Dynamic myopia management lens prescription customization method and system

Abstract

The invention provides a dynamic myopia management lens prescription customization method and system, and relates to the technical field of lens design and customization. According to the dynamic myopia management lens prescription customization technology provided by the invention, ophthalmic basic data of a patient is acquired; then periodically follow-up the patient to obtain ophthalmology measurement follow-up data of the patient; the joint evaluation model dynamically updates a myopia risk evaluation grade of the patient based on ophthalmology basic data and ophthalmology measurement follow-up visit data; and finally, dynamically updating the lens prescription based on the myopia risk assessment level of each time by the joint assessment model. According to the technical scheme, the combined evaluation model can perform dynamic lens prescription adjustment according to follow-up visit ocular axis progress and choroid change conditions, dynamic myopia management lens prescription customization, whole-course tracking of a myopia cycle and individualized prevention and control management are realized, and thus the effectiveness and continuity of myopia prevention and control are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lens design and customization, in particular to a dynamic myopia management lens prescription customization method and system. BACKGROUND

[0002] Children with myopia are at high risk of pathological myopia, and it is estimated that about 10% of children with high myopia will develop pathological myopia with continuous growth of the eye axis and retinal lesions, which requires lifelong prevention and control.

[0003] At present, common myopia prevention methods mainly include: 1) by increasing outdoor activity time, using sunlight to stimulate the release of dopamine in the eyes, reducing the growth rate of the eye axis, and at the same time, broadening the field of vision to reduce close-range eye use; 2) in terms of medical means, by wearing a corneal molding lens at night to temporarily change the shape of the cornea, clear naked eye vision can be obtained during the day, thereby controlling the growth of adolescent myopia; 3) using low-concentration atropine eye drops and other liquid medicines to delay the progression of myopia to some extent.

[0004] However, due to differences in age, myopia, and living habits of different patients, the one-size-fits-all myopia prevention method at least has the following problems: 1) it cannot follow the principle of individualization; 2) it cannot dynamically and flexibly adjust the myopia prevention strategy according to the eye development of the patient (especially children) in a period; 3) the production of myopia prevention devices (such as lenses) lacks the participation of medical optometry professionals and personnel.

[0005] Therefore, it can be seen that the existing myopia prevention intervention methods lack dynamics, individualization, and specialization, and cannot provide precise prevention and control schemes for patients at different stages of myopia progression. SUMMARY

[0006] (I) Technical problems solved

[0007] In view of the deficiencies in the prior art, the present application provides a dynamic myopia management lens prescription customization method and system, which at least solves one or more problems mentioned in the background.

[0008] (II) Technical solutions

[0009] To achieve the above purpose, the present application is realized by the following technical solutions:

[0010] In a first aspect, the present application first proposes a dynamic myopia management lens prescription customization method, which comprises:

[0011] Obtaining patient ophthalmic basic data;

[0012] Periodically follow up the patient, obtain patient ophthalmic measurement follow-up data, upload patient ophthalmic basic data and patient ophthalmic measurement follow-up data to the management database;

[0013] Grab the patient-associated parameters in the management database, pre-construct a joint evaluation model for evaluating the myopia risk assessment level of the patient and dynamically updating the lens prescription, and train the pre-constructed joint evaluation model to convergence using the patient-associated parameters;

[0014] The joint evaluation model dynamically updates the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data;

[0015] Obtain the myopia risk assessment level of the patient, and the joint evaluation model dynamically updates the lens prescription in combination with the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data.

[0016] In one embodiment, the ophthalmic basic data includes corneal topography data and OCT data, and the ophthalmic measurement follow-up data includes patient axial length change data and choroid change data.

[0017] In one embodiment, the joint evaluation model training method comprises:

[0018] Traverse the management database to grab the patient-associated parameters for training the joint evaluation model, wherein the patient-associated parameters include patient ophthalmic basic data, patient ophthalmic measurement follow-up data, lens customization and / or treatment scheme and / or axial length change curve, use Latin hypercube sampling method to extract a modeling sample set, normalize and enhance the modeling sample set, and divide the modeling sample set into a training set and a test set;

[0019] Load the pre-constructed joint evaluation model, pre-set the feature fusion engine loss function, the double evaluation module loss function, and embed the joint loss function in combination with the feature fusion engine loss function and the double evaluation module loss function, pre-set the initial hyperparameters of the risk assessment block and the lens design block in the double evaluation module, and set the early stopping mechanism and the training rounds;

[0020] Obtain the training set, freeze the feature fusion engine, and pre-train the risk assessment block and the lens design block using the training set, respectively, and use the joint fine-tuning strategy to fine-tune the initial hyperparameters of the risk assessment block and the lens design block during pre-training;

[0021] After pre-training, the feature fusion engine and the double evaluation module are alternately trained using the proximal optimization, and the proximal regularizer is used to stabilize the training process, and the converged joint evaluation model is output;

[0022] obtaining a test set, taking the test set as input, executing the joint evaluation model, and outputting test results, wherein the test results include a test risk assessment level and a test lens prescription, and comparing the test risk assessment level in the test results with the true risk assessment, the test lens prescription and the true lens prescription respectively to obtain a risk assessment F1 score and a lens prescription MSE weighted average, and output a comprehensive test value;

[0023] determining whether the comprehensive test value exceeds a preset test threshold, and if the comprehensive test value exceeds the preset test threshold, outputting the converged joint evaluation model;

[0024] If the comprehensive test value does not exceed the preset test threshold, the joint evaluation model is optimized using Bayesian optimization combined with a multi-objective optimization algorithm, and the optimized joint evaluation model is iteratively trained until convergence.

[0025] In one embodiment, when the joint evaluation model is pre-constructed, the joint evaluation model takes a long short-term memory network model as an initial model, and the joint evaluation model further includes a multi-modal input layer and an output layer. The multi-modal input layer is connected to the long short-term memory network model, and the long short-term memory network model is connected to the output layer. The long short-term memory network model is improved by adding three convolutional blocks before the long short-term memory network model and introducing a feature fusion layer after the long short-term memory network model. The three convolutional blocks, the long short-term memory network model and the feature fusion layer are embedded into a feature fusion engine. The feature fusion layer includes an up-sampling layer, a fully connected layer one, a first CSP layer, a convolutional layer, a fully connected layer two, a second CSP layer and a down-sampling layer. A double evaluation module is arranged between the feature fusion engine and the output layer. The double evaluation module includes a risk assessment block and a lens design block. The risk assessment block includes a feature pyramid and a risk evaluation layer. The risk evaluation layer is a Transformer cross-modal attention layer, and a risk clustering analysis method is introduced in the risk evaluation layer. The lens design block includes a physical optical simulation layer, a virtual aberration layer, a lens parameter mapping network, a critical threshold determination layer and a rich optimization layer. The lens parameter mapping network includes a defocus amount generator, an optical optimizer and a microlens array generator. The physical optical simulation layer includes an optical meta-network, an ISP proxy module and a simulation task network.

[0026] In one embodiment, the method for optimizing the joint evaluation model using Bayesian optimization combined with a multi-objective optimization algorithm includes:

[0027] A double-objective optimization function is constructed with risk assessment accuracy and lens prescription accuracy as constraints, and decision variables of the double-objective function are defined. The decision variables include input parameters and output parameters. The input parameters are axial change rate and choroidal thinning rate, and the output parameters include defocus amount, optical zone diameter, microlens density and asymmetry factor.

