Multimodal parameter fusion-based ophthalmic visual effect dynamic simulation method and system

Abstract

The application discloses a multi-modal parameter fusion ophthalmic visual effect dynamic simulation method and system, and relates to the biomedical technology field. The method comprises the following steps: acquiring target user eye condition information and a visual parameter set; determining a plurality of optional models of implantable artificial lenses; performing simulation state evaluation on the plurality of optional artificial lens models to determine a plurality of predicted simulation complexities; generating optimal operation parameters according to the visual parameter set; and controlling a virtual reality (VR) device to perform post-cataract surgery visual acuity recovery state evolution simulation. The technical problem that the prior art cannot intuitively show the postoperative visual effect difference of implanting different models of artificial lenses in cataract surgery to patients, resulting in the lack of subjective experience basis for the patients when selecting, and causing the mismatch between the selected artificial lenses and the actual needs of the patients is solved.

Description

Technical Field

[0001] This application relates to the field of biomedical technology, specifically to a method and system for dynamic simulation of ophthalmic visual effects using multimodal parameter fusion. Background Technology

[0002] The treatment of cataract diseases in ophthalmology has shifted from simple cataract surgery to refractive surgery. Phacoemulsification cataract extraction with one-stage intraocular lens implantation is an important treatment method. Currently, there are many optical design principles and models of intraocular lenses available in clinical practice, including monofocal, multifocal, depth-of-field extension, and continuous-range lenses. The personalized selection of intraocular lenses during the surgical planning stage is an important part of doctor-patient communication. Different cataract patients have individual differences in their optical performance requirements for intraocular lenses, which directly affects the patient's postoperative visual quality and satisfaction.

[0003] However, traditional intraocular lens (IOL) selection relies on doctors' experience-based judgment based on clinical data. Patients can only understand the characteristics of IOLs through textual descriptions or static pictures, making it difficult to perceive the visual performance of different IOLs in real-life scenarios, such as the degree of glare when driving at night, the clarity when reading small print, and the accuracy of color reproduction. During the preoperative communication stage, doctors lack useful patient education tools, such as dynamic simulation methods and systems for ophthalmic visual effects, and cannot intuitively show patients the differences in visual effects after implantation of different IOL models. As a result, patients lack subjective experience to make choices and find it difficult to fully understand the expected surgical outcomes.

[0004] On the other hand, traditional methods of selecting intraocular lenses often focus on matching ocular anatomical parameters, neglecting the individual visual function needs, eye habits, and the dynamic impact of other eye diseases and disease stages associated with cataracts on postoperative visual outcomes. This may result in the selected intraocular lens not matching the patient's actual needs, affecting postoperative visual quality and recovery. Summary of the Invention

[0005] This application provides a method and system for dynamic simulation of ophthalmic visual effects by fusing multimodal parameters. This solves the technical problem that existing technologies cannot intuitively demonstrate the differences in visual effects of different intraocular lens models to cataract patients, resulting in a lack of subjective experience for patients when making their selections and causing the selected intraocular lens to not match the patient's actual needs.

[0006] The technical solution to the above-mentioned technical problems in this application is as follows:

[0007] In a first aspect, this application provides a method for dynamic simulation of ophthalmic visual effects by multimodal parameter fusion, the method comprising:

[0008] Perform visual detection on the target user to obtain eye condition information and visual parameter set;

[0009] Based on the target user's eye disease type, stage of disease, and eye condition information, several selectable intraocular lens models for implantation are determined;

[0010] Based on the eye disease type, disease stage, eye condition information, and visual parameter set, the simulation status evaluation of the several optional intraocular lens models is carried out to determine several predictive simulation complexities;

[0011] The virtual parameter generator of the VR device is selectively activated based on the aforementioned predictive simulation complexity. Based on the visual parameter set, several optimal operating parameters for the aforementioned selectable intraocular lens models are generated to control the VR device to simulate the evolution of postoperative vision recovery.

[0012] Secondly, this application provides a dynamic simulation system for ophthalmic visual effects based on multimodal parameter fusion, including:

[0013] The data acquisition module is used to perform visual detection on target users and obtain eye condition information and visual parameter sets;

[0014] The artificial lens screening module is used to determine several selectable lens models that can be implanted based on the target user's eye disease type, stage of disease, and eye condition information.

[0015] The artificial lens evaluation module is used to evaluate the simulated state of the several selectable lens models based on the eye disease type, disease stage, eye condition information and visual parameter set, and to determine several predictive simulation complexities.

[0016] The simulation execution module is used to selectively activate the virtual parameter generator of the VR device based on the several predicted simulation complexities, generate several optimal operating parameters for several selectable intraocular lens models according to the visual parameter set, and control the VR device to simulate the evolution of postoperative vision recovery status.

[0017] This application provides one or more technical solutions, which have at least the following technical effects or advantages:

[0018] This application provides a method and system for dynamic simulation of ophthalmic visual effects using multimodal parameter fusion. First, visual detection is performed on the target user to obtain their ocular condition information and visual parameter set. Second, based on the target user's eye disease type, stage, and the obtained ocular condition information, several optional intraocular lens (IOL) models are determined, ensuring the targeted selection and rationality of the initial screening. Then, based on the eye disease type, stage, ocular condition information, and visual parameter set, the simulation state of each optional IOL model is evaluated, thereby determining several predicted simulation complexities. These complexities reflect the difficulty and resource requirements of different IOL models during the simulation process. Finally, based on the predicted simulation complexity, the virtual parameter generator of the VR device is selectively activated, and several optimal operating parameters corresponding to the optional IOL models are generated according to the visual parameter set. This controls the VR device to simulate the evolution of postoperative visual recovery, allowing patients to immerse themselves in the dynamic process of visual recovery after implantation of different IOL models through the VR device before surgery, intuitively experiencing the changes in visual effects at different postoperative stages.

