Multi-mode orthokeratology lens intelligent fitting method and system based on multiple models

Abstract

According to the multi-model-based intelligent fitting method and system for the multi-modal orthokeratology lens, parameters most matched with the corneal morphology of a patient are automatically screened out from candidate lenses through a recall model, and the defects of an experience-dominated traditional method are overcome; and optimizing lens selection by using a sorting model: further selecting the lens with the best shaping effect from the screened candidate lenses through the sorting model. The sorting model considers two standards, namely the eye axis growth speed and the type of a difference map of corneal topography before and after lens matching, so that the accuracy of personalized treatment is improved. Through objective evaluation of the eye axis growth speed and the difference map type of the corneal topography map, subjectivity in a traditional method is avoided, and the shaping effect can be evaluated more accurately.

Description

Technical Field

[0001] The present invention relates to the field of vision technology, and in particular to a multi-modal intelligent fitting method for orthokeratology lenses based on multiple models, a multi-modal intelligent fitting system for orthokeratology lenses based on multiple models, an electronic device, and a computer-readable storage medium. Background Art

[0002] Orthokeratology lenses, also known as OK lenses, are a special type of rigid gas-permeable contact lenses that temporarily correct myopia by changing the shape of the anterior corneal surface, especially suitable for children and adolescents to control the progression of myopia. The fitting process of orthokeratology lenses includes the following key steps: I. Pre-fitting examination for the first time Medical history inquiry: The doctor will inquire in detail about the patient's current medical history, past medical history, family history, allergy history, etc. to comprehensively understand the patient's ocular and systemic conditions.

[0003] Specialist examination: including visual acuity, slit lamp examination, intraocular pressure measurement, refractive power examination, fundus examination, corneal curvature measurement, corneal topography examination, corneal endothelial cell count, axial length measurement, corneal diameter measurement, pupil diameter measurement, tear test, etc. to evaluate the patient's eye structure, refractive state, and corneal health status.

[0004] II. Design parameters and trial fitting and ordering lenses Based on the results of the pre-fitting examination for the first time, the doctor will select a suitable brand of orthokeratology lenses for the patient and design lens parameters suitable for the patient. Subsequently, the doctor or optometrist will perform corneal fluorescein sodium staining and observe the lens fitting situation under the slit lamp to make refined parameter adjustments to determine the most suitable lens parameters. After the parent or patient confirms, an order will be sent to the manufacturer to enter the lens ordering process.

[0005] III. Examination on the day of picking up the lenses After the lenses arrive, the doctor will conduct an examination on the day of picking up the lenses. This includes wearing the lenses examination and removing the lenses examination to verify the lens information and confirm that the patient's ocular and systemic health conditions are good. At the same time, the doctor will guide the patient or parent on how to correctly and safely put on and take off the lenses, and how to perform daily care.

[0006] IV. First-day follow-up visit after wearing the lenses On the first day of wearing the lenses, the patient needs to have a follow-up visit within 2 hours after waking up in the morning. The doctor will check the patient's wearing lens visual acuity and naked eye visual acuity, as well as conduct a slit lamp examination and a corneal topography examination to evaluate the lens fitting situation and correction effect.

[0007] V. Post-fitting review After wearing corneal reshaping lenses, patients need to have regular follow-up examinations. Usually, it includes the 1st day, 1 week, and 1 month after the first wearing, and then once every 3 months. The follow-up content covers vision, refractive power, ocular surface condition, lens fitting status and lens condition, as well as corneal topography, etc. Regular follow-up of corneal thickness, axial length of the eye, corneal endothelial cells, etc. is carried out every 6 months to monitor the corneal health status. The doctor will give personalized guidance to the patient according to the follow-up results, including adjustment of wearing time, improvement of nursing methods, etc., to ensure the safety and effectiveness of the patient wearing corneal reshaping lenses.

[0008] However, at present, the parameter selection in the fitting process of corneal reshaping lenses (OK lenses) usually depends on the experience of ophthalmologists and the auxiliary diagnosis of some equipment, and there are the following deficiencies in its fitting method: The parameter selection is not precise enough: Due to the limited experience of doctors, all individual differences cannot be fully considered, which may lead to the selected lens parameters not fully meeting the needs of patients and affecting the treatment effect; Lack of precise evaluation of reshaping effect: Most of the existing reshaping effect evaluation methods are subjective evaluations, and the growth rate of the eye axis and the types of changes in corneal topography are not fully considered; Lack of individualized programs: The corneal morphology and eye axis growth of each patient are different, and traditional parameter selection cannot achieve a completely individualized treatment program. Summary of the Invention

[0009] In order to solve the technical problems existing in the prior art, the present invention provides the following technical solutions: On the one hand, a multi-modal intelligent fitting method for corneal reshaping lenses based on multiple models is provided. This method is implemented by an electronic device, and this method includes: Input the basic data of the patient, including corneal topography and the basic information of the patient. The basic information of the patient includes physiological parameters: axial length of the eye, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power; Input the corneal topography of the patient into a preset recall model. The recall model identifies the depth image features therein and predicts the lens parameters corresponding to the depth image features. Based on the predicted lens parameters, a candidate interval for the candidate lenses to be recalled is generated, and K pairs of candidate lenses are screened out according to the candidate interval; Input the lens parameters of each candidate lens, the corneal topography of the patient, and the basic information of the patient into a preset sorting model in sequence. The sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual growth rate of the eye axis and the type of reshaping effect difference map, and sorts out the candidate lenses whose reshaping effect difference map type is the 1st or 2nd type; Through the sorting scoring formula, the axial length annual growth rate and the orthokeratology effect difference map type of each sorted candidate lens are scored for the orthokeratology effect, and the candidate lens with the maximum orthokeratology effect score is selected as the final lens recommended to the patient; Output the lens parameters of the final lens.

[0010] Preferably, the generation method of the recall model includes: Collect the corneal topographies of several patients and at least the following lens parameters of the corneal orthokeratology lenses they wear: AC1, diameter or toricity; Pre-construct a multi-task regression model based on ResNet50 and perform training configuration. Among them, the multi-task regression model includes a backbone network and a multi-task regression head including multiple branch networks. Among them: the backbone network is used to learn the depth image features of the corneal topography using ResNet50, and the multi-task regression head is used to learn various lens parameters corresponding to the depth image features. The configuration information includes the loss function: , λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss terms respectively; , , are the Loss terms of AC1, diameter, and toricity respectively: is the total Loss term of the multi-task regression model; Add the corresponding various lens parameters to the depth image features, count and obtain a feature set composed of the depth image features of each patient, and divide it into a training set, a validation set, and a test set according to a ratio; Input the training set into the multi-task regression model, and perform regression training and learning on the depth image features of each patient and their corresponding lens parameters through the multi-task regression model to generate the initial recall model; Use the test set to test the recall model, and use the validation set to compare and verify the prediction results of the recall model: If the verification passes, deploy and apply the recall model; Otherwise, repeat the above steps to regenerate the recall model.

