Refractive error prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a refractive error prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a target machine learning model from a plurality of candidate machine learning models based on a target group of target objects, the plurality of candidate machine learning models are obtained by pre-training based on a plurality of training sample data corresponding to the plurality of groups, and each training sample data comprises sample gene data, sample environment data and an ametropia risk tag; obtaining to-be-predicted data of the target object based on the target machine learning model; and inputting the to-be-predicted data into the target machine learning model to obtain a refractive error prediction result. The method can improve the refractive error prediction accuracy.

Description

Technical Field

[0001] This disclosure relates to the field of bioinformatics, and in particular to a method, apparatus, electronic device, and storage medium for predicting refractive errors. Background Technology

[0002] Refractive errors are a common and increasingly serious visual health problem worldwide. They not only cause blurred vision but can also be accompanied by a range of serious eye complications and systemic health issues. Therefore, refractive errors have become a major challenge to society and public health. While refractive errors can be diagnosed when patients experience symptoms such as blurred vision, the most effective way to address this problem is to predict and intervene in its early stages, before symptoms even appear.

[0003] However, the accuracy of predicting refractive errors in related technologies is low. Summary of the Invention

[0004] This disclosure provides a method, apparatus, electronic device, and storage medium for predicting refractive errors, which can improve the accuracy of refractive error prediction.

[0005] According to one aspect of this disclosure, a method for predicting refractive errors is provided, comprising:

[0006] The target machine learning model is determined from multiple candidate machine learning models based on the target group of the target object. The multiple candidate machine learning models are pre-trained based on multiple training sample data corresponding to multiple groups. Each training sample data includes sample gene data, sample environmental data, and refractive error risk label.

[0007] Based on the target machine learning model, obtain the data to be predicted for the target object;

[0008] The data to be predicted is input into the target machine learning model to obtain the refractive error prediction result.

[0009] According to one aspect of this disclosure, a refractive error prediction device is provided, comprising:

[0010] The determining unit is used to determine the target machine learning model from multiple candidate machine learning models based on the target group of the target object. The multiple candidate machine learning models are pre-trained based on multiple training sample data corresponding to multiple groups. Each training sample data includes sample gene data, sample environmental data, and refractive error risk label.

[0011] The acquisition unit is used to acquire the data to be predicted of the target object based on the target machine learning model;

[0012] The prediction unit is used to input the data to be predicted into the target machine learning model to obtain the refractive error prediction result.

[0013] Optionally, in one implementation, the determining unit is specifically used for:

[0014] Obtain multiple validation sample data corresponding to the target group of the target object;

[0015] The multiple candidate machine learning models are validated based on multiple validation sample data to obtain multiple validation scores;

[0016] Based on the multiple validation scores, the target machine learning model is determined from the multiple candidate machine learning models.

[0017] Optionally, in one implementation, the training process of the plurality of candidate machine learning models includes:

[0018] Acquire the multiple training sample data corresponding to the multiple groups, wherein each training sample data includes the sample gene data, the sample environmental data, and the refractive error risk label;

[0019] Multiple first preset machine learning models are trained based on multiple sample gene data and corresponding multiple refractive error risk labels to obtain multiple first candidate machine learning models and prediction data obtained by each first candidate machine learning model for each sample gene data.

[0020] The second preset machine learning model is trained based on the multiple prediction data obtained by the first candidate machine learning models for each sample gene data, the corresponding sample environment data, and the refractive error risk label to obtain the second candidate machine learning model.

[0021] The plurality of first candidate machine learning models and the second candidate machine learning models are determined as the plurality of candidate machine learning models.

[0022] Optionally, in one implementation, the step of training a second preset machine learning model based on multiple predicted data obtained by multiple first candidate machine learning models for each sample gene data, the corresponding sample environment data, and the refractive error risk label to obtain a second candidate machine learning model includes:

[0023] Using multiple predicted data and corresponding sample environment data as input features, and inputting the input features into the second preset machine learning model, risk prediction results are obtained;

[0024] The parameters of the second preset machine learning model are updated based on the loss between the risk prediction result and the refractive error risk label.

[0025] Optionally, in one implementation, obtaining the plurality of training sample data corresponding to the plurality of groups includes:

[0026] Acquire multiple candidate gene data corresponding to the multiple populations, and screen the multiple candidate gene data based on preset quality assessment conditions to determine multiple sample gene data;

[0027] Obtain the sample environment data and the refractive error risk label of the sample object corresponding to each of the sample gene data;

[0028] The multiple training sample data are constructed based on multiple sample gene data, multiple corresponding sample environmental data, and multiple refractive error risk labels.

[0029] Optionally, in one embodiment, after screening the plurality of candidate gene data based on preset quality assessment conditions to determine the plurality of sample gene data, the method further includes:

[0030] Genome-wide association analysis was performed on gene data from multiple samples to obtain the results of refractive error-related loci analysis.

[0031] Genetic filtering was performed on the gene data of multiple samples based on the results of refractive error-related site analysis and a predetermined threshold.

[0032] Optionally, in one embodiment, each training sample data includes the sample gene data, the sample environment data, the sample object behavior data, and the refractive error risk label;

[0033] The second preset machine learning model is trained based on multiple predicted data obtained from multiple first candidate machine learning models for each sample gene data, the corresponding sample environment data, and the refractive error risk label to obtain a second candidate machine learning model, including:

[0034] The second preset machine learning model is trained based on the multiple prediction data obtained by the first candidate machine learning model for each sample gene data, the corresponding sample environment data, the sample object behavior data, and the refractive error risk label, to obtain the second candidate machine learning model.

