JP6361065B2 - Cataract inspection device and cataract determination program - Google Patents

Description

  The present invention relates to a cataract inspection apparatus and a cataract determination program for easily performing a cataract diagnosis.

Conventionally, as a device for performing a cataract diagnosis, those disclosed in the following Patent Document 1 and Patent Document 2 are known, and both are spectroscopic optical systems for accurately diagnosing the degree of progression of cataract. And special equipment such as light-sensitive detectors. For this reason, particularly when used in developing countries, there are problems such as equipment introduction costs and a lack of skilled doctors handling the equipment.
Further, there is known a cataract inspection apparatus that does not require a special device such as a spectroscopic optical system or a light detection detector, and that allows simple inspection of cataracts by anyone with a configuration using simple equipment (Patent Document 3). ).

JP 2002-224041 A Special table 2004-535880 gazette Japanese Patent No. 5305409

In the cataract inspection device disclosed in Patent Document 3 described above, an image of a subject's eye is captured using a point-shaped light source, and the presence or absence of cataract disease is determined by image analysis. For this reason, it is difficult to specify the pupil region, and it is difficult to determine the severity of the cataract by analyzing the pupil region image.
In view of such a situation, the present invention provides a cataract inspection apparatus and a cataract determination program capable of easily specifying a pupil region and determining a cataract level using the specified pupil region with a configuration using simple equipment. Objective.

The inventors of the present invention have completed the present invention as a result of intensive studies on an apparatus and a method for performing a simple test (screening) of cataract.
The cataract inspection device of the present invention includes the following means 1) to 5).
1) a light projecting means for projecting from the front substantially onto the pupil region of the eyeball as ring-shaped projection light;
2) An imaging unit that adjusts the position of the ring-shaped reflected light of the eyeball of the ring-shaped projection light from the center of the pupil to the inner side of the iris region in the alignment at the time of imaging , and uses it as image information .
3) Reflected light detection means for detecting ring-shaped reflected light ,
4) Color feature analysis means for specifying and analyzing the inner side of the inner contour of the reflected light having the maximum diameter and the maximum brightness among the ring-shaped reflected light,
5) Cataract determining means for determining nuclear cataract or posterior subcapsular cataract from the color characteristics of the specified pupil region .

  The cataract inspection device of the present invention is characterized in that light from the light source is ring-shaped projection light. By using the light from the light source as ring-shaped projection light, the pupil region can be easily specified in the image obtained by photographing the eyeball, and a simple examination (screening) of cataract by analyzing the pupil region can be performed. By determining the symptoms of cataract using the ring-shaped reflected light, the accuracy of determining the cataract can be improved as compared with the case where no ring-shaped reflected light is used.

  Here, the reason why the pupil region can be easily specified by the ring-shaped projection light will be described. Generally, it is possible to specify a pupil region by determining an inner contour of an iris in a photographed image of an eyeball. However, in the case of cataract eyes, if the inner contour of the iris is cloudy, the inner contour of the iris becomes unclear. In particular, in the case of a cortical cataract patient, it is a symptom of cloudiness from around the lens into the pupil, and the inner contour of the iris becomes unclear. For this reason, ring-shaped projection light is projected from the front of the eyeball to obtain concentric reflection light at the center of the pupil. Since the corneal surface is substantially hemispherical, the ring-shaped projection light projected from the front is strongly reflected from the corneal surface to become ring-shaped reflected light, and a circular image is obtained.

Since the ring-shaped reflected light is a region where the color is saturated in white on the image processing, that is, a region where the RGB value is as close as possible to 255, it is easy to extract the inner contour of the ring-shaped reflected light.
In addition, the ring-shaped reflected light is originally observed concentrically from the center of the pupil, and when the reflected light deviates from the center of the pupil, it is caused by a problem of alignment at the time of photographing. In normal eyes, the ring-shaped reflected light is observed on each of the four boundary surfaces (frontal cornea, rear surface of cornea, front surface of lens, and rear surface of lens), so that there are four ring-shaped reflected light in the photographed image. However, since the reflectance of the lens front surface is very low, it is difficult to observe the ring-shaped reflected light from the lens front surface. Accordingly, the ring-shaped reflected light in the normal eye is mainly observed with the strongest reflected light from the front surface of the cornea, and thereafter, three ring-shaped reflected lights of the reflected light from the rear surface of the cornea and the rear surface of the crystalline lens can be observed. On the other hand, in the case of a cataract eye, particularly a central nuclear cataract, only two ring-like reflected lights of the reflected light from the anterior cornea and the posterior cornea are observed. By measuring the number of ring-shaped reflected lights in the photographed image in this way, normal eyes and cataract eyes can be easily separated.
In order to improve the above-described discrimination accuracy, the pupil is photographed with the center from the front. By photographing the pupil with the center from the front, the ring-shaped reflected light is arranged concentrically, so that it is easy to identify the ring-shaped reflected light and the number of reflected lights.

