JP3448748B2 - Corneal endothelial cell measuring device and measuring method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a corneal endothelial cell measuring device and measuring method for inspecting corneal endothelial cells,
In particular, the ridge line is detected from the imaged data of the corneal cells, and the "line-likeness" is calculated from the angle formed by the two ridge directions to identify the contour line of the corneal endothelial cell and to trace the operator's traces. The present invention relates to a corneal endothelial cell measuring device and a measuring method capable of calculating the area of corneal endothelial cells with high accuracy without the need.

[0002]

2. Description of the Related Art The cornea of the outer eye is composed of a corneal epithelium (stratified squamous epithelium), Bowman's membrane, parenchyma (proprietary layer), Descemet's membrane (endothelial basement membrane), and corneal endothelium (monolayer squamous epithelium). It consists of 5 layers. The corneal endothelium is a hexagonal prismatic cell, lining the underside of the cornea with a single layer. The cells of the corneal endothelium have a diameter of about 20 μm and are relatively large cells. By grasping the condition of the corneal endothelium, it is possible to analyze the pathological condition and cause of the disease. In particular, when performing cataract surgery or artificial lens insertion surgery, it is necessary to quantitatively evaluate the degree of invasiveness of the corneal endothelium by comparing the state and cell area of the corneal endothelium before and after the surgery. Furthermore, corneal endothelial cells have an extremely low mitotic ability, and their cell density decreases with age. The corneal endothelial cells control the drainage function from the anterior chamber to the parenchyma, and have the function of preventing clouding of the cornea.

Therefore, the cell density of corneal endothelial cells, the coefficient of variation showing the variation in cell area, the appearance rate of hexagonal cells, etc. are measured to diagnose the pre-functional ability of the cornea, stress on the cornea, etc. It is being appreciated.

A specular microscope is used for observing the corneal endothelium at high magnification. This specular microscope has the property that most of the light incident at a specific angle is transmitted through the boundary surface of the two layers having different refractive indices, but part of the light is reflected at the boundary surface at an equal angle. Utilizing this, the boundary surface is observed by the boundary reflected light. As a method for calculating the average cell area of corneal cells using this specular microscope, a digitizer method or a grid method has been adopted.

The digitizer method is to input the coordinates of the start point or the end point of each side of the cell wall of each cell from a photographed image of a corneal cell to a computer using a digitizer to calculate the area of each cell.

In the grid method, a square grid (grid) of a predetermined size is superposed on a photographed image of corneal cells, the number of cells contained in the grid is measured, and the average cell area is calculated from the average cell number. It is to calculate backwards.

However, the above-mentioned digitizer method has a problem in that it requires manual tracing, requires a great deal of labor and time, and inevitably causes an analysis error by an operator. Further, the grid method has a problem that, for example, cells cut into the upper side and the left side of the lattice are not counted, and furthermore, the frequency distribution such as the cell area cannot be measured.

For these reasons, the advent of an automatic measuring device for corneal endothelial cells has been strongly enthusiastic in clinical practice, and various developments have been made. Image processing technology based on contrast is exclusively used for these automatic measurement methods.

In the image processing technique generally adopted, noise removal and local density gradient correction are performed as preprocessing steps. Next, as boundary extraction processing, binarization processing and thinning processing are performed. Here, after the boundary line is corrected by the operator, the area of each corneal cell is obtained.

Further, there have been devised devices for approximating the area of cells after coordinate input, and developed ones for simplifying manual input.

There is also a neural network in which the cell wall is identified. In this method, a cell membrane is roughly extracted using a neural network, and the cell wall is identified using a minimum / maximum average filter and a ridge line extraction method as post-processing.

[0012]

However, the above-mentioned conventional method that requires manual tracing requires a great deal of labor and time, and in addition, the analysis error by the operator cannot be avoided. However, even if the device is elaborated on the approximate calculation of the cell area or the manual input is simplified, it does not satisfy the original demand for automatic corneal endothelial cell measurement.

In the case of using the normal image processing technique based on contrast, there is a portion where the cell wall is not correctly extracted due to the local density distribution of the image, and if this data is binarized, local misrecognition is avoided. There was a problem that I could not do it. Further, there is a problem that erroneous recognition due to insufficient contrast or cell nuclei in cells cannot be avoided.

