Diopter Prediction Method, Storage Medium and Electronic Device after Intraocular Lens Implantation

Abstract

The present application relates to a method for predicting refractive power after intraocular lens implantation, a storage medium, and an electronic device, which includes: obtaining sample data in the posterior chamber of the eye with a lens; using the sample data as input parameters, and separately training and jointly training an intraocular contact lens (ICL)-type intraocular lens and a toric ICL (TICL)-type intraocular lens by using different preset machine learning prediction models, so as to obtain a postoperative separately-trained prediction result and a postoperative jointly-trained prediction result; using the postoperative separately-trained prediction result and the postoperative jointly-trained prediction result as new input parameters, and separately training ICL and TICL by using different machine learning prediction models, so as to obtain a trained model; using the trained model for prediction, obtaining a prediction result and comparing it with a preset experimental result to obtain a comparison result, and determining the best prediction models for ICL and TICL according to the comparison result. The present application has the effect of improving the accuracy of predicting refractive power after intraocular lens implantation through the best prediction models.

Description

Technical Field

[0001] The present application relates to the technical field of intraocular lenses, and particularly relates to a method for predicting refractive power after intraocular lens implantation, a storage medium, and an electronic device. Background Art

[0002] Currently, in the correction of ametropia of the eye, especially in the correction of aphakic ametropia after cataract surgery, the method of implanting an intraocular lens into the posterior chamber of the phakic eye is usually adopted, which not only improves the vision of patients, but also improves the life of patients. In order to achieve the best visual effect for patients after implanting an intraocular lens, the accuracy of postoperative refractive power prediction has attracted the attention of many scientists. Intraocular lenses include Implantable Collamer Lens (ICL) and Toric Implantable Collamer Lens (TICL). ICL is suitable for correcting high myopia, and TICL is suitable for correcting high myopia combined with astigmatism. Among them, ametropia refers to the condition where, when the eye is not using accommodation, parallel light rays, after passing through the refractive action of the eye, cannot form a clear image on the retina, but form an image in front of or behind the retina.

[0003] In related technologies, the most commonly used method for predicting refractive power after ICL / TICL implantation clinically is the vergence formula. The parameters included in such formulas are only parameters such as patient age, spherical lens, cylindrical lens, axis, anterior chamber depth, corneal thickness, white-to-white, etc. In addition, the accuracy of such formulas is easily affected by different ocular anatomical structures (axial length, corneal curvature).

[0004] Regarding the above related technologies, the inventors believe that there are the following defects: the parameters included in the calculation formula of the postoperative refractive power prediction method are not comprehensive enough, and the accuracy of the refractive power predicted by the calculation formula needs to be improved. Summary of the Invention

[0005] In order to improve the accuracy of predicting refractive power after intraocular lens implantation, the present application provides a method for predicting refractive power after intraocular lens implantation, a storage medium, and an electronic device.

[0006] In a first aspect, the method for predicting refractive power after intraocular lens implantation, the storage medium, and the electronic device provided by the present application adopt the following technical solution: The method for predicting refractive power after intraocular lens implantation includes:

[0007] Obtaining sample data in the posterior chamber of the phakic eye;

[0008] Taking the sample data as input parameters, separately training the intraocular lens of ICL type and the intraocular lens of TICL type using different pre-set machine learning prediction models to obtain the separate training prediction results after surgery;

[0009] Taking the sample data as input parameters, jointly training the intraocular lens of ICL type and the intraocular lens of TICL type using different pre-set machine learning prediction models to obtain the joint training prediction results after surgery;

[0010] Taking the separate training prediction results after surgery and the joint training prediction results after surgery as new input parameters, separately training the intraocular lens of ICL type and the intraocular lens of TICL type using different pre-set machine learning prediction models to obtain the trained models;

[0011] Using the trained models for prediction, obtaining the prediction results and comparing them with the pre-set experimental results to obtain the comparison results, and determining the best prediction models for ICL and TICL according to the comparison results.

[0012] By adopting the above technical solution, taking the sample data as the input parameters of the machine learning prediction model, separately and jointly training ICL and TICL successively, enabling the machine learning prediction model to learn the different features and common features of ICL and TICL, taking the separate training prediction results after surgery and the joint training prediction results after surgery as new input parameters, and separately training ICL and TICL again using the machine learning model to obtain the final trained models, comparing the obtained prediction results with the experimental results, and finding the best prediction models, so that the machine learning prediction model learns more fully and the prediction results are more accurate, thereby improving the accuracy of the refractive power prediction after intraocular lens implantation.