[0035] Optionally, in one embodiment, the refractive error prediction device further includes:

[0036] A receiving unit is configured to receive a prediction request for refractive error prediction of a target object, and to obtain the target group of the target object based on the prediction request;

[0037] The sending unit is used to send the refractive error prediction result to the initiator of the prediction request.

[0038] According to one aspect of this disclosure, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the refractive error prediction method as described above.

[0039] According to one aspect of this disclosure, a computer-readable storage medium is provided, the storage medium storing a computer program that, when executed by a processor, implements the refractive error prediction method as described above.

[0040] According to one aspect of this disclosure, a computer program product is provided, comprising a computer program that is read and executed by a processor of a computer device, causing the computer device to perform the refractive error prediction method as described above.

[0041] The refractive error prediction method in this embodiment determines a target machine learning model from multiple candidate machine learning models based on the target population of the target object. The multiple candidate machine learning models are pre-trained based on multiple training sample data corresponding to multiple populations. Each training sample data includes sample gene data, sample environmental data, and refractive error risk labels. The method obtains the data to be predicted for the target object based on the target machine learning model and inputs the data to be predicted into the target machine learning model to obtain the refractive error prediction result.

[0042] Therefore, in the refractive error prediction method of this disclosure, multiple candidate machine learning models trained based on sample genetic data, sample environmental data, and refractive error risk labels have the ability to predict refractive errors based on the object's genes and the object's environment. The causes of refractive errors differ for different groups; therefore, for different target groups, a target machine learning model with better prediction performance for the target group can be selected from multiple candidate machine learning models, and refractive error prediction is performed based on the data to be predicted required by the target machine learning model. Thus, the refractive error prediction method of this disclosure can improve the accuracy of refractive error prediction.

[0043] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objectives and other advantages of this disclosure may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0044] The accompanying drawings are provided to further understand the technical solutions of this disclosure and constitute a part of the specification. They are used together with the embodiments of this disclosure to explain the technical solutions of this disclosure and do not constitute a limitation on the technical solutions of this disclosure.

[0045] Figure 1 This is a system architecture diagram of a refractive error prediction method applied according to an embodiment of the present disclosure;

[0046] Figure 2 This is a schematic flowchart of a refractive error prediction method according to an embodiment of the present disclosure;

[0047] Figure 3 This is a schematic diagram of ROC curves and AUC according to an embodiment of the present disclosure;

[0048] Figure 4 This is a schematic diagram illustrating the training of a second preset machine learning model based on multiple prediction data and sample environment data according to one embodiment of the present disclosure;

[0049] Figure 5 This is a schematic diagram of a human body map interface according to one embodiment of the present disclosure;

[0050] Figure 6 This is a schematic diagram of the eye region interface according to one embodiment of the present disclosure;

[0051] Figure 7 This is a schematic diagram of a candidate machine learning model selection interface according to one embodiment of the present disclosure;

[0052] Figure 8 This is a schematic diagram of a model information interface according to one embodiment of the present disclosure;

[0053] Figure 9 This is a schematic diagram of a predicted data upload interface according to one embodiment of the present disclosure;

[0054] Figure 10 This is a schematic diagram of a refractive error prediction result display interface according to one embodiment of the present disclosure;

[0055] Figure 11 This is a schematic diagram of the structure of a refractive error prediction device according to an embodiment of the present disclosure;

[0056] Figure 12 This is a terminal structure diagram for implementing various methods according to an embodiment of the present disclosure;

[0057] Figure 13 This is a server structure diagram illustrating the implementation of various methods according to an embodiment of the present disclosure. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this disclosure.

[0059] Before providing a further detailed description of the embodiments of this disclosure, the terms and concepts used in these embodiments are explained, and they are subject to the following interpretations:

[0060] Refractive errors: Refractive errors refer to the inability of parallel light rays from more than 5 meters away to focus on the retina after being refracted by the eye's refractive system when the eye is at rest. Simply put, the eye cannot accurately focus light onto the retina, resulting in blurred vision. Refractive errors mainly include three types: myopia, hyperopia, and astigmatism.

[0061] Myopia: When looking at distant objects, the image falls in front of the retina, while when looking at near objects, the image falls on the retina, resulting in blurred vision when looking at distant objects and clear vision when looking at near objects.

[0062] Farsightedness: Regardless of distance, the image falls behind the retina and requires accommodation to see clearly; more accommodation is needed to see near objects.

[0063] Astigmatism: When looking at an object, multiple images fall on the vicinity of the retina at the same time, making the object appear partially blurred.

[0064] Machine learning: Machine learning is a branch of artificial intelligence that refers to the ability of computers to automatically improve and optimize algorithms by learning from data and patterns. Simply put, machine learning is a method that allows computers to learn from data to accomplish specific tasks.

[0065] High-throughput gene sequencing technology is a technique that enables the parallel sequencing of large amounts of DNA or RNA sequences in a short period of time. It is characterized by its high throughput and high resolution, allowing for the simultaneous sequencing of hundreds of thousands to millions of DNA molecules, thus providing a wealth of genetic information.

[0066] The system architecture used in the embodiments of this disclosure

[0067] Figure 1This is a system architecture diagram of a refractive error prediction method according to an embodiment of the present disclosure. It includes a terminal 140, an Internet 130, a gateway 120, a server 110, etc.

[0068] Terminal 140 can take various forms, including desktop computers, laptops, PDAs (personal digital assistants), mobile phones, in-vehicle terminals, smart terminals, and dedicated terminals. Furthermore, it can be a single device or a collection of multiple devices. Terminal 140 can communicate with the Internet 130 via wired or wireless means to exchange data.

