Myopia grading method driven by coupling of corneal cell mechanics and topographic features

Abstract

The application provides a myopia grading method coupled by corneal cell mechanics and topographic features, belongs to the technical field of myopia ophthalmic disease diagnosis and development mechanism based on artificial intelligence, adopts atomic force microscope nanomechanical characterization technology, measures the surface topography and mechanical properties such as Young's modulus and adhesion of corneal cells of low, moderate and high myopia patients, takes the corneal cell surface topography image obtained by the atomic force microscope as the cell skeleton quantitative analysis data set, utilizes edge extraction, brightness detection and multi-scale image segmentation method, quantifies the overall cell skeleton distribution characteristics through the cell skeleton area average complexity and cell skeleton area average similarity, the method does not depend on staining, can more comprehensively evaluate the organization structure of the cell skeleton, is suitable for more diversified cell skeleton structures including complex clusters, and establishes a myopia grading model, thereby providing a more direct cell-level experimental basis for the myopia development mechanism research.

Description

Technical Field

[0001] This invention belongs to the technical field of diagnosis and development mechanism of myopia based on artificial intelligence, and specifically relates to a myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics. Background Technology

[0002] Myopia is a prevalent eye disease worldwide, especially in East Asia. The World Health Organization has identified it as a leading cause of visual impairment in individuals with refractive errors, making it a serious public health problem. Currently, the exact causes of myopia remain unclear. Research indicates that both genetic and environmental factors influence the development of myopia, with its main pathological signs being increased axial length and changes in corneal curvature. Previous studies on the development of myopia have primarily focused on scleral biomechanical properties and retinal-fundus regulatory mechanisms, with relatively less attention paid to the cornea, particularly lacking systematic correlation analysis methods between the degree of myopia and corneal cell biomechanical properties and cytoskeleton morphology. Although some studies have indicated the correlation between corneal biomechanical properties and myopia, and have conducted preliminary explorations of the biomechanical properties of corneal cells at different degrees of myopia, the following problems still exist: (1) First, in studies on the correlation between cellular mechanical properties and myopia, corneal cells are mainly derived from animal models, lacking quantitative assessment of the mechanical behavior of human corneal cells. Most current corneal mechanics research focuses on the tissue level, rather than the single-cell level. The few existing corneal cell mechanics studies are also based on animal models, with almost no research on the mechanics of human corneal cells. Although these animal corneas share some structural and biomechanical characteristics with humans, significant differences remain. The cytoskeleton structure, biomechanical response, and adaptability of corneal cells may vary among different species. Animal corneas are typically thicker and have a more compact tissue structure, while human corneas may exhibit different cellular remodeling mechanisms during myopia development. Studies relying solely on animal models cannot directly reflect the true changes in the mechanical properties of human corneal cells, affecting the accurate understanding of the mechanisms underlying myopia development.

[0003] (2) Secondly, existing research typically focuses on the analysis of single features, failing to comprehensively consider the coupling relationship between corneal cytoskeleton morphology and its mechanical properties and myopia. Current research on corneal-related myopia tends to focus on single-feature analysis, such as solely investigating changes in corneal morphology or studying the correlation between corneal mechanical properties and the severity of myopia, lacking in-depth exploration of the coupling relationship between mechanics and morphology. However, corneal cells may undergo cytoskeleton remodeling at different degrees of myopia, such as the rearrangement of actin fibers or the enhancement / weakening of the microtubule network. These changes may affect the cell's mechanical adaptability, thereby altering the overall biomechanical properties of the cornea. Therefore, single-feature analysis is insufficient to fully reveal the cellular biophysical mechanisms of myopia development.

[0004] (3) Finally, most current myopia grading models are based on fundus images and retinal morphology datasets, or focus on macroscopic corneal morphology and mechanical characteristics changes, neglecting the important influence of cellular-level microstructure and cellular mechanical properties on myopia grading. Current myopia grading model research is mainly based on fundus images and retinal morphology datasets obtained using techniques such as optical coherence tomography (OCT) images and fundus color photography, as well as corneal macroscopic morphology and tissue mechanical property datasets obtained through corneal topography. The accuracy of these grading models mainly depends on changes in fundus structure, or mostly focuses on structural deformation at the corneal tissue scale, without deeply exploring the influence of corneal cellular-level micromechanical properties on the overall mechanical behavior of the cornea.

[0005] To address the aforementioned research gaps and deficiencies, exploring a myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics has significant scientific value and application prospects. Summary of the Invention

[0006] This invention proposes a myopia grading method driven by the coupling of corneal cell mechanics and morphology. Compared to traditional myopia grading methods, this research delves into the cellular level, focusing on the morphology and mechanical properties of corneal cells, and quantitatively analyzing the mechanical behavior of corneal cells and the adaptive changes in cytoskeleton structure under different degrees of myopia. By comprehensively analyzing the mechanical properties and morphological characteristics of corneal cells, this study provides a new perspective for exploring the mechanisms of myopia development and provides more precise biophysical evidence for early myopia screening and personalized treatment strategies.

[0007] To solve the above problems, the present invention adopts the following technical solution: A myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics, the method comprising the following steps: S1: Simultaneous Imaging of Corneal Cell Morphology and Mechanical Properties at Different Degrees of Myopia: Mechanical properties and morphological images of corneal cells at different degrees of myopia were acquired using atomic force microscopy. These corneal cell surface morphology images were then used as a dataset for cytoskeleton quantification analysis. S2: Quantification of corneal cytoskeleton morphology: Using edge extraction, brightness detection and multi-scale image segmentation methods, the overall cytoskeleton distribution characteristics are quantified by two parameters: average complexity of cytoskeleton regions and average similarity of cytoskeleton regions. Based on the idea of ​​renormalization group, the cytoskeleton image is uniformly quantified and evaluated, and the self-similarity error is calculated to evaluate the average similarity of cytoskeleton regions. S3: Development of a myopia grading model based on corneal cell mechanics and cytoskeleton morphology parameters: The cell mechanics parameters used include Young's modulus and adhesion force, and the cytoskeleton morphology quantification parameters used include fractal dimension and self-similarity error, which respectively describe the structural complexity and structural similarity of the overall distribution of the cytoskeleton. Two artificial intelligence models, KNN and decision tree, are used to achieve myopia grading.