[0019] Through the above technical solution, this application combines the multimodal ocular and visual parameters of the target user with VR technology, enabling patients to intuitively and dynamically experience the vision recovery status and visual effects after implantation of different optional intraocular lens models through VR devices before surgery. This allows patients to choose the most suitable intraocular lens model for their own needs, effectively avoiding the problem of inappropriate intraocular lens selection due to lack of intuitive experience. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the method for dynamic simulation of ophthalmic visual effects by multimodal parameter fusion provided in the embodiments of this application;

[0022] Figure 2 This is a schematic diagram of the structure of the ophthalmic visual effect dynamic simulation system with multimodal parameter fusion provided in the embodiments of this application.

[0023] The components represented by each number in the attached diagram are explained below:

[0024] Data acquisition module 11, artificial lens screening module 12, artificial lens evaluation module 13, simulation execution module 14. Detailed Implementation

[0025] This application provides a method and system for dynamic simulation of ophthalmic visual effects by fusing multimodal parameters. This is intended to address the technical problem that existing technologies cannot intuitively demonstrate the differences in visual effects between different intraocular lens models to patients, resulting in a lack of subjective experience for patients when making their selections and causing the selected lens to not match the patient's actual needs.

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0028] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0029] Example 1, as Figure 1 As shown in the embodiments of this application, a method for dynamic simulation of ophthalmic visual effects by multimodal parameter fusion is provided, including:

[0030] S10: Perform visual detection on the target user to obtain eye condition information and visual parameter set;

[0031] The eye condition information includes corneal reports, fundus reports, biometric reports, and overall health indicators, and the visual parameter set includes optical parameters, functional parameters, spatial visual parameters, and color vision parameters.

[0032] In this embodiment, visual detection is first performed on the target user, which includes the collection of multi-dimensional eye information. Specifically, the acquisition of eye condition information requires the integration of multiple examination results. Among them, the corneal report records morphological data such as corneal curvature, thickness, and topography; the fundus report includes the health status of fundus structures such as the retina, macula, and optic nerve; the biometry report reflects the measurement results of parameters such as axial length, anterior chamber depth, and lens thickness; and the overall health indicator considers whether the target user has systemic diseases such as diabetes and hypertension that may affect eye condition and postoperative recovery.

[0033] Simultaneously, a set of visual parameters is acquired, including optical parameters, functional parameters, spatial visual parameters, and color vision parameters. The core of the optical parameters is whole-eye wavefront aberration data, such as the Zernike coefficient, which covers higher-order aberrations that cause halos and starbursts. Functional parameters mainly consist of contrast sensitivity function curves and glare incapacitance values, quantifying the degree of visual impairment in foggy conditions and strong light. Spatial visual parameters include microfield light sensitivity maps and standard visual field indices, which can locate visual field defects and dark spots. Color vision parameters, such as hue misalignment scores or spectral sensitivity curves, objectively measure color vision shift and blue light perception attenuation. These four types of parameters together constitute a complete visual quality profile from physical optics to neural perception.

[0034] S20: Determine several selectable lens models for implantation based on the target user's eye disease type, stage of disease, and eye condition information;

[0035] In this embodiment of the application, after obtaining the target user's eye condition information and visual parameter set, the system analyzes the type and stage of the user's eye disease to determine several selectable lens models that can be implanted with an artificial lens.

[0036] Specifically, based on a clear understanding of the type and stage of the eye disease, the system further filters the data by combining key parameters from the eye condition information, thereby identifying several selectable lens models that can be implanted with artificial lenses.

[0037] Specifically, step S20 in the method includes:

[0038] Obtain the target user's eye disease type and stage based on clinical diagnostic reports;

[0039] The information on the type of eye disease, stage of the disease, and eye condition is output to a pre-built artificial lens database for matching, and several selectable artificial lens models are obtained. Each artificial lens model is identified by lens attribute information.

[0040] In this embodiment of the application, firstly, information is extracted from the clinical diagnosis report to clarify the type of eye disease suffered by the target user, such as cataract combined with refractive errors (myopia, hyperopia, astigmatism) or presbyopia, or combined with other complex eye diseases such as glaucoma and macular degeneration. At the same time, the stage of the cataract is determined, such as the initial stage, inflated stage, mature stage or hypermature stage of the cataract.

[0041] Subsequently, the information on the type and stage of the eye disease, along with previously acquired information on eye conditions including cornea, fundus, biometrics, and overall health status, is input into a pre-built artificial lens database for matching and retrieval.

[0042] The artificial lens database stores information on various artificial lenses available on the market. Each lens model is identified by its lens attribute information, which includes basic identification and physical properties, core optical performance parameters, light energy distribution and visual phenomenon characteristics, as well as clinical matching and safety parameters.

[0043] Specifically, basic identifiers and physical attributes include the intraocular lens model and manufacturer, material and refractive index, and optical design type, such as monofocal, multifocal, depth-of-field extended, and continuous-range lenses; core optical performance parameters include nominal refractive power, additional refractive power, spherical or aspherical design, astigmatism correction capability, and modulation transfer function; light energy distribution and visual phenomenon characteristics include energy star distribution maps and descriptions of known optical phenomena, the size and intensity rate models of halo effects that multifocal lenses may produce, and the focal shift curves of continuous-range lenses under different lighting conditions; clinical matching and safety parameters include the applicable ocular conditions range and biometric constants such as the A constant / SF formula.

[0044] The database uses a preset matching algorithm to compare various parameters in the user's eye disease type, stage of disease, and eye condition information with the intraocular lens attribute information in multiple dimensions. It then filters out intraocular lens models that meet the target user's current eye condition and treatment needs in terms of optical characteristics, physiological compatibility, and surgical adaptability, thus obtaining several selectable intraocular lens models.

