HAL + HAR) (Equation 10)

Hasym indicates that 0 is the highest symmetry, and the closer the value is to 1, the lower the symmetry.

[(5) Left-right amplitude (HA)]
FIG. 12 is a diagram for explaining the amplitude in the left-right direction. FIG. 12 (a) shows a waist trajectory captured from the dorsal frontal face of a healthy person and a left hemiplegic patient. As the paralysis becomes heavier, the left and right swing width tends to increase. FIG. 12B shows a definition formula of the amplitude (HA) in the horizontal direction, and FIG. 12C shows an evaluation method. Here, attention is paid to how much the waist swings in the left-right direction during walking in FIG. HA is defined as the movement distance from the position THL where the waist is closest to the left side to the position THR where the waist is closest to the right side. In hemiplegic patients, HA is significantly higher than that of healthy individuals, and it is said that there is a correlation between amplitude and walking speed.

The left-right amplitude (HA) is regarded as an important factor in evaluating hemiplegic walking, and the left-right amplitude (HA) is defined as shown in (Equation 11) (see FIG. 12B). ).

HA = HAL-ALR (Formula 11)

The HA evaluation standard is 4.5 cm as a standard value, and the closer to the standard value, the closer to the normal pattern.

[Extraction result of gait disturbance features]
In grouping subjects for each BS and performing comparison, pre-processing was performed as follows. In order to remove the influence of the sign of the feature value due to the difference on the paralyzed side, as a preprocessing, a representative value was calculated for each subject according to the following procedure and used in the subsequent analysis.
Process 1 Extract 10 gait cycles for analysis 2 Process 1 feature calculation for each gait feature for each gait cycle 3 Calculate median of 10 samples obtained 4 Process the absolute value of median as a representative value According to the above procedure, a total of 42 samples were obtained, one sample each from 32 hemiplegic patients and 10 healthy individuals for each feature quantity.

  FIG. 13 illustrates the relationship between each gait feature amount and the BS. BS. III-BS. VI, plots were made using a box chart for each group of 5 healthy subjects. FIG. 13A shows an example of LRdif, FIG. 13B shows Hasym, FIG. 13C shows VUasym, FIG. 13D shows VDasym, and FIG. The center line of the rectangle indicates the median value, the upper side indicates the third quartile, the lower side indicates the first quartile, and the upper side and the lower side correspond to data variations. For each gait disorder feature, a significant difference between groups was tested using a Kruskal-Wallis test and Steel-Dwass multiple comparison.

In LRdif, BS. IV-Healthy (healthy person), BS. A significant difference was found between V-Healty (p <0.05) and BS. III-BS. VI, BS. III-Healthy, BS. V-BS. There was a significant trend between VIs (p <.10).
In Hasym, BS. There is a significant difference between IV and Health (p <0.05), BS. III-BS. V, BS. III-BS. VI, BS. There was a significant trend between III and Health (p <.10.).
In VUasym, BS. IV-BS. VI (p <0.05), BS. There is a significant difference between IV-Healty (p <0.01), and BS. III-Healthy, BS. I-VBS. V, BS. There was a significant trend between V-Healthy (p <.10).
In VDasym, BS. IV-BS. VI, BS. IV-Healthy (p <0.01), BS. V-BS. VI, BS. A significant difference was observed in V-Healthy (p <0.05). III-BS. V, BS. III-BS. VI, BS. There was a significant trend between III and Health (p <.10.).
In HA, BS. V-Healthy, BS. IV-Healty (p <0.05), BS. IV-BS. There is a significant difference between VI (p <.01) and BS. III-BS. V, BS. III-BS. VI, BS. There was a significant trend between III and Health (p <.10.).
Here, p is a reference value that statistically indicates that there is a difference between two or more people being compared.

  A common feature of all gait features is BS with severe paralysis. It is the point which shows the tendency of the lowering of a right shoulder as it goes to a healthy person from III. As described above, when only one gait disturbance feature amount is associated with a BS, a relationship with the BS is suggested, but there is a case where no significant tendency is observed between the BSs.

