JPH10295673A - Disease seriousness diagnosis device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To objectively diagnose the seriousness and the state of the disease based on stabilometry data by extracting evaluation parameter from the stabilometry results so as to learn the diagnosis of the disease seriousness, and analyzing the evaluation parameter using the learning result so as to diagnose the seriousness of the disease. SOLUTION: When a testee is mounted on a detection table 2, respective loads detected by load sensors S1-S3 are outputted by a signal amplifier 3. The signal amplifier 3 amplifies the respective load signals to a prescribed level and an A/D converter 4 converts the respective load detection signals into digital data so as to output it to an device 5. The arithmetic device 5 extracts the evaluation parameter stored in the hard disc beforehand based on the stabilometry results so as to diagnose the disease seriousness based on the input data such as the extracted evaluation parameter, an input signal expressing the disease name inputted from the outside, and the learning results by an error inverse propagation method of a neural network, using the neural network.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for diagnosing the severity of a disease, and more particularly, to a method of analyzing a center of gravity sway test data by a sway meter using a neural network technique.
The present invention relates to a disease severity diagnosis apparatus having a severity diagnosis function for diagnosing the severity of a disease in the same category.

[0002]

2. Description of the Related Art It has been empirically known that data of a body sway test using a body sway meter is different depending on the severity and condition of each disease which complains of vertigo and imbalance.
For example, utilizing this fact, a patient diagnosed with Parkinson's disease is tested by a body sway meter, and a physician analyzes the data of the body sway test by the test. Can also be determined to be any of the positive syndrome, the negative symptom group a, or the negative symptom group b.

[0003]

However, the above-mentioned severe diagnosis relies on the experience of a physician who analyzes the data, and no objective criteria for diagnosing the severity or disease state has been established. For such analog problems of finding a pattern based on data, it is effective to use an approach from neural network technology that demonstrates excellent capabilities in pattern-related problems such as pattern mapping, pattern perfection, and pattern recognition. It is.

[0004] A neural network solves a problem of analog information processing that cannot be solved by a conventional sequential processing computer system or is inefficient even if it can be solved. It is intended to be solved by a parallel processing type computer system that performs tightly coupled parallel processing, especially pattern mapping, pattern perfection,
It excels in pattern-related issues such as pattern recognition. In the neural network technology, a processing method for engineeringly realizing a mechanism of a biological nervous system and, if possible, a mechanism of a human brain nervous system is used. That is, a large number of simple but super-parallel and super-dispersed neurons that perform signal processing are connected in a network form, and exchange information to perform information processing on the entire network. At this time, some sample input data are given to the network, and the network is spontaneously learned so that the desired output data is obtained, so that the network itself finds a processing procedure.

[0005] The learning ability by training is a major feature of neural networks. Among them, the error backpropagation method described below is the most widely used neural network paradigm, and has a wide range of fields. Successful in application.

[0006] The processing unit of the neural network performs an operation of obtaining a weighted sum according to the strength of each mutual coupling of multiple inputs from another plurality of units, and calculates an output by a nonlinear threshold function. It is sent to the target processing unit through the output side interconnect. The weight of the strength of the interconnection of the processing units is adjusted, i.e. the weight is corrected so that if the network gives the wrong answer, the weight is corrected so that the probability that the subsequent response of the network is correct will be correct. The ascending error is the error back propagation method. Given that the network produces an output pattern associated with it given an input pattern, this is called pattern mapping, and the backpropagation method can handle problems that require this pattern mapping.

[0007] The object of the present invention is to provide a method for the objective severity of a disease accompanied by a balance dysfunction in the otolaryngological region from test data of a body sway by a test using a body sway meter by using the back propagation method of the neural network. It is an object of the present invention to provide an apparatus for seriously diagnosing a disease, which is provided with a severely diagnosing function enabling diagnosis of a disease or a disease state.

