Lens recommendation method and apparatus, and electronic device and storage medium

Abstract

Provided in the present disclosure are a lens recommendation method and apparatus, and an electronic device and a storage medium. The method comprises: acquiring user information of a target user and lens parameters of a plurality of candidate lenses; inputting the user information and the lens parameters into a target deep learning model to perform prediction, so as to determine a predicted postoperative result corresponding to each candidate lens, wherein the target deep learning model is obtained by means of a preset deep learning model performing training on training data, and the training data comprises user information of sample users, the lens parameters, and sample postoperative results of the sample users after same are implanted with corresponding candidate lenses; and according to the predicted postoperative results, determining from among the candidate lenses a target lens that meets a preset condition. In this way, the user information and the lens parameters are modeled and analyzed by using a deep learning algorithm and by means of the pre-trained target deep learning model, and a target lens corresponding to the target user can be effectively and accurately selected, thereby meeting postoperative requirements of different groups of people in a personalized manner.

Description

Crystal recommendation method, device, electronic device and storage medium

[0001] This disclosure claims priority to the Chinese patent application filed with the China Patent Office on June 25, 2023, with application number 202310756105.9 and invention name “A crystal recommendation method, device, electronic device and storage medium”, the entire contents of which are incorporated by reference in this disclosure. Technical Field

[0002] The present disclosure relates to the field of data processing technology, and in particular to a crystal recommendation method, device, electronic device, and storage medium. Background Art

[0003] Phakic intraocular lens implantation (Implantable Collamer Lens, ICL) and toric implantable contact lens implantation (TICL) are surgical methods for correcting myopia. By adding a pair of "glasses" (ICL or TICL lens) inside the eye, the purpose of correcting myopia can be achieved. No corneal cutting is required during the whole process. Compared with other cutting surgeries, it is safer and reversible, and can be used in a wide range of refractive correction application scenarios.

[0004] Before ICL / TICL surgery, lens selection is required. Currently, traditional machine learning methods can be used to determine the lens model to be used during surgery. However, these methods are incapable of analyzing user characteristics, resulting in inaccurate selection accuracy. Consequently, the lens model determined may not meet user needs, leading to poor surgical outcomes.

[0005] Summary of the Invention

[0006] To address the above technical issues, this application presents a lens recommendation method, device, electronic device, and storage medium to at least address the problem in related technologies of insufficient analysis of user characteristics, resulting in insufficient selection accuracy, the failure of the determined lens model to meet user needs, and thus poor surgical results. The technical solutions disclosed in this disclosure are as follows:

[0007] In a first aspect, the present disclosure provides a crystal recommendation method, comprising:

[0008] Obtaining user information of a target user and crystal parameters of multiple candidate crystals;

[0009] Inputting the user information and the lens parameters into a target deep learning model for prediction to determine the predicted postoperative outcome corresponding to each candidate lens; the target deep learning model is obtained by training training data using a preset deep learning model, and the training data includes the user information of a sample user, the lens parameters, and the sample postoperative outcome of the sample user after implantation of the corresponding candidate lens;

[0010] According to the predicted postoperative result, a target crystal that meets preset conditions is determined from the candidate crystals.

[0011] Optionally, obtaining user information of the target user and crystal parameters of multiple candidate crystals includes:

[0012] Obtaining crystal parameters of multiple candidate crystals;

[0013] Obtain text data related to the target user and the characteristic data type corresponding to each candidate crystal;

[0014] Data corresponding to each characteristic data type is extracted from the text data, and user information corresponding to each candidate crystal is generated respectively.

[0015] Optionally, before acquiring text data related to the target user and the type of characteristic data corresponding to the candidate crystal, the method further includes:

[0016] Get training data;

[0017] The training data is classified according to the candidate lens to be implanted, and the user information of the sample users included in each classification is matched and analyzed to determine the characteristic data type required to obtain data before implanting each candidate lens; the data of the characteristic data type is used to indicate the preoperative condition of the sample user.

[0018] Optionally, inputting the user information and the lens parameters into a target deep learning model for prediction to determine a predicted postoperative outcome corresponding to each candidate lens includes:

[0019] In a case where the user information and / or the crystal parameter are discrete variables, performing one-hot encoding on the user information and / or the crystal parameter to obtain a first variable;

[0020] In a case where the user information and / or the crystal parameter are continuous variables, normalizing the user information and / or the crystal parameter to obtain a second variable;

[0021] Calling an activation function to process the first variable and the second variable to obtain an input variable;

[0022] The input variables are input into the target deep learning model for prediction to determine the predicted postoperative outcome corresponding to each candidate lens.

[0023] Optionally, before inputting the user information and the lens parameters into the target deep learning model for prediction and determining the predicted postoperative outcome corresponding to each candidate lens, the method further includes:

[0024] Acquire training data; the training data includes positive samples and negative samples;

[0025] Repeatedly sampling the negative samples to obtain training data;

[0026] Inputting the training data into a preset deep learning model for prediction, and obtaining the training postoperative results of the sample user after implantation of the corresponding candidate lens;

[0027] According to the training postoperative results, the sample postoperative results and the preset loss function, the model parameters of the preset deep learning model are iteratively adjusted to obtain the target deep learning model.

