JPH07136123A - Analyzer for disease of eyeground - Google Patents

Abstract

PURPOSE:To enable the determining of a retinal coordinate system automatically by arranging a feature extracting section to determine the position of yellow spot part and the position of a disk from a retinal image data of a testee to set a retinal coordinate system pertaining to the retinal image data based on the extracted positions and a specified distribution pattern of an optic nerve fasciculus. CONSTITUTION:A feature extraction section 110 of an arithmetic processor 100 determines the position of a yellow spot part and the position of a disk from a retinal image data of a tester. A retinal coordinate system is set pertaining to the retinal image data based on the position of the yellow spot part and the position of the disk extracted with the eyeground coordinate system determining section 120 and a specified distribution pattern of an optic nerve fasciculus. This analyzer enables the detection of optic nerve disk and yellow spot automatically from an image of the eyeground and also the determination of the retinal coordinate system automatically. Thus. a comprehensive judgment is possible by analyzing the retinal image data or the like according to the retinal coordinate system determined.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for analyzing fundus diseases, and more particularly to analysis of the condition of the optic nerve head and optic nerve fiber bundle defect obtained from color fundus images and measurement results of a perimeter based on a medical viewpoint. The present invention relates to an apparatus for analyzing a fundus disease that forms a fundus coordinate system useful for.

[0002]

2. Description of the Related Art In recent years, the number of patients with glaucoma has been rapidly increasing, and when the patient complains of subjective symptoms and visits an ophthalmologist, the stage of glaucoma is already close to the end stage, and 80% of all optic nerve fibers are present.
Most of them lost the above.

The interior of the eyeball is filled with the vitreous body and aqueous humor, and its shape is maintained by the pressure of them. The intraocular pressure, which is the pressure from the inside, has a certain physiological range. If this intraocular pressure is too high, the circulation in the eye is disturbed,
The atrophy occurs when the optic nerve is compressed. Glaucoma is a condition in which this intraocular pressure is too high to cause visual dysfunction, and it is a frequent cause of blindness.

[0004] An important point for this glaucoma is to prevent the occurrence and progression of optic neuropathy, and to achieve this purpose, make a diagnosis as early as possible, and perform appropriate management and treatment. .

Abnormal findings of glaucoma include various visual dysfunctions such as intraocular pressure, fundus and visual field, and each has its own characteristics. Therefore, the means for early detection of glaucoma is to comprehensively diagnose test results such as an early detection method by an intraocular pressure test, an early detection method by a fundus examination, and various visual function inspection methods including a visual field.

As a method for quantitatively grasping the position and size of the fundus lesion, an ophthalmologist has generally grasped the amount and the size of the eye lesion.

In addition, manually measuring the optic disc,
Some exist that automatically determine only the location of the optic disc. This automatic measurement of the optic disc position is performed by r (λ),
Detect the pixel with the highest g (λ) and b (λ),
Since that point is inside the optic disc, r (λ), g (λ), b (λ) in the vertical and horizontal directions based on this point.
The brightness values of are compared, and the portion where a change that exceeds a certain threshold value is generated is set as the position of the optic disc edge.

[0008]

However, the above-mentioned conventional technique has a problem that the optic disc and the macula cannot be automatically detected from the fundus image, and the coordinates cannot be automatically determined. . There is a serious problem that determination of coordinates by manual operation is extremely inefficient and poor in reproducibility.

There has been a strong demand for the appearance of a fundus disease analyzing apparatus capable of accurately considering the arrangement state of optic nerve fibers and incorporating the appearance frequency of each lesion site.

[0010]

The present invention has been devised in view of the above problems, and a feature extraction unit for determining a macular position and a nipple position from fundus image data of a subject is extracted. The fundus coordinate system setting unit sets a fundus coordinate system for the fundus image data based on the macula position, the papilla position, and a predetermined distribution pattern of the optic nerve fiber bundle.

The present invention also provides a feature extraction unit for determining the macular position and the nipple position from the fundus image data of the subject, and the optic nerve of the subject based on the extracted macular position and nipple position. An optic nerve fiber bundle distribution pattern forming unit that forms a distribution pattern of fiber bundles.

Further, the feature extraction unit of the present invention may be arranged such that the feature extraction unit extracts a blood vessel and extracts a focus position of the blood vessel as a papilla.

The feature extraction unit of the present invention can also be configured to detect the nipple position based on the lightness and saturation of the fundus image.

Further, the feature extraction unit of the present invention can be configured to determine the optic disc edge based on the amount of fall of the red component of the fundus image.

Further, the feature extraction unit of the present invention obtains a change in tone with reference to the papilla, the change in tone is linearly depressed, and the range of circular spread on the retina is the macula. It is also possible to extract as.

Further, the fundus coordinate system setting unit of the present invention uses the center of the papilla as a coordinate origin and scales a predetermined distribution pattern of the optic nerve fiber bundle according to the macula position and the papilla position to obtain a fundus coordinate system of curved coordinates. You can also set it.

[0017]

In the present invention configured as described above, the feature extraction unit determines the macula position and the nipple position from the fundus image data of the subject, and the fundus coordinate system setting unit determines the extracted macula position. , A fundus coordinate system relating to the fundus image data is set based on the papilla position and a predetermined distribution pattern of optic nerve fiber bundles.

Further, in the present invention, the feature extraction unit determines the macula position and the papilla position from the fundus image, and the optic nerve fiber bundle distribution pattern forming unit determines the subject based on the extracted macula position and the papilla position. Forming a distribution pattern of optic nerve fiber bundles.

The feature extraction unit of the present invention extracts a blood vessel and extracts the focus position of the blood vessel as a papilla.

The feature extraction unit of the present invention can also detect the nipple position based on the lightness and saturation of the fundus image.

Further, the feature extraction unit of the present invention can also determine the optic disc edge based on the amount of fall of the red component of the fundus image.

Further, the feature extraction unit of the present invention obtains a change in tone with the papilla as a reference, and the change in tone is linearly recessed, and a range of circular spread on the retina can be extracted as a macula. it can.

Further, the fundus oculi coordinate system setting unit of the present invention uses the nipple center as a coordinate origin and scales a predetermined distribution pattern of optic nerve fiber bundles according to the macula position and the nipple position to obtain a fundus coordinate system of curved coordinates. It is something to set.

[0024]

【Example】

A glaucoma analysis apparatus 1000 according to an embodiment of the present invention
Will be described with reference to the drawings.

The glaucoma analysis apparatus 1000 of the present embodiment corresponds to an analysis apparatus for fundus disease, and performs a processing calculation of fundus image data and visual field data of a subject by a computer, and an ophthalmologist comprehensively deals with glaucoma. It supports judgment.

As shown in FIG. 1, the glaucoma analysis apparatus 1000 comprises an arithmetic processing unit 100, a program memory 200 connected to the arithmetic processing unit 100, a data memory 300, and a display unit 400. There is. A perimeter 2000 and a fundus camera 3000 are connected to the glaucoma analysis apparatus 1000.

As shown in FIG. 2, the arithmetic processing means 100 includes a feature extraction unit 110 for determining a macular position and a papilla position from a fundus image, a predetermined optic nerve fiber bundle distribution pattern, and extracted macular position. And a fundus coordinate system setting unit 120 for setting a fundus coordinate system based on the nipple position, a visual field data conversion unit 130 for converting a given visual field measurement result based on the fundus coordinate system, and a fundus image data to the fundus. The influence part extraction part 140 for extracting the influence part affected by the disease and the analysis part 150 for associating the visual field data and the characteristic part based on the fundus coordinate system and analyzing the fundus disease. .

(Principle of Basic Configuration)

Next, the principle of the basic configuration used in the glaucoma analysis apparatus 1000 of this embodiment will be described in detail.

"A: fundus coordinate automatic creation block"

The fundus coordinate automatic creation block is composed of three steps of detecting the optic disc, detecting the macula, and setting the fundus coordinates.

