JP2007020802A - Method for extracting gradual phase and nystagmus data analytic system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for reliably extracting a gradual phase for precisely carrying out precise extraction of a gradual phase even from nystagmus data when rotation stimulation is not given. <P>SOLUTION: This method is for reliably extracting the gradual phase by applying fuzzy theory in nystagmus data, and the method has: a step of calculating difference data between the n-th (n is an integer of ≥2) piece of nystagmus data and (n-1)-th piece of nystagmus data; a step of deciding a membership value based on the absolute value of the difference data; and a step of extracting the gradual phase by comparing the membership value with a prescribed threshold. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a method for extracting a slow phase in nystagmus data and a nystagmus data analysis system using the same.

  One of the medical eye movement analysis methods is an analysis method using vestibulo-oculomotor reflex. The vestibulo-oculomotor reflex is a reflection in which the eyeball moves in the opposite direction to the rotation of the head. When the acceleration stimulus in a certain direction continues, the eyeball is gradually displaced in the opposite direction, and the movement of rapidly returning to the original is repeated regularly.

In the eye movement, a smooth component that gradually shifts is called a “slow phase”, and a component that quickly returns to the original is called a “rapid phase”. The phenomenon in which these two phases appear alternately is known as nystagmus. The data obtained from the image series in which the nystagmus is observed is hereinafter referred to as “nystagmus data”.
For the vestibulo-oculomotor reflex analysis, it is necessary to remove the rapid phase component caused by the central action such as the cerebellum and extract only the slow phase component caused by the vestibulo-oculomotor reflex.

  Conventionally, various methods for removing a rapid phase component from nystagmus data and extracting a slow phase component have been proposed. For example, Non-Patent Document 1 discloses a method of automatically extracting a slow phase by applying fuzzy set theory to nystagmus data. In the conventional method such as the method of Non-Patent Document 1, the nystagmus data obtained by observing nystagmus when a rotational stimulus (sine wave stimulus) is applied to the head of the subject is based on a sine wave approximation curve. The slow phase was extracted by applying fuzzy set theory and determining membership values.

Mohammad Arzi, Michel Magnin: "A Fuzzy Set Theoretical Approach to Automatic Analysis of Nystagmic Eye Movements" IEEE Trans. On Biomedical Engineering, vol.36, NO.9, pp.954-963

  However, nystagmus is also caused by diseases such as a semicircular canal and a brain disorder other than when a rotational stimulus (sine wave stimulus) is applied to the subject's head. The above conventional method can extract a slow phase from nystagmus data obtained by observing nystagmus when a rotational stimulus is applied to the subject's head, but nystagmus when no rotational stimulus is applied. There was a problem that the slow phase could not be extracted accurately from the nystagmus data observed.

  In view of the above problems, the present invention provides a reliable slow phase extraction method and a nystagmus data analysis system that extract a slow phase with high accuracy even from nystagmus data when no rotational stimulus is applied. Is an issue.

In order to solve the above problems, the present invention has the following configuration.
A slow phase extraction method according to one aspect of the present invention is a method for extracting a slow phase by applying fuzzy theory to nystagmus data, wherein n (n is an integer of 2 or more) nystagmus data and ( n-1) calculating difference data with the nystagmus data, determining a membership value based on an absolute value of the difference data, and comparing the membership value with a predetermined threshold Extracting a slow phase.
The predetermined threshold value can be determined by an arbitrary threshold value determination method, but, for example, the Kittler method is preferable because of its high accuracy.
According to the present invention, it is possible to extract a slow phase with high accuracy not only from nystagmus data when a rotational stimulus is applied but also from nystagmus data when a rotational stimulus is not applied. A phase extraction method can be realized.

Preferably, the nystagmus data further includes a step of selecting whether the nystagmus data when the rotational stimulus is given or the nystagmus data when the rotary stimulus is not given, The calculation step, the determination step, and the extraction step are executed when nystagmus data is selected when no rotational stimulus is applied in the selection step.
According to the present invention, it is possible to extract a slow phase with high accuracy not only from nystagmus data when a rotational stimulus is applied but also from nystagmus data when a rotational stimulus is not applied. A phase extraction method can be realized.

