JP2863938B2 - Visual field measuring device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a visual field measuring device for measuring a range of a visual field of a human eye, and in particular, to a visual field abnormal pattern estimating unit using a multilayer neural network. The present invention relates to a visual field measuring device that can estimate an abnormal pattern of the visual field of a person to be measured by analogy, and can automatically determine an additional visual target.

"Prior art" Perimetry is considered to be extremely important in ophthalmic clinical examinations, and is used for diagnosis of glaucoma, retinal detachment, brain disorder, hysteresis, and the like.

The field of view is "the area where you can look at one point and see with one eye"
However, since the sensitivity of the retina is not constant even within the range of the visual field, the visual field is accurately referred to as “visual sensitivity distribution”. The perimeter used for this clinical examination includes a dynamic quantitative perimeter and a static quantitative perimeter.

A line connecting the points of equal sensitivity in each part of the retina is called an isosensitivity line isopter, and optotypes of various luminances and sizes are moved from the periphery to the center, and the first time the optotype is confirmed (or disappeared) Each isopter can be obtained by obtaining a figure in which the points of (h) are connected. By appropriately selecting this isopter, what obtains a visual field table is a dynamic quantitative visual field.For example, a target is presented at various brightness on a semi-circular screen, and the subject is displayed at each position. The Goldman perimeter shows the brightness (threshold) that starts to be felt and displays this in a diagram or the like. By using this figure, the measurer judged an abnormality in the visual field of the subject (a site with low sensitivity).

Static quantitative perimetry measures the sensitivity of the retina as the reciprocal of the luminance when the target is recognized for the first time after fixing the position of the visual field and increasing the luminance.

These perimeters include two types: manual and automatic.
There was kind.

The manual measurement is a method in which a measurer presents an optotype and confirms a response to determine a subsequent measurement point.

On the other hand, the automatic measurement is a method in which the order of presenting the optotypes is determined in advance by a program or the like, and the presentation is executed according to the order. This automatic perimeter can also easily perform static quantitative measurement by suprathreshold stimulation. The static quantitative measurement by the suprathreshold stimulus is a method of presenting a stimulus light brighter than the threshold and confirming that it is visible.

"Problems to be Solved by the Invention" In order to accurately judge the abnormality of the visual field, it is desirable to efficiently measure at the inspection position according to the symptoms. This type of manual perimeter has the merit that it can be measured at the measurement position according to the symptoms. Attention is required, and there is a problem that the burden on the measurer is extremely large. If you use a Goldman perimeter to examine the 360-degree range of the visual field every 30 degrees, you need to move the target repeatedly in 12 directions, and if you find an abnormal part, it will be in the vicinity Has to be measured in more detail, which is extremely time-consuming. Therefore, there is a serious problem that not only requires a great deal of effort from the measurer but also gives the subject a feeling of fatigue.

With the automatic perimeter, which is an auto-type, the burden on the measurer is reduced because the device itself presents the target automatically, but the problem is that measurement cannot be performed at the measurement position according to the symptoms. there were.

Therefore, it is possible to reduce the burden on the measurer and the person to be measured by analogizing the pattern of the visual field abnormality of the person to be measured and assisting the diagnosis of the measurer. There has been a strong demand for a visual field measuring device capable of determining a measuring point in response.

"Means for solving the problem" The present invention has been devised in view of the above-described problems, and a subject presenting unit that presents a predetermined basic measurement target, and the subject has recognized the presented target. It has a response unit that responds whether or not, an input layer consisting of input cells of the number of basic measurement targets, an output layer consisting of a hidden layer and an output layer of the number of abnormal patterns of the visual field, and the visual field of the subject is previously determined. A multi-layer neural network having a nerve weight determined based on the typical response in the abnormal case and the normal response and a basic measurement target are presented, and when the response is obtained, this is input data. The analog input unit of the above-mentioned multilayer neural network is input as an analogy unit for inferring the abnormal pattern of the visual field of the subject from the output of the output layer at that time, and is added according to the abnormal pattern of the visual field analogized by the analogical unit. The measurement target is determined and presented to the target display section. And an additional instruction unit.

The present invention may also include an input data determining unit that determines input data of an input layer of the multilayer neural network in response to an additional measurement target response in response to a response of a nearby basic measurement target. it can.

Further, the present invention provides an optotype presenting section for presenting a predetermined basic measurement optotype, a response section for responding whether or not the subject has recognized the presented optotype, and inputting the number of basic measurement optotypes. A neural network which has an input layer consisting of cells, a hidden layer, and an output layer consisting of output cells of the number of abnormal patterns of the visual field, and which is determined in advance based on a typical response when the visual field of the subject is abnormal and a normal response. While presenting a multi-layer neural network having a weighting factor and a basic measurement target, the response up to that point is input to the input layer of the multi-layer neural network as input data, and the input layer at that time is input. From the output of the analogy part to analogize the abnormal pattern of the visual field of the person to be measured, and determine a basic measurement target to be presented later according to the abnormal pattern of the visual field analogized by the analogous part,
And an additional instructing unit to be presented to the optotype presenting unit.