[0028] Optionally, the preset deep learning model is a multi-layer perceptron model, a residual multi-layer perceptron model or a Transformer model;

[0029] The loss function is a mean square loss function, a mean absolute error loss function or a smoothed mean absolute error loss function.

[0030] Optionally, inputting the user information and the lens parameters into a target deep learning model for prediction to determine a predicted postoperative outcome corresponding to each candidate lens includes:

[0031] Determine the crystal parameters of the candidate crystals that match the user information as target crystal parameters, input the user information and the target crystal parameters into a target deep learning model for prediction, and determine the predicted postoperative outcome corresponding to each candidate crystal; or

[0032] According to the preset grouping, the reference crystal corresponding to each group is selected from a plurality of candidate crystals, the user information and the crystal parameters of the reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to each reference crystal is determined. According to the predicted postoperative result, the target reference crystal that meets the preset conditions is determined from the reference crystals, and the crystal parameters of the candidate crystals in the same group as the user information and the target reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to the candidate crystals in the same group is determined.

[0033] Optionally, determining a target crystal that meets preset conditions from the candidate crystals according to the predicted postoperative result includes:

[0034] determining a preset number of candidate crystals whose predicted postoperative results are closest to a preset value as target crystals; or,

[0035] A risk assessment is performed on the predicted postoperative outcome, and a preset number of candidate crystals are determined as target crystals in ascending order of the risk assessment results.

[0036] In a second aspect, the present disclosure provides a crystal recommendation device, comprising:

[0037] An acquisition module, used to obtain user information of a target user and crystal parameters of multiple candidate crystals;

[0038] A prediction module is configured to input the user information and the lens parameters into a target deep learning model for prediction, and determine a predicted postoperative outcome corresponding to each candidate lens; the target deep learning model is obtained by training training data using a preset deep learning model, and the training data includes the user information of a sample user, the lens parameters, and the sample postoperative outcome of the sample user after implantation of the corresponding candidate lens;

[0039] A recommendation module is used to determine a target crystal that meets preset conditions from the candidate crystals based on the predicted postoperative results.

[0040] Optionally, the acquisition module is specifically configured to:

[0041] Obtaining crystal parameters of multiple candidate crystals;

[0042] Obtain text data related to the target user and the characteristic data type corresponding to each candidate crystal;

[0043] Data corresponding to each characteristic data type is extracted from the text data, and user information corresponding to each candidate crystal is generated respectively.

[0044] Optionally, the acquisition module is specifically configured to:

[0045] Get training data;

[0046] The training data is classified according to the candidate lens to be implanted, and the user information of the sample users included in each classification is matched and analyzed to determine the characteristic data type required to obtain data before implanting each candidate lens; the data of the characteristic data type is used to indicate the preoperative condition of the sample user.

[0047] Optionally, the prediction module is specifically configured to:

[0048] In a case where the user information and / or the crystal parameter are discrete variables, performing one-hot encoding on the user information and / or the crystal parameter to obtain a first variable;

[0049] In a case where the user information and / or the crystal parameter are continuous variables, normalizing the user information and / or the crystal parameter to obtain a second variable;

[0050] Calling an activation function to process the first variable and the second variable to obtain an input variable;

[0051] The input variables are input into the target deep learning model for prediction to determine the predicted postoperative outcome corresponding to each candidate lens.

[0052] Optionally, the device further comprises a training module, configured to:

[0053] Acquire training data; the training data includes positive samples and negative samples;

[0054] Repeatedly sampling the negative samples to obtain training data;

[0055] Inputting the training data into a preset deep learning model for prediction, and obtaining the training postoperative results of the sample user after implantation of the corresponding candidate lens;

[0056] According to the training postoperative results, the sample postoperative results and the preset loss function, the model parameters of the preset deep learning model are iteratively adjusted to obtain the target deep learning model.

[0057] Optionally, the preset deep learning model is a multi-layer perceptron model, a residual multi-layer perceptron model or a Transformer model;

[0058] The loss function is a mean square loss function, a mean absolute error loss function or a smoothed mean absolute error loss function.

[0059] Optionally, the prediction module is specifically configured to:

[0060] Determine the crystal parameters of the candidate crystals that match the user information as target crystal parameters, input the user information and the target crystal parameters into a target deep learning model for prediction, and determine the predicted postoperative outcome corresponding to each candidate crystal; or

[0061] According to the preset grouping, the reference crystal corresponding to each group is selected from a plurality of candidate crystals, the user information and the crystal parameters of the reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to each reference crystal is determined. According to the predicted postoperative result, the target reference crystal that meets the preset conditions is determined from the reference crystals, and the crystal parameters of the candidate crystals in the same group as the user information and the target reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to the candidate crystals in the same group is determined.

