Orthokeratology lens parameter recommendation method

Predicting corneal resizing mirror parameters through machine learning methods with multi-model fusion, solving the problems of insufficient parameter recommendation and difficulty in adjustment in the existing technology, achieving more accurate and efficient corneal resizing mirror fitting, reducing patient risks and trials Number of times worn.

CN118939991AActive Publication Date: 2024-11-12SHANGHAI MEDIWORKS PRECISION INSTR CO LTD +1

Patent Information

Application Number
CN202410818286.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-24
Publication Date
2024-11-12
Estimated Expiration
2044-06-24

AI Technical Summary

Technical Problem

The existing corneal resizing fitting methods have problems such as insufficient parameter recommendation, difficulty in adjustment and single parameter recommendation, which leads to poor matching, affecting the efficacy of vision and myopia control, and there is a risk of cross-infection.

Method used

Using multiple biometric information as model inputs, a machine learning method for multi-model fusion is built to predict multiple parameters of corneal reshaping mirrors, including BCR, RZD1, RZD2, LZA1, LZA2 and TD, through support vector regression, Lasso regression and randomness Forest models make predictions and improve prediction accuracy through model fusion.

Benefits of technology

A more accurate and comprehensive prediction of corneal resizing lens parameters is achieved, which reduces the number of trials, improves fitting efficiency and accuracy, reduces the patient's burden and risk of cross-infection, and has a smaller difference between the recommended parameters and actual parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939991A_ABST
    Figure CN118939991A_ABST
Patent Text Reader

Abstract

The invention relates to an orthokeratology lens parameter recommendation method, which comprises the following steps of: 1, collecting data samples and incorporating characteristic parameters, and screening successfully fitted samples by adopting a corneal topographic map based on a Placido ring; step 2, data division and normalization: dividing a training verification set and a test set, and performing mean-variance normalization on each input feature; 3, modeling a machine learning model; 4, performing independent optimization on the hyper-parameters of the AI complete version and the AI simplified version, and searching an optimal hyper-parameter; 5, performing model training; and 6, carrying out model fusion prediction to obtain a final prediction value. The problems that a CRT lens fitting method is insufficient in fitting parameter, difficult to adjust and single in parameter recommendation are solved, multiple pieces of biological characteristic information and multi-model fusion prediction are included in the method, all orthokeratology lens parameters are predicted more accurately and more completely, related APP plug-ins can be generated by a currently used AI method and loaded on a corneal topography map, and the method has the advantages of being high in adaptability, high in adaptability and the like. And the orthokeratology lens is more efficiently and accurately fitted clinically.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field The present invention relates to the technical field of orthokeratology lenses, and in particular to a method for recommending orthokeratology lens parameters. Background Art Orthokeratology is one of the most effective methods for preventing and controlling myopia. Compared with traditional single-focus frame glasses and soft contact lenses, it can significantly slow down the elongation of the eye axis and the development of myopia. Paragon CRT (Corneal refractive therapy) lenses were approved by China's CFDA at the end of 2016 and are currently widely used in the fitting of orthokeratology lenses in clinical practice. The current commonly used clinical fitting method for CRT lenses is to use a pull-out caliper or parameter selection APP provided by the manufacturer to assist in the selection of trial lenses with two parameters. However, this method has many limitations. First, in clinical practice, the fitting doctor often needs to make adjustments based on experience evaluation, which requires a high level of practical experience from the fitter and a long learning curve. In addition, multiple lens changes during trial fitting also increase the complexity of the clinical process, increase the burden on patients and the risk of cross-infection. Secondly, the number of trial lens libraries is limited, and the lenses recommended by traditional methods may not have corresponding trial lenses, which further increases the limitations. In addition, traditional methods cannot directly provide some parameters such as lens diameter, toricity, and LZA2. The final fitting is greatly affected by the above factors, and poor fitting may lead to poor shaping, which not only affects the naked eye vision, visual quality, and myopia control efficacy during the day, but also leads to risks such as corneal damage. In addition, CN117826451A discloses a method for recommending parameters for orthokeratology lenses, which only calculates the landing zone parameters of orthokeratology lenses through the two parameters of pupil diameter and iris diameter. First, the individual characteristics of patients, such as corneal morphology, astigmatism, and corneal diameter, vary greatly. This method is not enough to only include pupil diameter and iris diameter to fit lenses. Secondly, the complete parameters of orthokeratology lenses usually include: BCR (central spherical region radius of curvature), RZD1 / RZD2 (central reversal zone), LZA1 / LZA2 (landing zone), and TD (lens diameter). However, this method only makes lens recommendations based on the landing zone parameters, and the lens parameter recommendations are single and insufficient. In addition, there are few studies on the use of artificial intelligence AI in the field of orthokeratology lens fitting, such as the following papers [1] and [2]. The results are all single lens parameters, which still need to be tried on and cannot directly form the final prescription required clinically.