[0008] Furthermore, S1 specifically includes the following steps: S101: Corneal cell sample extraction: collect corneal stromal lenses and extract corneal cells from the corneal stromal lenses; S102: Corneal cell sample culture and preparation: The corneal cells extracted in S101 were washed, cultured, and then used for atomic force microscopy measurement experiments; S103: Initialization of the detection system: Place the corneal cells in a culture dish, fix the culture dish on the atomic force microscope, and adjust the position of the probe and positioning stage; S104: Construction of liquid phase detection environment: Add an appropriate amount of culture medium to the culture dish and immerse the probe below the liquid surface for stable measurement; S105: Coarse alignment of probe and sample detection point: initially locate the corneal cells, select a suitable measurement area, move the measurement area to the center of the field of view, and move the probe tip above the position to be measured to complete the coarse alignment; S106: Measurement sensitivity and probe stiffness settings: Further reduce the Z-axis distance between corneal cells and probe to 10~20 μm, calibrate the measurement sensitivity through thermal tuning, and use the nominal stiffness value for probe stiffness. S107: Single-point detection of corneal morphology and mechanical properties: Set the voltage reference value at the moment when the probe contacts the corneal cell and produces a certain static indentation, adjust the distance between the probe and the corneal cell, record the probe deformation voltage signal and the displacement value of the positioning stage during the entire process of the probe approaching, contacting and moving away from the corneal cell, and obtain the force-displacement curve by using the measurement sensitivity and probe stiffness. S108: Simultaneous imaging of corneal morphology and mechanical properties: Adjust the position of the corneal cells to create a relative lateral displacement between them and the probe. Move the probe tip to the next detection point and repeat step S107. The corresponding static indentation remains unchanged, i.e., the contact voltage reference value is set to be constant. Then, measure the sample surface characteristics of several points in the area to be detected sequentially along the grating scanning path to achieve simultaneous imaging of corneal morphology and mechanical properties.

[0009] Furthermore, in S101, the corneal cells are divided into three groups: low myopia group: -3.00 D < SE ≤ -0.50 D; moderate myopia group: -6.00 D < SE ≤ -3.00 D; high myopia group: SE ≤ -6.00 D.

[0010] Furthermore, in S102, the culture conditions for the corneal cells are as follows: maintaining a constant temperature of 37 °C in an incubator with 5% CO2 until the cells attach to the culture dish substrate, rinsing the cells with phosphate buffer to remove surface proteins and dead cells, adding fresh culture medium, and culturing in the incubator for 1 hour.

[0011] Furthermore, the curve data obtained through S107, combined with the Hertz contact mechanics model, is used to calculate the nanomechanical properties of the corneal cell surface, including but not limited to adhesion force and Young's modulus.

[0012] Furthermore, S2 specifically includes the following steps: S201: Image preprocessing: Convert the RGB color image of corneal cell surface morphology obtained by atomic force microscopy into a grayscale image, use the average brightness of each image as a threshold, set pixels below this threshold to 0, and leave other pixels unchanged. S202: Calculation of average complexity of cytoskeleton region: Using the image preprocessed in step S201 as the dataset, the cytoskeleton structure is extracted using the adaptive Canny edge detection algorithm to obtain the corresponding binary image; during the adaptive edge detection process, the image is denoised by Gaussian filtering, the image gradient matrix is ​​calculated, the optimal threshold is obtained through iterative optimization, and further refined using the Otsu method to achieve dynamic adjustment of the Canny edge detection threshold; the fractal dimension D is calculated using the Box-counting method to evaluate the cytoskeleton complexity; S203: Calculation of average similarity of cytoskeleton region: Using the image preprocessed in step S201 as the dataset, the image is divided into sub-blocks of different scales. Zero-pixel sub-blocks are removed to reduce the influence of uneven background distribution on the calculation results. The idea of ​​renormalization group is used to evaluate the uniformity of cytoskeleton image.

[0013] Furthermore, in S202, an average complexity parameter for the cytoskeleton region is set to evaluate the degree of disorder in the overall distribution of the cytoskeleton, and in S203, an average similarity parameter for the cytoskeleton region is set to evaluate the regularity and uniformity of the overall distribution of the cytoskeleton.

[0014] Furthermore, the quantitative parameters for corneal cell morphology characteristics include: the average complexity of the cytoskeleton region (average fractal dimension) and the average similarity of the cytoskeleton region (average self-similarity error).

[0015] Furthermore, in S3, 80% of all data is used as the training set and 20% as the test set. The main steps include: reading the data, extracting features and labels, standardizing the data, dividing the training set and the test set, using cross-validation to select the K value, training the KNN classification model, predicting the test set data, and calculating the classification evaluation index.

[0016] Furthermore, in S3, two sorting rules for the datasets were established. One is to sort according to the correspondence between mechanical properties and morphological quantization parameters, that is, the mechanical properties and morphological quantization parameters of the Nth data point correspond to the same corneal cell. The other is to shuffle the sorting, that is, the mechanical properties and morphological quantization parameters of the Nth data point may not correspond to the same corneal cell, but the myopia degree classification label is correct. Based on the different sorting rules of each data point in the two columns of feature variables, the KNN classification model was trained twice.

[0017] Advantages of this invention: 1. By integrating corneal cell mechanical properties and cytoskeleton morphological parameters, this invention establishes a myopia grading model based on changes in multiple cell characteristics. Compared with traditional grading methods that rely on the macroscopic mechanical properties of the cornea, this model can more precisely analyze the corneal mechanical behavior and morphological changes at the cellular level under different degrees of myopia, improving the accuracy and reliability of myopia grading and providing more scientific evidence for basic research on the mechanisms of myopia development.

[0018] 2. This invention uses atomic force microscopy to measure key parameters such as Young's modulus and adhesion of human corneal cells, and combines this with cytoskeleton morphological characteristics to provide more in-depth cellular-level evidence for the biophysical mechanisms of myopia development, filling the gap in research on the differences in the mechanical properties of corneal cells with different degrees of myopia in humans.

[0019] 3. This invention uses atomic force microscopy combined with image processing algorithms to achieve a comprehensive assessment of the cytoskeleton without staining by using parameters such as the complexity and similarity of the cytoskeleton region. This avoids interference with the original state of living cells during the staining process and improves the accuracy of quantitative assessment results of cytoskeleton morphology for different degrees of myopia.