[0062] Optionally, the recommendation module is used to:

[0063] determining a preset number of candidate crystals whose predicted postoperative results are closest to a preset value as target crystals; or,

[0064] A risk assessment is performed on the predicted postoperative outcome, and a preset number of candidate crystals are determined as target crystals in ascending order of the risk assessment results.

[0065] In a third aspect, the present disclosure shows an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the crystal recommendation method as described in any one of the above items are implemented.

[0066] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of any one of the above-described crystal recommendation methods are implemented.

[0067] According to a fifth aspect of an embodiment of the present disclosure, a computer program is provided, comprising a computer-readable code, which, when executed on a computing and processing device, causes the computing and processing device to execute any one of the function control methods described above.

[0068] Compared with the prior art, the present disclosure has the following advantages:

[0069] In the present disclosure, user information of a target user and crystal parameters of a plurality of candidate crystals are obtained; the user information and crystal parameters are input into a target deep learning model for prediction to determine the predicted postoperative results corresponding to each candidate crystal; the target deep learning model is obtained by training training data with a preset deep learning model, and the training data includes user information, crystal parameters and sample postoperative results of sample users after the corresponding candidate crystals are implanted; based on the predicted postoperative results, a target crystal that meets the preset conditions is determined from the candidate crystals.

[0070] In this way, by using a deep learning algorithm and a pre-trained target deep learning model, user information and lens parameters can be modeled and analyzed. The training data of the target deep learning model includes the user information of sample users from a wide range of populations before lens transplant surgery and the sample postoperative results obtained by postoperative vision and eye physiological parameters. Therefore, it is possible to effectively and accurately select the target lens corresponding to the target user. Compared with traditional machine learning methods, the selection of lenses is more accurate and can be personalized to meet the postoperative needs of different groups of people.

[0071] The above description is only an overview of the technical solution of the present disclosure. In order to more clearly understand the technical means of the present disclosure, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present disclosure more obvious and easy to understand, the specific implementation methods of the present disclosure are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0073] FIG1 schematically shows a flow chart of the steps of a crystal recommendation method disclosed herein;

[0074] FIG2 schematically shows a schematic diagram of a data collection and cleaning process of the present disclosure;

[0075] FIG3 schematically shows a model diagram of a multilayer perceptron model of the present disclosure;

[0076] FIG4 schematically shows a model diagram of a residual multilayer perceptron model of the present disclosure;

[0077] FIG5 schematically shows a model diagram of a Transformer model of the present disclosure;

[0078] FIG6 schematically shows a structural block diagram of a crystal recommendation device disclosed in the present invention;

[0079] FIG7 schematically shows a structural block diagram of an electronic device disclosed herein;

[0080] FIG8 schematically shows a storage unit of the present disclosure for storing or carrying program codes for implementing the method according to the present disclosure. Specific embodiments

[0081] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0082] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0083] The crystal recommendation method provided by the embodiment of the present invention will be described in detail below through specific examples.

[0084] 1 , there is shown a flow chart of a crystal recommendation method disclosed herein. The method may specifically include the following steps:

[0085] S101: Obtain user information of a target user and crystal parameters of multiple candidate crystals.

[0086] ICL and TICL are common surgical methods for correcting myopia. They correct myopia by implanting an ICL or TICL lens inside the eye. Before ICL / TICL surgery, lens selection is required to determine the lens model to be used during the procedure.

[0087] This step requires obtaining the target user's information and the lens parameters of multiple candidate lenses. The target user is the person currently undergoing surgery and selecting a lens, and the candidate lenses are the lenses currently available to the target user. Only after obtaining the user information and lens parameters can the target user and candidate lenses be matched based on these two pieces of information.

[0088] For example, user information may include basic user information such as age, gender, and eye type; preoperative examination information such as preoperative sphere, preoperative cylinder, preoperative axial position, preoperative spherical equivalent, dark pupil, intraocular pressure, axial length, corneal curvature K1, K1 axial, corneal curvature K2, K2 axial, anterior chamber depth, corneal thickness (CT), white to white (WTW), anterior chamber volume, anterior chamber angle, pupil diameter, lens thickness, etc. Lens parameters may include lens size, lens sphere, lens cylinder, and lens axial position, etc., without limitation.

[0089] In one implementation, obtaining user information of a target user and crystal parameters of multiple candidate crystals includes:

[0090] Acquire crystal parameters of multiple candidate crystals; acquire text data related to the target user and the characteristic data type corresponding to each candidate crystal; extract data corresponding to each characteristic data type from the text data, and generate user information corresponding to each candidate crystal.

[0091] Among them, text data can include the target user's historical medical history, surgical records, and preoperative examination reports. For paper data, it can be digitized through image acquisition methods and converted into extractable text data through optical character recognition (OCR) technology.