The method using a neural network is effective for an image with extremely small noise, but it is not effective for analyzing an image containing noise.

[0015]

[0016]

SUMMARY OF THE INVENTION The present invention has been devised in view of the above-mentioned problems, and comprises an image input means for capturing an image signal of a corneal endothelium and an image input to the image input means. It comprises arithmetic processing means for identifying the outline of the corneal endothelial cells, and the arithmetic processing means comprises valley line detecting means for detecting concave portions from the image input to the image inputting means, and Moving-direction pixel density detecting means for calculating the density of a pixel in the radial direction centering on the pixel extracted as a valley line in the image detected by the valley-line detecting means, and this moving-direction pixel By differentiating the data obtained by the density detecting means in the angular direction with the pixel extracted as a valley line in the center, and examining the portion where the sign of the differential coefficient changes, a ridge line detecting means for detecting a ridge can be obtained. , This tail Based on the angle formed by the two ridge directions detected by the line detection means, a "line-likeness" detection means for detecting the "line-likeness" of the ridge, and a "line-likeness" detected by this "line-likeness" detection means The line image extraction means for identifying the line image by "likeness" and the outline of the corneal endothelial cell.

Further, the arithmetic processing means of the present invention can be provided with a shape arithmetic means for calculating the area and the like from the outline of the corneal endothelial cells obtained by the line drawing extracting means.

[0018]

Further, the corneal endothelial cell measuring method of the present invention comprises a first step of capturing an image signal of an image of the corneal endothelium, and a first step of detecting a valley line by detecting a concave portion from the image input in the first step. Data obtained in two steps, a third step for calculating the pixel density in the radial direction around the pixel extracted as a valley line for the image detected in the second step, and data obtained in the third step By differentiating in the angular direction centered on the pixel extracted as the valley line and examining the part where the sign of the differential coefficient changes, and the fourth step of detecting the ridge and the two ridges detected in the fourth step. Based on the angle formed by the directions, a line image is identified by the fifth step of detecting the "linelikeness" of the ridge and the "linelikeness" detected in the fifth step, and the outline of the corneal endothelial cell is identified. It can also be configured by the sixth step.

The method for measuring corneal endothelial cells of the present invention comprises:
A seventh step of calculating the area and the like from the contour line of the corneal endothelial cell obtained in the sixth step may be provided.

[0021]

[0022]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention configured as described above is
The image input means captures an image signal of the corneal endothelium, and the arithmetic processing means identifies the outline of the corneal endothelium cell from the image input to the image input means and calculates the area and the like. . In this arithmetic processing means, the valley line detecting means detects the concave portion from the image input to the image input means, and the moving-direction pixel density detecting means detects the valley line as the valley line with respect to the image detected by the valley line detecting means. The density of the pixel is calculated in the radial direction centered on the extracted pixel, and the ridge line detection means
Differentiate the data obtained by the moving direction pixel density detection means in the angular direction centered on the pixel extracted as a valley line,
The ridge is detected by examining the portion where the sign of the differential coefficient changes, and the "line-likeness" detecting means detects the "line-likeness" of the ridge based on the angle formed by the two ridge directions detected by the ridge line detecting means. Is detected and the line drawing extraction means
The "line-likeness" detected by the "line-likeness" detecting means makes it possible to identify the line image and the outline of the corneal endothelial cell.

The shape calculation means provided in the calculation processing means of the present invention can calculate the area and the like from the outline of the corneal endothelial cells obtained by the line drawing extraction means.

[0024]

Further, the corneal endothelium measuring method of the present invention captures an image signal of an image of the corneal endothelium, detects a concave portion from the captured image by detecting a valley line, and detects a valley line from the detected image. The density of the pixel is calculated in the radial direction centered on the pixel extracted as, and the calculated data is differentiated in the angular direction centered on the pixel extracted as the valley line, and the part where the sign of the differential coefficient changes is calculated. By examining the ridge, the "line-likeness" of the ridge is detected based on the angle formed by the two ridge directions, and the line image is identified by the "line-likeness" to identify the outline of the corneal endothelial cell. can do.

The method for measuring corneal endothelial cells of the present invention comprises:
The area and the like can be calculated from the outline of the corneal endothelial cell.