[0045] S30: Based on the eye disease type, disease stage, eye condition information and visual parameter set, perform simulation state evaluation on the several optional intraocular lens models respectively, and determine several predictive simulation complexities;

[0046] In this embodiment of the application, after determining several selectable intraocular lens models, a simulation state evaluation is performed on each model based on the type of eye disease, stage of disease, eye condition information, and visual parameter set, so as to determine the predictive simulation complexity.

[0047] Among them, the simulation state evaluation is a process that comprehensively considers the characteristics of intraocular lenses and individual patient differences, and assesses the computational resources, algorithm complexity and simulation accuracy requirements required to construct a dynamic simulation of postoperative visual effects for each available intraocular lens model.

[0048] Furthermore, the simulation complexity reflects the technical difficulty and resource consumption of different intraocular lens models in the simulation process. The higher the value, the more complex the computational model and the higher the hardware performance required to simulate the postoperative visual effect of that lens model.

[0049] Specifically, step S30 in the method includes:

[0050] The clinical pathological complexity is determined based on the type and stage of the eye disease.

[0051] Based on the eye condition information and visual parameter set, the lens fit of the several optional intraocular lens models is predicted, and the simulation complexity of several intraocular lenses is evaluated based on the fit prediction results.

[0052] The clinical pathological complexity is added to the simulation complexity of each of the several artificial lens models to obtain several predicted simulation complexities for the several selectable artificial lens models.

[0053] In this embodiment, the clinical pathological complexity is first determined based on the target user's eye disease type and stage. Specifically, corresponding complexity weighting coefficients are set for different eye disease types and stages.

[0054] For example, the clinicopathological complexity weight for mature simple age-related cataracts is a base value of 1.0; if combined with high myopia, the weight increases by 0.3; if diabetic retinopathy (DR) in the non-proliferative stage is also present, the weight increases by another 0.2; if DR has progressed to the proliferative stage and is accompanied by macular edema, the weight increases to 1.8. For glaucoma patients, the weight is graded according to the degree of visual field loss: MD > -6dB, the weight increases by 0.4; -12dB < MD ≤ -6dB, the weight increases by 0.7; and MD ≤ -12dB, the weight increases by 1.0. The dynamic changes in the disease stage also affect the complexity; for example, as cataracts progress from the initial stage to the hypermature stage, the weight linearly increases from 0.5 to 1.2. By quantifying and superimposing the weight coefficients, a specific value for clinicopathological complexity is finally obtained, which reflects the contribution of the patient's ocular pathology to the basic difficulty of the simulation process.

[0055] Secondly, based on ocular condition information and visual parameter sets, lens fit prediction was performed on several selectable intraocular lens (IOL) models. The simulation complexity of several IOLs was then evaluated based on the fit prediction results. First, an IOL fit prediction model was constructed. This model uses biometric parameters from ocular condition information, such as corneal curvature, anterior chamber depth, and axial length, as well as visual parameter sets, such as total wavefront aberration and contrast sensitivity, as input variables. Optical design parameters and material refractive index from IOL attribute information were used as reference variables. Multifactor regression analysis was used to calculate the fit index between each selectable IOL model and the target user. The fit index ranges from 0 to 100, with a higher index indicating better fit.

[0056] For example, when the target user has corneal astigmatism of 1.5D, and the intraocular lens has a matching cylinder power and an axis deviation of ≤5°, the fit index can be increased by 15 points; if the user's higher-order aberrations RMS > 0.3μm, choosing an aspherical intraocular lens can increase the fit index by 20 points.

[0057] Then, the fit index is converted into the intraocular lens (IOL) simulation complexity. The conversion formula is: IOL simulation complexity = 100 - fit index + IOL characteristic basic complexity. The IOL characteristic basic complexity is set according to the lens type: 10 for monofocal IOLs, 30 for multifocal IOLs, 40 for continuous range IOLs, and 35 for depth-of-field extended IOLs, etc.

[0058] For example, if the fit index of a certain multifocal intraocular lens is 75, then its intraocular lens simulation complexity = 100 - 75 + 30 = 55.

[0059] Finally, the clinical pathological complexity is added to the simulation complexity of several artificial lens models to obtain several predicted simulation complexities for several selectable artificial lens models.

[0060] For example, if the clinical pathology complexity of the target user is 2.5 and the crystal simulation complexity of a certain optional intraocular lens model is 55, then the predicted simulation complexity of this model is 2.5 + 55 = 57.5.

[0061] Through the above calculation method, each optional intraocular lens model obtains a predictive simulation complexity value that comprehensively reflects the patient's pathological condition and the characteristics of the intraocular lens itself.

[0062] Specifically, based on the ocular condition information and visual parameter set, the fit of several selectable intraocular lens (IOL) models is predicted. Based on the fit prediction results, the simulation complexity of several IOLs is evaluated, including:

[0063] Randomly select a first intraocular lens model from the plurality of optional intraocular lens models;

[0064] Using a dual-channel intraocular lens (IOL) fitting prediction system, the IOL fitting is predicted based on the eye condition information, visual parameter set, and first IOL model, and the first conditional fitting and first visual fitting are output.

[0065] The simulation complexity of the first artificial lens is determined by a weighted evaluation based on the first conditional fit and the first visual fit, and then added to the simulation complexity of the plurality of artificial lenses. The simulation complexity of the first artificial lens is negatively correlated with the first conditional fit and the first visual fit.

[0066] In this embodiment of the application, firstly, one of several optional intraocular lens models is randomly selected as the first intraocular lens model in order to predict the intraocular lens fit and evaluate the intraocular lens simulation complexity.

[0067] Secondly, a dual-channel prediction method is used to predict intraocular lens (IOL) fit. This dual-channel method considers both conditional fit and visual fit. The conditional fit channel primarily calculates the matching degree between ocular condition information and the lens attribute information of the first IOL model, outputting the first conditional fit score. Specifically, biometric parameters such as corneal curvature, anterior chamber depth, and axial length from the ocular condition information are compared with attributes such as the applicable range and size specifications of the first IOL model.