An nystagmus data analysis system according to one aspect of the present invention is an nystagmus data analysis system that extracts a slow phase by applying fuzzy theory to nystagmus data, wherein n (n is an integer of 2 or more). Difference data between the nystagmus data and the (n−1) th nystagmus data is calculated, a membership value is determined based on the absolute value of the difference data, and the membership value is compared with a predetermined threshold value. A data processing unit for extracting a slow phase.
The predetermined threshold value can be determined by an arbitrary threshold value determination method, but for example, the Kittler method is preferable because of its high accuracy.
According to the present invention, a reliable eye that extracts a slow phase with high accuracy not only from nystagmus data when a rotational stimulus is applied but also from nystagmus data when a rotational stimulus is not applied. A vibration data analysis system can be realized.

Preferably, the data processing unit, when the nystagmus data is nystagmus data when no rotational stimulus is applied, (n is an integer of 2 or more) nystagmus data and (n -1) calculating difference data with the first nystagmus data, determining a membership value based on the absolute value of the difference data, and extracting the slow phase by comparing the membership value with a predetermined threshold It is configured to
According to the present invention, a reliable eye that extracts a slow phase with high accuracy not only from nystagmus data when a rotational stimulus is applied but also from nystagmus data when a rotational stimulus is not applied. A vibration data analysis system can be realized.

  According to the present invention, it is possible to extract a slow phase with high accuracy not only from nystagmus data when a rotational stimulus is applied but also from nystagmus data when a rotational stimulus is not applied. There is an advantageous effect that the phase extraction method and the nystagmus data analysis system can be realized.

  Hereinafter, examples specifically showing the best mode for carrying out the present invention will be described with reference to the drawings.

With reference to FIGS. 1-19, the nystagmus data analysis system using the slow phase extraction method in the Example of this invention is demonstrated.
First, nystagmus data handled by the nystagmus data analysis system using the slow phase extraction method in this embodiment will be described with reference to FIG.

  FIG. 1 is a diagram for explaining a slow phase and a rapid phase in nystagmus data. As shown in FIG. 1, the nystagmus data usually includes a “slow phase” (indicated by reference numeral 1) that is a gentle displacement and a displacement that suddenly returns to the original direction in the opposite direction to the slow phase. There is "rapid phase" (indicated by 2).

  In this embodiment, nystagmus data observing nystagmus when rotational stimulation (sine wave stimulation) is applied to the subject's head, and nystagmus data observing nystagmus when no rotational stimulation is applied. The two types of nystagmus data are handled. Both are obtained by acquiring an image series at 30 frames per second from a moving image obtained by photographing the eye movement of the subject with a video camera under each condition described later, and calculating nystagmus data therefrom. (That is, 30 nystagmus position data can be obtained for one second.)

  As nystagmus data in this embodiment, nystagmus eye position data (hereinafter simply referred to as “eye position data”) indicating the position of the eyeball in each image in the image series, and the speed of eyeball displacement during nystagmus are shown. Nystagmus velocity data (hereinafter simply referred to as “velocity data”) is used.

  As a method for calculating the eye position data, for example, a method is used in which an iris crest pattern is extracted from an image at a certain moment, and the eye position difference is obtained by comparing the pattern with the pattern of the image when viewed from the front. . As a method for calculating the velocity data, for example, the displacement of the position of the eyeball between two consecutive images in the image series is measured, and the time between each data (in the case of 30 frames per second, (1/30) seconds) ) Is used to obtain the value.

The nystagmus data when a rotational stimulus is applied to the subject's head will be described.
The subject's head is bent 60 degrees and kept in a posture rotated 45 degrees in the right or left direction, and the subject is rotated clockwise or counterclockwise. In this way, by applying a rotational stimulus to the subject's head, the semicircular canal is stimulated to induce nystagmus. The eye movement that occurs during rotation is shot with a video camera.
The frequency of the rotational stimulus is, for example, 0.1 Hz (that is, one rotation for 10 seconds) and 0.5 Hz (that is, one rotation for 2 seconds). In general, when a rotational stimulus is applied to the subject's head in this way, it is known that the slow phase component in the obtained nystagmus data has a sinusoidal shape with a frequency close to the frequency of the rotational stimulus. This feature is particularly noticeable for speed data.

The nystagmus data when no rotational stimulus is applied will be described.
In patients with benign paroxysmal positional vertigo (BPPV) who has a disease in the semicircular canal, the head position of the subject is first changed only once in the second canal aspect to induce nystagmus. The resulting eye movement is then captured with a video camera.

  With reference to FIG.2 and FIG.19, the structure of the nystagmus data analysis system using the slow phase extraction method in a present Example is demonstrated. FIG. 19 is a block diagram showing the configuration of the nystagmus data analysis system using the slow phase extraction method in this example. FIG. 2 is a flowchart showing a process flow of the nystagmus data analysis system using the slow phase extraction method in the present embodiment.