[0028] Load the pre-constructed dual-objective optimization function, and preset the iteration number, exploration coefficient and search space of the Bayesian optimizer, initialize the Bayesian optimizer, use the proxy model and dual-objective optimization function of the Bayesian optimizer to select the next most promising super parameter combination, update the proxy model with the newly observed data, and obtain the super parameter updated proxy model;

[0029] Initialize the population size, crossover probability and mutation probability parameters of the multi-objective optimization algorithm NSGA-II, select individuals for crossover operation according to non-dominated sorting and crowding distance, and perform mutation operation on the newly generated individuals, and select the optimal solution from the Pareto front;

[0030] With the optimal solution as the physical feasibility constraint, process the physical constraint through the feasibility rule, and optimize the joint evaluation model super parameter, and output the super parameter optimized joint evaluation model.

[0031] In one embodiment, the method for dynamically updating the myopia risk assessment level of a patient based on the ophthalmic basic data and ophthalmic measurement follow-up data includes:

[0032] Load the ophthalmic basic data and ophthalmic measurement follow-up data, input the ophthalmic basic data and ophthalmic measurement follow-up data into the multi-modal input layer, normalize and reduce the dimension of the ophthalmic basic data and ophthalmic measurement follow-up data in the multi-modal input layer, and perform cross-modal alignment processing on the ophthalmic basic data and ophthalmic measurement follow-up data, and output the cross-modal alignment set;

[0033] Obtain the cross-modal alignment set, and the feature fusion engine extracts the features in the cross-modal alignment set, and outputs the feature fusion set;

[0034] Load the feature fusion set, and the feature pyramid further extracts the feature fusion set, extracts the risk impact factors and impact weights in the feature fusion set;

[0035] Determine whether the impact weight of the risk impact factor exceeds the preset impact threshold, if it exceeds the preset impact threshold, retain the current risk impact factor, and stack the risk impact factor to the feature fusion set;

[0036] Based on K-means, the risk impact factors in the feature fusion set are clustered, and the risk feature clustering results are imported into the Transformer cross-modal attention layer as prior knowledge;

[0037] The Transformer cross-modal attention layer performs cross-modal fusion on the feature fusion set, and calculates the myopia risk assessment level of the patient based on the prior knowledge risk feature clustering results.

[0038] In one embodiment, the joint evaluation model combines myopia risk assessment level, ophthalmic basic data and ophthalmic measurement follow-up data to dynamically update the lens prescription method, which comprises:

[0039] Obtain the myopia risk assessment level, ophthalmic basic data and ophthalmic measurement follow-up data, and the photophysical optical simulation layer generates a simulated optical model matched with the patient's eye based on the ophthalmic basic data and the ophthalmic measurement follow-up data;

[0040] Load the simulated optical model, simulate the phase difference existing in the patient's eye based on the virtual aberration layer, and correct the simulated optical model with the phase difference as the constraint;

[0041] With the myopia risk assessment level as the physical constraint, the defocus amount generator optimizes the simulated optical model based on the curvature optimization algorithm, determines the defocus amount distribution of the lens, and outputs the defocus amount distribution diagram;

[0042] Determine the position and width of the progressive channel of the lens using the optical optimizer, and perform three-dimensional environment modeling based on the position and width of the progressive channel of the lens to simulate and optimize the progressive channel design of the lens;

[0043] Obtain the simulated optical model, generate a microlens array with asymmetric gradient density distribution based on the simulated optical model, and optimize the arrangement and shape of the microlens using a genetic algorithm;

[0044] Obtain the defocus amount distribution diagram, the simulated and optimized progressive channel design of the lens, the arrangement and shape of the microlens, and identify the design lens parameters based on the defocus amount distribution diagram, the simulated and optimized progressive channel design of the lens, the arrangement and shape of the microlens. Map the design lens parameters to the lens parameter mapping network;

[0045] The critical threshold determination layer identifies the design lens parameters in the lens parameter mapping network, adjusts the critical threshold of the design lens parameters based on the myopia risk assessment level of the physical constraint, optimizes the design lens parameters in the lens parameter mapping network through the critical threshold, and integrates the design lens parameters after the parameter optimization processing. Generate feedback optical quality indicators through the design lens parameters;

[0046] The rich optimization layer responds to the feedback optical quality indicators, executes the design lens parameter execution instructions based on the feedback optical quality indicators, and generates a dynamically updated lens prescription.

[0047] In one embodiment, the photophysical optical simulation layer generates a simulated optical model matched with the patient's eye based on ophthalmic basic data and ophthalmic measurement follow-up data, which comprises:

[0048] Extract the patient's eye optical feature vector based on the ophthalmic basic data and the ophthalmic measurement follow-up data, and the optical element network constructs the spatially varying point spread function of the simulated simulation optical system based on the eye optical feature vector;

[0049] generating an initial optical model matching the patient's eye optical feature vector through a spatially varying point spread function;

[0050] The ISP proxy module performs image signal processing proxy operations on the initial optical model, simulates ISP effects in the real imaging process, and performs color correction and denoising processing on the initial optical model to output a corrected optical model;

[0051] The simulation task network corrects the defocus uniformity of the corrected optical model in combination with the myopia risk assessment level to generate a simulated optical model matching the patient's eye.

[0052] In one embodiment, the method further comprises visualizing the ophthalmic basic data and / or ophthalmic measurement follow-up data of the patient;

[0053] visualizing and / or generating lens design drawings based on the lens prescription;

[0054] automatically generating a prevention and control effect tracking report based on the ophthalmic basic data and ophthalmic measurement follow-up data;

[0055] automatically or based on a user request, pushing a lens customization and / or treatment plan and / or eye axis change curve;

[0056] automatically or based on a user request, pushing a follow-up visit reminder to the patient;

[0057] automatically or based on a user request, pushing a myopia prevention and control teaching video to the patient.

[0058] In a second aspect, the present application further provides a dynamic myopia management lens prescription customization system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, for implementing the steps of the dynamic myopia management lens prescription customization method as described above.

[0059] (III) Beneficial effects

[0060] The present application provides a dynamic myopia management lens prescription customization method and system. Compared with the prior art, the following beneficial effects are achieved:

[0061] The application provides a dynamic myopia management lens prescription customization method, which comprises the following steps: acquiring ophthalmic basic data of a patient; regularly following up the patient to acquire ophthalmic measurement follow-up data of the patient; dynamically updating a myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data by a joint assessment model; and dynamically updating a lens prescription based on the myopia risk assessment level each time by the joint assessment model. According to the application, the joint assessment model can dynamically adjust the lens prescription according to the follow-up axial progression and choroidal change, realizes dynamic myopia management lens prescription customization, and realizes whole-process tracking and individualized prevention and control management of the myopia period, so that the effectiveness and continuity of myopia prevention and control are improved. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0063] Figure 1 The flowchart of the dynamic myopia management lens prescription customization method in the embodiments of the present application is shown.

[0064] Figure 2 The implementation flowchart of the joint assessment model training method in the embodiments of the present application is shown.

[0065] Figure 3 The implementation flowchart of the joint assessment model optimization method using Bayesian optimization combined with a multi-objective optimization algorithm in the embodiments of the present application is shown.

[0066] Figure 4 The implementation flowchart of the method for dynamically updating the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data by the joint assessment model in the embodiments of the present application is shown.

[0067] Figure 5 The implementation flowchart of the method for dynamically updating the lens prescription by the joint assessment model combined with the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data in the embodiments of the present application is shown.

