Multi-model-based multimodal orthokeratology lens intelligent fitting method and system

Abstract

The present invention provides a multi-model-based intelligent fitting method and system for multimodal orthokeratology lenses. Through the recall model, the parameters that best match the patient's corneal morphology are automatically screened out from candidate lenses, making up for the shortcomings of traditional experience-based methods; a ranking model is used to optimize lens selection: among the screened candidate lenses, the lenses with the best orthokeratology effect are further selected through the ranking model. 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 lenses, 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 of traditional methods is avoided, and the orthokeratology effect can be evaluated more accurately.

Description

Technical Field

[0001] The present invention relates to the field of visual technology, and in particular to a multi-model-based multi-modal orthokeratology lens intelligent fitting method and a multi-model-based multi-modal orthokeratology lens intelligent fitting system, electronic equipment and computer-readable storage medium. Background Art

[0002] Orthokeratology lenses, also known as OK lenses, are a special type of rigid gas-permeable contact lens that temporarily corrects myopia by changing the shape of the front surface of the cornea. They are particularly suitable for children and adolescents to control the progression of myopia. The fitting process of orthokeratology lenses includes the following key steps:

[0003] 1. Inspection before the first fitting

[0004] Medical history inquiry: The doctor will ask in detail about the patient's current medical history, past medical history, family history, allergy history, etc., to fully understand the patient's eye and systemic condition.

[0005] Specialized examinations: including visual acuity, slit lamp examination, intraocular pressure measurement, refractive power examination, fundus examination, keratometric 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 status and corneal health.

[0006] 2. Design parameters and trial fitting

[0007] Based on the results of the initial pre-fitting examination, the doctor will select an appropriate brand of orthokeratology lenses for the patient and design lens parameters tailored to the patient. The doctor or optometrist will then perform corneal fluorescein sodium staining, observe the lens fit under a slit lamp, and make detailed adjustments to determine the most appropriate lens parameters. Once the parent or patient confirms the selection, an order is placed with the manufacturer, and the lens ordering process begins.

[0008] 3. Examination on the day of lens removal

[0009] After your lenses arrive, your doctor will conduct a lens removal day examination. This includes both a wear and removal exam to verify the lens information and confirm your eye and general health. Your doctor will also instruct you or your parents on how to properly and safely remove and wear your lenses, as well as on daily lens care.

[0010] 4. Follow-up on the first day of wearing glasses

[0011] On the first day of wearing glasses, patients need to return for a follow-up appointment within 2 hours of waking up in the morning. The doctor will check the patient's visual acuity with and without glasses, and perform a slit lamp examination and corneal topography to assess the lens fit and correction effect.

[0012] 5. Post-match review

[0013] After wearing orthokeratology lenses, patients need to undergo regular follow-up examinations. This usually includes the first day, one week, and one month after the first wearing of the lenses, and then every three months. The review covers visual acuity, refractive power, ocular surface condition, lens fitting and lens condition, as well as corneal topography. Corneal thickness, axial length, corneal endothelial cells, etc. are regularly reviewed every six months to monitor the health of the cornea. Based on the review results, the doctor will provide personalized guidance to the patient, including adjustments to the wearing time, improvements to care methods, etc., to ensure the safety and effectiveness of the patient's wearing of orthokeratology lenses.

[0014] However, the current selection of parameters for orthokeratology lenses (OK lenses) during fitting usually relies on the experience of ophthalmologists and the auxiliary diagnosis of some equipment. This fitting method has the following shortcomings:

[0015] Inaccurate parameter selection: Due to the doctor's limited experience, it is impossible to fully consider all individual differences, which may result in the selected lens parameters not fully meeting the patient's needs and affecting the treatment effect;

[0016] Lack of accurate assessment of reshaping effects: Existing methods for assessing reshaping effects are mostly subjective and fail to fully consider the rate of axial growth and the type of corneal topography changes;

[0017] Lack of individualized solutions: The corneal morphology and axial length growth of each patient are different, and traditional parameter selection cannot achieve a completely personalized treatment plan. Summary of the Invention

[0018] In order to solve the technical problems existing in the prior art, the present invention provides the following technical solutions:

[0019] On the one hand, a multi-model-based multimodal orthokeratology lens intelligent fitting method is provided, which is implemented by an electronic device and includes:

[0020] Input basic patient data, including corneal topography and basic patient information. 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 and history of keratoconus; optometry data: naked eye visual acuity, corrected visual acuity, and refractive power;

[0021] Inputting the patient's corneal topography into a preset recall model, the recall model identifying depth image features therein and predicting lens parameters corresponding to the depth image features, generating candidate intervals for candidate lenses to be recalled based on the predicted lens parameters, and screening out K pairs of candidate lenses based on the candidate intervals;

[0022] The lens parameters of each candidate lens, the patient's corneal topography, and the patient's basic information are sequentially input into a preset ranking model, the ranking model identifies the corneal basic feature vector therein and outputs the corresponding annual axial length growth rate and shaping effect difference map type, and sorts the candidate lenses whose shaping effect difference map type is Category 1 or Category 2;

[0023] Using a ranking and scoring formula, scoring the annual growth rate of the axial length and the type of the shaping effect difference graph of each of the sorted candidate lenses is performed, and the candidate lens with the maximum shaping effect score is selected as the final lens recommended to the patient;

[0024] Output the lens parameters of the final lens.

[0025] Preferably, the recall model is generated by:

[0026] Corneal topography of several patients and at least the following lens parameters of the orthokeratology lenses they wore were collected: AC1, diameter or toricity;

[0027] A multi-task regression model based on ResNet50 is pre-built and trained. The multi-task regression model includes a backbone network and a multi-task regression head including multiple branch networks. 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 multiple lens parameters corresponding to the depth image features. The configuration information includes the loss function:

[0028] Loss Total =λ1×Loss AC1 +λ2×Loss Diameter +λ3×Loss PC ,

[0029] λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss items;

[0030] Loss AC1 、Loss Diameter 、Loss PC The Loss items are AC1, diameter, and annular curvature: Loss Total is the total Loss term of the multi-task regression model;

[0031] Adding the corresponding multiple lens parameters to the depth image features, counting and obtaining a feature set consisting of the depth image features of each patient, and dividing the feature set into a training set, a validation set, and a test set according to a proportion;

[0032] Inputting the training set into the multi-task regression model, performing regression training on the depth image features of each patient and the corresponding lens parameters through the multi-task regression model to generate the initial recall model;

[0033] The recall model is tested using the test set, and the prediction results of the recall model are compared and verified using the validation set:

[0034] If the verification passes, the recall model is deployed and applied;

[0035] Otherwise, repeat the above steps to regenerate the recall model.