[0027]

【Example】

An embodiment of the present invention will be described with reference to the drawings.

FIG. 1 shows a corneal endothelial cell measuring device 1 of this embodiment.
It shows the configuration of. The corneal endothelial cell measuring device 1 is
An image input interface 11, an AD converter 12,
The field memory 13 and the arithmetic processing means 14 are included.

The image input interface 11 corresponds to an image input means for taking in an image signal of an image of the corneal endothelium, to which the image pickup means 20 is connected. Although the CCD camera is adopted as the imaging means 20 in this embodiment, any imaging means capable of imaging the corneal endothelium can be adopted. It is also possible to use a system for taking a photograph of the corneal endothelium and converting this photographic image into an image signal. When the output signal of the image pickup means 20 is an NTSC composite signal, the image input interface 11 has a built-in clamp circuit for extracting a luminance signal and preventing a change in DC level, a sync separation circuit for separating a sync signal, and the like. Has been done.

The A / D converter 12 is for digitizing the luminance signal, and is adapted to perform A / D conversion from the black level to the white level of the luminance signal.

The field memory 13 includes the A / D converter 1
It is for storing the image signal digitized by 2.

The arithmetic processing means 14 is for identifying the outline of the corneal endothelial cell from the image signal and calculating the area and the like. As shown in FIG. 2, the arithmetic processing means 14 of the present embodiment has a valley line detecting means 141, a moving direction pixel density detecting means 142, a ridge line detecting means 143, a line likeness detecting means 144, and a line drawing extracting means. 145 and shape calculation means 1
And 46.

The valley line detecting means 141 is for detecting the concave portion from the image input to the image input interface 11. The moving direction pixel density detecting means 142 is for calculating the density of the pixel in an appropriate radial direction. The ridge line detecting means 143 is for detecting a ridge by differentiating the data obtained by the moving-direction pixel density detecting means 142 and examining the portion where the sign of the differential coefficient changes.

The "line-likeness" detecting means 144 is for detecting the "line-likeness" from the angle formed by the two ridge directions detected by the ridge line detecting means 143. The line drawing extracting means 145 is for extracting the line drawing by identifying the line image based on the “line likeness” detected by the “line likeness” detecting means 144. The shape calculation means 146 is for calculating the area and the like from the outline of the corneal endothelial cells obtained by the line drawing extraction means 145.

The arithmetic processing means 14 is composed of an electronic computer or the like including a CPU, and has a valley line detecting means 141, a moving direction pixel density detecting means 142 and a ridge line detecting means 143.
, Line-likeness detection means 144, and line drawing extraction means 14
5 and the shape calculation means 146 are executed by a program stored in the calculation processing means 14.

The output means 30 is for displaying the calculation result of the calculation processing means 14, and the shape calculation means 146.
It is possible to display the area of the corneal endothelial cells calculated by A display device, a printer, or the like can be applied to the output unit 30.

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

First, when performing image processing, it is necessary to perform preprocessing. The process is started in step 1 (hereinafter abbreviated as S1), and the smoothing process is performed in S2. The smoothing process of this embodiment employs a 5 * 5 median filter to remove spike noise and the like caused by electrical noise of the CCD camera and the like included in the original image. Note that not only the median filter but also a moving average can be used. These S1 and S2 are executed by the preprocessing program in the arithmetic processing means 14.

Next, in S3, the valley line detecting means 141 in the arithmetic processing means 14 detects the valley line. Here, the valley line detection is a filter for detecting a concave portion of the original image. This process will be described in detail with reference to FIGS. 4 and 5. First, 5 * 5 differential operations are performed twice in the x direction and the y direction separately. The first differentiation operation is x in FIG.
And in both directions of y, calculated by Sigma (C _n -b _n) (Note Sigma is the sum symbol). Then, the degree of the concave portion of the pixel is obtained by adding the twice differential value in the x direction and the y direction. This method is suitable for extracting the cell wall portion of the corneal cell image, and finally the calculated value is corrected to a value in the range of 0 to 255. The result is shown in FIG.

After the valley line detection in S3, the target processed image is not the original image but the image (I) that has passed through the valley line detection filter. In this image (I), the value of the pixel most likely to be the valley line is 255 and the value of the pixel most unlikely to be the valley line is 0 within the detection capability of the valley line detection filter. Therefore, only the part of the valley line becomes a white image.