[Walking disorder learning part]
The gait obstacle learning unit 7 causes the gait obstacle degree determination unit 6 to learn as learning data by associating the gait obstacle degree of the subject 2 whose gait obstacle degree is known with the gait disorder feature amount.
The gait disorder learning unit 7 stores the gait disorder degree BS in the database as learning data in association with the five gait disorder feature amounts extracted by the gait disorder feature extraction unit 5 for the 32 subjects (hemiplegic patients) described above. The gait disturbance degree determination unit 6 was made to learn.

[Walking disorder degree determination unit]
The walking disorder degree determination unit 6 determines the walking disorder degree of the subject 2 based on the walking disorder feature amount extracted by the walking disorder feature extraction unit 5. Further, the walking obstacle degree determination unit 6 determines the walking obstacle degree with reference to the learning data of the walking obstacle learning unit 7.
In the present embodiment, a method of configuring a multi-value classifier by combining a plurality of support vector machines (SVM) is adopted for determining the walking disorder degree BS. As the technique, the One-against-all method and the One-against-one method are common, but in this embodiment, paying attention to the order of BS, a method of constructing a multi-value classifier using a binary tree is used. Using. As an advantage, it is expected to reduce the calculation cost and increase the efficiency of learning. Here, an example will be described in which feature quantities are described as multidimensional vectors in a multidimensional space (five gait disturbance feature quantities are represented as five-dimensional vectors), and a plurality of SVMs and binary trees are used in combination as a pattern classifier. Other pattern classifiers such as a linear classifier, quadratic classifier, k approximation method, boosting, decision tree, neural network, Bayesian network, and hidden Markov model may be used. Multidimensional means two or more dimensions.

FIG. 14 shows an outline of the SVM algorithm in this embodiment. In this method, BS. III vs Others, BS. IV vs Others (excluding BS.III), BS. VvsBS. Three binary classifiers of VI (except BS.III, BS.IV) were trained.
An ordered pair of data vector x _i and corresponding label y _i

(X ₁ , y ₁ )... (X ₁ , y ₁ ) εR ^N × {−1, 1}

And Here, l is the size of the training data. The discriminator determines a hyperplane based in some sense on an appropriate optimization criterion. In the test phase, a new data vector label

f (x) = ω · x + b (Equation 12)

It is determined by the sign of The hyperplane is uniquely determined by the margin maximization criterion. The optimum hyperplane can be described only by a vector on a margin called a support vector. The problem of obtaining the hyperplane can be formulated as the following optimization problem by introducing a slack variable ξ _i (soft margin).

The size of C corresponds to the size of the penalty when the constraint condition is violated. To solve this problem, the Lagrange multiplier α _i can be rewritten as follows.

Here, NS is the number of support vectors. (Expression 17) is derived from (Expression 15) and (Expression 16). When ω in (Expression 12) is replaced with (Expression 17),

Here, replacing the inner product x · _{x i} symmetric kernel function K _(x, x _i) and. This replacement is equivalent to an implicit conversion from data space to feature space. As a result, the following nonlinear discriminant function is constructed.

In this embodiment, a Gaussian kernel is used as the kernel function.

  As a data set, 10 samples of feature vectors having five feature values as dimensions were extracted from walking data of 32 hemiplegic patients, and all 320 samples were used for training and evaluation of SVM. Moreover, each feature-value was normalized to [0-1] as preprocessing. It is necessary to select appropriate values for the learning parameter C and the Gaussian kernel parameter σ. Optimal parameters were determined systematically using 10-fold cross validation.

  Table 2 shows the result of BS determination by the walking disorder degree determination unit 6. Prediction error estimation by 10-fold cross validation was performed using 320 samples obtained from walking data of 32 hemiplegic patients. An example of the results of splitting the data set in two, training on one hand and testing on the other is shown. As a result, the prediction error was 11.37%, and it was confirmed that the BS could be estimated with an accuracy of about 88%. BS. The correct answer rate of V was 83.7%, which was slightly low, but the correct answer rate exceeded 85% in the other groups, and a good classification of 88: 96% was achieved as a whole. The above results indicate that the gait disturbance features obtained from the waist trajectory are effective in quantitatively evaluating hemiplegic walking, and are the same as BS that is often used as an evaluation index for motor paralysis in clinical practice. It shows that it has a degree of evaluation ability.