  The nystagmus data analysis system using the slow phase extraction method in this embodiment shown in FIG. 19 includes an nystagmus data extraction unit 2000, an operation unit 2001, a data processing unit 2002, a storage unit 2003, an image processing unit 2004, and a display. Part 2005.

The nystagmus data extraction unit 2000 inputs captured image data and extracts nystagmus data.
The operation unit 2001 inputs the type of captured image data, that is, whether the data is when a rotational stimulus is applied, whether the data is eye position data, or speed data. It is an operation input part for doing.

  The data processing unit 2002 reads out a flowchart (processing procedure) corresponding to each type from the storage unit 2003 using the nystagmus data type input from the operation unit 2001, and the nystagmus data input from the nystagmus data extraction unit 2000. A process for extracting the slow phase is performed (details will be described later).

The image processing unit 2004 creates display image data by superimposing the slow phase extracted by the data processing unit 2002 on the original nystagmus data extracted by the nystagmus data extracting unit 2000. Specifically, display image data is created by superimposing an image in which a slow phase portion is emphasized on an image obtained by graphing nystagmus data.
A display unit 2005 outputs the display image data created by the image processing unit 2004 to a display or the like. An operator (for example, a doctor or the like) diagnoses BPPV based on image data that is a slow phase extraction result displayed on the display unit 2005.

  In this example, BPPV is diagnosed by extracting the slow phase of nystagmus data when the subject's head position is first changed only once in terms of the latter canal, but not limited to this, Another disease can be diagnosed by inducing nystagmus by another method based on the characteristics of nystagmus (gaze nystagmus, spontaneous nystagmus, etc.) that appear due to another disease.

With reference to FIG. 2, the flow of processing of the nystagmus data analysis system using the slow phase extraction method in the present embodiment will be described.
In step S201, the nystagmus data extraction unit 2000 extracts nystagmus data from an image series obtained by photographing eye movements with a video camera.
In step S202, the data processing unit 2002 checks based on the nystagmus data type input from the operation unit 2001 whether or not the extracted nystagmus data is nystagmus data when a rotational stimulus is given. . If the nystagmus data is nystagmus data when a rotational stimulus is applied, the process proceeds to step S203. Otherwise, the process proceeds to step S206.

  In step S203, the data processing unit 2002 checks whether the extracted nystagmus data is speed data or eye position data based on the nystagmus data type from the operation unit 2001. If the nystagmus data is speed data, the process proceeds to step S204. If it is eye position data, the process proceeds to step S205.

  In each of steps S204, S205, and S206, the data processing unit 2002 executes a flowchart read from the storage unit 2003 based on the nystagmus data type (described later with reference to FIGS. 3, 8, and 9). And extract the slow phase. A detailed description of the slow phase extraction method in steps S204, S205, and S206 will be described later.

  In step S207, the image processing unit 2004 creates a display image by superimposing the slow phase extracted in steps S204, S205, and S206 on the original nystagmus data extracted by the nystagmus data extracting unit 2000.

  In step S208, the display unit 2005 outputs the display image data created by the image processing unit 2004.

With reference to FIGS. 3-9, the slow phase extraction method of a present Example is demonstrated.
First, with reference to FIG. 3 to FIG. 7, the slow phase extraction method of speed data in the case where the rotational stimulus in step S204 is given will be described in detail. FIG. 3 is a flowchart showing a slow phase extraction method for speed data when a rotational stimulus is applied.

In step S301, a sudden spike in the speed data is removed.
For example, there is a method using a low-pass filter. The difference between the original data y _i and the data y _i ′ after passing through the low-pass filter is taken as a noise component q _i . The noise component q _i is expressed by the following equation (1).
q _i = y _i −y _i ′ (i = 2, 3,..., N) (1)

The absolute value of each noise component q _i is compared with the threshold value q _ref, and data satisfying the following equation (2) is regarded as an abrupt spike and removed.
| Q _i |> q _ref (2)

In step S302, the data after spike removal in step S301 is periodically divided into a plurality of fragments.
In the present embodiment, first, a curve f ₁ (x _i ) represented by the following equations (3) and (4) is defined for periodic division. In the following equation (3), A, B, and C are parameters for fitting. ω is an angular frequency with respect to the frequency F of the rotational stimulus when 30 pieces of data are obtained per second, and is expressed by the following equation (4).
f ₁ (x _i ) = A · sin (ω · x _i ) + B · cos (ω · x _i ) + C (3)
ω = 2πF / 30 (4)

For the velocity data (x _i , y _i ), the sum S of the square errors between the defined curve f ₁ (x _i ) and each data y _i shown by the following equation (5) is minimized. , Parameters A, B, and C are determined. N indicates the total number of speed data.