[0036] Preferably, the generation method of the ranking model includes:

[0037] A multi-task regression model based on ResNet50 is pre-built and trained and configured, wherein the multi-task regression model includes a backbone network and a task regression head of a branch network, wherein: the backbone network is used to learn the annual growth rate of the axial length corresponding to the basic feature vector of the cornea using ResNet50, and the task regression head is used to learn the type of the shaping effect difference map corresponding to the basic feature vector of the cornea. The configuration information includes the loss function:

[0038]

[0039] Among them, α and β are adjustable weights, For MSE, is the cross entropy loss;

[0040] Collect lens parameters of orthokeratology lenses worn by several patients, their corneal topography after orthokeratology, and basic patient information, including physiological parameters: axial length, corneal diameter, central corneal thickness; demographic data: age, gender; genetic and medical history information: including family myopia and history of keratoconus; optometry data: uncorrected visual acuity, corrected visual acuity, and refractive power;

[0041] Extracting corneal topography features from the corneal topography of each patient after wearing orthokeratology lenses and data features from the patient's basic information, wherein the data features at least include the annual growth rate of the axial length over time;

[0042] Performing feature splicing on the lens parameters, the corneal topography features, and the data features to generate a basic corneal feature vector for each patient;

[0043] According to the corneal topography features after the patient wears orthokeratology lenses, the orthokeratology effect of each patient is evaluated and a corresponding orthokeratology effect difference map type is generated. The corneal basic feature vector is labeled with the corresponding orthokeratology effect difference map type. The rules for evaluating the orthokeratology effect difference map type according to the orthokeratology effect of each patient are as follows:

[0044] Category 1: The defocus ring in the image is located in the center of the cornea, centered and completely intact, indicating a good reshaping effect and an ideal correction effect;

[0045] Category 2: The defocus ring in the image may be slightly offset, but the overall structure is still intact, indicating that the correction effect is acceptable, but there is slight uneven shaping;

[0046] Category 3: The out-of-focus ring in the image is significantly off-center or its structure is incomplete, indicating poor correction or significant deviation;

[0047] Collecting the corneal basic feature vectors of each patient to form a feature set, and dividing it into a training set, a validation set, and a test set according to a preset ratio;

[0048] Inputting the training set into the multi-task regression model, performing regression training on the corneal basic feature vectors of each patient through the multi-task regression model to generate the initial ranking model;

[0049] The ranking model is tested using the test set, and the prediction results of the ranking model are compared and verified using the validation set:

[0050] If the verification passes, the sorting model is deployed and applied;

[0051] Otherwise, repeat the above steps to regenerate the sorting model.

[0052] Preferably, when comparing and verifying the prediction results of the ranking model using the validation set, the validation indicators include:

[0053] Metric 1: Mean squared error (MSE) and mean absolute error (MAE) between validation set labels and test set labels;

[0054] as well as,

[0055] Indicator 2: Lens shaping effect score,

[0056] When both indicators meet the preset values, the verification is qualified.

[0057] Preferably, the ranking scoring formula is:

[0058]

[0059] in:

[0060] GrowthRate is the annual growth rate of the eye axis, in mm / year. The lower the value, the better the reshaping effect. The reciprocal is taken to reflect the minimization goal.

[0061] TypeScore is the score of the shaping effect difference map type: Category 1: TypeScore = 1.0 (best effect); Category 2: TypeScore = 0.7 (second best effect); Category 3: TypeScore = 0.0 (poor effect, not included in the score);

[0062] ω1 and ω2 are corresponding weight parameters, indicating the importance of the annual growth rate of the axial length and the type of the difference map of the reshaping effect, and are adjusted according to clinical needs: if the control of axial length growth is more important, ω1 is larger (0.7); if the type of the difference map is more important, ω2 is larger (0.3);

[0063] Score is the shaping effect rating of the lens.

[0064] Preferably, the method further comprises:

[0065] Inputting the lens parameters of the final lens recommended to the patient, the patient's corneal topography, and the patient's basic information into a preset sorting model, the sorting model identifying the corneal basic feature vector therein and outputting the corresponding annual axial growth rate and reshaping effect difference map type;

[0066] generating a shaping effect score for the final lens according to the annual growth rate of the axial length and the type of the shaping effect difference graph;

[0067] According to the shaping effect score, the orthokeratology lens wearing treatment plan corresponding to the shaping effect score of the final lens is retrieved from the background database and output.

[0068] On the other hand, a multi-model-based multimodal orthokeratology lens intelligent fitting system is provided. The multi-model-based multimodal orthokeratology lens intelligent fitting system is used to implement the multi-model-based multimodal orthokeratology lens intelligent fitting method described above. The system includes:

[0069] An input module is used to input basic patient data, including corneal topography and basic patient information. 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 and history of keratoconus; optometry data: naked eye visual acuity, corrected visual acuity, and refractive power;

[0070] a recall module, configured to input the patient's corneal topography into a preset recall model, have the recall model identify depth image features therein and predict lens parameters corresponding to the depth image features, generate candidate intervals for candidate lenses to be recalled based on the predicted lens parameters, and screen out K pairs of candidate lenses based on the candidate intervals;

[0071] a sorting module, configured to sequentially input the lens parameters of each candidate lens, the patient's corneal topography, and the patient's basic information into a preset sorting model, wherein the sorting model identifies the corneal basic feature vector therein and outputs the corresponding annual axial length growth rate and shaping effect difference map type, and sorts the candidate lenses whose shaping effect difference map type is Category 1 or Category 2;

[0072] a recommendation module, configured to score the shaping effect of each of the sorted candidate lenses based on the annual growth rate of the eye axis and the type of the shaping effect difference graph using a sorting and scoring formula, and select the candidate lens with the maximum shaping effect score as the final lens recommended to the patient;

[0073] An output module is used to output lens parameters of the final lens.

[0074] On the other hand, an electronic device is provided, comprising: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned multi-model-based multimodal orthokeratology lens intelligent fitting methods is implemented.

[0075] On the other hand, a computer-readable storage medium is provided, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned multi-model-based multimodal orthokeratology lens intelligent fitting methods.