As shown in FIG. 3, the fundus coordinate automatic generation block first starts execution in step 1 (hereinafter abbreviated as S1), detects the optic disc (FIG. 4) in S2, and then in S3. The macula is detected (FIG. 14). Further, the fundus coordinates are set in S4, and the execution is ended in S5.

(1) "Detection of optic nerve head (S in FIG. 3)
2) "

Next, the detection of the above-described optic disc will be described with reference to FIG.

First, the detection is started in S1, and the optic disc is detected by utilizing the color in S2. Next, in S3, the optic disc is detected using the vertical blood vessels. Then, in S4, the optic disc is detected by using the color in S2, and S3 is detected.
The result is compared with the result of the detection of the optic disc using the vertical blood vessel according to the above, and it is determined whether there is a misrecognition. That is, if the optic disc areas detected in S2 and S3 partially overlap, the optic disc detected by using the color in S2 is saved as a correct one. When the optic disc areas detected in S2 and S3 do not partially overlap with each other, the examiner confirms on the screen to detect the optic disc using color and detect the optic disc using the longitudinal blood vessels. It is determined whether any of the results of (1) is correct, or the examiner manually specifies it on the screen. By performing the determination process on the area of the optic disc in S4, the reliability of the detection result can be improved. Then, the process proceeds to S5 to end the detection of the optic disc.

"Detection of the optic disc by color utilization (Fig. 4
S2) "

Now, the detection of the optic nerve head by using colors will be described in detail with reference to FIG. First, in S1, detection is started, and in S2, fundus image data is converted from RGB space to HSI space, and hue (H: hue) and saturation (S: satur) are converted.
ratio) and a lightness (I: intensity) component are obtained. That is, the HSI conversion obtains data of hue, saturation, and brightness from the three primary colors of RGB (red, green, blue). Although there are various conversion methods for this HSI conversion, the bihexagonal pyramid model is adopted in this embodiment. This bi-hexagonal pyramid model has a structure in which two hexagonal pyramids are bonded together. In the lightness axis I forming the central axis of the hexagonal pyramid, one vertex is black with I = 0 and the other vertex is I. = 1 white

I = (max {R, G, B} + min
{R, G, B}) / 2

It is defined as

Here, the HSI conversion is

I = (Imax + Imin) / 2

However, Imax = max {R, G, B} Imin = min {R, G, B}

Then, Imax, Imin, R, G, B
Can also be used to determine S and H.

Next, in S3, since the optic nerve head is brighter than other portions, the maximum value of the lightness I is searched from the fundus image data to be searched, and is set as the high lightness point. Next, in S4, a labeling search range for extracting a bright portion as a region is determined. A minimum brightness value is selected from the eight pixels around the high brightness point, and the brightness range between the high brightness point and the point having the minimum brightness among the eight pixels is set as the labeling target brightness range.

Labeling is a work in which pixels adjacent to the high-brightness point for which the labeling search range is determined are in the same group and a label (number) is assigned to this group. If there are two or more high-brightness points that have the maximum value, labeling is performed for that number.

Then, in S5, the pixels included in the lightness range to be labeled from the high lightness point are labeled. Furthermore S
In 6, it is judged whether or not it has been repeated 8 times, and if it has not been repeated 8 times, the process returns to S3. The search for the high-brightness point in S3 at this time is performed by excluding the labeled fundus image data from the labeling processing target. If it is determined in S6 that the process is repeated eight times, the process proceeds to S7. In S7, focusing on the fact that the optic disc is highly bright and highly saturated, the average saturation is calculated for each label,
The label with the highest average saturation is selected as the optic disc.

Further, in S8, as shown in FIG. 6, a point having the highest lightness (high lightness point) among the labels having the highest average saturation among the labels selected in S7 is set as the optic disc innermost point.

(Determination of optic disc edge)

Next, the process of determining the optic disc edge will be described. First, we will extract the schematic shape of the optic disc.
In S9, a radial profile is obtained in steps of 360 degrees with the optic disc inner point selected in S8 as the center.
Here, the profile is a polygonal line that represents changes in brightness of continuous image data. This profile is R
Of the (red), G (green) and B (blue) components,
FIG. 7 shows the R component. Looking at this Figure 7, R
With the (red) component, the depression of the optic nerve head is larger than the depression of the blood vessel, so this R component is used. In addition, the amount of depression is as shown in FIG.

Then, in S10, the falling start position and the falling end position of the maximum dip amount are detected from each profile obtained in S9 in steps of 1 degree, and these ranges are set as the optic disc edge detection range. In S11, S1
The point having the maximum differential value is detected within the optic disc edge detection range set at 0, and this point is set as the optic disc edge. Further, the detected optic disc margins are linearly connected to all the profiles to form a schematic shape of the optic disc edge.

Since the approximate shape of the optic disc edge is determined in S11, the detailed optic disc edge is determined by utilizing this. In order to determine the optic disc edge, pay attention to the fact that the R component of the optic disc portion has a particularly large reflection, and in principle use the R image, but if the R image is overexposed or otherwise unusable, It is also possible to adopt a G image or a B image.

First, in S12, the approximate shape of the optic disc edge obtained in S11 is expanded to set a detailed search region for the optic disc edge. Here, by using the information of the fundus coordinate automatic creation function, the search area is limited, and the accuracy of the histogram analysis is improved by preventing the inclusion of extra information in other areas that adversely affect the analysis.

Next, in S13, histogram analysis is performed in order to determine the threshold value for binarization. Histogram analysis is for examining the distribution of brightness values of an image, and can be analyzed by measuring how many pixels having each brightness value are present in the image. As for the histogram of the optic papilla, as shown in FIG. 9 in order of increasing lightness when the region is appropriately set, a peak of the histogram showing the vicinity of the outer side of the optic disc edge and a peak of the histogram showing the vicinity of the inner side of the optic disc edge,
Divide into three peaks of the histogram showing the pale part.
Therefore, the lightness corresponding to the valley bottom existing between the peak indicating the outer side of the optic disc edge and the peak indicating the inner side of the optic disc edge is determined as the threshold value for binarization.

Next, in S14, the threshold value determined in S13 is used to binarize the image data, and noise appearing outside the optic disc due to reflection of the retina and blood vessels is removed. In this embodiment, the noise is removed based on the area and 200 pixels or less are processed as noise, but the noise may be removed by other methods.

Then, in S15, the image data binarized in S14 is subjected to expansion and contraction processing twice, so that the inside of the binarized optic disc area caused by the blood vessels inside the optic disc is processed. Can fill small holes.
Here, the contraction process is a process of removing all the boundary points of a given connected component to reduce it by one layer, and the expansion process is
On the contrary, it is a process for thickening one layer. By combining this expansion / contraction processing, small components and small holes in the binarized image can be detected and eliminated.

Further, in S16, only the edge of the binarized image data is extracted, the process proceeds to S17, the data extracted in S16 is labeled, and it is recognized as the optic disc edge.

Next, in S18, the center is determined from the optic disc edge recognized in S17. Although the center of gravity is determined by using the center of gravity in this embodiment, the intersection of line segments having the maximum longitudinal and transverse diameters of the optic disc edge can also be adopted.

Then, in S19, the detailed shape and coordinate information of the optic disc edge obtained by the above processing are stored, and then the process proceeds to S20 to end the execution.

"Optical nerve head detection process using blood vessels (S3 in FIG. 4)"

Next, the optic disc detection processing using blood vessels will be described with reference to FIG. In the optic disc, if the macula direction is the lateral direction from the optic disc, the blood vessels have a characteristic of branching in the vertical direction. A blood vessel that branches vertically from the optic nerve head is defined as a vertical blood vessel, and the vertical blood vessel is used to determine the optic nerve head.