[0020] 4. This invention proposes a quantitative method for cytoskeleton uniformity based on the concept of renormalization groups. It employs techniques such as multi-scale segmentation and self-similarity error calculation to accurately assess the regularity of cytoskeleton distribution, providing a new quantitative analysis method for corneal cytoskeleton remodeling during myopia progression.

[0021] 5. The myopia grading model of the present invention can be applied to individualized myopia development prediction and intervention. Combined with artificial intelligence algorithms, it can provide more accurate myopia progression assessment based on the patient's cellular biomechanical characteristics, and can provide a scientific basis for myopia correction and prevention strategies based on cellular biomechanics in the future.

[0022] In summary, this invention has significant scientific value and application prospects in areas such as precise myopia grading based on multiple corneal cell types, research on the mechanical differences of corneal cells in myopic patients, quantitative assessment of the structure of unstained protocellular skeletons, assessment of the overall distribution similarity of the cytoskeleton, and personalized intervention based on myopia grading models. It provides new technical means for basic medical research and clinical diagnosis and treatment of myopia. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the structure of an atomic force microscope corneal cell mechanics and morphology simultaneous imaging characterization system; Figure 2 This is a schematic diagram illustrating the construction process of the environment for detecting nanomechanical properties using liquid phase atomic force microscopy. Figure 3 This is a schematic diagram of the force-displacement curve; Figure 4 This is a statistical chart of Young's modulus data for corneal cells at different degrees of myopia; Figure 5 This is a statistical chart of corneal cell adhesion data for different degrees of myopia; Figure 6 This is a schematic diagram illustrating the method for calculating the average complexity of the cytoskeleton region. Figure 7 This is a schematic diagram illustrating the method for calculating the average similarity of cytoskeleton regions. Figure 8 This is a statistical chart of the average complexity data of the cytoskeleton region; Figure 9 This is a statistical graph of the average self-similarity error data of the cytoskeleton region; Figure 10 This is the classification result of the KNN model when the mechanical and morphological data do not correspond; Figure 11 It is the classification result of the KNN model when mechanical and morphological data correspond; Figure 12 This is the classification result of the decision tree model when the mechanical and topographic data do not correspond; Figure 13 The classification result of the decision tree model when mechanical and topographic data correspond. Among them, 1: XYZ micron positioning stage; 2: XY nanon positioning stage; 3: temperature-controlled sample holder; 4: culture dish; 5: corneal cells; 6: probe; 7: probe holder; 8: Z nanon positioning stage; 9: laser; 10: optical microscope; 11: displacement sensor PSD; 12: exciter; 13: controller. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] This invention employs atomic force microscopy (AFM) to characterize the surface morphology and mechanical properties of corneal cells from patients with low, moderate, and high myopia in a natural cell culture environment. Unlike previous studies that primarily relied on animal models for studying the mechanical behavior of corneal cells, this invention uniquely focuses on human corneal cells, providing more direct cellular-level experimental evidence for the study of the mechanisms of myopia development. Using corneal cell surface morphology images acquired by AFM as a dataset, edge extraction, brightness detection, and multi-scale image segmentation methods are employed to quantify the overall cytoskeleton distribution characteristics using two parameters: average complexity and average similarity of the cytoskeleton region.

[0027] Unlike traditional methods, this approach is stain-free, allowing for a more comprehensive assessment of the cytoskeleton's organizational structure. Furthermore, by quantifying the overall morphological arrangement of the natural cytoskeleton across multiple dimensions, it is applicable to a wider variety of cytoskeleton structures, including complex clusters. Based on cytoskeleton morphological parameters and cellular mechanical properties of corneal cells at different levels of myopia, the correlation between cytoskeleton morphology and mechanical changes during the development of myopia is analyzed. Combined with artificial intelligence methods, a myopia grading model is established based on the coupling relationship between morphology and mechanics.

[0028] In summary, this invention has achieved breakthroughs in experimental subject selection, data analysis methods, and myopia grading model datasets, providing a new perspective for understanding the biophysical mechanisms of myopia. This invention overcomes the limitations of traditional myopia grading methods, which primarily rely on macroscopic tissue mechanical properties or fundus images while neglecting cellular-level biomechanical characteristics. It employs atomic force microscopy combined with cytoskeleton morphology analysis, achieving quantitative analysis of the mechanical properties and skeletal structure of corneal cells at different degrees of myopia through the coupling of cellular mechanics and morphological features. Furthermore, by incorporating artificial intelligence methods, a myopia grading model is established, providing more precise basic medical research evidence for studying the mechanisms of myopia development.

[0029] Example 1 This invention proposes a myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics, specifically including the following steps: S1: Simultaneous imaging of corneal cell morphology and mechanical properties at different degrees of myopia.

[0030] S101: Extraction of corneal cell samples.

[0031] Corneal stromal microlenses were collected, and corneal cells were extracted from the corneal stromal microlenses. The corneal cells were divided into three groups: low myopia group: -3.00 D < SE ≤ -0.50 D; moderate myopia group: -6.00 D < SE ≤ -3.00 D; high myopia group: SE ≤ -6.00 D.

[0032] S102: Corneal cell sample culture and preparation.

[0033] After rinsing and culturing the corneal cells extracted from S101, they were used for atomic force microscopy measurements. The culture conditions were as follows: the cells were kept at a constant temperature of 37 °C in an incubator with 5% CO2 until they adhered to the substrate of the culture dish. The cells were then rinsed with phosphate buffer to remove surface proteins and dead cells, and fresh culture medium was added. The cells were then cultured in the incubator for 1 hour.

[0034] S103: Detection system initialization.

[0035] The corneal cells were placed in a culture dish, which was then fixed on the atomic force microscope, and the positions of the probe and positioning stage were adjusted.

[0036] S104: Construction of liquid phase detection environment.

[0037] Add an appropriate amount of culture medium to the culture dish and immerse the probe below the surface of the liquid for stable measurement.

[0038] S105: Coarse alignment of the probe with the sample detection point.

[0039] The corneal cells are initially located, a suitable measurement area is selected, and the measurement area is moved to the center of the field of view. The probe tip is moved above the position to be measured to complete the coarse alignment.

[0040] S106: Measurement sensitivity and probe stiffness settings.

[0041] The Z-axis distance between corneal cells and the probe was further reduced to 10-20 μm, and the measurement sensitivity was calibrated by thermal tuning. The probe stiffness was set to the nominal stiffness value.