  For example, FIG. 5 shows curves when the above-described processing is performed on the speed data shown in FIG. 4 and the parameters A, B, and C of the above equation (3) are determined. The speed data is divided according to the period of this curve. FIG. 6 shows how the speed data is divided according to the period of the curve in FIG. By performing subsequent analysis using fuzzy set theory for each segment, even if the nystagmus data has a sinusoidal waveform up and down as shown in FIG. I can do it.

In step S303, the first fragment periodically divided in step S302 is selected. Further, loop = 1 is set as the initial parameter setting.
In step S304, a curve f ₂ (x _i ) approximating the velocity data in the selected fragment is determined.
First, a curve (hereinafter referred to as “inertia curve”) f ₂ (x _i ) represented by the following equation (6) is defined for a rotational stimulus having a frequency F (unit: Hz). In the following equation (6), P, Q, and R are fitting parameters. ω is an angular frequency with respect to the frequency F of the rotational stimulus when 30 pieces of data are obtained per second, and is expressed by the above equation (4).
f ₂ (x _i ) = P · sin (ω · x _i ) + Q · cos (ω · x _i ) + R (6)

For the velocity data in the selected fragment, the sum S of the squared errors between the defined inertial curve f ₂ (x _i ) and each data y _{i as} shown by the following equation (7) is minimized: Parameters P, Q, and R of the inertial curve f ₂ (x _i ) are determined. In the following formula (7), a and b are the start and end points of the periodically divided fragments, and μ _i is the membership value. In the initial state, μ _i = 1 at all points in the velocity data in the selected fragment.

In step S305, the membership value is determined using the inertial curve f ₂ (x _i ) for which the parameter has been determined.
First, the velocity data _(x i, _{y i)} with respect to determining the value of _{n i} satisfy the following equation (8). Δy is a resolution defined by the following equation (9). max (y _i ) indicates data that is the maximum value among the data y _i . min (y _i ) indicates data that is the minimum value among the data y _i .
Although an arbitrary value can be set as Δy, in this embodiment, as an example, a value obtained by dividing the difference between the maximum value max (y _i ) and the minimum value min (y _i ) of eye position data by 20 is used. n _i is a value indicating the degree of separation of the velocity data y _i in units of Δy with respect to the inertial curve f ₂ (x _i ) at each point of the velocity data.
n _i · Δy ≦ (y _i −f ₂ (x _i )) <(n _i +1) · Δy (8)
Δy = (max (y _i ) −min (y _i )) / 20 (9)

An example of how to obtain the membership value will be described with reference to FIG.
First, as shown in FIG. 7, two windows W1 having a width w corresponding to a predetermined number of speed data (in FIG. 7, three speed data) and an overlapping width of (w / 3) and Define W2. M1 _i (= n _i−2 + n _i−1 + n _i ), 2 is the sum of the degrees of separation of each point (current attention point and previous two points) in the velocity data in the first window W1. M2 _i (= n _i + n _{i + 1} + n _{i + 2} ) is the total separation degree of each point (current attention point and the following two points) of the nystagmus data in the second window W2. Let m3 _i (= n _i ) be the sum of the degrees of separation of nystagmus data points (current attention points) in the overlapping portion of W1 and the second window W2. m1 _i , m2 _i , and m3 _i are calculated from the above equation (8).

The calculated m1 _i , m2 _i , and m3 _i are all multiplied as shown in the following equation (10). As a result, the weighted membership value μ _i ′ at the point (current attention point) belonging to the overlapping portion of the first window W1 and the second window W2 is obtained.
μ _i ′ = m1 _i · m2 _i · m3 _i (i = a + 2, a + 3,..., b-3, b-2) (10)

By sequentially shifting the two windows W1 and W2 in the periodically divided fragment, the membership value μ _i ′ at each point of the velocity data is obtained. Finally, the obtained membership value μ _i ′ is normalized as shown in the following equation (11) to calculate the membership value μ _i at each point of the velocity data. max (μ _i ′) indicates the membership value which is the maximum value among the membership values μ _i ′ obtained as described above in the selected fragment. The closer to 1 the membership value μ _i , the closer to the inertial curve f ₂ (x _i ), and the higher the possibility of a slow phase.
μ _i = 1− (μ _i ′ / max (μ _i ′)) (11)

In step S306, it is checked whether or not the parameter loop is 3. If loop = 3, the process proceeds to step S308. Otherwise, 1 is added to the parameter loop (step S307), the process returns to step S304, and steps S304 and S305 for obtaining the membership value μ _i are repeated. By repeating steps S304 and S305 a plurality of times, an optimum membership value μ _i is obtained. In this embodiment, steps S304 and S305 are repeated three times, but the number of repetitions may be any number of one or more.