First, detection is started in S1 and RG is detected in S2.
Color conversion is performed from the B image to an image in which the blood vessel level is higher than the retina. This color conversion is performed by, for example, R / G, r, 255-g,
Images such as R-G and r-g. In this embodiment, R / G
However, the treatment can also be performed using r, 255-g, R-G, r-g, or the like. Where r, g, b
Means that the brightness is excluded from the RGB image data,

R = R / (R + G + B)

G = G / (R + G + B)

B = B / (R + G + B)

It can be expressed as

Next, in S3, a vertical blood vessel is extracted. The extraction method will be described in detail later.

Then, in S4, the longitudinal blood vessel obtained in S3 is thinned as shown in FIG.

Further, in S5, the vertical blood vessels thinned in S4 are labeled. This labeling is performed from the end point or the intersection to the end point or the intersection as one label.

Then, in S6, the results labeled in S5 are linearly approximated by using the method of least squares or the like. If the number of pixels included in one label is large, the label may not be a straight line, and a straight line approximation may not be possible. That is, the approximated straight line is separated from the arrangement of blood vessels, and the approximated straight line does not pass on the optic disc. For this reason,
Although there is a method of adopting a higher-order least squares method, in the present embodiment, the label is divided into 20 pixels and linear approximation is performed so that the approximated straight line matches the arrangement of blood vessels in a small region, and the approximation is performed. The straight line makes it easy to pass over the optic disc. The labels with 20 pixels or less are left as they are.

Next, in S7, the straight line resulting from the linear approximation in S6 has a width around the center, and the thickened results are accumulated and added. Further, in S8, the place where the straight line with the largest cumulative addition result in S7 intersects the best is set as the top 3
The first to the optic disc areas are selected. Then, in S9, the optic disc area candidates obtained in S8 are stored. After saving the optic disc area candidates in S9,
The process proceeds to S10 and ends the process.

(Vertical blood vessel extraction processing)

Next, the vertical blood vessel extraction processing of this embodiment will be described with reference to FIG.

First, the detection is started in S1, and the vertical line component is detected in S2. This vertical line component is detected by using only the vertical line detection filter, and all pixels that have reacted with this filter are extracted. Next, in S3, line tracing is performed. In the line tracking of the present embodiment, when the entire image is scanned to search for a line element, the line tracking is performed with the search portion as the target pixel. In line tracing, as shown in FIG. 13, if there is a line element in the tracking image direction from the target image, line tracing is performed with the line element as the next target pixel. As a result of line tracing, if there is no corresponding line element, stop line tracing,
The number of pixels traced by the line is counted and stored. Then, the scanning is started from the point where the line tracing of the image is started, and after the entire image is scanned, the line tracing is completed.

Then, in S4, noise removal is performed. Since the image contains noise, the number of pixels Pi obtained by the line tracking processing in S3 and P of the following equation
If the number of pixels obtained in S3 is smaller than that of n,
It is determined to be noise and removed. In this example, a = 4096.

Pn = a / (20 * (R / G) +1)

(2) "Detection of macula (S3 in FIG. 3)"

The macula means the center of the center of the fundus and the surrounding area of a radius of about 2 nipples. The macula is a dense arrangement of cones and is responsible for visual acuity and color vision in the light. This macula is slightly darker than the surrounding retina, has a certain area, and its shape is a cone, and the macula is detected.

Next, the detection of the macula of this embodiment will be described with reference to FIG.

First, in S1, detection is started, and in S2, a profile is created at intervals of 360 degrees from the center of the optic disc. The range of this profile is as shown in FIG.
It is set between 100 and 270 pixels on the image. This is because the macula is about 2 papillary ears from the optic disc (about 170 in the image).
Since it is in the area of (pixels), it is possible to detect the fundus image data by limiting the range instead of detecting the entire fundus image data. Note that if the magnification of the captured image changes, the range also changes. Moreover, if there is left-right information of the imaged eye, the detection range can be further narrowed.

Next, in S3, a process of linearly approximating the lightness gradient is performed for each profile created in S2. This linear approximation of the lightness gradient is to remove the influence of shading. Note that shading refers to a phenomenon in which an edge of an image is darker, and shading correction is to perform density correction so that the entire image has uniform brightness. The slope of the straight line is calculated by calculating the difference value of all points in each profile and using the average value of the difference values. Note that the linear approximation of the lightness gradient can adopt not only the method using this difference value but also the least square method.

Then, in S4, a process for obtaining a threshold straight line is performed for each profile created in S2. In this processing, the standard deviation is obtained from the shading straight line obtained in S3 and the original data, and the straight line obtained by lowering the shading straight line by 2.5 times the standard deviation as shown in FIG. 16 is used as the threshold straight line. There is. This threshold curve is difficult to compare with a constant threshold, so
The comparison is made with a threshold straight line along the gradient of brightness.

Further, in S5, comparison processing and area addition are performed. Blood vessels that are darker than the retina must be removed for accurate detection of the macula. For this purpose, the macula and the blood vessel must be separated by utilizing the characteristics that the macula is round and takes a certain area and the feature that the blood vessel is elongated. That is, when the threshold straight line and the pixels lying on the threshold straight line are compared, the average brightness of the area 11 * 11 around the pixel to be compared is compared. If the comparison result is less than or equal to the threshold straight line, 1 is added to the area 11 * 11 around the comparison pixel. This process is performed for all the profiles obtained in S2.

Next, in S6, based on the result of cumulative addition in S5, as shown in FIG. 17, the maximum point of the cumulative addition result is determined as the macula central portion. By using the maximum point, it is possible to prevent erroneous detection of a small area such as an intersection of blood vessels. Then, the process proceeds to S7, the center position coordinates of the macula determined in S6 are stored, and the process ends at S8.

(3) "Determination of coordinate axes (fundus coordinate setting)"
(S4 in Fig. 3) "

The coordinate axes will be determined based on the results of the above-described detection of the optic disc and the macular portion. As for the coordinate axes, as shown in FIG. 18, the horizontal axis is a straight line passing through the center of the optic disc and the macula, and the vertical axis is a straight line perpendicular to the horizontal axis and passing through the center of the optic disc. The orthogonal coordinate axes are set with the macular-optic disc distance as a unit distance. However, this Cartesian coordinate system is almost always tilted with respect to the image,
Image transformations when matching Cartesian coordinates to images can also be included in the process.

A curvilinear coordinate system in which the coordinate axes are connected to the optic nerve fiber bundle distribution pattern as described in "D: Optimal nerve fiber bundle defect automatic analysis block" described later corresponds to the fundus coordinate system.

At the same time, with the center of the optic disc as the reference, the horizontal axis opposite to the macula is 0 degrees, the vertical axis upper side is 90 degrees, the macula side horizontal axis is 180 degrees, and the vertical axis lower side is 270 degrees. Set the angle of coordinates. In this specification, the angle of this coordinate system will be referred to as "intrapapillary angle" hereinafter.

"B: Pallet part automatic analysis block" (corresponding to one of the affected part extraction parts)

In glaucoma, when the optic nerve fiber bundle is deficient, the shape of the inside of the optic disc where the optic nerve fiber bundle is converged has an effect of enlarging the recess. The pale part does not mean a direct depression, but it is known to be highly related and is considered to be a clinically important factor.

Since the pallor is located on the optic disc margin, the pallor can be extracted by the same method as the principle of extracting the optic disc margin used in the fundus coordinate automatic creation block. In this embodiment, the image used for detecting the pale part is a G (green) image that is less likely to be saturated even in the bright optic disc.

In addition to the G (green) image, an image of lightness or saturation after conversion into the HSI space can be used.

Next, the detection of the pale part of this embodiment will be described with reference to FIG.