[0069] Server 110 refers to a computer system that can provide certain services to terminal 140. Compared to ordinary terminal 140, server 110 has higher requirements in terms of stability, security, and performance. Server 110 can be a single high-performance computer in a network platform, a cluster of multiple high-performance computers, a portion of a single high-performance computer (e.g., a virtual machine), or a combination of portions of multiple high-performance computers (e.g., virtual machines).

[0070] Gateway 120, also known as an internetwork connector or protocol converter, is a computer system or device that acts as a translator, enabling network interconnection at the transport layer. It bridges the gap between two systems using different communication protocols, data formats, languages, or even completely different architectures. Gateways can also provide filtering and security functions. Messages sent from terminal 140 to server 110 are forwarded to the corresponding server 110 via gateway 120. Messages sent from server 110 to terminal 140 are also forwarded to the corresponding terminal 140 via gateway 120.

[0071] The refractive error prediction method of this disclosure can be implemented entirely on the terminal 140; it can be implemented entirely on the server 110; or it can be implemented partly on the terminal 140 and partly on the server 110.

[0072] General Description of Embodiments in this Disclosure

[0073] According to one embodiment of this disclosure, a method for predicting refractive errors is provided.

[0074] like Figure 2 The diagram shown is a flowchart of a refractive error prediction method provided in this disclosure. This method can be applied to a refractive error prediction device. The refractive error prediction method may include:

[0075] Step 210: Determine the target machine learning model from multiple candidate machine learning models based on the target group of the target object;

[0076] Step 220: Obtain the data to be predicted for the target object based on the target machine learning model;

[0077] Step 230: Input the data to be predicted into the target machine learning model to obtain the refractive error prediction result.

[0078] In step 210, the target machine learning model is determined from multiple candidate machine learning models based on the target population of the target object. The multiple candidate machine learning models are pre-trained based on multiple training sample data corresponding to multiple populations, and each training sample data includes sample genetic data, sample environmental data, and refractive error risk labels.

[0079] In one implementation, the training process for multiple candidate machine learning models includes:

[0080] Acquire multiple training sample data corresponding to multiple groups. Each training sample data includes sample genetic data, sample environmental data, and refractive error risk labels.

[0081] Multiple first-preset machine learning models are trained based on multiple sample gene data and corresponding multiple refractive error risk labels to obtain multiple first candidate machine learning models and the prediction data obtained by each first candidate machine learning model for each sample gene data.

[0082] The second preset machine learning model is trained based on multiple prediction data obtained from the gene data of each sample by multiple first candidate machine learning models, the corresponding sample environment data, and the refractive error risk label, to obtain the second candidate machine learning model.

[0083] Multiple first-candidate machine learning models and second-candidate machine learning models are identified as multiple candidate machine learning models.

[0084] A population can refer to the social group represented by the sample objects corresponding to the training sample data. For example, different cities or ethnic groups. The genetic makeup of individuals from different populations often shares some commonalities. For instance, the genetic makeup of individuals in population G1 makes them significantly more likely to have refractive errors than individuals in population G2. Furthermore, individuals from different populations exposed to the same environment may have an increased risk of certain diseases. Therefore, machine learning models can learn the associations between genetic and environmental data and refractive errors in different populations by using training sample data corresponding to multiple populations. This allows machine learning models to solve the problem of predicting refractive errors in different populations.

[0085] The sample genetic data in the training sample data can be the genetic data of the sample subjects, which may contain millions of single nucleotide polymorphism sites. Sample genetic data can be obtained from open-source biomedical databases, such as the UK Biobank database, which collects and analyzes health data from a large number of sample subjects worldwide, including genomic data, phenotypic data, lifestyle data, and medical records. It covers a wide range of diseases, such as cardiovascular disease, cancer, and refractive errors. Based on open-source biomedical databases, a large amount of genomic data corresponding to sample subjects can be obtained. Then, high-throughput gene sequencing technology can be used to sequence this genomic data to obtain the corresponding sample genetic data.

[0086] The environmental data in the training sample data can be data related to the sample object but unrelated to genetics. Specifically, it can include: macroscopic data of the sample object itself and environmental data of the sample object. Macroscopic data of the sample object itself can include characteristics of the sample object itself, such as age and gender. Environmental data of the sample object can include information about family members, family income, and characteristics of the surrounding natural environment. Environmental data can be obtained through surveys of the sample objects; it can also be collected from open-source databases. For example, environmental data for refractive error prediction can be collected from records of environmental data of sample objects in the aforementioned UK Biobank database.

[0087] Refractive error risk labels indicate the refractive error status of a sample object. A refractive error risk label can be a binary label, indicating whether the sample object has a refractive error. When the sample object has a refractive error, the refractive error risk label can be 1; otherwise, it can be 0. Refractive error risk labels can also be multi-class labels, indicating the refractive error status of the sample object. For example, refractive error risk labels can include: myopia, hyperopia, and normal; more specifically, refractive error risk labels can include: high myopia, mild myopia, high hyperopia, mild hyperopia, and normal. Refractive error risk labels can be determined based on the sample object's visual acuity. Storing the sample object's corresponding genetic data, environmental data, and refractive error risk labels can constitute the training sample data for the sample object.

[0088] In another implementation, each of the training sample data includes sample genetic data, sample environmental data, sample object behavior data, and refractive error risk labels.

[0089] Sample subject behavioral data refers to features related to the behavior of the sample subjects, such as their computer usage time, outdoor activity time, and sleep time. This data can be obtained through surveys of the sample subjects or collected from open-source databases; for example, the aforementioned UK Biobank database also records behavioral data related to the sample subjects. Training a model based on sample genetic data, sample environmental data, sample subject behavioral data, and refractive error risk labels allows the model to further learn the relationship between subject behavior and refractive error risk, thereby improving the accuracy of refractive error prediction.