[0008]

According to the first aspect of the present invention,
A detection table for detecting the load of the subject, calculating means for sequentially obtaining the center of gravity of the subject from the load detected by the detection table as a result of calculating the center of gravity by a predetermined calculation process; A gravity center calculation result storage means for sequentially storing a gravity center calculation result, and a disease severity diagnosis apparatus for diagnosing the severity of a disease based on a gravity center calculation result obtained from a gravity center sway meter, wherein a disease name of a subject is input. Input means, extraction means for extracting an evaluation parameter used for the severity diagnosis of a disease from the center of gravity calculation result stored in the center of gravity calculation result storage means, based on the evaluation parameter extracted by the extraction means, Learning means for learning the severity diagnosis of the disease relating to the disease name inputted from the input means,
Using the learning result of the severity diagnosis of the disease learned by the learning means, analyzing the evaluation parameters extracted by the extraction means, and performing a severity diagnosis of the disease associated with the disease name input from the input means. A diagnostic means; and an output means for outputting a result of the severe diagnosis of the disease performed by the severe diagnostic means.

According to the first aspect of the present invention, the extracting means extracts an evaluation parameter used for diagnosing the severity of a disease from the center-of-gravity calculation result stored in the center-of-gravity calculation result storing means. Based on the evaluation parameters extracted by the extracting means, learning of a severity diagnosis of the disease associated with the disease name input from the input means is performed, and the severity diagnosis means learns the severity diagnosis of the disease learned by the learning means. Using the result, the evaluation parameters extracted by the extraction means are analyzed, and a severity diagnosis of the disease associated with the disease name input from the input means is performed. The result of the severe diagnosis is output.

Therefore, it is possible to output the result of objectively diagnosing the severity of the disease using the result of the calculation of the center of gravity by the sway meter.

According to a second aspect of the present invention, in the apparatus for diagnosing a disease severity according to the first aspect, the learning means performs learning of the severity diagnosis of the disease associated with the input disease name based on a predetermined neural network. The method is characterized in that the severity diagnosis means performs a severity diagnosis of the disease using a neural network learned by the learning means.

According to the second aspect of the present invention, the learning means performs the learning of the severity diagnosis of the disease associated with the input disease name using a learning method based on a predetermined neural network. Performs a severe diagnosis of the disease using the neural network learned by the learning means.

Therefore, a highly reliable objective diagnosis result of a disease can be output using the established learning rule of the neural network.

According to a third aspect of the present invention, in the apparatus for diagnosing a disease severity according to the first or second aspect, the output means is provided when the result of the severity diagnosis of the disease performed by the severeness diagnosis means covers various symptoms. In addition, an output indicating the ratio is performed for each diagnostic symptom.

According to the third aspect of the present invention, when the result of the severe diagnosis of the disease performed by the severe diagnostic means covers a wide range of symptoms, the output means outputs an indication indicating a ratio for each diagnostic symptom. .

Therefore, it is possible to output the result of the serious diagnosis of the disease in an easily understandable expression.

[0017]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIGS. 1 to 15 are diagrams showing an embodiment of a disease severity diagnosis apparatus to which the present invention is applied.

First, the configuration will be described. FIG. 1 is a block diagram showing a main configuration of a disease severity diagnosis apparatus 1 according to the present embodiment. In this FIG.
Comprises a detection table 2, a signal amplifier 3, an A / D converter 4, an arithmetic unit 5, a display unit 6, and a printing unit 7.

The detection table 2 is provided with load sensors S1 to S3 for detecting a load at each apex when the subject rides on the detection table 2, and is detected by the load sensors S1 to S3. Each load is output to the signal amplifier 3 as a load detection signal.

The signal amplifier 3 amplifies each load detection signal to a predetermined level and converts the load detection signal input from the detection table 2 into a signal level that can be processed by each subsequent block. 4 is output.

The A / D converter 4 A / D converts each load detection signal input from the signal amplifier 3 into digital data that can be processed by the arithmetic unit 5, and converts each load detection signal into predetermined load data. The data is converted and output to the arithmetic unit 5.