[0068] For example, if the minimum anterior chamber depth required by the first intraocular lens model is 2.8 mm, while the anterior chamber depth of the target user is 2.7 mm, then points will be deducted accordingly in the condition fit score; conversely, if all biometric parameters are within the ideal fit range of the intraocular lens model, then the condition fit score will be higher.

[0069] Furthermore, the visual adaptation channel focuses on evaluating the correlation between the visual parameter set and the optical performance of the first artificial lens model, and outputs the first visual adaptation degree.

[0070] Next, after obtaining the first conditional fit and the first visual fit, a weighted evaluation is performed based on their importance to determine the simulation complexity of the first intraocular lens. Since the simulation complexity of the first intraocular lens is negatively correlated with both the first conditional fit and the first visual fit, a higher fit indicates a better match between the intraocular lens model and the target user, resulting in fewer parameters requiring additional correction and adjustment during the simulation, and consequently, lower simulation complexity. Conversely, a lower fit requires more complex algorithms and more computational resources to simulate the visual effects the lens might produce in the target user's eyes, naturally increasing the simulation complexity.

[0071] Finally, the calculated simulation complexity of the first crystal is added to a set of simulation complexities for several crystals to complete the simulation complexity evaluation for the first crystal model. For the remaining optional crystal models, the same process is followed to obtain their respective crystal simulation complexities.

[0072] Furthermore, the method for constructing the dual-channel crystal fit prediction includes:

[0073] Based on historical ophthalmology treatment records, several samples of ocular condition information, several samples of visual parameter sets, and several samples of intraocular lens models were collected. The historical postoperative visual success rate of the sample intraocular lens model corresponding to different ocular condition information was obtained as the sample condition fit, and the historical postoperative visual success rate of the sample intraocular lens model corresponding to different sample visual parameter sets was obtained as the sample visual fit, resulting in several sample condition fits and several sample visual fits.

[0074] Using the aforementioned sample eye condition information, sample visual parameter sets, and sample intraocular lens models as input data, and the aforementioned sample conditional fit and sample visual fit as label data, a deep learning model is trained until convergence, generating a dual-channel intraocular lens fit prediction.

[0075] In this embodiment of the application, firstly, cases with complete data are selected from historical ophthalmological treatment records as training samples. Each sample contains information on the sample's ocular conditions, such as corneal curvature, axial length, anterior chamber depth, corneal topography, fundus examination results, etc. The sample visual parameter set includes whole-wavefront aberration data, contrast sensitivity function curve, glare disability value, microfield light sensitivity map, color vision parameters, etc., and also includes the model of the sample's actual implanted intraocular lens.

[0076] Simultaneously, postoperative visual effect evaluation indicators corresponding to the model of the intraocular lens (IOL) in the samples were extracted from historical records and converted into sample conditional fit and sample visual fit. For example, the absence of complications such as IOL dislocation, tilting, eccentricity, or secondary cataracts within 3 months postoperatively, along with stable biometric parameters, was quantified as a high sample conditional fit. Postoperative uncorrected visual acuity, corrected visual acuity, spectacle rejection rate, improvement in contrast sensitivity, and subjective scores for visual disturbances such as halos and starbursts were comprehensively converted into sample visual fit as label data.

[0077] Subsequently, a deep learning model architecture was constructed, comprising two parallel processing channels: a conditional adaptation channel and a visual adaptation channel. The input layer of the conditional adaptation channel receives sample ocular condition information and lens attribute information of the sample intraocular lens model, such as the applicable ocular condition range and size specifications. Features are extracted using a convolutional neural network, such as the matching features between corneal curvature and intraocular lens optical design, and the compatibility features between anterior chamber depth and intraocular lens haptic design. The input layer of the visual adaptation channel receives sample visual parameter sets and optical performance parameters of the sample intraocular lens model, such as wavefront aberration correction capability and light energy distribution characteristics. Sequential data is processed using a recurrent neural network to capture the dynamic correlation between visual parameters and intraocular lens optical properties.

[0078] The two output layers output the predicted sample conditional fit and the predicted sample visual fit, respectively. During training, the mean squared error loss function is used to compare the model's predicted sample conditional fit and sample visual fit with the actual labeled data. The model parameters are continuously adjusted through the backpropagation algorithm until the model converges, i.e., the prediction error reaches below the preset threshold.

[0079] Ultimately, the trained deep learning model constitutes a dual-channel prediction system for intraocular lens fit, which can automatically output the corresponding conditional fit and visual fit based on the input eye condition information, visual parameter set, and specific intraocular lens model.

[0080] For example, a dual-channel prediction system for artificial lens fitness is built and trained based on a deep learning model. The specific steps are as follows:

[0081] First, data preparation: The input nodes for the dual-channel intraocular lens fit prediction are several samples of ocular condition information, several samples of visual parameter sets, and several samples of intraocular lens models, which are collected based on historical ophthalmological treatment records.

[0082] Secondly, in model construction, the number of nodes in the input layer equals the dimension of the input features, such as corneal curvature, axial length, anterior chamber depth, corneal topography, fundus examination results, whole-wavefront aberration data, contrast sensitivity function curve, glare disabling value, micro-field light sensitivity map, color vision parameters, and the model of the sample intraocular lens, totaling 11 features, thus the input layer contains 11 nodes; 1-3 hidden layers are set, and the number of nodes in each layer is adjusted experimentally, such as 64, 32, etc., and the activation function is ReLU; the output layer generally does not use an activation function, such as outputting continuous values ​​directly if the output takes 2 nodes.