[0068] Figure 6 The implementation flowchart of the method for generating a simulated optical model matched with the patient's eye based on the ophthalmic basic data and the ophthalmic measurement follow-up data by the photophysical optical simulation layer in the embodiments of the present application is shown.

[0069] Figure 7 The visualization interface of the axial length trend graph in the embodiments of the present application is shown.

[0070] Figure 8 The visualization interface of the lens based on the lens prescription in the embodiment of the application is shown.

[0071] Figure 9 The nearsightedness prevention and control teaching video visualization interface in the embodiment of the application is shown. DETAILED DESCRIPTION

[0072] In order to make the objects, technical solutions and advantages of the embodiments of the application clearer, the technical solutions in the embodiments of the application are described clearly and completely. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0073] The embodiment of the application provides a dynamic nearsightedness management lens prescription customization method and system, which solves the problem that the existing nearsightedness prevention and control cannot provide accurate prevention and control schemes for patients in different nearsightedness progress stages, and achieves the purpose of dynamic lens prescription customization of "one person, one strategy".

[0074] The technical solution in the embodiment of the application is as follows to solve the above technical problems:

[0075] In order to solve the above problems of the prior art, the dynamic nearsightedness management lens prescription customization technology provided by the application first acquires patient ophthalmic basic data including corneal topographic data and OCT data, then acquires patient eyeball parameters including axial progress and choroid changes and time-series follow-up data of nearsightedness prevention and control progress according to a preset period, then generates a nearsightedness risk assessment level of the patient based on the above ophthalmic basic data and time-series follow-up data, and finally generates a lens prescription corresponding to the nearsightedness risk assessment level of the patient. The technical solution of the application realizes dynamic nearsightedness management lens prescription customization of "one person, one strategy", performs whole-process tracking and individualized prevention and control management in a nearsightedness period, and improves the effectiveness and continuity of nearsightedness prevention and control.

[0076] It should be noted that the user of the dynamic nearsightedness management lens prescription customization technology provided by the embodiment of the application includes but is not limited to an ophthalmologist, and can also be personnel of other medical optometry professional teams, etc. The related description of the use object in the following embodiments is not regarded as a limitation on the protection scope of the application.

[0077] In order to better understand the above technical solution, the above technical solution will be described in detail in combination with the drawings in the specification and specific implementation manners.

[0078] Embodiment 1:

[0079] In a first aspect, the present application first proposes a dynamic myopia management lens prescription customization method, as shown in Figure 1 The method comprises:

[0080] S1, obtaining patient ophthalmic basic data, wherein the ophthalmic basic data comprises but is not limited to corneal topography data and OCT data;

[0081] S2, regularly following up the patient to obtain patient ophthalmic measurement follow-up data, and uploading the patient ophthalmic basic data and the patient ophthalmic measurement follow-up data to a management database;

[0082] S3, capturing patient associated parameters in the management database, pre-constructing a joint evaluation model for evaluating a myopia risk assessment level of the patient and dynamically updating a lens prescription, and training the pre-constructed joint evaluation model to convergence using the patient associated parameters;

[0083] S4, dynamically updating the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data by the joint evaluation model;

[0084] S5, obtaining the myopia risk assessment level of the patient, and dynamically updating the lens prescription by the joint evaluation model in combination with the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data.

[0085] The dynamic myopia management lens prescription customization method proposed in the embodiment first obtains patient ophthalmic basic data, then regularly follows up the patient to obtain patient ophthalmic measurement follow-up data, dynamically updates the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data, and finally dynamically updates the lens prescription based on the myopia risk assessment level each time. The technical solution proposed in the embodiment can dynamically update the lens prescription by the joint evaluation model in combination with the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data, and realize individualized dynamic myopia management lens prescription customization for different patients.

[0086] The implementation process of one embodiment of the present application will be described in detail below in combination with the accompanying Figures 1-9 and specific explanations of steps S1-S5.

[0087] S1, obtaining patient ophthalmic basic data, wherein the ophthalmic basic data comprises but is not limited to corneal topography data and OCT data;

[0088] In order to preliminarily understand the myopia condition of the patient and better grasp the myopia condition of the patient, in an embodiment, the ophthalmic basic data of the patient including the corneal topography data and the OCT data is selected to be obtained and inputted.

[0089] Corneal topography is a technique that measures the surface of the cornea by taking multiple points of measurement, obtaining information such as the curvature and height of the anterior and posterior surfaces of the cornea, and presenting these data in the form of a topographic map with color coding. Corneal topography data has high resolution and strong visuality, and can comprehensively and accurately display the morphological characteristics of the corneal surface. It can clearly show the type of corneal astigmatism, the axis of corneal astigmatism, and the irregularity of the corneal surface, etc.

[0090] OCT (Optical Coherence Tomography) is a technique that uses the principle of light interference to obtain tomographic images of the eye by measuring the interference signals of reflected light. The OCT system emits a beam of low coherence light, which is reflected when it encounters tissue interfaces with different refractive indices. By detecting the interference signals of the reflected light, the optical path difference is calculated, and the depth information of the eye tissue is obtained, generating high-resolution two-dimensional or three-dimensional images. OCT data can provide tomographic images of the eye tissue, clearly showing the fine structure of each layer of tissue in the eye, with high resolution and sensitivity.

[0091] In addition, the patient's ophthalmic basic data also includes the axial length of the left and right eyes (mm) under the current examination state, the refractive power of the left and right eyes (D), and the astigmatism of the left and right eyes, etc.

[0092] S2, regularly follow up the patient, obtain the patient's ophthalmic measurement follow-up data, and upload the patient's ophthalmic basic data and the patient's ophthalmic measurement follow-up data to the management database.

[0093] In the prior art, refractive power detection is usually used to judge the myopia of the patient, but false myopia may occur in refractive power detection, and only relying on refractive power detection cannot accurately reflect the changes in the patient's eyeball parameters, the progress and effectiveness of myopia prevention and control, etc. In order to more accurately monitor the myopia prevention and control of the patient, in this embodiment, the axial progression and choroidal changes of the patient are monitored to reflect the myopia of the patient.

[0094] Axial length and choroid play a key role in the occurrence and development of myopia, and changes in the two will directly or indirectly affect the refractive state of the eyeball and the health of the eye. Among them:

[0095] Axial length represents the length of the eyeball from front to back. Under normal physiological conditions, the axial length will gradually develop with age, but if the axial length develops excessively or grows too fast, it will cause myopia. Generally speaking, for every 1mm of axial length, the myopia degree will increase by about 200-300 degrees. In addition, axial progression reflects true myopia, which is the latest standard for judging myopia, while refractive power detection may result in false myopia. Therefore, in this embodiment, regular monitoring of the patient's axial data can reflect the approximate situation of the patient.

[0096] The choroid is located between the retina and the sclera, and is rich in blood vessels, providing nutrition and oxygen to the retina. Changes in choroidal thickness are closely related to the development of myopia, and it can affect the normal metabolism and function of the retina, thereby promoting the progression of myopia. In myopic patients, especially in high myopic patients, the choroidal thickness will be thinner. When the choroid is thin, its support and nutrition supply to the retina will decrease, and the structure and function of the retina may be damaged, further affecting the refractive state of the eyeball. Therefore, in the embodiment, the approximate condition of the patient can be reflected by regularly monitoring the axial data of the patient.

[0097] In a preferred embodiment, the time interval for following up the patient is set to 3-6 months, that is, the patient's axial progression and choroidal changes are followed up every 3-6 months, and the ophthalmic measurement follow-up data is recorded.

