A method for delineating fungal lesions in fungal keratitis under a microscope

Abstract

The present invention relates to the field of image recognition technology, and specifically to a method for delineating fungal lesions in fungal keratitis under a microscope, comprising: obtaining microscopic images of the epithelium, stroma, and endothelium at different depths of a corneal sample; analyzing the influence range of fungal spores based on the grayscale distribution characteristics of the microscopic image of the epithelium to obtain a connected domain of pending lesions in the epithelium; analyzing the interweaving of highlight portions based on the edge distribution characteristics of the microscopic image of the stroma to obtain a connected domain of pending lesions in the stroma; analyzing the regularity and permeability of cells based on the transparency characteristics of the microscopic image of the endothelium to obtain a connected domain of pending lesions in the endothelium; and fusing the connected domains of the pending lesions of the epithelium, stroma, and endothelium to delineate the fungal lesion area corresponding to the corneal sample. The present invention can avoid the influence of nerve fibers on fungal lesion identification and improve the accuracy of fungal lesion delineation results.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a method for outlining fungal lesions of fungal keratitis under a microscope. Background Art

[0002] Fungal keratitis is an eye disease caused by fungal infection, which typically leads to corneal inflammation. In severe cases, it can affect vision and even cause vision loss. Typical features of this disease include damage to the corneal surface (such as small ulcers or erosions), corneal edema, and corneal opacity. These lesions hinder the normal transmission of light, thereby affecting visual function. To help medical students better understand the pathological characteristics of fungal keratitis, optical anatomical image acquisition using a corneal active optical confocal microscope (SL-OCT, Slit-scan Optical Coherence Tomography) is commonly used in teaching to obtain clear images of corneal lesions, assisting medical students in intuitively learning the pathological manifestations of fungal infections.

[0003] However, in corneal lesion images, fungal hyphae are mainly distributed in the corneal stroma, and their morphology and orientation are highly similar to corneal nerve fibers. Examples of nerve fibers are shown in the figure below. Figure 1 An example of fungal hyphae is shown in Figure 2 This similarity can cause confusion among medical students when observing and identifying fungal lesions, especially when accurately outlining and defining them, as corneal nerve fibers can interfere with their ability to accurately identify and assess their extent. This interference not only affects students' ability to accurately identify and assess the extent of fungal lesions, but can also lead to errors in lesion identification during teaching, ultimately impacting students' development of their skills. Summary of the Invention

[0004] In order to solve the technical problem that the fungal hyphae in the corneal lesion images are highly similar in morphology and orientation to corneal nerve fibers, which causes confusion among medical students when observing and identifying fungal lesions, the present invention aims to provide a method for delineating fungal lesions of fungal keratitis under a microscope. The technical solution adopted is as follows:

[0005] One embodiment of the present invention provides a method for delineating fungal lesions of fungal keratitis under a microscope, the method comprising the following steps:

[0006] Obtain microscopic images of the epithelium, stroma, and endothelium at different depths of corneal samples;

[0007] The influence range of fungal spores was analyzed based on the grayscale distribution characteristics of the microscopic image of the epithelium, and the connected domain of the suspected lesion in the epithelium was obtained;

[0008] The interweaving of the highlight parts is analyzed according to the edge distribution characteristics of the microscopic image of the matrix layer, and the connected domain of the undetermined lesion in the matrix layer is obtained;

[0009] The regularity and permeability of the cells were analyzed based on the transparency characteristics of the microscopic images of the endothelial layer, and the connected domain of the endothelial lesion to be determined was obtained;

[0010] The connected domains of the suspected lesions in the epithelium, stroma, and endothelium were fused to outline the fungal lesion area corresponding to the corneal sample.

[0011] Furthermore, the analysis of the influence range of the fungal spores based on the grayscale distribution characteristics of the microscopic image of the epithelial layer to obtain the connected domain of the undetermined lesion in the epithelial layer includes:

[0012] Performing superpixel segmentation on the microscopic image of the epithelium to obtain various superpixel clusters; performing fitting analysis on each superpixel block in the superpixel cluster to obtain a fitting evaluation value of each superpixel cluster;

[0013] Classify each superpixel cluster according to its grayscale value to obtain various superpixel clusters; the superpixel cluster with the highest grayscale is recorded as the spore cluster;

[0014] The connected domain of the pending lesion in the epithelial layer is obtained according to the fitting evaluation values ​​of each superpixel cluster and each spore cluster.

[0015] Furthermore, obtaining the connected domain of the pending lesion in the epithelial layer according to the fitting evaluation values ​​of each superpixel cluster and each spore cluster includes:

[0016] Obtain an enclosed area of ​​all the spore clusters in the microscopic image of the epithelial layer, and continuously adjust the area of ​​the enclosed area through an optimization algorithm until the area of ​​the enclosed area is maximized and the cumulative value of the fitting evaluation values ​​of all superpixel clusters in the enclosed area is minimized, and the final enclosed area is determined as the connected domain of the pending lesion in the epithelial layer; wherein the enclosed area is the area containing all the spore clusters.

[0017] Furthermore, the method of analyzing the interweaving of the highlight portion according to the edge distribution characteristics of the microscopic image of the matrix layer to obtain the connected domain of the undetermined lesion in the matrix layer includes:

[0018] Performing edge detection on the microscopic image of the matrix layer to obtain edge pixel points, and performing threshold segmentation on the microscopic image of the matrix layer to obtain high-threshold pixel points;

[0019] The edge pixel points selected from all high-threshold pixel points are used as feature pixel points;

[0020] Determining a first significance level of hyphae features of each characteristic line according to each characteristic pixel point; wherein the first significance level of hyphae features is used to characterize the straightness of the characteristic line;

[0021] Determining the second significance level of hyphae characteristics of each characteristic line based on the interlaced characteristics between the characteristic lines;

[0022] Combining the first and second significant degrees of the hyphae characteristics of each characteristic line, screening out the hyphae segments to be determined;

[0023] The minimum connected domain containing all the hyphae segments to be determined in the microscopic image of the matrix layer is used as the connected domain of the lesion to be determined in the matrix layer.