[0042] S107: Single-point detection of corneal morphology and mechanical properties.

[0043] A voltage reference value is set at the moment when the probe contacts the corneal cell and generates a certain static indentation. The distance between the probe and the corneal cell is adjusted, and the probe deformation voltage signal and the displacement value of the positioning stage are recorded throughout the entire process of the probe approaching, contacting, and moving away from the corneal cell. The force-displacement curve is obtained by measuring the sensitivity and probe stiffness. By using the obtained curve data and combining the Hertz contact mechanics model, the nanomechanical properties of the corneal cell surface are calculated, including but not limited to adhesion force and Young's modulus.

[0044] S108: Simultaneous imaging of corneal morphology and mechanical properties.

[0045] The position of the corneal cells is adjusted to create a relative lateral displacement between them and the probe. The probe tip moves to the next detection point. Step S107 is repeated, and the corresponding static indentation remains unchanged, i.e., the contact voltage reference value is set to be constant. Subsequently, the sample surface characteristics of several points in the area to be detected are measured sequentially along the grating scanning path to achieve simultaneous imaging of corneal morphology and mechanical properties.

[0046] S2: Quantification of corneal cell cytoskeleton morphology.

[0047] S201: Image preprocessing.

[0048] The RGB color images of corneal cell surface morphology obtained by atomic force microscopy are converted into grayscale images. The average brightness of each image is used as a threshold, and pixels below this threshold are set to 0, while other pixels remain unchanged.

[0049] S202: Calculation of average complexity of cytoskeleton region.

[0050] An average complexity parameter for the cytoskeleton region is set to evaluate the disorder of the overall cytoskeleton distribution. Using the image preprocessed in step S201 as the dataset, the cytoskeleton structure is extracted using the adaptive Canny edge detection algorithm to obtain the corresponding binary image. During the adaptive edge detection process, Gaussian filtering is applied to the image for noise reduction, the image gradient matrix is ​​calculated, and the optimal threshold is obtained through iterative optimization. The Otsu method is used for further refinement to achieve dynamic adjustment of the Canny edge detection threshold. The fractal dimension D is calculated using the Box-counting method to evaluate the cytoskeleton complexity.

[0051] S203: Calculation of average similarity of cytoskeleton regions.

[0052] A cytoskeleton region average similarity parameter is set to evaluate the regularity and uniformity of the overall distribution of the cytoskeleton. Using the image preprocessed in step S201 as the dataset, the image is divided into sub-blocks of different scales. Zero-pixel sub-blocks are removed to reduce the influence of uneven background distribution on the calculation results. The idea of ​​renormalization group is used to evaluate the uniformity of the cytoskeleton image.

[0053] S3: Development of a myopia grading model based on corneal cell mechanics and skeletal morphology parameters.

[0054] All data is used as 80% of the training set and 20% as the test set. The main steps include: reading the data, extracting features and labels, standardizing the data, dividing the training and test sets, selecting the K value using cross-validation, training the KNN classification model, predicting the test set data, and calculating the classification evaluation index.

[0055] Two sorting rules were established for the datasets. One rule was to sort the datasets according to the correspondence between mechanical properties and morphological quantization parameters, meaning that the mechanical properties and morphological quantization parameters of the Nth data point correspond to the same corneal cell. The other rule was to shuffle the datasets, meaning that the mechanical properties and morphological quantization parameters of the Nth data point do not necessarily correspond to the same corneal cell, but the myopia degree classification label is correct. Based on the different sorting rules for each data point in the two columns of feature variables, the KNN classification model was trained twice.

[0056] Example 2 As a preferred embodiment of the present invention, such as Figure 1 As shown, the atomic force microscope corneal cell mechanical and morphological synchronous imaging characterization system used in this embodiment includes an XYZ micro-stage (1), an XY nano-stage (2) and a temperature-controlled sample holder (3) installed sequentially on the XYZ micro-stage (1), corneal cells (5) attached to the bottom of the culture dish (4) and installed on the temperature-controlled sample holder (3), a probe (6) installed on a probe holder (7), and the probe holder (7) installed on the Z nano-stage (8). The laser force measurement system consists of a laser (9) and a displacement sensor PSD (11). In addition, an optical microscope (10) forms a top-down field of view to assist in positioning.

[0057] During the detection process, the detection location point is coarsely located using the XYZ micron positioning stage (1), and the relative position between the probe (6) and the corneal cell (5) is adjusted. After coarse positioning, a voltage reference value is set, and a driving voltage is generated by the exciter (12) to cause the distance between the probe (6) and the corneal cell (5) to change periodically, i.e., a periodic approach-contact-distance process; combined with the Z nanon positioning stage (8), the probe (6) is controlled to contact the corneal cell (5), and the peak contact force between the probe (6) and the corneal cell (5) is kept constant during this period, i.e., the peak voltage of the displacement sensor PSD (11) during the periodic contact is maintained at the set reference value; based on the change in peak voltage, the current displacement compensation error is obtained, and the Z nanon positioning stage (8) is driven by the controller (13) to achieve displacement compensation, and the magnitude of the compensated displacement is the morphological change value. During the periodic approach-contact-distance process, the probe cantilever deformation signal and displacement change are obtained, and the force-displacement curve can be obtained by combining the system measurement sensitivity and probe stiffness; according to a suitable contact mechanics model, the corresponding corneal cell mechanical properties can be calculated.

[0058] use Figure 1 The atomic force microscope shown acquires images of the mechanical properties and morphology of corneal cells with different degrees of myopia. Subsequently, based on the cytoskeleton features reflected in the corneal cell surface morphology images, cytoskeleton morphological parameters are quantified. The quantification parameters used include fractal dimension and self-similarity error, which respectively describe the structural complexity and structural similarity of the overall distribution of the cytoskeleton. Among them, the uniformity quantification evaluation of the cytoskeleton images based on the idea of ​​renormalization group and the calculation of self-similarity error is a unique feature that distinguishes it from previous quantification methods.

[0059] This yields a dataset of cellular mechanical properties: surface adhesion force and Young's modulus of corneal cells at different degrees of myopia, and a dataset of cytoskeleton morphological parameters: fractal dimension and self-similarity error of corneal cell skeletons at different degrees of myopia. Based on these two types of parameters, training and testing datasets for a myopia grading model are established. Two classic classification models, KNN and decision trees, are applied to the myopia grading problem, classifying high, moderate, and low myopia. The model training effect is evaluated using metrics such as accuracy, precision, recall, and F1 score.