  It is preferable that the light projecting unit is a camera flash interlocked with the image capturing unit. A general-purpose digital camera can be used as an imaging means, and cataract inspection (screening) can be performed with a built-in or external flash.

In the cataract inspection device of the present invention , the cataract determination means determines nuclear cataract or posterior subcapsular cataract from the color characteristics of the pupil region.
Nuclear cataract is a state in which the middle (center) of the crystalline lens called the nucleus is cloudy, and can be determined from the clear difference in color characteristics between the peripheral part and the central part of the pupil region. On the other hand, of the cataract, the most common is a cortical cataract is a condition turbid plaques toward the circumference of the lens in the pupil. In addition, posterior subcapsular cataract is a condition in which cloudy as frosted glass from the middle at close on the back side of the lens.

  Further, the cataract inspection device of the present invention further comprises opacity analysis means for discriminating eyelashes from the pupil region and analyzing the degree of turbidity from the area and color where the pattern is present except for the eyelashes. Cortical cataract is determined from the degree of opacity of the pupil region. Cortical cataract is a symptom of cloudiness from around the lens into the pupil, and the area of turbidity is widespread, and the turbid area tends to be distributed around the lens. Determine.

In addition, since the eyelashes have a characteristic shape, the pupils near the inner contour of the iris are removed by removing the eyelash image and distinguishing the eyelash image from the opacity of the cortical cataract. It eliminates the opacity caused by eyelashes from the opacity of the area and improves the accuracy of cortical cataract determination.
Here, the pattern refers to color unevenness, and color unevenness occurs depending on the degree of turbidity.

The cataract inspection device of the present invention further includes the following means a) to d), and the cataract determination means uses a classifier by machine learning to obtain a feature vector composed of a color feature parameter and a opacity parameter. It is preferable to determine the presence and extent of nuclear cataract, cortical cataract, and other cataracts .
a) means for converting an RGB image in the pupil region into an HSV image ;
b) means for converting an RGB image into a binarized image ;
c) Color feature amount calculating means for calculating H, S, V and maximum, average, and median color feature parameters of the HSV image ;
d) Calculate the number of pixels of the part that is suspected of turbidity for each area divided into N equal parts (N is 2 or more) according to the distance from the center in the binarized image, and calculate the turbidity degree parameter Means for calculating turbidity .

Further, in the cataract determining means of the cataract inspection device of the present invention, it is possible to determine a cataract by the number of ring-shaped reflected light reflected at the boundary surface of each translucent body in the eye.
As described above, the ring-shaped reflected light is observed concentrically from the center of the pupil, and in the case of a normal eye, the ring-shaped reflected light is mainly composed of three rings of light reflected by the front surface of the cornea, light reflected by the back surface of the cornea, and light reflected by the back surface of the crystalline lens. On the other hand, in the case of a cataract eye, particularly a central nuclear cataract, only two ring-like reflected light reflected by the anterior and corneal surfaces are observed. Therefore, normal eyes and cataractous eyes can be distinguished by measuring the number of ring-shaped reflected lights. The pupil region is identified using the ring-shaped reflected light reflected from the front surface of the cornea from the reflected light reflected from the boundary surface of each translucent body in the eye, and the presence or absence of cataract is determined by the number of reflected lights in the pupil region. And the presence and extent of nuclear cataract, cortical cataract, and other cataracts can also be determined using the above-mentioned means a) to c).
Table 1 below shows the reflected image luminance ratios of the respective reflecting surfaces of the anterior corneal surface (corneal epidermis), the posterior corneal surface (corneal endothelium), the front surface of the crystalline lens, and the rear surface of the crystalline lens. It can be seen that the brightness of the reflected image on the front surface of the crystalline lens is one digit or more lower than the other reflecting surfaces, the brightness is low, and the reflection is low.

The cataract inspection device of the present invention is a housing including an imaging unit and a light projecting unit at one end, and the housing projects from the light source when the other end is opened and covered around the eyeball of the subject. It is configured to block incidents other than light, and is provided with a position adjustment index that allows the subject to see the light source direction, and the end of the housing has a mechanism that deforms in the vertical direction of the eyelid so as to open the subject's eyelid It is preferable.
If an iris is reflected in the ring-shaped reflected light, it is determined to be cloudy. To avoid this, it is necessary to adjust the position of the ring-shaped reflected light so that it is inside the iris area. is there. Therefore, in order to make the pupil wide open, a dark place is created by the housing, and imaging is performed with the pupil open.

By using the housing, it is possible to block the incident of external light at the housing of the housing and make a dark room state without making the surroundings dark at the time of inspection. Here, the shape of the housing may be, for example, a binocular shape, a cylindrical shape, or a box shape. More preferably, the imaging means and the light projecting means are operated after a predetermined number of seconds have elapsed with the housing in a dark box state. This is to make the pupil open greatly. The predetermined number of seconds is, for example, the passage of time of 3 to 5 seconds or 5 seconds or more.
Moreover, it is preferable to project the ring-shaped projection light from the front of the eyeball to obtain concentric ring-shaped reflected light at the center of the pupil. Therefore, a position adjustment index is provided so that the subject looks at the light source direction. Here, the position adjustment index is, for example, an LED blinking marker that cannot be seen unless it is in front of the eyeball.