First, the detection is started in S1, and the G (green) image of the fundus image is called in S2. Next, in S3, the optic disc center coordinates are read from the setting of the fundus coordinates described above,
In S4, the optic disc edge coordinates are also read from the setting of the fundus coordinates described above. Next, proceeding to S5, the optic disc edge is set as the search area for the pallor area, histogram analysis is performed, and the threshold value is determined. If the region is set appropriately, the histogram of the pale part will be divided into three peaks, a peak of the histogram representing the vicinity of the nasal optic disc edge, a peak mainly representing the vicinity of the optic disc edge, and a peak representing the pale part. . In view of this point, the binarization threshold is used to determine the gray level corresponding to the valley bottom between the peak of the pale part where the gray level is the largest and the peak mainly in the middle and near the optic disc margin. Determine as a value.

Next, in S6, binarization is performed using the threshold value determined in S5, and the process proceeds to S7. In S7, a process of removing the noise caused by the reflection of the blood vessel or the like on the condition of the area is performed. In this embodiment, the binarized small area of 200 pixels or less is used as noise. Note that this condition can be changed and set as appropriate.

Then, in S8, it is determined whether or not to fill the binarization hole caused by the blood vessel existing inside the pale part. S
When it is determined in 8 that the hole for binarization is to be filled, the process proceeds to S9, and expansion / contraction processing is performed. By this expansion / contraction process, the holes in the binarized image can be eliminated. When the expansion / contraction process is completed in S9, the process proceeds to S10, and only the edge of the binarized region is extracted. Further, in S8, when it is determined that the binarization hole is not filled from the viewpoint that the pallor does not include the blood vessel region, the process proceeds to S10, and only the edge of the binarization region is extracted.

After extracting only the edge of the binarized area in S10, the flow proceeds to S11, and labeling is performed to recognize the edge of the pale part. As a result, data of the optic disc center, the optic disc margin, and the pallor of the pallor can be obtained.

Next, in S12, DD is performed for each optic disc angle.
Calculate (θ) and DP (θ). Here, DD (θ) is the distance from the optic disc center to the optic disc edge, and DP (θ) is the distance from the optic disc center to the pallor edge.

Next, in S13, based on the medical viewpoint that the enlargement of the pallor is related to the depression of the optic papilla,
We will calculate the change in the shape of the pale part of the optic nerve head. That is, assuming that the enlargement of the pallor is related to the defect of the optic nerve fiber bundle, the amount representing the change in shape of the pallor with respect to the optic disc is determined based on the values of DD (θ) and DP (θ). Calculate for each θ. In this embodiment, DD is used as an amount representing the change in the shape of the pale part of the optic nerve head.
(Θ) −DP (θ), DP (θ) / DD (θ), etc. are calculated for each optic disc angle θ.

Further, in S14, the area of the pallor and the area of the optic nerve head are calculated by counting the number of pixels, and the area of the pallor / the area of the optic nerve head is calculated. Then, the process proceeds to S15, and various calculated values are output so that the glaucoma can be analyzed based on the calculation results so far. In this embodiment, DD (θ) -DP (θ), DP (θ) / DD
(Θ), DD (θ), DP (θ), etc. are output for each optic disc angle, and also a suggestion degree indicating the degree of suggesting glaucoma is output. In addition, the pallor area / optic disc area can also be output.

In S16, the data output in S15 is DD (θ) -DP (θ), DP (θ) / DD.
(Θ), DD (θ), DP (θ), etc. are saved, and the process proceeds to S17 and ends.

In this embodiment, as shown in FIG. 20, the optic disc edge and the pale part edge are automatically extracted, and the optic disc internal angle θ is obtained.
It is also possible to output the glaucoma suggestion degree, the pallor area / the optic disc area, etc. for each.

[C: Optimal nerve disc edge blood vessel curvature automatic analysis block] (corresponding to one of the affected portion extraction units)

Since glaucoma lacks the optic nerve fiber layer, it also affects the shape of the inside of the optic disc where the fiber layer is converged, by expanding the recess. It is said that the optic disc blood vessels arranged on the optic disc tend to bend greatly near the optic disc edge as the degree of depression increases.

The image used takes advantage of the characteristic that the blood vessel portion has a large r component and a small g component, and in the present embodiment, the r-g component of 256 gradations is used. Where r,
g and b are obtained by excluding the brightness from the RGB image data,

R = R / (R + G + B)

G = G / (R + G + B)

B = B / (R + G + B)

It can be expressed as Therefore, r component, g
The components represent the ratio of the red component and the ratio of the green component on the chromaticity coordinate, and have a relationship of r + g + b = 1.

Next, the analysis of the degree of curvature of the optic disc edge in this example will be described with reference to FIG.

First, detection is started in S1, and in S2, the optic disc coordinates are read from the fundus coordinate automatic creation block.
Further, in S3, the detailed optic disc edge coordinates are read from the fundus coordinate automatic creation block. Next, in S4, a blood vessel is extracted, and a set of pixels connected in four directions (up and down, left and right) (near 4) is labeled. The blood vessel extraction method will be described in detail later.

Then, in S5, the peaks of the profile in the rg image of the portion where the blood vessel label extracted in S4 and the optic disc edge intersect are extracted and labeled from 1 to Nvd. Here, Nvd is the number of extracted optic disc border vessels. Here, the defect labeled as a defect so as to intersect with the optic disc edge is determined as the optic disc blood vessel.

In S6, each pixel extracted in S5 is set as the coordinate of the midpoint C shown in FIG. 22 in each optic disc edge blood vessel. Then, in S7 and S8, 1 to Nv
For the labels of the optic disc edges up to d, the blood vessel pixels are tracked inward and outward with respect to the center of the optic disc, and the internal and external points extracted by tracking a certain distance are respectively point I. Set to O. In this embodiment, the blood vessel is traced along the peak of the rg image, and the distance to be traced is within a range in which the distance from the nipple changes by ± 5 pixels from the midpoint C. If blood vessel tracking fails within this range,
Based on clinical knowledge, it is considered that the optic disc edge vessels wrap around to the back side of the retina due to the extreme enlargement of the depression. Together with the degree of curvature of the optic disc edges, the diagnosis of glaucoma is made. It can be useful information for support.

Next, in S9, with respect to the labels of the optic disc edges from 1 to Nvd, a straight line I-C which is a straight line connecting the midpoint C and the point I and a straight line which is a straight line connecting the midpoint C and the point O. The inclination on the coordinates of OC is calculated. Further, in S10, for the labels of the optic disc marginal blood vessels from 1 to Nvd, the intersection angle, which is the angle at which the straight line I-C and the straight line O-C intersect, is detected as the blood vessel bending degree VC (θ) in the intra-papillary angle θ direction. To do. This optic papilla margin degree VC (θ) is defined as the degree of curvature in the direction of point C.

Then, the optic nerve head margin VC (θ) obtained in S10 is stored in S11 so as to facilitate the integrated analysis of glaucoma, and the process ends in S12.

In S10, in order to utilize the detected data alone, the degree of optic disc margin VC (θ) and the degree of optic disc margin VC of each optic disc angle θ direction
It is also possible to calculate the average value of (θ) and display the output. Further, in the present example, the optic disc edge vascular tortuosity was determined based on the optic disc edge, but by using a similar method, it is also possible to determine the optic disc edge vascular flexibility to the pallor edge. Is.

"D: Automatic analysis block for optic nerve fiber bundle defect" (corresponding to one of the affected part extraction units)

As the optic nerve fiber layer is lost due to glaucoma or the like, the reflectance decreases and the color tone becomes darker. A glaucoma specialist can detect extremely subtle changes, but is generally said to be extremely difficult. This optic nerve fiber layer is also called an optic nerve fiber bundle defect because it is lost along the fiber bundle.

Next, the analysis of the optic nerve fiber bundle defect of this embodiment will be described with reference to FIG.