[0001] Contact Lens and Anterior Eye 44(2021)240–269.

[0002] Chan KY,Cheung SW,Cho P.Clinical performance of an orthokeratologylens fitted with the aid of a computer software in Chinese children.Cont LensAnterior Eye.2012Aug;35(4):180-4. Summary of the invention Aiming at the problems of insufficient adaptation parameters, difficult adjustment and single parameter recommendation in the fitting method of CRT lenses, a method for recommending parameters of orthokeratology lenses is proposed. The method for recommending parameters of orthokeratology lenses of the present invention selects multiple biometric information as model input, builds a machine learning method of multi-model fusion, and predicts multiple parameters of orthokeratology lenses respectively. The technical solution of the present invention is: A method for recommending orthokeratology lens parameters, comprising the following steps: Step 1: Data sample collection and feature parameter inclusion: Professional doctors manually marked the successful fitting samples of orthokeratology lenses (CRT). The process of marking the successful fitting is as follows: using the corneal topography based on the Placido ring, selecting the tangential difference map, and dividing it into the following three categories according to the center positioning: Category 1: eccentricity less than 0.5mm and complete defocus ring are ideal; Category 2: 0.5mm≤eccentricity≤1.0mm and complete defocus ring are not ideal but acceptable; Category 3: eccentricity>1.0mm Or an incomplete defocus ring is unacceptable. Two or more ideal topography maps must be obtained within one year after the first glasses are fitted, which is marked as a successful fitting. Each sample contains the following biometric information: optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, SD8, and orthokeratology lens CRT parameter information: BCR, RZD1, RZD2, LZA1, LZA2, TD; Step 2: Data division and normalization: First, divide the training validation set and the test set into 4:1; then divide the training validation set into 5 equal parts, and the input features include optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, SD8; perform mean-variance normalization on each input feature; Step 3: Machine learning model building: In order to regress and predict the parameters of orthokeratology lenses, a machine learning regression model was adopted, and the following three were selected: support vector regression SVR, Lasso regression and random forest Random Forest; according to the different number of feature inputs, two AI versions, AI full version and AI simplified version, were established to predict the six orthokeratology lens CRT parameters of BCR, RZD1, RZD2, LZA1, LZA2, and TD; the input features of the AI ​​full version are optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, and SD8; the input features of the AI ​​simplified version are optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, and corneal diameter; Step 4: Hyperparameter optimization: The hyperparameters of the AI ​​full version and AI simplified version were optimized independently. For each orthokeratology lens parameter, the three machine learning models, SVR, Lasso and RF, were cross-validated at five folds, and the GridSearchCV grid search method was used to find the optimal hyperparameters. Step 5: Model training: The AI ​​full version and simplified version are trained independently. The optimal hyperparameters found in step 4 are used to perform 5-fold cross-training on SVR, Lasso and RF for each orthokeratology lens parameter. Four copies of the data are used for training each time, and the remaining one copy is used for verification, for a total of 5 training times. After each training, the trained model is saved. Each orthokeratology lens parameter obtains 5 trained models of the SVR, Lasso and RF models, so there are a total of 15 models for predicting a orthokeratology lens parameter. Step 6: Model fusion prediction: The AI ​​full version and simplified version are trained to make independent predictions. For the prediction of each orthokeratology lens parameter, the 15 trained models are used to predict the test set. Each test sample will get 15 prediction values, which are finally averaged and fused to get the final prediction value. Furthermore, it also includes step 7: comparative evaluation of prediction results, comparing the AI ​​method and the traditional clinical method of doctors on the test set, the AI ​​method includes the AI ​​full version and the AI ​​simplified version, and the traditional clinical method of doctors is the parameter selection APP; linear fitting is performed between the parameters recommended by different methods and the actual fixed lens parameters of OK lens wearers, and the R value is used to measure the correlation degree and R between the two variables. 2 The value is used to describe the degree to which the regression model explains the changes in the dependent variable to compare the strength of the correlation with the actual parameter. Furthermore, the method may use machine learning models not limited to SVR, Lasso and RF, but may also use other machine learning regression models. The beneficial effects of the present invention are: The present invention provides a method for recommending orthokeratology lens parameters. Compared with traditional manual methods and methods in published patents and papers, this method incorporates multiple biometric information and multi-model fusion predictions, and can more accurately and completely predict all orthokeratology lens parameters. In addition, the AI ​​method currently used can generate relevant APP plug-ins that are loaded on corneal topography to assist clinicians in more efficient and accurate fitting of orthokeratology lenses. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 A graph showing the proportion of the difference between the parameters recommended by different methods of the present invention and the actual lens parameters of the OK lens wearer; DETAILED DESCRIPTION The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments. Glossary: Corneal topography: The corneal morphology is digitally analyzed through a computer image processing system, and the information obtained is presented in a color map with different characteristics. It is called a corneal topography because it resembles the natural geographical description of an area in geology and expresses the ups and downs of the terrain surface. Tangential difference map: The tangential difference map represents the local curvature of each point on the cornea, and is sometimes called a "local curvature map" or "instant curvature map." It is more sensitive to changes in local curvature and more accurately reflects the shape of the corneal periphery. Center positioning: refers to whether the optical center of the lens is accurately aligned with the center of the eye's visual axis, that is, the center of the pupil, when performing orthokeratology or wearing contact lenses. IS: average diopter difference between upper and lower sides; SAI: corneal asymmetry index; SRI: corneal regularity index; SD8: Corneal height difference of 8mm diameter. BCR: Base Curvature Radius, that is, the base curvature radius, refers to the central curvature radius of the contact lens. RZD1 and RZD2: The first and second cut-off zone diameters (Return Zone Diameter) of the orthokeratology lens, that is, the diameters of the orthokeratology lens in the central lens area and the edge transition area. LZA1 and LZA2: The offset of the center and transition area of ​​the adapter lens (Lens Zone Angle), that is, the offset angle of the transition area between the center and edge of the contact lens relative to the base curve. TD: Total Diameter of the orthokeratology lens, that is, the diameter of the entire contact lens. The present invention provides a method for recommending orthokeratology lens parameters, which selects multiple biometric information as model input, builds a multi-model fusion machine learning method, and predicts multiple orthokeratology lens parameters respectively. The specific steps are as follows: Step 1: Data sample collection and feature parameter inclusion: 798 samples of successful orthokeratology lenses (CRT) fitting were selected through manual annotation by professional doctors. The process of marking successful fitting is as follows: using corneal topography based on Placido rings, selecting tangential difference maps, and dividing them into the following three categories according to the center positioning: Category 1: ideal (eccentricity less than 0.5 mm and complete defocus ring); Category 2: unsatisfactory but acceptable (0.5 mm ≤ eccentricity ≤ 1.0 mm and complete defocus ring); Category 3: unacceptable (eccentricity > 1.0 mm or incomplete defocus ring). Within one year after the first fitting, there must be two or more ideal topography maps to be marked as successful fitting. Each sample contains the following biometric information: optometry degree, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, SD8, and orthokeratology lens CRT parameter information: BCR, RZD1, RZD2, LZA1, LZA2, TD. Step 2: Data division and normalization: First, the training validation set and the test set were divided into 4:1, with 600 samples in the training validation set and 198 samples in the test set. Then the training validation set was divided into 5 equal parts, and the input features included optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, and SD8. Each input feature was normalized by mean-variance. Step 3: Machine learning model building: In order to regress and predict the parameters of orthokeratology lenses, we adopted a machine learning regression model and selected the following three: support vector regression (SVR), Lasso regression and random forest (RandomForest). According to the number of feature inputs, we established two AI versions, AI full version and AI simplified version, to predict the six orthokeratology lens CRT parameters of BCR, RZD1, RZD2, LZA1, LZA2, and TD. The input features of the AI ​​full version are optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, SD8, and the input features of the AI ​​simplified version are optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, and corneal diameter. Step 4: Hyperparameter optimization: The hyperparameters of the AI ​​full version and AI simplified version are optimized independently. For each orthokeratology lens parameter, the three machine learning models SVR, Lasso and RF are cross-validated with five folds, and the GridSearchCV grid search method is used to find the optimal hyperparameters. Step 5: Model training: The AI ​​full version and simplified version are trained independently. Using the optimal hyperparameters