[0013] Optionally, before taking the sample data as input parameters, it includes:

[0014] Cleaning the sample data to obtain cleaned data;

[0015] Dividing the cleaned data into a training and validation set and a test set according to a pre-set division ratio;

[0016] The step of taking the sample data as input parameters, separately training the intraocular lens of ICL type and the intraocular lens of TICL type using different pre-set machine learning prediction models to obtain the separate training prediction results after surgery includes:

[0017] Based on the training validation set, different preset machine learning prediction models are used to separately train ICL-type intraocular lenses and TICL-type intraocular lenses to obtain separate postoperative training prediction results;

[0018] The sample data in the posterior chamber of the phakic eye is used as a model input parameter, and different preset machine learning prediction models are used to perform combined training on the ICL type intraocular lens and the TICL type intraocular lens, respectively, to obtain a postoperative combined training prediction result, including:

[0019] Based on the training validation set, different preset machine learning prediction models are used to perform combined training on ICL-type artificial lenses and TICL-type artificial lenses respectively to obtain postoperative combined training prediction results.

[0020] By adopting the above technical solution, after obtaining the sample data, some unqualified data in the sample data is cleaned up, so as to ensure the accuracy and objectivity of the final sample data, and then the cleaned data obtained after cleaning is divided into a training validation set and a test set. Different machine learning prediction models are used for training based on the data in the training validation set. The training validation set can train and verify the model, and the test set can perform a final evaluation of the model to determine the best model.

[0021] Optionally, the step of dividing the cleaned data into data sets according to a preset division ratio to obtain a training validation set and a test set comprises:

[0022] The training validation set is divided into a training set and a validation set according to a preset division ratio.

[0023] By adopting the above technical solution, the divided training set and validation set are convenient for the implementation of cross-training validation. A fixed proportion of data in the training validation set is selected in turn as the training set, and the rest is used as the validation set, so that the machine learning prediction model can be trained multiple times, which helps to select the most appropriate model parameters for the machine learning prediction model.

[0024] Optionally, based on the training validation set, the ICL type intraocular lens and the TICL type intraocular lens are trained separately using different preset machine learning prediction models to obtain separate postoperative training prediction results, including:

[0025] Based on the training set and the validation set and using multi-fold cross validation, the ICL type intraocular lens and the TICL type intraocular lens are trained separately to determine the separate optimal hyperparameters of the different machine learning prediction models in each experiment, wherein the hyperparameters are model parameters set for each machine learning prediction model;

[0026] Obtain a machine learning separate training prediction model according to the separately optimal hyperparameters;

[0027] Perform prediction on the validation set using the machine learning separate training prediction model to obtain the postoperative separate training prediction result.

[0028] Through the above technical solution, the validation set data is input into the machine learning prediction model. By continuously setting the respective model parameters within the machine learning prediction model, separate training with multi-fold cross-validation is performed on ICL and TICL and postoperative prediction is carried out. The postoperative prediction of the prediction model is verified on the validation set to find the best-performing model parameters, thereby obtaining the machine learning separate training prediction model, so that the machine learning separate training prediction model can obtain higher accuracy of the postoperative separate prediction results for ICL and TICL respectively.

[0029] Optionally, based on the training validation set, and using different preset machine learning prediction models to perform combined training on the ICL type intraocular lens and the TICL type intraocular lens respectively, to obtain the postoperative combined training prediction result, including:

[0030] Based on the training set and the validation set, perform combined training on the ICL type intraocular lens and the TICL type intraocular lens using multi-fold cross-validation to determine the combined optimal hyperparameters of different machine learning prediction models for each experiment;

[0031] Obtain a machine learning combined training prediction model according to the combined optimal hyperparameters;

[0032] Perform prediction on the validation set using the machine learning combined training prediction model to obtain the postoperative combined training prediction result.

[0033] Through the above technical solution, the validation set data is input into the machine learning prediction model. By continuously setting the respective model parameters within the machine learning prediction model, combined training with multi-fold cross-validation is performed on ICL and TICL and postoperative prediction is carried out. The postoperative prediction of the prediction model is verified on the validation set to find the best-performing model parameters, thereby obtaining the machine learning combined training prediction model, so that the machine learning combined training prediction model can obtain higher accuracy of the postoperative combined prediction results for ICL and TICL respectively.

[0034] Optionally, using the postoperative separate training prediction result and the postoperative combined training prediction result as new input parameters, and using different preset machine learning prediction models to perform separate training on the ICL type intraocular lens and the TICL type intraocular lens respectively to obtain the trained model, further includes:

[0035] Combine the new input parameters with the sample data to obtain a new training set;

[0036] Perform multi-fold cross-validation on the new training set to determine the final optimal hyperparameters of different machine learning prediction models;

[0037] Obtain the trained model according to the final optimal hyperparameters.