After the detection of the valley line in S3, the process proceeds to S4 and binarization is performed. This binarization is such that first, the average value of the densities of the pixels of the original image is obtained, and an appropriate value is added to the average value to obtain a threshold value. In this embodiment, since it is a pretreatment, it is necessary to leave many samples as cell wall candidates, and 20 is subtracted from the average value to obtain a threshold value. Next, S5 is a target pixel rough extraction, and the process target pixels are narrowed down only to the cell wall candidate group larger than the threshold value.

As described above, the preprocessing is completed by S1 to S5.

Next, the process for extracting the outer shape of the cell wall is started.
First, in S6, the moving direction pixel density detecting means 142 calculates the density distribution for each target pixel in each direction. That is, as shown in FIG. 6, on the radial direction of 24 directions around the image of interest,
The average value of the density values of the pixels included in the range from the radius r to the radius R is calculated. The point to get this data is usually
It is not an integer but a decimal. Then, the data at each location is calculated from the values of the four surrounding pixels using the Lagrange interpolation method, and 24 calculated values are obtained. This S6
If the data obtained in step 1 is plotted on the ordinate and the direction is plotted on the abscissa, a graph such as shown in FIG. 7 is obtained. In addition to the cell wall concentration difference, this data includes many fine vertical movements.

Therefore, in S7, the moving direction pixel density detecting means 142 calculates an average of all five directions of the corresponding direction and the two directions before and after the average in all directions obtained in S6. Reduce small fluctuations.

Further, in S8, the ridge line detecting means 143 causes the S
Each data obtained in 7 is differentiated with respect to the direction, and the process proceeds to S9 to check the portion where this differential coefficient changes from a positive value to a negative value to be a ridge, and that direction is recorded as the ridge direction.

Next, in S10, the ridge line detecting means 143
The number of recorded ridges is checked, and it is determined in S11 whether the number of ridges is two. If the number of ridges is two in S10, the place is regarded as a line and the process proceeds to S11. S1
When it is 0 and the number of ridges is not two, the process returns to S6.

In S11, the line likeness detecting means 144
The first-order line likeness is calculated only for the pixels regarded as lines in 9. The "line-likeness" index is an angle formed by two ridge directions of the pixel.

That is, in S6 to S11, the angle formed by the two ridge directions is obtained by applying the line detection filter in 24 directions to the pixels remaining as the image to be processed in the preprocessing of S1 to S5. is there. Where "line direction"
With respect to, with respect to two directions, forming an angle a is equivalent to forming an angle of 360 degrees-a. So 180
For angles larger than 360 degrees, when it is displayed at an angle of 180 degrees or less by subtracting from 360 degrees, it is 165 degrees in most cases as shown in FIG. Absent. Further, regarding the angles other than 180 degrees, there are two, one larger than 180 degrees and the other smaller than 180 degrees, and therefore the number is doubled. That is, in the actual line portion of the image, the most "line-like portion" is at the center of the line, and two lines are provided on both sides of the same "line-likeness". There 1
If the data of 65 degrees or less is divided by 2 and this value is displayed in FIG. 8, it becomes like a dotted line. Therefore, in order to judge the "linelikeness", it may be judged whether or not it is appropriate by using the dotted line in FIG.

Here, the angle detection capability of the filter will be described in detail. In order to calculate the difference between the angles of the two directions, the filter starts from a predetermined angle, sequentially calculates the two directions to be detected, and calculates the difference between the two angles. Therefore, for example, if the angle is 165 degrees, there are 11 cases where it is calculated as 195 degrees, and 1
There are 13 cases where it is calculated as 65 degrees. If this is expressed by a general formula,

Angle = 180-15 × N (degree)

At the time of,

When there are 180 + 15 × N (degrees), there are (12−N) pieces,

There are (12 + N) cases where 180-15 × N (degrees).

Therefore, although it is detected asymmetrically with respect to 180 degrees, in this filter, 165 degrees and 195 degrees are detected.
There is no point in distinguishing from degrees. Therefore, the angle of 180 degrees or more is also displayed.