found in step 4, five-fold cross-training is performed on SVR, Lasso, and RF for each orthokeratology lens parameter. Four copies of the data are used for training each time, and the remaining one is used for verification, for a total of five training sessions. After each training session, the trained model is saved, and each orthokeratology lens parameter obtains five trained models of the SVR, Lasso, and RF models, so that there are a total of 15 models for predicting a orthokeratology lens parameter. Step 6: Model fusion prediction: The AI ​​full version and simplified version are trained and predicted independently. For the prediction of each orthokeratology lens parameter, the 15 trained models are used to predict the test set. Each test sample will get 15 prediction values, and finally the final prediction value is obtained by taking the average value and fusion. Step 7: Comparative evaluation of prediction results: Compare the AI ​​method (AI full version, AI simplified version) and the traditional clinical method of doctors (parameter selection APP) on a test set of 198 samples. 83.3% to 85.4% of the RZD1 recommended by AI are completely consistent with the clinical parameters, while the traditional method is completely consistent with 84.3%. The vast majority (99.5%) of the RZD1 recommended by the AI ​​full version and AI simplified version are within 25μm (i.e., the value of the lowest change in RZD) of the actual parameters. However, the traditional method shows an uneven distribution and a tendency to underestimate RZD1. Among the results with a difference of ≤25μm from the actual parameters, the proportion of low RZD1 values ​​recommended by the traditional method is as high as 90.0%, which is significantly higher than the AI ​​method (56.3-64.3%). For other parameters, including BCR, RZD2, LZA1 and TD, the values ​​recommended by the AI ​​method are better than those recommended by the traditional method. Specifically, the difference between the parameters recommended by the AI ​​method and the actual parameters is smaller than that recommended by the traditional method. The performance of the AI ​​method in RZD2 is particularly outstanding. In RZD2, the difference between the AI ​​method recommended parameters and the actual parameters is ≤25μm (1 level) in 90.4% to 93.4%, which is much higher than the 69.7% of the traditional method (such as Figure 1 The figure of RZD2 is shown, and the other figures are not attached one by one). Similar to RZD1, the traditional method is also prone to underestimate RZD2. Among the results with a difference of ≤25μm from the actual parameters, the proportion of the RZD1 recommended by the traditional method is as high as 91.1%, which is higher than the AI ​​method (51.0~53.8%). In BCR, the difference between the parameters recommended by the AI ​​method and the actual parameters is ≤0.1mm (1 level) in the proportion of 93.9% to 95.5%, while the traditional method is 92.93%. The difference between the LZA1 and LZA2 recommended by the AI ​​method and the actual fixed film is ≤1° (1 level) in the proportion of 97.5% and 97.0% respectively, which is higher than the traditional recommended LZA1 (90.4%), and the traditional method lacks LZA2 recommendation. In TD, the TD recommended by the AI ​​method is completely consistent with the actual fixed film TD in the proportion of 80.8% to 82.3%, which is higher than the traditional recommendation (75.3%). There was no significant difference between the parameters predicted by AI method and the actual parameters (p>0.05). Except for BCR and TD (p>0.05), the predicted RZD1 (p=0.022) and RZD2 (p<0.001) of APP parameters were significantly lower than the actual parameters, and the predicted LZA1 (p=0.002) was significantly higher than the actual parameters. Table 1 shows the linear correlation between the parameters recommended by different methods and the actual film parameters. The parameters recommended by the AI ​​method showed significant linear correlation with the actual film parameters (all p < 0.001), and the correlation coefficients between RZD2, LZA1, TD, etc. recommended by AI and the actual parameters were higher than those of traditional methods. Table 1. Linear fitting relationship between parameters recommended by different methods and actual lens parameters of OK lens wearers, Liner regression, **P < 0.001. NA = not available Among them, ** indicates that the statistical P value of the item is <0.001, and we will perform hypothesis testing with a significance level (alpha) of 0.05. If the calculated p value is less than 0.05, we usually consider the observed difference to be statistically significant and reject the null hypothesis; R value (correlation coefficient): R value is an indicator used to measure the strength of the linear relationship between two variables. Its value range is between -1 and 1: when the R value is close to 1, it means that there is a strong positive correlation between the two variables; when the R value is close to -1, it means that there is a strong negative correlation between the two variables; when the R value is close to 0, it means that there is no linear correlation between the two variables; R 2 Value (coefficient of determination): R 2 The coefficient of determination (also called the coefficient of determination) is an indicator used to describe the degree to which the independent variable explains the variation of the dependent variable. Its value range is between 0 and 1; R 2 The closer the value is to 1, the higher the degree to which the independent variable explains the change of the dependent variable, and the better the fitting effect of the regression model. 