[0038] Through the above technical solution, the data in the new training set is used for multi-fold cross-training of different machine learning prediction models, and the predictions obtained from the training are verified on the validation set to determine the optimal hyperparameters of each machine learning prediction model, and finally the corresponding trained machine learning prediction model is obtained, so that the prediction results of the model are more accurate.

[0039] Optionally, before using the trained model for prediction, obtaining the prediction result and comparing it with the preset experimental result to obtain a comparison result, and determining the best prediction models for ICL and TICL according to the comparison result, further includes:

[0040] Use the machine learning separate training prediction model to perform average prediction on the test set to obtain separate test set parameters;

[0041] Use the machine learning combined training prediction model to perform average prediction on the test set to obtain combined test set parameters;

[0042] Combine the separate test set parameters, the combined test set parameters and the sample data to obtain a new test set.

[0043] Through the above technical solution, based on the parameter data of the test set, use the machine learning separate training prediction model and the machine learning combined training prediction model after multi-fold cross-training to perform result prediction respectively, and take the average value of the predictions respectively. Combine the average value and the original sample data to obtain the final new test set, so that the accuracy of the new test set is higher and the generalization ability of the evaluation model is better.

[0044] Optionally, using the trained model for prediction, obtaining the prediction result and comparing it with the preset experimental result to obtain a comparison result, and determining the best prediction models for ICL and TICL according to the comparison result, includes:

[0045] Use the trained model to perform prediction on the new test set to obtain the prediction result;

[0046] When the MEA index of the prediction result is smaller than the MEA index of the preset experimental result in the comparison result, determine that the model corresponding to the minimum MEA index is the best prediction model.

[0047] Through the above technical solution, the trained model obtained by training with the new training set has a relatively high degree of optimization. Finally, the result prediction is carried out on the new test set with a relatively high degree of optimization, and the prediction result is compared with the real experimental data. The trained model with the smallest MEA index of the prediction result of the ICL type intraocular lens and the TICL type intraocular lens is respectively selected as the best prediction model for the refractive power after intraocular lens implantation, which can make the predicted data of the refractive power after surgery more accurate.

[0048] In summary, the present application includes at least one of the following beneficial technical effects:

[0049] 1. By separately and jointly training the ICL type intraocular lens and the TICL type intraocular lens, the obtained prediction result is used as a new input parameter to determine the final best prediction model, thereby improving the accuracy when using this model to predict the refractive power after intraocular lens implantation;

[0050] 2. By cleaning the obtained sample data to obtain the cleaned data, the accuracy and objectivity of the final sample data are better;

[0051] 3. Through the obtained new test set, the machine learning prediction model after continuous debugging can be finally evaluated, so as to determine the respective best models of the ICL type intraocular lens and the TICL type intraocular lens. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a schematic flowchart of the method for predicting the refractive power after intraocular lens implantation according to an embodiment of the present application.

[0053] Figure 2 is a schematic flowchart of step S14 according to an embodiment of the present application.

[0054] Figure 3 is a schematic flowchart of step S15 according to an embodiment of the present application.

[0055] Figure 4 is a schematic flowchart of step S16 according to an embodiment of the present application.

[0056] Figure 5 is a schematic flowchart of step S20 according to an embodiment of the present application.

[0057] Figure 6 is a comparison table of the prediction results and experimental results of the best prediction models of ICL and TICL in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] The following will further describe the present application in detail Figure 1-6 with reference to the accompanying drawings.

[0059] The embodiments of the present application disclose a method for predicting refractive power after intraocular lens implantation, a storage medium, and an electronic device.

[0060] Reference Figure 1 , the method for predicting refractive power after intraocular lens implantation includes:

[0061] S10: Obtain sample data in the posterior chamber of the phakic eye;

[0062] Specifically, the sample data in this embodiment contains multiple parameters. There are 24 specific model input parameters as follows: age at the time of surgery, preoperative spherical lens, preoperative cylindrical lens, preoperative axis, preoperative equivalent spherical lens, dark pupil, intraocular pressure, axial length of the eye, K1, K1 axis, K2, K2 axis, average of K1 and K2, anterior chamber depth (excluding corneal endothelial thickness), central corneal thickness, corneal diameter, anterior chamber volume, anterior chamber angle, pupil diameter, ICL spherical lens (or TICL spherical lens), ICL cylindrical lens (or TICL cylindrical lens), ICL (or TICL equivalent spherical lens), ICL size (or TICL size), axis. These 24 input parameters are not fixed and can be appropriately reduced in other embodiments, such as 23, 25, etc. The specific model prediction parameters are: equivalent spherical lens from one week to one month after surgery and spherical lens from one week to one month after surgery. In other embodiments, the specific model prediction parameters can also be equivalent spherical lens from one week to one year after surgery and spherical lens from one week to one year after surgery.