[0083] Next, during model training, the first conditional fit and the first visual fit are used as outputs. Several sample conditional fits and several sample visual fits are used as label data. The Adam optimizer and mean squared error loss function are used to construct the training framework. The batch size is set to 32, the total number of training epochs is 50, and an early stopping mechanism with a patience of 5 is introduced. When the validation set loss does not decrease for 5 consecutive epochs, the training process is automatically terminated, resulting in a trained dual-channel prediction of intraocular lens fit. This effectively avoids overfitting while ensuring the model reaches convergence.

[0084] S40: Based on the aforementioned predictive simulation complexity, the virtual parameter generator of the VR device is selectively activated, and several optimal operating parameters for the aforementioned selectable intraocular lens models are generated according to the visual parameter set, thereby controlling the VR device to simulate the evolution of postoperative vision recovery status.

[0085] In this embodiment, after obtaining the predicted simulation complexity of several selectable artificial lens models, the virtual parameter generator of the VR device is selectively activated. Specifically, a complexity threshold range is preset. When the predicted simulation complexity of a certain selectable artificial lens model is lower than the lower limit of this range, it indicates that its simulation process is relatively simple, and the basic version of the virtual parameter generator can be called to quickly generate core operating parameters using a simplified parameter generation algorithm. When the predicted simulation complexity is within the threshold range, the standard version of the virtual parameter generator is activated, and a conventional multi-parameter collaborative optimization algorithm is enabled. When the predicted simulation complexity is higher than the upper limit of the threshold range, the advanced version of the virtual parameter generator is triggered, calling a parameter generation module that includes advanced features such as adaptive mesh subdivision and ray tracing acceleration to meet the high-precision simulation requirements.

[0086] Secondly, after activating the virtual parameter generator, the target user's visual parameter set will be used as the core input. Combined with the intraocular lens attribute information of the selectable intraocular lens model and eye condition information, several optimal operating parameters of the selectable intraocular lens model will be generated, and the VR device will be controlled to simulate the evolution of postoperative vision recovery status.

[0087] The virtual parameter generator, which selectively activates the VR device based on the aforementioned predictive simulation complexity, generates several optimal operating parameters for the several selectable artificial lens models according to the visual parameter set, including:

[0088] Based on the historical operation records of similar VR devices, several sample visual parameter sets were collected, and the historical qualified operation parameters of VR devices corresponding to different sample visual parameter sets were used as sample operation parameters to obtain several sample operation parameters.

[0089] A generative adversarial network is trained using the aforementioned set of visual parameters and the set of operational parameters of several samples as training data to construct a virtual parameter generator;

[0090] Based on the aforementioned predictive simulation complexity, the virtual parameter generator is selectively activated to generate several optimal operating parameters for the several selectable artificial lens models according to the visual parameter set.

[0091] In this embodiment, firstly, cases meeting clinical simulation standards are selected from the historical operation records of similar VR devices. Sample visual parameter sets are then extracted from these cases. These sets encompass data on preoperative whole-eye wavefront aberration, contrast sensitivity, glare threshold, and retinal photoreceptor distribution density for patients of different ages, refractive states, and ocular pathological characteristics. Simultaneously, for each sample visual parameter set, the various operating parameters used by the VR device when successfully completing postoperative visual simulation are collected as sample operating parameters. These include rendering resolution, field of view, light sampling frequency, corneal scattering coefficient, lens diffraction model parameters, and retinal imaging algorithm version. This ensures that all historical operating parameters have passed clinical validity verification, meaning the deviation between the simulation results and the actual postoperative visual effect is within a preset acceptable range.

[0092] Subsequently, a training framework for a virtual parameter generator is constructed using the sample visual parameter set as input data to the generative adversarial network (GAN) and the corresponding sample runtime parameters as output targets. The GAN comprises two core modules: a generator and a discriminator. The generator receives the sample visual parameter set and attempts to generate simulated runtime parameters, while the discriminator evaluates the similarity between the generated runtime parameters and the real sample runtime parameters. By alternately training the generator and discriminator, the generator gradually learns the mapping relationship from visual parameters to optimal runtime parameters.

[0093] Furthermore, during training, a hybrid loss function is employed, combining mean squared error loss and perceptual loss to improve the accuracy and clinical relevance of the generated parameters. After training, the virtual parameter generator possesses the ability to autonomously generate the optimal parameter combination adapted to VR device operation based on the input visual parameter set.

[0094] Finally, based on the previously calculated predicted simulation complexity of several selectable intraocular lens (IOL) models, the activation mode of the virtual parameter generator is selected. For example, when the predicted simulation complexity is low, such as ≤30, the generator's fast generation mode is activated to reduce the number of network computation layers and prioritize parameter output speed; when the predicted simulation complexity is medium, such as 30 < complexity ≤70, the standard generation mode is enabled, calling the complete generation network structure; when the predicted simulation complexity is high, such as >70, the enhanced generation mode is activated, increasing the number of parameter optimization iterations on the basis of the standard generation network and introducing a Monte Carlo sampling mechanism to fine-tune the generated parameters in multiple rounds, ensuring that the generated optimal operating parameters can support the VR device's simulation of the postoperative vision recovery evolution of complex ocular conditions and highly adaptable IOL models.

[0095] Specifically, a generative adversarial network is trained using the aforementioned sets of visual parameters and operational parameters of several samples as training data to construct a virtual parameter generator, including:

[0096] The aforementioned visual parameter sets and operational parameters of several samples are used as training data, and the training data is subjected to Q-fold cross-partitioning to obtain Q training sets of samples, where Q is an integer greater than 5;

[0097] Using the sample visual parameter set as input and the sample running parameters as labels, the adversarial network is trained to converge using the Q sample training sets, generating Q virtual parameter generation units, which are then combined to obtain a virtual parameter generator.

[0098] In this embodiment, the generative adversarial network is first trained using Q-fold cross-validation. Specifically, the training data, consisting of all collected sample visual parameter sets and corresponding sample running parameters, is divided into a training set and a validation set according to a preset ratio, such as 8:2. The training set is then further randomly and uniformly divided into Q subsets, where Q is an integer greater than 5, for example, Q=10. Each subset contains a similar number of different types of visual parameter samples, ensuring that the data distribution characteristics of each subset are consistent with the overall training set.