[Operation] In the present invention configured as described above, the index presenting unit presents a predetermined basic measurement target, and the responding unit responds as to whether or not the subject has recognized the measurement index. A multi-layer neural network having an input layer, a hidden layer, and an output layer having output cells with an abnormal number of patterns in the visual field, calculates a nerve weight determined based on the response between a normal visual field and an abnormal visual field. By inputting a response to the basic measurement target from the response unit to the input layer and sending an output from the output layer to the analogization unit, the analogization unit causes an abnormality in the visual field of the subject. The pattern can be inferred. Further, the additional instruction unit can determine an additional measurement target according to the abnormal pattern of the visual field analogized by the analogization unit, and can present the additional measurement target to the target presentation unit. Further, the input data determination unit can determine the input data of the input layer of the multilayer neural network in response to the response of the basic measurement target in the vicinity when there is a response of the additional measurement target.

Further, according to the present invention, while the analog inference unit presents the basic measurement target, the response up to that time is input as input data to the input layer of the multilayer neural network, and the output of the output layer at that time is used as input data. Estimate the abnormal pattern of the field of view of the measurer. Then, the additional instructing unit can determine a basic measurement target to be presented subsequently according to the abnormal pattern of the visual field inferred by the analogization unit, and can present the basic measurement target to the target presentation unit.

"Example" An example of the present invention will be described with reference to the drawings. 1 and 2 are views for explaining the mechanical configuration of the visual field measuring device 1 and the visual field measuring device 1 includes a visual field measuring device main body 100 and a gantry 200. The visual field measuring device main body 100 includes a housing 110, and the housing 110 is provided with a substantially hemispherical dome 120, and the inner surface of the dome 120 is a projection screen 130. A panel 140 is provided on the front part of the dome 120, and the panel 140 has a face receiving hole.
The face receiving hole 150 is provided with a face receiving member 160 for fixing the face. The forehead rest 170 and the chin rest 180 are attached to the face receiving member 160. The eye to be measured is a projection screen 130.
Is located near the substantially spherical center including the spherical center O.

As shown in FIG. 2, the housing 110 is provided with a projection optical system 300 on the rear side of the dome 120. The projection optical system 300 includes an illumination light source 310, a condenser lens 320, and an index 330.
, A collimator lens 340, a projection lens 350, and a reflection mirror 360. Here, the illumination light source 310, the condenser lens 320, the optotype 330, and the collimator lens 340,
It constitutes a first optical system for emitting the target light as a parallel light beam. The projection lens 350 and the reflection mirror 360, together with the rotating mirror 400 described later, receive a parallel light beam and pass the target projection light through the vicinity of the spherical center excluding the spherical center O, so that the target presenting point I on the projection screen 130 is displayed. The dome 120 has a guide hole 121 formed therein. The target projection light reflected by the reflection mirror 360 is
The light is guided to the rotating mirror 400 through the guide hole 121.

The rotating mirror 400 is provided in the vicinity of the spherical center excluding the spherical center O. When the rotating mirror 400 is rotated, the optotype presenting location I is changed.

Next, an electrical configuration of the visual field measuring device 1 of the present embodiment will be described with reference to FIG. The electric configuration of the visual field measuring device 1 includes a target projection control unit 2, a response switch 3, a BP (back propagation) execution unit 4, a target projection program storage unit 5, a response storage unit 6, It comprises an operation unit 7, a display unit 8, and a CPU 9. The optotype projection control unit 2
Target drive mirror drive motor 21 and visual field focus drive motor
22 and for controlling the index projection light source 23,
This corresponds to the optotype presenting unit. The optotype projection rotating mirror drive motor 21 drives the rotary mirror 400, and the optotype presenting location I can be changed by rotating the rotary mirror 400. The optotype focusing drive motor 22 is for moving the imaging lens 350, and is a distance L from the rotation center of the rotating mirror 400 to the optotype presenting position I.
Can be changed. That is, when the rotating mirror 400 is rotated by driving the target mirror rotating mirror drive motor 21, the distance L from the rotation center of the rotating mirror 400 to the target presenting position I changes. In order to correct the change in the distance L, the target focusing drive motor 22 is driven, and the imaging lens is driven.
It is necessary to move 350 in the optical axis direction. As a result, regardless of the change of the target presenting position I, an image of the target whose size and luminance hardly change can be displayed on the projection screen 130. The index projection light source 23 controls the lighting of the illumination light source 310, and can change the brightness (intensity) of the light source by adjusting the light source voltage as well as switching between lighting and extinguishing. Then, by adjusting the intensity of the illumination light source 310, the luminance of the target on the projection screen 130 can be changed.