[0011] Preferably, the generation method of the sorting model includes: Pre - construct a multi - task regression model based on ResNet50 and perform training configuration. Among them, the multi - task regression model includes a backbone network and a task regression head of a branch network, where: the backbone network is used to learn the annual axial length growth rate corresponding to the basic corneal feature vector by using ResNet50, and the task regression head is used to learn the type of reshaping effect difference map corresponding to the basic corneal feature vector. The configuration information includes a loss function: , where α and β are adjustable weights, Lgrowth is MSE, and Lclassification is cross - entropy loss; Collect the lens parameters of the corneal reshaping lenses worn by several patients, their corneal topographies after corneal reshaping, and the basic patient information. Among them, the basic patient information includes physiological parameters: axial length, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power; Extract the corneal topography features in the corneal topographies of each patient after wearing the corneal reshaping lens and the data features in the basic patient information. Among them, the data features at least include the annual axial length growth rate of the axial length changing with time; Perform feature splicing on the lens parameters, the corneal topography features, and the data features to generate the basic corneal feature vector of each patient; According to the corneal topography features of each patient after wearing the corneal reshaping lens, evaluate the corneal reshaping effect of each patient and generate the corresponding type of reshaping effect difference map, and label the corresponding type of reshaping effect difference map for the basic corneal feature vector. The rules for evaluating the type of reshaping effect difference map according to the corneal reshaping effect of each patient are as follows: Type 1: The defocus ring in the image is located at the center of the cornea, centered and completely intact, indicating good reshaping effect and relatively ideal correction effect; Type 2: The defocus ring in the image may have a slight deviation, but the overall structure is still intact, indicating that the correction effect is acceptable, but there is slight uneven reshaping; Type 3: The defocus ring in the image is significantly deviated from the center, or its structure is incomplete, indicating poor correction effect or significant deviation; Collect the basic corneal feature vectors of each patient to form a feature set, and divide it into a training set, a validation set, and a test set according to a preset ratio; Input the training set into the multi - task regression model, and perform regression training and learning on the basic corneal feature vectors of each patient through the multi - task regression model to generate the initial ranking model; Test the sorting model using the test set, and use the validation set to compare and verify the prediction results of the sorting model: If the verification passes, deploy and apply the sorting model; Otherwise, repeat the above steps to regenerate the sorting model.

[0012] Preferably, when using the validation set to compare and verify the prediction results of the sorting model, the verification metrics include: Metric 1: Mean Squared Error (MSE) and Mean Absolute Error (MAE) between the validation set label and the test set label; And, Metric 2: The shaping effect score of the lens, When both metrics meet the preset values, the verification is qualified.

[0013] Preferably, the sorting score formula: , Where: GrowthRate is the annual eye axis growth rate, with the unit of mm / year. The lower the value, the better the shaping effect. Take the reciprocal to reflect the minimization goal; TypeScore is the score of the shaping effect difference map type: Type 1: TypeScore = 1.0 (best effect); Type 2: TypeScore = 0.7 (sub-optimal effect); Type 3: TypeScore = 0.0 (poor effect, not included in the score); ω1 and ω2 are the corresponding weight parameters, representing the importance of the eye axis growth rate and the difference map type, and are adjusted according to clinical needs: If eye axis growth control is more important, ω1 is larger (preferably 0.7); if the difference map type is more important, ω2 is larger (preferably 0.3); is the shaping effect score of the lens.

[0014] Preferably, the method further includes: Input the lens parameters of the final lens recommended to the patient, the corneal topographic map of the patient, and the basic information of the patient into a preset sorting model. The sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual eye axis growth rate and the type of shaping effect difference map; Generate a shaping effect score for the final lens according to the annual eye axis growth rate and the type of shaping effect difference map; Retrieve and output the corneal reshaping lens wearing treatment plan corresponding to the shaping effect score of the final lens from the background database according to the shaping effect score.

[0015] On the other hand, a multi-model based multi-modal orthokeratology lens intelligent fitting system is provided. The multi-model based multi-modal orthokeratology lens intelligent fitting system is used to implement the above-mentioned multi-model based multi-modal orthokeratology lens intelligent fitting method. The system includes: An input module, configured to input basic patient data, including corneal topographies and basic patient information. The basic patient information includes physiological parameters: axial length of the eye, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power. A recall module, configured to input the corneal topography of the patient into a preset recall model. The recall model identifies depth image features therein and predicts lens parameters corresponding to the depth image features, generates a candidate interval of candidate lenses to be recalled based on the predicted lens parameters, and screens out K pairs of candidate lenses according to the candidate interval. A sorting module, configured to sequentially input the lens parameters of each candidate lens, the corneal topography of the patient, and the basic patient information into a preset sorting model. The sorting model identifies corneal basic feature vectors therein and outputs corresponding annual axial growth rate of the eye and types of shaping effect difference maps, and sorts out the candidate lenses whose types of shaping effect difference maps are the first type or the second type. A recommendation module, configured to perform a shaping effect score on the annual axial growth rate of the eye and the types of shaping effect difference maps of each sorted candidate lens through a sorting score formula, and screen out the candidate lens with the maximum shaping effect score as the final lens recommended to the patient. An output module, configured to output the lens parameters of the final lens.

[0016] On the other hand, an electronic device is provided. The electronic device includes: a processor; a memory, on which computer-readable instructions are stored. When the computer-readable instructions are executed by the processor, any one of the methods in the above-mentioned multi-model based multi-modal orthokeratology lens intelligent fitting method is implemented.

[0017] On the other hand, a computer-readable storage medium is provided. At least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement any one of the methods in the above-mentioned multi-model based multi-modal orthokeratology lens intelligent fitting method.

[0018] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include: The present invention selects appropriate lens parameters and evaluates the shaping effect by combining a recall model and a ranking model. The recall model automatically selects the parameters that best match the patient's corneal morphology from the candidate lenses, making up for the shortcomings of the traditional experience-based method. The ranking model is used to optimize lens selection: among the selected candidate lenses, the ranking model is used to further select the lens with the best shaping effect. The ranking model takes into account two major criteria: the axial growth rate and the type of corneal topography difference map before and after fitting the glasses, thereby improving the accuracy of personalized treatment. Through objective evaluation of the axial growth rate and the type of corneal topography difference map, the subjectivity in the traditional method is avoided, and the shaping effect can be evaluated more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 It is a schematic diagram of corneal topography; Figure 2 is a structural diagram of a recall model provided by an embodiment of the present invention; Figure 3 is a schematic diagram of screening of multimodal orthokeratology lenses based on multiple models provided by an embodiment of the present invention; Figure 4 is a structural schematic diagram of a sorting model provided by an embodiment of the present invention; Figure 5 It is a block diagram of a multi-model-based multi-modal orthokeratology lens intelligent fitting system provided by an embodiment of the present invention; Figure 6 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0022] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0023] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0024] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are consistent.