The arithmetic unit 5 has a CPU (Central Processi
ng Unit), ROM (Read Only Memory), RAM (Ra
2 and 3 and has the functions shown in FIGS. FIG. 2 is a diagram showing a functional configuration of the arithmetic unit 5. In FIG. 2, the arithmetic unit 5 detects a load from each load data input from the A / D converter 4, and detects the load. The center of gravity calculation function for calculating the center of gravity from each of the calculated loads, the center of gravity calculation result storage function for storing the calculated center of gravity calculation results, the analysis function for analyzing the stored center of gravity calculation results, and the analysis results are displayed. It has a display function for displaying on the device 6 and a printing function for printing the analysis result by the printing device 7.

FIG. 3 is a diagram showing details of the analysis function of FIG. In FIG. 3, the arithmetic unit 5 takes in the center-of-gravity calculation results stored in the RAM, extracts the evaluation parameters stored in the hard disk in advance, and extracts the evaluation parameters. A neural network for diagnosing the severity of a disease by using a neural network as an input of an input signal representing a disease name to be performed and a learning result of the neural network by an error backpropagation method.

As the display device 6, a CRT (Cathode Ra) is used.
y Tube) or the like is used, and the calculation device 5 analyzes the result of the calculation of the center of gravity, and displays a result of a serious diagnosis in which the severity of the disease is diagnosed. The printing device 7 analyzes the result of the calculation of the center of gravity by the calculation device 5, and prints a result of the severity diagnosis or the like in which the severity of the disease is diagnosed.

Next, the operation of this embodiment will be described. First, the learning function of the neural network shown in FIG. 3 which is executed by the arithmetic unit 5 at the time of constructing an optimal neural network before severe diagnosis is shown in FIGS.
This will be described with reference to FIG. FIG. 4 is a flowchart illustrating a learning function of the neural network.
5 and 6 are diagrams showing a neural network structure, and FIGS. 7 to 12 are graphs showing examples of learning results when an optimal neural network is constructed.

First, the arithmetic unit 5 calculates a severe (for example, middle cerebellar, cerebellar, etc.) data from past sway test data for diseases belonging to one category (for example, cerebellar disease) stored in a storage device such as a hard disk. Hemispheres, spinocerebellar degeneration, etc.) were collected for known evaluation data (FIG. 4).
Step S1) This data is input to the neural network whose structure is not optimized (Step S2 in FIG. 4).

Next, the operation shifts to the operation of determining the structure of the neural network (step S3 in FIG. 4). As the neural network model, a hierarchical neural network as shown in FIG. 5 is employed. In this neural network, a unit, which is a signal processing element that receives a plurality of inputs, performs an operation using a threshold function, and outputs the signals, is arranged in three types of independent layers: an input layer, an intermediate layer, and an output layer. The input layer is the first,
The intermediate layer is the second, the output layer is the third, and so on. The input units U1 to U6 constituting the input layer and the intermediate units U7 to U1 constituting the intermediate layer.
1, and output units U12 to U1 forming an output layer
Four.

The strength of the connection between the unit Ui of the (k-1) th layer and the unit Uj of the kth layer is called a load and is called w ^k-1 _i ^k _j
It expresses. For example, the input unit U1 of the input layer (first layer)
And the load between the intermediate unit U7 of the intermediate layer (second layer) is w
¹ ₁ ² ₇ Each layer is supposed to receive input only from an earlier layer. Therefore, since there is no feedback-type coupling back, the input pattern given to the first layer, that is, the input layer, is successively converted through the layers, and reaches the final layer, which is the output layer, so that one time information is obtained. The process will end. In FIG. 5, the load between the units is represented by the thickness of the line connecting the units (the thicker the load is, the greater the load) and the type of line (the dotted line is a positive value, and the solid line is a negative value).

[0029] Here, the sum of the input to the unit Uj of the k layer i ^k _j, when the sum of the output from the unit Ui of the (k-1) layer o ^k-1 _i, that, i ^k _j Is the sum of the inputs from all units Ui in the (k-1) layer multiplied by the respective loads,

(Equation 1) Further, the output values from unit Uj of k layer and o ^k _j, threshold function f (x) is common to all units of formula (2) f (x) = 1 / (1 + exp (-x) ) assuming that given by ··· (2), o ^k _j is expressed as _{^{o k j = f (i k}} j) ··· (3).