In this embodiment, the dotted line portion in FIG. 8 is approximated by a straight line, and the angle and the "linelikeness" are associated with each other by the proportional relationship. Therefore, the case of 180 degrees is the most "linear"
The "line-likeness" decreases as the distance from 180 degrees increases, and the "line-likeness" becomes 0 at 90 degrees.

As described above, in S11, the line likelihood detecting means 1
After the 44 calculates the first-order "line-likeness", the process proceeds to S12, and the line-likeness detection means 144 further calculates the second-order "line-likeness". The secondary "line-likeness" in S12 is to obtain the final "line-likeness" from the primary "line-likeness" in S11 and the pixel value of the image (I). The first-order "line-likeness" calculated in S11 may give good results with the true cell wall, and at the same time, the "line-likeness" may be high even in a place other than the cell wall. Good results can be obtained by multiplying by the value of the image (I) and then normalizing appropriately in order to prevent the "line-likeness" from increasing even in a place other than the cell wall.

Since the line portion of the image has been identified in the steps up to S12 as described above, the line drawing extracting means 145 extracts it as a line drawing in the steps from S13 to S20.

First, in S13 to S15, the line drawing extracting means 145 corrects a portion having a large difference in density by the relaxation method. That is, in S13, the deviation p of each pixel is calculated from the average value of the image, and this p is set as an initial value. The deviation p is standardized to be a decimal number from 0 to 1. Next, in S14, the consistency with surrounding pixels is checked.
In the present embodiment, the average value of p is taken for a total of 8 pixels in the vicinity,
Let this value be q. This q is also a decimal number from 0 to 1. Then, in S15, a new p ′ is calculated by the following formula.

P '= (p * q) / ((p * q) + ((1
-P) * (1-q)))

Equation 1

Now, explaining the characteristics of the first equation, p and q in this equation are completely symmetrical. Now, when p is 0, the new p'is 0 regardless of the value of q. Also, when p is 1, p ′ is 1 regardless of the value of q. Further, when p is 0.5, p ′ becomes the value of q. As a result, p
Alternatively, when it is not clear whether q is a line or not, the attribute of one's pixel can be determined with reference to data other than one's own.

As described above, if the operations from S13 to S15 are repeated an appropriate number of times N, the pixels having the density isolated from the surroundings are removed and the cell wall portions are smoothly connected. In the present embodiment, N = 3 is set to repeat three times. In S17, it is determined whether or not three times have been repeated. If less than three times, the process returns to S14 and after three times, It is designed to proceed to S18.

In S18, binarization is performed, and in S19, the line drawing extracting means 145 thins the line and extracts the line drawing. The threshold for binarization in S18 is the average value of the image in this embodiment. The thinning of S19 degenerates the cell boundary line to a line having a width of 1 pixel because the cell boundary line has various widths. The thinning is performed while preserving the geometrical features (connected state and intersecting state) so that the line is not shortened or become a point.

After the thinning is performed in S19, the shape calculating means 146 calculates the area of the cell and the like in S20. Since the cell wall is composed of continuous lines, the area and the like can be calculated by an appropriate calculation means.

In the present embodiment configured as described above, the area of corneal endothelial cells and the like can be calculated easily and with high accuracy.

FIG. 9 shows corneal endothelial cells obtained by conventional image processing with contrast, and the accuracy thereof is 8
Only about 7.5%. FIG. 10 shows corneal endothelial cells obtained by the image processing of this example. In the corneal endothelial cell measurement result of the present example, there is only one false recognition, the accuracy reaches 99.1%, and there is an effect that the false recognition is extremely small.

[0068]

[Effect] According to the present invention configured as described above, the contour line of the corneal endothelium is identified from the image input means for capturing the image signal of the corneal endothelium and the image input to the image input means. And a valley line detecting means for detecting a concave portion from the image input to the image input means, and an image detected by the valley line detecting means. On the other hand, the moving direction pixel density detecting means for calculating the density of the pixel in the radial direction centered on the pixel extracted as the valley line, and the data obtained by this moving direction pixel density detecting means Differentiate in the angle direction with the pixel extracted as a valley line as the center, and examine the part where the sign of the differential coefficient changes to detect the ridge, and the ridge line detection means detects the ridge. 2 Based on the angle formed by the ridge direction of the ridge, the line image is identified by the "line-likeness" detection means for detecting the "line-likeness" of the ridge and the "line-likeness" detected by this "line-likeness" detection means. However, since it is composed of a line drawing extracting means for identifying the outline of the corneal endothelial cell, it is possible to recognize the morphology of the line unique to the cell wall in the image, and thereby to determine the cell wall. Therefore, there is an effect that erroneous recognition of corneal endothelial cells can be extremely reduced.