  Furthermore, the end of the housing is provided with a mechanism that deforms the eyelid in the vertical direction so as to open the subject's eyelid. If the opening of the eyelid is small, the pupil region is hidden by the eyelid, and more easily affected by eyelashes. Therefore, in order to forcefully open the subject's eyelid, the eyelid is opened by a mechanism for lifting / lowering in contact with the upper and lower skin of the eyelid.

Next, the cataract determination program of the present invention will be described.
The cataract determination program of the present invention is a program for determining a cataract of an eyeball by causing a computer to function as the following means (1) to (4).
(1) The ring-shaped reflected light from the cornea and the lens of the eyeball when projected onto the pupillary area of the eyeball from the front as ring-shaped projection light is concentrically circled from the pupil center to the inside of the iris area at the time of imaging. Image information acquisition means for adjusting the position and acquiring image information;
(2) Reflected light detection means for detecting ring-shaped reflected light,
(3) Color feature analysis means for analyzing the color feature by identifying the inner side of the inner contour of the reflected light having the maximum diameter and the maximum luminance among the ring-shaped reflected light,
( 4 ) Cataract determining means for determining nuclear cataract or posterior subcapsular cataract from the color characteristics of the specified pupil region .

In cataract determination program of the present invention, cataracts determining means preferably determine the nuclear cataracts or posterior subcapsular cataract from the color characteristics of the pupil region.

  In the cataract determination program of the present invention, the computer further includes eyelash determination means for determining eyelashes from the pupil region, and opacity degree analysis means for analyzing the degree of turbidity from the area area and color where the pattern exists except for the eyelashes. It is preferred that the cataract determination means determine cortical cataract from the degree of opacity of the pupil region.

In the cataract determination means in the cataract determination program of the present invention, it is preferable that the computer functions as the following means (a) to (e) .
(A) means for converting an RGB image in the pupil region into an HSV image ;
(B) means for converting an RGB image into a binarized image ;
(C) Color feature amount calculating means for calculating H, S, V, and maximum, average, and median color feature parameters in the HSV image ;
(D) Calculate the number of pixels of a portion suspected of being cloudy for each region divided into N equal parts (N is 2 or more) according to the distance from the center of the binarized image, and set the parameter of the cloudiness degree. Turbidity degree calculation means to calculate ,
(E) Means for determining the presence and extent of nuclear cataracts, cortical cataracts, and other cataracts using a feature vector composed of color feature parameters and opacity parameters using a machine learning classifier .

In the cataract determination means in the cataract determination program of the present invention, it is preferable that the computer functions as a means for determining a cataract based on the number of ring-shaped reflected lights reflected from the boundary surface of each translucent body in the eye. The number of ring-shaped reflected lights in the image is measured by computer image processing to discriminate between normal eyes and cataract eyes.
That is, in the cataract determination means in the cataract determination program of the present invention, the pupil region is specified using the ring-shaped reflected light reflected from the front surface of the cornea from the reflected light reflected from the boundary surface of each translucent body in the eye. Then, the presence / absence of cataract is determined based on the number of reflected light, and the presence / absence and degree of nuclear cataract, cortical cataract, and other cataracts are determined using the means (a) to (d) described above.

According to the cataract inspection apparatus and the cataract determination program of the present invention, it is possible to easily specify the pupil region by using the ring-shaped light, and to determine the level of the cataract using the color characteristics and the degree of opacity of the pupil region. is there.
Even in places where there is a shortage of ophthalmologists in developing countries and where there is a shortage of diagnostic equipment, it is possible to automatically estimate the type of cataract for patients with cataracts using a simple device such as a combination of a digital camera and a computer. Will be able to.

Conceptual diagram of the cataract inspection device of Example 1 Cross-sectional view of the configuration of the cataract inspection device of Example 1 Explanatory drawing of observation image of ring-shaped reflected light in front of cornea and lens Process flow diagram of the cataract inspection device of Example 1 Color feature analysis processing flow chart Illustration of pupil region image and polar coordinate development image Explanatory diagram of RGB image and HSV color space of pupil region image Explanatory drawing of the three feature values of the polar coordinate development image Illustration of polar coordinate development image An example of polar coordinate image Explanatory diagram about discrimination between eyelashes and opacity Illustration of turbidity analysis Illustration of feature vector Conceptual diagram of determination using classifier Explanatory drawing of evaluation method by 10-fold cross validation Visualization of decision flow by decision tree (1) Visualization of decision flow by decision tree (1)

  Hereinafter, an example of an embodiment of the present invention will be described in detail with reference to the drawings. The scope of the present invention is not limited to the following examples and illustrated examples, and many changes and modifications can be made.