[0063] S11: Clean the sample data to obtain cleaned data;

[0064] Specifically, the cleaning of the sample data includes discarding samples with incomplete data and excluding some sample data according to preset exclusion criteria. The preset exclusion criteria are as follows: having any previous ocular complications (corneal or fundus lesions) or surgical history, trauma history; having complications after surgery; postoperative corrected visual acuity < 0.5 or losing more than 1 line; to avoid astigmatism corrected through the surgical incision, excluding ICL preoperative astigmatic cylindrical lens > 0.5 D; any inspection or follow-up medical history data is incomplete. Finally, the cleaned data is obtained, which better ensures the accuracy and objectivity of the data.

[0065] S12: Divide the cleaned data into a training and validation set and a test set according to a preset division ratio;

[0066] Specifically, the preset division ratio of this embodiment adopts a 4:1 ratio of people. For example, if the number of people involved in the collected data is 100, then after the division, there are 80 people in the training and verification set, and 20 people in the test set, and it is ensured that the people in the test set are only monocular, and are randomly included in the left eye or right eye. Because the collected samples may collect monocular data or binocular data. People in the training and verification set can be randomly included in monocular or binocular.

[0067] S13: Divide the training validation set into a training set and a validation set according to a preset division ratio;

[0068] Specifically, the division ratio in this embodiment adopts a 4:1 ratio of the number of people, that is, the training verification set is divided into 5 equal parts according to the number of people, and 4 of them are selected in turn as training sets, and the remaining 1 is used as a verification set, so as to perform cross-validation training later. The training set is used to train the parameters in the model, and the verification set is used to test the state of the model during training. It should be noted that the verification set here is similar to the above-mentioned test set, and only a single eye is included, and the left eye or the right eye is randomly included.

[0069] S14: Based on the training validation set, different preset machine learning prediction models are used to train ICL-type intraocular lenses and TICL-type intraocular lenses separately to obtain postoperative separate training prediction results;

[0070] Specifically, the different machine learning prediction models preset in this embodiment use four machine learning prediction models, namely, lasso regression, support vector machine regression, random forest regression, and XGBoost regression models. In other embodiments, other machine learning prediction models such as logistic regression models may also be used. The number of the four machine learning prediction models is not fixed, and five machine learning prediction models may also be used in other embodiments. The four machine learning prediction models are used to separately train ICL-type intraocular lenses and TICL-type intraocular lenses. According to specific prediction parameters: equivalent spherical lens one week to one month after surgery and spherical lens one week to one month after surgery, the prediction results of ICL equivalent spherical lens separate training after surgery, ICL spherical lens separate training results, TICL equivalent spherical lens separate training results, and TICL spherical lens separate training results are finally obtained.

[0071] refer to Figure 2 , S141: Based on the training set and the validation set and using multi-fold cross validation, the ICL type intraocular lens and the TICL type intraocular lens are trained separately to determine the separate optimal hyperparameters of different machine learning prediction models in each experiment, wherein the hyperparameters are model parameters set for each machine learning model;

[0072] Specifically, the embodiment of the present application adopts a five-fold cross validation, where the training validation set is divided into five equal parts according to the number of people, and four of them are selected in turn as training sets, and the remaining one is used as a validation set, so as to perform a five-fold cross validation. In other embodiments, six-fold cross training can also be adopted, that is, the training validation set is divided into five equal parts according to the number of people, and five of them are selected in turn as training sets, and the remaining one is used as a validation set. Hyperparameters are parameters set in each machine learning prediction model, such as parameter X and parameter XX. First, the machine learning prediction model with parameter X is set to train ICL and TICL separately to obtain prediction result X, and then the parameters of the machine learning model are adjusted to parameter XX, and then ICL and TICL are trained separately to obtain prediction result XX. The prediction result X and the prediction result XX are verified on the validation set, and the best prediction of the two is obtained by comparison. The parameters corresponding to the best prediction are determined as the separate best hyperparameters of the machine learning prediction model.

[0073] refer to Figure 2 , S142: obtaining a machine learning separate training prediction model according to separate optimal hyper parameters;

[0074] Specifically, the originally set parameters in the corresponding machine learning prediction model are adjusted according to the obtained optimal hyperparameters, so as to obtain a machine learning separate training prediction model, which is more accurate in postoperative prediction than the previous machine learning prediction model.