[0098] S3, grab the patient associated parameters in the management database, pre-construct a joint evaluation model for evaluating the myopia risk assessment level of the patient and dynamically updating the lens prescription, and train the pre-constructed joint evaluation model to convergence using the patient associated parameters.

[0099] In the embodiment of the application, as shown in Figure 2 The joint evaluation model training method comprises the following steps:

[0100] S31, traverse the management database to grab the patient associated parameters for training the joint evaluation model, wherein the patient associated parameters comprise patient ophthalmic basic data, patient ophthalmic measurement follow-up data, lens customization and / or treatment scheme and / or axial change curve, use Latin hypercube sampling method to extract a modeling sample set, normalize and enhance the modeling sample set, and divide the modeling sample set into a training set and a test set, wherein the ratio of the training set and the test set is 3:1 or 4:1.

[0101] When the Latin hypercube sampling method is used to extract the modeling sample set, in order to ensure that each sample point is not repeated, the sampling formula is represented as:

[0102]

[0103] Wherein, i is the index of the sample point, N is the number of samples, d represents the dimension index, a represents the hypercube sampling coefficient, and A is the number of parameter types.

[0104] S32, load the pre-constructed joint evaluation model, preset the feature fusion engine loss function, the double evaluation module loss function, and combine the feature fusion engine loss function, the double evaluation module loss function to embed the joint loss function, preset the initial hyperparameters of the risk evaluation block and the lens design block in the double evaluation module, and set the early stopping mechanism and the training round, wherein the initial hyperparameters of the risk evaluation block and the lens design block can be set to 0.02 and 0.001, and the early stopping mechanism can be that the joint loss function does not decrease for 10 rounds of verification, and the training round can be 150-180 times;

[0105] wherein the joint loss function is represented as:

[0106] L total =λ1L f +λ2L c +λ3L t

[0107] λ1+λ2+λ3=1

[0108] wherein L total represents the joint loss function, L f , L c , and L t represent the risk constraint loss function, the optical design performance loss function, and the feature fusion engine loss function, respectively, the risk constraint loss function can be a cross-entropy loss function, and the optical design performance loss function and the feature fusion engine loss function can be a contrast loss function and a center loss function, respectively. In this embodiment, λ1, λ2, and λ3 are the weight coefficients of the risk constraint loss function, the optical design performance loss function, and the feature fusion engine loss function, respectively. In a preferred embodiment, they can be set to 0.5, 0.2, and 0.3, respectively.

[0109] S33, obtain the training set, freeze the feature fusion engine, and pre-train the risk evaluation block and the lens design block using the training set, and use the joint fine-tuning strategy to fine-tune the initial hyperparameters of the risk evaluation block and the lens design block during pre-training.

[0110] S34, after pre-training, use alternating proximal optimization to jointly train the feature fusion engine and the double evaluation module, and use proximal regularizer to stabilize the training process, and output the converged joint evaluation model.

[0111] S35, obtain the test set, input the test set, execute the joint evaluation model, and output the test result, wherein the test result includes the test risk evaluation level and the test lens prescription, and the test risk evaluation level in the test result is compared with the real risk evaluation, the test lens prescription, and the real lens prescription, respectively, to obtain the risk evaluation F1 score and the lens prescription MSE weighted average, and output the comprehensive test value.

[0112] S36, judge whether the comprehensive test value exceeds the preset test threshold, wherein in an embodiment, the preset test threshold can be 0.9-0.95, if the comprehensive test value exceeds the preset test threshold, output the converged joint evaluation model.

[0113] S37, if the comprehensive test value does not exceed the preset test threshold, use Bayesian optimization combined with multi-objective optimization algorithm to optimize the joint evaluation model, return to S13, continue to iteratively train the optimized joint evaluation model to convergence; if the comprehensive test value exceeds the preset test threshold, output the converged joint evaluation model.

[0114] It should be noted that when the joint evaluation model is pre-constructed, the joint evaluation model takes a long short-term memory network model as an initial model, and the joint evaluation model further includes a multi-modal input layer and an output layer. The multi-modal input layer is connected with the long short-term memory network model, and the long short-term memory network model is connected with the output layer. The long short-term memory network model is improved by adding 3 convolutional blocks in front of the long short-term memory network model and introducing a feature fusion layer after the long short-term memory network model. The 3 convolutional blocks, the long short-term memory network model, and the feature fusion layer are embedded into a feature fusion engine. The feature fusion layer includes an up-sampling layer, a full connection layer one, a first CSP layer, a convolutional layer, a full connection layer two, a second CSP layer, and a down-sampling layer. A double evaluation module is arranged between the feature fusion engine and the output layer. The double evaluation module includes a risk evaluation block and a lens design block. The risk evaluation block includes a feature pyramid and a risk evaluation layer. The risk evaluation layer is a Transformer cross-modal attention layer, and a risk clustering analysis method is introduced in the risk evaluation layer. The lens design block includes a physical optical simulation layer, a virtual aberration layer, a lens parameter mapping network, a critical threshold determination layer, and a rich optimization layer. The lens parameter mapping network includes a defocus amount generator, an optical optimizer, and a microlens array generator. The physical optical simulation layer includes an optical meta network, an ISP proxy module, and a simulation task network. The physical optical simulation layer in the lens design block is used to simulate the spatially varying point spread function (PSF) of the composite optical system, thereby providing an optical basis for lens design. This simulation technology can consider the optical characteristics of the patient's eyeball, generate an optical model that matches the patient's eye condition, and improve the scientificity and accuracy of lens design.

[0115] In the embodiment of the present application, the joint evaluation model integrates corneal topography, OCT image, axial length, diopter, astigmatism, axial length change data, choroid change data and other basic ophthalmic data and follow-up data. The fusion of such multi-modal data enables the model to comprehensively evaluate the myopia risk and eye condition of the patient from multiple angles, and the risk evaluation block can dynamically update the myopia risk evaluation level according to the latest ophthalmic data of the patient. Such dynamic evaluation method can timely reflect the myopia progression of the patient, and provide more accurate decision support for clinicians, and the physical optical simulation layer in the lens design block simulates the spatial variation point spread function (PSF) of the compound optical system, providing an optical basis for lens design. Such simulation technology can consider the optical properties of the patient's eyeball, generate an optical model that matches the patient's eye condition, and improve the scientificity and accuracy of lens design.

[0116] As Figure 3 shown, the embodiment of the present application also provides a joint evaluation model optimization method using Bayesian optimization combined with a multi-objective optimization algorithm, which specifically comprises:

[0117] S371, taking the risk assessment accuracy and the lens prescription accuracy as constraints, constructing a double-objective optimization function, defining the decision variables of the double-objective function, the decision variables including input parameters and output parameters, wherein the input parameters are the axial length change rate and the choroid thinning rate, and the output parameters include the defocus amount, the optical zone diameter, the miotic lens density and the asymmetry factor.

[0118] It should be noted that the traditional single-objective optimization can only optimize a single index (only improve the risk assessment accuracy or only reduce the lens cost), but in the embodiment of the present application, the joint evaluation model needs to balance the two conflicting objectives of risk assessment accuracy and lens prescription accuracy, and the multi-objective conflict can be converted into a mathematical optimization problem by constructing a double-objective optimization function.

[0119] The decision variable is represented as:

[0120]

[0121] The objective function is represented as:

[0122]

[0123] Wherein, ΔL, dC dt , ΔJ, φ, ρ t , κ represent the axial length change rate, the choroid thinning rate, the defocus amount, the optical zone diameter, the miotic lens density and the asymmetry factor, respectively, and s1, s2 are the risk assessment accuracy constraint function and the lens prescription accuracy constraint function, respectively.