[0076] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0077] 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 the objective evaluation of the axial growth rate and the type of corneal topography difference map, the subjectivity of the traditional method is avoided, and the shaping effect can be evaluated more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] 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.

[0079] Figure 1 It is a schematic diagram of corneal topography;

[0080] Figure 2 is a structural diagram of a recall model provided by an embodiment of the present invention;

[0081] Figure 3 1 is a schematic diagram of screening of multimodal orthokeratology lenses based on multiple models provided by an embodiment of the present invention;

[0082] Figure 4 This is a schematic diagram of the structure of a sorting model provided by an embodiment of the present invention;

[0083] Figure 5 This is a block diagram of a multi-model-based multimodal orthokeratology lens intelligent fitting system provided by an embodiment of the present invention;

[0084] Figure 6 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0085] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0086] 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 an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0087] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0088] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0089] Explanation of terms:

[0090] Corneal topography, or computer-assisted corneal topographic analysis system, is a diagnostic technique that uses a computer image processing system to digitally analyze corneal morphology. The information obtained is presented as a color map with distinct characteristics. Because it resembles the undulating topographical structure of a geological surface, it is called corneal topography. Corneal topography can accurately measure and analyze the curvature and refractive power of any point on the entire anterior corneal surface, providing a systematic and comprehensive quantitative analysis of anterior corneal surface morphology. Its principle is to utilize a computer-assisted image processing system to photograph or capture images or interference fringes projected onto the corneal surface. After automated processing using specialized imaging software, the desired images are intelligently digitized and then color-coded to create a corneal topography map. This provides intuitive, detailed, and accurate qualitative and quantitative information on the anterior corneal curvature. A corneal topographer typically consists of three components: a Placido disk projection system, a real-time image monitoring system, and a computer image processing system. During use, the corneal topographer projects a series of concentric rings evenly onto the corneal surface. These ring images are then captured by a real-time image monitoring system and stored in a computer. A computer image processing system then digitizes these images and analyzes them using pre-set mathematical formulas and programs. The results are then displayed on a screen as color images.

[0091] AC1: Usually expressed in "mm", refers to the radius of curvature of the central area of ​​the lens. The smaller the unit, the greater the curvature.

[0092] Diameter: The diameter of an OK lens refers to the overall diameter of the lens, usually measured in millimeters (mm). It determines the extent of the lens' coverage of the cornea. The appropriate lens diameter is usually selected based on the patient's corneal diameter and eye shape.

[0093] Annular curvature refers to the gradual change in the curvature of the front surface of an OK lens from the center to the edge of the lens. It controls the change in curvature from the center to the edge of the lens and generally includes several areas: the central area (AC1), the transition zone (BC, base curve), and the annular zone (PC, peripheral curve).

[0094] Example: As attached Figure 1The following measurements can be made on corneal topography for a patient with mild myopia (-1.00D to -2.50D):

[0095] AC1: 40.50mm

[0096] Explanation: Suitable for patients with mild myopia. It has a larger AC1 and a gentler curvature in the center of the lens, providing a gentle shaping effect and avoiding overcorrection.

[0097] Diameter: 10.6mm

[0098] Explanation: The smaller diameter accommodates smaller corneas, ensuring the lens can comfortably cover the cornea and avoid interference from the lens edge.

[0099] Peripheral Curve: PC1 (Peripheral Curve 1): 42.50mm

[0100] Explanation: The gentler toric design is suitable for patients with less curved corneas. It can provide higher wearing comfort, ensure the stability of the lens edge and not affect the shaping effect.

[0101] This invention mainly uses AI models to recommend personalized orthokeratology lenses to patients.

[0102] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0103] The embodiment of the present invention provides a multi-modal orthokeratology lens intelligent fitting method based on multiple models, which can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart of the multi-model-based multimodal orthokeratology lens intelligent fitting method shown in FIG. 1 may include the following steps:

[0104] S1. Input basic patient data, including corneal topography and basic patient information. 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 and history of keratoconus); optometry data (uncorrected visual acuity, corrected visual acuity, and diopter);

[0105] S2. Inputting the patient's corneal topography into a preset recall model, the recall model identifying depth image features therein and predicting lens parameters corresponding to the depth image features, generating candidate intervals for candidate lenses to be recalled based on the predicted lens parameters, and screening out K pairs of candidate lenses based on the candidate intervals;

[0106] S3. Inputting the lens parameters of each candidate lens, the patient's corneal topography, and the patient's basic information into a preset ranking model in sequence, the ranking model identifies the corneal basic feature vector therein and outputs the corresponding annual axial length growth rate and orthokeratology effect difference map type, and ranking the candidate lenses whose orthokeratology effect difference map type is Category 1 or Category 2;

[0107] S4. Using a ranking and scoring formula, score the annual growth rate of the axial length and the type of the shaping effect difference graph for each of the sorted candidate lenses, and select the candidate lens with the maximum shaping effect score as the final lens recommended to the patient;

[0108] S5. Output the lens parameters of the final lens.

[0109] The present invention mainly consists of the following steps: input, recall of candidate lenses, sorting of candidate lenses, recommendation of final lenses, and output:

[0110] Input: Responsible for receiving the patient's basic data, including corneal topography, axial length, age, gender and other information.

[0111] Recall model: Using the corneal topography before fitting as input, a multi-task regression model is designed based on ResNet50 to predict the parameters of VST lenses (AC1, diameter, toricity), and recall the candidate parameters of orthokeratology lenses that meet the patient's needs.

[0112] Sorting model: Further screen the recalled lens candidates. The model input includes user information (basic information such as axial length, age, gender, etc.), corneal topography, and lens parameters. The ranking model is used to evaluate the reshaping effect of the lens (annual growth rate of the axial length and the type of reshaping effect difference graph) and select the lens with the best effect.

[0113] Effect evaluation: The reshaping effect of each lens is evaluated based on the axial length growth rate and the type of corneal topography difference map.

[0114] Output: Output the final selected lens parameters and recommended treatment plan.

[0115] Co-optimization between steps:

[0116] 1. Input and Model (Recall and Ranking): Data preprocessing optimizes the input quality of the recall model and reduces model training errors.

[0117] 2. Recall model and ranking model: The quality of the candidate lens pool directly affects the output of the ranking model.

[0118] 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.

[0119] 4. Output and effect evaluation: The output module can dynamically display the evaluation results, providing doctors with intuitive decision-making basis.