Particularly, in the present invention, since the direction of the line and the “line-likeness” are defined and calculated, the weak cell wall data part is
It has the outstanding effect that it can be correctly recognized as a cell wall and noise data can be reliably removed.

[0070]

[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 an arithmetic processing unit 14 according to the present exemplary embodiment.

FIG. 3 is a diagram illustrating the operation of the present embodiment.

FIG. 4 is a diagram illustrating the principle of valley line detection according to the present embodiment.

FIG. 5 is a diagram illustrating valley line detection according to the present embodiment.

FIG. 6 is a diagram illustrating detection of pixel density in a moving direction according to the present exemplary embodiment.

FIG. 7 is a diagram illustrating an average pixel density in a moving direction according to the present exemplary embodiment.

FIG. 8 is a diagram illustrating “linelikeness” of the present embodiment.

FIG. 9 is a diagram showing corneal endothelial cells obtained by conventional image processing using contrast.

FIG. 10 is a diagram showing corneal endothelial cells obtained in this example.

[Explanation of symbols]

1 Corneal endothelial cell measurement device 11 Image input interface 12 A / D converter 13 field memory 14 Arithmetic processing means 141 valley line detection means 142 Moving direction pixel density detecting means 143 Ridge line detection means 144 Line-likeness detection means 145 Line drawing extraction means 146 Shape calculation means 20 Imaging means 30 Output means

─────────────────────────────────────────────────── --Continued from the front page (56) References JP 62-17659 (JP, A) JP 4-361733 (JP, A) JP 50-11332 (JP, A) JP 4- 94884 (JP, A) JP-A-5-324803 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) A61B 3/00-3/16

Claims (4)

(57) [Claims] 1. An image input unit for capturing an image signal of an image of a corneal endothelium, and an arithmetic processing unit for identifying an outline of the corneal endothelial cell from an image input to the image input unit. This arithmetic processing means
A valley line detecting means for detecting a concave portion from the image input to the image input means, and a pixel extracted as a valley line with respect to the image detected by the valley line detecting means
The moving direction pixel density detecting means for calculating the density of the pixel in the radial direction, and the pixel extracted with the data obtained by the moving direction pixel density detecting means as the valley line
Based on the angle formed by the ridge line detecting means for detecting a ridge and the two ridge directions detected by the ridge line detecting means by differentiating in the angle direction and examining the part where the sign of the differential coefficient changes , "Line-likeness" detection means for detecting the "line-likeness" of the ridge, and this "line-likeness"
A corneal endothelial cell measuring device, comprising line drawing extracting means for identifying a line image based on the "line-likeness" detected by the detecting means and for identifying the outline of the corneal endothelial cell. 2. The corneal endothelial cell measuring device according to claim 1, wherein the arithmetic processing means is provided with a shape arithmetic means for calculating the area and the like from the outline of the corneal endothelial cells obtained by the line drawing extracting means. . 3. A first step of capturing an image signal of an image of a corneal endothelium, a second step of detecting a concave line by detecting a concave portion from an image input in the first step, and a second step of detecting the valley line. Image extracted as a valley line from the captured image
The third step of calculating the pixel density in the radial direction around the element and the data obtained in the third step are extracted as valley lines.
Ridge detection by differentiating in the angular direction with respect to the pixel and examining the part where the sign of the differential coefficient changes
Based on the step and the angle formed by the two ridge directions detected in the fourth step, the line image is obtained by the fifth step of detecting the "line-likeness" of the ridge and the "line-likeness" detected in the fifth step. And a sixth step of identifying the outline of the corneal endothelial cell. 4. The method for measuring corneal endothelial cells according to claim 3, further comprising a seventh step of calculating the area and the like from the outline of the corneal endothelial cells obtained in the sixth step.