In the input units U1 to U6, the evaluation parameters are extracted from the measurement values of the disease severity diagnostic apparatus 1 by the function of extracting the evaluation parameters shown in FIG. The data stored as evaluation data is input.

Specifically, the outer peripheral area E.I., which is the inner area surrounded by the outermost contour of the locus of sway of the center of gravity. AREA
Is input to the input unit U1, and the unit time trajectory length LNG / TIM, which is the average value of the moving speed of the center of gravity within the measurement time
E is input to the input unit U2, and the unit area locus length LN, which is the length of the center of gravity moved in the unit area within the measurement time,
G / AREA is input to the input unit U3, and the average X-direction fluctuation center displacement MX, which is the average value of the vibration in the X (left and right) direction, is input to the input unit U4. A certain Y-direction fluctuation average center displacement MY is input to the input unit U5, and the closed eye /
The Romberg rate ROM, which is the ratio of the eye opening sway, is input to the input unit U6.

[0032] The input unit U1 of the input layer (first layer), i ^{₁ 1} = E. As AREA, outputs o ¹ ₁ = f a (i ¹ ₁₎ intermediate layer for each of the intermediate unit U7~U11 (second layer). Similar processing is performed in the input units U2 to U6.

In the intermediate unit U7 of the intermediate layer (second layer),

(Equation 2) As for output to the respective output units U12~U14 output layer _{^{o 2 7 = f (i 2}} 7) ( third layer). Similar processing is performed in the intermediate units U8 to U11.

In the output unit U12 of the output layer (third layer),

(Equation 3) Is output as o ³ ₁₂ = f (i ³ ₁₂ ). Here, the output unit U12, is given a teacher signal y ₁₂ from the neural network outside, the output unit U12
Calculates the learning signal d ³ ₁₂ and calculates the connection weight correction amount Δw ² _i ³
₁₂ and determine the intermediate units U7 to U of the intermediate layer (second layer)
11 is corrected to w ² _i ³ ₁₂ + Δw ² _i ³ ₁₂ .

Here, the teacher signal is a signal representing a known result (here, the severity of cerebellar disease), and the optimization of the neural network structure is such that the neural network eventually converges to this teacher signal. The teacher signal y _j is given to the output unit Uj of the output layer.

The learning signal is a signal representing an error between the output value of the neural network and the teacher signal, and is represented by the following equation (6) in the output unit Uj of the m-th output layer. _{^{d m j = 2 (o m}} j -y j) f'(i m j) ··· (6) 1 -order differential function f'for f (x) of equation (2) (x) is expressed by the following equation (7 ), f'(x) = f (x) (1-f (x) because represented by) (7), the following equation equation (6) (8), d m j = 2 (o _{^{m j -y j) o m j}} (1-o m j) can also be expressed as (8).

The connection weight correction amount is a signal for correcting the weight value in order to construct a neural network having an optimum structure, and is a signal between the unit Ui of the (k-1) th layer and the unit Uj of the kth layer. Is given by the following equation (9). ^{_{Δw k-1 i k j =}} -εd k j o k-1 i ··· (9) (ε is a constant)

In the output units U13 and U14, all loads between the intermediate layer and the output layer are corrected by the same processing, and the load between the input layer and the intermediate layer is determined by the corrected load value. To correct.

That is, the unit Uj of the k-th layer and the unit (k
+1) The learning signal in the unit Uj of the k-th layer, taking into account the modified load w ^k _j ^{k + 1} _h between all the units Uh of the layer, is represented by the following equation (10):

(Equation 4) And by substituting d ^k _j obtained by equation (10) into equation (9), the load between all units Ui on the (k-1) th layer and units Uj on the kth layer is obtained. Is calculated.