First, detection is started in S1, and in S2, the optic disc center coordinates are read from the fundus coordinate automatic creation block. Further, in S3, the macular center coordinates are read from the fundus coordinate automatic creation block, the optic disc-macular line is set as the horizontal axis, and the line orthogonal to the horizontal axis at the optic disc center is set as the vertical axis and the coordinate axis is set. This coordinate axis is linked to the optic nerve distribution pattern shown by Hogan as shown in FIG. 24 with the optic disc-macular distance as a reference length, and the curve coordinates along the optic nerve distribution pattern that matches the image data of this subject. A system is formed. Therefore, with respect to the optic nerve distribution pattern of the subject, a general optic nerve distribution pattern as shown in FIG. 24 is calculated from the macular position and the papilla position in the fundus image data of the subject on the fundus image data of the subject. It is formed by correcting the scale and inclination. This portion corresponds to the fundus coordinate system setting unit together with the fundus coordinate automatic creation block, and also corresponds to the optic nerve fiber bundle distribution pattern forming unit by itself.

Compared with the conventional method of searching radially from the center of the optic disc, it is possible to search directly on the optic nerve arrangement, noise is reduced, and a defect in the optic nerve fiber bundle is detected sensitively. can do. Furthermore, medical information such as the probability of the appearance of glaucoma scotoma in the visual field can be positively introduced into the image processing. In addition, by projecting the information of the optic nerve fiber layer to the information of the corresponding intra-papillary angle θ, it is possible to match the information with the intra-optic disc information, and it is possible to construct an integrated glaucoma analysis system. .

In S4, the optic disc edge shape is read from the fundus coordinate automatic creation block, and the inside of the optic disc edge is labeled as the optic disc portion. Further, in S5, blood vessels are extracted from the image data and labeled as blood vessel portions. Since the blood vessel part has a large r component and a small g component in chromaticity coordinates, in this embodiment, the input image signal is converted into an image of 256 gradations representing the r-g component,
The blood vessel portion is extracted from this converted image. In this case, the density difference between the retina portion and the blood vessel portion is about 20.
It is known to be about gradation.

Extraction of blood vessels will be described in detail with reference to FIG. In this embodiment, the line component in the image is defined as the blood vessel part. That is, a set of pixels forming a bright line drawn on a dark background of image density is used as a blood vessel.

First, detection is started in S1a, and in S2a,
Operate the bridge filter for all target pixels,
Extract ridges. The bridge filter used here is an extension of the general bridge filter (operation in two directions) shown in Material e3. It is operated in multiple directions so that it can be extracted regardless of the shape of the blood vessel. A threshold is provided to improve accuracy. In the present embodiment, with respect to the circumference of 7 pixels in diameter centered on the target pixel (L =
3) The ridge is extracted by operating in 10 directions and setting the threshold value to 15. That is, 2 * f- (a + b)> 15 in either direction out of 10 directions
When satisfying, this pixel is extracted as a ridge.

Then, in S3a, an expansion / contraction operation is performed to fill the fine holes in the ridge image to smooth the ridge image,
Proceed to S4a. In S4a, the direction components of all the ridges are extracted. That is, the distribution of ridges in 20 directions is searched within the range of 20 pixels in diameter centered on the target ridge, and the component in the direction with 90 degrees added to the direction with the smallest ridge distribution is extracted. It has become like.

Next, in S5a, S
The directional component extracted in 4a and the ridge distribution in the direction intersecting at right angles are searched to determine whether or not each ridge is a blood vessel.
That is, since the ridge distributed in the center of this ridge distribution represents the cross section of the width of the blood vessel portion, those which do not fit the expected width of the blood vessel are excluded. In the present embodiment, those having a constant angle of view and a maximum width exceeding 12 pixels are excluded.

Further, in S6a, the expansion / contraction operation is performed again to fill the fine holes in the ridge image and smooth it. Then, the process proceeds to S7a, and in S7a, the image smoothed in S6a is used as a blood vessel image and is labeled as a blood vessel portion. Then, the process proceeds to S8a to end the blood vessel extraction process.

By the above processing, blood vessels can be extracted. Here, the analysis of the optic nerve fiber bundle defect in FIG. 23A will be described again. In S5, blood vessels were extracted from the image data and labeled as blood vessel portions.
Proceeding to step 6, the peripheral portion of the blood vessel label is labeled as the peripheral portion of the blood vessel. The peripheral portion of the blood vessel label of this embodiment is 5
It is assumed that the distance is within pixels.

Next, in step S7, the pixels that have an adverse effect on the analysis are excluded. In this embodiment, S4, S5, S
The optic papilla, the blood vessel portion, and the blood vessel peripheral portion extracted in 6 are excluded, and the subsequent processing is performed so that the pixels target the retina portion. Then, in S8, the input image is set to B + of 64 gradations.
Convert to G image. This B + G image corresponds to a commonly used red-free projection photograph, in which the optic nerve fiber bundle defect portion is emphasized and the influence of choroidal blood vessels is small.

Further, in S9, while simulating the optic nerve fiber bundle on the image coordinates, the image density is scanned along the scanning path to obtain the average image density for each fiber bundle. FIG. 25 is an original image converted so that the horizontal axis represents the intra-papillary angle and the vertical axis represents the distance along the optic nerve fiber bundle, and the optic nerve fiber bundle defect converted into a rectangle appears near the left quarter. ing.

Here, the optic nerve fiber bundle will be simulated based on the optic nerve diagram by Hogan.
This simulation of the optic nerve fiber bundle is performed with the intra-papillary angle θ.
It is executed such that the coordinates of the input image are made to correspond to each other.

As shown in FIG. 24, the optic nerve fiber bundle has a peculiar point at the point A, and the optic nerve fiber bundle has a curved shape. Therefore, from the macula to the point A, the optic nerve fiber bundle can be approximated by a relatively simple equation of motion, but to approximate the optic nerve fiber bundle after the point A by a mathematical expression, a slightly complicated equation is used. Must be used. However, in a fundus photograph with a general angle of view and composition, only the region from the macula to the point A is photographed, and therefore the following two optic nerve fiber simulations are used properly.

That is,

(1) It is approximated by a motion equation relating to the intra-nipple angle θ. That is, since the arrangement of the optic nerve fiber bundle resembles the trajectory of the ball when the ball is rolled on the floor depressed like a mortar, it is possible to create a motion equation. This equation can be calculated for any intra-nipple angle θ.

(2) Each function for each intra-papillary angle θ is approximated by the method of least squares. In this case, the optic nerve fiber bundle after the point A can be accurately approximated.

First, the equation of motion regarding the intra-nipple angle θ of (1) will be described. The locus of the ball when the ball is rolled from the center of the optic disc at an initial velocity V ₀ and an angle θ corresponds to the optic nerve fiber bundle on the floor where the macula is depressed like a mortar.

The trajectory of the ball at that time can be obtained based on the following equation of motion.

F = ma = mg / r ² −μmV

When the nipple portion is regarded as a circle having a certain area, the following approximation can be performed. A mortar-shaped floor surface that produces an acceleration of G / R ³ toward the macula, and a floor surface having friction that is inversely proportional to the square of the distance r from the depression, and the optic disc at an initial velocity V _0. The trajectory of the ball when the ball is rolled corresponds to the optic nerve fiber bundle.

The trajectory of the ball is obtained based on the following equation of motion.

Ma = mg / r ³ −μ ′ mV / r ²

Where m is the mass of the ball, μ and μ ′ are friction coefficients, g is gravitational acceleration, r is the distance from the center of the depression,
v is the velocity.

Next, in (2), the required optic nerve fiber bundle for each intra-papillary angle θ is read from the optic nerve diagram by Hogan, and this is applied to the 8th order equation in this embodiment, as shown in FIG. , Each coefficient of the function for each intra-nipple angle θ is determined by the least squares method. Originally, the entire range of the image can be expressed by the above formula,
Due to uneven lighting and the like, the reliability of the density information of the peripheral image is low, even if the image is of high quality. Furthermore, the retina itself has a color distribution, for example, the area around the macula is dark and the area along the arcade is bright. Therefore, when it is desired to widen the range to which the above equation is applied, it is necessary to measure the illumination light distribution and the retina color distribution in advance and perform correction. In this embodiment, the above formula is applied up to 0.4 times the distance from the center of the optic disc to the fovea.