[0120] The specific implementation of each step will be described in detail below.

[0121] 1. Input

[0122] Collect comprehensive eye data of patients:

[0123] Corneal topography reflects the curvature distribution and morphological characteristics of the cornea.

[0124] Physiological parameters: axial length, corneal diameter, central corneal thickness.

[0125] Demographic data: basic information such as age and gender.

[0126] Genetic and medical history information: including family myopia degree, whether there is a history of keratoconus, etc.

[0127] Optometry data: naked eye vision, corrected vision, refraction, etc.

[0128] Data preprocessing

[0129] Image features: Standardized preprocessing (such as denoising and resolution unification) of corneal topography is performed to ensure the consistency of subsequent model input.

[0130] The corneal topography is an image input (such as a 224×224×3 RGB image), and a convolutional neural network (CNN) will be used to extract image features.

[0131] Preprocessing:

[0132] Image normalization: Normalize the pixel values ​​of the corneal topography to the range of [0, 1] to reduce the impact of brightness and contrast differences on model training.

[0133] Data augmentation: To improve the robustness and generalization ability of the model, the images are subjected to enhancement operations such as rotation, translation, scaling, and mirror flipping to simulate the changes in the corneas of different patients.

[0134] 1) Feature extraction:

[0135] 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.

[0136] Output: By removing the fully connected layer of ResNet50 (i.e. extracting the output of the penultimate layer of ResNet50), its output feature vector (usually 2048 dimensions) is used as the image feature.

[0137] Schematic diagram:

[0138] Input: Corneal topography (224×224×3)

[0139] Feature extraction: ResNet50 network (2048-dimensional features)

[0140] Numerical features: Continuous variables (such as axial length) are normalized; discrete variables (such as gender) are one-hot encoded.

[0141] For numerical features (such as axial length, age, etc.), complex image processing is not required, but it is necessary to ensure that they can be effectively input into the neural network.

[0142] Standardization / normalization:

[0143] For continuous variables (such as axial length and age), standardization or normalization is usually required. Standardization can transform features into data with a mean of 0 and a standard deviation of 1, thus preventing scale differences between different features from affecting the training process.

[0144] For example, the standardized formula for axial length is:

[0145]

[0146] Where μ is the mean axial length and σ is the standard deviation.

[0147] Normalization compresses the data into the range of [0,1], which is suitable for input into the model.

[0148] Discrete feature processing (such as gender, family medical history):

[0149] 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 to [1, 0] (male) or [0, 1] (female).

[0150] Example:

[0151] Input: axial length, age, gender (normalized / standardized and one-hot encoded);

[0152] Output: processed numerical feature vector (e.g. [0.32, 0.45, 1, 0]).

[0153] 2) Data integration: All input data is uniformly converted into matrix and tensor formats to ensure compatibility with subsequent deep learning models.

[0154] Fusion of image features and numerical features:

[0155] The output of the image feature extraction part is a high-dimensional vector (1024 dimensions). Numerical features (such as age, axial length, etc.) are usually a low-dimensional vector (10 dimensions, etc.).

[0156] Concatenation: Concatenate the image feature vector and the numerical feature vector 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 concatenation results in a feature vector of 1024 + 10 = 1034 dimensions.

[0157] EmbeddingLayer:

[0158] For discrete category features (such as gender, family medical history, etc.), use the EmbeddingLayer to convert them into dense vectors, and then concatenate these dense vectors with other features.

[0159] For example, gender (2 categories) can be mapped to a low-dimensional vector (2 dimensions) through the embedding layer. In this way, the low-dimensional embedding of discrete features and the concatenation of continuous features can be input into the subsequent network.

[0160] Features after fusion:

[0161] Concatenated vector: a fusion vector containing image features and numerical features.

[0162] Schematic diagram:

[0163] Input: image feature vector (1024 dimensions) + numerical feature vector (10 dimensions) + discrete feature vector (2 dimensions);

[0164] Output: concatenated fusion vector (1036 dimensions).

[0165] After preprocessing, the input data can be fed into the subsequent recall model and ranking model for further feature recognition and prediction. Data preprocessing optimizes the input quality of the recall model and reduces model training errors.

[0166] 2. The recall model is mainly used to identify the depth features of the patient's corneal topography and predict the lens parameters of the orthokeratology lens with corresponding features. The system generates a candidate interval of the candidate lenses to be recalled based on the predicted lens parameters (predicted values) of the orthokeratology lens, and selects K pairs of candidate lenses based on the candidate interval.

[0167] The recall model uses a multi-task regression model based on ResNet50 to predict and output candidate lenses for patients. The main contents are as follows:

[0168] 1) Model structure of recall model

[0169] As attached Figure 2 As shown:

[0170] Network Architecture: Multi-task regression model based on ResNet50. Backbone Network: ResNet50 is used to extract deep features of corneal topography and generate multi-level convolutional feature maps.

[0171] Multi-taskHeads: It consists of two branches, which predict the lens parameters AC1, diameter and annulus respectively.

[0172] Parameter adjustment: Adjust the model depth, learning rate, and activation function through hyperparameter optimization to improve prediction accuracy.

[0173] 2) Working Principle

[0174] Model structure design

[0175] Backbone network: ResNet50 is used for deep feature extraction of corneal topography.

[0176] Input: Preprocessed corneal topography (224×224×3).

[0177] Convolutional module: ResNet50 residual block, used to capture multi-scale features.

[0178] Feature output: Flattened feature vector (128 dimensions).

[0179] Multi-task regression head:

[0180] Branch 1: Predict lens parameter AC1.

[0181] Branch 2: Predict lens parameter diameter.

[0182] Branch 3: Predict lens parameter curvature.

[0183] Activation function: Linear activation is used, which is suitable for regression tasks.

[0184] Loss function: The two branches use mean square error (MSE) respectively, and the final loss is:

[0185] Loss Total =λ1×Loss AC1 +λ2×Loss Diameter +λ3×Loss PC,

[0186] λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss items;

[0187] Loss AC1 、Loss Diameter 、Loss PC The Loss items are AC1, diameter, and annular curvature: Loss Total is the total Loss term of the multi-task regression model;

[0188] λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss items, which are used to balance the impact of different objectives on optimization.

[0189] 3) Model training:

[0190] Corneal topography of several patients and at least the following lens parameters of the orthokeratology lenses they wore were collected: AC1, diameter or toricity;

[0191] For details, please refer to the previous section "Input and Preprocessing".