For example, when data for which the result of the severe diagnosis is the middle part of the cerebellum for cerebellar disease, the teacher signal is y
₁₂ = 1, y ₁₃ = 0, and y ₁₄ = 0 (the signal is “1” to indicate the middle of the cerebellum, and the signal “0” is not the cerebellar hemisphere or spinal cerebellar degeneration). , the output value o ³ ₁₂ output units ^{_{U12~U14, o 3 13, o 3}} 14 neural networks outside extraction (Fig. 4 step S
4) It is determined whether an optimal neural network has been constructed by comparing y ₁₂ = 1, y ₁₃ = 0, and y ₁₄ = 0, respectively (step S5 in FIG. 4). If the error from the teacher signal is large, the process returns to step S3 in FIG. 4 for further optimization, and the neural network by the above-described weight correction is used by using the weight value corrected by the first learning. Repeat optimization.

If it is determined in step S5 in FIG. 4 that the error between the output value of the output unit and the teacher signal is sufficiently small, the optimization of the neural network is terminated, and the optimized neural network structure is processed by the arithmetic unit. 5 is stored in a storage device such as a hard disk (step S6 in FIG. 4).

In FIG. 5, the construction of a neural network for performing a severe diagnosis of a disease belonging to the category of cerebellar disease has been described. As shown in FIG. 6, the construction of a neural network for performing a severe diagnosis of a disease belonging to the category of Parkinson's disease Can be performed similarly. 5 and 6
Shows a trained and optimized neural network, which clearly shows that the weights of each neural network converged to different values and a unique neural network was constructed. .

FIGS. 7 to 9 show examples of graphs used to determine whether or not an optimal neural network has been constructed in step S6 in the case of cerebellar disease (when the neural network shown in FIG. 5 is used). It shows about.

FIG. 7 shows the teacher signal y ₁₂ (FIG. 7) in the output unit U12 which outputs an output value corresponding to the central part of the cerebellum.
(A)) and the neural network output value o ³ ₁₂ (FIG. 7)
It is a graph which compares (b)).

FIG. 8 shows a teacher signal y ₁₃ (FIG. 8) in an output unit U13 that outputs an output value corresponding to the cerebellar hemisphere.
(A)) and the neural network output value o ³ ₁₃ (FIG. 8)
It is a graph which compares (b)).

FIG. 9 shows a teacher signal y in an output unit U14 for outputting an output value corresponding to spinocerebellar degeneration.
₁₄ (FIG. 9A) and neural network output value o ³
₁₄ is a graph comparing FIG. 9 (b).

The horizontal axis in FIGS. 7 to 9 is common, and the horizontal axis in FIG. 7A is the teacher signal y (n) (n is the data number |
n = 0, 1, 2,..., 13), and the horizontal axis of (b) in each figure is the neural network output value o (n) (n is the data number | n = 0, 1, 2,. .., 13). That is, y (0) and y (1) are y in FIG.
_{1 2 = 1, y 13 =} 0 in FIG. 8, y ₁₄ = 0 in FIG. 9
Indicates that the severity is in the middle part of the cerebellum, and y (2) to y (7) are y ₁₂ = 0 and y ₁₃ =
A pattern of 1, y ₁₄ = 0 indicates that the severity is in the cerebellar hemisphere, and y (8) to y (13) are y
It indicates that severe is spinocerebellar degeneration by _{12 = 0, y 13 = 0} , y 14 = 1 that pattern.

The neural network output values for these teacher signals y (n) are o (0) and o (1) for y (0) and y (1), and both o ³ ₁₂ ≒ 1, o ³ ₁₃
≒ 0, o ³ ₁₄ ≒ 0, and it can be said that as a result of the learning, an almost correct result is output with respect to the input signal. Similarly, y
It can be said that o (2) to o (13) also output almost correct results for (2) to y (13), and among these output values, o (5) and o (5) in FIG. O in FIG.
As shown in (10), there is data whose error from the value of the teacher signal is so large that it cannot be ignored.

However, the neural network increases reliability by learning more patterns without bias in data, and thus the above-described error decreases as the learning amount increases.