Next, in S9, the average image density for each fiber bundle is obtained, and then the process proceeds to S10, in which a graph showing the relationship between the intra-nipple angle θ and the average image density is created. Depending on the intra-papillary angle θ, there may be no retinal image on the arranged route, and in this case, the average image densities at both ends are linearly complemented.

Further, in S11, the relationship between the intra-nipple angle θ and the average image density obtained in S9 is converted into the relationship between the intra-nipple angle θ and the optic nerve fiber bundle defect suggestion degree NFBD (θ). It is supposed to be displayed in the graph as shown. This conversion is for introducing the appearance probability of the optic nerve fiber bundle defect location obtained by the study of Aulhorn et al.

The concrete conversion method is as follows. First, the intra-nipple angle θ
Invert the graph of the relationship between and the average image density. Width 4
The range of the portion corresponding to 5 o'clock to 8 o'clock and the portion corresponding to 10 o'clock to 1 o'clock of the difference graph with the expected graph shape are smoothed by applying an average filter of 0 unit θ and the intra-nipple angle θ and the optic nerve It is a graph showing the relationship with the fiber bundle defect suggestion degree NFBD (θ).

Then, in S12, a single-direction bridge filter is operated on the previously calculated graph of the optic nerve fiber bundle defect suggestion degree NFBD (θ), and the fibers showing a concentration drop above the threshold value are detected. The bundle is extracted as an optic nerve fiber bundle defect portion, and the result is output as an image as shown in FIG. 28. In the present embodiment, the threshold is set to 20 gradations by operating the bridge filter of width 40 unit θ and unidirectional.

Further, in S13, the relationship between the intra-papillary angle θ and the optic nerve fiber bundle defect suggestion degree NFBD (θ) obtained in S11 is stored for application to the glaucoma integrated analysis, and S14 is stored.
Ends the process.

[E: Field of view automatic analysis block]

When the optic nerve fiber bundle is lost due to glaucoma,
Abnormality of visual field peculiar to the defective shape is recognized. Since the visual field measurement is a subjective test, it is unavoidable that noise is mixed in, the reproducibility is slightly poor, and there is also the problem that the interpretation of the visual field is empirical, but in the clinical examination of glaucoma, along with the fundus photograph, This is a very important test method.

Here, as shown in FIG. 34 (a), the visual field data generally has the line-of-sight position (macular position) as the coordinate origin, the horizontal axis passing through the line-of-sight position is the horizontal axis, and the vertical axis is the vertical axis. It is displayed on the longitude and latitude lines.

On the other hand, the fundus image is as shown in FIG.
As shown in Fig. 3, the image center is generally located around the middle position between the optic disc center and the macula, which is different from the center of the visual field data. An automatic analysis was tried in view of the difficulty of making a judgment.

Next, the automatic analysis of the visual field of this embodiment will be described with reference to FIG.

First, the processing is started in S1, and the visual field data is input in S2. The visual field data is input from the automatic perimeter, but in this embodiment, the screening program for the automatic perimeter is used. Note that the automatic perimeter can use not only the screening program but also a detailed measurement program that is a threshold measurement.

In S3, unit elements (optic disc center Ap, macula position Bp, and their interval Lp) necessary for matching the visual field data to the fundus coordinate system set based on the fundus image are extracted. Specifically, the blind spot position Ap corresponds to the optic disc center, the visual field center position Bp corresponds to the macula position, and the interval Lp between them is determined.

At S4, the scales of the positions of the fundus image and the visual field data are adjusted. Specifically, among the unit elements obtained from the visual field data, the distance L between the optic disc center and the macula position
If p is different from the distance Li between the optic disc center Ai and the macula position Bi in the fundus image data in which the fundus coordinate system is set, the mutual position of the visual field data is enlarged or reduced as a whole. , And the distance Li is adjusted to match the scale. For example, the position of the visual field data is the coordinate value (x, y),
If it is displayed, the coordinate value is (x * (Li / L
p), y * (Li / Lp)), the scale can be adjusted.

In S5, the respective positions of the fundus image data and the visual field data whose scales are matched are matched. Here, the visual field data corresponds to an inverted image obtained by vertically inverting the fundus image data. Therefore, the visual field data is re-arranged rotationally symmetrically about the horizontal axis passing through the line-of-sight position.

Specifically, first, the visual field data is converted into rotational symmetry about the horizontal axis passing through the line-of-sight position. And
Regarding the fundus image data and the visual field data scaled to this, the line that connects the optic disc center Ai and the macular portion position Bi in the fundus image data with the optic disc center Ap that is the origin of the fundus coordinate system is the same in the visual field data. The optic disc center Ap and the line connecting the macula position Bp are matched so that they are combined. As a result, the fundus image data and the visual field data are expressed on the common fundus coordinate system.

In S6, the independently obtained fundus image data and visual field data are associated with each other on the common fundus coordinate system, and the combined data is displayed as shown in FIG.

Next, in S8, the optic nerve fiber bundle simulation used in the optic nerve fiber bundle defect automatic analysis block is called, and the line connecting the Marriott blind spot center and the macula is set as the horizontal axis, and the Marriott blind spot center is set as the horizontal axis. A coordinate system is set up with the orthogonal line as the vertical axis.

Then, in S8, as shown in FIG. 30, the optic nerves in the optic disc intra-papillary angle θ direction are associated with the visual field data. Further, in S9, the degree of continuous defect of the optic nerve fiber bundle is calculated. Here, the degree of continuous defect will be described in detail. To determine the degree of consecutive defects, if the values at adjacent measurement points both indicate visual field defects, use a calculation formula that calculates a positive contribution corresponding to the size of the mutual defects and determines the placement of the optic nerve fibers. The sum is calculated by sequentially calculating the measurement points appearing along with. That is,

Degree of continuous defect = sigma (visual field defect (J,
θ) * visual field loss (J + 1, θ) / ((number of visual field data −
1) * Maximum visual field loss * Maximum visual field loss))

However, sigma is a summation symbol, θ is an intraoptic disc angle of each optic nerve fiber bundle, J is a position of a visual field measurement point corresponding to the optic nerve fiber, and numbers are sequentially given from the center of the optic disc to the far direction. There is. Further, the average degree of deficiency may be used in addition to the degree of continuous deficiency.

As described above, by projecting the information of the optic nerve fiber obtained from the visual field to the information of the corresponding intraoptic disc angle θ of the optic nerve fiber bundle, the information can be matched with the intraoptic disc information of the optic nerve fiber bundle and integrated glaucoma analysis can be performed. It can be performed. Note that the visual field measurement is a subjective test and requires measurement time.
There is a concern that the person to be measured may suffer from the measurement for a long time. Therefore, in many cases, the number of measurement points is limited in order to reduce the pain caused by long-term measurement. Therefore, the number of measurement points for the arrangement of one optic nerve fiber must be reduced to a smaller number, and when the defect is short, the continuity calculation may become too rough and it may be impossible to deal with it. In this example, by introducing a “optic nerve fiber band” model in which optic nerve fibers are summarized to some extent, and increasing the density by increasing the number of visual field measurement points corresponding to this band, it is possible to cope with short defects. It is configured.

Next, in S10, from the optic nerve fiber bundle simulation used in the analysis of the optic nerve visual field defect, a setting in which the arrangement of the optic nerve fiber bundle as shown in FIG. 30 is divided into bands in advance is called. With respect to the band division model called in S10.

Then, in S12, the degree of continuous defect of the optic nerve fiber bundle is calculated for each band section, as in S9. In addition, when the values of adjacent measurement points are both visual field defects, the continuous defect degree is calculated by using a calculation formula that positively contributes to the size of the mutual defect and calculates The measurement points appearing along the arrangement are sequentially calculated and the sum is calculated.