[0060] This embodiment specifically includes the following steps: S1: Simultaneous imaging of corneal cell morphology and mechanical properties at different degrees of myopia.

[0061] S101: Extraction of corneal cell samples.

[0062] In this embodiment, corneal stromal microlenses were collected from patients who underwent SMILE refractive surgery to correct myopia at the Refractive Surgery Center of Tianjin Eye Hospital, and corneal cells were extracted from these corneal stromal microlenses. To minimize the influence of confounding factors such as age, gender, and myopia progression rate, the test population was selected from young people around 30 years old, with an equal male-to-female ratio. The collected corneal cells were divided into three groups according to the degree of myopia: low myopia group: -3.00 D < SE ≤ -0.50 D; moderate myopia group: -6.00 D < SE ≤ -3.00 D; high myopia group: SE ≤ -6.00 D.

[0063] It should be noted that this study was approved by the Ethics Committee of Tianjin Eye Hospital, with ethics approval certificate number: KY-2024058; all patients were informed of the risks of the surgery and the benefits of the study and signed informed consent forms; all surgeries were performed in accordance with the principles of the Declaration of Helsinki.

[0064] S102: Corneal cell sample culture and preparation.

[0065] The extracted corneal cells were cultured in modified Eagle medium supplemented with 10% fetal bovine serum and kept at a constant temperature of 37 °C in an incubator with 5% CO2 until the cells adhered to the culture dish substrate. The cells were then rinsed with phosphate buffer to remove surface proteins and dead cells, and fresh medium was added. After culturing for another hour in the incubator, the cells were ready for atomic force microscopy measurements.

[0066] S103: Detection system initialization.

[0067] Prepared corneal cells (5) are attached to the bottom of a culture dish (4). The culture dish (4) is fixed to a temperature-controlled sample holder (3). The probe (6) is installed on a probe holder (7). The temperature-controlled sample holder (3) and the probe holder (7) are then installed on the XY nanostage (2) and the Z nanostage (8), respectively. The pre-set wiring terminals inside the probe holder (7) can then be electrically connected to the corresponding data acquisition device.

[0068] S104: Construction of liquid phase detection environment.

[0069] like Figure 2As shown, corneal cells (5) are attached to the bottom of the culture dish (4). At this time, there is an appropriate amount of culture medium in the culture dish (4). After the probe (6) is installed on the probe holder (7), 2-3 drops of culture medium, about 100-150 μL, are carefully added to the front end of the probe holder (7) with a pipette so that the droplets cover the probe (6). Then, the XYZ micron positioning stage (1) is moved so that the probe (6) gradually approaches the liquid surface in the culture dish along the Z direction, forming a liquid bridge at the initial contact. Then, the probe (6) is immersed in the liquid surface for stable measurement.

[0070] S105: Coarse alignment of the probe with the sample detection point.

[0071] Move the XYZ micro-positioning stage (1), preliminarily locate the corneal cells (5) using the optical microscope (10), select a suitable measurement area, and move the area to the center of the field of view of the optical microscope (10). Use the XYZ micro-positioning stage (1) to place the tip of the probe (6) above the position to be measured, complete the coarse alignment, and adjust the laser spot of the probe (6) to the center of the front end of the probe microcantilever.

[0072] S106: Measurement sensitivity and probe stiffness settings.

[0073] The XYZ micrometer positioning stage (1) is moved to further reduce the Z-direction distance between the corneal cells (5) and the probe (6) to 10~20 μm. The measurement sensitivity is calibrated by thermal tuning, and the probe stiffness is based on the nominal stiffness value.

[0074] S107: Single-point detection of corneal morphology and mechanical properties.

[0075] A voltage reference value is set at the moment when the probe (6) contacts the corneal cell (5) and generates a certain static indentation. The Z-nano positioning stage (8) is driven to descend along the Z-axis until the deformation voltage signal of the probe (6) reaches the set value. The displacement value of the Z-nano positioning stage (8) at this moment is recorded as the morphological height value of that point. Then, the Z-nano positioning stage (8) is driven to rise along the Z-axis to the initial point. The deformation voltage signal of the probe (6) and the displacement value of the Z-nano positioning stage (8) are recorded throughout the process of the probe (6) approaching, contacting, and moving away from the corneal cell (5). The force-displacement curve is obtained by measuring the sensitivity and probe stiffness, as shown in the figure. Figure 3 As shown.

[0076] By combining the curve data with the Hertz contact mechanics model, the nanomechanical properties of the corneal cell surface, including adhesion force and Young's modulus, were calculated.

[0077] S108: Simultaneous imaging of corneal morphology and mechanical properties.

[0078] Move the XY nanostage (2) to generate a relative lateral displacement between the corneal cells (5) and the probe (6). Move the tip of the probe (6) to the next detection point and repeat step S107. The corresponding static indentation remains unchanged, that is, the contact voltage reference value is set to be constant. Then, measure the sample surface characteristics of several points in the area to be detected along the grating scanning path in sequence to realize the synchronous imaging of corneal morphology and mechanical properties.

[0079] like Figure 4 and Figure 5 As shown, Figure 4 (a) Figure 5 (a) are the statistical histograms of the Gaussian fit. Figure 4 (b) Figure 5 (b) are scatter plots. The atomic force microscopy results include: surface morphology of corneal cells, surface Young's modulus map and surface adhesion force image of patients with low, moderate and high myopia.

[0080] S2: Quantification of corneal cell cytoskeleton morphology.

[0081] S201: Image preprocessing.

[0082] The RGB color images of corneal cell surface morphology acquired by atomic force microscopy were converted into grayscale images with pixel values ​​ranging from 0 to 255. The average brightness of each image was used as a threshold; pixels below this threshold were set to 0, while other pixels remained unchanged.

[0083] S202: Calculation of average complexity of cytoskeleton region.

[0084] A parameter for the average complexity of cytoskeleton regions is set to evaluate the degree of disorder in the overall distribution of the cytoskeleton.