[0024] Furthermore, determining the first significance level of the hyphae feature of each characteristic line according to each characteristic pixel point includes:

[0025] Connect adjacent feature pixel points within the eight neighborhoods to form feature lines, and obtain several feature lines;

[0026] For any characteristic line, obtain the length of the line connecting the two end points of the characteristic line, and compare the characteristic line length with the length of the line connecting to obtain the deviation;

[0027] Draw a perpendicular line from each characteristic pixel point on the characteristic line to the line segment of the connecting length, with the direction of the perpendicular line being perpendicular to the line segment of the connecting length, and determine the sharpness based on the difference in the number of perpendicular lines on both sides of the line segment of the connecting length;

[0028] Obtain edge chain codes of characteristic pixel points on the characteristic line, and obtain the tortuosity based on the difference between adjacent code numbers in the edge chain codes;

[0029] Obtaining a first significance level of hyphae characteristics of the characteristic line according to the deviation, the sharpness, and the tortuosity of the characteristic line;

[0030] Among them, the deviation, the sharpness, and the tortuosity are all negatively correlated with the first significant degree of the mycelium characteristics.

[0031] Furthermore, the determining of the second significance level of the hyphae characteristics of each characteristic line based on the interlaced characteristics between the characteristic lines includes:

[0032] For any feature line, the feature line closest to the feature line endpoint is recorded as the first reference line, and the feature line second closest to the feature line endpoint is recorded as the second reference line;

[0033] Extending the characteristic line in the direction of the undetermined reference line and then connecting it with the undetermined reference line to obtain a new characteristic line; wherein the undetermined reference line is the first reference line or the second reference line;

[0034] Obtain each inflection point on the pending reference line, and then obtain the angle formed by the line connecting the two end points of the pending reference line and any inflection point, and use the minimum angle as the target angle corresponding to the pending reference line;

[0035] The second significance level of the hyphae feature of the feature line is determined according to the similarity degree of grayscale distribution between the feature line and the new feature line and the target angle corresponding to the first reference line and the second reference line of the feature line.

[0036] Furthermore, determining the second significance level of the hyphae feature of the characteristic line according to the degree of similarity in grayscale distribution uniformity between the characteristic line and the new characteristic line and the target angle corresponding to the first reference line and the second reference line of the characteristic line includes:

[0037] Obtaining a first grayscale variance of the characteristic line, a second grayscale variance of the new characteristic line when the pending reference line is the first reference line, and a third grayscale variance of the new characteristic line when the pending reference line is the second reference line;

[0038] determining a first similarity based on a difference between the first grayscale variance and the second grayscale variance, and determining a second similarity based on a difference between the first grayscale variance and the third grayscale variance;

[0039] The second significance level of the hyphae feature of the characteristic line is determined according to the first similarity, the second similarity, and the target angle between the first reference line and the second reference line.

[0040] Furthermore, determining a second significance level of the hyphae feature of the characteristic line by using the first similarity, the second similarity, and a target angle between the first reference line and the second reference line includes:

[0041] Calculate the ratio of the second similarity to the average target angle corresponding to the second reference line, and calculate the ratio of the first similarity to the average target angle corresponding to the first reference line. Perform a difference calculation on the two ratios and then normalize them to determine the second significance level of the hyphae feature of the characteristic line.

[0042] Furthermore, the analysis of the regularity and permeability of cells based on the transparency characteristics of the microscopic image of the endothelial layer to obtain the connected domain of the pending lesion in the endothelial layer includes:

[0043] Acquiring a high-threshold region of the microscopic image of the endothelial layer, performing connected domain detection on the high-threshold region to obtain each cell connected domain; performing hexagonal fitting on each cell connected domain to obtain a loss value and a fitting side length;

[0044] Determining the deformation coefficient of each of the cell-connected domains according to the loss value of the cell-connected domain and the difference between the fitted side length and the average side length; wherein the average side length is the average of the fitted side lengths of all the cell-connected domains;

[0045] According to the distribution of pixel points outside the cell connected domain in the microscopic image of the endothelial layer and the deformation coefficient, the microscopic image of the endothelial layer is segmented to obtain the undetermined lesion connected domain of the endothelial layer.

[0046] Furthermore, the image segmentation of the microscopic image of the endothelial layer is performed based on the distribution of pixel points outside the cell connected domain in the microscopic image of the endothelial layer and the deformation coefficient to obtain the undetermined lesion connected domain of the endothelial layer, including:

[0047] Pixels outside the cell connected domain in the microscopic image of the endothelial layer are used as pixels to be detected, and the minimum distance value between the pixel to be detected and all cell connected domains is determined in the microscopic image of the endothelial layer;

[0048] Determine a window area with the pixel to be detected as the center, and determine each maximum point in the window area; wherein the maximum point is the pixel point with the largest grayscale value within its eight neighborhoods;

[0049] Performing information entropy calculation on the distance between each maximum point and the pixel to be detected in the window area, and obtaining the fungal infection characteristic value of each pixel to be detected according to the minimum distance value, the information entropy and the deformation coefficient;

[0050] Threshold segmentation is performed based on the fungal infection characteristic value of each pixel to be detected in the microscopic image of the endothelial layer to obtain high-threshold pixel points;

[0051] The high-threshold pixel point, the cell connected domain adjacent to the high-threshold pixel point, and the cell connected domain surrounded by the high-threshold pixel point are merged to obtain the undetermined lesion connected domain of the endothelial layer.

[0052] The present invention has the following beneficial effects:

[0053] Compared with the existing fungal lesion delineation method that ignores the influence of corneal nerve fibers, the present invention provides a fungal lesion delineation method for fungal keratitis under a microscope. This method determines the connected domains of the pending lesions in the epithelial layer, stromal layer and endothelial layer by analyzing the distinguishing features of fungal hyphae and nerve fibers in fiber images of different depth layers, and then fuses the connected domains of the three layers of pending lesions to delineate the fungal lesion area corresponding to the corneal sample. This method can achieve accurate identification and range assessment of fungal lesions, and provide more reliable analysis data for the teaching of fungal keratitis. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 This is an example diagram of a nerve fiber;

[0056] Figure 2 This is an example diagram of fungal hyphae;

[0057] Figure 3 This is a flowchart of the steps of a method for delineating fungal lesions of fungal keratitis under a microscope according to an embodiment of the present invention;

[0058] Figure 4 This is an example of a microscopic image of the epithelium;

[0059] Figure 5 This is an example diagram of a microscopic image of the epithelium after filtering and morphological processing in an embodiment of the present invention;

[0060] Figure 6 This is a flowchart of step S2 in an embodiment of the present invention;

[0061] Figure 7 This is an example of the result of superpixel segmentation of a microscopic image of the epithelium in an embodiment of the present invention;

[0062] Figure 8 This is a flowchart of step S3 in an embodiment of the present invention;

[0063] Figure 9 This is a flowchart of step S4 in an embodiment of the present invention;

[0064] Figure 10 This is an example diagram of the fungal lesion area corresponding to the corneal sample in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] To further illustrate the technical means and effects employed by the present invention to achieve its intended objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementations, structures, features, and effects of the technical solutions proposed by the present invention. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0066] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0067] The application scenarios targeted by the present invention may be:

[0068] Fungal hyphae observed under a microscope are mainly distributed in the corneal stroma, and their morphology and direction are highly similar to corneal nerve fibers. Traditional methods based on the morphological characteristics of hyphae may misjudge some nerve fibers as fungal hyphae, resulting in interference in the accurate delineation and definition of fungal lesions, thereby affecting the precise identification and range assessment of fungal lesions, posing a challenge for medical students to accurately identify fungal lesions.