Similarly, FIG. 10 to FIG.
Shows an example of a graph used for determining whether or not an optimal neural network has been constructed in the case of Parkinson's disease (when the neural network shown in FIG. 6 is used).

FIG. 10 shows a teacher signal y in an output unit U15 for outputting an output value corresponding to Parkinson positive.
₁₅ (Fig. 10 (a)) and neural network output value o
It is a graph which compares ³ ₁₅ (FIG. 10 (b)).

[0052] Figure 11 is a teacher signal y ₁₆ in the output unit U16 for outputting an output value corresponding to Parkinson negative a (FIG. 11 (a)) and the neural network output value o ³ _{1 6} (FIG. 11 (b)) the It is a graph to be compared.

[0053] Figure 12 compares the teacher signal y ₁₇ in the output unit U17 for outputting an output value corresponding to Parkinson negative b (FIG. 12 (a)) and the neural network output value o ³ ₁₇ (see FIG. 12 (b)) It is a graph to do.

The horizontal axis in FIGS. 10 to 12 is common, and the horizontal axis in FIG. 10A is the teacher signal y (n) (n is the data number | n = 0, 1, 2,...) , 13) and (b) in each figure.
Represents the neural network output value o (n) (n is the data number | n = 0, 1, 2,..., 13). That, y (0), y ( 1) is, y ₁₅ = 1 in FIG. 10, y ₁₆ = 0 in FIG. 11, it represents the severity is Parkinson's positive by pattern of y ₁₇ = 0 12 , Y (2) to y (7) are y
A pattern of ₁₅ = 0, y ₁₆ = 1, y ₁₇ = 0 indicates that the severity is Parkinson negative a.
(8) to y (13) are y ₁₅ = 0, y ₁₆ = 0, y ₁₇ = 1
Indicates that the severity is Parkinson-negative b.

From the learning results shown in FIGS. 10 to 12, it can be seen that the neural network outputs substantially correct results with respect to the input by learning, as in FIGS. 7 to 9.

Next, the severity diagnosis by the neural network shown in FIG. 3 executed after the optimal neural network is constructed by the arithmetic unit 5 will be described with reference to FIG.
This will be described with reference to FIGS. FIG. 13 is a flowchart illustrating the severity diagnosis by the neural network. FIGS. 14 and 15 perform the severity diagnosis by the neural network from the sway test data of the sickness of the patient whose severity is unknown. 7 is a graph of the resulting expression.

First, the arithmetic unit 5 displays a patient data input screen on the display unit 6, and displays the patient's severe diagnosis device 1 such as a doctor.
In addition to prompting the operator to enter necessary information (name, gender, age, etc.) regarding the patient (step S10), in addition to the disease name (category name of disease such as cerebellar disease), the operator inputs the disease name. Is read from a storage device such as a hard disk in the arithmetic unit 5 into the RAM inside the arithmetic unit 5 (step S1).
1) Further, in accordance with the input of the information on the patient from the operator, the patient's center of gravity sway test data to be subjected to the severe diagnosis is read from a storage device such as a hard disk in the arithmetic unit 5 (step S12).

Here, the neural network used in accordance with the disease name learns according to the flowchart of FIG. 4 to construct an optimal structure, and is stored in a storage device such as a hard disk in the arithmetic unit 5. It is. For example, if the input of the disease name from the operator is a cerebellar disease, a neural network corresponding to the cerebellar disease shown in FIG. 5 is read in response to the input.

Next, the patient's center-of-gravity sway test data read in step S12 and the outer peripheral area E.P. AREA, unit time locus length LNG / TIME, unit area locus length LNG / AR
EA, X-direction average oscillation displacement MX, Y-direction average oscillation center displacement MY, FIG.
Are input to the input units U1 to U6 of the neural network, and are calculated by the neural network of FIG. 5 (step S13), and the outputs from the output units U12 to U14 by this calculation are taken out of the neural network (step S14). The output result is displayed on the display device 6 and printed by the printing device 7 (step S15).