  FIG. 15 shows an example of a processing flow of walking obstacle automatic analysis according to this embodiment. First, the walking motion measurement unit 3 measures the walking motion of the subject 2 (walking motion measurement step: S001). Next, the walking motion tracking unit 4 obtains a waist trajectory from the walking motion measured in the walking motion measurement step (S001) (walking motion tracking step: S002). Next, the gait disorder feature extraction unit 5 extracts the gait disorder feature quantity as the feature quantity related to the gait disorder from the waist trajectory obtained in the walking movement tracking process 4 (S001) (gait disorder feature extraction process: S003). ). On the other hand, the gait obstacle learning unit 7 causes the gait obstacle degree determination unit 6 to learn the gait disorder degree of the subject 2 whose gait obstacle degree is known as the learning data in association with the gait disorder feature amount (gait disorder learning step: S004). Then, the walking disorder degree determination unit 6 determines the walking disorder degree of the subject 2 based on the walking disorder feature amount extracted by the walking disorder feature extraction unit 5 and with reference to the learning data (walking disorder determination). Step: S005).

  As described above, according to the gait disorder automatic analysis system according to the present embodiment, it is possible to easily perform measurement and quantitatively evaluate the degree of gait disorder at the clinical site.

In the second embodiment, an example will be described in which the walking disorder degree determination unit 6 extracts five walking disorder feature quantities from the waist trajectory and determines the walking disorder degree of the subject 2 using principal component analysis. Differences from the first embodiment will be mainly described, and redundant description will be omitted.
The configuration of the gait disturbance automatic analysis system 1A (not shown) in the second embodiment is shown in FIG. However, the walking obstacle degree determination unit 6 is different in that the principal component analysis is performed instead of the analysis by the support vector machine of the first embodiment. Further, a processing flow example of the gait disturbance automatic analysis is also shown in FIG. However, the walking obstacle determination step (S005) is different in that principal component analysis is performed instead of analysis by the support vector machine of the first embodiment.

  BS focuses on the regularity of the recovery process of motor paralysis, and has received very high evaluation in terms of proposing its most basic classification, but on the other hand, because the classification is qualitative by observation The low resolution for evaluating the patient's recovery process is regarded as a problem. Moreover, it has been reported that even when no change is observed in the BS, the muscle strength and walking ability on the affected side are improved. Therefore, there is a demand for an index that can exceed the discrete classification framework of the BS and can evaluate the patient condition that changes continuously every day in more detail. Therefore, in this example, no change was observed in the BS during the experiment period in Example 1 using the five gait disorder features that were shown to be effective as evaluation indices for gait impairment in Example 1. We performed continuous gait measurement for a certain hemiplegic patient for a certain period and tried to quantitatively grasp the temporal change of the gait function of the patient.

  Table 3 shows the attributes of the subject 2. The subject 2 was a hemiplegic patient of Example 1 who was hospitalized and expected to recover walking function by rehabilitation intervention based on the findings of a physical therapist. Seven subjects participated, and all subjects were able to walk without assistance, but in order to ensure the safety of the subjects, measurements were performed in the presence of a physical therapist. In addition, all subjects were those who had no change in BS according to the findings of the physical therapist. The subject was given an outline of the experiment in advance and the experiment was conducted after obtaining consent.

  FIG. 16 exemplifies a change with time for each walking disorder feature amount. The result of subject B (BS.III) is shown. FIG. 16A shows an example of LRdif, FIG. 16B shows Hasym, FIG. 16C shows VUasym, FIG. 16D shows VDasym, and FIG. 16E shows an example of HA. The superiority of the change in each feature amount was tested using Friedman's test and Scheffe's paired comparison. As a result, a significant change was observed in all gait disturbance features (P <.01). However, the trend of change is not uniform. According to the findings of the physical therapist, no change in BS occurred during the experimental period, but the change in walking was clear. Moreover, the change of BS was not recognized from the result of the analysis by the support vector machine.