  In step S308, it is checked whether the currently selected fragment is the last fragment. If it is not the last fragment, the next fragment is selected and the parameter loop is initialized to 1 (step S309), the process returns to step 304, and the same processing is repeated for the next fragment. If it is the last fragment, go to step 310.

  In step S310, a threshold value for extracting a slow phase is determined. In this embodiment, the Kittler method is used as the threshold value determination method. Kittler's method divides the membership value distribution into two classes according to a certain threshold. The ambiguity indicating which class a membership value belongs to which class belongs to the worst distribution (normal distribution). It is a way to minimize.

For a threshold k (0 <k <1), a class having a membership value greater than or equal to the threshold k is C ₀ , and a class having a membership value less than the threshold k is C ₁ . A value k that minimizes the degree of ambiguity E (k) belonging to one of the two classes represented by the following equation (12) is selected as a threshold value. N is the number of all membership values (that is, the number of all speed data), N ₀ is the number of membership values belonging to class C ₀ , N ₁ is the number of membership values belonging to class C ₁ , and σ ₀ ² ( k) and σ ₁ ² (k) denote the variance of the respective membership values in classes C ₀ and C ₁ . Further, the following equations are established: ω ₀ = N ₀ / N, ω ₁ = N ₁ / N.
E (k) = ω ₀ log {σ ₀ / ω ₀ } + ω ₁ log {σ ₁ / ω ₁ } (12)

  Instead of the Kittler method, other threshold determination methods such as the known Otsu method may be used. However, the Kittler method is preferred because of the high accuracy of slow phase extraction.

In step S311, slow phase extraction is performed. Compares the threshold k determined in step S310, the a membership value mu _i at the point of each speed data, if the membership value mu _i is the threshold value k or more, considered that data point as being slow phase . The data points extracted as the slow phase are reflected on the original data by marking them as black spots on the graph of the original data, and display image data is created.
After the processing in steps S301 to S311 is completed, the process proceeds to step S207 in FIG. Step S207 in FIG. 2 has already been described.

  Next, with reference to FIG. 8, the slow phase extraction method of eye position data when the rotational stimulus of step S205 is given will be described in detail. FIG. 8 is a flowchart showing a slow phase extraction method for eye position data when a rotational stimulus is applied.

  Since the eye position data when the rotational stimulus is applied does not maintain a sine waveform as compared with the speed data, a step of calculating difference data (step S801) is added to the step of the flowchart shown in FIG. To do.

In step S801, for each point in the eye position data, the difference between the eye position data and the previous eye position data is obtained to obtain difference data. Since the eye position data when a rotational stimulus is applied has a large displacement, the eye position data can be made closer to a sine wave shape by obtaining difference data. For the eye position data (x _i , y _i ), the difference data d _i is defined by the following equation (1). N is the number of all eye position data.
d _i = y _i −y _i−1 (i = 2, 3,..., N) (13)

  Steps S802 to S812 are the same as steps S301 to S311 in FIG. 3 described above except that they are performed on the difference data obtained in step S801 instead of the speed data, and thus description thereof is omitted. After the processing in steps S801 to S812 is completed, the process proceeds to step S207 in FIG. Step S207 in FIG. 2 has already been described. Step S801 is executed immediately before step S802, but may be executed at a timing later than step S802.

In addition, in the flowchart shown in FIG.3 and FIG.8, you may add the step which calculates | requires the final fitting curve with respect to the extracted slow phase. This is because the amplitude and phase of the final fitting curve are medically useful. In that case, a curve of the following equation (14) is defined for the extracted slow phase, and parameters a, b, c, and d of this equation are determined so as to minimize G of the following equation (15). To obtain the final fitting curve. Note that μ _i is the finally obtained membership value, and z _i is the extracted slow phase component.
f (x _i ) = a + b · x _i + c · sin (ω · x _i + d) (14)

  Next, with reference to FIG. 9, the method of extracting the slow phase of nystagmus data when the rotational stimulus is not given in step S206 will be described in detail. FIG. 9 is a flowchart showing a slow phase extraction method for nystagmus data when no rotational stimulus is given.