[0192] 1.1 Data preparation:

[0193] a. Input: corneal topography and its corresponding real lens parameters (AC1, diameter, toricity).

[0194] b. Data enhancement: Rotate, scale, and color normalize the terrain map to improve the robustness of the model.

[0195] c. Data division: 70% training set, 15% validation set, and 15% test set.

[0196] Features and training process

[0197] 1.2 Training configuration:

[0198] a. Optimizer: Adam optimizer, learning rate set to 1e-4, using learning rate decay strategy.

[0199] b. Batch size: 32.

[0200] c. Training rounds: 50 rounds, monitoring the validation set loss.

[0201] 1.3 Model Evaluation

[0202] a. Indicators: mean square error (MSE) and mean absolute error (MAE).

[0203] b. Evaluation method:

[0204] y i is the validation set label (representing a data in the set), is the test set label, and N is the total number of evaluation samples:

[0205]

[0206] c. Recall rate: For all candidate lenses within the true parameter range, evaluate the candidate coverage of the recall model.

[0207] The patient's corneal topography can then be input and recognized by the recall model. The recall model identifies the depth image features (including the corresponding lens parameters: AC1 (and optionally AC2), diameter, and curvature) and predicts the lens parameters corresponding to the depth image features. Based on the predicted lens parameters, a candidate interval for the candidate lens to be recalled is generated, and K pairs of candidate lenses are selected based on the candidate intervals. After the model recognizes the features, it outputs the predicted value for the corresponding lens.

[0208] 1.4 Candidate generation: Generate candidate intervals based on the predicted values ​​(such as AC1±0.25, AC2±0.25), and select K pairs of candidate lenses. Figure 3 As shown in Figure 2, the recall model can recall multiple candidate lenses. 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, which will sort the lens according to the input candidate lens parameters and the patient's basic information (see Appendix 2). Figure 6 The user features in the lens are recalled and sorted to predict the patient's personalized features "annual growth rate of eye axis" and "type of reshaping effect difference map", and the final lens is selected and recommended to the patient.

[0209] When generating candidate intervals based on the predicted values, fluctuation factors can be adjusted based on the patient's basic information. The candidate interval AC1±0.25 (0.25 is the fluctuation value) is just an example. For example, the fluctuation value of the predicted values ​​"AC1, diameter, and annulus" can be adjusted based on personalized factors in the patient's basic information. The following example is an example:

[0210] make:

[0211] Fluctuation value = P rec / 10,P rec The corneal comprehensive recommendation index is calculated as follows:

[0212]

[0213] The definitions of each character are shown in the following table:

[0214]

[0215]

[0216] By considering the patient's physiological parameters, corneal data, medical history, and optometry data, a comprehensive corneal recommendation index is generated and the fluctuation value is calculated, so that the final recommendation can be made comprehensively to recommend candidate lenses that meet the patient's individual characteristics. For example, the axial length may affect the lens power, and 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 and a history of keratoconus may require adjustment of the lens design to avoid risks. The refractive power in the optometry data is directly related to the lens power, and naked eye and corrected vision may affect whether a special design is needed (such as a defocus lens).

[0217] The above weights can be designed based on clinical experience and are only examples.

[0218] Therefore, we shared the feature extraction layer of ResNet50 and optimized multiple objectives simultaneously to improve the efficiency and accuracy of the model; the prediction results are adaptive and can dynamically adjust the screening range of candidate lenses based on patient characteristics.

[0219] Next, the ranking 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.

[0220] 3. The sorting model is also built based on the ResNet50 multi-task regression model.

[0221] The multi-task regression model includes a backbone network and a task regression head of a branch network, wherein: the backbone network is used to use ResNet50 to learn the annual growth rate of the axial length corresponding to the basic feature vector of the cornea, and the task regression head is used to learn the type of shaping effect difference map corresponding to the basic feature vector of the cornea.

[0222] The backbone network ResNet50 can learn the image features of corneal topography, and the branch network can learn other parameters, thereby forming a multi-task regression model composed of the backbone network and the branch network, which will be used as the training model of the sorting model for initial training.

[0223] 1) The model architecture of the sorting model is as follows:

[0224] Multimodal input layer: processes different features through multiple independent embedding layers;

[0225] Feature fusion layer: Fully connected layers merge image features with embedded features;

[0226] Multi-task branch: Design independent output branches for regression and classification tasks respectively.

[0227] 2) Input features of the sorting model (implemented in combination with the previous input and its preprocessing):

[0228] Discrete features: including age, gender, family medical history, etc. (encoded into low-dimensional vectors by the embedding layer).

[0229] Lens parameters: AC1 (direct input, AC2 can also be added).

[0230] Image features: corneal topography features (extracted by ResNet50).

[0231] Fusion features: Combine the above three types of features to generate a unified feature vector (manually labeled difference map type).

[0232] 3) Multi-task output:

[0233] Task 1: Predict the annual growth rate of the axial length (continuous value).

[0234] Task 2: Predict the type of shaping effect difference map (classification task, 3 categories).

[0235] Activation function: Task 1: linear activation, for regression;

[0236] Task 2: Softmax activation for classification;

[0237] Loss function:

[0238]

[0239] Among them, α and β are adjustable weights, is the MSE (refer to the mean square error above), is the cross entropy loss (calculated by the system);

[0240] 4) Training Process

[0241] 1. Data preparation:

[0242] a. Input: candidate lenses, corneal topography, and patient characteristics to the recall model.

[0243] b. Data enhancement: Perform operations such as horizontal flipping and random cropping on the corneal topography.

[0244] 2. Training configuration:

[0245] a. Optimizer: SGD (with momentum), learning rate is set to 0.01.

[0246] b. Regularization: Dropout and weight decay (WeightDecay) prevent overfitting.

[0247] c. Batch size: 64.

[0248] 3. Model Evaluation:

[0249]

[0250] in:

[0251] GrowthRate is the annual growth rate of the eye axis, in mm / year. The lower the value, the better the reshaping effect. The reciprocal is taken to reflect the minimization goal.

[0252] TypeScore is the score of the shaping effect difference map type: Category 1: TypeScore = 1.0 (best effect); Category 2: TypeScore = 0.7 (second best effect); Category 3: TypeScore = 0.0 (poor effect, not included in the score);

[0253] ω1 and ω2 are corresponding weight parameters, indicating the importance of the annual growth rate of the axial length and the type of the difference map of the reshaping effect. They are adjusted according to clinical needs: if the control of axial length growth is more important, ω1 is larger (preferably 0.7); if the difference map type is more important, ω2 is larger (preferably 0.3);

[0254] Score is the shaping effect rating of the lens.