[0075] refer to Figure 2 , S143: predicting the machine learning separate training prediction model on the validation set, obtaining the postoperative separate training prediction results, and forming new input parameters on the corresponding training set;

[0076] Specifically, in this embodiment, the prediction of ICL equivalent spherical lens after surgery is taken as an example. The four machine learning separately trained prediction models of ICL equivalent spherical lens are: ICL equivalent spherical lens lasso regression separately trained prediction model, ICL equivalent spherical lens support vector machine regression separately trained prediction model, ICL equivalent spherical lens random forest regression separately trained prediction model and ICL equivalent spherical lens XGBoost regression separately trained prediction model. The four machine learning separately trained prediction models are sequentially rolled predicted on the validation set of five-fold crossover. Because the five-fold cross-validation requires five experiments, five validation sets will be generated. The equivalent spherical lens postoperative prediction results of the four machine learning separately trained prediction models on the five validation sets are combined to obtain the four ICL equivalent spherical lens postoperative separately trained prediction results corresponding to the four machine learning separately trained prediction models, forming four new input parameters on the corresponding training set.

[0077] refer to Figure 1, S15: Based on the training validation set, use different pre-set machine learning prediction models to separately perform combined training on intraocular lenses of the ICL type and intraocular lenses of the TICL type, and obtain combined separate training prediction results;

[0078] Specifically, in this embodiment, the different pre-set machine learning prediction models also use four machine learning prediction models: lasso regression, support vector machine regression, random forest regression, and XGBoost regression models. In other embodiments, other machine learning prediction models such as the logistic regression model can also be used. Use these 4 machine learning prediction models to perform combined training on intraocular lenses of the ICL type and intraocular lenses of the TICL type. The prediction parameters are still: the equivalent spherical lens from one week to one month after surgery and the spherical lens from one week to one month after surgery. Finally, obtain the combined training prediction results of the ICL+TICL equivalent spherical lens after surgery and the combined training prediction results of the ICL+TICL spherical lens after surgery.

[0079] Reference Figure 3 , S151: Based on the training set and the validation set, and using multi-fold cross-validation, perform combined training on intraocular lenses of the ICL type and intraocular lenses of the TICL type to determine the combined optimal hyperparameters of different machine learning prediction models for each experiment;

[0080] Specifically, the five-fold cross-validation is used in the embodiments of this application. In other embodiments, six-fold cross-training can also be used. Continuously set the parameters within each machine learning prediction model. Under different parameters, the machine learning prediction model performs combined training on ICL and TICL. Compare which parameter of the machine learning prediction model gives better prediction results on the validation set, and determine the combined optimal hyperparameters of the machine learning prediction model.

[0081] Reference Figure 3 , S152: Obtain the machine learning combined training prediction model according to the combined optimal hyperparameters;

[0082] Specifically, adjust the originally set parameters within the corresponding machine learning prediction model according to the obtained combined optimal hyperparameters to obtain the machine learning combined training prediction model, which is more accurate in postoperative prediction than the previous machine learning prediction model.

[0083] Reference Figure 3 , S153: Predict the machine learning combined training prediction model on the validation set to obtain the postoperative combined training prediction results;

[0084] Specifically, in this embodiment, the prediction after ICL equivalent spherical power surgery is still taken as an example. The four machine learning combined training prediction models for ICL equivalent spherical power are: the ICL + TICL equivalent spherical power lasso regression combined training prediction model, the ICL + TICL equivalent spherical power support vector machine regression combined training prediction model, the ICL + TICL equivalent spherical power random forest regression combined training prediction model, and the ICL + TICL equivalent spherical power XGBoost regression combined training prediction model. The four machine learning combined training prediction models are sequentially and iteratively predicted on the validation set of five-fold cross-validation. Since five-fold cross-validation requires 5 experiments, 5 validation sets will be generated. The prediction results after ICL equivalent spherical power surgery of the four machine learning combined training prediction models on the 5 validation sets are combined to obtain the four ICL + TICL equivalent spherical power postoperative combined training prediction results corresponding to the four machine learning combined training prediction models.

[0085] Reference Figure 1 , S16: Use the postoperative separately trained prediction results and the postoperative combined trained prediction results as new input parameters, and separately train the ICL type intraocular lens and the TICL type intraocular lens using different preset machine learning prediction models to obtain the trained models;

[0086] Specifically, in this embodiment, the postoperative separately trained prediction results and the postoperative combined trained prediction results are used as new model input parameters. Still taking the prediction after equivalent spherical power surgery as an example, then the four ICL equivalent spherical power postoperative separately trained prediction results, the four TICL equivalent spherical power postoperative separately trained prediction results, and the four ICL + TICL equivalent spherical power postoperative combined trained prediction results are used as new model input parameters and incorporated into the four machine learning prediction models for another round of separate training. After the model undergoes the final learning and training, the trained models are obtained. It should be noted that in other embodiments, either the postoperative separately trained prediction results or the postoperative combined trained prediction results can also be incorporated into the four machine learning prediction models.