[0092] The acquired text data can be used for structured information extraction, extracting data corresponding to each characteristic data type, including indicators and indications related to lens implantation. Since structured information can be broken down into multiple interconnected components after analysis, each with a clear hierarchical structure, the information extracted from the text data can be stored in a structured manner, storing the data in a database according to its corresponding hierarchical structure to standardize data management, use, and maintenance. Furthermore, by analyzing the stored data, the data that the target user needs to refer to when implanting each candidate lens can be determined, namely, the user information corresponding to each candidate lens.

[0093] Data of different characteristic data types can be extracted from different files. For example, information such as corneal curvature, axis, axial length, anterior chamber angle, and white-to-white distance can be extracted from the optical coherence biometer (IOLMaster) examination report; information such as age, gender, eye type, and common medical history can be extracted from medical records; and information such as incision size, incision direction, postoperative diagnosis, and implanted lens can be extracted from surgical records.

[0094] In this way, rich and high-quality user information can be obtained from various text data, which is crucial for building a target deep learning model with reliable performance. It can further improve the accuracy of crystal selection and meet the needs of different users.

[0095] Before obtaining the text data related to the target user and the feature data type corresponding to the candidate crystal, the following steps are also included:

[0096] Acquire training data; classify the training data according to the candidate lens to be implanted, and perform matching analysis on the user information of the sample users included in each classification to determine the characteristic data type required to obtain data before implanting each candidate lens; the data of the characteristic data type is used to indicate the preoperative condition of the sample user.

[0097] It's understandable that different types of lens implant surgeries require slightly different types of user information, including the types of feature data. For example, in postoperative follow-up, in addition to visual acuity and the time since surgery, ICL lens implantation also focuses on postoperative ocular structural parameters such as arch height. Therefore, by classifying and matching the training data, we can determine the required feature data types for each candidate lens, thereby improving the richness and quality of user information.

[0098] For example, the training data can be classified according to key fields such as medical record number, and then the user information of the sample users included in each classification can be matched and analyzed. According to the preoperative parameters and postoperative conditions corresponding to each candidate lens, the characteristic data type required to obtain data before implanting each candidate lens is determined; the data of the characteristic data type is used to indicate the preoperative condition of the sample user. For example, the characteristic data type may include age, gender, eye type, common medical history, corneal curvature and corresponding axis, eye axis, anterior chamber angle, white to white, arch height, etc., without specific limitation.

[0099] Furthermore, the determined entries can also be manually verified first, and screened according to the user information to meet the medical conditions to obtain the final feature data type, which can further improve the safety and accuracy of crystal selection.

[0100] As shown in Figure 2, it is a schematic diagram of a data collection and cleaning process. First, the original data is obtained, including paper documents or text data such as historical medical records, surgical records and preoperative test reports. Among them, for the paper documents in the original data, you can first take a photo of them, and then use OCR technology to convert the image data into extractable text data. Then, structured information extraction is performed on the text data converted from the original data, and indicators, indications and other information related to lens implantation are extracted, and this information is structured and stored. Then, multi-source data matching is performed on different types of information, and the final available data entries are obtained through manual verification to determine the characteristic data type corresponding to each candidate lens.

[0101] S102: Input user information and crystal parameters into the target deep learning model for prediction to determine the predicted postoperative results corresponding to each candidate crystal; the target deep learning model is obtained by training the training data with a preset deep learning model, and the training data includes the user information, crystal parameters and sample postoperative results of the sample user after the corresponding candidate crystal is implanted.

[0102] In this disclosure, user information and crystal parameters of sample users are used during training to predict sample postoperative outcomes and adjust the model parameters of the deep learning model. During inference, by traversing the crystal parameters of a limited number of candidate crystals, the target deep learning model can obtain the predicted postoperative outcome corresponding to each candidate crystal.

[0103] Among them, the predicted postoperative results are the results of predicting relevant information of the target user after implantation of a certain candidate lens, which can reflect the postoperative condition of the target user. For example, it can include postoperative vision and eye parameters. Eye parameters include arch height, etc., which are not specifically limited.

[0104] Since deep learning models have stronger analysis capabilities for input user information and crystal parameters than traditional machine learning models, the prediction results based on the target deep learning model are more accurate, making crystal selection more accurate.

[0105] In one implementation, user information and lens parameters are input into a target deep learning model for prediction, and the predicted postoperative outcome corresponding to each candidate lens is determined, including:

[0106] When the user information and / or crystal parameters are discrete variables, the user information and / or crystal parameters are one-hot encoded to obtain a first variable; when the user information and / or crystal parameters are continuous variables, the user information and / or crystal parameters are normalized to obtain a second variable; an activation function is called to process the first variable and the second variable to obtain an input variable; the input variable is input into the target deep learning model for prediction to determine the predicted postoperative outcome corresponding to each candidate crystal.

[0107] It can be understood that the user information and crystal parameters acquired in step S101 include different types. For example, a certain data may be a discrete category or a continuous numerical type.