[0069] One embodiment of the present invention provides a method for delineating fungal lesions of fungal keratitis under a microscope, such as Figure 3 As shown, the following steps are included:

[0070] S1, Microscopic images of the epithelium, stroma, and endothelium at different depths of a corneal sample were obtained.

[0071] In order to outline the fungal lesions of fungal keratitis, it is first necessary to collect a microscopic image of the corneal sample. The image property of the microscopic image here is a grayscale image, and each pixel in the grayscale image has a corresponding grayscale value.

[0072] Specifically, the eye is correctly positioned under the microscope, and a special bracket or device is usually used to ensure that the eye remains stable during positioning. To avoid the effects of eye discomfort or involuntary blinking, local anesthetic eye drops are usually instilled into the eye to keep the eye calm and reduce interference. The microscope scans the different depths of the cornea layer by layer using optical sectioning technology. Each layer of the cornea is imaged by stimulating different layers of the cornea with laser light and detecting the corresponding light signals that have been reflected or scattered back. The light signal after laser irradiation is captured by a detector, which receives the reflected signal and converts it into a digital image. The digital image is processed to obtain microscopic images of the epithelium, stroma, and endothelium. Image processing includes but is not limited to filtering, morphological processing, and grayscale processing.

[0073] The microscope described above can be a confocal microscope, which uses a laser as a light source. The laser used is typically a low-power, single-wavelength laser (e.g., 488nm or 785nm). The laser passes through the microscope's optical system to illuminate the surface of the cornea. It is worth noting that the laser beam is concentrated only at the focal point and does not diffuse into the surrounding area.

[0074] It is worth noting that, based on the intensity and properties of reflected light, confocal microscopy can obtain information about different layers of the cornea. Because each layer is scanned independently, the acquired images are continuous slices, reflecting the details and hierarchical structure of the cornea. This allows for microscopic images of the epithelium, stroma, and endothelium at different depths of the corneal sample.

[0075] So far, this embodiment has obtained a microscopic image of the epithelial layer, a microscopic image of the stroma layer, and a microscopic image of the endothelial layer of the corneal sample.

[0076] It should be noted that when fungal keratitis occurs in corneal specimens, the fungal infection typically involves multiple layers of the cornea, particularly the epithelium, stroma, and endothelium. Therefore, microscopic images of corneal specimens at different depths are obtained as basic image data to analyze the image features of fungal hyphae in each layer to obtain the connected domain of the unidentified lesions in each layer. Ultimately, the connected domains of the unidentified lesions identified in the three layers are superimposed and analyzed. Compared to identifying the connected domain of the fungal lesions in only one layer, this embodiment can delineate a more accurate connected domain of the fungal lesions for medical students' teaching and analysis.

[0077] S2, analyzing the affected range of fungal spores according to the grayscale distribution characteristics of the microscopic image of the epithelium, and obtaining the connected domain of the pending lesion in the epithelium.

[0078] The fiber image of the epithelium contains epithelial cells. The epithelium infected by fungi often has a rough, discontinuous, and irregular morphology. Especially in the lesion area, a large range of epithelial defects or ulcers may be observed. Therefore, when fungal infection occurs, epithelial cells in some areas will fall off, causing some local fungal spores and irregular changes in the cell edges in the epithelium. Fungal spores will appear in the form of highlights between cells. Therefore, by analyzing the influence range of fungal spores, the connected domain of the epithelial lesion can be obtained. Among them, the example of the microscopic image of the epithelium is shown in the figure below. Figure 4 As shown in the figure, the example of the microscopic image of the epithelium after filtering and morphological processing is shown in Figure 5 shown.

[0079] The above step S2 can be Figure 6 The steps S21 to S23 shown implement:

[0080] S21, performing superpixel segmentation on the microscopic image of the epithelium to obtain various superpixel clusters; performing fitting analysis on each superpixel block in the superpixel cluster to obtain a fitting evaluation value of each superpixel cluster.

[0081] Here, the fitting evaluation value can characterize the regularity of superpixel distribution in a superpixel cluster. The larger the fitting evaluation value, the more regular the superpixel distribution in the cluster. By segmenting the microscopic image of the epithelium into regions with similar color characteristics, the image structure can be simplified and the efficiency of subsequent image processing can be improved. In addition, step S2 is to obtain the connected domain of the undetermined lesion in the epithelium. Superpixel clustering can generally better preserve the boundary information of the image, which is helpful for connected domain analysis. Fungal infections in the epithelium usually manifest as large-scale defects or ulcers, which result in poor regularity of image pixel distribution. Therefore, when determining the connected domain of the undetermined lesion in the epithelium, the regularity of the superpixel distribution of the lesion connected domain should be poor, that is, small.

[0082] In this embodiment, the implementation process of superpixel segmentation is prior art and will not be described in detail here. The superpixel clusters are fitted using a chi-square test, and the fitting effect is evaluated by calculating the chi-square statistic between the color distribution of each superpixel and the expected distribution to obtain a fitting evaluation value. The high or low fitting evaluation value can be determined by the chi-square statistic. The better the fitting effect, the more regular the distribution of superpixels within the superpixel cluster. The process of fitting superpixel clusters using the chi-square test is an existing basis and will not be described in detail here. Of course, the implementer can also use other fitting methods to implement it, and there is no specific limitation on the fitting method of superpixel clusters, such as Jensen-Shannon divergence or Bhattacharyya distance.

[0083] Among them, the results of superpixel segmentation of the microscopic image of the epithelium are shown in the figure below. Figure 7 As shown, Figure 7 This is a superpixel region annotation map. The annotation color does not represent the actual grayscale of the superpixel.

[0084] S22, classifying various superpixel clusters according to the grayscale values ​​of each superpixel cluster; recording the superpixel cluster with the highest grayscale as a spore cluster.

[0085] Here, the spore cluster refers to the cluster with the largest cumulative grayscale value of all superpixels. The larger the grayscale value, the more likely it is that the fungal spores are presented in a highlighted form. Therefore, the superpixel cluster with the highest grayscale is recorded as the spore cluster.