In this embodiment, the principal component analysis is used to reduce the five gait disturbance feature amounts to the principal components and use them as an index for quantitatively evaluating walking changes. Principal component analysis is effective as a method used to reduce the information held by a large number of variables and grasp the whole picture with a small number of information. In clinical research, it is a method for evaluating multiple variables in an integrated manner. It is used as.
Ten samples were newly extracted from the data for each measurement date of the seven subjects who newly participated in the 320 samples used in Example 1 to obtain a total of 470 samples, and further extracted from the 10 healthy subjects in Example 1 The principal component analysis was performed on a total of 570 samples including the 100 samples, and extraction of the principal components was attempted. Note that, as preprocessing, standardization was performed so that an average of 0 and a standard deviation of 1 were obtained for each feature amount.

  Table 4 shows the eigenvalue, contribution rate, and cumulative contribution rate of each principal component obtained as a result of the principal component analysis. Conventionally, a principal component having an eigenvalue of 1 or more is a meaningful principal component, and the principal component having an eigenvalue of 1 or more in Table 4 is only the first principal component. The factor loadings for each gait feature amount of the first principal component are: LRdif is 0.852, Hasym is 0.532, VUasym is 0.762, VDasym is 0.891, HA is 0.855, and Hasym However, since the correlation with other feature quantities is strong, the first principal component can be used as a scale that can evaluate the five feature quantities in an integrated manner to some extent.

  FIG. 17 is a diagram illustrating the relationship between the first principal component and the BS. The median value is calculated as the representative value of the first principal component for each subject, and the BS. III-BS. VI, a box plot for every 5 groups of healthy subjects. BS. III-BS. Since the categories of VI and healthy subjects can be regarded as ordinal scales from the scale of the degree of motor paralysis, BS. III to 1, BS. IV to 2, BS. V to 3, BS. 4 was assigned to VI and 5 were assigned to healthy subjects, and the presence or absence of correlation with the first principal component was tested using Spearman's rank correlation coefficient. As a result, BS. IV-BS. VI, BS. IV-Healthy, BS. A significant correlation was observed with V-Healty (p <0.01), and BS. III-BS. V, BS. III-BS. VI, BS. There was a significant trend between III and Health (p <.10.). The correlation coefficient was -0.888, indicating a strong negative correlation. Therefore, it can be judged that the first principal component is effective in capturing the subject's changes in an integrated manner.

FIG. 18 is a diagram in which changes with time of the first main component are plotted for each subject. Each point shows the median value extracted from the 10 primary sample principal components obtained for each measurement date. In healthy persons, the first principal component is distributed almost in the vicinity of -2, and therefore, as an evaluation method, the approach to -2 was improved, and the separation from -2 was defined as deterioration.
From FIG. 18, it is possible to quantitatively observe the tendency of the change in the first principal component even in the test subject whose BS did not change during the experimental period. Subjects A, C, and D are common in that they change once in the direction of deterioration and then shift in the direction of improvement, and subjects F and G also change in the direction of improvement during the two measurement periods. I understand that Subject E showed little noticeable change, but Subject B showed a trend of change that was different from other patients, with repeated deterioration and improvement. In this way, even when the BS was the same, the temporal change in the walking function of the patient could be quantitatively analyzed because the analysis of the waist trajectory was close to the center of gravity of the body near the lumbar spine, which was the measurement location. The point that is predicted not to be included. The body's center of gravity is located in the center of both hip joints, and the purpose of walking is to efficiently carry the body's center of gravity, and the lower limbs perform complex and delicate control of each joint and muscle to accomplish this purpose. Need. Since the center of gravity of the body and the lower limb are closely related, the analysis of the waist trajectory, which is equivalent to the analysis of the movement of the body centroid, was effective in evaluating the lower limb disorders.