[0124] S372, load the pre-constructed double-target optimization function, and preset the iteration number, exploration coefficient and search space of the Bayesian optimizer, initialize the Bayesian optimizer, select the next most promising hyperparameter combination using the proxy model of the Bayesian optimizer and the double-target optimization function, update the proxy model using the newly observed data, and obtain the updated proxy model of the hyperparameters. It should be noted that the number of hyperparameters of the joint evaluation model is large and the dimension is high, and the traditional grid search needs to evaluate multiple groups of parameters, which makes the calculation cost extremely high, while in the present application, the proxy model of the Bayesian optimizer can be fitted by Gaussian process (GP) or random forest (RF) to historical evaluation data, thereby constructing a probabilistic proxy model of the objective function.

[0125] S373, initialize the population size, crossover probability and mutation probability parameters of the multi-objective optimization algorithm NSGA-II, select individuals for crossover operation according to non-dominated sorting and crowding distance, and perform mutation operation on the newly generated individuals, and select the optimal solution from the Pareto front, wherein the Pareto front generated by NSGA-II contains different hyperparameter combinations, thereby covering the needs of different clinical scenarios, and further enhancing the generalization ability of the model.

[0126] S374, taking the optimal solution as the physical feasibility constraint, processing the physical constraint through the feasibility rule, and optimizing the hyperparameters of the joint evaluation model, outputting the joint evaluation model after hyperparameter optimization.

[0127] In the embodiments of the present application, the Bayesian optimization combined with the multi-objective optimization algorithm can more accurately adjust the hyperparameters of the model, so that the F1 score and other indicators of the risk assessment block when evaluating the myopia risk level are significantly improved, which helps to more accurately identify the myopia risk of patients and provides more reliable basis for clinical decision-making. In the optimization process, the model can dynamically adjust the hyperparameters of risk assessment and lens design according to the input parameters, and the dynamic adjustment capability makes the model better adapt to the myopia progression of different patients at different stages, providing personalized risk assessment and lens design services for patients.

[0128] S4, the joint evaluation model dynamically updates the myopia risk assessment level of the patient based on the ophthalmic basic data and ophthalmic measurement follow-up data.

[0129] First, according to the first ophthalmic basic data and the first ophthalmic measurement follow-up data, the joint evaluation model can determine the myopia risk assessment level of the patient, and then based on each ophthalmic measurement follow-up data, continuously update the myopia risk assessment level of the patient at the current data measurement time.

[0130] As shown in Figure 4 the method for dynamically updating the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data includes:

[0131] S41, load the ophthalmic basic data and the ophthalmic measurement follow-up data, input the ophthalmic basic data and the ophthalmic measurement follow-up data into a multi-modal input layer, and the multi-modal input layer normalizes and reduces the dimension of the ophthalmic basic data and the ophthalmic measurement follow-up data, wherein the normalization processing mode can be a maximum-minimum normalization processing, the dimension reduction processing can be a principal component analysis dimension reduction processing, and the ophthalmic basic data and the ophthalmic measurement follow-up data are cross-modally aligned, the cross-modal alignment processing mode can be a timestamp-based or anatomical structure-based spatial registration, and an aligned cross-modal set is output.

[0132] S42, obtain the aligned cross-modal set, and a feature fusion engine extracts features from the aligned cross-modal set, and outputs a feature fusion set.

[0133] In the embodiment of the application, when the feature fusion set is generated: the 3-layer convolution block extracts the local curvature anomaly of the corneal topography and the inter-retinal layer structure anomaly of the OCT through different receptive fields (the sizes of the convolution kernels of the 3-layer convolution block are 3x3, 5x5 and 7x7 kernels), and the long short-term memory network model can model the monthly / quarterly change trend of the axial length, and the feature fusion set covering complete features is obtained after the fusion of the two.

[0134] S43, load the feature fusion set, and a feature pyramid further extracts the risk influencing factors and the influence weights in the feature fusion set.

[0135] S44, determine whether the influence weight of the risk influencing factor exceeds a preset influence threshold.

[0136] S45, if the preset influence threshold is exceeded, the current risk influencing factor is retained, and the risk influencing factor is stacked to the feature fusion set.

[0137] S46, if the preset influence threshold is not exceeded, wherein the influence threshold can be 0.5-0.6, the current risk influencing factor is discarded, the high-weight factor is retained, the risk influencing factor highly related to the progression of myopia is screened, and the sensitivity of the model to the core driving factor is improved.

[0138] S47, risk feature clustering is performed on the risk influencing factors in the feature fusion set based on K-means, and the risk feature clustering result is imported into a Transformer cross-modal attention layer as prior knowledge.

[0139] S48, the Transformer cross-modal attention layer performs cross-modal fusion on the feature fusion set, combines the prior knowledge risk feature clustering result to calculate the myopia risk assessment level of the patient, and updates the calculated myopia risk assessment level to the myopia risk assessment level of the patient at the current data measurement time; wherein the Transformer cross-modal attention layer can perform cross-modal fusion on the feature fusion set, fully consider the correlation and complementary information between different modal data, combine the prior knowledge risk feature clustering result, and more accurately calculate the myopia risk assessment level of the patient, update the calculated myopia risk assessment level to the level at the current data measurement time of the patient, realize dynamic tracking and real-time updating of the myopia risk of the patient, can timely reflect the myopia progress of the patient, provide more timely and accurate decision support for clinicians, and help better manage and control the myopia development of the patient.

[0140] The myopia risk assessment level is calculated by the following formula:

[0141]

[0142] Wherein Q, K, and V represent the feature fusion set input representation, the risk feature clustering center feature vector, and the historical risk feature vector corresponding to the risk feature clustering center, respectively, d k represents the risk feature clustering center feature vector dimension, QK T represents the matrix multiplication of query and key.

[0143] In the embodiment of the application, the feature fusion engine extracts features from the cross-modal aligned data, which can integrate key information in different modal data, thereby generating a more representative and comprehensive feature set. These features can better reflect the overall condition of the patient's eye and provide more information for risk assessment. The output feature fusion set as the input of the subsequent model can make the model better understand and learn the complex features of the patient's eye compared to single modal data or unmerged multi-modal data, thereby improving the accuracy of risk assessment. Combined with the prior knowledge risk feature clustering result, the myopia risk assessment level of the patient can be more accurately calculated, and the calculated myopia risk assessment level is updated to the level at the current data measurement time of the patient, realizing dynamic tracking and real-time updating of the myopia risk of the patient, which can timely reflect the myopia progress of the patient, provide more timely and accurate decision support for clinicians, and help better manage and control the myopia development of the patient.

[0144] In one embodiment, the myopia risk assessment level can be determined according to the output result of the Transformer cross-modal attention layer.

[0145] In a preferred embodiment, the correspondence between the output of the Transformer cross-modal attention layer and the myopia risk assessment level of the patient is set. For example, when the output of the Transformer cross-modal attention layer x < 0.1, it can be determined as low risk; when the output of the Transformer cross-modal attention layer 0.1 ≤ x < 0.5, it can be determined as medium risk, and when the output of the Transformer cross-modal attention layer x ≥ 0.5, it can be determined as high risk.

[0146] S5, obtaining the myopia risk assessment level of the patient, the joint assessment model combines the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data to dynamically update the lens prescription.