[0099] Subsequently, Q-1 subsets are selected sequentially as training subsets, and the remaining subset is selected as validation subset, constructing Q different training-validation data pairs. For each data pair, the generative adversarial network is trained independently using the training subset. During training, the model performance is monitored in real time using the validation subset. Training for that set is stopped when the validation loss no longer decreases, resulting in a trained virtual parameter generation unit. This process is repeated Q times to obtain Q virtual parameter generation units with identical structures but different parameters. Each virtual parameter generation unit corresponds to the optimal parameter mapping capability under different data partitioning scenarios.

[0100] Finally, the Q virtual parameter generation units are integrated to form the final virtual parameter generator. When a new set of visual parameters is input, the virtual parameter generator calls all Q generation units to generate runtime parameters. The Q outputs are then fused using a weighted average or majority voting method to obtain the optimal runtime parameters with better overall performance and stronger robustness. Furthermore, the weights can be dynamically adjusted based on the performance of each unit on the validation set. These steps effectively reduce model performance fluctuations caused by biases in single data partitioning.

[0101] Furthermore, the virtual parameter generator is selectively activated based on the aforementioned several prediction simulation complexities, and several optimal operating parameters are generated according to the visual parameter set, including:

[0102] Randomly select the first prediction simulation complexity from the plurality of prediction simulation complexities;

[0103] The ratio of the first prediction simulation complexity to the historical maximum prediction simulation complexity recorded within the historical time period is multiplied by Q and rounded to obtain K, where K is an integer greater than 1 and less than or equal to Q.

[0104] K virtual parameter generation units are randomly selected from the Q virtual parameter generation units of the virtual parameter generator. K predicted operating parameters are obtained based on the visual parameter set. The average value of the K predicted operating parameters is selected as the first optimal operating parameter, and several optimal operating parameters are obtained by sequential analysis.

[0105] In this embodiment of the application, firstly, one prediction simulation complexity is randomly selected from the prediction simulation complexity corresponding to several optional artificial lens models as the first prediction simulation complexity.

[0106] Secondly, obtain the maximum value of all predicted simulation complexities recorded within the historical time period. For example, select the maximum value of the predicted simulation complexity over the past 6 months or 1 year, and record it as the historical maximum predicted simulation complexity. Calculate the ratio of the first predicted simulation complexity to the historical maximum predicted simulation complexity. Multiply this ratio by the total number Q of virtual parameter generation units in the virtual parameter generator, and round the product to obtain an integer K. K satisfies the condition of being greater than 1 and less than or equal to Q, ensuring that at least two generation units participate in parameter generation to reflect the ensemble concept, while not exceeding the total number of generation units.

[0107] For example, if Q=10, the complexity of the first prediction simulation is 60, and the historical maximum prediction simulation complexity is 80, then the ratio is 0.75, 0.75×10=7.5, and after rounding, K=7.

[0108] Subsequently, among the Q virtual parameter generation units of the virtual parameter generator, K generation units are selected by random sampling. The visual parameter set of the target user is input into the K selected generation units respectively. Each generation unit independently predicts and outputs a set of running parameters based on the parameter mapping relationship it has trained, thus obtaining K predicted running parameters.

[0109] Finally, the mean of the K predicted operating parameters is calculated, and the average value of each parameter dimension is taken as the first optimal operating parameter for the available artificial lens model.

[0110] Furthermore, by following the same process described above, all available intraocular lens models can be processed sequentially to obtain the optimal operating parameters for each available intraocular lens model.

[0111] By dynamically adjusting the number of generating units involved in the decision-making process based on the complexity of the prediction simulation, the reliability of parameter generation can be improved by utilizing the wisdom of more units when the complexity is high, while maintaining a certain level of computational efficiency when the complexity is moderate. At the same time, randomly selecting generating units also avoids the systematic bias that may be caused by fixed combinations.

[0112] In summary, compared with existing technologies, this application provides ophthalmologists and patients with a more comprehensive basis for selecting intraocular lenses by combining the conditional fit and visual fit of the dual-channel output predicted by intraocular lens fit, as well as the postoperative visual recovery state evolution simulation results generated by VR devices.

[0113] In summary, the embodiments of this application have at least the following technical effects:

[0114] This application provides a method for dynamic simulation of ophthalmic visual effects using multimodal parameter fusion. First, visual detection is performed on the target user to obtain their ocular condition information and visual parameter set. Second, based on the target user's eye disease type, stage, and the obtained ocular condition information, several optional intraocular lens (IOL) models suitable for cataract surgery are determined, ensuring the targeted selection and rationality of the initial screening. Then, based on the eye disease type, stage, ocular condition information, and visual parameter set, the simulation state of each optional IOL model is evaluated, thereby determining several predicted simulation complexities. This complexity reflects the difficulty and resource requirements of different IOL models during the simulation process. Finally, based on the predicted simulation complexity, the virtual parameter generator of the VR device is selectively activated, and several optimal operating parameters corresponding to the optional IOL models are generated according to the visual parameter set. This controls the VR device to simulate the evolution of postoperative visual recovery, allowing patients to immerse themselves in the dynamic process of visual recovery after implantation of different IOL models through the VR device before surgery, intuitively experiencing the changes in visual effects at different stages postoperatively.

[0115] Through the above technical solution, this application combines the multimodal ocular and visual parameters of the target user with VR technology, enabling patients to intuitively and dynamically experience the vision recovery status and visual effects after implantation of different optional intraocular lens models through VR devices before surgery. This allows patients to choose the most suitable intraocular lens model for their own needs, effectively avoiding the problem of inappropriate intraocular lens selection due to lack of intuitive experience.