  FIG. 1 shows a conceptual diagram of the cataract inspection apparatus of the first embodiment. FIG. 2 is a sectional view showing the structure of the cataract inspection device. As shown in FIGS. 1 and 2, the cataract inspection device of Example 1 includes a ring-shaped flash 22 that is a light projecting unit that projects light from a light source onto the eyeball 1 as ring-shaped projection light 20, and a ring-shaped flash 22. It is composed of a camera 21 that is an image pickup means for picking up ring-shaped reflected light from the crystalline lens 2 and cornea 6 of the light eyeball 1 and using it as image information, and a computer 23. The computer 23 includes an image information acquisition unit 23a for acquiring image information, a ring-shaped reflected light detection unit 23b for detecting the ring-shaped reflected light 10, and an inner contour of the reflected light having the maximum diameter and the maximum luminance among the ring-shaped reflected light. Color feature analyzing means 23c for analyzing the color characteristics of the inner pupil region, turbidity analyzing means 23d for analyzing the degree of opacity of the pupil region, and ring-like reflected light number analyzing means 23e for analyzing the number of ring-like reflected lights 10 And the cataract determining means 23f for determining the cataract using the ring-shaped reflected light 10. The camera 21, the ring-shaped flash 22, and the eyeball 1 of the subject are covered with the housing 7. The housing 7 blocks outside light such as a dark box or a black curtain.

  As shown in FIGS. 1 and 2, the ring-shaped flash 22 includes a ring-shaped slit 22 a having a ring-shaped opening, a shielding plate 22 b, a light emitting unit 22 d, and a mount unit 22 c for mounting the light emitting unit. There are a light shielding plate 22b and a mount portion 22c so as to cover the light emitting portion 22d of the ring-shaped flash 22, and a gap between the light shielding plate 22b and the flash body is a ring-shaped slit 22a. The optical center of the camera lens 21a and the geometric center of the ring light flash 22 are arranged on the same line. That is, the optical center of the camera lens and the optical center of the ring light projection light are set on the same line. Thereby, the ring-shaped projection light 20 can be projected from substantially the front of the eyeball, and the frontal reflection of the cornea and the crystalline lens can be obtained. The power of the light emitting unit 22d is supplied from the battery unit 25 via the power cable 26.

The camera 21 can be a general-purpose digital camera. The ring flash 22 may be a flash external to the camera 21 or may be guided to the ring flash 22 by an optical fiber cable. Then, ring-shaped projection light is projected from the light source of the ring-shaped flash 22 in conjunction with the shutter of the camera 21.
The computer 23 is connected to the camera 21 via a USB (Universal Serial Bus) cable (data communication interface 24) so that image information data can be captured from the camera 21.

When the ring-shaped reflected light is observed with the camera 21, for example, an observation image as shown in FIG. 3 (1) or FIG. 3 (2) is obtained.
The ring-shaped reflected light 10 that is the reflected light having the maximum diameter and the maximum luminance is reflected light from the front surface 6 a of the cornea 6 of the eyeball 1. In the normal eye, as shown in FIG. 2, the ring-shaped reflected light is observed at each of four boundary surfaces (frontal cornea 6a, rear cornea 6b, front lens surface 2a, rear lens surface 2b), and the captured image has four ring shapes. There is reflected light. However, since the lens front surface 2a has a very low reflectance, it is generally difficult to observe the ring-shaped reflected light from the lens front surface 2a. As shown in FIG. 2 and FIG. Three ring-shaped reflected lights (the corneal front surface 6a, the corneal rear surface 6b, the lens rear surface 2b) (the ring-shaped reflected light 10 of the cornea front surface 6a, the ring-shaped reflected light 11 of the corneal rear surface 6b, and the ring-shaped reflected light 12 of the lens rear surface 2b). ) Can be observed. On the other hand, in the case of a cataract eye, particularly a central nuclear cataract, two ring-like reflected lights (ring-like reflected light 10 on the anterior corneal face 6a, posterior cornea 6b) on two boundary surfaces (the corneal front face 6a and the posterior corneal face 6b). Only the ring-shaped reflected light 11) is observed.
In this embodiment, as shown in FIG. 3 (1), only the ring-shaped reflected light 10 that is the reflected light of the corneal front surface 6a having the maximum diameter and the maximum luminance is captured, and the color characteristics of the pupil region inside the inner contour are Cataract is determined by the degree of opacity.

FIG. 4 shows a process flow diagram of the cataract inspection apparatus of the first embodiment.
First, the ring-shaped flash is turned on, the ring-shaped projection light is projected onto the eyeball, and the eyeball is photographed (step S01). Next, the pupil region is detected from the captured image of the ring-shaped reflected light on the front surface of the cornea (step S02). Then, using the ring-shaped reflected light on the front surface of the cornea, a region inside the inner contour is cut out as a pupil region image (step S03).
Specifically, the pupil region is detected by converting a captured image into a black and white image, binarizing, thinning, and detecting a circular shape of the ring-shaped reflected light by Hough conversion.