FIGS. 14 and 15 show examples of the result expressions printed by the printing device 7 in step S15. FIG.
5 is an example of a severe diagnosis result of a patient with a cerebellar disease diagnosed by the neural network structure illustrated in FIG. In FIG. 14, on the vertical axis, the middle part of the cerebellum is the value of o ³ ₁₂ output from the output unit U12, the cerebellar hemisphere is the value of o ³ ₁₃ output from the output unit U13, and the spinocerebellar degeneration is the output unit U14. is expressed as a number indicated on the horizontal axis the value of o ³ ₁₄ output from, in FIG. 14, the value of the cerebellum Central _{^{(= o 3 12) (≒}} 0.85) is at a maximum, severe cerebellar It is most likely to be in the middle, with spinocerebellar degeneration (≒ 0.3) and cerebellar hemisphere (≒ 0.
1), the likelihood decreases in order.

FIG. 15 shows an example of the result of a severe diagnosis of a patient with Parkinson's disease, which was diagnosed by the neural network structure shown in FIG. In this FIG.
On the vertical axis, Parkinson positive indicates the value of o ³ ₁₅ output from the output unit U15, Parkinson negative a indicates the value of o ³ ₁₆ output from the output unit U16, and Parkinson negative b indicates o ³ output from the output unit U17. The value of ₁₇ is represented by the numerical value shown on the horizontal axis. In FIG. 15, the value of Parkinson negative b (= o ³ ₁₇ ) (≒ 0.7) is the maximum, so the severity may be Parkinson negative b. Sex is the highest, and the likelihood decreases in the order of Parkinson negative a (ソ ン 0.2) and Parkinson positive (≒ 0.05).

As described above, in the apparatus for diagnosing a disease severity 1 according to the present embodiment, the arithmetic unit 5 includes the CPU, ROM, RA
M, and a hard disk, etc., as shown in FIG.
A load detecting function for detecting each load from each load data input from the A / D converter 4, a center of gravity calculating function for calculating a center of gravity from each of the detected loads, and a result of the calculated center of gravity calculation is stored. A center-of-gravity calculation result storage function, an analysis function of analyzing the stored center-of-gravity calculation result, a display function of displaying the analysis result on the display device 6, and a printing function of printing the analysis result by the printing device 7 are provided. The analysis function further includes an evaluation parameter extraction function shown in FIG. 3 for taking in the center-of-gravity calculation result stored in the RAM and extracting an evaluation parameter stored in the hard disk in advance. A neural network is input by using the input signal representing the disease name input from outside and the learning result by the back propagation method of the neural network as input. A severe diagnosis function by a neural network that have been diagnosed with severe disease were to have.

Therefore, the apparatus for diagnosing disease severity 1 according to the present embodiment makes it possible to perform a severity diagnosis by calculating a neural network using the data extracted by the evaluation parameter extraction function of the arithmetic unit 5, and Can output an objective severe diagnosis result without relying on the diagnosis by.

The center-of-gravity sway test data input to the input unit of the neural network according to the present embodiment includes an outer peripheral area E.V. AREA, unit time locus length LNG / TIME, unit area locus length LNG / ARE
A, the X-direction average oscillating center displacement MX, the Y-direction oscillating average central displacement MY, and the Romberg rate ROM are not limited, and can be appropriately set depending on the disease to be diagnosed severely. The number of intermediate layers, the number of units in each layer, and the like are also set, and an appropriate output for each disease can be obtained by a neural network having a structure reflecting these settings.

[0065]

According to the first aspect of the present invention, it is possible to output a result of objectively diagnosing the severity of a disease using the center of gravity calculation result by the center of gravity sway meter.

According to the second aspect of the present invention, a highly reliable objective diagnosis result of a disease can be output by using the established learning rule of the neural network.

According to the third aspect of the present invention, it is possible to output the result of the serious diagnosis of the disease in an easily understandable expression.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a main part of a disease severity diagnosis apparatus 1 according to an embodiment of the present invention.

FIG. 2 is a diagram showing a functional configuration of an arithmetic unit 5 in FIG. 1;

FIG. 3 is a diagram showing details of an analysis function block in FIG. 2;

FIG. 4 is a flowchart illustrating a learning function of the neural network.