The response switch 3 corresponds to a response unit, and inputs whether or not the subject has recognized the target. It is desirable that the response switch 3 be operated by the person to be measured. However, when the person to be measured is a child or an elderly person, the measurer receives recognition from the person to be measured and is operated by the person to be measured. Good.

The BP execution unit 4 is for executing a determination by back propagation using a predetermined nerve weight based on a typical response when the visual field is abnormal or normal and a normal response. Therefore, a typical response when the visual field is abnormal and the response of the subject are matched by the neural network, and an abnormal pattern of the visual field of the subject can be inferred.

The nerve weight used in the BP execution unit 4 is the following BP
Using the same neural network as the execution unit 4, it is predetermined based on a typical response when the visual field is abnormal and a normal response.

The optotype projection program storage section 5 stores a program for determining optotype presenting conditions for visual field measurement, and controls the optotype luminance, optotype position, lighting time, and the like to desired values. Can be done. As the target projection program, for example, a screening measurement program, a meridional program, a glaucoma program, or the like can be adopted.

The response storage unit 6 is for storing whether or not the subject has recognized the optotype. That is, this is for storing whether or not the response switch 3 has been activated for the presented target. The operation unit 7 is for performing the entire operation of the visual field measurement device 1, and includes various control switches, a light pen, and the like. The display unit 8 includes a TV monitor, a printer, and the like.
It is possible to display the type of target and the target distribution projected on 0, the command input from the control switch, etc., and also to display the result of the analogy of the abnormal pattern of the visual field of the subject. it can.

The CPU 9 controls various arithmetic operations of the visual field measuring device 1.

In the present embodiment configured as described above, when a measurement start command is input through the operation unit 7, the CPU 9 reads a control program from the optotype projection program storage unit 5 and executes it. CPU9
Controls the target projection control unit 2 according to the control program to drive the index projection light source 23. Further, by controlling the target projection rotating mirror drive motor 21 and rotating the rotating mirror 400, the position of the target is determined. Further, the CPU 9 drives the optotype focusing drive motor 22 in accordance with the rotation of the rotary mirror 400, and adjusts so that the luminance and size of the optotype do not change. The response storage unit 6 stores the response of the subject. And CPU9
Performs matching with a typical abnormal pattern and a normal pattern in the visual field, and infers an abnormal pattern in the visual field. Then, the display unit 8 displays the inferred abnormal pattern of the visual field and the like.

Next, the configuration of the multilayer neural network constituting the neural network will be described in detail with reference to FIG.

A neural network is composed of a plurality of nerve cells (neurons). One neuron consists of a cell body, dendrites (signal input part), and axons (signal output part).
It is composed of The axon (signal output part) is synapse-coupled with dendrites of other neurons, forming a network.

The learning method applied to the neural network is called a back propagation method. The configuration of the neural network is composed of an input layer A, an intermediate layer B, and an output layer C as shown in FIG. It has a multilayer structure. Although there is a connection between layers, there is no connection between units in the layer.

Since a neuron can be regarded as a multi-input / single-output nonlinear element, in other words, it can be regarded as an element having a “threshold action”. That is, the output pulse is turned on when the total amount of input signals is higher than the threshold, and the output is turned off when the total amount is less than the threshold.

Thus, the input signal _{S 1, S 2, S 3} , the output signal net against · · · S _n, the weighted sum of products It is described as follows. That is, the structure of the network can be changed by changing the weight (W).
The weight (W) takes positive, negative, and zero values, where zero indicates that there is no connection. The input / output characteristic function is sigmoi
d function is applied. This sigmoid function is a differentiable pseudo-linear function, for example, Can be adopted. The value range of this function is 0 to 1, becoming 1 as the input value increases, and becoming 0 as the input value decreases. And when the input value is 0, it is 0.
It is set to 5.

Next, the algorithm of the back propagation learning rule will be described. The intermediate layer B may be any number of layers, and assumes a network without feedback coupling (interlayer coupling). Here, the intermediate layer B may be called a (hidden layer), that is, a hidden layer.

(A) First, an input signal such as an image pattern is input to an input layer.

(B) Next, from the input layer A to the output layer C, the state change of each neuron accompanying the signal transmission process is sequentially calculated.

(C) The output of the j-th neuron of the output layer C obtained in (b) is _{defined as Opj} , the desired output (teacher signal) of the neuron with respect to the input signal is _{defined as Tpj,} and the square error of the following equation is evaluated. Define and operate as a function. The given image pattern is defined as p.

(D) changing the synaptic connections or weights of the network such that the cost function is at a local minimum (preferably a minimum) (ie, so that the actual output is as close as possible to the desired output).

That is, the strength of all the couplings may be changed so as to reduce the output error. Here, the change amount of the weight _Wji when the image pattern p is given is Is determined. If further _{modified, Δ p W ji = ηδ pj} O pj ··· (5) In addition, O _pi is the input value from unit i to unit j, [delta] _pj the unit j is either the output unit or the intermediate unit In the case of an output unit, δ _pj = (t _pj −O _pj ) f ′ _j (net _pj ) (6) Equation (7) is a recursive function.