2The closer the value is to 0, the less the independent variable explains the change of the dependent variable, and the poorer the fit of the regression model; In general, the R value is used to measure the degree of correlation between two variables. 2 The value is used to describe the degree to which the regression model explains the variation of the dependent variable. Compared with the parameters recommended by traditional methods, the parameters such as BCR, RZD2, LZA1 and TD recommended by the AI ​​method have smaller differences from the actual film setting, are more accurate, and are more obvious in RZD2. There is no significant difference between the parameters predicted by the AI ​​method and the actual film setting parameters, while RZD1 and RZD2 predicted by APP are significantly lower than the actual film setting parameters, and LZA1 is significantly higher than the actual film setting parameters. Linear analysis also shows that all parameters recommended by AI show significant correlation with the actual film setting parameters, and in terms of parameters such as RZD2, LZA1 and TD, the correlation between the values ​​recommended by AI and the actual parameters is stronger than that recommended by traditional methods. This method can use machine learning models not limited to SVR, Lasso and RF, but can also use other machine learning regression models. The above-mentioned embodiment only expresses one implementation mode of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A method for recommending orthokeratology lens parameters, characterized in that: The following steps are involved: Step 1: Data sample collection and feature parameter inclusion: Professional doctors manually marked the successful fitting samples of orthokeratology lenses (CRT). The process of marking the successful fitting is as follows: using the corneal topography based on the Placido ring, selecting the tangential difference map, and dividing it into the following three categories according to the center positioning: Category 1: eccentricity less than 0.5mm and complete defocus ring are ideal; Category 2: 0.5mm≤eccentricity≤1.0mm and complete defocus ring are not ideal but acceptable; Category 3: eccentricity>1.0mm Or an incomplete defocus ring is unacceptable. Two or more ideal topography maps must be obtained within one year after the first glasses are fitted, which is marked as a successful fitting. Each sample contains the following biometric information: optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, SD8, and orthokeratology lens CRT parameter information: BCR, RZD1, RZD2, LZA1, LZA2, TD; Step 2: Data division and normalization: First, divide the training validation set and the test set into 4:1; then divide the training validation set into 5 equal parts, and the input features include optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, SD8; perform mean-variance normalization on each input feature; Step 3: Machine learning model building: In order to regress and predict the parameters of orthokeratology lenses, a machine learning regression model was adopted, and the following three were selected: support vector regression SVR, Lasso regression and random forest Random Forest; according to the different number of feature inputs, two AI versions, AI full version and AI simplified version, were established to predict the six orthokeratology lens CRT parameters of BCR, RZD1, RZD2, LZA1, LZA2, and TD; the input features of the AI ​​full version are optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, corneal diameter, IS, SAI, SRI, and SD8; the input features of the AI ​​simplified version are optometry, flat K value, flat K angle, steep K value, steep K angle, deltaK, flat E value, flat E angle, oblique E value, oblique E angle, and corneal diameter; Step 4: Hyperparameter optimization: The hyperparameters of the AI ​​full version and AI simplified version were optimized independently. For each orthokeratology lens parameter, the three machine learning models, SVR, Lasso and RF, were cross-validated at five folds, and the GridSearchCV grid search method was used to find the optimal hyperparameters. Step 5: Model training: The AI ​​full version and simplified version are trained independently. The optimal hyperparameters found in step 4 are used to perform 5-fold cross-training on SVR, Lasso and RF for each orthokeratology lens parameter. Four copies of the data are used for training each time, and the remaining one copy is used for verification, for a total of 5 training times. After each training, the trained model is saved. Each orthokeratology lens parameter obtains 5 trained models of the SVR, Lasso and RF models, so there are a total of 15 models for predicting a orthokeratology lens parameter. Step 6: Model fusion prediction: The AI ​​full version and simplified version are trained to make independent predictions. For the prediction of each orthokeratology lens parameter, the 15 trained models are used to predict the test set. Each test sample will get 15 prediction values, which are finally averaged and fused to get the final prediction value.