[0147] As shown in Figure 5 The embodiment of the present application also provides a method for dynamically updating a lens prescription by a joint assessment model combining a myopia risk assessment level, ophthalmic basic data and ophthalmic measurement follow-up data, and the method specifically comprises the following steps:

[0148] S51, obtaining the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data, and the photophysical optical simulation layer generates a simulated optical model matched with the patient's eye based on the ophthalmic basic data and the ophthalmic measurement follow-up data, wherein the simulated optical model can accurately reflect the optical characteristics of the patient's eye, providing a reliable basis for subsequent optical optimization and lens design, and improving the scientificity and accuracy of lens design;

[0149] S52, loading the simulated optical model, simulating the phase difference existing in the patient's eye based on the virtual aberration layer, and correcting the simulated optical model with the phase difference as a constraint, wherein the virtual aberration layer can simulate the phase difference existing in the patient's eye, more truly reflect the optical environment of the patient's eye, make the designed lens better correct the patient's vision problem, and improve the visual quality. Correcting the simulated optical model with the phase difference as a constraint can correct the error in the model, further improve the accuracy of the simulated optical model, and provide more accurate optical parameters for subsequent lens design;

[0150] S53, taking the myopia risk assessment level as a physical constraint, the defocus amount generator optimizes the simulation optical model based on a curvature optimization algorithm to determine the defocus amount distribution of the lens and output a defocus amount distribution map, wherein taking the myopia risk assessment level as a physical constraint can ensure that the lens design matches the myopia risk condition of the patient, making the designed lens more targeted in controlling myopia progression. The defocus amount generator optimizes the simulation optical model based on a curvature optimization algorithm to accurately determine the defocus amount distribution of the lens and output a defocus amount distribution map, providing key optical parameters for subsequent lens design, which helps to achieve retinal defocus control and delay myopia progression;

[0151] S54, the optical optimizer is used to determine the position and width of the progressive channel of the lens, and a three-dimensional environment model is established based on the position and width of the progressive channel of the lens, the simulation optimization lens progressive channel design is simulated, and the progressive channel design is optimized based on the simulation optical model, which can fully consider the eye features and myopia risk level of the patient, realize personalized lens design, and meet the needs of different patients;

[0152] S55, the simulation optical model is obtained, the microlens array with asymmetric gradient density distribution is generated based on the simulation optical model, and the arrangement and shape of the microlens are optimized using a genetic algorithm, which can find better microlens design parameters to further improve the optical performance and visual quality of the lens and provide better visual experience for the patient;

[0153] S56, the defocus amount distribution map, the simulation optimized lens progressive channel design, and the arrangement and shape of the microlens are obtained, and the design lens parameters are identified based on the defocus amount distribution map, the simulation optimized lens progressive channel design, and the arrangement and shape of the microlens, and the design lens parameters are mapped to a lens parameter mapping network;

[0154] S57, the critical threshold determination layer identifies the design lens parameters in the lens parameter mapping network, adjusts the critical threshold of the design lens parameters based on the physical constraint myopia risk assessment level, optimizes the design lens parameters in the lens parameter mapping network through the critical threshold, integrates the design lens parameters after the parameter optimization processing, generates a feedback optical quality index through the design lens parameters, and needs to be explained that adjusting the critical threshold of the design lens parameters based on the physical constraint myopia risk assessment level can ensure that the designed lens parameters meet the actual requirements of physical optics and match the myopia risk condition of the patient, improving the feasibility and effectiveness of lens design. And it can further improve the accuracy and quality of lens design, ensure the scientificity and reliability of lens design;

[0155] S58, the rich optimization layer responds to the feedback optical quality index, executes the design lens parameter execution instruction based on the feedback optical quality index, and generates a dynamically updated lens prescription.

[0156] In the embodiments of the present application, when the joint evaluation model combines the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data to dynamically update the lens prescription, the process of dynamic development of myopia can be considered, and the problem that the traditional lens cannot adjust the parameters with the progress of myopia due to the use of static template design can be overcome. The present method realizes the synchronous update of the lens prescription and the eye state of the patient through three-dimensional environment modeling and dynamic parameter optimization, can cover more people, and thus better serves different people, avoids the one-sidedness of relying only on the axial length in the traditional evaluation, and avoids visual fatigue caused by excessive intervention, thereby ensuring that the progress of myopia of patients with different risk levels is effectively prevented and controlled.

[0157] As shown in Figure 6 The method for generating the simulated optical model matched with the patient's eye based on the ophthalmic basic data and the ophthalmic measurement follow-up data includes:

[0158] S511, extracting the eyeball optical feature vector of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data, and constructing the spatially varying point spread function of the simulated optical system by the optical meta-network based on the eyeball optical feature vector;

[0159] S512, generating the initial optical model matched with the eyeball optical feature vector of the patient by the spatially varying point spread function;

[0160] S513, performing image signal processing proxy operation on the initial optical model by the ISP proxy module, simulating the ISP effect in the real imaging process, and performing color correction and denoising processing on the initial optical model, to output the corrected optical model. Through color correction and denoising processing, the image quality of the generated optical model can be significantly improved, noise and color deviation can be reduced, and the model can more accurately reflect the optical characteristics of the patient's eye, providing a clearer and more accurate image basis for subsequent defocus correction;

[0161] S514, performing defocus uniformity correction on the corrected optical model by the simulation task network combined with the myopia risk assessment level, to generate the simulated optical model matched with the patient's eye.

[0162] In the embodiments of the present application, the defocus correction combined with the myopia risk assessment level can ensure that the corrected optical model not only matches the patient in terms of optical characteristics, but also meets the needs of myopia prevention and control. The correction method based on the risk level can make the lens design more targeted and improve the prevention and control effect. After the defocus uniformity correction, the generated simulated optical model can more accurately match the optical characteristics of the patient's eye, provide a high-quality optical basis for subsequent lens design, and ensure that the designed lens can better adapt to the specific needs of the patient, thereby improving the effect and sustainability of myopia prevention and control.

[0163] The joint evaluation model combines myopia risk assessment levels, ophthalmic basic data, and ophthalmic measurement follow-up data to dynamically update lens prescriptions. The lens prescriptions include setting prescription types, lens parameters, etc.

[0164] Specifically, the prescription types include basic version, advanced version, flagship version, etc. Among them:

[0165] The basic version is suitable for children with mild to moderate myopia, and has a basic myopia control function. Suitable patients: mild myopia (-1.00D to -3.00D); functional design includes: standard micro-lens design, conventional micro-lens spacing, basic myopia control effect, etc.

[0166] The advanced version is suitable for children with moderate to high myopia, and has enhanced myopia control effect. Suitable patients: moderate myopia (-3.00D to -6.00D); functional design includes: enhanced micro-lens design, astigmatism compensation function, larger optical zone diameter, etc.

[0167] The flagship version is suitable for children with high myopia or rapid progression, and has the best myopia control effect. Suitable patients: high myopia (-6.00D or more) or rapid progression patients; functional design includes: advanced micro-lens design, non-periodic micro-lens arrangement, transparent acid gradient control, high-order aberration compensation, etc.

[0168] The lens parameters include myopia control strength, optical quality score, defocus amount, optical zone diameter, micro-lens type, micro-lens number, etc.

[0169] In one embodiment, the method further comprises:

[0170] S61, visualizing the ophthalmic basic data and / or ophthalmic measurement follow-up data of the patient.

[0171] In order to facilitate ophthalmologists and other medical and optometric professional team personnel to timely view the myopia situation and myopia prevention and follow-up situation of the patient, including but not limited to visualizing the ophthalmic basic data and / or ophthalmic measurement follow-up data of the patient, for example, feeding back the patient's eye axis length trend through a line chart, etc., as shown in Figure 7 Figure 7 The eye axis length trend chart visualization interface in the embodiment of the application is shown.