Then, the number of ring-shaped reflected light is analyzed (step S04), the color feature is analyzed (step S05), or the turbidity is analyzed (step S06). By analyzing only the number of ring reflected lights (step S04) and detecting only the number of ring reflected lights, the presence or absence of cataract disease can be determined easily. However, in order to accurately perform automatic determination by image processing, color feature analysis (step S05) and turbidity analysis (step S06) are also used.
The result of the color feature analysis (step S05) or the turbidity degree analysis (step S06) is compared with the determination criteria previously extracted from the learning data to determine the occurrence of cataract (cataract eye) (step S07). ). As a result of the determination, a cataract onset (cataract eye) or a healthy eye is determined (step S08, 09).

  In the present embodiment, a system is assumed in which only the presence or absence of the occurrence of cataract is identified, and in the case of the onset, a medical examination is encouraged to an ophthalmologist. That is, the presence / absence of cataract is estimated from the photographed image of the eye of the subject, and the medical institution (with the subject's identifier (ID) is integrated into the medical institution as information obtained by integrating the image information of the eyeball and the estimation result information. This system is assumed to be transmitted to (under the doctor) and encourage the patient to see the cataract.

Next, a color feature analysis processing flow will be described with reference to FIG.
An RGB image is acquired from the cut out pupil region image (step S21), converted into an HSV image, and an HSV image is acquired (step S22). Then, the average value, maximum value, and median value of each HSV image are measured (step S23). The measured average value, maximum value, and median are used to classify and analyze the color features (step S24).

Next, with reference to FIGS. 6 to 9, analysis of the cut out pupil region image will be described.
FIG. 6A shows a state in which the pupil region is detected based on the shape and position of the ring-shaped reflected light in the captured image of the eyeball.
FIG. 6 (2) shows a cut out pupil region image. Color features are analyzed from this pupil region image. FIG. 6 (3) shows a polar coordinate developed image obtained by developing a clipped pupil region image (circular shape) in polar coordinates. The ring-like flash pattern of the ring-like reflected light is analyzed from the polar coordinate developed image.
In the pupil region image, the ring-shaped reflected light has a circular shape for a healthy person. On the other hand, in the case of a patient with cataract, the ring-shaped reflected light has a remarkably reduced brightness or a partially missing shape.

First, a color feature detection method will be described.
FIG. 7 (1) shows a cut out pupil region image (RGB image). As described above, an RGB image is converted into an HSV image to obtain an HSV image, and color features are analyzed in the HSV color space.
The HSV color space is a color space including three components of hue (H), saturation (S), and lightness (V), and can be visualized using a cone. In the visual representation using a cone, the hue (H) is drawn in a three-dimensional conical structure of the color wheel, the saturation (S) is the distance from the center of the cone, and the lightness (V) is from the vertex of the cone. Expressed in distance. R, G, and B in the pupil region image (RGB image) are in the range of 0.0 to 1.0, respectively. If a color defined by (R, G, B) is given, the corresponding color of (H, S, V) can be derived by a mathematical formula. On the other hand, the hue (H) changes from 0.0 to 360.0, and the hue (H) is expressed by an angle along the color circle indicated. Saturation (S) and brightness (V) vary from 0.0 to 1.0.

In the RGB image (RGB color space), since brightness and color are expressed in a mixed manner, for example, when the brightness changes due to a subtle change in illumination, all RGB are affected. On the other hand, in the HSB color space, since it is decomposed into hue (H), saturation (S), and brightness (V), it is possible to correctly detect a change in pupil color.
By detecting color features and determining whether or not the entire lens is white, nuclear cataracts and posterior subcapsular cataracts are determined. Also, cortical cataract is discriminated by detecting a pattern from color features and quantifying the degree of turbidity inside the lens. When detecting a pattern, as will be described later, eyelash removal is performed to remove noise.

As shown in FIG. 8, the polar coordinate developed image is analyzed by paying attention to three feature amounts. In FIG. 8, the feature v of the horizontal line, the feature h of the vertical line, and the feature vh are analyzed as three feature quantities.
Here, in order to derive the feature quantities v and h, the horizontal and vertical image filters shown in FIG. 8 are used. In a polar coordinate developed image, an image filter is applied around a certain point to calculate the average brightness value m ₁ of pixels that are white pixels and the average brightness value m ₂ of pixels that are black pixels. Finally, m ₁ −m ₂ is calculated as feature quantities v and h.

In the polar coordinate development image on the left in FIG. 8, in the portion surrounded by the square frame, a large number of horizontal stripe patterns can be detected, so the feature value v takes a large value. Since it is not detected, the feature amount h is expected not to be as large as the feature amount v.
In this way, candidate points are calculated using the size of the feature quantity v and the feature quantity vh.
Then, using the number of pixels calculated as candidate points, that is, the number of pixels determined to be ring-shaped reflected light, and the average value of the results calculated by the image filter in the horizontal direction for each pixel The degree of clarity of reflected light is digitized, and this is used as a feature amount indicating the intensity of ring-shaped reflected light.