[0116] Example 2, as Figure 2 As shown, based on the same inventive concept as the multimodal parameter fusion method for dynamic simulation of ophthalmic visual effects provided in Embodiment 1, this application also provides a multimodal parameter fusion system for dynamic simulation of ophthalmic visual effects, including:

[0117] Data acquisition module 11 is used to perform visual detection on the target user and obtain eye condition information and visual parameter set;

[0118] The artificial lens screening module 12 is used to determine several optional artificial lens models that can be implanted based on the target user's eye disease type, disease stage and eye condition information;

[0119] The artificial lens evaluation module 13 is used to perform simulated state evaluation on the several selectable artificial lens models based on the eye disease type, disease stage, eye condition information and visual parameter set, and to determine several predicted simulation complexities.

[0120] The simulation execution module 14 is used to selectively activate the virtual parameter generator of the VR device based on the plurality of predicted simulation complexities, generate a plurality of optimal operating parameters for the plurality of selectable intraocular lens models according to the visual parameter set, and control the VR device to perform postoperative vision recovery state evolution simulation.

[0121] In one embodiment, visual testing is performed on the target user according to a preset standardized comprehensive ophthalmological testing procedure to obtain eye condition information and a set of visual parameters. The eye condition information includes corneal reports, fundus reports, biometric reports, and overall health indicators. The set of visual parameters includes optical parameters, functional parameters, spatial visual parameters, and color vision parameters.

[0122] In one embodiment, the artificial lens screening module 12 is specifically used for:

[0123] Obtain the target user's eye disease type and stage based on clinical diagnostic reports;

[0124] The information on the type of eye disease, stage of the disease, and eye condition is output to a pre-built artificial lens database for matching, and several selectable artificial lens models are obtained. Each artificial lens model is identified by lens attribute information.

[0125] In one embodiment, the artificial lens evaluation module 13 is specifically used for:

[0126] The clinical pathological complexity is determined based on the type and stage of the eye disease.

[0127] Based on the eye condition information and visual parameter set, the fit of several optional intraocular lens models is predicted, and the simulation complexity of several intraocular lenses is evaluated based on the fit prediction results.

[0128] The clinical pathological complexity is added to the simulation complexity of each of the several artificial lens models to obtain several predicted simulation complexities for the several selectable artificial lens models.

[0129] Furthermore, in one embodiment, the lens fit prediction is performed on the plurality of selectable intraocular lens models based on the eye condition information and visual parameter set, and the simulation complexity of several intraocular lenses is evaluated based on the fit prediction results, including:

[0130] Randomly select a first intraocular lens model from the plurality of optional intraocular lens models;

[0131] Using a dual-channel artificial lens fit prediction method, the lens fit is predicted based on the eye condition information, visual parameter set, and first artificial lens model, and the first conditional fit and first visual fit are output.

[0132] The simulation complexity of the first artificial lens is determined by a weighted evaluation based on the first conditional fit and the first visual fit, and then added to the simulation complexity of the plurality of artificial lenses. The simulation complexity of the first artificial lens is negatively correlated with the first conditional fit and the first visual fit.

[0133] Furthermore, in one embodiment of the application, the method for constructing the dual-channel intraocular lens fit prediction includes:

[0134] Based on historical ophthalmology treatment records, several samples of ocular condition information, several samples of visual parameter sets, and several samples of intraocular lens models were collected. The historical postoperative visual success rate of the sample intraocular lens model corresponding to different ocular condition information was obtained as the sample condition fit, and the historical postoperative visual success rate of the sample intraocular lens model corresponding to different sample visual parameter sets was obtained as the sample visual fit, resulting in several sample condition fits and several sample visual fits.

[0135] Using the aforementioned sample eye condition information, sample visual parameter sets, and sample intraocular lens models as input data, and the aforementioned sample conditional fit and sample visual fit as label data, a deep learning model is trained until convergence, generating a dual-channel intraocular lens fit prediction.

[0136] Furthermore, in one embodiment, a virtual parameter generator that selectively activates the VR device based on the plurality of predictive simulation complexities generates several optimal operating parameters for the plurality of selectable artificial lens models according to the visual parameter set, including:

[0137] Based on the historical operation records of similar VR devices, several sample visual parameter sets were collected, and the historical qualified operation parameters of VR devices corresponding to different sample visual parameter sets were used as sample operation parameters to obtain several sample operation parameters.

[0138] A generative adversarial network is trained using the aforementioned set of visual parameters and the set of operational parameters of several samples as training data to construct a virtual parameter generator;

[0139] Based on the aforementioned predictive simulation complexity, the virtual parameter generator is selectively activated to generate several optimal operating parameters for the several selectable artificial lens models according to the visual parameter set.

[0140] Furthermore, a generative adversarial network is trained using the aforementioned set of visual parameters and the set of operational parameters as training data to construct a virtual parameter generator, including:

[0141] The aforementioned visual parameter sets and operational parameters of several samples are used as training data, and the training data is subjected to Q-fold cross-partitioning to obtain Q training sets of samples, where Q is an integer greater than 5;

[0142] Using the sample visual parameter set as input and the sample running parameters as labels, the adversarial network is trained to converge using the Q sample training sets, generating Q virtual parameter generation units, which are then combined to obtain a virtual parameter generator.

[0143] Furthermore, in one embodiment, the virtual parameter generator is selectively activated based on the plurality of prediction simulation complexities to generate a plurality of optimal operating parameters according to the visual parameter set, including:

[0144] Randomly select the first prediction simulation complexity from the plurality of prediction simulation complexities;

[0145] The ratio of the first prediction simulation complexity to the historical maximum prediction simulation complexity recorded within the historical time period is multiplied by Q and rounded to obtain K, where K is an integer greater than 1 and less than or equal to Q.