  For nystagmus data when no rotational stimulus is given, it is difficult to predict the shape of the slow phase because a sinusoidal waveform cannot be obtained, and an inertial curve is defined as when a rotational stimulus is given Can not. Therefore, in this embodiment, attention is paid to the absolute value of the difference between the nystagmus data and the previous nystagmus data for each point of the nystagmus data, and the absolute value of the difference is used instead of the inertia curve. “The absolute value of the difference is small” means that the nystagmus data belongs to the slow phase, and “the absolute value of the difference is large” means that the degree of the nystagmus data belongs to the rapid phase. high.

  This method can be applied to both eye position data and speed data. Hereinafter, only the case of applying to eye position data will be described, and the description of speed data will be omitted because it is the same.

In step S901, for each point in the eye position data, the difference between the eye position data and the previous eye position data is calculated to obtain difference data. For the eye position data (x _i , y _i ), the difference d _i is defined by the following equation (16). N is the number of all eye position data.
d _i = y _i −y _i−1 (i = 2, 3,..., N) (16)

In step S902, the membership value is determined using the absolute value of the difference data d _i calculated in step S901.
First, at each point of difference data to determine the value of n _i of the following equation (17). Δy is a resolution defined by the following equation (18). max (d _i ) indicates the difference data which is the maximum value among the difference data d _i obtained as described above. min (d _i ) indicates difference data which is the minimum value among the difference data d _i obtained as described above.
Δy can be set to an arbitrary value, but in this embodiment, as an example, a difference between the maximum value max (d _i ) and the minimum value min (d _i ) of the difference data is divided by 20. n _i is a value indicating the degree of separation of the absolute value of the difference data d _i in units of Δy.
n _i · Δy ≦ | d _i | <(n _i +1) · Δy (17)
Δy = (max (d _i ) −min (d _i )) / 20 (18)

An example of a method of obtaining the distribution of separability n _i will be described.
First, two windows W1 and W2 having a width w corresponding to a predetermined number of difference data (for example, three) and overlapping each other (w / 3) are defined. M1 _i (= n _i−2 + n _i−1 + n _i ) is the sum of the degrees of separation of the absolute value points of the difference data in the first window W1 (current attention point and previous two points). ) M2 _i (= n _i + n _{i + 1} + n _{i + 2} ), the sum of the degrees of separation of the absolute value points of the difference data in the second window W2 (current attention point and the following two points), Let m3 _i (= n _i ) be the sum of the degrees of separation of the absolute value points (current attention points) of the difference data in the overlapping portion of the first window W1 and the second window W2. m1 _i , m2 _i , and m3 _i are calculated from the above equation (17).

The calculated m1 _i , m2 _i , and m3 _i are all multiplied as shown in the following equation (19). As a result, the weighted membership value μ _i ′ at the absolute value point (current attention point) of the difference data belonging to the overlapping portion of the first window W1 and the second window W2 is obtained.
μ _i ′ = m 1 _i · m 2 _i · m 3 _i (i = 3, 4,..., N−1, N−2) (19)

By sequentially shifting the two windows W1 and W2, the membership value μ _i ′ at each point of the absolute value of the difference data is obtained. Finally, the obtained membership value μ _i ′ is normalized as shown in the following equation (20), and the membership value μ _i at each point of the absolute value of the difference data is calculated. max (μ _i ′) represents the membership value that is the maximum value among the membership values μ _i ′ determined as described above. The closer the membership value μ _i is to 1, the higher the possibility of a slow phase.
μ _i = 1− (μ _i ′ / max (μ _i ′)) (20)

  Steps S903 and S904 are the same as steps S310 and S311 in FIG. 3 and steps S811 and S812 in FIG. After the processing in steps S901 to S904 is completed, the process proceeds to step S207 in FIG. Step S207 in FIG. 2 has already been described.

  As described above, according to the slow phase extraction method in the present embodiment, it is possible to extract the slow phase even for nystagmus data when no rotational stimulus is given. Note that the slow phase extraction method for nystagmus data when no rotational stimulus is given, as shown in FIG. 9, can also be applied to nystagmus data when a rotational stimulus is given. In this case, the processing according to the flowchart of FIG. 9 has an effect that the processing time can be shortened, as is apparent from the comparison with the processing according to the flowcharts of FIG. 3 and FIG.