[0090] In one implementation, acquiring multiple training sample data corresponding to multiple groups includes:

[0091] Acquire multiple candidate gene data corresponding to multiple populations, and screen the multiple candidate gene data based on preset quality assessment conditions to determine multiple sample gene data;

[0092] Obtain the sample environment data and refractive error risk label of the sample object corresponding to each sample gene data;

[0093] Multiple training sample data were constructed based on multiple sample genetic data, corresponding multiple sample environmental data, and multiple refractive error risk labels.

[0094] Multiple candidate gene data from multiple populations can also be obtained from open-source biomedical databases, such as the UK Biobank database. Preset quality assessment criteria can be used to screen data from multiple candidate gene datasets that do not meet predetermined quality requirements.

[0095] In one implementation, multiple candidate gene data are screened based on preset quality assessment conditions to determine multiple sample gene data. This includes: performing variation quality control on gene loci in the multiple candidate gene data according to predetermined conditions. Variation quality control refers to assessing and controlling the accuracy, validity, and reliability of variation data in the candidate gene data. A variation at a gene locus can lead to changes in protein structure or function, thereby causing some diseases. Therefore, obtaining multiple sample gene data by performing variation quality control on multiple candidate gene data can ensure the accuracy, validity, and reliability of the sample gene data, thereby improving the accuracy of refractive error prediction.

[0096] Variation quality control of gene loci in multiple candidate gene datasets can be performed using open-source gene analysis tools, such as PLINK, a powerful gene analysis software for quality control of gene data. In PLINK, parameters can be set to achieve variation quality control based on predetermined conditions. Configurable parameters may include: minimum allele frequency: 0.01; genotype deletion rate: 0.1; Hardy-Weinberg equilibrium: 1e-5; minimac3-r2 filtering threshold: 0.3; retain only single nucleotide polymorphisms; maximum number of alleles: 2.

[0097] In another implementation, multiple candidate gene data are screened based on preset quality assessment conditions to determine multiple sample gene data, including: performing sample quality control on the multiple candidate gene data according to predetermined conditions. Sample quality control can be an assessment and control of the accuracy, validity, and reliability of the overall candidate gene data sample. Performing quality control on the overall candidate gene data sample can further improve the accuracy, validity, and reliability of the sample gene data, thereby improving the accuracy of refractive error prediction.

[0098] Sample quality control for multiple candidate gene datasets can include: removing candidate gene datasets with abnormal heterozygosity and deletion rates; removing candidate gene datasets with abnormal genomic analysis results; removing candidate gene datasets with abnormal genetic relevance; and removing candidate gene datasets with corresponding erroneous sample data. Sample quality control of candidate gene datasets can also be performed using open-source gene analysis tools, such as PLINK.

[0099] In another implementation, multiple candidate gene data are screened based on preset quality assessment conditions to determine multiple sample gene data, including: screening multiple candidate gene data based on refractive error diseases to determine multiple sample gene data. When candidate gene data lacks data for assessing refractive errors, the candidate gene data can be removed. The data used to assess refractive errors can be the mean equivalent spherical refractive error (avMSE), a parameter used to describe the refractive state of the eye, which is the average value of the equivalent spherical power. The process of calculating asMSE may be affected by other diseases or events, so candidate gene data corresponding to diseases or events that would affect asMSE measurement (such as traumatic cataract, posterior cataract, amblyopia, corneal degeneration, paralytic strabismus, corneal edema, etc.) can also be removed.

[0100] In one implementation, after screening the multiple candidate gene data based on preset quality assessment conditions and determining multiple sample gene data, the method further includes: normalizing the multiple sample gene data.

[0101] Normalization can eliminate the effects of batch effects and physiological differences between individuals, ensuring the accuracy of prediction results. Probabilistic reference normalization can be used to normalize gene data from multiple samples. This transforms the data into a standard daily distribution, resulting in a mean of 0 and a standard deviation of 1.

[0102] In one implementation, after screening multiple candidate gene data based on preset quality assessment conditions and determining multiple sample gene data, the method further includes:

[0103] Genome-wide association analysis was performed on gene data from multiple samples to obtain the results of refractive error-related loci analysis.

[0104] Genetic filtering was performed on gene data from multiple samples based on the results of refractive error-related site analysis and a predetermined threshold.

[0105] Since candidate gene data can include genetic loci associated with various diseases, in order to make the prediction of refractive error risk more accurate, the genetic loci in the candidate gene data can be filtered to retain those that are significantly associated with refractive error.

[0106] Genome-wide association studies (GWAS) identify genetic loci significantly associated with diseases or traits across multiple sample gene datasets by comparing genotype data from case and control groups. GWAS can utilize software tools, such as Genome-wide Complex Trait Analysis (GCTA), a tool for genome-wide association studies that assesses the proportion of single nucleotide polymorphisms to the variance of complex trait phenotypes. Using the fastGWA tool within GCTA, a mixed linear model is employed for rapid analysis of sample gene data to obtain results on refractive error-related loci.

[0107] Genetic filtering of multiple sample gene data is performed based on the results of refractive error-related locus analysis and a predetermined threshold. The predetermined threshold can be a significance P-value, which is used to assess the strength of the association between an observed genetic variant and a specific phenotype. The P-value reflects the probability of obtaining the current or more extreme result if the null hypothesis (i.e., that the genetic variant is not associated with the phenotype) is true. The smaller the P-value, the less likely the observed association is to be accidental, and thus the more likely it is to reject the null hypothesis and conclude that there is a real association between the genetic variant and the phenotype. The refractive error-related locus analysis results include a significance P-value for each genetic locus. Filtering genetic loci based on significance P-values ​​helps improve the accuracy of filtering for genetic loci significantly associated with refractive errors. The predetermined threshold corresponding to the significance P-value can be between 1e-3 and 1e-8.