The above is the basic algorithm of the back propagation method. Learning of each synaptic connection (correction of the weight) proceeds from the output layer to the input layer in the direction opposite to the signal propagation. This is why it is called back propagation. In this back propagation method, the calculation of ΔW is started from the output layer and proceeds to the unit of the intermediate layer. In the intermediate unit, the calculation cannot be performed unless the preceding stage ΔW is determined. Therefore (because it is recursive), calculations cannot be made without going back to the last input layer. Therefore, in the back propagation method, learning data is input, and the result is output (forward). Next, the strength of the bond is changed (backward) to reduce the resulting error.
Then, the learning data is input again. By repeating these, ΔW is determined so that the error is minimized.

Here, if indicated general formula _{ΔW, ΔW ji (n + 1} ) = ηδ pj O pj + αΔW ji (n) ···· (8) n is the number of times of learning, the first term on the right side [Delta] W, second The term is
This is an additional term for preventing error oscillation and accelerating convergence.

Next, a back propagation method employed for learning a multilayer neural network will be described with reference to FIG.

FIG. 5 illustrates the back propagation execution subroutine (SUBROUTINE BP). First, in S51, an initial value of the nerve weight is set by a random number. Then, in S52, the calculation is repeatedly performed for the number of learning data. Then S5
Calculate the output of each cell in step 3. In this embodiment, the above equation (2) is employed for calculating the output. And S5
In step 4, an evaluation function is calculated by the above equation (3). Next, in order to minimize the evaluation function value, a nerve weight correction amount is calculated in S55 according to the above equation (8). Note that the learning constant η and the stability constant α are empirically determined and, for example, η = 0.4 and α = 0.6. Then, in S56, the nerve weight is corrected based on the nerve weight correction amount, and the process returns to S52 to repeatedly calculate. Therefore, in this embodiment, the nerve weight is corrected for each learning data. Further, in this embodiment, S5
At 7, the sum of the evaluation functions of the respective data is calculated, and at S58, it is determined whether the sum of the evaluation functions is smaller than a predetermined threshold value. That is, S5
In step 8, it is determined whether the sum of the errors is equal to or less than a predetermined value. If the sum of the evaluation functions is not equal to or less than the threshold value, the calculation is performed again using the nerve weight at this time. When the sum of the evaluation functions is equal to or smaller than the predetermined value, the final nerve weight is output in S59. As a result, (SUBROUTI
NE BP) ends.

Next, the analogy of the abnormal pattern of the visual field using the above-described multilayer neural network and learning by the back propagation method will be described in detail.

First, a typical pattern when the visual field is abnormal will be described with reference to FIG. FIG. 6 shows the dynamic amount visual field of the right eye. In addition, since it is a right eye, the right outside is wide.

FIG. 6A shows a typical visual field pattern in the case of a central dark spot. This is the case when there is a scotoma in the center of the visual field, and is seen in cases of macular retina disorders, such as macular degeneration and central retinitis, and diseases of the papillary macular fiber bundle, such as acute retrobulbar optic neuritis. It is something that can be done. FIG. 6 (b) shows a typical visual field pattern in the case of an annular dark spot. This is the case where the scotoma surrounds the center and assumes a ring shape, and is also seen in glaucoma where Seidel-Bjerrum scotoma (a phenomenon in which the Marriott scotoma extends and expands vertically) and retinitis pigmentosa is there. FIG. 6 (c) shows a typical visual field pattern in the case of visual field narrowing. In the case of afferent stenosis,
The visual field in each direction is narrowed to the same extent from the periphery toward the center. For example, it is found at the end of glaucoma, retinitis pigmentosa and the like. In some cases, stenosis is partially observed (subsidence), and in retinal detachment, stenosis in a portion corresponding to the detached portion is recognized. FIG. 6D shows a typical visual field pattern in the case of semi-blindness. The semi-blindness is a state in which the visual field on the half side is lost from a vertical line passing through the fixation point due to an obstacle in a visual conduction path. FIG. 6E shows a typical visual field pattern in the case of a Bierm scotoma. The Bierm scotoma is a scotoma that is found in a region (Bierm region) in the middle from the top and bottom of the Marriott blind spot and is characteristic of the glaucoma visual field. It is caused by damage to nerve fibers in the vertical direction of the optic disc. FIG. 6 (f) is an arcuate scotoma. Bierm scotoma usually appears in one of the up and down directions, and an extreme case of this is an arcuate scotoma. It occurs in the case of glaucoma. FIG. 6 (g) is an intensity dark spot. The intensity scotoma is a shape in which the arcuate scotoma extends in either of the upper and lower directions to form a donut horizontally divided into two parts, and reaches the horizontal meridian opposite to the macula. The field of view at this time is also called the Renne stairs. FIG. 6 (h) shows the visual field in the case of normal. There is an oblong dark spot, which is a Marriott blind spot. Marriott blind spots occur because there is no photoreceptor cell in the optic disc and the corresponding field of vision becomes a scotoma.