In this way, even when the BS is the same, a plurality of appropriate gait disorder features are extracted from the waist trajectory and contracted by principal component analysis to quantitatively evaluate the degree of gait disturbance that the BS has evaluated. It has been shown that this can be done, and it has been found that it is possible to quantitatively evaluate the temporal changes in the gait function of patients who change continuously every day.
Moreover, according to the walking disorder automatic analysis system which concerns on a present Example, it can measure simply and can evaluate a walking disorder degree quantitatively in a clinical field.

In the first and second embodiments, the example in which the geometric walking obstacle feature amount is extracted from the waist trajectory and the walking obstacle degree is determined has been described. In the third embodiment, the temporal walking obstacle from the waist trajectory is described. An example will be described in which a feature amount is extracted and a walking obstacle degree is determined.
In Example 1 and Example 2, although the example which applies a gait disorder automatic analysis system to the gait disorder by hemiplegia was demonstrated, Example 3 demonstrates the example applied to the gait disorder by Parkinson's disease. Parkinson's disease is a disorder of the basal ganglia system, causing problems in the rhythm generation of walking movements. In particular, from the viewpoint of walking assistance, “smooth legs” at the beginning of walking and “accelerated walking” that cannot be stopped once walking starts.

  The configuration of the gait disorder automatic analysis system 1B (not shown) in the third embodiment is shown in FIG. However, the gait disturbance feature extraction unit 5 instead of extracting the five feature quantities that are geometric features from the waist trajectory in the first embodiment as the gait disturbance feature quantities, instead of extracting five feature quantities that are temporal features. The difference is that the gradient is extracted. Further, a processing flow example of the gait disturbance automatic analysis is also shown in FIG. However, in the gait obstacle feature extraction step (S003), instead of extracting the five feature quantities that are geometric features from the waist trajectory in the first embodiment, a gait period gradient that is a temporal feature is used. The point of extraction is different.

  The inventors applied a walking assistance system Walk-Mate, which stabilizes walking movement based on mutual synchronization of walking rhythms, to Parkinson's disease patients (see Patent Document 1). This Walk-Mate is a virtual robot configured on a PC and a human walking rhythm that draws each other by exchanging the rhythm sound corresponding to the foot contact timing, and the mutual synchronization of the walking rhythm and dynamic stabilization of the movement It is a system to realize. The phase relationship between the two walking rhythms can be controlled by the mutual adaptation process between the walking rhythm of the human and the walking rhythm of the assistance system. It is considered that the walking motion is promoted by applying a sound stimulus at a phase preceding the leg contact, and the walking motion is suppressed by applying a sound stimulus at the opposite phase. As a result, it is possible to promote the movement to the slack leg state and to suppress the movement to the accelerated walking state.

  FIG. 19 shows an example of the relaxation effect of accelerated walking by the walking assistance system. The vertical axis represents the walking cycle, and the horizontal axis represents time. The algorithm used for the gait control is an algorithm for controlling the timing shift at the time of synchronization, and the control device is controlled so that the rhythm on the assisting device side is synchronized slightly later than the patient side. Before the use of the walking assistance system, there is an accelerated walking in which the walking cycle fluctuates and the walking cycle gradually decreases. From the time when the cooperative walking with the assistance system is started (indicated by the use start arrow in the figure) It can be seen that the fluctuation stabilizes in about one minute and the fluctuation becomes small. In another measurement, it has been confirmed that a stable state continues for a while even if the use of the walking assistance system is stopped. This indicates that the use of this walking assistance system is effective in preventing accelerated walking in Parkinson's disease.

  This also means that a feature amount related to accelerated walking can be extracted as a characteristic gait disorder feature amount of Parkinson's disease. Furthermore, acceleration walking can be regarded as a process of decreasing the walking cycle, the gradient of the time change can be calculated, and the walking cycle gradient can be extracted as a walking disorder feature quantity. In addition, periodic fluctuations, left-right asymmetry of the waist trajectory, and the like can be extracted as gait disturbance feature values. Therefore, walking period gradients and fluctuations that are temporal features from the waist trajectory are extracted as gait disturbance feature values, and are associated with Yar's indices used as the degree of gait disturbance, as learning data. It can be used to determine the degree of gait disturbance. For the determination of the degree of gait disturbance, a support vector machine may be used as in the first embodiment or the second embodiment, and principal component analysis may be used. Alternatively, a pattern classifier other than the support vector machine may be used.