[0025] Terminology explanation: Corneal topography, the full name of which is computer-assisted corneal topographic analysis system, is an inspection technology that uses a computer image processing system to digitally analyze the corneal morphology. It presents the information obtained in color images with different characteristics. Because it looks like the ups and downs of the terrain surface in geology, it is called corneal topography. Corneal topography can accurately measure and analyze the curvature and refractive power of any point on the entire anterior surface of the cornea. It is a systematic and comprehensive quantitative analysis method for studying the anterior surface morphology of the cornea. Its principle is mainly to use a computer full-scale image processing system to photograph or capture the image or interference fringes projected onto the corneal surface. After automatic processing by professional image program software, the required image is intelligently digitized, and then drawn into an eye corneal topography with color coding, so that qualitative and quantitative information on the curvature of the anterior surface of the cornea can be obtained intuitively, thoroughly and accurately. Corneal topographers usually consist of three parts: Placido's disk projection system, real-time image monitoring system and computer image processing system. When used, the corneal topographer projects a series of concentric rings evenly onto the corneal surface, and then captures these ring images through a real-time image monitoring system and stores them in a computer. The computer image processing system then digitizes these images and applies preset mathematical calculation formulas and programs for analysis, and finally displays the analysis results on the screen in the form of color images.

[0026] AC1: Usually expressed in "mm", refers to the radius of curvature of the center area of ​​the lens. The smaller the unit, the greater the curvature.

[0027] Diameter: The diameter of an OK lens refers to the overall diameter of the lens, usually measured in millimeters (mm). It determines the range of the lens covering the cornea, and the appropriate lens diameter is usually selected based on the patient's corneal diameter and the shape of the eyeball.

[0028] Toricity: Refers to the gradually changing part of the anterior surface curvature of the OK lens from the center to the edge of the lens. It controls the curvature change of the lens from the center to the edge and usually includes several regions: the central region (AC1), the transition zone (BC, Base Curve), and the peripheral curve (PC, Peripheral Curve).

[0029] Example: As shown in the Figure 1 topography of a patient with mild myopia (-1.00D to -2.50D), the following can be measured on the corneal topography: AC1: 40.50mm Explanation: Applicable to patients with mild myopia, with a larger AC1, the curvature of the central region of the lens is flatter, providing a mild reshaping effect and avoiding overcorrection.

[0030] Diameter: 10.6mm Explanation: A smaller diameter is suitable for a smaller cornea, ensuring that the lens can comfortably cover the cornea and avoiding interference from the lens edge.

[0031] Toricity: PC1 (Peripheral Curve 1): 42.50mm Explanation: A flatter toricity design is suitable for patients with a less curved cornea, providing higher wearing comfort and ensuring the stability of the lens edge without affecting the reshaping effect.

[0032] The present invention mainly uses an AI model to recommend personalized corneal reshaping lenses for patients.

[0033] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0034] An embodiment of the present invention provides a multi-model based multi-modal intelligent fitting method for corneal reshaping lenses. This method can be implemented by an electronic device, which can be a terminal or a server. As shown in the Figure 1 flowchart of the multi-model based multi-modal intelligent fitting method for corneal reshaping lenses, the processing flow of this method can include the following steps: S1. Input the basic data of the patient, including the corneal topography and the basic information of the patient. The basic information of the patient includes physiological parameters: axial length of the eye, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power; S2. Input the corneal topographic map of the patient into a preset recall model. The recall model identifies the depth image features therein and predicts the lens parameters corresponding to the depth image features, generates a candidate interval for the candidate lenses to be recalled based on the predicted lens parameters, and screens out K pairs of candidate lenses according to the candidate interval; S3. Input the lens parameters of each of the candidate lenses, the corneal topographic map of the patient, and the basic information of the patient into a preset sorting model in sequence. The sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual axial length growth rate and the type of reshaping effect difference map, and sorts out the candidate lenses whose type of reshaping effect difference map is the first type or the second type; S4. Through a sorting scoring formula, perform a reshaping effect score on the annual axial length growth rate and the type of reshaping effect difference map of each sorted candidate lens, and screen out the candidate lens with the maximum reshaping effect score as the final lens recommended to the patient; S5. Output the lens parameters of the final lens.

[0035] The present invention mainly includes several steps: input, recall candidate lenses, sort the candidate lenses, recommend the final lens, and output. Input: Responsible for receiving the basic data of the patient, including information such as corneal topographic map, axial length, age, gender, etc.

[0036] Recall model: Use the corneal topographic map before fitting as input, design a multi-task regression model based on resnet50, predict each parameter (AC1, diameter, toricity) of the vst lens, and recall the candidate parameters of the corneal reshaping lens that meet the patient's needs.

[0037] Sorting model: Further screen the recalled lens candidates. The model inputs include user information (basic information such as axial length, age, gender, etc.), corneal topographic map, and lens parameters. Evaluate the reshaping effect (annual axial length growth rate and type of difference map) of the lens through the sorting model, and select the lens with the best effect.

[0038] Effect evaluation: Evaluate the reshaping effect of each lens according to the annual axial length growth rate and the type of corneal topographic map difference map.

[0039] Output: Output the lens parameters of the finally selected lens and the recommended treatment plan.

[0040] Collaborative optimization between steps: 1. Input and models (recall and sorting): Data preprocessing optimizes the input quality of the recall model and reduces the model training error.

[0041] 2. Recall model and sorting model: The quality of the candidate lens pool directly affects the output effect of the sorting model.

[0042] 3. Sorting Model and Effect Evaluation: The evaluation module verifies the accuracy of the sorting results and improves the reliability of the final lens selection.

[0043] 4. Output and Effect Evaluation: The output module can dynamically display the evaluation results, providing an intuitive decision-making basis for doctors.

[0044] The specific implementation plans for each step will be described in detail below.

[0045] 1. Input Collect comprehensive eye data of patients: Corneal topographies, which reflect the curvature distribution and morphological characteristics of the cornea.

[0046] Physiological parameters: axial length of the eye, corneal diameter, central corneal thickness.

[0047] Demographic data: basic information such as age, gender, etc.

[0048] Genetic and medical history information: including family myopia degree, whether there is a history of keratoconus, etc.

[0049] Refraction data: uncorrected visual acuity, corrected visual acuity, refractive power, etc.