2. The method for recommending orthokeratology lens parameters according to claim 1, characterized in that: It also includes step 7: comparative evaluation of prediction results, comparing the AI ​​method and the traditional clinical method of doctors on the test set. The AI ​​method includes the complete version of AI and the simplified version of AI, and the traditional clinical method of doctors is a parameter selection APP; linear fitting is performed between the parameters recommended by different methods and the actual parameters of the OK lens wearers, and the R value is used to measure the correlation between the two variables, R 2 The value is used to describe the degree to which the regression model explains the changes in the dependent variable to compare the strength of the correlation with the actual parameter.

3. The method for recommending orthokeratology lens parameters according to claim 1, characterized in that: This method can use machine learning models not limited to SVR, Lasso and RF, but can also use other machine learning regression models.

Citation Information

Patent Citations

  • Orthokeratology lens fitting method, device and equipment and readable storage medium

    CN113378414A

  • Diagnostic classification of corneal shape abnormalities

    US20180000342A1

  • Method and electronic device for selecting deep neural network hyperparameters

    US20210174210A1

  • Improvements to fitting orthokeratology lenses

    WO2019000041A1

Cited By

  • A method for predicting corneal orthokeratology parameters based on artificial neural network

    CN119786068A

  • Intelligent cornea difference recognition system and method based on deep learning

    CN120236154A

  • A corneal difference intelligent recognition system and method based on deep learning

    CN120236154B