[0085] like Figure 6 As shown in (a), Figure 6 (a) is the flowchart of adaptive edge detection. First, using the preprocessed image of S201 as the dataset, the cytoskeleton structure is extracted using the adaptive Canny edge detection algorithm to obtain the corresponding binary image. In the adaptive edge detection process, Gaussian filtering is applied to the image for noise reduction, the image gradient matrix is ​​calculated, and the optimal threshold is obtained through iterative optimization. The Otsu method is used for further refinement to achieve dynamic adjustment of the Canny edge detection threshold. Then, the fractal dimension is calculated using the Box-counting method. D To assess the complexity of the cytoskeleton.

[0086] like Figure 6 As shown in (b), Figure 6 (b) is a schematic diagram of the Box-counting method, where the analyzed image is divided by a side length of... rThe box covers the area, depending on the size. r The change calculation requires the minimum number of boxes to cover the entire image. N r And plotted in a double logarithmic coordinate system r With minimum number of boxes N r The slope value obtained by fitting the relationship curve using the least squares method is the fractal dimension. D In a double logarithmic coordinate system r With minimum number of boxes N r The relationship formula is shown below:

[0087] Average complexity of cytoskeleton regions RAC Represented as fractal dimension D Number of pixels with brightness greater than zero S The ratio, i.e. RAC = D / S .

[0088] This embodiment uses a size of 2 1 ~2 8 Eight bounding boxes of varying side lengths are used to scan a binary image set of the cytoskeleton, and the average complexity parameter of the cytoskeleton region is calculated.

[0089] S103: Calculation of average similarity of cytoskeleton regions.

[0090] A parameter for the average similarity of cytoskeleton regions is set to evaluate the regularity and uniformity of the overall distribution of the cytoskeleton.

[0091] like Figure 7 As shown, Figure 7 (A) is a schematic diagram of multi-scale sub-block division. (a) is the original image. (b) to (d) show the image divided into 4, 16, and 64 sub-blocks, respectively. Figure 7 (B) illustrates the processing steps of three computational methods. The renormalization group concept is used to evaluate the uniformity of the cytoskeleton image. Using the image preprocessed in step S201 as the dataset, the image is divided into sub-blocks of different scales. Zero-pixel sub-blocks are removed to reduce the impact of uneven background distribution on the computational results. Three methods are employed: simultaneously fixing two sub-blocks, rotating one sub-block, and rotating a single sub-block, to calculate the self-similarity error of the image pixel values. MSE The calculation formula is as follows:

[0092] in, B 1 and B 2 represents any two sub-blocks, where N The number of pixels for each sub-block;B 1,i and B 2,i These represent the first and second sub-blocks respectively. i Each pixel value.

[0093] Calculated self-similarity error MSE It is negatively correlated with similarity, and the smaller the value, the better. MSE This indicates that the similarity between sub-blocks is high, while larger ones... MSE This indicates low similarity between sub-blocks; the sum of self-similarity error values ​​between sub-blocks is calculated. MSE sum Compared with the total number of comparisons T The ratio between them is the average similarity of the cytoskeleton regions. RASE ( RASE = MSE sum / T In this embodiment, each image is divided into three different numbers of sub-blocks: 4, 16, and 64. Sub-blocks with more than 80% of non-zero pixels are selected, and the self-similarity error of the cytoskeleton region is calculated.

[0094] like Figure 8 and Figure 9 As shown, the quantitative parameters of corneal cell morphology characteristics include: average complexity of the cytoskeleton region (average fractal dimension) and average similarity of the cytoskeleton region (average self-similarity error).

[0095] S3: Development of a myopia grading model based on corneal cell mechanics and skeletal morphology parameters.

[0096] For a sample point to be classified, calculate its Euclidean distance to all samples in the training set.

[0097] The K nearest training samples to the sample to be classified are selected as references. According to the majority voting principle, the category that appears most frequently among the K nearest training samples is taken as the predicted category of the sample to be classified. Cross-validation is used to select the optimal K value to avoid the classification effect and overfitting problem caused by K being too large or too small. The Z-score normalization method is used to ensure the balance of feature influence and avoid the large impact of different feature value ranges on the distance calculation results. That is, normalize the data so that different features are in a similar numerical range.

[0098] The objective of this embodiment is to classify myopia severity using KNN based on the mechanical properties of corneal cells and skeletal morphology parameters, resulting in three categories: high myopia, moderate myopia, and low myopia. Specifically, the data includes two features: Young's modulus / adhesive force, and fractal dimension / self-similarity (the inverse of the self-similarity error). 80% of the data is used as the training set, and 20% as the test set. The classification model is evaluated using the following metrics: accuracy, precision, recall, and F1 score. The implementation is as follows: The main steps include: reading data, extracting features and labels, standardizing data, dividing the training set and the test set, selecting the K value using cross-validation, training the KNN classification model, predicting the test set data, and calculating the classification evaluation index.

[0099] First, the dataset of corneal cell mechanical properties and skeletal morphological parameters was read and loaded as a data frame. Then, feature variables were extracted from the dataset, including the Young's modulus / adhesion ratio and the average fractal dimension / self-similarity (the reciprocal of the average self-similarity error), along with the corresponding myopia degree label. Since the numerical ranges of different features may vary significantly, the feature data were Z-score standardized to a mean of 0 and a standard deviation of 1 to avoid certain features dominating distance calculations. Next, the dataset was divided into an 80% training set and a 20% test set to facilitate... The model can learn data patterns and be evaluated on independent test sets. To select the optimal K value, 5-fold cross-validation is used to test different K values, and the K value with the best classification performance is selected. Subsequently, the KNN classification model is trained using the training set data to learn the feature distributions corresponding to different degrees of myopia. After training, the test set data is input into the model for prediction to obtain the myopia classification result for each sample. Finally, the classification performance of the KNN model is evaluated by calculating classification evaluation metrics such as accuracy, precision, recall, and F1 score, and its performance on different degrees of myopia is analyzed.

[0100] Since the mechanical properties and morphological features of corneal cells obtained using atomic force microscopy are in a one-to-one correspondence, two dataset sorting rules were established. One rule is to sort according to the correspondence between mechanical properties and morphological quantification parameters, meaning the mechanical properties and morphological quantification parameters of the Nth data point correspond to the same corneal cell. The other rule is to shuffle the sorting, meaning the mechanical properties and morphological quantification parameters of the Nth data point may not necessarily correspond to the same corneal cell, but the myopia degree classification label is correct. Based on the different sorting rules for each data point in the two columns of feature variables, two KNN classification model training operations were performed, such as... Figures 10-13 As shown, the model training results are described below: (1) When the two columns of feature variable data correspond, the classification model training results are analyzed as follows: The classification results obtained by the KNN class model are: accuracy 86.67%, precision 87.69%, recall 86.67%, and F1 score 86.57%.