[0146] K virtual parameter generation units are randomly selected from the Q virtual parameter generation units of the virtual parameter generator. K predicted operating parameters are obtained based on the visual parameter set. The average value of the K predicted operating parameters is selected as the first optimal operating parameter, and several optimal operating parameters are obtained by sequential analysis.

[0147] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0148] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0149] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for dynamic simulation of ophthalmic visual effects by multimodal parameter fusion, characterized in that the method... include: Perform visual detection on the target user to obtain eye condition information and visual parameter set; Based on the target user's eye disease type, stage of disease, and eye condition information, several selectable intraocular lens models for implantation are determined; Based on the aforementioned eye disease type, disease stage, eye condition information, and visual parameter set, a simulation state evaluation is performed on the several selectable lens models to determine several predictive simulation complexities, including: The clinical pathological complexity is determined based on the type and stage of the eye disease. Based on the aforementioned eye condition information and visual parameter set, lens fit prediction is performed on the several selectable lens models. Based on the fit prediction results, the simulation complexity of several lenses is evaluated, including: Randomly select a first crystal model from the plurality of optional crystal models; Using a dual-channel lens fit prediction method, lens fit prediction is performed based on the eye condition information, visual parameter set, and first lens model, and the first conditional fit and first visual fit are output. The first crystal simulation complexity is determined by weighted evaluation based on the first conditional fit and the first visual fit, and then added to the plurality of crystal simulation complexities. The first crystal simulation complexity is negatively correlated with the first conditional fit and the first visual fit. The clinical pathological complexity is added to each of the several crystal simulation complexities to obtain several predicted simulation complexities for the several selectable crystal models; A virtual parameter generator for the VR device is selectively activated based on the aforementioned predictive simulation complexity. This generator generates several optimal operating parameters for several selectable lens models according to the visual parameter set, controlling the VR device to simulate the evolution of postoperative visual recovery. This includes: Based on the historical operation records of similar VR devices, several sample visual parameter sets were collected, and the historical qualified operation parameters of VR devices corresponding to different sample visual parameter sets were used as sample operation parameters to obtain several sample operation parameters. Using the aforementioned set of visual parameters and the set of operational parameters from several samples as training data, a generative adversarial network is trained to construct a virtual parameter generator, including: The aforementioned visual parameter sets and operational parameters of several samples are used as training data, and the training data is subjected to Q-fold cross-partitioning to obtain Q training sets of samples, where Q is an integer greater than 5; Using the sample visual parameter set as input and the sample running parameters as labels, the Q sample training sets are used to train the Generative Adversarial Network until convergence, generating Q virtual parameter generation units, which are then combined to obtain a virtual parameter generator. Based on the aforementioned predictive simulation complexity, the virtual parameter generator is selectively activated to generate several optimal operating parameters for the several selectable crystal models according to the visual parameter set, including: Randomly select the first prediction simulation complexity from the plurality of prediction simulation complexities; The ratio of the first prediction simulation complexity to the historical maximum prediction simulation complexity recorded within the historical time period is multiplied by Q and rounded to obtain K, where K is an integer greater than 1 and less than or equal to Q. K virtual parameter generation units are randomly selected from the Q virtual parameter generation units of the virtual parameter generator. K predicted operating parameters are obtained based on the visual parameter set. The average value of the K predicted operating parameters is selected as the first optimal operating parameter, and several optimal operating parameters are obtained by sequential analysis.

2. The method for dynamic simulation of ophthalmic visual effects by multimodal parameter fusion according to claim 1, characterized in that, Visual examination is conducted on target users according to a pre-set standardized comprehensive ophthalmological examination procedure to obtain eye condition information and a set of visual parameters. The eye condition information includes corneal reports, fundus reports, biometric reports, and overall health indicators. The set of visual parameters includes optical parameters, functional parameters, spatial visual parameters, and color vision parameters.

3. The method for dynamic simulation of ophthalmic visual effects by multimodal parameter fusion according to claim 1, characterized in that, Based on the target user's eye disease type, stage, and eye condition information, several selectable intraocular lens models are determined, including: Obtain the target user's eye disease type and stage based on clinical diagnostic reports; The eye disease type, disease stage, and eye condition information are input into a pre-built artificial lens database for matching to obtain several selectable lens models, each of which is identified by lens attribute information.

4. The method for dynamic simulation of ophthalmic visual effects by multimodal parameter fusion according to claim 1, characterized in that, The method for constructing the dual-channel crystal fit prediction includes: Based on historical ophthalmology treatment records, several samples of ocular condition information, several samples of visual parameter sets, and several samples of lens models were collected. The historical postoperative visual success rate of the lens models corresponding to different ocular condition information was obtained as the sample condition fit, and the historical postoperative visual success rate of the lens models corresponding to different visual parameter sets was obtained as the sample visual fit, resulting in several sample condition fits and several sample visual fits. Using the aforementioned sample eye condition information, sample visual parameter sets, and sample lens models as input data, and the aforementioned sample condition fit and sample visual fit as label data, a deep learning model is trained until convergence, generating a dual-channel lens fit prediction.

5. A dynamic simulation system for ophthalmic visual effects based on multimodal parameter fusion, characterized in that, The method for performing the multimodal parameter fusion dynamic simulation of ophthalmic visual effects according to any one of claims 1-4 includes: The data acquisition module is used to perform visual detection on target users and obtain eye condition information and visual parameter sets; The lens screening module is used to determine several selectable lens models that can be implanted with an artificial lens based on the target user's eye disease type, stage of disease, and eye condition information; The lens evaluation module is used to evaluate the simulated state of the several selectable lens models based on the eye disease type, disease stage, eye condition information and visual parameter set, and to determine several predictive simulation complexities. The simulation execution module is used to selectively activate the virtual parameter generator of the VR device based on the several predicted simulation complexities, generate several optimal operating parameters for several selectable lens models according to the visual parameter set, and control the VR device to simulate the evolution of postoperative vision recovery status.

Images