[0108] In this step, one-hot encoding can be performed for discrete data. One-hot encoding, also known as single-bit effective encoding, mainly uses an N-bit state register to encode N states. Each state corresponds to an independent register bit, and at any time, only one bit of the N-bit state register is effective. For example, for two different crystal ICL and TICL, as two different characteristics, a 2-bit state register can be used to encode them, for example, they can be recorded as [1,0] and [0,1] respectively;

[0109] For continuous variables, you can set the value range of the variable, subtract the lower limit of the value range from each variable value, and divide it by the difference between the upper and lower limits of the value range to transform it into the range of [0,1].

[0110] Then, the activation function is called to perform a Sigmoid transformation on the first and second variables to obtain the input variable. The Sigmoid function, also known as the S-shaped growth curve, has properties such as monotonic increase and inverse monotonic increase, and can map the input variable to a value between 0 and 1.

[0111] Among them, the call of the Sigmoid function ensures that even if the input variable is out of range, it can still be converted into the range of values ​​learned by the target deep learning model, thus avoiding unknown unexpected situations and further standardizing the data.

[0112] Typically, in medical diagnosis, the number of positive samples showing abnormalities is far smaller than the number of negative samples. This will cause the trained target deep learning model to be dominated by negative samples with a large amount of data, thereby reducing model performance and failing to accurately select crystals.

[0113] Then, in one implementation, before inputting user information and lens parameters into the target deep learning model for prediction and determining the predicted postoperative outcome corresponding to each candidate lens, the following steps are performed:

[0114] Obtain training data, which includes positive samples and negative samples; repeatedly sample negative samples to obtain training data; input the training data into a preset deep learning model for prediction to obtain the training postoperative results after the sample user implants the corresponding candidate lens; based on the training postoperative results, the sample postoperative results and the preset loss function, iteratively adjust the model parameters of the preset deep learning model to obtain the target deep learning model.

[0115] In this way, by adopting the method of repeated sampling, the frequency of minority category samples in training can be increased, and data amplification of rare positive samples can be achieved, so that the training data is relatively balanced, and the situation where the trained model is dominated by negative samples due to the fact that the number of abnormal positive samples is far less than the number of negative samples is reduced, thereby better adapting to the needs of model training, improving model performance and the accuracy of crystal selection for positive samples.

[0116] In the present disclosure, the preset deep learning model can be a multi-layer perceptron model, a residual multi-layer perceptron model, a Transformer model, or any other deep learning model; the preset loss function can be a mean square error (MSE) loss function, a mean absolute error (L1) loss function, or a smoothed L1 loss function, etc., without specific limitation.

[0117] For example, as shown in FIG3 , which is a schematic diagram of a multilayer perceptron model, the multilayer perceptron model includes three fully connected layers, and each two fully connected layers include two hidden layers. The hidden layers can perform nonlinear transformation on the input user information and crystal parameters through activation functions to obtain predicted postoperative results. The activation functions may include Sigmoid activation functions, Dropout activation functions, and Tanh activation functions, etc., without specific limitation.

[0118] As shown in Figure 4, it is a schematic diagram of the residual multilayer perceptron model. The residual multilayer perceptron model is based on the multilayer perceptron model and adds a residual perceptron module. The residual perceptron module includes 2 fully connected layers and 4 hidden layers. By adding the input of the residual perceptron module with the output of the 2 fully connected layers and the 4 hidden layers, the input of the third fully connected layer outside the residual perceptron module is obtained. In this way, the depth of the model is further increased and the training effect is better.

[0119] Figure 5 shows a schematic diagram of the Transformer model. The Transformer model consists of two fully connected layers and a Transformer encoding layer. The Transformer encoding layer encodes and learns input user information and crystal parameters based on a self-attention mechanism, which has higher parallelism and computational efficiency.

[0120] In addition, the present disclosure may also adopt other deep learning models for training and use for crystal selection, without specific limitation.

[0121] S103: According to the predicted postoperative results, a target lens that meets the preset conditions is determined from the candidate lenses.

[0122] After obtaining the predicted postoperative results, the various data in the predicted postoperative results can be analyzed to determine the target crystal that meets the preset conditions from the candidate crystals, and try to make the various data in the predicted postoperative results distributed in a reasonable range to reduce the probability of complications.

[0123] For example, we can first select candidate lenses that meet the postoperative refractive requirements in the predicted postoperative results, and then analyze other parameters of the postoperative eye structure, such as the arch height, to ensure that they are also distributed in a reasonable range, and then determine the target lens.

[0124] This method can intuitively show the distribution of crystal recommendations, making it easier to evaluate the reliability of the method. At the same time, doctors can also use their clinical experience to further optimize crystal selection parameters based on the user's characteristics.

[0125] In this step, based on the predicted postoperative outcome, a target lens that meets the preset conditions is determined from the candidate lenses, including:

[0126] Determine a preset number of candidate crystals whose predicted postoperative results are closest to the preset value as target crystals; or, perform a risk assessment on the predicted postoperative results, and determine a preset number of candidate crystals as target crystals in order of risk assessment results from small to large.