[0087] Reference Figure 4 , S161: Combine the new input parameters with the sample data to obtain a new training set;

[0088] Specifically, since the sample data includes 24 model input parameters in the previous example, taking the prediction after ICL equivalent spherical power surgery as an example, these 24 model input parameters, the four ICL equivalent spherical power postoperative separately trained prediction results, and the four ICL + TICL equivalent spherical power postoperative combined trained prediction results constitute the new training set.

[0089] Reference Figure 4 , S162: Perform multi-fold cross-validation on the new training set to determine the final optimal hyperparameters of different machine learning prediction models;

[0090] Specifically, the new training set replaces the original training set, and training is carried out on the new training set. The parameters set in the four machine learning prediction models are continuously adjusted, and the four machine learning prediction models obtain corresponding prediction results. Based on the best prediction with higher accuracy in the prediction results on the validation set, the final best hyperparameters are determined.

[0091] Reference Figure 4 , S163: Obtain the trained model according to the final best hyperparameters.

[0092] Specifically, according to the obtained final best hyperparameters, the parameters originally set in the corresponding machine learning prediction model are adjusted to obtain the trained model. According to the predicted parameters of the equivalent spherical lens from one week to one month after the operation and the spherical lens from one week to one month after the operation, specifically, the four models for predicting the ICL equivalent spherical lens after the operation, the four models for predicting the ICL spherical lens after the operation, the four models for predicting the TICL spherical lens after the operation, and the four models for predicting the TICL equivalent spherical lens after the operation are obtained. The four models here correspond to the four machine learning separate training prediction models.

[0093] Reference Figure 1 , S17: Use the machine learning separate training prediction model to perform an average prediction on the test set to obtain the separate test set parameters;

[0094] Specifically, four machine learning separate prediction models are used to train on the test set through five-fold cross-validation, taking turns 5 times. For the 5 prediction results obtained by each machine learning separate prediction model, the average value of the 5 prediction results is taken, and finally 4 separate test set parameters are obtained. Here, taking the prediction of the ICL equivalent spherical lens after the operation as an example, finally 4 separate test set parameters of the ICL equivalent spherical lens are obtained.

[0095] Reference Figure 1 , S18: Use the machine learning combined training prediction model to perform an average prediction on the test set to obtain the combined test set parameters;

[0096] Specifically, four machine learning combined prediction models are used to train on the test set through five-fold cross-validation, taking turns 5 times. For the 5 prediction results obtained by each machine learning combined prediction model, the average value of the 5 prediction results is taken, and finally 4 combined test set parameters are obtained. Here, taking the prediction of the ICL equivalent spherical lens after the operation as an example, finally 4 combined test set parameters of the ICL equivalent spherical lens are obtained.

[0097] Reference Figure 1 , S19: Combine the separate test set parameters, the combined test set parameters, and the sample data to obtain a new test set;

[0098] Specifically, the 4 separate test set parameters, the 4 combined test set parameters, and the 24 model input parameters included in the original sample data constitute a new test set with 32 parameters.

[0099] Reference Figure 1 , S20: Use the trained model for prediction, obtain the prediction result, compare it with the preset experimental result, obtain the comparison result, and determine the best prediction models for ICL and TICL according to the comparison result;

[0100] Specifically, the preset experimental results here are the results predicted by the currently commonly used postoperative refractive power calculation formula in clinical practice. The 4 machine learning separate training prediction models corresponding to ICL equivalent spherical lens, ICL spherical lens, TICL equivalent spherical lens, and TICL spherical lens are respectively used for result prediction, and the prediction results are respectively compared with the experimental results. Finally, the best prediction models for ICL equivalent spherical lens, ICL spherical lens, TICL equivalent spherical lens, and TICL spherical lens are obtained.

[0101] Reference Figure 5 , S201: Use the trained model to make predictions on the new test set to obtain the prediction results;

[0102] Specifically, the 4 machine learning separate training prediction models corresponding to ICL equivalent spherical lens, ICL spherical lens, TICL equivalent spherical lens, and TICL spherical lens are subjected to five-fold cross-training on the new test set to obtain the corresponding prediction results.