With reference to FIGS. 10-18, the precision of the slow phase extraction method of a present Example is evaluated.
First, the accuracy when the slow phase extraction method of this embodiment is applied to a data model will be described with reference to FIG. FIG. 10 is a diagram showing four data models.

FIG. 10A shows a data model when a rotational stimulus is applied, and shows a velocity data model in which the amplitude of the slow phase component is constant and there is no displacement in the horizontal direction. FIG. 10 (b) shows a data model in the case where a rotational stimulus is applied, and shows a velocity data model in which the slow phase component has a constant amplitude but is inclined obliquely. FIG. 10C shows a data model in the case where a rotational stimulus is given, and a velocity data model in which the amplitude of the slow phase component is not constant. FIG. 10D shows an eye position data model in which the shape of the nystagmus data is sawtooth.
The data models shown in FIGS. 10A, 10B, 10C, and 10D are hereinafter referred to as models A, B, C, and D, respectively.

  The models A, B, and C were created by combining the slow phase component as a sine wave and synthesizing random noise corresponding to the rapid phase component into the sine wave signal. For model D, a curve with a gentle slope as a slow phase and a curve with a steep slope as a rapid phase were connected together.

By comparing the slow phase data extracted by applying the slow phase extraction method described above to models A, B, C, and D and the known slow phase data in the modeled nystagmus data, Quantify the accuracy of slow phase extraction. The slow phase extraction accuracy p is defined by the following equation (21) (unit of p is%). Note that the number of extracted slow phases is U, the total number of data is N, and the number of artificially created rapid phases is V.
p = (U / (N−V)) × 100 (21)

  The slow phase extraction method using the flowchart of FIG. 3 and the flowchart of FIG. 8 is referred to as algorithm 1, and the slow phase extraction method using the flowchart of FIG.

Result of calculating slow phase extraction accuracy p by applying algorithm 1 to models A, B, and C, and slow phase extraction accuracy by applying algorithm 2 to models A, B, C, and D The results of calculating p are shown in Table 1. The reason why the algorithm 1 is not applied to the model D is that the algorithm 1 performs analysis under the assumption that the slow phase is sinusoidal, and is not applicable to the model D.
As is clear from Table 1, in each case, a slow phase extraction accuracy of 85% or more was obtained. For model D, a slow phase extraction accuracy of 90% or more was obtained.

Next, with reference to FIGS. 11 to 18, the accuracy when the slow phase extraction method of the present embodiment is applied to actual data will be described.
First, the result of applying the slow phase extraction method of the present embodiment to nystagmus data when a rotational stimulus is applied will be described with reference to FIGS. FIG. 11 and FIG. 12 are diagrams respectively showing measured eye position data and a slow phase extraction result when rotational stimuli of 0.1 Hz and 0.5 Hz are given. 11 and 12, (a) is the original eye position data, (b) is the result of extracting the slow phase using the slow phase extraction method of this example, and (c) is a partially enlarged view of (b). Each figure is shown.
As shown in FIGS. 11 and 12 (b) and 12 (c), in both the data of 0.1 Hz and 0.5 Hz, a gentle displacement portion in the eye position data is extracted, and a good slow motion is obtained. Phase extraction results were obtained.

With reference to FIG. 13 and FIG. 14, the result of applying the slow phase extraction method of the present embodiment to the speed data when a rotational stimulus is applied will be described. FIG. 13 and FIG. 14 are diagrams respectively showing measured speed data and a slow phase extraction result when rotational stimuli of 0.1 Hz and 0.5 Hz are given. 13 and 14, (a) is the original velocity data, (b) is the result of extracting the slow phase using the slow phase extraction method of this example, and (c) is a partially enlarged view of (b). Respectively. For the velocity data, a process for obtaining a final fitting curve was performed on the extracted slow phase.
As shown in FIGS. 13 and 14 (b) and (c), in both 0.1 Hz and 0.5 Hz data, a moderately displaced portion is extracted from the velocity data, and a good slow phase is obtained. Extraction results were obtained.