[0127] In other words, based on the required postoperative refractive target value, a series of candidate lenses with similar refractive ranges for predicted postoperative results can be automatically screened out and recommended as target lenses. Alternatively, the predicted postoperative results can be analyzed in combination with postoperative eye parameter predictions, and risk assessments can be performed on each candidate lens, with certain risk warnings given. For example, selecting a certain lens may result in a postoperative arch height that is too high or too low, increasing the user's risk of postoperative complications, and so on.

[0128] In this disclosure, user information and lens parameters are input into the target deep learning model for prediction to determine the predicted postoperative outcome for each candidate lens, including:

[0129] Determine the crystal parameters of the candidate lens that matches the user information as the target crystal parameters, input the user information and the target crystal parameters into the target deep learning model for prediction, and determine the predicted postoperative outcome corresponding to each candidate lens;

[0130] Among them, the candidate crystals that match the user information can be all possible candidate crystals, or they can be possible candidate crystals selected by the doctor based on traditional experience. There is no specific limitation. In this way, the target deep learning model needs to traverse the matching candidate crystals to find the optimal solution.

[0131] Alternatively, according to preset grouping, a reference crystal corresponding to each group is selected from a plurality of candidate crystals, user information and crystal parameters of the reference crystal are input into a target deep learning model for prediction, and the predicted postoperative outcome corresponding to each reference crystal is determined; based on the predicted postoperative outcome, a target reference crystal that meets preset conditions is determined from the reference crystals, and the crystal parameters of the candidate crystals in the same group as the user information and the target reference crystal are input into the target deep learning model for prediction, and the predicted postoperative outcome corresponding to the candidate crystals in the same group is determined;

[0132] That is to say, by grouping the candidate crystals, the candidate crystals can be traversed at coarser intervals first. The candidate crystals included in the same group have similar characteristics. In this way, after determining the better group, a detailed traversal is performed within the group to determine the predicted postoperative results corresponding to each candidate crystal in the group, so as to select the best crystal for surgery, thereby reducing the number of traversals and improving the efficiency of model processing and crystal selection.

[0133] As can be seen from the above, the solution provided by the present disclosure uses a deep learning algorithm. Through a pre-trained target deep learning model, user information and lens parameters can be modeled and analyzed. The training data of the target deep learning model includes user information of sample users from a wide range of populations before lens transplant surgery and sample postoperative results obtained by postoperative vision and ocular physiological parameters. Therefore, it is possible to effectively and accurately select the target lens corresponding to the target user. Compared with traditional machine learning methods, the deep learning model has a stronger ability to analyze the input user information and lens parameters. Therefore, the prediction results based on the target deep learning model are more accurate, making the lens selection more accurate. Furthermore, the postoperative needs of different groups of people can be met in a personalized manner.

[0134] It should be noted that for the method embodiments, for simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present disclosure is not limited by the order of the actions described, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions involved are not necessarily required by the present disclosure.

[0135] 6 , a block diagram of a crystal recommendation device according to the present disclosure is shown. The device may include the following modules:

[0136] An acquisition module 201 is used to acquire user information of a target user and crystal parameters of multiple candidate crystals;

[0137] Prediction module 202 is configured to input the user information and the lens parameters into a target deep learning model for prediction, and determine a predicted postoperative outcome corresponding to each candidate lens; the target deep learning model is obtained by training training data using a preset deep learning model, and the training data includes the user information of a sample user, the lens parameters, and the postoperative outcome of the sample user after implantation of the corresponding candidate lens;

[0138] The recommendation module 203 is configured to determine a target lens that meets preset conditions from the candidate lenses based on the predicted postoperative results.

[0139] In one implementation, the acquisition module 201 is specifically configured to:

[0140] Obtaining crystal parameters of multiple candidate crystals;

[0141] Obtain text data related to the target user and the characteristic data type corresponding to each candidate crystal;

[0142] Data corresponding to each characteristic data type is extracted from the text data, and user information corresponding to each candidate crystal is generated respectively.

[0143] In one implementation, the acquisition module 201 is specifically configured to:

[0144] Get training data;

[0145] The training data is classified according to the candidate lens to be implanted, and the user information of the sample users included in each classification is matched and analyzed to determine the characteristic data type required to obtain data before implanting each candidate lens; the data of the characteristic data type is used to indicate the preoperative condition of the sample user.

[0146] In one implementation, the prediction module 202 is specifically configured to:

[0147] In a case where the user information and / or the crystal parameter are discrete variables, performing one-hot encoding on the user information and / or the crystal parameter to obtain a first variable;

[0148] In a case where the user information and / or the crystal parameter are continuous variables, normalizing the user information and / or the crystal parameter to obtain a second variable;

[0149] Calling an activation function to process the first variable and the second variable to obtain an input variable;

[0150] The input variables are input into the target deep learning model for prediction to determine the predicted postoperative outcome corresponding to each candidate lens.