[0255] As attached Figure 4 The ranking model structure shown in the figure takes as input the parameters of the candidate lenses of the recall model, the user characteristics of the patient, and the corneal topography characteristics. 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 and whether there is a history of keratoconus; optometry data: naked eye visual acuity, corrected visual acuity, and refractive power.

[0256] Extracting corneal topography features from the corneal topography of each patient after wearing orthokeratology lenses and data features (user features) from the patient's basic information, wherein the data features at least include the annual growth rate of the axial length over time (the annual growth rate of the axial length over time of the patient can be obtained through statistical analysis);

[0257] The lens parameters, the corneal topography features, and the data features are subjected to feature concatenation to generate a basic corneal feature vector for each patient; an example is as follows:

[0258] Input: image feature vector (1024 dimensions) + numerical feature vector (10 dimensions) + discrete feature vector (2 dimensions);

[0259] Output: The concatenated fusion vector F fused (1036 dimensions).

[0260] Feature splicing methods, such as:

[0261] F fused =Concat(F img ,F num ,F disc )∈R di+dn+dd ,

[0262] F img ,F num ,F disc are the feature vectors of each dimension mentioned above, di+dn+dd are the dimensions respectively, Concat is the sequential concatenation operation along the feature dimension, and the pyramid fusion strategy or attention mechanism weighted optimization information integration can also be used:

[0263] F enhanced =MLP(F fused )(MLP stands for Multi-layer Perceptron).

[0264] Through the above indicators and sorting performance evaluation, dual indicator evaluation is achieved:

[0265] Metric 1: Mean squared error (MSE) and mean absolute error (MAE) between validation set labels and test set labels;

[0266] as well as,

[0267] Indicator 2: Lens shaping effect score,

[0268] When both indicators meet the preset values, the verification is qualified.

[0269] Therefore, the accuracy of the model ranking results can be verified, and the reliability of the final lens selection can be improved.

[0270] Combined with attachment Figure 4 As shown in the figure, the sorting model is a multimodal multitask model, which uses corneal topography and patient personal characteristics to determine whether the lens can make the corneal topography after fitting reach Class 1 / 2 and the axial length growth rate is the slowest.

[0271] By sequentially inputting the lens parameters of each candidate lens, the patient's corneal topography and the patient's basic information into a preset sorting model, the sorting model identifies the basic corneal feature vector and outputs the corresponding annual axial growth rate and shaping effect difference map type, and sorts the candidate lenses whose shaping effect difference map type is Class 1 or Class 2; judge all the lenses in the recalled lenses in turn, and select the best lens as the final lens parameter.

[0272] The rules for evaluating the orthokeratology effect difference graph types based on the orthokeratology effect of each patient are as follows:

[0273] Category 1: The defocus ring in the image is located in the center of the cornea, centered and completely intact, indicating a good reshaping effect and an ideal correction effect;

[0274] Category 2: The defocus ring in the image may be slightly offset, but the overall structure is still intact, indicating that the correction effect is acceptable, but there is slight uneven shaping;

[0275] Category 3: The out-of-focus ring in the image is significantly off-center or its structure is incomplete, indicating poor correction or significant deviation.

[0276] 4. Effect evaluation

[0277] Based on the output of the ranking model, the optimal lens parameters are verified twice:

[0278] Axial growth rate: The lower the annual growth rate, the better the correction effect;

[0279] Difference graph type: Category 1 or Category 2 is preferred, indicating a good shaping effect.

[0280] After sorting, the candidate lenses can predict their effects on the patient's annual axial length growth rate and the type of reshaping effect difference map, so the final lenses are screened by the reshaping effect score.

[0281] For the scoring of the shaping effect of lenses, please refer to the ranking scoring formula in the previous section.

[0282] Sorting and scoring formula, where Score is the lens' shaping effect score. This method can be used to find the best lens from the candidate lenses.

[0283] Here, the candidate lens corresponding to the maximum value is set as the best lens.

[0284] 5. Visual output

[0285] Dynamically generate data reports, including:

[0286] Trend chart of annual growth rate of axial length;

[0287] Comparison of corneal topography before and after fitting glasses;

[0288] Example of the difference in sculpting effects.

[0289] Comprehensive indicator verification: Evaluate lens effects from multiple dimensions to ensure the scientific nature of the recommended results.

[0290] Visual support: Provides doctors with intuitive results display to assist clinical decision-making.

[0291] Through the predicted output of the final lens, intelligent OK lens fitting can be achieved for patients. It can also analyze and predict the basic characteristic vectors of the patient's cornea and output the corresponding annual growth rate of the axial length and the type of shaping effect difference map.

[0292] This department can also further recommend corresponding glasses treatment plans for patients.

[0293] Preferably, the method further comprises:

[0294] Inputting the lens parameters of the final lens recommended to the patient, the patient's corneal topography, and the patient's basic information into a preset sorting model, the sorting model identifying the corneal basic feature vector therein and outputting the corresponding annual axial growth rate and reshaping effect difference map type;

[0295] A shaping effect score is generated for the final lens based on the annual growth rate of the axial length and the type of the shaping effect difference graph.

[0296] Please understand the steps for scoring the patient's personalized shaping effect based on the lens parameters of the final lens in conjunction with the previous analysis steps for scoring the shaping effect.

[0297] The system can retrieve and output the orthokeratology lens wearing treatment plan corresponding to the shaping effect score of the final lens from the background database based on the shaping effect score.

[0298] The backend database is pre-configured with orthokeratology lens wearing treatment plans corresponding to different reshaping effect scores, including: wearing time, follow-up, precautions, etc., which can be decided by the doctor.

[0299] Therefore, the above method can bring the following technical effects:

[0300] 1. Improve the accuracy of parameter selection

[0301] Traditional methods rely on the doctor's experience and are prone to overlooking minor individual differences. This invention combines machine learning models, uses large-scale data to train a recall model, and automatically selects parameters from the lens library to ensure a high degree of compatibility between the lens and the patient's cornea.

[0302] This improved accuracy is reflected in a significant reduction in the number of adjustments required due to improper fitting.

[0303] 2. Personalized treatment plan

[0304] 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.