[0172] In one embodiment, the method further comprises:

[0173] S62, visualizing the lens and / or generating lens design drawings based on the lens prescription.

[0174] In order to facilitate ophthalmologists, other medical and optometric professional team personnel, lens manufacturing personnel to view, modify, and produce lenses, visualize the lens based on the lens prescription of the patient, and generate corresponding lens design drawings, as shown in​Figure 8 As shown.

[0175] In one embodiment, the method further comprises:

[0176] S63, automatically generating a prevention and control effect tracking report based on the ophthalmic basic data and ophthalmic measurement follow-up data.

[0177] In order for ophthalmologists and other medical and optometric professional team personnel to trace back and view the myopia prevention and control data and effect of patients, based on the ophthalmic basic data and ophthalmic measurement follow-up data, a prevention and control effect tracking report is automatically generated on the doctor side, for ophthalmologists and other medical and optometric professional team personnel to call and review.

[0178] In one embodiment, the method further comprises:

[0179] S64, automatically or based on a user end request to push a lens customization and / or treatment scheme and / or ocular axis change curve.

[0180] In order to facilitate patients to view myopia prevention and control related data and understand lens customization schemes, treatment schemes, etc., a lens customization and / or treatment scheme and / or ocular axis change curve can be automatically pushed or pushed based on a user end request. For example, at the patient end, the patient scans the code to view the exclusive lens design and treatment scheme, ocular axis change curve, etc., as shown in Figure 9 As shown.

[0181] In one embodiment, the method further comprises:

[0182] S65, automatically or based on a user end request to push a re-examination reminder to the patient.

[0183] In order to facilitate patients to re-examine and treat in a timely manner according to the myopia prevention and control situation, a re-examination reminder can be automatically pushed or pushed based on a user end request. For example, at the patient end, the patient scans the code to obtain the intelligent push function of the re-examination reminder.

[0184] In one embodiment, the method further comprises:

[0185] S66, automatically or based on a user end request to push a myopia prevention and control teaching video to the patient.

[0186] Figure 9 A myopia prevention and control teaching video visualization interface in the embodiment of the application is shown, in order to facilitate patients to scientifically treat according to the myopia prevention and control situation, a myopia prevention and control teaching video can be automatically pushed or pushed based on a user end request. For example, at the patient end, the patient scans the code to obtain the myopia prevention and control teaching video.

[0187] Embodiment 2:

[0188] In a second aspect, the present application also provides a dynamic myopia management lens prescription customization system, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, for implementing the steps of the dynamic myopia management lens prescription customization method as described in the above embodiments and preferred embodiments, which mainly comprises:

[0189] S1, obtaining patient ophthalmic basic data, the ophthalmic basic data comprising corneal topography data and OCT data;

[0190] S2, regularly following up the patient to obtain patient ophthalmic measurement follow-up data, and uploading the patient ophthalmic basic data and the patient ophthalmic measurement follow-up data to a management database;

[0191] S3, grabbing patient-associated parameters in the management database, pre-constructing a joint evaluation model for evaluating the myopia risk assessment level of the patient and dynamically updating the lens prescription, and training the pre-constructed joint evaluation model to convergence using the patient-associated parameters;

[0192] S4, the joint evaluation model dynamically updating the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data;

[0193] S5, obtaining the myopia risk assessment level of the patient, and the joint evaluation model dynamically updating the lens prescription in combination with the myopia risk assessment level, the ophthalmic basic data, and the ophthalmic measurement follow-up data.

[0194] It can be understood that the dynamic myopia management lens prescription customization system provided by the embodiments of the present application corresponds to the above-mentioned dynamic myopia management lens prescription customization method, and the explanation, examples, beneficial effects, etc. of the related content can refer to the corresponding content in the dynamic myopia management lens prescription customization method, which will not be repeated here.

[0195] In summary, compared with the prior art, the following beneficial effects are achieved:

[0196] 1. The dynamic myopia management lens prescription customization method provided by the present application obtains patient ophthalmic basic data, then regularly follows up the patient to obtain patient ophthalmic measurement follow-up data, the joint evaluation model dynamically updates the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data, and finally the joint evaluation model dynamically updates the lens prescription based on the myopia risk assessment level each time. The technical solution of the present application can dynamically adjust the lens prescription according to the follow-up axial progression and choroidal changes, realize dynamic myopia management lens prescription customization, and track and individualize the prevention and control management of the myopia period, thereby improving the effectiveness and sustainability of myopia prevention and control.

[0197] 2、In the embodiment of the application, the joint evaluation model integrates corneal topography, OCT images, axial length, refractive power, astigmatism, axial length change data, choroidal change data and other basic ophthalmic data and follow-up data. The fusion of such multi-modal data enables the model to comprehensively evaluate the myopia risk and eye condition of the patient from multiple angles, and the risk evaluation block can dynamically update the myopia risk evaluation level according to the latest ophthalmic data of the patient. Such a dynamic evaluation method can reflect the myopia progression of the patient in a timely manner, and provide more accurate decision support for clinicians, and the physical optical simulation layer in the lens design block simulates the spatial variation point spread function (PSF) of the compound optical system, providing an optical basis for lens design. Such simulation technology can consider the optical properties of the patient's eyeball, generate an optical model that matches the patient's eye condition, and improve the scientificity and accuracy of lens design.

[0198] 3、In the embodiment of the application, the Bayesian optimization combined with the multi-objective optimization algorithm can more accurately adjust the hyperparameters of the model, so that the F1 score and other indicators of the risk evaluation block in evaluating the myopia risk level are significantly improved, which helps to more accurately identify the myopia risk of the patient and provide more reliable basis for clinical decision-making. In the optimization process, the model can dynamically adjust the hyperparameters of risk evaluation and lens design according to the input parameters, and the dynamic adjustment capability enables the model to better adapt to the myopia progression of different patients at different stages, providing personalized risk evaluation and lens design services for patients.

[0199] 4、In the embodiment of the application, the feature fusion engine extracts features from the cross-modal aligned data, which can integrate key information in different modal data to generate a more representative and comprehensive feature set. These features can better reflect the comprehensive condition of the patient's eye and provide more information for risk assessment. The output feature fusion set as the input of the subsequent model can make the model better understand and learn the complex features of the patient's eye compared to single modal data or un-fused multi-modal data, thereby improving the accuracy of risk assessment. Combined with the prior knowledge risk feature clustering result, the myopia risk evaluation level of the patient can be more accurately calculated, and the calculated myopia risk evaluation level is updated to the level at the time of measuring the current data of the patient, realizing dynamic tracking and real-time updating of the myopia risk of the patient, which can timely reflect the myopia progression of the patient and provide more timely and accurate decision support for clinicians, which helps to better manage and control the myopia development of the patient.

[0200] 5、In the embodiment of the present application, when the combined evaluation model combines myopia risk evaluation grades, ophthalmic basic data and ophthalmic measurement follow-up data to dynamically update the lens prescription, the process of dynamic development of myopia can be considered, and the traditional lens using static template design cannot adjust the parameters with the progress of myopia. The method realizes the synchronous update of the lens prescription and the eye state of the patient through three-dimensional environment modeling and dynamic parameter optimization, can cover more people, and better serve different people, can avoid the one-sidedness of only relying on the axial length in the traditional evaluation, and can also avoid visual fatigue caused by excessive intervention, so as to ensure that the progress of myopia of patients with different risk grades is effectively prevented and controlled.