[0151] In one implementation, the apparatus further includes a training module configured to:

[0152] Acquire training data; the training data includes positive samples and negative samples;

[0153] Repeatedly sampling the negative samples to obtain training data;

[0154] Inputting the training data into a preset deep learning model for prediction, and obtaining the training postoperative results of the sample user after implantation of the corresponding candidate lens;

[0155] According to the training postoperative results, the sample postoperative results and the preset loss function, the model parameters of the preset deep learning model are iteratively adjusted to obtain the target deep learning model.

[0156] In one implementation, the preset deep learning model is a multi-layer perceptron model, a residual multi-layer perceptron model, or a Transformer model;

[0157] The loss function is a mean square loss function, a mean absolute error loss function or a smoothed mean absolute error loss function.

[0158] In one implementation, the prediction module 202 is specifically configured to:

[0159] Determine the crystal parameters of the candidate crystals that match the user information as target crystal parameters, input the user information and the target crystal parameters into a target deep learning model for prediction, and determine the predicted postoperative outcome corresponding to each candidate crystal; or

[0160] According to the preset grouping, the reference crystal corresponding to each group is selected from a plurality of candidate crystals, the user information and the crystal parameters of the reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to each reference crystal is determined. According to the predicted postoperative result, the target reference crystal that meets the preset conditions is determined from the reference crystals, and the crystal parameters of the candidate crystals in the same group as the user information and the target reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to the candidate crystals in the same group is determined.

[0161] In one implementation, the recommendation module 203 is configured to:

[0162] determining a preset number of candidate crystals whose predicted postoperative results are closest to a preset value as target crystals; or,

[0163] A risk assessment is performed on the predicted postoperative outcome, and a preset number of candidate crystals are determined as target crystals in ascending order of the risk assessment results.

[0164] As can be seen from the above, in the solution provided by the present invention, a deep learning algorithm is adopted, and user information and lens parameters can be modeled and analyzed through a pre-trained target deep learning model. The training data of the target deep learning model includes user information of sample users sampled from a wide range of people before lens transplant surgery and sample postoperative results obtained by postoperative vision review and eye physiological parameters. Therefore, the target lens corresponding to the target user can be effectively and accurately selected. Compared with traditional machine learning methods, the selection of lens is more accurate and can be personalized to meet the postoperative needs of different groups of people.

[0165] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0166] The device embodiments described above are merely illustrative. 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, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0167] The various component embodiments of the present disclosure can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the computing processing equipment according to the embodiments of the present disclosure. The present disclosure can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing part or all of the methods described herein. Such a program implementing the present disclosure can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0168] For example, FIG7 illustrates a computing processing device that can implement the methods of the present disclosure. The computing processing device conventionally includes a processor 1010 and a computer program product or computer-readable medium in the form of a memory 1020. Memory 1020 can be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, a hard disk, or ROM. Memory 1020 has storage space 1030 for program code 1031 for executing any of the method steps described above. Optionally, the computer-readable storage medium is RAM (generally referred to as DDR) for internal memory; for ROM, eMMC or UFS may be selected, depending on the technology. For example, the storage space 1030 for program code can include individual program codes 1031 for implementing various steps of the method described above. These program codes can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards, or floppy disks. Such computer program products are typically portable or fixed storage units, as described with reference to FIG8 . The storage unit may have storage segments, storage spaces, etc. arranged similarly to the memory 1020 in the computing and processing device of FIG7 . The program code may be compressed, for example, in an appropriate form. Typically, the storage unit includes computer-readable code 1031′, i.e., code that can be read by a processor such as 1010, which, when executed by the computing and processing device, causes the computing and processing device to perform the various steps of the method described above.

[0169] The communication bus mentioned in the terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0170] The communication interface is used for communication between the above terminal and other devices.

[0171] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0172] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0173] As can be seen from the above, in the solution provided by the present invention, a deep learning algorithm is adopted, and user information and lens parameters can be modeled and analyzed through a pre-trained target deep learning model. The training data of the target deep learning model includes user information of sample users sampled from a wide range of people before lens transplant surgery and sample postoperative results obtained by postoperative vision review and eye physiological parameters. Therefore, the target lens corresponding to the target user can be effectively and accurately selected. Compared with traditional machine learning methods, the selection of lens is more accurate and can be personalized to meet the postoperative needs of different groups of people.

[0174] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions that, when executed on a computer, enable the computer to execute the crystal recommendation method described in any one of the above embodiments.

[0175] In another embodiment of the present invention, a computer program product comprising instructions is provided. When the computer program product is run on a computer, the computer is enabled to execute the crystal recommendation method described in any one of the above embodiments.

[0176] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. 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 (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. 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 includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0177] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0178] Each embodiment in this specification is described in a related manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so the description is relatively simple. For related parts, refer to the description of the method embodiment.