[0101] The above results indicate that the accuracy of 86.67% means that KNN can correctly classify the degree of myopia in 86.67% of cases; the precision of 87.69% means that most of the samples classified as a certain degree of myopia do indeed belong to that category; the recall of 86.67% means that most of the true class samples are correctly classified; and the F1 score of 86.57% indicates that the model has found a good trade-off between precision and recall.

[0102] According to the results shown in the confusion matrix, the values ​​on the diagonal are the correct classification results, indicating a good classification effect; the misclassifications include 2 moderate myopia cases that were misclassified as high myopia, and 2 low myopia cases that were misclassified as moderate myopia.

[0103] (2) When the two columns of feature variable data do not correspond, the classification model training results are analyzed as follows: The classification results obtained by the KNN class model are: accuracy 96.67%, precision 96.94%, recall 96.67%, F1 score 96.64%. The physical meaning of the four index results is the same as above. According to the results shown by the confusion matrix, the values ​​on the diagonal are the correct classification results, and the classification effect is good. The misclassified cases include one moderate myopia that was misclassified as high myopia.

[0104] In this classification task, in addition to using KNN to classify the degree of myopia, a decision tree model was also trained for comparative analysis: First, similar to KNN, the data is read and preprocessed, including extracting Young's modulus / adhesion ratio and fractal dimension / self-similarity as features, and myopia degree as a label. Then, the data is Z-score standardized and divided into training and test sets. For KNN, cross-validation is used to select the optimal K value before training, while for decision trees, information gain and other features are used as the feature splitting criteria, and pruning strategies are used to prevent overfitting. After training, the decision tree model is used to predict the test set data, and classification evaluation metrics such as accuracy, precision, recall, and F1 score are calculated.

[0105] Finally, by comparing the classification results of the KNN and decision tree models, we analyzed their performance in classifying different degrees of myopia, and explored their advantages, disadvantages, and applicable scenarios. The specific implementation was as follows: Using a decision tree supervised learning algorithm, data is progressively divided through a series of decision rules to ultimately determine the category of each sample: First, an optimal feature is selected as the root node, and then the data is divided based on different values ​​of this feature; information gain is used to measure the reduction in uncertainty of the data before and after the division, and the greater the information gain, the greater the contribution of the feature to the classification; the optimal feature is selected as the classification basis. In this embodiment, the mechanical properties of corneal cells and the quantification parameters of skeletal morphology at different degrees of myopia are selected as feature variables, and the dataset is divided according to the different values ​​of these two features to form sub-nodes; feature selection and data division are repeated for each sub-node until the stopping condition is met (such as all samples belonging to the same category or the tree reaching its maximum depth).

[0106] In this embodiment, the optimal maximum tree depth and minimum number of sample splits are selected through cross-validation to avoid overfitting. The stopping conditions of the decision tree are: 1) all samples belong to the same category and no further splitting is needed; 2) the features of the current node have been used up and further splitting is not possible; 3) the preset constraints such as tree depth and minimum number of samples are met to prevent overfitting.

[0107] Subsequently, a decision tree classification model was trained using the training set data, enabling it to learn feature partitioning rules corresponding to different degrees of myopia. After training, the test dataset was input into the model for classification. When a new sample was input into the decision tree, starting from the root node, the model traversed downwards along the branches based on the sample's feature values ​​until a leaf node was reached. The category of the leaf node was the final classification result. Classification evaluation metrics, including accuracy, precision, recall, and F1 score, were calculated to evaluate the classification performance of the decision tree in this task and to analyze the model's performance in classifying different degrees of myopia. Figures 10-13 As shown, the training results of the decision tree classification model are described below: (1) Analysis of the classification model training results when the two columns of feature variable data do not correspond: The classification results obtained by the decision tree model are: accuracy 96.67%, precision 96.94%, recall 96.67%, and F1 score 96.64%. The above indicators show that the accuracy of 96.67% means that the decision tree can correctly classify the degree of myopia in 96.67% of cases; the precision of 96.94% means that most of the samples classified as a certain degree of myopia do indeed belong to that category; the recall of 96.67% means that most of the true category samples are correctly classified; and the F1 score of 96.64% shows that the model has found a good balance between precision and recall. In addition, according to the results shown in the confusion matrix, the values ​​on the diagonal are the correct classification results, and the classification effect is good. The misclassified cases include one moderate myopia that was misclassified as high myopia.

[0108] (2) When the two columns of feature variable data correspond, the classification model training results are analyzed as follows: The classification results obtained by the decision tree model are: accuracy 90%, precision 90.94%, recall 90%, F1 score 89.97%, and the physical meaning of the four indicators is the same as above; According to the results shown by the confusion matrix, the values ​​on the diagonal are the correct classification results, and the classification effect is good. The misclassifications include 1 moderate myopia that is misclassified as high myopia and 2 low myopia that are misclassified as moderate myopia.

[0109] This method was used to classify corneal cells of different degrees of myopia; Figure 10 and Figure 11 The classification results of the KNN model are shown for cases where the mechanical and morphological data correspond and cases where they do not. (A) shows the results of the four classification evaluation indicators: accuracy, precision, recall, and F1 score; (B) shows the confusion matrix corresponding to (A), which shows the results of correct and misclassification in the test data; (C) shows the ROC curve. Figure 12 and Figure 13 The classification results of the decision tree model are shown when the mechanical and morphological data correspond and when they do not correspond. (A) shows the results of the four classification evaluation indicators: accuracy, precision, recall and F1 score; (B) is the confusion matrix corresponding to (A), which shows the results of correct classification and misclassification in the test data; (C) is the ROC curve.