  When configured in this way, it is possible to extract gait disturbance features not only for the geometric characteristics of the waist trajectory but also for the temporal characteristics, make it easy to measure, and quantitatively evaluate the degree of gait disturbance in the clinical setting. Can do.

  Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and it is obvious that various modifications can be made to the embodiments.

  For example, in the above embodiment, the example in which the present invention is applied to hemiplegia and Parkinson's disease has been described. However, the present invention is also applicable to gait disorders related to other diseases such as knee osteoarthritis, rheumatism, and leg fracture. There is. Moreover, although the above embodiment demonstrated the example which uses a three-dimensional acceleration sensor as a motion sensor, you may use the position sensor using GPS. In the above embodiment, the configuration in which the walking motion tracking unit is provided in the personal computer has been described. However, the waist trajectory is obtained by a microcomputer in which the walking motion tracking unit is attached to the motion sensor, and wirelessly transmitted to the personal computer. It is good also as a structure. Further, in the above embodiment, the example of extracting the vertical and horizontal geometric feature values and temporal feature values extracted from the waist trajectory as walking obstacle feature values has been described. A difference between left and right stride that is a scientific feature amount may be extracted. Also, the number of extracted walking obstacle feature quantities is not limited to five or one, and can be changed according to the degree of correlation with the degree of walking obstacles. In addition, the walking disorder learning unit explained an example in which the learning data is collectively learned by the walking disorder degree determination unit. However, if the physical therapist newly recognizes the walking disorder degree, the walking disorder feature amount is newly added. Can be extracted and added to the learning data. Moreover, although the above embodiment demonstrated the example which uses a support vector machine and a principal component analysis for a walking disorder degree determination, you may use pattern classifiers, such as a linear classifier. In addition, the number of sampling cycles for data acquisition, the number of data acquisition cycles, the number of learning data, the kernel, and the like can be variously changed.

  The present invention is used for analyzing the degree of gait disturbance.

DESCRIPTION OF SYMBOLS 1 Walking obstacle automatic analysis system 2 Test subject 3 Walking movement measurement part 4 Walking movement tracking part 5 Walking obstacle feature extraction part 6 Walking obstacle degree determination part 7 Walking obstacle learning part 8 Control part 10 Personal computer 31 Lumbar trajectory measuring device 32 Foot contact timing Detection device

Claims (6)

A walking motion measuring unit that measures the walking motion of the subject by means of a motion sensor;
A walking motion tracking unit for obtaining a waist trajectory from the walking motion measured by the walking motion measuring unit;
A gait disorder feature extraction unit that extracts a gait disorder feature quantity as a feature quantity related to a gait disorder from the waist trajectory obtained by the walking movement tracking unit;
A gait disorder determination unit that determines the gait disorder degree of the subject based on the gait disorder feature amount extracted by the gait disorder feature extraction unit;
A gait disorder learning unit that causes the gait disorder degree determination unit to learn the gait disorder degree of a subject whose degree of gait disorder is known as learning data in association with the gait feature amount;
The gait disturbance degree determination unit performs the determination with reference to the learning data;
Gait disorder automatic analysis system. The gait impairment determination unit determines the gait impairment of the subject using a pattern classifier that classifies a vector in a multidimensional space into a plurality of patterns;
The gait disorder automatic analysis system according to claim 1. The gait disturbance determination unit determines the gait disturbance of the subject using principal component analysis;
The gait disorder automatic analysis system according to claim 1. As the gait obstacle feature amount, a geometric feature amount related to the left-right asymmetry of the waist movement is used;
The gait disorder automatic analysis system according to any one of claims 1 to 3. As the gait disorder feature amount, a temporal feature amount of hip exercise is used;
The gait disorder automatic analysis system according to any one of claims 1 to 3. The motion sensor is a three-dimensional acceleration sensor mounted near the lumbar spine;
The gait disorder automatic analysis system according to any one of claims 1 to 5.