[0050] Data Preprocessing Image features: Standardize the corneal topographies through preprocessing (such as denoising, resolution unification) to ensure the consistency of subsequent model input.

[0051] The corneal topography is an image input (such as an RGB image of 224×224×3), and a convolutional neural network (CNN) will be used to extract image features.

[0052] Preprocessing: Image normalization: Normalize the pixel values of the corneal topographies to the range of [0,1] to reduce the impact of brightness and contrast differences on model training.

[0053] Data augmentation: To improve the robustness and generalization ability of the model, perform augmentation operations such as rotation, translation, scaling, and mirror flipping on the images to simulate the changes of the corneas of different patients.

[0054] 1) Feature Extraction: Use ResNet50 (or other deep convolutional networks such as VGG16, DenseNet, etc.) as the feature extractor. ResNet50 can effectively extract deep spatial and texture features through residual blocks.

[0055] Output: By removing the fully connected layer of ResNet50 (i.e., extracting the output of the penultimate layer of ResNet50), use its output feature vector (usually 2048-dimensional) as the image feature.

[0056] Schematic diagram: Input: Corneal topographic map (224×224×3) Feature extraction: ResNet50 network (2048 - dimensional features) Numerical features: Normalize continuous variables (such as axial length of the eye); perform one - hot encoding on discrete variables (such as gender).

[0057] For numerical features (such as axial length of the eye, age, etc.), complex image processing is not required, but it is necessary to ensure that they can be effectively input into the neural network.

[0058] Standardization / Normalization: For continuous variables (such as axial length of the eye, age, etc.), standardization or normalization is usually required. The standardization operation can transform the features into data with a mean of 0 and a standard deviation of 1, thus avoiding the impact of scale differences between different features on the training process.

[0059] For example, the standardization formula for the axial length of the eye is: , where μ is the mean of the axial length of the eye and σ is the standard deviation.

[0060] Normalization compresses the data into the range [0, 1], which is suitable for input into the model.

[0061] Discrete feature processing (such as gender, family medical history): For discrete features (such as gender, family medical history, etc.), One - Hot Encoding is usually used to convert them into binary features of 0 and 1. For example, the gender feature can be converted into [1, 0] (male) or [0, 1] (female).

[0062] Example: Input: Axial length of the eye, age, gender (after standardization / normalization and one - hot encoding); Output: Processed numerical feature vector (such as [0.32, 0.45, 1, 0]).

[0063] 2) Data integration: Uniformly convert all input data into matrix and tensor formats to ensure compatibility with subsequent deep learning models.

[0064] Fusion of image features and numerical features: The part of image feature extraction outputs a high - dimensional vector (1024 - dimensional). Numerical features (such as age, axial length of the eye, etc.) are usually a low - dimensional vector (10 - dimensional, etc.).

[0065] Concatenation: Concatenate the image feature vector and the numerical feature vector dimension - by - dimension to form a larger feature vector. For example, if the image feature has 1024 dimensions and the numerical feature has 5 dimensions, the concatenated feature vector has 1024 + 5 = 1029 dimensions.

[0066] EmbeddingLayer: For discrete categorical features (such as gender, family medical history, etc.), use the EmbeddingLayer to transform them into dense vectors, and concatenate these dense vectors with other features.

[0067] For example, gender (with 2 categories) can be mapped by the EmbeddingLayer into a low - dimensional vector (2 - dimensional). In this way, the low - dimensional embeddings of discrete features concatenated with continuous features can be input into the subsequent network.

[0068] Fused features: Concatenated vector: A fused vector containing image features and numerical features.

[0069] Schematic diagram: Input: Image feature vector (1024 dimensions)+Numerical feature vector (10 dimensions)+Discrete feature vector (2 dimensions); Output: Concatenated fused vector (1036 dimensions).

[0070] After the input data is pre - processed, it can be input into the subsequent recall model and ranking model for the next step of feature recognition and prediction. Data pre - processing optimizes the input quality of the recall model and reduces the model training error.

[0071] 2. The recall model is mainly used to identify the depth features of the patient's corneal topographic map and predict the lens parameters of the orthokeratology lens corresponding to the features. Let the system generate a candidate interval of the candidate lenses to be recalled according to the predicted lens parameters (predicted values) of the orthokeratology lens, and screen out K pairs of candidate lenses according to the candidate interval.

[0072] The recall model uses a multi - task regression model based on ResNet50 to predict and output the candidate lenses for the patient. The main content is as follows: 1) Model structure of the recall model As shown in the appendix Figure 2 Shown below: Network architecture: A multi - task regression model based on ResNet50. Backbone network: ResNet50 is used to extract the depth features of the corneal topographic map and generate multi - level convolutional feature maps.

[0073] Multi-task Heads: It includes two branches, which respectively predict the lens parameters AC1, diameter, and toricity.

[0074] Parameter adjustment: The model depth, learning rate, and activation function are adjusted through hyperparameter optimization to improve the prediction accuracy.

[0075] 2) Working principle Model structure design Backbone network: ResNet50 is adopted for extracting the depth features of corneal topography.

[0076] Input: The preprocessed corneal topography (224×224×3).

[0077] Convolution module: The residual blocks of ResNet50 are used to capture multi-scale features.

[0078] Feature output: The flattened feature vector (128-dimensional).

[0079] Multi-task regression head: Branch 1: Predicts the lens parameter AC1.

[0080] Branch 2: Predicts the lens parameter diameter.

[0081] Branch 3: Predicts the lens parameter toricity.

[0082] Activation function: Linear Activation is used, which is suitable for regression tasks.

[0083] Loss function: The mean squared error (MSE) is adopted for the two branches respectively. The final loss is: , λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss terms respectively; , , The Loss terms for AC1, diameter, and toricity are respectively: is the total Loss term of the multi-task regression model; λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss terms, which are used to balance the influence of different objectives on optimization.

[0084] 3) Model training: Collect the corneal topographies of several patients and at least the following lens parameters of the orthokeratology lenses they wear: AC1, diameter, or toricity; For details, please refer to the previous section "Input and Its Preprocessing".

[0085] 1. 1 Data Preparation: a. Input: Corneal topographies and their corresponding true lens parameters (AC1, diameter, toricity).

[0086] b. Data Augmentation: Rotate, scale, and normalize the color of the topographies to improve the robustness of the model.

[0087] c. Data Partitioning: 70% training set, 15% validation set, 15% test set.

[0088] Features and Training Process 1.2 Training Configuration: a. Optimizer: Adam optimizer, learning rate set to 1e−4, using a learning rate decay strategy.

[0089] b. Batch Size: 32.

[0090] c. Number of Training Epochs: 50 epochs, monitoring the loss of the validation set.