[0201] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0202] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A dynamic myopia management lens prescription customization method, characterized in that, The method comprises: acquiring patient ophthalmic basic data; regularly follow-up the patient, acquire patient ophthalmic measurement follow-up data, upload the patient ophthalmic basic data and the patient ophthalmic measurement follow-up data to a management database; grabbing patient associated parameters in the management database, pre-building a joint evaluation model for evaluating myopia risk assessment level of the patient and dynamically updating a lens prescription, and training the pre-built joint evaluation model to convergence using the patient associated parameters; the joint evaluation model dynamically updates the myopia risk assessment level of the patient based on the ophthalmic basic data and the ophthalmic measurement follow-up data; acquiring the myopia risk assessment level of the patient, and the joint evaluation model dynamically updates the lens prescription in combination with the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data.

2. The method of claim 1, wherein, The ophthalmic basic data includes corneal topography data and OCT data, and the ophthalmic measurement follow-up data includes patient eye axis change data and choroid change data.

3. The method of claim 2, wherein, The joint evaluation model training method comprises: traversing the management database, extracting a modeling sample set using a Latin hypercube sampling method, and dividing the modeling sample set into a training set and a test set; presetting initial hyperparameters of a risk assessment block and a lens design block in a double evaluation module; acquiring the training set, freezing a feature fusion engine, and pre-training the risk assessment block and the lens design block using the training set respectively; after pre-training, performing alternating proximal optimization joint training on the feature fusion engine and the double evaluation module to output a converged joint evaluation model; acquiring the test set, executing the joint evaluation model, and outputting test results of the joint evaluation model to obtain a risk assessment F1 score and a lens prescription MSE weighted average, and output a comprehensive test value; determining whether the comprehensive test value exceeds a preset test threshold, and if the comprehensive test value exceeds the preset test threshold, outputting the converged joint evaluation model; if the comprehensive test value does not exceed the preset test threshold, optimizing the joint evaluation model using Bayesian optimization combined with a multi-objective optimization algorithm, and continuing to iteratively train the optimized joint evaluation model to convergence.

4. The method of claim 3, wherein, When pre-building the joint evaluation model, the joint evaluation model takes a long short-term memory network model as an initial model, and further comprises a multi-modal input layer and an output layer, embeds 3-layer convolutional blocks, the long short-term memory network model and a feature fusion layer into a feature fusion engine, and sets a double evaluation module between the feature fusion engine and the output layer, wherein the double evaluation module comprises a risk assessment block and a lens design block.

5. The method of claim 3, wherein, The method for optimizing the joint evaluation model using Bayesian optimization combined with a multi-objective optimization algorithm comprises: constructing a double-objective optimization function with risk assessment accuracy and lens prescription accuracy as constraints, and defining decision variables of the double-objective function; initializing a Bayesian optimizer, selecting the next most promising hyperparameter combination using a surrogate model of the Bayesian optimizer and the double-objective optimization function, updating the surrogate model using newly observed data, and obtaining a surrogate model after updating the hyperparameters; selecting individuals for crossover operation according to non-dominated sorting and crowding distance, and performing mutation operation on the newly generated individuals, and selecting optimal solutions from a Pareto frontier. The optimal solution is taken as a physical feasibility constraint, the physical constraint is processed through a feasibility rule, and the hyperparameters of the joint evaluation model are optimized, and the joint evaluation model after hyperparameter optimization is output.

6. The method of claim 4, wherein, The method for dynamically updating the myopia risk assessment level of a patient based on the ophthalmic basic data and ophthalmic measurement follow-up data by the joint evaluation model comprises: The ophthalmic basic data and ophthalmic measurement follow-up data are input into a multi-modal input layer to output a cross-modal alignment set; The cross-modal alignment set is obtained, and a feature fusion engine extracts features from the cross-modal alignment set to output a feature fusion set; The feature fusion set is loaded, and a feature pyramid further extracts the feature fusion set to extract risk impact factors and impact weights in the feature fusion set; It is determined whether the impact weight of the risk impact factor exceeds a preset impact threshold, and if so, the current risk impact factor is retained and the risk impact factor is stacked to the feature fusion set; A Transformer cross-modal attention layer performs cross-modal fusion on the feature fusion set, and a myopia risk assessment level of a patient is calculated based on the prior knowledge risk feature clustering result.

7. The method of claim 6, wherein, The method for dynamically updating the lens prescription by combining the myopia risk assessment level, the ophthalmic basic data and the ophthalmic measurement follow-up data by the joint evaluation model comprises: An optical physics and optics simulation layer generates a simulated optical model matched with the patient's eye based on the ophthalmic basic data and the ophthalmic measurement follow-up data; A virtual aberration layer simulates the phase difference existing in the patient's eye, and corrects the simulated optical model based on the phase difference as a constraint; A defocus amount generator optimizes the simulated optical model based on a curvature optimization algorithm to determine the defocus amount distribution of the lens and output a defocus amount distribution map, taking the myopia risk assessment level as a physical constraint; An optical optimizer determines the position and width of the gradient channel of the lens, and performs three-dimensional environment modeling based on the position and width of the gradient channel of the lens to simulate and optimize the gradient channel design of the lens; A micro-lens array with an asymmetric gradient density distribution is generated based on the simulated optical model, and a genetic algorithm is used to optimize the arrangement and shape of the micro-lens; Based on the defocus amount distribution map, the simulated and optimized gradient channel design of the lens, and the arrangement and shape of the micro-lens, the lens parameters are identified, and the lens parameter mapping network is mapped; A critical threshold determination layer identifies the lens parameters in the lens parameter mapping network, and adjusts the critical threshold of the lens parameters based on the physical constraint of the myopia risk assessment level; A rich optimization layer generates a dynamically updated lens prescription in response to a feedback optical quality index, and executes the design lens parameter execution instruction based on the feedback optical quality index.

8. The method of claim 7, wherein, The method for generating a simulated optical model matched with the patient's eye based on the ophthalmic basic data and the ophthalmic measurement follow-up data by the optical physics and optics simulation layer comprises: An eye optical feature vector of the patient is extracted based on the ophthalmic basic data and the ophthalmic measurement follow-up data, and an optical element network constructs a spatially varying point spread function of a simulated optical system based on the eye optical feature vector; An initial optical model matched with the eye optical feature vector of the patient is generated through the spatially varying point spread function; The ISP proxy module performs image signal processing proxy operations on the initial optical model, simulates the ISP effect in the real imaging process, and outputs a corrected optical model; The simulation task network corrects the defocus uniformity of the corrected optical model in combination with the myopia risk assessment level, and generates a simulated optical model matched with the patient's eye.

9. The method of claim 4, wherein, The method further comprises: visualizing the ophthalmic basic data and / or ophthalmic measurement follow-up data of the patient; and / or visualizing and / or generating lens design drawings based on the lens prescription; and / or automatically generating a prevention and control effect tracking report based on the ophthalmic basic data and ophthalmic measurement follow-up data; and / or automatically or based on user-side request, pushing lens customization and / or treatment scheme and / or ocular axis change curve; and / or automatically or based on user-side request, pushing a re-examination reminder to the patient; and / or automatically or based on user-side request, pushing a myopia prevention and control teaching video to the patient.

10. A dynamic myopia management lens prescription customization system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, Steps for implementing the dynamic myopia management lens prescription customization method according to any one of claims 1-9.

Images