[0086] Using a grayscale similarity threshold to group superpixel clusters with similar grayscale values, several classes of superpixel clusters can be obtained. Specifically, the grayscale value difference between each superpixel cluster is calculated, and superpixel clusters with a grayscale value difference less than the grayscale similarity threshold are grouped together. The grayscale similarity threshold and the number of classes can be set by the implementer based on experience and actual conditions, and are not specifically limited.

[0087] S23, obtaining the connected domain of the pending lesion in the epithelial layer according to the fitting evaluation value of each superpixel cluster and each spore cluster.

[0088] Specifically, the enclosed area of ​​all spore clusters in the microscopic image of the epithelial layer is obtained, and the area of ​​the enclosed area is continuously adjusted through an optimization algorithm until the area of ​​the enclosed area is maximized and the cumulative value of the fitting evaluation values ​​of all superpixel clusters in the enclosed area is minimized. The final enclosed area is determined as the connected domain of the pending lesion in the epithelial layer.

[0089] The enclosing region is the area that contains all spore clusters. It needs to be as large as possible to cover more possible lesion areas. The cumulative value of the fitting evaluation values ​​of all superpixel clusters within the enclosing region should be as small as possible to ensure high consistency within the region. The optimization algorithm can be a greedy algorithm, dynamic programming, or genetic algorithm.

[0090] It is worth noting that for superpixel clusters where some superpixel blocks in the enclosing area are not in the enclosing area, and some superpixel blocks are not in the enclosing area, the fitting evaluation value is recalculated based on some superpixel blocks in the enclosing area.

[0091] So far, this embodiment has obtained the connected domain of the pending lesion in the epithelial layer.

[0092] S3, analyzing the interweaving of the highlight portion according to the edge distribution characteristics of the microscopic image of the matrix layer, and obtaining the connected domain of the pending lesion in the matrix layer.

[0093] When fungal keratitis occurs in corneal specimens, obvious hyphal branches appear in the stroma. In extreme cases, fungal hyphae may penetrate the stroma into the endothelium, causing damage to endothelial cells or dysfunction. Therefore, identification of the stroma layer showing fungal manifestations is necessary when performing fungal identification.

[0094] The stroma is the most common layer of corneal infection. Fungal infections in this layer manifest as distinct, elongated, branching fungal hyphae, with distinct curvature or branching, and their reflectivity is prominent. Therefore, based on the edge distribution characteristics of microscopic images of the stroma, we analyze the interweaving of highlights and identify the connected domains of the potential lesions in the stroma.

[0095] The above step S3 can be Figure 8The steps S31 to S33 shown implement:

[0096] S31, edge detection is performed on the microscopic image of the matrix layer to obtain edge pixel points, and threshold segmentation is performed on the microscopic image of the matrix layer to obtain high-threshold pixel points; the edge pixel points selected from all the high-threshold pixel points are used as feature pixel points.

[0097] Here, the feature pixel points are pixel points that contain two pixel properties, namely high threshold pixel points and edge pixel points. The feature pixel points can be used to analyze the distribution of edge pixel points located in the highlight area of ​​the fiber image of the matrix layer, which can be used to express the interweaving characteristics of the highlight part.

[0098] In this embodiment, the edge detection method is canny edge detection, and the threshold segmentation method is Otsu threshold. The implementation processes of both are existing technologies and will not be described in detail here. Of course, the implementer can also adopt other edge detection and threshold segmentation methods, which are not specifically limited here.

[0099] S32, determining a first significance level of the hyphae feature of each characteristic line according to each characteristic pixel point; and determining a second significance level of the hyphae feature of each characteristic line according to the interlaced features between each characteristic line.

[0100] Here, the first significance level of the hyphae feature is used to indicate how fast and straight the characteristic lines are. The greater the first significance level of the hyphae feature, the more the characteristic lines fit the branching characteristics of the fungal hyphae. The second significance level of the hyphae feature is used to indicate the degree of interweaving of the characteristic lines. The greater the second significance level of the hyphae feature, the more the characteristic lines fit the characteristics of the accumulation of fungal hyphae.

[0101] First, the first significance level of the hyphae feature of each feature line is determined based on each feature pixel point, including:

[0102] In the first step, adjacent feature pixel points within the eight neighborhoods are connected to form feature lines, thereby obtaining several feature lines.

[0103] In this embodiment, characteristic lines are obtained to analyze the straightness or curvature of the branches of the highlighted part in the microscopic image of the matrix layer. The number of characteristic pixels contained in the characteristic lines must not be less than 5. That is, for characteristic lines with a length that is too short, this embodiment does not perform feature analysis on them; adjacent characteristic pixels refer to the characteristic pixels that are closely adjacent to the previous and next ones. There is a possibility that the characteristic lines are the edges of fungal hyphae or nerve fibers.

[0104] The second step is to obtain the length of the line connecting the two end points of any characteristic line for any characteristic line, and compare the characteristic line length with the length of the line connecting the two end points to obtain the deviation.

[0105] In this embodiment, the length refers to the number of pixels, and the deviation can be the difference between the length of the feature line and the length of the connecting line, or the ratio of the length of the feature line to the length of the connecting line; the greater the length of the feature line is than the length of the connecting line, the more serious the deviation of the feature line from the straight line is, that is, the greater the deviation of the feature line, the smaller the first significance of the mycelium feature will be.

[0106] The third step is to draw a perpendicular line from each feature pixel point on the feature line to the line length segment. The direction of the perpendicular line is perpendicular to the line length segment. The sharpness is determined according to the difference in the number of perpendicular lines on both sides of the line length segment.

[0107] In this embodiment, the number of perpendicular lines on both sides of a line segment is calculated. If the line segment is parallel to the horizontal line, the number of perpendicular lines on the upper and lower sides of the line segment is counted. If the line segment is perpendicular to the horizontal line, the number of perpendicular lines on the left and right sides of the line segment is counted. The absolute value of the difference between the number of perpendicular lines on both sides is used as the sharpness. The sharpness can indicate whether the characteristic line is bent toward one side. The greater the degree of bending toward one side, the greater the sharpness.

[0108] The fourth step is to obtain the edge chain code of the feature pixel points on the feature line, and obtain the tortuosity based on the difference between adjacent code numbers in the edge chain code.

[0109] In this embodiment, the process of determining the tortuosity by edge chain code is a prior art and is not within the scope of protection of the present invention, and will not be elaborated here.

[0110] The fifth step is to obtain the first significance degree of the hyphae characteristics of the characteristic lines according to the deviation, sharpness and tortuosity of the characteristic lines.