[0091] 1.3 Model Evaluation: a. Metrics: Mean Squared Error (MSE) and Mean Absolute Error (MAE).

[0092] b. Evaluation Method: is the label of the validation set (representing a data in the set), is the label of the test set, and N is the total number of evaluation samples: .

[0093] c. Recall Rate: For all candidate lenses within the range of true parameters, evaluate the candidate coverage (Coverage) of the recall model.

[0094] Subsequently, the corneal topography of the patient can be input, and the input corneal topography of the patient can be recognized by the recall model. The recall model recognizes the depth image features therein (adding the corresponding multiple lens parameters: AC1 (AC2 can also be included), diameter, toricity) and predicts the lens parameters corresponding to the depth image features. Based on the predicted lens parameters, a candidate interval for the candidate lenses to be recalled is generated, and K pairs of candidate lenses are selected according to the candidate interval. After the model recognizes the features, the predicted values of the corresponding lenses can be output.

[0095] 1.4 Candidate Generation: Generate a candidate interval based on the predicted values (such as AC1±0.25, AC2±0.25), and select K pairs of candidate lenses. As shown in the appendix Figure 3As shown, through the recall model, multiple candidate lenses can be recalled. Here, 6 pairs of candidate lenses are taken as an example. The parameters of these candidate lenses will be input into the subsequent sorting model, and the sorting model will perform recall sorting based on the parameters of the input candidate lenses and the patient's basic information (user characteristics Figure 6 in the attachment), predict the patient's personalized characteristics "annual axial length growth rate" and "orthokeratology effect difference map type", select the final lens and recommend it to the patient.

[0096] When generating the candidate interval based on the predicted value here, the fluctuation factor can be adjusted according to the patient's basic information. Here, the candidate interval is such as AC1±0.25 (0.25 is the fluctuation value), which is just an example. For example, the fluctuation values of the predicted values "AC1, diameter, toricity" can be adjusted by combining the personalized factors in the patient's basic information. The example is as follows: Let: Fluctuation value = / 10, is the corneal comprehensive recommendation index, and the calculation is as follows: , where the definitions of each character and letter are shown in the following table: Symbol Parameter Name Definition and Calculation Clinical Weight Range AXL Axial Length Anterior-posterior diameter of the eyeball (unit: mm), normal range 22 - 26mm α=0.3 CD Corneal Diameter Horizontal visible iris diameter (unit: mm), affecting the selection of lens diameter α=0.3 AGE Age Actual age of the patient (unit: years), defocus design needs to be considered for adolescents β=0.2 FD_{risk} Genetic Risk Factor Family average myopia degree / 100 + keratoconus history marker (yes = 1, no = 0) γ = 0.15 SE Spherical Equivalent Diopter (spherical lens + 0.5 × cylindrical lens), unit: D δ = 0.35 UCVA Uncorrected Visual Acuity Uncorrected vision (such as 0.5) δ = 0.35 BCVA Best Corrected Visual Acuity Best corrected vision (such as 1.0) δ = 0.35 By considering the patient's physiological parameters, corneal data, medical history, and optometry data, etc., generating the corneal comprehensive recommendation index and calculating the fluctuation value, the final recommendation can be obtained comprehensively, so as to recommend candidate lenses that meet the patient's individual characteristics. For example, the axial length may affect the lens power, the corneal diameter and central thickness may involve the adjustment of the lens diameter and sagittal height. Age and gender may affect the choice of lens type (such as progressive lenses for teenagers). Family myopia degree and keratoconus history may require adjustment of the lens design to avoid risks. The diopter in the optometry data is directly related to the lens power, and the uncorrected and corrected visual acuity may affect whether special design is needed (such as defocus lenses).

[0097] The above weights can be designed from clinical experience values, which are just examples.

[0098] Therefore, here, by sharing the feature extraction layer of ResNet50 and optimizing multiple objectives at the same time, the model efficiency and accuracy are improved; the prediction result has self-adaptability and can dynamically adjust the screening range of candidate lenses according to the patient's characteristics.

[0099] Next, the sorting model will comprehensively predict the best lens recommended for the patient based on the patient's basic information and the parameters of the above candidate lenses.

[0100] 3. The sorting model is also constructed based on the multi-task regression model of ResNet50.

[0101] The multi-task regression model includes a backbone network and a task regression head of a branch network, where: the backbone network is used to learn the annual axial length growth rate corresponding to the basic corneal feature vector by using ResNet50, and the task regression head is used to learn the type of reshaping effect difference map corresponding to the basic corneal feature vector.

[0102] The backbone network ResNet50 can perform image feature learning of corneal topography, and the branch network can learn other parameters, thereby forming a multi-task regression model composed of a backbone network and a branch network, which will be used as the training model of the ranking model for initial training.

[0103] 1) The model architecture of the ranking model is as follows: Multi-modal input layer: Different features are processed through multiple independent embedding layers; Feature fusion layer: The fully connected layers (FullyConnectedLayers) merge the image features and the embedding features; Multi-task branches: Independent output branches are designed for regression and classification tasks respectively.

[0104] 2) Input features of the ranking model (implemented in combination with the previous input and its preprocessing): Discrete features: including age, gender, family medical history, etc. (encoded as low-dimensional vectors through the embedding layer).

[0105] Lens parameters: AC1 (directly input, and AC2 can also be added).

[0106] Image features: Corneal topography features (extracted by ResNet50).

[0107] Fusion features: The above three types of features are concatenated to generate a unified feature vector (manually labeled difference map type).

[0108] 3) Multi-task output: Task 1: Predict the annual axial length growth rate (continuous value).

[0109] Task 2: Predict the type of reshaping effect difference map (classification task, 3 classes).

[0110] Activation function: Task 1: Linear activation, used for regression; Task 2: Softmax activation, used for classification; Loss function: , where α and β are adjustable weights, Lgrowth is MSE (refer to the above mean square error), and Lclassification is the cross-entropy loss (calculated by the system itself); 4) Training process 1. Data Preparation: a. Input: Candidate lenses of the recall model, corneal topographies, and patient personal characteristics.

[0111] b. Data Augmentation: Operations such as horizontal flipping and random cropping are performed on the corneal topographies.

[0112] 2. Training Configuration: a. Optimizer: SGD (with momentum), learning rate is set to 0.01.

[0113] b. Regularization: Dropout and Weight Decay are used to prevent overfitting.

[0114] c. Batch Size: 64.

[0115] 3. Model Evaluation: Among them, α and β are adjustable weights, Lgrowth is the MSE (refer to the mean squared error above), and Lclassification is the cross-entropy loss (calculated by the system itself); 4) Training Process 1. Data Preparation: a. Input: Candidate lenses of the recall model, corneal topographies, and patient personal characteristics.