  The result of applying the slow phase extraction method of the present embodiment to nystagmus data when no stimulus is given will be described with reference to FIGS. 15, 16, 17, and 18. FIG. 15, FIG. 16, and FIG. 17 show the case of the latter half canal type benign paroxysmal positional vertigo (case kt) and the case of the latter half canal type benign paroxysmal positional vertigo having a particularly strong rotation component (case gt). FIG. 5 is a diagram showing measured eye position data of a case of a benign paroxysmal positional vertigo of the outer semicircular canal type (case ried) and a slow phase extraction result corresponding thereto. FIG. 18 is a diagram showing the measured speed data of the case ried and the slow phase extraction result corresponding thereto. 15, FIG. 16, FIG. 17 and FIG. 18, (a) is the original data, (b) is the result of the slow phase extraction using the slow phase extraction method of this example, and (c) is one of (b). Each enlarged view is shown.

  As shown in FIGS. 15, 16, 17, and 18, in any case data, there is a part in which the rapid phase is not removed or the slow phase is not extracted, Overall, good results were obtained. In particular, in the case ried, it is known that the slow phase component is exponentially attenuated, and this and the slow phase extraction result shown in FIG. 18 agree with each other, and a good result is shown. It was.

  As described above, the slow phase extraction method of the present invention automatically extracts a slow phase with high accuracy not only from nystagmus data when a stimulus is given, but also from nystagmus data when a stimulus is not given. It is possible to realize a reliable slow-phase extraction method and nystagmus data analysis system.

  The slow phase extraction method and the nystagmus data analysis system according to the present invention can be used for, for example, a medical nystagmus data analysis system.

Diagram for explaining slow phase and rapid phase in nystagmus data The flowchart which shows the slow phase extraction method in the Example of this invention. The flowchart which shows the slow phase extraction method about the speed data when the rotation stimulus is given in the Example of this invention. The figure which shows the speed data of the state where the sine waveform is displaced up and down The figure which shows the curve at the time of processing with respect to the speed data of FIG. 4, and determining the parameter The figure which shows a mode that the speed data was divided | segmented according to the period of the curve of FIG. Illustration to explain how to find membership values The flowchart which shows the slow phase extraction method about the eye position data in the Example of this invention when rotational stimulation is given. The flowchart which shows the slow phase extraction method about the eye position data and speed data when the irritation | stimulation is not given in the Example of this invention. The figure which shows four data models of the nystagmus data in the Example of this invention. The figure which shows the measurement eye position data when the test subject is given the rotational stimulation of 0.1 Hz, and the slow phase extraction result with respect to it The figure which shows the measurement eye position data when the test subject is given the rotational stimulus of 0.5 Hz, and the slow phase extraction result with respect to it The figure which shows the measurement speed data when the test subject is given the rotational stimulation of 0.1 Hz, and the slow phase extraction result with respect to it The figure which shows the measurement speed data when the test subject is given the rotational stimulus of 0.5 Hz, and the slow phase extraction result with respect to it The figure which shows the measurement eye position data of case kt, and the slow phase extraction result with respect to it The figure which shows the measurement eye position data of case gt, and the slow phase extraction result with respect to it The figure which shows the measurement eye position data of case ried, and the slow phase extraction result with respect to it The figure which shows the measurement speed data of case ried, and the slow phase extraction result with respect to it The block diagram which shows the structure of the nystagmus data analysis system using the slow phase extraction method in the Example of this invention.

Explanation of symbols

1 Slow phase 2 Rapid phase

Claims (4)

A method for extracting a slow phase by applying fuzzy theory in nystagmus data,
calculating difference data between n (n is an integer of 2 or more) nystagmus data and (n−1) nystagmus data;
Determining a membership value based on an absolute value of the difference data; and extracting a slow phase by comparing the membership value with a predetermined threshold;
A slow phase extraction method characterized by comprising: Selecting whether the nystagmus data is nystagmus data when a rotational stimulus is given or nystagmus data when a rotary stimulus is not given,
The calculation step, the determination step, and the extraction step are executed when nystagmus data is selected when no rotational stimulus is given in the selection step. 2. The slow phase extraction method according to 1. A nystagmus data analysis system that extracts a slow phase by applying fuzzy theory to nystagmus data,
Calculate difference data between n (n is an integer of 2 or more) nystagmus data and (n−1) nystagmus data, and determine a membership value based on the absolute value of the difference data A nystagmus data analysis system comprising: a data processing unit that extracts a slow phase by comparing the membership value with a predetermined threshold value. When the nystagmus data is nystagmus data when no rotational stimulus is applied, the data processing unit includes n (n is an integer of 2 or more) nystagmus data and (n−1) Calculating difference data from the first nystagmus data, determining a membership value based on an absolute value of the difference data, and extracting a slow phase by comparing the membership value with a predetermined threshold value The nystagmus data analysis system according to claim 3, wherein

Images