FIG. 5 is a diagram showing a structure of a neural network for diagnosing the severity of cerebellar disease.

FIG. 6 is a diagram showing a structure of a neural network for diagnosing the severity of Parkinson's disease.

FIG. 7 is a graph comparing a teacher signal y ₁₂ (FIG. 7 (a)) and a neural network output value o ³ ₁₂ (FIG. 7 (b)) in an output unit U12 which outputs an output value corresponding to the central part of the cerebellum.

8 is a graph comparing a teacher signal y ₁₃ (FIG. 8A) and a neural network output value o ³ ₁₃ (FIG. 8B) in an output unit U13 that outputs an output value corresponding to a cerebellar hemisphere.

9 shows a teacher signal y ₁₄ in an output unit U14 that outputs an output value corresponding to spinocerebellar degeneration (FIG. 9A)
And neural network output value o ³ ₁₄ (Fig. 9 (b))
Graph to compare.

10 shows a teacher signal y ₁₅ in an output unit U15 that outputs an output value corresponding to Parkinson positive (FIG. 10
(A)) and a graph comparing the neural network output value o ³ ₁₅ (FIG. 10 (b)).

11 shows a teacher signal y ₁₆ in an output unit U16 that outputs an output value corresponding to Parkinson negative a (FIG. 11).
(A)) and the neural network output value o ³ ₁₆ (FIG. 1)
1 (b) is a graph for comparison.

12 shows a teacher signal y ₁₇ in an output unit U17 that outputs an output value corresponding to Parkinson negative b (FIG. 12)
(A)) and the neural network output value o ³ ₁₇ (FIG. 1)
2 (b) is a graph for comparison.

FIG. 13 is a flowchart illustrating a severity diagnosis using a neural network.

14 is a graph showing an example of a result of a severe diagnosis of a patient with a cerebellar disease that has been subjected to a severe diagnosis using the neural network structure shown in FIG. 5;

FIG. 15 is a graph showing an example of a result of a severe diagnosis of a Parkinson's disease patient who has been subjected to a severe diagnosis using the neural network structure shown in FIG. 6;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Severity diagnostic device of disease 2 Detection stand 3 Signal amplifier 4 A / D converter 5 Arithmetic device 6 Display device 7 Printing device

Claims (3)

[Claims] 1. A detecting table for detecting a load of a subject, a calculating means for sequentially obtaining a center of gravity of the subject from a load detected by the detecting table as a result of calculating a center of gravity by a predetermined calculating process, and the calculating means A gravity center calculation result storage means for sequentially storing the barycenter calculation results sequentially obtained by the method, and a disease severity diagnostic apparatus for diagnosing the severity of the disease based on a gravity center calculation result obtained from a gravity center sway meter, comprising: Input means for inputting a disease name of the patient; extracting means for extracting an evaluation parameter used for the severity diagnosis of a disease from the center of gravity calculation result stored in the center of gravity calculation result storage means; and an evaluation parameter extracted by the extraction means. Learning means for learning the severity diagnosis of the disease associated with the disease name input from the input means, based on the learning means; and learning results of the severity diagnosis of the disease learned by the learning means. By analyzing the evaluation parameters extracted by the extraction means, and performing a severity diagnosis of the disease associated with the disease name input from the input means; and a diagnosis of the disease performed by the severity diagnosis means. An output means for outputting a result of the severity diagnosis, comprising: a disease severity diagnosis apparatus. 2. The learning means performs learning of a severity diagnosis of a disease associated with the input disease name by using a learning method based on a predetermined neural network, and the severity diagnosis means learns by the learning means. The disease severity diagnosis apparatus according to claim 1, wherein the disease severity diagnosis is performed using the neural network. 3. The apparatus according to claim 1, wherein the output means outputs a ratio indicating each diagnostic symptom when the result of the severe diagnosis of the disease performed by the severe diagnostic means covers various symptoms. Alternatively, the disease severity diagnostic apparatus according to 2.