[0111] In this embodiment, the deviation, sharpness, and tortuosity are all negatively correlated with the first significant degree of hyphae characteristics.

[0112] As an example, the calculation formula for the first significance level of the hyphae feature of the i-th feature line may be:

[0113] Where, Indicates the first significant degree of hyphae characteristics of the i-th characteristic line, represents the deviation of the i-th characteristic line, represents the sharpness of the i-th feature line, It represents the tortuosity of the ith characteristic line, m represents a non-zero constant, and its empirical value is 0.01, which is used to prevent the denominator of the fraction from being zero.

[0114] In the calculation formula for the first significance degree of mycelial characteristics, the greater the deviation, the greater the curvature of the characteristic line; the greater the sharpness, the greater the curvature of the characteristic line; the greater the tortuosity, the greater the curvature of the characteristic line; and the greater the curvature, the less the characteristic line conforms to the straight and sharp morphological characteristics of fungal mycelium, and the smaller the first significance degree of the mycelial characteristics; deviation, sharpness, and tortuosity are usually dimensionless indicators used to describe curve characteristics, and therefore have no clear units.

[0115] It should be noted that quantifying the straightness and sharpness of the characteristic lines from three different aspects can more comprehensively analyze the degree of curvature of the characteristic lines, making the numerical accuracy of the first significance degree of the hyphae characteristics that characterize the hyphae morphological characteristics higher, and thus making the subsequent connected domain of the undetermined lesions in the matrix layer obtained based on the first significance degree more reliable.

[0116] Secondly, based on the interlaced features between the characteristic lines, the second significance level of the hyphae features of each characteristic line is determined.

[0117] First of all, it should be noted that, since some nerve fibers may have similarities to the straight and sharp features of hyphae, it is inaccurate to detect hyphae only according to the first significant degree of hyphae features. Therefore, in order to improve the recognition accuracy of the connected domain of the pending lesions in the matrix layer, it is necessary to further analyze the fungal hyphae features of the characteristic lines. For the fungal hyphae in the matrix layer, they are bifurcated, disordered, and bamboo-like or branch-like. The refractive index and morphology of the hyphae of different species are different. Therefore, the fungal hyphae are piled up with each other due to their disordered arrangement and bifurcation, resulting in the situation where the hyphae are intertwined with each other.

[0118] Here, the second significance level of the hyphae feature is used to indicate the degree of interlacing of the characteristic lines. The more severe the interlacing, the more obvious the mutual accumulation feature between the fungal hyphae is, and the higher the possibility that the characteristic lines belong to the fungal hyphae.

[0119] In the first step, for any feature line, the feature line closest to the feature line endpoint is recorded as the first reference line, and the feature line second closest to the feature line endpoint is recorded as the second reference line.

[0120] It should be noted that the purpose of obtaining the first reference line and the second reference line of the characteristic line is to analyze the interlacing relationship between the characteristic line and other characteristic lines that are closer to it, so as to quantify whether the characteristic line is in an interlaced state and the degree of interlacing, thereby improving the data analysis basis for the subsequent quantification of the interlacing coefficient.

[0121] In this embodiment, each feature line has two corresponding endpoints, each with a first reference line and a second reference line. The distance between an endpoint and feature lines other than the feature line it is on can be calculated using existing point-to-line distance calculations, which will not be described in detail here. If an endpoint corresponds to two feature lines with the same distance, one of the two feature lines is used as the first reference line, and the other as the second reference line.

[0122] The second step is to extend the feature line in the direction of the pending reference line and then connect it to the pending reference line to obtain a new feature line, where the pending reference line is the first reference line or the second reference line.

[0123] In this embodiment, the feature line extension refers to extending in the direction of the pending reference lines along the endpoints on both sides until the feature line is connected with the pending reference lines at both ends to obtain the extended feature line. The extended feature line is used as the new feature line, and each feature line can correspond to two different new feature lines.

[0124] The third step is to obtain the inflection points on the pending reference line, and then obtain the angle formed by the line connecting the two end points of the pending reference line and any inflection point, and use the minimum angle as the target angle corresponding to the pending reference line.

[0125] Specifically, a second-order derivative calculation is performed on a reference line to detect inflection points, thereby obtaining each inflection point on the reference line. For each inflection point, the angle formed by the line connecting the two endpoints of the reference line and the inflection point is obtained, thereby obtaining all angles corresponding to the reference line, with the minimum angle being used as the target angle for the reference line. The implementation of the second-order derivative calculation and inflection point detection are both prior art and fall outside the scope of this invention, and will not be elaborated upon here.

[0126] When determining the target angle, the smaller the angle corresponding to the pending reference line is, the greater the possibility of intersection between fungal hyphae. Therefore, when determining the target angle, the minimum angle is used as the target angle corresponding to the pending reference line.

[0127] The fourth step is to determine the second significance level of the hyphae feature of the feature line according to the similarity degree of grayscale distribution between the feature line and the new feature line, and the target angle corresponding to the first reference line and the second reference line of the feature line.

[0128] Here, the uniform grayscale distribution is used to analyze the disorder of fungal hyphae arrangement, while the target angle can analyze the bifurcation of fungal hyphae. By combining the interlaced characteristics of the two aspects, the second significant degree of the hyphae characteristics of the characteristic lines can be determined.

[0129] As an optional embodiment, determining the second significance level of the hyphae characteristics of the characteristic line includes:

[0130] The first sub-step is to obtain the first grayscale variance of the characteristic line, the second grayscale variance of the new characteristic line when the pending reference line is the first reference line, and the third grayscale variance of the new characteristic line when the pending reference line is the second reference line.

[0131] In this embodiment, the grayscale variance can be calculated by the grayscale value of each characteristic pixel point on the characteristic line and the new characteristic line. For the sake of distinction, the grayscale variance corresponding to the characteristic line is recorded as the first grayscale variance, the grayscale variance corresponding to the new characteristic line when the pending reference line is the first reference line is recorded as the second grayscale variance, and the grayscale variance corresponding to the new characteristic line when the pending reference line is the second reference line is recorded as the third grayscale variance.

[0132] In the second sub-step, a first similarity is determined according to a difference between the first grayscale variance and the second grayscale variance, and a second similarity is determined according to a difference between the first grayscale variance and the third grayscale variance.

[0133] In this embodiment, the absolute value of the difference between two grayscale variances is processed using the exponential function exp(-) with a natural constant as the base to obtain a uniform and similar grayscale distribution. Of course, implementers may also use other methods to quantify similarity, which are not specifically limited here, such as using the Bhattacharyya distance to analyze the distribution similarity of two grayscale sets.