Therefore, learning can be performed by using the above-described typical abnormal pattern of the visual field and the teacher signal. Here, the teacher signal is composed of bits corresponding to the number of abnormal patterns in the visual field. , 0, 0 ".

Next, an example in the case where an actual measurement is performed will be described with reference to FIG. First, FIG. 7 (a) shows an average value of a subject having a normal visual field and its standard deviation. Based on this,
Set four levels. The level 5 dB brighter than the normal average is LEVEL0 (normal), the level 3 dB brighter than this is LEVEL1, and the level 3 dB brighter is LE.
VEL2, 3 dB brighter are LEVEL3, and the maximum brightness is LEVEL4.

First, the target is presented at the luminance of LEVEL0, and if it is not visible, the luminance is increased to LEVEL1 and LEVEL2. Then, the mark of the level at which the subject is seen is plotted, and this plotting is performed for each target. FIG. 7 (b) shows an example of the case where the half-blindness and the visual field narrowing appear.

When using actual measurement results instead of typical abnormal patterns, the teacher signal may be, for example, "0.1, 0.0,
0.4, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0 ".

First Embodiment Next, the operation of the visual field measuring device 1 of the present embodiment will be described with reference to the flowcharts shown in FIGS.

FIG. 8 is a view for explaining the operation of the first embodiment in which all the basic measurement targets are presented, an abnormal pattern of the visual field is inferred, and an additional measurement target is determined. First, in step 1 (hereinafter abbreviated as S1), a program instructed by the measurer is called from the target projecting program storage unit 5 and executed. For example, a 71-point quick screening program can be executed. Next, in S2, the data of the program read in S1 is input as an input value. next,
The subroutine decision BP is read out in S3 and executed in S4. The SUBROUTINE judgment BP will be described later in detail with reference to FIG. By executing S4, the abnormal pattern of the visual field is determined. Next, in S5, it is determined whether or not the determination certainty in the determination in S4 exceeds a predetermined value. In the present embodiment, this is described as 50%. If the determination certainty is 50% or more, the process proceeds to S6. In S6, it is determined whether the determination name of the visual field abnormality is determined. If this judgment name is acceptable, the process proceeds to S7, where the judgment name of the abnormality of the visual field is determined, and the process ends. If it is determined in S6 that the visual field abnormality is undetermined, the process proceeds to S8. In S8, SU
Load BROUTINE part BP. And SUBROUTINE in S9
Execute the site BP. By executing the SUBROUTINE site BP, additional measurement sites to be added later are predicted. The SUBROUTINE site BP will be described later in detail with reference to FIG. That is, a portion to be additionally measured is predicted based on the abnormal pattern of the visual field determined in S4. For example, the abnormal pattern of the visual field determined in S4 is the sixth
If it is determined that it is likely to be a central scotoma shown in FIG.
Additional measurement sites to be added are densely arranged near the boundary region of the central scotoma. Further, when the abnormal pattern of the visual field determined in S4 is like an annular dark spot shown in FIG. Similarly, if it is determined that the visual field seems to be narrowed as shown in FIG. 6 (c), it is arranged at high density near the boundary area of the surrounding dark spots. If it is determined that it is likely to be a Bierm scotoma (probably due to glaucoma) shown in FIG. 6 (e), it is densely arranged near the Marriott blind spot and near the Bierm area (annular area near 10 degrees). If it is determined that the arc-shaped scotoma (probably due to glaucoma) shown in FIG. 6 (f) is likely, the dots are arranged at high density around the Bierm area. Further, if it is determined that the intensity scotoma (probably caused by glaucoma) shown in FIG. 6 (g) is likely, the scotoma is arranged more densely around the arcuate scotoma.
Then, in the case of the normal visual field shown in FIG. 6 (h), it is presented according to the basic program.

Next, in S10, a measurement point is set at an additional measurement site at random. Then, the measurement is executed in S11. Next, in S12, the data measured in S11 is added to the result of the previous visual field measurement and used as input data. Further, in S13, the subroutine determination BP is executed again to determine the abnormality of the visual field. Then, in S15, the determination certainty of the determination in S14 and the determination in S4 are compared, and if the determination certainty in S14 is low, the process proceeds to S16.
In step S16, the operator is prompted to determine whether to continue the measurement.If not, the process proceeds to step S17 to end the measurement using the current determination name as the visual field abnormality determination name and display the determination name, etc. 8 is displayed. If the measurement is to be continued, the process returns to S8 and executes the SUBROUTINE site BP.

In S15, if the determination certainty in S14 is high, the process proceeds to S18, and it is determined whether the determination certainty exceeds a predetermined value. If the certainty is exceeded, the process proceeds to S17 to end the measurement. If the diagnostic certainty does not exceed the predetermined value in S18, the process returns to S8 and executes the SUBROUTINE part BP.