[0116] b. Data Augmentation: Operations such as horizontal flipping and random cropping are performed on the corneal topographies.

[0117] 2. Training Configuration: a. Optimizer: SGD (with momentum), learning rate is set to 0.01.

[0118] b. Regularization: Dropout and Weight Decay are used to prevent overfitting.

[0119] c. Batch Size: 64.

[0120] 3. Model Evaluation: , Where: GrowthRate is the annual eye axis growth rate, with the unit of mm / year. The lower the value, the better the shaping effect. The reciprocal is taken to reflect the minimization target; TypeScore is the score for the type of shaping effect difference map: Type 1: TypeScore = 1.0 (best effect); Type 2: TypeScore = 0.7 (sub-optimal effect); Type 3: TypeScore = 0.0 (poor effect, not included in the score); ω1 and ω2 are the corresponding weight parameters, representing the importance of the axial length growth rate and the type of difference map, which are adjusted according to clinical needs: if the control of axial length growth is more important, ω1 is larger (preferably 0.7); if the type of difference map is more important, ω2 is larger (preferably 0.3); is the shaping effect score of the lens.

[0121] As shown in the Figure 4 sorting model structure shown, its input is the parameters of the candidate lenses of the recall model, the user characteristics of the patient, and the corneal topographic features. The user characteristics include physiological parameters: axial length, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power; Extract the corneal topographic features in the corneal topography after each patient wears the orthokeratology lens and the data features (user characteristics) in the basic information of the patient. Among them, the data features at least include the annual axial length growth rate of the axial length changing with time (the annual axial length growth rate of the patient changing with time can be obtained through statistical analysis); Perform feature splicing on the lens parameters, the corneal topographic features, and the data features to generate the basic corneal feature vectors of each patient; the example is as follows: Input: image feature vector (1024 dimensions) + numerical feature vector (10 dimensions) + discrete feature vector (2 dimensions); Output: the fused vector Ffused (1036 dimensions) after splicing.

[0122] Feature splicing methods, such as: Ffused = Concat(Fimg, Fnum, Fdisc) ∈ R di+dn+dd , Fimg, Fnum, and Fdisc are the feature vectors of the above-mentioned respective dimensions, di + dn + dd are the respective dimensions, Concat is the splicing operation along the feature dimension order, and the pyramid fusion strategy or attention mechanism weighting can also be used to optimize information integration: Fenhanced = MLP(Ffused) (MLP is a multi-layer perceptron).

[0123] Through the above indicators and sorting performance evaluation, double-index evaluation is achieved: Indicator 1: the mean square error (MSE) and mean absolute error (MAE) between the validation set label and the test set label; And, Indicator 2: the shaping effect score of the lens, When both indicators meet the preset values, the verification is qualified.

[0124] Therefore, it is possible to verify the accuracy of the model sorting result and improve the reliability of the final lens selection.

[0125] Combined with the attached Figure 4 As shown, the sorting model is a multi-modal multi-task model. Through corneal topography and patient personal characteristics, it is judged whether the lens can make the corneal topography after fitting reach Class 1 / 2 and the axial length growth rate is the slowest.

[0126] By sequentially inputting the lens parameters of each of the candidate lenses, the corneal topography of the patient, and the basic information of the patient into a preset sorting model, the sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual axial length growth rate and the type of reshaping effect difference map, and sorts out the candidate lenses whose reshaping effect difference map type is Class 1 or Class 2; sequentially judge all the lenses in the recalled lenses, and select the best lens as the final lens parameter.

[0127] And the rules for evaluating the type of reshaping effect difference map according to the corneal reshaping effect of each patient are as follows: Class 1: The defocus ring in the image is located at the center of the cornea, centered and completely intact, indicating good reshaping effect and relatively ideal correction effect; Class 2: The defocus ring in the image may have a slight deviation, but the overall structure is still intact, indicating that the correction effect is acceptable, but there is slight uneven reshaping; Class 3: The defocus ring in the image is significantly deviated from the center, or its structure is incomplete, indicating poor correction effect or significant deviation.

[0128] 4. Effect evaluation Based on the output of the sorting model, perform a secondary verification on the best lens parameters: Axial length growth rate: The lower the annual growth rate value, the better the correction effect; Difference map type: Preferably Class 1 or Class 2, representing good reshaping effect.

[0129] After sorting each candidate lens, it is possible to predict its impact on the annual axial length growth rate and the type of reshaping effect difference map of the patient. Therefore, the final lens is selected through the reshaping effect score.

[0130] For the reshaping effect score of the lens, please refer to the previous sorting score formula.

[0131] Sorting score formula, is the reshaping effect score of the lens. Through the above method, the best lens can be found from the candidate lenses.

[0132] Here, the candidate lens corresponding to the maximum value is set as the optimal lens.

[0133] 5. Visual output Dynamically generate a data report, including: A trend chart of the annual axial length growth rate; Comparison of corneal topographies before and after glasses fitting; An example of a difference chart of orthokeratology effects.

[0134] Comprehensive index verification: Evaluate the lens effect from multiple dimensions to ensure the scientific nature of the recommended results.

[0135] Visualization support: Provide doctors with an intuitive result display to assist in clinical decision-making.

[0136] Through the predicted output of the final lens, intelligent orthokeratology fitting for patients can be achieved, and the basic corneal feature vectors of the patients can be analyzed and predicted, and the corresponding annual axial length growth rate and the type of difference chart of orthokeratology effects can be output.

[0137] Here, a corresponding glasses fitting treatment plan can be further recommended for the patient.

[0138] Preferably, the method further includes: Input the lens parameters of the final lens recommended to the patient, the corneal topography of the patient, and the basic patient information into a preset sorting model. The sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual axial length growth rate and the type of difference chart of orthokeratology effects; Generate an orthokeratology effect score for the final lens according to the annual axial length growth rate and the type of difference chart of orthokeratology effects.

[0139] The step of performing a personalized orthokeratology effect score for the patient according to the lens parameters of the final lens should be understood in combination with the previous analysis steps of the orthokeratology effect score.

[0140] The system can retrieve and output a corneal orthokeratology lens wearing treatment plan corresponding to the orthokeratology effect score of the final lens from the background database according to the orthokeratology effect score.

[0141] Corneal orthokeratology lens wearing treatment plans corresponding to different orthokeratology effect scores are pre-configured in the background database, including: wearing time, reexamination, precautions, etc., which can be specifically determined by the doctor.

[0142] Therefore, through the above method, the following technical effects can be achieved: 1. Improve the accuracy of parameter selection Traditional methods rely on doctors' experience and tend to ignore minor individual differences. The present invention combines machine learning models, uses large-scale data to train recall models, and automatically screens parameters from the lens library to ensure high compatibility between the lens and the patient's cornea.