[0103] Reference Figure 5 and Figure 6 , S202: When the MEA index of the prediction result is smaller than the MEA index of the preset experimental result in the comparison result, determine that the model corresponding to the minimum MEA index is the best prediction model.

[0104] Specifically, when the 4 prediction results obtained by the 4 machine learning separate training prediction models of ICL equivalent spherical lens are smaller than the MEA index of the ICL equivalent spherical lens STAAR experimental result, take the one with the smallest MEA index among these 4 prediction results. Here, STAAR is the currently commonly used postoperative refractive power calculation formula in clinical practice. As shown in Table 1, the prediction result of the lasso separate training prediction model has the smallest MEA index, so this lasso separate training prediction model is determined as the best prediction model for ICL equivalent spherical lens. By analogy, the comparison of ICL spherical lens is shown in Table 2, the comparison of TICL equivalent spherical lens is shown in Table 3, and the comparison of TICL spherical lens is shown in Table 4. It should be noted that SVR refers to the machine learning prediction model of support vector regression mentioned above. It can be seen from Tables 1-4 that the MAE of our model is relatively small in all prediction comparisons, which further shows that the accuracy of our machine learning-based prediction model is better.

[0105] The implementation principle of the diopter prediction method after intraocular lens implantation in the embodiments of this application is as follows: After obtaining the sample data covering multiple parameters in the posterior chamber of the eye with a lens, it is necessary to clean the sample data. The obtained cleaned data is divided into a training and validation set and a test set. The training and validation set is further divided into a training set and a validation set. Based on the training set and the validation set and using the multi-fold cross-validation method, separate training and combined training are performed on the intraocular lenses of the ICL and TICL types respectively through a machine learning prediction model. The prediction results obtained from the separate training and the combined training are used as new input parameters and combined with the original sample data to form a new training set. Separate training is performed again through the machine learning prediction model. The trained model is determined by rolling validation on the validation set. Finally, the trained model is used to predict the results on the new test set, and the results are compared with the experimental data predicted by the STAAR postoperative diopter calculation formula to determine the best prediction models for ICL and TICL.

[0106] The embodiments of this application also disclose a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the diopter prediction method after intraocular lens implantation in the above embodiments is adopted.

[0107] Among them, the computer program can be stored in a computer-readable medium. The computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some middleware form, etc. The computer-readable medium includes any entity or device, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the computer-readable medium includes, but is not limited to, the above components.

[0108] Among them, through this computer-readable storage medium, the diopter prediction method after intraocular lens implantation in the above embodiments is stored in the computer-readable storage medium and is loaded and executed on the processor to facilitate the storage and application of the above method.

[0109] The embodiments of this application also disclose an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, the diopter prediction method after intraocular lens implantation in the above embodiments is adopted.

[0110] Among them, the electronic device can be a computer device such as a desktop computer, a laptop computer, or a cloud server, and the electronic device includes, but is not limited to, a processor and a memory. For example, the electronic device may also include input and output devices, network access devices, and a bus, etc.

[0111] Among them, the processor may adopt a central processing unit (CPU). Of course, according to the actual usage, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. may also be adopted. The general-purpose processor may adopt a microprocessor or any conventional processor, etc. This application does not make any restrictions in this regard.

[0112] Among them, the memory may be an internal storage unit of the electronic device. For example, the hard disk or memory of the electronic device. It may also be an external storage device of the electronic device. For example, a plug-in hard disk, a smart media card (SMC), a secure digital card (SD), or a flash card (FC) equipped on the electronic device, etc. Moreover, the memory may also be a combination of the internal storage unit and the external storage device of the electronic device. The memory is used to store computer programs and other programs and data required by the electronic device. The memory may also be used to temporarily store the data that has been output or will be output. This application does not make any restrictions in this regard.

[0113] Among them, through this electronic device, the diopter prediction method after the implantation of the intraocular lens in the above embodiment is stored in the memory of the electronic device, and is loaded and executed on the processor of the electronic device, which is convenient for use.

[0114] The embodiments of this specific implementation manner are all preferred embodiments of this application, and do not limit the protection scope of this application accordingly. Therefore, all equivalent changes made according to the structure, shape, and principle of this application should be covered within the protection scope of this application.