[0108] Genetic filtering based on genome-wide association analysis results can retain gene features significantly associated with refractive errors and remove features with low association with refractive error prediction, which helps improve the accuracy of refractive error prediction.

[0109] In another implementation, in addition to genome-wide association analysis, principal component analysis and partial least squares regression analysis can be used to screen key genetic features that are significantly associated with refractive errors from multiple sample gene data.

[0110] After screening multiple candidate gene data to determine multiple sample gene data, the sample environment data and refractive error risk labels of the sample objects corresponding to each sample gene data can be obtained. The process of obtaining the sample environment data and refractive error risk labels of the sample objects has been described in detail in the aforementioned embodiments and will not be repeated here. Training sample data is constructed based on multiple sample gene data and the corresponding multiple sample environment data and multiple refractive error risk labels.

[0111] By screening candidate gene data based on preset quality assessment conditions, the quality of sample gene data can be guaranteed, thereby improving the accuracy of subsequent model training based on sample gene data.

[0112] After obtaining multiple training sample data corresponding to multiple groups, multiple first-preset machine learning models can be trained based on multiple sample gene data and corresponding multiple refractive error risk labels to obtain multiple first-candidate machine learning models and the prediction data obtained by each first-candidate machine learning model for each sample gene data.

[0113] Multiple first-preset machine learning models are not limited to a specific machine learning model, but can be a variety of different machine learning algorithms, such as BayesA, LASSOSUM, LDpred2, PRS-CS, random forest algorithm, support vector machine, neural network, etc.

[0114] For different first-preset machine learning models, training can be performed based on multiple sample gene data and corresponding multiple refractive error risk labels.

[0115] In one implementation, a first preset machine learning model is trained based on multiple sample gene data and corresponding multiple refractive error risk labels to obtain a first candidate machine learning model, including:

[0116] Multiple sample gene data are input into the first preset machine learning model to obtain multiple refractive error risk prediction values ​​corresponding to the multiple sample gene data;

[0117] The loss value is calculated based on multiple predicted values ​​of refractive error risk and corresponding multiple refractive error risk labels, and the parameters of the first preset machine learning model are updated based on the loss value to obtain the corresponding first candidate machine learning model.

[0118] The loss value between the predicted refractive error risk value and its corresponding multiple refractive error risk labels describes the difference between the predicted value and the true value of the first preset machine learning model. The smaller the loss value, the closer the predicted value of the first preset machine learning model is to the true value, and the better the prediction effect of the first preset machine learning model. After calculating the loss value, it can be backpropagated to the first preset machine learning model to update the parameters of the first preset machine learning model in the direction of gradient descent of the loss value.

[0119] Based on the above process, multiple first-preset machine learning models are trained to obtain multiple corresponding first-candidate machine learning models.

[0120] After a predetermined number of iterations, the first preset machine learning model can be validated. When the first preset machine learning model meets the predetermined validation conditions, the first candidate machine learning model can be obtained.

[0121] In one implementation, validating the first preset machine learning model includes: acquiring multiple validation sample data corresponding to multiple groups, wherein the validation sample data includes sample gene data, sample environmental data, and refractive error risk label data; inputting the sample gene data into the first preset machine learning model to obtain the refractive error prediction result corresponding to the first preset machine learning model; and calculating the prediction accuracy score based on the refractive error prediction result and the refractive error risk label.

[0122] After obtaining sample data from multiple groups, the sample data can be divided into training sample data and validation sample data according to a predetermined ratio. Training sample data can be used for model training, and validation sample data can be used for model validation. For example, the sample data can be divided in an 8:2 ratio, meaning 80% of the sample data is used as training sample data and 20% as validation sample data.

[0123] The prediction accuracy score can be obtained by calculating the area under the curve (AUC) based on the refractive error prediction results and refractive error risk labels corresponding to multiple validation sample data.

[0124] AUC is the area under the Receiver Operating Characteristic (ROC) curve. The ROC curve plots the True Positive Rate (TPR) on the vertical axis and the False Positive Rate (FPR) on the horizontal axis. The TPR represents the percentage of times the model correctly identifies a positive class, while the FPR represents the percentage of times the model incorrectly identifies a negative class as a positive class. For example... Figure 3 The ROC curve shown starts at (0,0) and extends upwards to the right. The area under the ROC curve is the AUC value. The AUC value ranges from 0 to 1, and a higher value indicates better classification performance of the model.

[0125] When the prediction accuracy score reaches a predetermined threshold, the first preset machine learning model can be identified as the first candidate machine learning model.

[0126] In another implementation, the first preset machine learning model can be cross-validated based on multiple training sample data and multiple validation sample data corresponding to multiple groups, thereby determining the first candidate machine learning model.

[0127] The cross-validation process includes: determining a predetermined number of training iterations, dividing the original sample dataset into multiple sample data subsets according to the predetermined number of training iterations, and using multiple sample data from the target sample data subset as multiple validation sample data; using multiple sample data from other sample data subsets besides the target sample data subset as multiple training sample data; training a first preset machine learning model based on the multiple training sample data in the target training iteration, and validating the first preset machine learning model based on the multiple validation sample data to obtain the prediction accuracy score corresponding to the target training iteration. The above process is repeated in each target training iteration to obtain the corresponding prediction accuracy score. The calculation process of the prediction accuracy score has been described in detail in the aforementioned embodiments and will not be repeated here. The average prediction accuracy score of the multiple prediction accuracy scores is calculated. When the average prediction accuracy score reaches a predetermined threshold, the first preset machine learning model can be determined as the first candidate machine learning model.