Degree of continuous defect = sigma (visual field defect (J, b
and) * visual field loss (J + 1, band) / ((number of visual field data-1) * maximum visual field loss * maximum visual field loss))

However, sigma is a summation symbol, and band
Is the optic nerve fiber bundle intra-papillary angle θ corresponding to the optic nerve fiber representing each optic nerve fiber band of the optic nerve fiber bundle simulation used in the analysis of the optic nerve defect. Further, J is the position of the visual field measurement point corresponding to the optic nerve fiber band, and is sequentially numbered from the center of the optic nerve head in the far direction. Further, the average degree of deficiency may be used in addition to the degree of continuous deficiency.

As described above, in S12, the degree of suggestion as to whether or not the optic nerve fiber bundle and the optic nerve fiber band are continuously lost can be calculated, that is, the degree of suggestion of glaucoma can be calculated from visual field data. You can Here, the degree of suggestion of a continuous loss of the optic nerve fiber bundle or the like is defined as a continuous loss degree VD (θ). Then, in S13, the optic nerve fiber bundle obtained in S12,
The continuous loss information for each optic nerve fiber band, the optic nerve fiber bundle intra-papillary angle θ corresponding to this continuous loss information, and the like are saved for integrated analysis of glaucoma, and the process proceeds to S14 and ends.

In S9, the continuous defect degree of the optic nerve fiber bundle discrete in the angle direction is continuously calculated in the angle direction by using the interpolation method so that the glaucoma can be analyzed only by analyzing the visual field. It can be converted into a continuous defect degree of the optic nerve fiber bundle and output.

[F: Automatic integrated analysis block]

Next, the automatic integrated analysis of this embodiment will be described with reference to FIG.

First, in S1, the process is started, and in S2, information is input from the pale-white part automatic analysis block. That is, the optic disc center-optic disc edge distance DD for each optic disc angle θ
(Θ), optic disc center-pallor edge distance DP (θ), D
Input D (θ) -DP (θ) and DP (θ) / DD (θ).

Next, in S3, information is input from the optic disc edge blood vessel curvature automatic analysis block. That is, the degree of blood vessel curvature V for each optic disc angle θ where blood vessels exist at the optic disc edge
C (θ) is called and input.

Further, in S4, information is input from the automatic analysis block for optic nerve fiber bundle defect. That is, the optic nerve fiber bundle defect suggestion degree NFBD (θ) for each optic disc angle θ is called and input.

Then, in S5, information is input from the visual field automatic analysis block. That is, the continuous defect degree for each optic disc angle θ corresponding to the arrangement of the optic nerve fiber bundle where the visual field measurement point exists, and the continuous defect degree VD (θ) for each optic disc angle θ corresponding to the called optic nerve fiber bundle are called. , Is designed to be input.

Next, in S6, the various kinds of glaucoma suggestion data input in S2 to S5 are aligned and output for each optic disc angle θ as shown in FIG. The direction of the optic nerve fiber bundle defect can be inferred from the degree. In FIG. 32, the Biermu region (110 degrees to 140 degrees) where glaucoma symptoms are likely to appear
, And 220 to 250 degrees). In addition, the extent of the optic nerve fiber bundle defect can be inferred from the extent of the overlapping peaks. Further, the degree of progression of optic nerve fiber bundle defect can be analogized from the intensity of peak overlap.

That is, by using the computer to collectively process the above-mentioned suggestion data, which are multiple and generally have different coordinate systems, as information for each optic disc angle θ, glaucoma can be seen in a fundus photograph. Although a little sex is seen, even if it is hardly seen in the visual field inspection,
Glaucoma represented by optic nerve fiber bundle defect can be grasped comprehensively.

Even when an ophthalmologist observes a fundus photograph with the naked eye, or when the possibility of glaucoma is not found only by the visual field examination result, the optic nerve fiber bundle is processed by performing image processing on the fundus image. In some cases, it is possible to understand the effects of glaucoma represented by a defect, and it is possible to assist an ophthalmologist in diagnosing glaucoma. As the perimeter, not only a general automatic perimeter but also a flicker perimeter for detecting glaucoma, a pyramid perimeter, a combination with a contrast sensitivity measurement and the like can be used.

Then, in S7, the degree of overlap in the peak direction of the glaucoma suggestion degree obtained in S6, the width of the overlapped peaks, and the strength of the peak overlaps are automatically calculated and displayed by an appropriate method. It has become.

(Operation of Example)

Next, the operation of the glaucoma analysis apparatus 1000 of this embodiment will be described with reference to FIG. First, the perimeter 2000 and the fundus camera 3000 are connected to the glaucoma analysis apparatus 1000.

Then, the analysis is started in S1, and the arithmetic processing unit 100 reads the program from the program memory 200 and starts the execution. Next, in S2, the fundus camera 3
From 000, the fundus image of the subject is input. The fundus camera 3000 of this embodiment employs an image sensor using a CCD. The image data of the fundus camera 3000 is
An analog signal or a digital signal digitized by an A / D converter is sent to the glaucoma analysis apparatus 1000.

After inputting the image data in S2, the process proceeds to S3, in which the arithmetic processing unit 100 activates the fundus coordinate automatic creation block. The fundus coordinate automatic creation block is
As shown in FIG. 3, it has an optic disc detection function (FIG. 4), a macula detection function (FIG. 14), and a fundus coordinate setting function.
The processing is performed by executing the feature extraction unit 110 of the arithmetic processing device 100. Then, the calculation result is saved and stored in the data memory 300.

Next, in S4, the arithmetic processing unit 100 activates the pale part automatic analysis block. The pale part automatic analysis block detects the pale part by a method as shown in FIG.
DD (θ) -DP (θ), DP (θ) / DD (θ), D
D (θ), DP (θ), etc. are saved and stored in the data memory 300.

Then, in S5, the arithmetic processing unit 100 activates the optic disc edge blood vessel curvature automatic analysis block.
The optic disc edge blood vessel curvature automatic analysis block calculates the optic disc edge blood vessel curvature by a method as shown in FIG. 21, and stores and stores the optic disc edge curvature VC (θ) and the like in the data memory 300. It has become like.

Further, in S6, the arithmetic processing unit 100 activates an automatic analysis block for optic nerve fiber bundle defect. The optic nerve fiber bundle defect automatic analysis block detects the optic nerve fiber bundle defect by a method as shown in FIG. The optic nerve fiber bundle automatic analysis block reads the optic disc center coordinates and the macula center coordinates from the fundus coordinate automatic creation block,
The horizontal axis is the optic disc-macular line, the vertical axis is the line at the center of the optic disc, which is orthogonal to the horizontal axis, and the coordinate system is set with the optic disc-macular distance as the reference length.
The optic nerve arrangement shown by Ogan is simulated on coordinates. The setting of this coordinate system is executed by the fundus coordinate system setting unit 120 for setting the fundus coordinate system based on the predetermined distribution pattern of the optic nerve fiber bundle, the extracted macula position, and the extracted papilla position. The optic nerve fiber bundle defect suggestion degree NFBD (θ) and the like calculated by the optic nerve fiber bundle defect automatic analysis block are stored and stored in the data memory 300.

Then, in S7, the visual field data of the subject is input from the perimeter 2000. Perimeter 2000 of this embodiment
Uses an automatic perimeter, and visual field data is connected to the glaucoma analysis apparatus 1000 according to an appropriate protocol.

Next, in S8, the processor 100 activates the visual field automatic analysis block. The visual field automatic analysis block associates each optic nerve in the optic disc intra-papillary angle θ direction with visual field data by a method as shown in FIG. 29, and calculates the continuous defect degree VD (θ) of the optic nerve fiber bundle. The calculated continuous defect degree VD (θ) and the like are stored in the data memory 3
00 and stored. The visual field automatic analysis block corresponds to the visual field data conversion unit 130 for converting a given visual field measurement result based on the fundus coordinate system.