[0305] The ranking model considers multiple input variables and provides a better solution through weighted evaluation, so that each patient receives the most suitable lens.

[0306] 3. Provide objective evaluation criteria

[0307] Traditional assessment of eye contouring results relies heavily on subjective judgment and is easily influenced by experience. By using two objective indicators, the annual growth rate of the eye axis and the type of eye contouring effect difference graph, we can avoid human error.

[0308] Utilize multimodal data combined with deep learning models to establish an efficient and standardized evaluation system.

[0309] 4. Improve fitting efficiency and reduce costs

[0310] The automated recall and ranking process significantly reduces physician selection time.

[0311] Accurate first-time fitting results reduce the chance of repeat fittings and reduce material and labor costs.

[0312] 5. Adapt to different patient scenarios

[0313] The system supports multimodal input, which is suitable for rapid intervention in children and can also serve the long-term correction needs of adults.

[0314] 6. Easy to expand and upgrade

[0315] The model is designed based on a deep learning framework and supports training optimization with more data.

[0316] The modular design allows for the addition of new diagnostic dimensions (such as ocular biomechanics data) to further enhance system performance.

[0317] Figure 5 This is a block diagram of a multi-model based multi-modal orthokeratology lens intelligent fitting system according to an exemplary embodiment. The system is used for a multi-model based multi-modal orthokeratology lens intelligent fitting method. Figure 5 ,in:

[0318] An input module is used to input basic patient data, including corneal topography and basic patient information. 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 and history of keratoconus; optometry data: naked eye visual acuity, corrected visual acuity, and refractive power;

[0319] a recall module, configured to input the patient's corneal topography into a preset recall model, have the recall model identify depth image features therein and predict lens parameters corresponding to the depth image features, generate candidate intervals for candidate lenses to be recalled based on the predicted lens parameters, and screen out K pairs of candidate lenses based on the candidate intervals;

[0320] a sorting module, configured to sequentially input the lens parameters of each candidate lens, the patient's corneal topography, and the patient's basic information into a preset sorting model, wherein the sorting model identifies the corneal basic feature vector therein and outputs the corresponding annual axial length growth rate and shaping effect difference map type, and sorts the candidate lenses whose shaping effect difference map type is Category 1 or Category 2;

[0321] a recommendation module, configured to score the shaping effect of each of the sorted candidate lenses based on the annual growth rate of the eye axis and the type of the shaping effect difference graph using a sorting and scoring formula, and select the candidate lens with the maximum shaping effect score as the final lens recommended to the patient;

[0322] An output module is used to output lens parameters of the final lens.

[0323] Please understand and implement the above modules in conjunction with the steps in the above method, and will not be repeated here.

[0324] Figure 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention, such as Figure 6 As shown, the electronic device may include the above Figure 5 The multi-modal orthokeratology lens intelligent fitting system based on multiple models is shown. Optionally, the electronic device 410 may include a first processor 2001.

[0325] Optionally, the electronic device 410 may further include a memory 2002 and a transceiver 2003 .

[0326] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0327] The following combination Figure 6 The components of the electronic device 410 are described in detail.

[0328] The first processor 2001 is the control center of the electronic device 410 and can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement an embodiment of the present invention, such as one or more digital signal processors (DSPs) or one or more field programmable gate arrays (FPGAs).

[0329] 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.

[0330] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 6 CPU0 and CPU1 are shown in FIG.

[0331] In a specific implementation, as an embodiment, the electronic device 410 may also include multiple processors, such as Figure 6 1 and 2. The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0332] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled by the first processor 2001 for execution. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0333] Alternatively, 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 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 disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store 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 accessed through the interface circuit ( Figure 6 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0334] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0335] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 6 (not shown separately in the figure). The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0336] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 6 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0337] It should be noted that Figure 6 The structure of the electronic device 410 shown in the figure 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 arrange the components differently.

[0338] In addition, the technical effects of the electronic device 410 can refer to the technical effects of the multi-modal corneal refractive therapy lens intelligent fitting method based on multiple models described in the above method embodiment, and will not be repeated here.

[0339] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0340] 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 read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (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 and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0341] The above embodiments can be implemented in whole or in part through 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 program are loaded or executed on a computer, the processes or functions described in accordance with 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 system. 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 a wired method (such as infrared, wireless, microwave, etc.). 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 a collection of one or more available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, or magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0342] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0343] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0344] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0345] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0346] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0347] In the 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 merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of the system or unit, which can be electrical, mechanical or other forms.

[0348] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0349] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0350] If the 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, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0351] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A multi-model-based multimodal orthokeratology lens intelligent fitting method, characterized in that: The method comprises: Input basic patient data, including corneal topography and basic patient information. 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 and history of keratoconus; optometry data: naked eye visual acuity, corrected visual acuity, and refractive power; Inputting the patient's corneal topography into a preset recall model, the recall model identifying depth image features therein and predicting lens parameters corresponding to the depth image features, generating candidate intervals for candidate lenses to be recalled based on the predicted lens parameters, and screening out K pairs of candidate lenses based on the candidate intervals; The lens parameters of each candidate lens, the patient's corneal topography, and the patient's basic information are sequentially input into a preset sorting model, the sorting model identifies the corneal basic feature vector therein and outputs the corresponding annual axial length growth rate and shaping effect difference map type, and sorts the candidate lenses whose shaping effect difference map type is Class 1 or Class 2; the generation method of the sorting model includes: A multi-task regression model based on ResNet50 is pre-built and trained and configured, wherein the multi-task regression model includes a backbone network and a task regression head of a branch network, wherein: the backbone network is used to learn the annual growth rate of the axial length corresponding to the basic feature vector of the cornea using ResNet50, and the task regression head is used to learn the type of the shaping effect difference map corresponding to the basic feature vector of the cornea. The configuration information includes the loss function: Among them, α and β are adjustable weights, For MSE, is the cross entropy loss; Collect lens parameters of orthokeratology lenses worn by several patients, their corneal topography after orthokeratology, and basic patient information; Extracting corneal topography features from the corneal topography of each patient after wearing orthokeratology lenses and data features from the patient's basic information, wherein the data features at least include the annual growth rate of the axial length over time; Performing feature splicing on the lens parameters, the corneal topography features, and the data features to generate a basic corneal feature vector for each patient; According to the corneal topography features, the orthokeratology effect of each patient is evaluated and a corresponding orthokeratology effect difference map type is generated, and the corneal basic feature vector is labeled with the corresponding orthokeratology effect difference map type; Collecting the corneal basic feature vectors of each patient to form a feature set, and dividing it into a training set, a validation set, and a test set according to a preset ratio, performing regression training on the multi-task regression model to obtain the ranking model; Using a ranking and scoring formula, scoring the annual growth rate of the axial length and the type of the shaping effect difference graph of each of the sorted candidate lenses is performed, and the candidate lens with the maximum shaping effect score is selected as the final lens recommended to the patient; Output the lens parameters of the final lens.