[0134] In the third sub-step, a second significance level of the hyphae feature of the characteristic line is determined based on the first similarity, the second similarity, and the target angle between the first reference line and the second reference line.

[0135] In this embodiment, the greater the first similarity corresponding to the first reference line and the smaller the target angle, and the smaller the second similarity corresponding to the second reference line and the larger the target angle, the greater the possibility that the first reference line and the second reference line of the characteristic line intersect, and the greater the second significance of the mycelium feature of the characteristic line.

[0136] Specifically, the ratio of the second similarity to the average target angle corresponding to the second reference line is calculated, and the ratio of the first similarity to the average target angle corresponding to the first reference line is calculated. The two ratios are difference-calculated and then normalized to determine the second significance level of the mycelial feature of the characteristic line.

[0137] As an example, the calculation formula for the second significance level of the hyphae feature of the i-th feature line may be:

[0138] Where, Indicates the second most significant degree of hyphae characteristics of the i-th characteristic line, Represents a hyperbolic function, which is used to normalize data. represents the second similarity, represents the mean value of the target angle corresponding to the second reference line, represents the first similarity, represents the mean value of the target angle corresponding to the first reference line, Represents the absolute value function.

[0139] In the calculation formula of the second significance of hyphae features, for the i-th feature line, the greater the second similarity corresponding to the second reference line, the smaller the mean value of the target angle, that is, The larger the value is, the smaller the second similarity corresponding to the first reference line is, and the larger the mean value of the target angle is. The smaller the value, the greater the possibility of the intersection between the first reference line and the second reference line, and the greater the second significance of the hyphae feature.

[0140] S33, combining the first and second significance levels of hyphae features of each characteristic line to screen out the hyphae segments to be determined; and taking the minimum connected domain containing all the hyphae segments to be determined in the microscopic image of the matrix layer as the connected domain of the pending lesion in the matrix layer.

[0141] In this embodiment, for any characteristic line, the product of the first significance degree of the hyphae feature and the second significance degree of the hyphae feature of the characteristic line is calculated as the hyphae judgment index, and the hyphae judgment index of each characteristic line is obtained; a judgment threshold is set as the hyphae judgment standard, and the empirical value is taken as 0.7. In implementation, the size of the judgment threshold can be set according to specific actual conditions and requirements; the characteristic lines whose hyphae judgment index is greater than the judgment threshold of 0.7 are used as the pending hyphae segments. After obtaining all the pending hyphae segments corresponding to the matrix layer, a connected domain analysis is performed on all the pending hyphae segments, and the minimum connected domain is used as the pending lesion connected domain of the matrix layer, that is, the fungal infection area of ​​the matrix layer.

[0142] So far, this embodiment has obtained the connected domain of the undetermined lesion in the matrix layer.

[0143] S4, analyzing the regularity and permeability of the cells based on the transparency characteristics of the microscopic image of the endothelial layer, and obtaining the connected area of ​​the pending lesion in the endothelial layer.

[0144] When fungal infection penetrates deep into the endothelium, there are usually more severe lesions and structural damage. Since the main function of the endothelium is to maintain corneal transparency, when the infection penetrates deep into the endothelium, it will affect the fluid balance and cause corneal turbidity. When the cornea is severely turbid, the penetration of light and the reflected light will be significantly reduced, which will significantly affect the imaging quality of the microscopic image of the endothelium. Therefore, it is necessary to analyze the regularity and permeability of cells through the transparency characteristics of the microscopic image of the endothelium to obtain the connected domain of the undetermined lesion in the endothelium.

[0145] The above step S4 can be Figure 9 The steps S41 to S43 shown implement:

[0146] S41, obtaining a high-threshold region of the microscopic image of the endothelial layer, performing connected domain detection on the high-threshold region to obtain each cell connected domain; performing hexagonal fitting on each cell connected domain to obtain a loss value and a fitting side length.

[0147] In this embodiment, when the fungal infection penetrates deep into the endothelial layer, the structure of some cells is damaged. Therefore, it is necessary to obtain the loss value and fitting side length of each cell connection domain in order to quantify the degree of damage to the cell structure and then obtain the undetermined lesion connection domain of the endothelial layer.

[0148] Specifically, the microscopic image of the endocortex is segmented using Otsu threshold segmentation to obtain high-threshold areas; connected domain detection is performed on the high-threshold areas to obtain several high-threshold connected domains, which are used as cell connected domains; for each cell connected domain, the least squares method is used to perform hexagon fitting on each cell connected domain to obtain the loss value of each cell connected domain and the total side length of the hexagon, and the total side length of the hexagon is used as the fitted side length of the cell connected domain.

[0149] Among them, the implementation process of Otsu threshold segmentation and the acquisition of loss value and fitting side length in the implementation process of least squares method are both existing technologies and are not within the scope of protection of the present invention, and will not be elaborated here.

[0150] S42, determining the deformation coefficient of each cell connected domain according to the loss value of the cell connected domain and the difference between the fitted side length and the average side length; wherein the average side length is the average of the fitted side lengths of all cell connected domains.

[0151] In this embodiment, both the loss value and the difference value represent the degree of loss. The difference between the fitted side length and the average side length is normalized to obtain a normalized value of the difference, such as a norm function. Therefore, the product of the loss value and the normalized difference value can be used as the deformation coefficient of the corresponding cell connectivity domain. A larger loss value and a larger difference between the fitted side length and the average side length indicate a more severe deformation loss in the cell connectivity domain and a larger deformation coefficient.

[0152] S43 , performing image segmentation on the microscopic image of the endocortex according to the distribution of pixel points outside the cell connected domain in the microscopic image of the endocortex and the deformation coefficient, to obtain a connected domain of the pending lesion in the endocortex.

[0153] Here, pixels outside the cell connectivity domain in the microscopic image of the endothelial layer are likely extracellular infected pixels. Edema caused by endothelial infection can lead to uneven water distribution within the endothelial layer, which in turn causes uneven scattering of light within the corneal endothelial layer, resulting in uneven brightness. Therefore, we analyze the light structure characteristics of the area surrounding the pixel to be detected.

[0154] As an optional embodiment, obtaining the connected domain of the undetermined lesion in the endothelial layer includes:

[0155] In the first step, pixels outside the cell connected domain in the microscopic image of the endothelial layer are taken as pixels to be detected, and the minimum distance value between the pixel to be detected and all cell connected domains is determined in the microscopic image of the endothelial layer.

[0156] In this embodiment, for any pixel to be detected, the distance between the pixel to be detected and each connected cell domain is calculated, and the minimum distance value is determined from all distances to obtain the minimum distance value corresponding to the pixel to be detected. The distance between the pixel to be detected and the connected cell domain can be obtained by finding the position on the connected domain boundary closest to the pixel to be detected and calculating the distance between these two points.