Here, the SUBROUTINE site BP will be described in detail with reference to FIG. First, in step S1, the visual field measurement value is assigned to the extended input cell group corresponding to the position. Unlike the conventional back propagation configuration, this extended input cell group has a radial network configuration locally (small area around the measurement point of basic measurement) in the input layer. . Therefore, cells other than the cells corresponding to the basic measurement will be referred to as expanded input cell groups.

Next, in S2, the weight of the horizontal network is determined. Further, in S3, the output of the input cell to the additional input target is received, and the basic measurement cell input value is reset. Next, in S4, cells are prepared for the number of input cells, the number of intermediate cells, and the number of output cells used during learning. Then, in step S5, the neural weights connecting the cells, which are calculated in advance by the learning back propagation system, are read from the file, and the neural network is set. Further, in S6, each cell of the neural network configured in S5 outputs a response according to the input / output characteristic function. In S7, the output value of each cell in S6 and the corresponding abnormal pattern name in the visual field (judgment)
And exit.

Next, the SUBROUTINE site BP will be described with reference to FIG. This subroutine predicts an additional measurement site to be measured using a learning method based on back propagation. Therefore, it has the same configuration as that of the determination of the abnormal pattern of the visual field shown in FIG. That is, S1 to S6 are the same as the subroutine judgment BP in FIG. Then, in S7, the output value of each cell in S6 and the designated value of an additional measurement site to be measured thereafter corresponding thereto are returned as a return value, and the process ends.

Next, the handling of the response of the additional measurement point will be described in detail with reference to FIG. It is necessary to devise the response of the additional measurement point to give the response of the basic measurement target (the measurement target before the addition) near the additional measurement point. Therefore, an additional measurement target No. 1 as shown in FIG. 11 (a) can be set. Basic Measurement optotypes NO.1 around this additional measurement optotype NO.1 in the radius r _m
0, NO.11 and NO.12 are set so as to exist. And
Additional measurement target No.1 to basic measurement target No.10, NO.11, NO.1
If the distance to 2 is r ₁ , r ₂ , r ₃ respectively, then r ₁ ≦ r ₂ and r
And it has a ₁ ≦ r _3.

Then, the response of the additional measurement target and the response and the response of the basic measurement target are connected by a radial network. The weight of the network is determined according to the distance. As an example, a method of determining the weight in inverse proportion to the distance will be described. That is,
With the shortest distance as the minimum unit, the weight is determined according to the inverse ratio of the distance.

For example, in the case where the response of the basic measurement target No. 10 is 2, the response of NO.11 is 2, the response of NO.12 is 3, and the response of the additional measurement target No. 1 is 4, .10 input data, And the input data of NO.11 is And the input data of NO.12 is Can be determined.

That is, if the input data is represented by a general formula, Note that, as shown in FIG. 11 (b), areas centered on the basic measurement target are slightly overlapped. The response of the additional measurement target is devised so as to affect the response of the basic measurement target in the area in which it is included. That is, when a plurality of additional measurement targets exist in the area of a certain basic measurement target,
The response of the basic measurement target is corrected according to the inverse ratio of the distance with the shortest distance as the unit distance. The basic general formula is the same as the ninth formula.

The area is not limited to a circle, but may be a rectangle or the like. Although the size of the area can be determined as appropriate, the area near the center needs to be inspected precisely, so it is desirable to reduce the area.

Second Embodiment Next, a second embodiment of the visual field measuring device 1 will be described with reference to FIG. In the second embodiment, an abnormal pattern of a visual field is analogized while presenting a basic measurement target, and a basic measurement target to be presented is determined.

First, in S1, a measurement program instructed by the measurer is read from the optotype projection program storage unit 5, and several measurement points are executed. For example, perform the first 5 points of a 71-point quick screen. Next, in S2, the current data is added to the result of the previous visual field measurement to be used as input data. If the data is new, the data is used as it is. Then, in S3, the subroutine decision BP is called, and in S4, the SUBROU
Execute TINE Judgment BP to judge abnormal pattern of visual field by output cell value response value. Further, in S5, the judgment certainty is compared with the previous judgment certainty, and if it is smaller than the previous judgment certainty, the process proceeds to S6, and the examiner is prompted to determine whether to continue the measurement. If the examiner does not want to continue the measurement in S6, in S7,
The process ends with the current decision name as the final decision.

If the diagnosis certainty is larger than the previous time in S5, the process proceeds to S8. In S8, it is determined whether or not the determination certainty exceeds a predetermined value, and if it exceeds, the process proceeds to S7 to end the measurement.
If the determination certainty does not exceed the predetermined value in S8, the process proceeds to S9. In S9, call the SUBROUTINE site BP, and in S10,
ROUTINE Execute the part BP, predict the additional measurement part to be measured, and set a new measurement point. Then, returning to S2, the measurement is continued.