[0179] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. A crystal recommendation method, wherein: include: Obtaining user information of a target user and crystal parameters of multiple candidate crystals; The user information and the crystal parameters are input into a target deep learning model for prediction, and the predicted postoperative results corresponding to each candidate crystal are determined; the target deep learning model is obtained by training training data with a preset deep learning model, and the training data includes the user information of a sample user, the crystal parameters, and the sample postoperative results of the sample user after the corresponding candidate crystal is implanted; According to the predicted postoperative result, a target crystal that meets preset conditions is determined from the candidate crystals.

2. The method according to claim 1, wherein: The step of obtaining user information of a target user and crystal parameters of multiple candidate crystals includes: Obtaining crystal parameters of multiple candidate crystals; Obtain text data related to the target user and the type of characteristic data corresponding to each candidate crystal; Data corresponding to each characteristic data type is extracted from the text data, and user information corresponding to each candidate crystal is generated respectively.

3. The method according to claim 2, wherein: Before acquiring the text data related to the target user and the type of characteristic data corresponding to the candidate crystal, the method further includes: Get training data; The training data is classified according to the implanted candidate lens, and the user information of the sample users included in each classification is matched and analyzed to determine the characteristic data type required to obtain data before implanting each candidate lens; the data of the characteristic data type is used to indicate the preoperative condition of the sample user.

4. The method according to claim 1, wherein: The step of inputting the user information and the crystal parameters into a target deep learning model for prediction, and determining a predicted postoperative outcome corresponding to each candidate crystal, includes: In the case where the user information and / or the crystal parameter are discrete variables, one-hot encoding is performed on the user information and / or the crystal parameter to obtain a first variable; In the case where the user information and / or the crystal parameter are continuous variables, normalizing the user information and / or the crystal parameter to obtain a second variable; Calling an activation function to process the first variable and the second variable to obtain an input variable; The input variables are input into the target deep learning model for prediction to determine the predicted postoperative outcome corresponding to each candidate lens.

5. The method according to claim 1, wherein: Before inputting the user information and the crystal parameters into the target deep learning model for prediction and determining the predicted postoperative result corresponding to each candidate crystal, the method includes: Acquire training data; the training data includes positive samples and negative samples; Repeatedly sampling the negative samples to obtain training data; Inputting the training data into a preset deep learning model for prediction, and obtaining the training postoperative results of the sample user after the corresponding candidate lens is implanted; According to the training postoperative results, the sample postoperative results and the preset loss function, the model parameters of the preset deep learning model are iteratively adjusted to obtain the target deep learning model.

6. The method according to claim 5, wherein: The preset deep learning model is a multi-layer perceptron model, a residual multi-layer perceptron model or a Transformer model; The loss function is a mean square loss function, a mean absolute error loss function or a smoothed mean absolute error loss function.

7. The method according to claim 1, wherein: The step of inputting the user information and the crystal parameters into a target deep learning model for prediction, and determining a predicted postoperative outcome corresponding to each candidate crystal, includes: Determine the crystal parameters of the candidate crystals that match the user information as the target crystal parameters, input the user information and the target crystal parameters into the target deep learning model for prediction, and determine the predicted postoperative outcome corresponding to each candidate crystal; or, According to the preset grouping, a reference crystal corresponding to each group is selected from a plurality of candidate crystals, the user information and the crystal parameters of the reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to each reference crystal is determined; based on the predicted postoperative result, a target reference crystal that meets the preset conditions is determined from the reference crystals, and the crystal parameters of the candidate crystals in the same group as the user information and the target reference crystal are input into the target deep learning model for prediction, and the predicted postoperative result corresponding to the candidate crystals in the same group is determined.

8. The method according to claim 1, wherein: Determining a target crystal that meets preset conditions from the candidate crystals according to the predicted postoperative result includes: Determine a preset number of candidate crystals whose predicted postoperative results are closest to a preset value as target crystals; or, A risk assessment is performed on the predicted postoperative outcome, and a preset number of candidate crystals are determined as target crystals in order of risk assessment results from small to large.

9. A crystal recommendation device, wherein: The method comprises: An acquisition module, used to acquire user information of a target user and crystal parameters of multiple candidate crystals; A prediction module is used to input the user information and the crystal parameters into a target deep learning model for prediction, and determine the predicted postoperative outcome corresponding to each candidate crystal; the target deep learning model is obtained by training training data with a preset deep learning model, and the training data includes the user information of a sample user, the crystal parameters, and the sample postoperative outcome of the sample user after the corresponding candidate crystal is implanted; A recommendation module is used to determine a target crystal that meets preset conditions from the candidate crystals according to the predicted postoperative results.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the crystal recommendation method according to any one of claims 1 to 8 are implemented.

11. A computer program comprising computer readable code, wherein: When the computer readable code is executed on a computing and processing device, the computing and processing device is caused to execute the crystal recommendation method according to any one of claims 1 to 8.

12. A computer-readable medium, wherein: The crystal recommendation method according to any one of claims 1 to 8 is stored.

Images