[0157] In the second step, a window area is determined with the pixel to be detected as the center, and each maximum point is determined within the window area; the maximum point is the pixel with the largest grayscale value within its eight neighborhoods.

[0158] In this embodiment, the size of the window area is determined according to the size of the microscopic image of the endothelial layer. The larger the size of the microscopic image, the larger the size of the window area. The window area can be set to For any pixel, if the grayscale value of the pixel is the largest within its eight neighborhoods, the pixel is taken as the maximum point, and all the maximum points in the window area are obtained.

[0159] In the third step, the information entropy of the distance between each maximum point and the pixel to be detected in the window area is calculated, and the fungal infection characteristic value of each pixel to be detected is obtained according to the minimum distance value, information entropy and deformation coefficient.

[0160] In this embodiment, the distances between each maximum point and the pixel to be detected within the window area are calculated, the distances are sorted, and then information entropy is calculated based on the sorted distances. The information entropy calculation process is prior art and falls outside the scope of this invention, so it will not be elaborated on here. A greater information entropy indicates a more diverse variation in the distances between the maximum points and the pixel to be detected in the image, and a more complex image structure.

[0161] As an example, the calculation formula for the fungal infection characteristic value of the jth pixel to be detected can be:

[0162] Where, represents the fungal infection characteristic value of the jth pixel to be detected, and th represents a hyperbolic function, which is used to normalize the data so that the value range of the fungal infection characteristic value is limited to between 0 and 1. Indicates the minimum distance value of the j-th pixel to be detected, Represents the deformation coefficient of the jth pixel to be detected, Represents the information entropy of the j-th pixel to be detected.

[0163] In the calculation formula of the fungal infection characteristic value, the minimum distance value The smaller the value, the closer the jth pixel to be detected is to the cell connectivity domain, the greater the possibility of infection, and the larger the fungal infection characteristic value of the jth pixel to be detected. The larger the value is, the greater the deformation loss of the cell connectivity domain corresponding to the j-th pixel to be detected is, and the greater the impact of fungal infection is; information entropy The larger the value is, the more complex the structural changes in the area around the j-th pixel to be detected are, the greater the difference in light changes is, and the more likely it is that the scattering is caused by uneven brightness, which is consistent with the uneven scattering of light in the corneal endothelium caused by fungal infection.

[0164] In the fourth step, threshold segmentation is performed based on the fungal infection feature value of each pixel to be detected in the microscopic image of the endothelium to obtain high-threshold pixels.

[0165] In this embodiment, the fungal infection characteristic value of each pixel to be detected in the microscopic image of the endothelium is subjected to Otsu threshold segmentation to obtain high-threshold pixel points, so as to filter out the pixel points to be detected with more serious fungal infection, and then a connected domain analysis is performed based on the filtered high-threshold pixel points.

[0166] In the fifth step, the high-threshold pixel point, the cell connected domain adjacent to the high-threshold pixel point, and the cell connected domain surrounded by the high-threshold pixel point are merged to obtain the undetermined lesion connected domain of the endothelial layer.

[0167] In this embodiment, the merging process refers to performing a connected domain analysis on the high-threshold pixel point, the cell connected domain adjacent to the high-threshold pixel point, and the cell connected domain surrounded by the high-threshold pixel point, and finally obtaining a connected domain, which is used as the connected domain of the pending lesion in the endothelial layer.

[0168] Among them, the cell connected domains adjacent to the high threshold pixel points and the cell connected domains surrounded by the high threshold pixel points are both areas infected by fungi, so they need to be merged.

[0169] So far, this embodiment has obtained the connected domain of the pending lesion in the endothelial layer.

[0170] S5, fuse the connected domains of the suspected lesions in the epithelium, stroma, and endothelium to outline the fungal lesion area corresponding to the corneal sample.

[0171] In this embodiment, the microscopic images of the epithelium, stroma, and endothelium are magnified and reduced in the same proportion to obtain adjusted microscopic images; the connected domains of the undetermined lesions in the three adjusted microscopic images are superimposed to obtain a superimposed connected domain of lesions, which is used as the fungal lesion area corresponding to the corneal sample and marked in the image. Among them, the example of the fungal lesion area corresponding to the corneal sample is shown in the figure below. Figure 10 shown.

[0172] So far, this embodiment has completed the accurate delineation of the fungal lesion area, that is, the fungal lesion area has been accurately identified.

[0173] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for delineating fungal lesions of fungal keratitis under a microscope, characterized in that: The following steps are involved: Obtain microscopic images of the epithelium, stroma, and endothelium at different depths of corneal samples; The influence range of fungal spores was analyzed based on the grayscale distribution characteristics of the microscopic image of the epithelium, and the connected domain of the suspected lesion in the epithelium was obtained; The interweaving of the highlight parts is analyzed according to the edge distribution characteristics of the microscopic image of the matrix layer, and the connected area of ​​the undetermined lesion in the matrix layer is obtained; The regularity and permeability of the cells were analyzed based on the transparency characteristics of the microscopic images of the endothelial layer, and the connected area of ​​the endothelial layer with unknown lesions was obtained; The connected domains of the suspected lesions in the epithelium, stroma, and endothelium were merged to outline the fungal lesion area corresponding to the corneal sample; The method of analyzing the influence range of the fungal spores according to the grayscale distribution characteristics of the microscopic image of the epithelial layer to obtain the connected domain of the undetermined lesion in the epithelial layer includes: Performing superpixel segmentation on the microscopic image of the epithelium to obtain various superpixel clusters; performing fitting analysis on each superpixel block in the superpixel cluster to obtain a fitting evaluation value of each superpixel cluster; Classify each superpixel cluster according to its grayscale value to obtain various superpixel clusters; the superpixel cluster with the highest grayscale is recorded as the spore cluster; Obtaining a connected domain of the epithelial lesion according to the fitting evaluation values ​​of each superpixel cluster and each spore cluster; The method of analyzing the interweaving of the highlight portion according to the edge distribution characteristics of the microscopic image of the matrix layer to obtain the connected region of the pending lesion in the matrix layer includes: Performing edge detection on the microscopic image of the matrix layer to obtain edge pixel points, and performing threshold segmentation on the microscopic image of the matrix layer to obtain high-threshold pixel points; The edge pixel points selected from all high-threshold pixel points are used as feature pixel points; Determining a first significance level of hyphae features of each characteristic line according to each characteristic pixel point; wherein the first significance level of hyphae features is used to characterize the straightness of the characteristic line; Determining the second significance level of hyphae characteristics of each characteristic line based on the interlaced characteristics between the characteristic lines; Combining the first and second significant degrees of the hyphae characteristics of each characteristic line, screening out the hyphae segments to be determined; The minimum connected domain containing all the hyphae segments to be determined in the microscopic image of the matrix layer is used as the connected domain of the lesion to be determined in the matrix layer; The analysis of the regularity and permeability of cells based on the transparency characteristics of the microscopic image of the endothelial layer to obtain the connected domain of the pending lesion in the endothelial layer includes: Acquiring a high-threshold region of the microscopic image of the endothelial layer, performing connected domain detection on the high-threshold region to obtain each cell connected domain; performing hexagonal fitting on each cell connected domain to obtain a loss value and a fitting side length; Determining the deformation coefficient of each of the cell-connected domains according to the loss value of the cell-connected domain and the difference between the fitted side length and the average side length; wherein the average side length is the average of the fitted side lengths of all the cell-connected domains; According to the distribution of pixel points outside the cell connected domain in the microscopic image of the endothelial layer and the deformation coefficient, the microscopic image of the endothelial layer is segmented to obtain the undetermined lesion connected domain of the endothelial layer.