[0128] For example, performing five-fold cross-validation on multiple pre-defined machine learning models involves dividing the original sample datasets corresponding to multiple groups into five subsets: U1, U2, U3, U4, and U5. In the first round of training, the sample data in subset U1 is used as validation samples, and the sample data in subsets U2 through U5 is used as training samples. The training samples are used to train the multiple pre-defined machine learning models, and the validation samples are used to validate the trained models, yielding prediction accuracy scores. Then, in the second round of training, the sample data in subset U2 is used as validation samples, and the sample data in subsets U1 and U3 through U5 are used as training samples. After performing the above training and validation processes, the prediction accuracy scores for the second round of training are obtained. Similarly, in the subsequent three rounds of training, multiple samples from sample data subsets U3, U4, and U5 are used as multiple validation samples, and multiple samples from the remaining multiple sample data subsets are used as multiple training samples. The training process is performed using the training samples, and the validation process is performed using the validation samples, resulting in three prediction accuracy scores for the three rounds of training. The average prediction accuracy score is obtained by averaging the prediction accuracy scores obtained from the five rounds of training.

[0129] After training multiple first-candidate machine learning models, multiple sample gene data can be input into these models to obtain the prediction data for each model for each sample gene data. For example, if a sample gene data is input into multiple first-candidate machine learning models, and the first-candidate machine learning model trained using the random forest algorithm predicts a prediction data of 0.8 for that sample gene data, this indicates that the random forest algorithm predicts an 80% probability that the corresponding sample subject has refractive errors; if the first-candidate machine learning model trained using the LDpred2 algorithm predicts a prediction data of 0.7 for that sample gene data, this indicates that the LDpred2 algorithm predicts a 70% probability that the corresponding sample subject has refractive errors.

[0130] After obtaining the prediction data of each first candidate machine learning model for each sample gene data, the second preset machine learning model can be trained based on the multiple prediction data obtained by multiple first candidate machine learning models for each sample gene data, the corresponding sample environment data, and the refractive error risk label to obtain the second candidate machine learning model.

[0131] The second pre-defined machine learning model can be an algorithm capable of integrating multiple machine learning models. This allows the second pre-defined machine learning model to learn the ability to predict refractive error risk based on sample genetic data from multiple first candidate machine learning models. In one implementation, the second pre-defined machine learning model is a CatBoost model, a gradient boosting machine learning algorithm.

[0132] In one implementation, a second preset machine learning model is trained based on multiple prediction data obtained from multiple first candidate machine learning models for each sample's gene data, corresponding sample environment data, and refractive error risk labels to obtain a second candidate machine learning model, including:

[0133] Using multiple prediction data and corresponding sample environment data as input features, and inputting the input features into a second preset machine learning model, risk prediction results are obtained;

[0134] The parameters of the second preset machine learning model are updated based on the loss between the risk prediction results and the refractive error risk label.

[0135] The second candidate machine learning model makes predictions based on multiple refractive error maple trees and sample environment data corresponding to multiple first candidate machine learning models, obtaining risk prediction results for the training sample data. The risk prediction results indicate whether the corresponding sample object has refractive errors. The parameters of the second preset machine learning model are updated based on the loss between the risk prediction results and the refractive error risk label. The process of updating the parameters of the second preset machine learning model is the same as the process of updating the parameters of the first preset machine learning model mentioned above, and will not be repeated here.

[0136] like Figure 4 As shown, the sample gene data are input into the first candidate machine learning model M1, the first candidate machine learning model M2, the first candidate machine learning model M3, and the first candidate machine learning model M4, respectively, to obtain predicted data s1, predicted data s2, predicted data s3, and predicted data s4. The multiple predicted data and the sample environment data are used as input features and input into the second preset machine learning model N1 to obtain the risk prediction result. The parameters in the second preset machine learning model N1 are updated based on the loss between the risk prediction result and the refractive error risk label.

[0137] By using multiple prediction data and sample environment data as input features, the second preset machine learning model can combine the risk prediction results corresponding to multiple first candidate machine learning models with the sample environment data to predict the refractive error risk of the sample object, which is beneficial to improving the accuracy of the second preset machine learning model in refractive error risk analysis.

[0138] Training a second pre-set machine learning model with multiple prediction data, sample environment data, and corresponding refractive error risk labels corresponding to multiple first candidate machine learning models allows the second pre-set machine learning model to quickly learn the ability to predict refractive error risk based on sample gene data from multiple first candidate machine learning models. By combining sample environment data, the accuracy of sample refractive error risk prediction is further improved, while the training efficiency of the second candidate machine learning model is also improved.

[0139] In one of the aforementioned embodiments, each training sample data includes not only sample gene data, sample environment data, and refractive error risk labels, but also sample object behavior data.

[0140] Based on this, in one implementation, a second preset machine learning model is trained using multiple prediction data obtained from multiple first candidate machine learning models for each sample's gene data, corresponding sample environment data, and refractive error risk labels to obtain a second candidate machine learning model, including:

[0141] The second preset machine learning model is trained based on multiple prediction data obtained from the gene data of each sample by multiple first candidate machine learning models, the corresponding sample environment data, sample object behavior data, and refractive error risk labels, to obtain the second candidate machine learning model.

[0142] In one implementation, a second preset machine learning model is trained using prediction data, sample environment data, sample object behavior data, and refractive error risk labels. Multiple prediction data, corresponding sample environment data, and sample object behavior data can be used as input features. The input features are then fed into the second preset machine learning model to obtain risk prediction results. The parameters of the second preset machine learning model are then updated based on the loss between the risk prediction results and the refractive error risk labels.