Next, a method for detecting ring-shaped reflected light will be described.
The ring-shaped reflected light appears concentrically with the center of the pupil. Therefore, the pupil region image is developed into a polar coordinate developed image, and the ring-shaped reflected light is detected as a horizontal stripe pattern. That is, as shown in FIG. 9, it is verified whether or not each point constitutes a circle from the polar coordinate development image, and the center of the circle corresponding to each candidate point is “voted”. To detect.
Specifically, the following (Process 1) to (Process 4) are performed.

(Process 1) First, a voting space M (x, y, r) for voting the center position and radius of a circle is prepared and initialized to M (x, y, r) = 0. The value of M becomes a score, and the larger the score is, the more ring-like reflected light is present there.

(Process 2) The following (Process 2-1) and (Process 2-2) are executed for candidate points (curve in FIG. 9) detected in advance.
(Processing 2-1) The polar coordinates P (r, θ) are converted into the plane coordinates I (x, y).
(Process 2-2) (Process 2-2-1) and (Process 2-2-2) are executed by changing the radius r within a predetermined range [r1, r2].
(Process 2-2-1) A point at a distance r from I is calculated.
(Processing 2-2-2) At that point, 1 is added to the score M (x, y, r).

(Process 3) After the above process is repeated, (x, y, r) = (xm, ym, rm) where M (x, y, r) is maximized is calculated.
(Process 4) The candidate points voted for (xm, ym, rm) are assumed to be “consistent”.

FIGS. 10A to 10F show examples of various polar coordinate development images. In FIGS. 10A, 10C, 10E, and 10F, many thin lines are seen in the vertical direction because the reflected image of “eyelashes” is reflected.
Since the reflected image of the “eyelash” becomes noise when detecting the opacity of the pupil region, it is excluded from the image in the analysis.

With reference to FIG. 11, the identification of eyelashes and turbidity will be described. FIG. 11A shows an example of a polar coordinate development image of a pupil region image in which a reflection image of “eyelashes” is reflected. A region where a pattern is present is detected using fast Fourier transform (FFT) for each block of the pupil region image. Here, the image is divided into 64 × 64 image blocks, and the fast Fourier transform (FFT) is performed by dividing the image block into a horizontal component and a vertical component. In this case, the DC component is equivalent to the average value (average brightness), so it should be ignored, and the horizontal high-frequency component when expanded in the polar coordinate system is strongly affected by the reflection of the eyelashes, so it should be ignored. To. Then, if the average value of the remaining components is larger than the threshold value, it is determined as a candidate area where turbidity exists.
FIG. 11 (2) shows a block portion that is a candidate for turbidity in the above processing. Then, binarization processing is performed on each block portion that is a candidate for turbidity.

In order to avoid false detection of eyelashes, the following processing is further performed to remove eyelashes.
1) A region where eyelashes may exist in an image is set in advance, and a binarization result is extracted.
2) Since the eyelashes are not scattered but continuous, a connected region is derived.
3) “Eccentricity (a) of ellipse having the same moment” and “angle (b) of the ellipse” are obtained in the connected region.
4) If a> a ₀ and b> b ₀ , remove as eyelashes. Here, a ₀ and b ₀ are eyelash shape determination thresholds.
In this way, the area where the pattern exists is detected using Fast Fourier Transform (FFT) for each block of the pupil area image, and “eyelash” and “turbidity” are identified from the shape characteristics of the binarization result. .

Regarding “turbidity”, since the presence or absence of cataract and the degree of cataract differ depending on which part of the pupil region is clouded, the degree of turbidity is examined according to the distance from the center of the pupil region. For example, the pattern like the pupil region shown in FIG. 12A is divided into _five donuts (P _{1 to} P ₅ ) according to the distance from the center of the pupil region as shown in FIG. To evaluate the “degree of turbidity”.

Then, the color feature parameters are the average value, maximum value, and median value of H, S, and V, that is, 3 × 3 nine parameters (f _{1 to} f ₉ ). Further, the parameters of the opacity are set to five parameters (f _{10 to} f ₁₄ ) divided into five donuts (P _{1 to} P ₅ ) according to the distance from the center of the pupil region. The number of pixels of the ring-shaped reflected light (r_1), the filter response value, the parameters of the color features, and the parameters of the turbidity are set as one feature vector v = (f ₁ , f ₂ , f ₃ ,..., F ₁₆ ). (See FIG. 13).

Then, the above feature vector v = (f ₁ , f ₂ , f ₃ ,..., F _N ) and the result of image diagnosis by an ophthalmologist are combined into a database as learning data, and a support vector machine ( A determination rule is learned using a machine learning method such as SVM) to discriminate between a cataract eye and a healthy eye.