2. The method for delineating fungal lesions of fungal keratitis under a microscope according to claim 1, characterized in that: The method of obtaining the connected domain of the epithelial lesion to be determined according to the fitting evaluation values ​​of each superpixel cluster and each spore cluster comprises: Obtain an enclosed area of ​​all the spore clusters in the microscopic image of the epithelial layer, and continuously adjust the area of ​​the enclosed area through an optimization algorithm until the area of ​​the enclosed area is maximized and the cumulative value of the fitting evaluation values ​​of all superpixel clusters in the enclosed area is minimized, and the final enclosed area is determined as the connected domain of the pending lesion in the epithelial layer; wherein the enclosed area is the area containing all the spore clusters.

3. The method for delineating fungal lesions of fungal keratitis under a microscope according to claim 1, characterized in that: The determining of the first significance level of the hyphae feature of each characteristic line according to each characteristic pixel point includes: Connect adjacent feature pixel points within the eight neighborhoods to form feature lines, and obtain several feature lines; For any characteristic line, obtain the length of the line connecting the two end points of the characteristic line, and compare the characteristic line length with the length of the line connecting to obtain the deviation; Draw a perpendicular line from each characteristic pixel point on the characteristic line to the line segment of the connecting length, with the direction of the perpendicular line being perpendicular to the line segment of the connecting length, and determine the sharpness based on the difference in the number of perpendicular lines on both sides of the line segment of the connecting length; Obtain edge chain codes of characteristic pixel points on the characteristic line, and obtain the tortuosity based on the difference between adjacent code numbers in the edge chain codes; Obtaining a first significance level of hyphae characteristics of the characteristic line according to the deviation, the sharpness, and the tortuosity of the characteristic line; Among them, the deviation, the sharpness, and the tortuosity are all negatively correlated with the first significant degree of the mycelium characteristics.

4. The method for delineating fungal lesions of fungal keratitis under a microscope according to claim 1, characterized in that: The step of determining the second significance level of the hyphae characteristics of each characteristic line based on the interlaced characteristics between the characteristic lines includes: For any feature line, the feature line closest to the feature line endpoint is recorded as the first reference line, and the feature line second closest to the feature line endpoint is recorded as the second reference line; Extending the characteristic line in the direction of the undetermined reference line and then connecting it with the undetermined reference line to obtain a new characteristic line; wherein the undetermined reference line is the first reference line or the second reference line; Obtain each inflection point on the pending reference line, and then obtain the angle formed by the line connecting the two end points of the pending reference line and any inflection point, and use the minimum angle as the target angle corresponding to the pending reference line; The second significance level of the hyphae feature of the feature line is determined according to the similarity degree of grayscale distribution between the feature line and the new feature line and the target angle corresponding to the first reference line and the second reference line of the feature line.

5. The method for delineating fungal lesions of fungal keratitis under a microscope according to claim 4, characterized in that: The determining of the second significance level of the hyphae feature of the feature line according to the degree of uniform similarity of the grayscale distribution between the feature line and the new feature line and the target angle corresponding to the first reference line and the second reference line of the feature line includes: Obtaining a first grayscale variance of the characteristic line, a second grayscale variance of the new characteristic line when the pending reference line is the first reference line, and a third grayscale variance of the new characteristic line when the pending reference line is the second reference line; determining a first similarity based on a difference between the first grayscale variance and the second grayscale variance, and determining a second similarity based on a difference between the first grayscale variance and the third grayscale variance; The second significance level of the hyphae feature of the characteristic line is determined according to the first similarity, the second similarity, and the target angle between the first reference line and the second reference line.

6. The method for delineating fungal lesions of fungal keratitis under a microscope according to claim 5, characterized in that: The determining, based on the first similarity, the second similarity, and a target angle between the first reference line and the second reference line, a second significance level of the hyphae feature of the feature line includes: Calculate the ratio of the second similarity to the average target angle corresponding to the second reference line, and calculate the ratio of the first similarity to the average target angle corresponding to the first reference line. Perform a difference calculation on the two ratios and then normalize them to determine the second significance level of the hyphae feature of the characteristic line.

7. The method for delineating fungal lesions of fungal keratitis under a microscope according to claim 1, characterized in that: The method of segmenting the microscopic image of the endothelial layer according to the distribution of pixel points outside the cell connected domain in the microscopic image of the endothelial layer and the deformation coefficient to obtain the undetermined lesion connected domain of the endothelial layer includes: Pixels outside the cell connected domain in the microscopic image of the endothelial layer are used as pixels to be detected, and the minimum distance value between the pixel to be detected and all cell connected domains is determined in the microscopic image of the endothelial layer; Determine a window area with the pixel to be detected as the center, and determine each maximum point in the window area; wherein the maximum point is the pixel point with the largest grayscale value within its eight neighborhoods; Performing information entropy calculation on the distance between each maximum point and the pixel to be detected in the window area, and obtaining the fungal infection characteristic value of each pixel to be detected according to the minimum distance value, the information entropy and the deformation coefficient; Threshold segmentation is performed based on the fungal infection characteristic value of each pixel to be detected in the microscopic image of the endothelial layer to obtain high-threshold pixel points; The high-threshold pixel point, the cell connected domain adjacent to the high-threshold pixel point, and the cell connected domain surrounded by the high-threshold pixel point are merged to obtain the undetermined lesion connected domain of the endothelial layer.

Images