[0143] After a predetermined number of iterations of parameter updates, the second preset machine learning model can be validated. When the second preset machine learning model meets the predetermined validation conditions, the second candidate machine learning model can be obtained.

[0144] In one of the aforementioned embodiments, the first preset machine learning model is validated using multiple validation sample data corresponding to multiple groups, and the validation sample data also includes sample environment data. Therefore, validating the second preset machine learning model includes:

[0145] The prediction data obtained by multiple first candidate machine learning models based on the sample gene data in the validation sample data and the sample environment data are input into the second preset machine learning model to obtain the refractive error prediction result corresponding to the second preset machine learning model; the prediction accuracy score is calculated based on the refractive error prediction result and the refractive error risk label.

[0146] The prediction accuracy score can also be an AUC index calculated based on multiple refractive error prediction results obtained from the second prediction machine learning model and the corresponding refractive error risk labels, which will not be elaborated here.

[0147] When the prediction accuracy score reaches a predetermined threshold, the second preset machine learning model can be identified as the second candidate machine learning model.

[0148] In another implementation, the second preset machine learning model can be cross-validated based on multiple training sample data and multiple validation sample data corresponding to multiple groups to determine the second candidate machine learning model. The method of cross-validating the second preset machine learning model is the same as that of cross-validating the first preset machine learning model, and will not be described again here.

[0149] It should be noted that when the first and second candidate machine learning models predict refractive errors, they can predict whether the object has refractive errors; or they can make multi-class predictions for subcategories of refractive errors, such as whether the object has high myopia, low myopia, high hyperopia, and low hyperopia.

[0150] After training and obtaining multiple first-candidate machine learning models and second-candidate machine learning models, these multiple first-candidate machine learning models and second-candidate machine learning models can be identified as multiple candidate machine learning models.

[0151] When it is necessary to predict refractive errors in a target object, the target machine learning model can be determined from multiple candidate machine learning models based on the target object's target population. The target object can be the object used for refractive error prediction, and the target population can be the group to which the target object belongs.

[0152] In one implementation, before determining the target machine learning model from multiple candidate machine learning models based on the target group of the target object, the method further includes: receiving a prediction request for refractive error prediction of the target object, and obtaining the target group of the target object based on the prediction request.

[0153] When it is necessary to predict refractive errors in a target subject, a prediction request can be initiated. The prediction request can include target group information for the target subject.

[0154] Multiple candidate machine learning models are trained on training sample data from multiple populations. However, the target machine learning model that yields more accurate predictions may differ for different target populations. In some target populations, refractive errors are primarily caused by genetic factors, and even with fewer external environmental factors contributing to refractive errors, there is still a high probability of developing them. Thus, the target machine learning model for a target population can be determined from multiple models trained on sample genetic data. Furthermore, different candidate machine learning models have different architectures and data processing methods; therefore, for multiple populations significantly influenced by genetic factors, the genes causing refractive errors may differ, and the applicable candidate machine learning models may also differ. Similarly, in some target populations, refractive errors are less affected by genetic factors; therefore, the target machine learning model can be determined from models trained on sample genetic and environmental data, allowing for refractive error risk prediction based on genetic and environmental factors.

[0155] In one implementation, determining a target machine learning model from multiple candidate machine learning models based on a target population of the target object includes:

[0156] Obtain multiple validation sample data corresponding to the target group of the target object;

[0157] Multiple candidate machine learning models are validated based on multiple validation sample data, resulting in multiple validation scores;

[0158] The target machine learning model is determined from multiple candidate machine learning models based on multiple validation scores.

[0159] Multiple validation sample data can be multiple sample data corresponding to the target population. Validation sample data may also include sample genetic data, sample environmental data, and refractive error risk labels.

[0160] When acquiring sample data, data can be obtained for each group and divided into training and validation data according to a predetermined ratio. The training data from multiple groups can be aggregated and used to train multiple candidate machine learning models. The validation data for each group can be used to validate the predictive performance of multiple candidate machine learning models within that group.

[0161] Multiple candidate machine learning models can be validated using multiple validation sample data, resulting in multiple validation scores.

[0162] In one of the aforementioned embodiments, the multiple candidate machine learning models include a first candidate machine learning model trained based on sample genetic data and refractive error risk labels, and a second candidate machine learning model trained based on sample genetic data, sample environmental data, and refractive error risk labels.

[0163] Therefore, when validating multiple first-candidate machine learning models, for each first-candidate machine learning model, the sample gene data from the validation sample data can be input into the first-candidate machine learning model to obtain the first refractive error risk prediction result; the validation score corresponding to the first-candidate machine learning model is calculated based on the first refractive error risk prediction result and the refractive error risk label. When validating the second-candidate machine learning model, multiple first refractive error risk prediction results output by multiple first-candidate machine learning models based on sample gene data and sample environmental data can be input into the second-candidate machine learning model to obtain the second refractive error risk prediction result; the validation score corresponding to the second-candidate machine learning model is calculated based on the second refractive error risk prediction result and the refractive error risk label.

[0164] For each first candidate machine learning model, the AUC value corresponding to the first candidate machine learning model can be determined as the validation score based on the first refractive error risk prediction result and the refractive error risk label corresponding to each validation sample data. For the second candidate machine learning model, the AUC value corresponding to the second candidate machine learning model can be determined as the validation score based on the second refractive error risk prediction result and the refractive error risk label corresponding to each validation sample data.

[0165] The validation score can also be the Positive Predictive Value (PPV), which measures the proportion of samples predicted as positive that are actually positive; or the Negative Predictive Value (NPV), which measures the proportion of samples detected as negative that are actually negative.