Note that the second embodiment has a simpler configuration than the first embodiment and has an effect that the measurement time can be shortened because automatic measurement is performed from the beginning.

Next, a modified example of the present embodiment will be described. First, the fixed target type visual field measuring apparatus 1 will be described with reference to FIG. The visual field measuring apparatus 1 of the above-described embodiment moves the optotype presenting position I by rotating the rotating mirror 400.
A light source like D can be embedded as a target. Thirteenth
FIG. 1A shows the appearance of an LED-type visual field measuring device 1. The visual field measuring device 1 is provided with a housing 500 and a panel provided with a circular hole 510 for receiving the face of a person to be measured and attached to the front side of the housing 500. 520, a recognition switch 530, and a measured eye fixing unit 540 for fixing the subject's eye to be examined in a predetermined position.
And an operation display device 550 formed on a side wall of the housing 500.
And a plurality of LEDs 560, 560,... Formed in the housing 500.
・ It is composed of a hemispherical dome with attached.

The operation display 550 includes a TV monitor 551 and a light pen 552.
And a printer 553, a control switch 554, and the like.

Next, the configuration of the electric system will be described with reference to FIG. The I / O interface 600 includes a light pen 552 operated by a measurer, an output signal from a control switch 554, a recognition signal from a recognition switch 530 for inputting whether or not the subject has recognized the target, and an initial signal. An initial measurement program selection signal from the measurement program selection unit 610 is input, the input signal is converted into a signal suitable for processing of an internal device, and the measurement result is converted into a signal that can be easily printed by the printer 553. Also, a response storage unit 645 for storing a response based on the recognition signal from the recognition switch 530 is provided.

The CPU 620 performs main control and the like of the visual field measuring device 1 of the present modification.

The LED matrix interface 630 is an interface for lighting the corresponding LED 560 under a predetermined condition in accordance with the optotype presenting condition read by the CPU 620, and forms a matrix with at least two transistor arrays.

The presentation condition storage unit 640 stores a program for realizing a target presentation condition (combination of lighting intensity, position, lighting time, and the like) that determines how the LED 560 of the hemispherical dome 570 is lit. For example, a screening measurement program and the like are stored.

GDC (Graphics Display Controller)
Inputs the LED array signal, the selected measurement program, the signal indicating the position of the LED being lit, and the response signal, and forms an image signal for displaying the information on the TV monitor 551. Then, this is output to the video memory 660. The timing control circuit 670 includes the clock oscillator 6
An appropriate timing signal is formed from the clock signal output from 80 and output to the GDC 650, the CPU 620, and the video memory 660.

The P / S converter 690 converts a parallel digital signal from the video memory 660 from parallel to serial to form a video signal, and outputs the video signal to the TV monitor 551.

An abnormal pattern analogy function of the visual field using the multilayer neural network of the present embodiment can be added to the LED-type visual field measuring apparatus 1 configured as described above.

Next, a modified example related to learning of a multilayer neural network will be described with reference to FIG. The above-described back propagation subroutine calculates an evaluation function for each learning data, corrects the nerve weight based on the evaluation function, and repeatedly performs the calculation. On the other hand, in the present embodiment, the nerve weight correction amount is calculated and stored in S55. Further, in S57, the sum of the evaluation function of each data is obtained, and it is determined in S58 whether the sum (Error) is smaller than a predetermined threshold value. If the sum is larger than the threshold value, each nerve weight correction is performed in S581. The sum of the quantities or the average value of each nerve weight correction amount is calculated. And in S56, S581
The respective nerve weights are corrected by the correction amount calculated in (1), and the calculation is repeated again using the nerve weights. Therefore, in this embodiment, after the nerve weight correction amount of all the learning data is calculated, it is corrected collectively. And S58
If it is determined that the sum of the evaluation functions has become smaller than the threshold value, the process proceeds to S59. The other steps are the same as those in the subroutine of FIG.

In this embodiment, an example in which the present invention is applied to a static perimeter is described. However, it is needless to say that the present invention is also applicable to a dynamic perimeter.

[Effect] The present invention configured as described above has a target presenting unit that presents a predetermined basic measurement target, and a responding unit that responds as to whether the subject has recognized the presented target. Has an input layer consisting of input cells of the number of basic measurement targets, a hidden layer, and an output layer consisting of output cells of the number of abnormal patterns of the visual field, and a typical response and normality when the visual field of the subject is abnormal in advance. When a multi-layer neural network having a nerve weight determined based on a response and a basic measurement target is presented and the response is obtained,
This is input to the input layer of the multilayer neural network as input data, and an analogy section for analogizing the abnormal pattern of the visual field of the subject from the output of the output layer at that time, and an abnormal pattern of the visual field analogized by the analogous section And an additional instructing unit that determines an additional measurement target in accordance with the above, and presents it to the target presenting unit. There is an outstanding effect that it is possible to assist the judgment of the abnormal pattern. Since the labor and time of the measurer are also reduced, the burden on the measurer and the person to be measured can be significantly reduced. Then, according to the inferred visual field abnormal pattern, an additional measurement target is automatically presented to the target presenting unit, so that a highly reliable judgment can be made and the measurement time is shortened, so that the measurement target can be shortened. It has the remarkable effect of reducing pain.