[0143] The improved accuracy is reflected in a significant reduction in the number of adjustments caused by improper fitting.

[0144] 2. Personalized treatment plan This system comprehensively analyzes the patient's individual characteristics (such as corneal topography and axial length growth trend) and the physical properties of the lens to design a fully customized fitting plan.

[0145] The ranking model considers multiple input variables and provides a better solution through weighted evaluation, so that each patient can get the most suitable lens.

[0146] 3. Provide objective evaluation criteria Traditional evaluation of eye shaping effects relies on subjective judgment and is easily influenced by experience. Human error can be avoided by using two objective indicators, namely, the growth rate of the eye axis and the type of difference graph.

[0147] Utilize multimodal data combined with deep learning models to establish an efficient and standardized evaluation system.

[0148] 4. Improve fitting efficiency and reduce costs The automated recall and sorting process significantly reduces the time it takes to select a doctor.

[0149] Accurate first-time fitting results reduce the chance of repeat fittings and reduce material and labor costs.

[0150] 5. Adapt to different patient scenarios The system supports multimodal input, which is suitable for both rapid intervention in children and long-term correction needs in adults.

[0151] 6. Easy to expand and upgrade The model is designed based on a deep learning framework and supports training optimization with more data.

[0152] The modular design allows for the addition of new diagnostic dimensions, such as ocular biomechanics data, to further enhance system performance.

[0153] Figure 5 The block diagram of a multi-model based multi-modal orthokeratology lens intelligent fitting system according to an exemplary embodiment is shown, and the system is used for a multi-model based multi-modal orthokeratology lens intelligent fitting method. Figure 5 ,in: An input module for inputting basic patient data, including corneal topography and basic patient information. The basic patient information includes physiological parameters: axial length of the eye, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power. A recall module for inputting the corneal topography of the patient into a preset recall model. The recall model identifies the depth image features therein and predicts the lens parameters corresponding to the depth image features, generates a candidate interval for the candidate lenses to be recalled based on the predicted lens parameters, and screens out K pairs of candidate lenses according to the candidate interval. A sorting module for sequentially inputting the lens parameters of each of the candidate lenses, the corneal topography of the patient, and the basic patient information into a preset sorting model. The sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual axial growth rate and the type of orthokeratology effect difference map, and sorts out the candidate lenses whose type of orthokeratology effect difference map is the first type or the second type. A recommendation module for performing an orthokeratology effect score on the annual axial growth rate and the type of orthokeratology effect difference map of each sorted candidate lens through a sorting score formula, and screening out the candidate lens with the maximum orthokeratology effect score as the final lens recommended to the patient. An output module for outputting the lens parameters of the final lens.

[0154] Please understand and implement the above-mentioned modules in combination with the respective steps in the above method, which will not be elaborated here.

[0155] Figure 6 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention, as Figure 6 shown. The electronic device may include the above-mentioned Figure 5 multi-model multi-modal orthokeratology intelligent fitting system shown. Optionally, the electronic device 410 may include a first processor 2001.

[0156] Optionally, the electronic device 410 may further include a memory 2002 and a transceiver 2003.

[0157] Among them, the first processor 2001 is connected to the memory 2002 and the transceiver 2003, such as through a communication bus.

[0158] Next, in combination with Figure 6 each component of the electronic device 410 will be specifically introduced: Among them, the first processor 2001 is the control center of the electronic device 410, which can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or can be an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. For example: one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs).

[0159] Optionally, the first processor 2001 can execute various functions of the electronic device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0160] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 6 the CPU0 and CPU1 shown in

[0161] In a specific implementation, as an embodiment, the electronic device 410 may also include multiple processors, such as Figure 6 the first processor 2001 and the second processor 2004 shown in

[0162] Among them, the memory 2002 is used to store software programs for implementing the solutions of the present invention and is controlled by the first processor 2001 for execution. The specific implementation method can refer to the above method embodiments and will not be elaborated here.

[0163] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or may also be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or any other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently and be coupled to the first processor 2001 through an interface circuit ( Figure 6 not shown) of the electronic device 410. The embodiments of the present invention do not make specific limitations thereto.

[0164] The transceiver 2003 is used to communicate with a network device or with a terminal device.

[0165] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 6 not shown separately). Among them, the receiver is used to implement the receiving function, and the transmitter is used to implement the sending function.

[0166] Optionally, the transceiver 2003 may be integrated with the first processor 2001 or may exist independently and be coupled to the first processor 2001 through an interface circuit ( Figure 6 not shown) of the electronic device 410. The embodiments of the present invention do not make specific limitations thereto.

[0167] It should be noted that Figure 6 the structure of the electronic device 410 shown in does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0168] In addition, the technical effects of the electronic device 410 may refer to the technical effects of the multi-model multi-modal orthokeratology lens intelligent fitting method described in the above method embodiments, and will not be elaborated here.

[0169] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0170] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM) or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM) and direct rambus RAM (DR RAM).

[0171] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0172] It should be understood that the term "and / or" in this document is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. Additionally, the character " / " in this document generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood by referring to the context before and after.

[0173] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0174] It should be understood that in various embodiments of the present invention, the magnitudes of the sequence numbers of the above processes do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0175] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0176] Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, the specific working processes of the devices, systems, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0177] In several embodiments provided by the present invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of systems or units can be in electrical, mechanical, or other forms.

[0178] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0179] In addition, the functional units in various embodiments of the present invention can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0180] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0181] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A multi-modal orthokeratology intelligent fitting method based on multiple models, characterized in that, The method includes: Inputting the patient's basic data, including corneal topography and the patient's basic information, where the patient's basic information includes physiological parameters: axial length of the eye, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power; Inputting the corneal topography of the patient into a preset recall model, where the recall model identifies the depth image features therein and predicts the lens parameters corresponding to the depth image features, generating a candidate interval of candidate lenses to be recalled based on the predicted lens parameters, and screening out K pairs of candidate lenses according to the candidate interval; Sequentially inputting the lens parameters of each of the candidate lenses, the corneal topography of the patient, and the patient's basic information into a preset sorting model, where the sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual axial growth rate and the type of reshaping effect difference map, and sorts out the candidate lenses whose type of reshaping effect difference map is the first type or the second type; Through a sorting scoring formula, performing a reshaping effect score on the annual axial growth rate and the type of reshaping effect difference map of each sorted candidate lens, and screening out the candidate lens with the maximum reshaping effect score as the final lens recommended to the patient; Outputting the lens parameters of the final lens.