2. The multi-model-based multimodal orthokeratology lens intelligent fitting method according to claim 1, characterized in that: Based on the corneal topography features, the rules for evaluating the orthokeratology effect of each patient and generating the corresponding orthokeratology effect difference map type are as follows: Category 1: The defocus ring in the image is located in the center of the cornea, centered and completely intact, indicating a good reshaping effect and an ideal correction effect; Category 2: The defocus ring in the image is slightly offset, but the overall structure is still intact, indicating that the correction effect is acceptable, but there is slight uneven shaping; Category 3: The out-of-focus ring in the image is significantly off-center or its structure is incomplete, indicating poor correction or significant deviation.

3. The multi-model-based multimodal orthokeratology lens intelligent fitting method according to claim 1, characterized in that: Performing regression training on the multi-task regression model to obtain the ranking model includes: Inputting the training set into the multi-task regression model, performing regression training on the corneal basic feature vectors of each patient through the multi-task regression model to generate the initial ranking model; The ranking model is tested using the test set, and the prediction results of the ranking model are compared and verified using the validation set: If the verification passes, the sorting model is deployed and applied; Otherwise, repeat the above steps to regenerate the sorting model.

4. The multi-model-based multimodal orthokeratology lens intelligent fitting method according to claim 1, characterized in that: The generation method of the recall model includes: Corneal topography of several patients and at least the following lens parameters of the orthokeratology lenses they wore were collected: AC1, diameter or toricity; A multi-task regression model based on ResNet50 is pre-built and trained. The multi-task regression model includes a backbone network and a multi-task regression head including multiple branch networks. 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 multiple lens parameters corresponding to the depth image features. The configuration information includes the loss function: Loss Total =λ1×Loss AC1 +λ2×Loss Diameter +λ3×Loss PC , λ1, λ2, and λ3 are the weight coefficients of the corresponding Loss items; Loss AC1 、Loss Diameter 、Loss PC The Loss items are AC1, diameter, and annular curvature: Loss Total is the total Loss term of the multi-task regression model; Adding the corresponding multiple lens parameters to the depth image features, counting and obtaining a feature set consisting of the depth image features of each patient, and dividing the feature set into a training set, a validation set, and a test set according to a proportion; Inputting the training set into the multi-task regression model, performing regression training on the depth image features of each patient and the corresponding lens parameters through the multi-task regression model to generate the initial recall model; The recall model is tested using the test set, and the prediction results of the recall model are compared and verified using the validation set: If the verification passes, the recall model is deployed and applied; Otherwise, repeat the above steps to regenerate the recall model.

5. The multi-model-based multimodal orthokeratology lens intelligent fitting method according to claim 3, characterized in that: When comparing and verifying the prediction results of the ranking model using the validation set, the validation indicators include: Metric 1: Mean square error (MSE) and mean absolute error (MAE) between validation set labels and test set labels; as well as, Indicator 2: Lens shaping effect score, When both indicators meet the preset values, the verification is qualified.

6. The multi-model-based multimodal orthokeratology lens intelligent fitting method according to claim 1, characterized in that: The ranking scoring formula is: in: GrowthRate is the annual growth rate of the eye axis, in mm / year. The lower the value, the better the reshaping effect. The reciprocal is taken to reflect the minimization goal. TypeScore is the score of the type of shaping effect difference map: Category 1: TypeScore = 1.0, the best effect; Category 2: TypeScore = 0.7, the second best effect; Category 3: TypeScore = 0.0, the effect is poor and not included in the score; ω1 and ω2 are corresponding weight parameters, which represent the importance of the annual growth rate of the axial length and the type of the reshaping effect difference map. They are adjusted according to clinical needs: if the control of axial length growth is more important, ω1 is larger; if the type of the reshaping effect difference map is more important, ω2 is larger. Score is the shaping effect rating of the lens.

7. The multi-model-based multimodal orthokeratology lens intelligent fitting method according to claim 1, characterized in that: The method further comprises: Inputting the lens parameters of the final lens recommended to the patient, the patient's corneal topography, and the patient's basic information into a preset sorting model, the sorting model identifying the corneal basic feature vector therein and outputting the corresponding annual axial growth rate and reshaping effect difference map type; generating a shaping effect score for the final lens according to the annual growth rate of the axial length and the type of the shaping effect difference graph; According to the shaping effect score, the orthokeratology lens wearing treatment plan corresponding to the shaping effect score of the final lens is retrieved from the background database and output.

8. A multi-model-based multi-modal orthokeratology lens intelligent fitting system, wherein the multi-model-based multi-modal orthokeratology lens intelligent fitting system is used to implement the multi-model-based multi-modal orthokeratology lens intelligent fitting method according to any one of claims 1 to 7, characterized in that: The system comprises: An input module is used to input basic patient data, including corneal topography and basic patient information. 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 and history of keratoconus; optometry data: naked eye visual acuity, corrected visual acuity, and refractive power; a recall module, configured to input the patient's corneal topography into a preset recall model, have the recall model identify depth image features therein and predict lens parameters corresponding to the depth image features, generate candidate intervals for candidate lenses to be recalled based on the predicted lens parameters, and screen out K pairs of candidate lenses based on the candidate intervals; a sorting module, configured to sequentially input the lens parameters of each candidate lens, the patient's corneal topography, and the patient's basic information into a preset sorting model, wherein the sorting model identifies the corneal basic feature vector therein and outputs the corresponding annual axial length growth rate and shaping effect difference map type, and sorts the candidate lenses whose shaping effect difference map type is Category 1 or Category 2; a recommendation module, configured to score the shaping effect of each of the sorted candidate lenses based on the annual growth rate of the eye axis and the type of the shaping effect difference graph using a sorting and scoring formula, and select the candidate lens with the maximum shaping effect score as the final lens recommended to the patient; An output module is used to output lens parameters of the final lens.

9. An electronic device, characterized in that: The electronic device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 7.

Images