Cross-validation is a method for evaluating a classifier model in machine learning. The given data is divided and used as learning data. This test verifies and verifies the validity of learning, and verifies and confirms how well the classifier (classifier) can cope with the population with a good approximation.
Among the evaluation methods based on cross-validation, in K-division cross-validation, the sample group is divided into K pieces, one of which is taken as a test case, the remaining K-1 pieces are taken as learning cases, and the combination is changed K times. Repeat the evaluation. FIG. 15 shows evaluation by 10-fold cross validation.

  Tables 2 to 4 below show the results of determination / evaluation on healthy subjects and cataract patients using the cataract inspection apparatus of Example 1. The table shows the number of correctly determined cataract eyes (True-Positive) and the number of falsely determined cataract eyes as healthy eyes (False-Positive). Here, what was determined to be cataract includes both nuclear cataract and cortical cataract. In addition, the number of correctly determined healthy eyes (True-Negative) and the number of falsely determined healthy eyes as cataract eyes (False-Negative) are shown.

  In Tables 2 to 4 below, “Recall” is the recall, which means the proportion of what is actually positive, predicted to be positive, and quantifies how much was found The minimum value is 0, and the closer to 1 or 1, the better. “Precision” is an index representing the accuracy of what is determined to be cataract, and is calculated from TP / (TP + FP). “Precision” is also better as the minimum value is 0 and closer to 1. The “F value (F-number)” is one of evaluation scales of prediction results in machine learning, and indicates a harmonic average of accuracy and recall. Specifically, the F value (F-number) is calculated from 2 × B × A / (A + B) where the value of Recall is “A” and the value of Precision is “B”. The “F value (F-number)” is also better as the minimum value is 0 and closer to 1.

  Table 2 shows the result of the analysis of color characteristics and turbidity using a support vector machine (SVM) (when the reflection image of the ring flash is not included in the feature amount). On the other hand, Table 3 shows the results of analyzing color characteristics and turbidity using a support vector machine (SVM) by observing ring-shaped reflected light. Table 4 compares “Recall”, “Precision”, and “F value (F-number)” when the ring flash is not used and when the ring flash is used.

Here, the NO threshold value and the C threshold value shown in the left column of Tables 2 to 4 are boundaries for determining the presence or absence of the occurrence of cataract used when providing teacher data in learning. First, for all the images used for the evaluation, NO and C were determined according to LOCSIII (The Lens Opacities Classification System III) cataract classification by an eyesight trainer skilled in cataract classification. NO is an index indicating the degree of turbidity of the lens, and C is an index indicating the degree of cortical cataract. The larger the value, the higher the degree.
Since these indices are subjective and there is no uniform standard for cataract if they are larger than a certain threshold, ranges that can be judged to be generally appropriate, for example, 2 to 3 for NO and 2 to 2 for C .4, learning data was created and tested using each value as a boundary value.

  Moreover, learning of a support vector machine (SVM) will be described. Support vector machines are known as very accurate classifiers, but parameters need to be set carefully to achieve sufficient performance. Here, a widely used radial basis function (RBF) kernel is used, and various parameters are set to obtain the best result.

  From Table 4, it was confirmed that the results of “Recall”, “Precision”, and “F value (F-number)” were improved when ring flash was used for all threshold values. That is, it was shown that the accuracy of cataract determination was improved by using a ring flash and observing the ring-shaped reflected light.

Tables 5 and 6 below show the results of identification using decision trees. By using a decision tree, it is possible to know which feature works in a color pattern, turbidity or ring flash. From Tables 5 and 6, it was found that the F value (F-number) was improved in almost all cases using the ring flash.
In Tables 5 and 6, the best case (the row shown in gray) was visualized with a decision tree. As shown in FIGS. 16 and 17, when the decision tree was visualized, it was found that the identification using the color feature mainly works, and the identification using other information works to compensate for it.

  In the visualization, learning is performed using all data without using cross-validation. In FIGS. 16 and 17, the nodes (nodes) of the decision tree correspond to the round frame, and the leaves (identification result) of the decision tree correspond to the square frame. The input data is input to the node (root node) located at the top, and it is determined which child node to proceed according to the determination condition at each node. The feature of the round frame pattern indicates which feature is used for determination.

  In the above-described cataract inspection apparatus according to the first embodiment, the camera 21, the ring flash 22, and the computer 23 can be easily configured. However, the hardware of the camera 21, the ring flash 22, and the computer 23 already exists. In addition, by providing the cataract determination program, the camera, the ring flash, and the computer hardware can operate as a cataract inspection device.

  As shown in FIG. 1, the cataract determination program comprises image information acquisition means 23a, reflected light detection means 23b, color feature analysis means 23c, turbidity analysis means 23d, ring-shaped reflected light number analysis means 23e, and cataract determination means 23f. Is done. The image information acquisition unit 31 captures an image captured by the camera 21. For example, a general-purpose digital camera has a data communication interface 24 such as a USB so that it can be connected to a computer. The image information acquisition means 23a performs data communication with the camera, takes in the image information in the camera, and allows the image information to be analyzed in the computer.