The pallor part automatic analysis block and the optic disc edge blood vessel curvature automatic analysis block, which correspond to the affected part extraction part, extract the characteristic part due to the effect of the fundus disease from the fundus image data. It corresponds to 140.

Further, in S9, the arithmetic processing unit 100 activates the automatic integrated analysis block. The automatic integrated analysis block can display the suggestion data of various glaucoma as shown in FIG. 32 by the method shown in FIG. It is also possible to automatically calculate and display the degree of overlap in the direction of the optic nerve glaucoma suggestion in the peak direction, the extent of the overlapped peaks, the degree of peak overlap, etc., so read these indications. Thereby, the ophthalmologist can know the state of glaucoma, the degree of progress, etc., and can assist the diagnosis. The automatic integrated analysis block corresponds to the analysis unit 150 for associating the visual field data and the characteristic portion based on the fundus coordinate system and analyzing the fundus disease.

Then, in step S10, the glaucoma analysis by the glaucoma analysis apparatus 1000 ends. The present invention is not limited to analysis of glaucoma, but can be applied to other diseases, complications with glaucoma, and the like.

[0193]

[Effects] The present invention configured as described above, the feature extraction unit for determining the macular position and the nipple position from the fundus image data of the subject, the extracted macular position, the nipple position and the predetermined position. Based on the distribution pattern of the optic nerve fiber bundle, since it is composed of a fundus coordinate system setting unit that sets the fundus coordinate system for the fundus image data, based on the distribution pattern of the macula position, the papilla position and a predetermined optic nerve fiber bundle. The fundus coordinate system can be automatically set, and by analyzing the fundus image data and the like according to this fundus coordinate system, a more comprehensive judgment can be made, which is an outstanding effect.

Further, according to the present invention, the feature extraction unit for determining the macular position and the nipple position from the fundus image data of the subject, and the optic nerve of the subject based on the extracted macular position and the nipple position. Since it is composed of the optic nerve fiber bundle distribution pattern forming unit that forms the distribution pattern of the fiber bundle, it is possible to automatically set the distribution pattern of the optic nerve fiber bundle of the subject based on the macular position and the papilla position. By analyzing fundus image data and the like according to the distribution pattern of the optic nerve fiber bundle, there is an outstanding effect that a more comprehensive judgment becomes possible.

Furthermore, the present invention has the effect that the color image data can be normalized and can be compared with the data of other measuring devices, and that changes over time can be observed with high accuracy.

Further, the present invention has an excellent effect that the reliability of the appearance position data of the optic nerve fiber bundle defect can be improved by comparing with the data of the perimeter.

[0197]

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of an exemplary embodiment of the present invention.

FIG. 2 is a diagram illustrating a configuration of a calculation processing unit according to the present exemplary embodiment.

FIG. 3 is a diagram illustrating a fundus coordinate automatic creation block according to the present embodiment.

FIG. 4 is a diagram illustrating detection of the optic disc of the present embodiment.

FIG. 5 is a diagram illustrating detection of an optic disc by using colors in the present embodiment.

FIG. 6 is a diagram illustrating the inside of the optic disc determined in this example.

FIG. 7 is a diagram illustrating a profile of this embodiment.

FIG. 8 is a diagram illustrating the amount of depression due to the teat edge of the present embodiment.

FIG. 9 is a diagram illustrating a histogram of this embodiment.

FIG. 10 is a diagram illustrating a optic disc detection process using blood vessels according to the present embodiment.

FIG. 11 is a diagram illustrating a vertical blood vessel.

FIG. 12 is a diagram illustrating a vertical blood vessel extraction process according to the present embodiment.

FIG. 13 is a diagram illustrating line tracking according to the present embodiment.

FIG. 14 is a diagram illustrating detection of the macula of the present embodiment.

FIG. 15 is a diagram illustrating a profile in macula detection according to the present embodiment.

FIG. 16 is a diagram illustrating a threshold straight line according to the present embodiment.

FIG. 17 is a diagram for explaining the central portion of the macula detected in this example.

FIG. 18 is a diagram illustrating fundus coordinates of the present embodiment.

FIG. 19 is a diagram illustrating a pale-white part automatic analysis block according to the present embodiment.

FIG. 20 is a diagram illustrating a result output of a pale-white part automatic analysis according to the present embodiment.

FIG. 21 is a diagram illustrating an automatic optic papilla edge blood vessel flexion degree analysis block according to the present embodiment.

FIG. 22 is a diagram for explaining the degree of optic papilla vascular tortuosity in this example.

FIG. 23 (a) is a diagram for explaining an automatic analysis block for optic nerve fiber loss according to the present embodiment.

FIG. 23 (b) is a diagram for explaining an automatic analysis block for optic nerve fiber loss in the present embodiment.

FIG. 24 is a diagram illustrating a simulation of optic nerve placement according to the present embodiment.

FIG. 25 is an original image converted so that the vertical axis represents the intra-papillary angle and the horizontal axis represents the distance along the optic nerve fiber bundle.

FIG. 26 is a diagram illustrating a simulation of optic nerve placement according to the present embodiment.

FIG. 27: Intrapapillary angle θ and optic nerve fiber bundle defect suggestion degree NF
It is a figure explaining the relationship with BD ((theta)).

[Fig. 28] Fig. 28 is a diagram illustrating the analysis result of an optic nerve fiber bundle defect.

FIG. 29 is a diagram illustrating an automatic field-of-view analysis block according to the present embodiment.

FIG. 30 is a diagram illustrating a state in which optic nerves in the intra-papillary angle θ direction of each optic nerve fiber bundle are associated with visual field data.

FIG. 31 is a diagram illustrating an automatic integrated analysis block according to the present embodiment.

FIG. 32 is a diagram illustrating output of results of automatic integrated analysis according to the present embodiment.

FIG. 33 is a diagram for explaining the operation of this embodiment.

FIG. 34 (a) is a diagram showing visual field data.

FIG. 34 (b) is a diagram showing a fundus image.

FIG. 35 is a diagram for explaining what is obtained by combining the visual field data and the fundus image. 100 arithmetic processing device 110 feature extraction unit 120 fundus coordinate system setting unit 130 visual field data conversion unit 140 affected portion extraction unit 150 analysis unit 200 program memory 300 data memory 400 display unit 1000 glaucoma analyzer 2000 perimeter 3000 fundus camera

Claims (7)

[Claims] 1. A feature extraction unit for determining a macular position and a nipple position from fundus image data of a subject, and an extracted macular position, a nipple position, and a predetermined optic nerve fiber bundle distribution pattern, An apparatus for analyzing a fundus disease, comprising: a fundus coordinate system setting unit that sets a fundus coordinate system for the fundus image data. 2. A feature extraction unit for determining a macular position and a nipple position from fundus image data of the subject, and a optic nerve fiber bundle of the subject based on the extracted macular position and nipple position. An apparatus for analyzing a fundus disease, comprising an optic nerve fiber bundle distribution pattern forming unit that forms a distribution pattern. 3. The fundus disease analyzing apparatus according to claim 1, wherein the feature extraction unit is configured to extract a blood vessel and to extract a focus position of the blood vessel as a papilla. 4. The fundus disease analyzing apparatus according to claim 1, wherein the feature extracting unit is configured to detect the nipple position based on the lightness and the saturation of the fundus image. 5. The fundus disease analyzing apparatus according to claim 1, wherein the feature extracting unit is configured to determine the optic disc edge based on the amount of fall of the red component of the fundus image. 6. The feature extraction unit is configured to obtain a change in tone with the papillae as a reference, and the change in tone is linearly depressed, and extracts a range of circular spread on the retina as a macula. The apparatus for analyzing a fundus disease according to claim 1 or 2. 7. The fundus coordinate system setting unit sets the fundus coordinate system of curvilinear coordinates by scaling a predetermined optic nerve fiber bundle distribution pattern according to the macula position and the papilla position, with the center of the papilla as a coordinate origin. Claims 1 to 6 configured as follows.
The fundus disease analyzer described.