2. The multi-modal orthokeratology intelligent fitting method based on multiple models according to claim 1, wherein The generation method of the recall model includes: Collecting the corneal topographies of several patients and at least the following lens parameters of the corneal reshaping lenses they wear: AC1, diameter or toricity; Pre-constructing a multi-task regression model based on ResNet50 and performing training configuration, where the multi-task regression model includes a backbone network and a multi-task regression head including multiple branch networks, where: the backbone network is used to learn the depth image features of the corneal topography using ResNet50, and the multi-task regression head is used to learn various lens parameters corresponding to the depth image features, and the configuration information includes a loss function: ， λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss terms respectively; , , are the Loss terms of AC1, diameter, and toric respectively: is the total Loss term of the multi-task regression model; Adding the corresponding various lens parameters to the depth image features, statistically obtaining a feature set composed of the depth image features of each patient, and dividing it into a training set, a validation set, and a test set according to a ratio; Inputting the training set into the multi-task regression model, and performing regression training and learning on the depth image features of each patient and their corresponding lens parameters through the multi-task regression model to generate the initial recall model; Testing the recall model using the test set, and comparing and validating the prediction results of the recall model using the validation set: If the validation passes, then deploy and apply the recall model; Otherwise, repeat the above steps to regenerate the recall model.

3. The multi-modal orthokeratology intelligent fitting method based on multiple models according to claim 1, wherein The generation method of the sorting model includes: Pre - construct a multi - task regression model based on ResNet50 and perform training configuration. Among them, the multi - task regression model includes a backbone network and a task regression head of a branch network, where: the backbone network is used to learn the annual axial length growth rate corresponding to the corneal basic feature vector by using ResNet50, and the task regression head is used to learn the type of reshaping effect difference map corresponding to the corneal basic feature vector. The configuration information includes a loss function: ， where α and β are adjustable weights, Lgrowth is MSE, and Lclassification is cross - entropy loss; Collect the lens parameters of the corneal reshaping lenses worn by several patients, their corneal topographies after corneal reshaping, and the basic patient information. Among them, the basic patient information includes physiological parameters: axial length, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power; Extract the corneal topography features in the corneal topographies of each patient after wearing the corneal reshaping lens and the data features in the basic patient information. Among them, the data features at least include the annual axial length growth rate of the axial length changing with time; Perform feature splicing on the lens parameters, the corneal topography features, and the data features to generate the corneal basic feature vector of each patient; According to the corneal topography features of each patient after wearing the corneal reshaping lens, evaluate the corneal reshaping effect of each patient and generate the corresponding type of reshaping effect difference map, and label the corresponding type of reshaping effect difference map for the corneal basic feature vector. Among them, the rules for evaluating the type of reshaping effect difference map according to the corneal reshaping effect of each patient are as follows: Type 1: The defocus ring in the image is located at the center of the cornea, centered and completely intact, indicating good reshaping effect and relatively ideal correction effect; Type 2: The defocus ring in the image may have a slight deviation, but the overall structure is still intact, indicating that the correction effect is acceptable, but there is slight uneven reshaping; Type 3: The defocus ring in the image is significantly deviated from the center, or its structure is incomplete, indicating poor correction effect or significant deviation; Collect the corneal basic feature vectors of each patient to form a feature set, and divide it into a training set, a validation set, and a test set according to a preset ratio; Input the training set into the multi - task regression model, and perform regression training and learning on the corneal basic feature vectors of each patient through the multi - task regression model to generate the initial sorting model; Use the test set to test the sorting model, and use the validation set to compare and verify the prediction results of the sorting model: If the verification passes, then deploy and apply the sorting model; Otherwise, repeat the above steps to regenerate the sorting model.

4. The multi-modal orthokeratology intelligent fitting method based on multiple models according to claim 3, characterized in that When using the validation set to compare and verify the prediction results of the sorting model, the verification metrics include: Metric 1: The mean square error (MSE) and mean absolute error (MAE) between the validation set labels and the test set labels; And, Index 2: The shaping effect score of the lens, When both indicators meet the preset values, the verification is qualified.

5. The multi-modal orthokeratology intelligent fitting method based on multiple models according to claim 1, characterized in that The sorting and scoring formula: ， Where: GrowthRate is the annual eye axis growth rate, in units of mm / year. The lower the value, the better the shaping effect. Take the reciprocal to reflect the minimization goal; TypeScore is the score for the type of shaping effect difference map: Type 1: TypeScore = 1.0 (optimal effect); Type 2: TypeScore = 0.7 (sub-optimal effect); Type 3: TypeScore = 0.0 (poor effect, not included in the score); ω1 and ω2 are the corresponding weight parameters, representing the importance of the eye axis growth rate and the difference map type, which are adjusted according to clinical needs: If eye axis growth control is more important, ω1 is larger (preferably 0.7); if the difference map type is more important, ω2 is larger (preferably 0.3); Score the shaping effect of the lens.

6. The multi-modal orthokeratology intelligent fitting method based on multiple models according to claim 1, wherein, The method further includes: Inputting the lens parameters of the final lens recommended to the patient, the corneal topographic map of the patient, and the basic information of the patient into a preset sorting model. The sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual eye axis growth rate and the type of shaping effect difference map; Generating a shaping effect score for the final lens according to the annual eye axis growth rate and the type of shaping effect difference map; Retrieving from the background database and outputting the corneal reshaping lens wearing treatment plan corresponding to the shaping effect score of the final lens according to the shaping effect score.

7. A multi-modal orthokeratology intelligent fitting system based on multiple models, which is used to implement the multi-modal orthokeratology intelligent fitting method based on multiple models according to any one of claims 1-6, characterized in that The system includes: An input module for inputting basic patient data, including the corneal topographic map and basic patient information. The basic patient information includes physiological parameters: eye axis length, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia degree, whether there is a history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, refractive power; A recall module for inputting the corneal topographic map of the patient into a preset recall model. The recall model identifies the depth image features therein and predicts the lens parameters corresponding to the depth image features, generates a candidate interval for the candidate lenses to be recalled based on the predicted lens parameters, and screens out K candidate lenses according to the candidate interval; A sorting module for sequentially inputting the lens parameters of each candidate lens, the corneal topographic map of the patient, and the basic information of the patient into a preset sorting model. The sorting model identifies the basic corneal feature vectors therein and outputs the corresponding annual eye axis growth rate and the type of shaping effect difference map, and sorts out the candidate lenses whose shaping effect difference map type is Type 1 or Type 2; A recommendation module for performing a shaping effect score on the annual eye axis growth rate and the type of shaping effect difference map of each sorted candidate lens through the sorting and scoring formula, and screening out the candidate lens with the maximum shaping effect score as the final lens recommended to the patient; An output module for outputting the lens parameters of the final lens.

8. An electronic device, characterized in that, The electronic device includes: A processor; A memory, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that, Program code is stored in the computer-readable storage medium, and the program code can be called by the processor to execute the method according to any one of claims 1 to 6.

Images