Claims

1. A method for predicting the refractive power after intraocular lens implantation, characterized in that, Including: Obtaining sample data in the posterior chamber of the phakic eye; Taking the sample data as input parameters, and separately training the intraocular lens of the ICL type and the intraocular lens of the TICL type using different preset machine learning prediction models to obtain the separate training prediction results after surgery; Taking the sample data as input parameters, and jointly training the intraocular lens of the ICL type and the intraocular lens of the TICL type using different preset machine learning prediction models to obtain the joint training prediction results after surgery; Taking the separate training prediction results after surgery and the joint training prediction results after surgery as new input parameters, and separately training the intraocular lens of the ICL type and the intraocular lens of the TICL type using different preset machine learning prediction models to obtain the trained models, including: combining the new input parameters with the sample data to obtain a new training set; Performing multi-fold cross-validation on the new training set to determine the final optimal hyperparameters of different machine learning prediction models; Obtaining the trained models according to the final optimal hyperparameters; Using the trained models for prediction, obtaining the prediction results and comparing them with the preset experimental results to obtain the comparison results, and determining the best prediction models for ICL and TICL according to the comparison results; wherein, the different preset machine learning prediction models used above are four machine learning prediction models: lasso regression, support vector machine regression, random forest regression, and XGBoost regression models.

2. The method for predicting the diopter after intraocular lens implantation according to claim 1, wherein Before taking the sample data as input parameters, including: Cleaning the sample data to obtain cleaned data; Dividing the cleaned data into a training and validation set and a test set according to a preset division ratio; Taking the sample data as input parameters, and separately training the intraocular lens of the ICL type and the intraocular lens of the TICL type using different preset machine learning prediction models to obtain the separate training prediction results after surgery, including: Based on the training and validation set, and separately training the intraocular lens of the ICL type and the intraocular lens of the TICL type using different preset machine learning prediction models to obtain the separate training prediction results after surgery; Taking the sample data as model input parameters, and jointly training the intraocular lens of the ICL type and the intraocular lens of the TICL type using different preset machine learning prediction models to obtain the joint training prediction results after surgery, including: Based on the training and validation set, and jointly training the intraocular lens of the ICL type and the intraocular lens of the TICL type using different preset machine learning prediction models to obtain the joint training prediction results after surgery.

3. The diopter prediction method after intraocular lens implantation according to claim 2, wherein After dividing the cleaned data into a training and validation set and a test set according to a preset division ratio, including: Dividing the training and validation set into a training set and a validation set according to a preset division ratio.

4. The diopter prediction method after intraocular lens implantation according to claim 3, wherein Based on the training and validation set, separately train the intraocular lenses of ICL type and the intraocular lenses of TICL type using different preset machine learning prediction models, and obtain the postoperative separate training prediction results, including: Based on the training set and the validation set, and using multi-fold cross-validation to separately train the intraocular lenses of ICL type and the intraocular lenses of TICL type, determine the separate optimal hyperparameters of different machine learning prediction models for each experiment, where the hyperparameters are the model parameters set by each machine learning model; Obtain the machine learning separate training prediction model according to the separate optimal hyperparameters; Predict the machine learning separate training prediction model on the validation set to obtain the postoperative separate training prediction results.

5. The method for predicting the diopter after intraocular lens implantation according to claim 4, wherein, Based on the training and validation set, and using different preset machine learning prediction models to jointly train the intraocular lenses of ICL type and the intraocular lenses of TICL type, obtain the postoperative joint training prediction results, including: Based on the training set and the validation set, and using multi-fold cross-validation to jointly train the intraocular lenses of ICL type and the intraocular lenses of TICL type, determine the joint optimal hyperparameters of different machine learning prediction models for each experiment; Obtain the machine learning joint training prediction model according to the joint optimal hyperparameters; Predict the machine learning joint training prediction model on the validation set to obtain the postoperative joint training prediction results.

6. The method for predicting the refractive power after intraocular lens implantation according to claim 5, wherein Before using the trained model for prediction, obtaining the prediction results and comparing them with the preset experimental results to obtain the comparison results, and determining the best prediction models for ICL and TICL based on the comparison results, further includes: Use the machine learning separate training prediction model to perform average prediction on the test set to obtain separate test set parameters; Use the machine learning joint training prediction model to perform average prediction on the test set to obtain joint test set parameters; Combine the separate test set parameters, the joint test set parameters and the sample data to obtain a new test set.

7. The refractive power prediction method after intraocular lens implantation according to claim 6, wherein, Using the trained model for prediction, obtaining the prediction results and comparing them with the preset experimental results to obtain the comparison results, and determining the best prediction models for ICL and TICL based on the comparison results, includes: Use the trained model to perform prediction on the new test set to obtain the prediction results; When the MEA index of the prediction result in the comparison result is smaller than the MEA index of the preset experimental result, determine that the model corresponding to the minimum MEA index is the best prediction model.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is loaded and executed by the processor, the method described in any one of claims 1-7 is adopted.

9. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor loads and executes the computer program, the method described in any one of claims 1-7 is adopted.

Images