The present invention may also include an input data determining unit that determines input data of an input layer of the multilayer neural network in response to an additional measurement target response in response to a response of a nearby basic measurement target. Since it is possible to use the data of the additional measurement target appropriately, there is an effect that highly accurate measurement can be performed.

Further, the present invention provides an optotype presenting section for presenting a predetermined basic measurement optotype, a response section for responding whether or not the subject has recognized the presented optotype, and inputting the number of basic measurement optotypes. A neural network which has an input layer consisting of cells, a hidden layer, and an output layer consisting of output cells of the number of abnormal patterns of the visual field, and which is determined in advance based on a typical response when the visual field of the subject is abnormal and a normal response. While presenting a multi-layer neural network having a weight and a basic measurement target, the response up to that point is input to the input layer of the multi-layer neural network as input data, and the output layer at that time is input. From the output of the analogy part to analogize the abnormal pattern of the visual field of the person to be measured, and determine a basic measurement target to be presented later according to the abnormal pattern of the visual field analogized by the analogous part,
Since it consists of an additional instructing unit to be presented to the optotype presenting unit, the basic measurement optotypes are all presented, and the measurement time is shortened as compared with a visual field measuring device to which additional measuring optotypes are added. There is an outstanding effect that you can do.

[Brief description of the drawings]

FIG. 1 shows a visual field measuring apparatus according to one embodiment of the present invention. FIG. 1 is a diagram showing an external configuration of the present embodiment. FIG. 3 is a diagram for explaining an electrical configuration of the present embodiment, FIG. 4 is a diagram for explaining a multilayer neural network, FIG. 5 is a diagram for explaining a back propagation method, and FIG. FIG. 7 illustrates a typical example, FIG. 7 illustrates a specific measurement result, FIG. 8 illustrates an operation of the first embodiment, FIG. 9 illustrates a subroutine of determination BP, FIG. FIG. 11 is a diagram for explaining a subroutine of a part BP, FIG. 11 is a diagram for explaining an additional measurement target, FIG. 12 is a diagram for explaining the operation of the second embodiment, and FIG. 13 is a screen.
FIG. 14 is a diagram illustrating a modification using an LED, and FIG. 14 is a diagram illustrating a modification of the back propagation method. DESCRIPTION OF SYMBOLS 1 ... Visual field measuring device 2 ... Target projection control part 3 ... Response switch 4 ... BP execution part 5 ... Target target projection program storage part 6 ... Response storage part 7 ... Operation part 8 ... Display part 9 CPU 130 Shooting screen 300 Imaging optical system 330 Target 400 Rotating mirror

Claims (3)

(57) [Claims] 1. An optotype presenting section for presenting a predetermined basic measurement optotype, a response section for responding as to whether or not the subject has recognized the presented optotype, and inputting the number of basic measurement optotypes. A neural network which has an input layer consisting of cells, a hidden layer, and an output layer consisting of output cells of the number of abnormal patterns of the visual field, and which is determined in advance based on a typical response when the visual field of the subject is abnormal and a normal response. When a response is obtained by presenting a multi-layer neural network having a weight and a basic measurement target, this is input to the input layer of the multi-layer neural network as input data, and the output layer at that time is input. An analogy unit for analogizing the abnormal pattern of the visual field of the subject from the output, and an additional measurement target determined in accordance with the abnormal pattern of the visual field analogized by the analogous unit, and an additional instruction to be presented to the target display unit And a field of view measuring device. 2. An input data determination unit for determining input data of an input layer of the multilayer neural network in response to a response of an additional measurement target in response to a response of a basic measurement target in the vicinity thereof. Item 7. A visual field measuring device according to Item 1. 3. A target presenting unit for presenting a predetermined basic measurement target, a response unit responding whether or not the subject has recognized the presented target, and inputting the number of basic measurement targets. A neural network which has an input layer consisting of cells, a hidden layer, and an output layer consisting of output cells of the number of abnormal patterns of the visual field, and which is determined in advance based on a typical response when the visual field of the subject is abnormal and a normal response. While presenting a multi-layer neural network having a weight and a basic measurement target, the response up to that point is input to the input layer of the multi-layer neural network as input data, and the output layer at that time is input. An analogy part that infers the abnormal pattern of the visual field of the subject from the output of the subject, and determines a basic measurement target to be presented later according to the abnormal pattern of the visual field analogized by the analogous part, and determines the visual target presentation part. A visual field measuring device comprising an additional instruction unit to be presented.