  The present invention is useful as a medical diagnosis / inspection (screening) device for an eyeball.

DESCRIPTION OF SYMBOLS 1 Eyeball 2 Crystal 3 Iris 4 Retina 5 Pupil 6 Cornea 7 Housing 10-12 Ring-shaped reflected light 20 Ring-shaped projection light 21 Camera 21a Camera lens 21b Lens main body 21c CCD image sensor 22 Ring-shaped flash 22a Ring-shaped slit 22b Shielding plate 22c Mount portion 22d Light emitting portion 23 Computer 24 Data communication interface 25 Battery unit 26 Power cable

Claims (9)

A light projecting means for projecting from the substantially front to the pupil region of the eyeball as ring-shaped projection light;
An imaging means for adjusting the position of the ring-shaped reflected light of the eyeball of the ring-shaped projection light from the center of the pupil to the inner side of the iris region in the alignment at the time of imaging and imaging it as image information;
Reflected light detecting means for detecting the ring-shaped reflected light;
Color feature analysis means for identifying the inner side of the inner contour of the reflected light having the maximum diameter and the maximum luminance among the ring-shaped reflected light and analyzing the color feature,
Cataract determination means for determining nuclear cataract or posterior subcapsular cataract from the color characteristics of the identified pupil region;
A cataract inspection device comprising: Further comprising turbidity analysis means for discriminating eyelashes from the pupil region and analyzing the degree of turbidity from the area and color of the region where the pattern is present except for the eyelashes,
The cataract inspection device according to claim 1 , wherein the cataract determination means determines cortical cataract from the degree of opacity of the pupil region. Means for converting an RGB image in the pupil region into an HSV image;
Means for converting the RGB image into a binarized image;
Color feature amount calculating means for calculating H, S, V, and maximum, average, and median color feature parameters of the HSV image;
Calculate the number of pixels of the part suspected of being turbid for each region divided into N equal parts (N is 2 or more) according to the distance from the center in the binarized image, and calculate the opacity parameter. Turbidity calculation means,
Further comprising
The cataract determining means determines whether or not there is a nuclear cataract, cortical cataract, or other cataract using a classifier based on a machine learning based on a feature vector composed of the color feature parameter and the opacity parameter. The cataract inspection device according to claim 2 , wherein In the cataract determining means, is reflected at the boundary surface of each transparent body in the eye, the number of the ring-shaped reflected light, according to any one of claims 1 to 3, wherein determining that cataracts Cataract inspection device. A housing including the imaging unit and the light projecting unit at one end, wherein the housing is configured to block incident light other than the projection light from the light source when the other end is opened and covered around the eyeball of the subject. Configured,
A position adjustment index is provided to allow the subject to see the light source direction,
The cataract inspection device according to any one of claims 1 to 4 , wherein an end portion of the housing includes a mechanism that deforms the eyelid vertically so as to open the eyelid of the subject. A cataract determination program for determining a cataract of an eyeball using a computer,
The computer,
The ring-shaped reflected light from the cornea and crystalline lens of the eyeball when projected onto the pupillary area of the eyeball from the front as ring-shaped projection light is positioned inside the iris area concentrically from the pupil center at the time of imaging. Image information acquisition means for acquiring image information;
Reflected light detecting means for detecting the ring-shaped reflected light;
Color feature analysis means for identifying the inner side of the inner contour of the reflected light having the maximum diameter and the maximum luminance among the ring-shaped reflected light and analyzing the color feature,
Cataract determining means for determining nuclear cataract or posterior subcapsular cataract from the color characteristics of the identified pupil region,
Cataract determination program to function as. Said computer further
Eyelash discriminating means for discriminating eyelashes from the pupil region,
It functions as a turbidity analysis means for analyzing the turbidity from the area and color of the area where the pattern exists except for the eyelashes,
The cataract determination program according to claim 6 , wherein the cataract determination means determines cortical cataract from the degree of opacity of the pupil region. In the cataract determining means,
The computer,
Means for converting an RGB image in the pupil region into an HSV image;
Means for converting the RGB image into a binarized image;
Color feature amount calculating means for calculating H, S, V, and maximum, average, and median color feature parameters in the HSV image;
Calculate the number of pixels of the part suspected of being turbid for each region divided into N equal parts (N is 2 or more) according to the distance from the center in the binarized image, and calculate the opacity parameter. Turbidity calculation means,
A feature vector comprising the color feature parameter and the opacity parameter is made to function as a means for determining the presence and extent of nuclear cataract, cortical cataract, and other cataracts using a machine learning classifier. The cataract determination program according to claim 7 . 9. The cataract determination unit according to claim 6 , wherein the cataract determination unit determines a cataract based on the number of the ring-shaped reflected lights that are reflected at a boundary surface of each translucent body in the eye. Cataract inspection program.