[0110] The present invention has been described in detail above through embodiments, but the content is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics, characterized in that: The method includes the following steps: S1: Simultaneous Imaging of Corneal Cell Morphology and Mechanical Properties at Different Degrees of Myopia: Mechanical properties and morphological images of corneal cells at different degrees of myopia were acquired using atomic force microscopy. These corneal cell surface morphology images were then used as a dataset for cytoskeleton quantification analysis. S2: Quantification of corneal cytoskeleton morphology: Using edge extraction, brightness detection and multi-scale image segmentation methods, the overall cytoskeleton distribution characteristics are quantified by two parameters: average complexity of cytoskeleton regions and average similarity of cytoskeleton regions. Based on the idea of ​​renormalization group, the cytoskeleton image is uniformly quantified and evaluated, and the self-similarity error is calculated to evaluate the average similarity of cytoskeleton regions. S3: Development of a myopia grading model based on corneal cell mechanics and cytoskeleton morphology parameters: The cell mechanics parameters used include Young's modulus and adhesion force, and the cytoskeleton morphology quantification parameters used include fractal dimension and self-similarity error, which respectively describe the structural complexity and structural similarity of the overall distribution of the cytoskeleton. Two artificial intelligence models, KNN and decision tree, are used to achieve myopia grading. Two sorting rules for the datasets are formulated: one is to sort according to the correspondence between mechanical properties and morphology quantification parameters, that is, the mechanical properties and morphology quantification parameters of the Nth data point correspond to the same corneal cell; the other is to shuffle the sorting, that is, the mechanical properties and morphology quantification parameters of the Nth data point do not necessarily correspond to the same corneal cell, but the myopia degree classification label is correct. Based on the different sorting rules of each data point in the two columns of feature variables, the KNN classification model is trained twice.

2. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 1, characterized in that: S1 specifically includes the following steps: S101: Corneal cell sample extraction: collect corneal stromal lenses and extract corneal cells from the corneal stromal lenses; S102: Corneal cell sample culture and preparation: The corneal cells extracted in S101 were washed, cultured, and then used for atomic force microscopy measurement experiments; S103: Initialization of the detection system: Place the corneal cells in a culture dish, fix the culture dish on the atomic force microscope, and adjust the position of the probe and positioning stage; S104: Construction of liquid phase detection environment: Add an appropriate amount of culture medium to the culture dish and immerse the probe below the liquid surface for stable measurement; S105: Coarse alignment of probe and sample detection point: initially locate the corneal cells, select a suitable measurement area, move the measurement area to the center of the field of view, and move the probe tip above the position to be measured to complete the coarse alignment; S106: Measurement sensitivity and probe stiffness settings: Further reduce the Z-axis distance between corneal cells and probe to 10~20 μm, calibrate the measurement sensitivity through thermal tuning, and use the nominal stiffness value for probe stiffness. S107: Single-point detection of corneal morphology and mechanical properties: Set the voltage reference value at the moment when the probe contacts the corneal cell and produces a certain static indentation, adjust the distance between the probe and the corneal cell, record the probe deformation voltage signal and the displacement value of the positioning stage during the entire process of the probe approaching-contacting-moving away from the corneal cell, and obtain the force-displacement curve by using the measurement sensitivity and probe stiffness. S108: Simultaneous imaging of corneal morphology and mechanical properties: Adjust the position of the corneal cells to create a relative lateral displacement between them and the probe. Move the probe tip to the next detection point and repeat step S107. The corresponding static indentation remains unchanged, i.e., the contact voltage reference value is set to be constant. Then, measure the sample surface characteristics of several points in the area to be detected sequentially along the grating scanning path to achieve simultaneous imaging of corneal morphology and mechanical properties.

3. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 2, characterized in that: In S101, the corneal cells are divided into three groups: low myopia group: -3.00 D < SE ≤ -0.50 D; moderate myopia group: -6.00 D < SE ≤ -3.00 D; high myopia group: SE ≤ -6.00 D.

4. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 2, characterized in that: In S102, the culture conditions for the corneal cells are as follows: maintain a constant temperature of 37 °C in an incubator with 5% CO2 until the cells attach to the substrate of the culture dish, rinse the cells with phosphate buffer to remove surface proteins and dead cells, add fresh culture medium, and incubate for 1 hour.

5. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 2, characterized in that: The curve data obtained through S107, combined with the Hertz contact mechanics model, are used to calculate the nanomechanical properties of the corneal cell surface, including but not limited to adhesion force and Young's modulus.

6. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 1, characterized in that: S2 specifically includes the following steps: S201: Image preprocessing: Convert the RGB color image of corneal cell surface morphology obtained by atomic force microscopy into a grayscale image, use the average brightness of each image as a threshold, set pixels below this threshold to 0, and leave other pixels unchanged. S202: Calculation of average complexity of cytoskeleton region: Using the image preprocessed in step S201 as the dataset, the cytoskeleton structure is extracted using the adaptive Canny edge detection algorithm to obtain the corresponding binary image; during the adaptive edge detection process, the image is denoised by Gaussian filtering, the image gradient matrix is ​​calculated, the optimal threshold is obtained through iterative optimization, and further refined using the Otsu method to achieve dynamic adjustment of the Canny edge detection threshold; the fractal dimension D is calculated using the Box-counting method to evaluate the cytoskeleton complexity; S203: Calculation of average similarity of cytoskeleton region: Using the image preprocessed in step S201 as the dataset, the image is divided into sub-blocks of different scales. Zero-pixel sub-blocks are removed to reduce the influence of uneven background distribution on the calculation results. The idea of ​​renormalization group is used to evaluate the uniformity of cytoskeleton image.

7. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 6, characterized in that: In S202, the average complexity parameter of the cytoskeleton region is set to evaluate the degree of disorder in the overall distribution of the cytoskeleton. In S203, the average similarity parameter of the cytoskeleton region is set to evaluate the regularity and uniformity of the overall distribution of the cytoskeleton.

8. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 6, characterized in that: The quantification parameters of corneal cell morphology characteristics include the average complexity of the cytoskeleton region and the average similarity of the cytoskeleton region. The average complexity of the cytoskeleton region is evaluated by the average fractal dimension, and the average similarity of the cytoskeleton region is evaluated by the average self-similarity error.

9. The myopia grading method driven by the coupling of corneal cell mechanics and morphological characteristics according to claim 1, characterized in that: In S3, 80% of all data is used as the training set and 20% as the test set. The main steps include: reading the data, extracting features and labels, standardizing the data, dividing the training set and the test set, using cross-validation to select the K value, training the KNN classification model, predicting the test set data, and calculating the classification evaluation index.

Images