[0166] In one implementation, the first candidate machine learning model includes BayesA, Lassosum, Ldpred2, Random forest, Plink, and Prs_cs, and the second candidate machine learning model is an ensemble model based on CatBoost. When multiple first and second candidate machine learning models are used to predict multiple refractive error subclasses (high myopia, low myopia, high hyperopia, and low hyperopia) for three groups, the corresponding validation results are shown in Table 2.

[0167]

[0168]

[0169]

[0170]

[0171]

[0172]

[0173]

[0174]

[0175]

[0176]

[0177]

[0178]

[0179]

[0180]

[0181]

[0182]

[0183]

[0184] Table 1

[0185] Finally, based on multiple validation scores, the target machine learning model is determined from multiple candidate machine learning models. Specifically, the target machine learning model with the highest validation score among the multiple candidate machine learning models can be determined as the target machine learning model corresponding to the target group.

[0186] Multiple candidate machine learning models are validated based on validation sample data corresponding to the target group, and the target machine learning model is determined based on the validation results. The target machine learning model can be determined based on the refractive error prediction performance of different models in the target group, which improves the accuracy of determining the target machine learning model.

[0187] In another implementation, the target machine learning model is determined from multiple candidate machine learning models based on the target population of the target object, including:

[0188] The candidate machine learning model selection interface displays multiple model name controls corresponding to multiple candidate machine learning models;

[0189] In response to a trigger on the model name control, display the model validation data corresponding to the model name control;

[0190] The target machine learning model is determined based on the target group and model validation data of the target object.

[0191] The candidate machine learning model selection interface can be a model selection interface for predicting refractive errors in a disease prediction platform. In one implementation, the disease prediction platform can also construct candidate machine learning models for various diseases of different human tissues and organs to complete risk prediction for diseases of different tissues and organs. For example... Figure 5 As shown, the human body map interface 510 in the disease prediction platform can display human tissue statistics 511, which includes detailed descriptions of different tissues and organs in the human body. When an object clicks on a target tissue or organ, it can jump to the candidate machine learning model selection interface corresponding to the target tissue or organ, and then select a model to perform disease risk assessment for the target tissue or organ.

[0192] In another implementation, the disease prediction platform can predict diseases for different regions of the eye. Since different regions of the eye correspond to different diseases, the disease prediction platform can also develop candidate machine learning model structures for different regions of the eye to predict the disease risk for those regions. For example... Figure 6 As shown, the eye region interface 610 in the disease prediction platform displays eye tissue statistics 611, which includes detailed descriptions of different regions within the eye. When an user clicks on a target region, they are redirected to a candidate machine learning model selection interface corresponding to that region, allowing them to choose a model to assess disease risk for that target region of the eye.

[0193] The candidate machine learning model selection interface can display multiple model name controls corresponding to several candidate machine learning models used for refractive error prediction. For example... Figure 7As shown, the candidate machine learning model selection interface 710 can display multiple model name controls 712 corresponding to multiple candidate machine learning models, and can also display a filter condition input box 711. The filter condition input box 711 can be used to input model filter conditions, which can include algorithm type, data source, significance p-value, etc.

[0194] When the model name control is triggered, the corresponding model validation data will be displayed. Model validation data can include detailed model information and model evaluation results. Detailed model information may include the domain in which the model is applied, the samples used, the algorithm employed, and the significance level p-values ​​employed. Model evaluation results may include evaluation results obtained after validating the model on different populations, such as ROC curves. Figure 8 As shown, after triggering the model name control 712 corresponding to the candidate machine learning model M1, the model information interface 810 can display the model details 811 and the model evaluation results 812.

[0195] Based on the target group corresponding to the target object and the model validation data, the model that performs better for the target group can be selected from multiple candidate machine learning models and used as the target machine learning model for predicting the risk of refractive errors in the target object.

[0196] In step 220, the data to be predicted for the target object is obtained based on the target machine learning model.

[0197] The data used for refractive error prediction may differ for different candidate machine learning models. Therefore, the data to be predicted for the target object can be obtained based on the target machine learning model.

[0198] In one of the aforementioned embodiments, the multiple candidate machine learning models include a first candidate machine learning model trained based on sample genetic data and refractive error risk labels, and a second candidate machine learning model trained based on sample genetic data, sample environmental data, and refractive error risk labels.

[0199] Therefore, when the target machine learning model is one of several first-candidate machine learning models, the data to be predicted can be the target gene data corresponding to the target object. When the target machine learning model is a second-candidate machine learning model, the data to be predicted can be both the target gene data and the target environment data corresponding to the target object.

[0200] In step 230, the data to be predicted is input into the target machine learning model to obtain the refractive error prediction result.

[0201] Based on the above implementation method, when the target machine learning model is a first candidate machine learning model, inputting the target prediction data into the target machine learning model can yield the refractive error prediction result corresponding to the target object. When the target machine learning model is a second candidate machine learning model, the target gene data can be input into multiple first candidate machine learning models to obtain multiple refractive error risk scores, and the multiple refractive error risk scores and target environmental data can be input into the second candidate machine learning model to obtain the refractive error prediction result corresponding to the target object.

[0202] In one of the aforementioned embodiments, after receiving a prediction request for refractive error prediction of a target object, refractive error prediction is performed on the target object based on steps 210-230. Based on this, after inputting the data to be predicted into the target machine learning model and obtaining the refractive error prediction result, the method further includes: sending the refractive error prediction result to the party initiating the prediction request.

[0203] By sending the refractive error prediction results to the initiator of the prediction request, feedback on the refractive error prediction results of the target object can be achieved, thereby improving the usability and interactivity of the refractive error prediction method.

[0204] In one of the aforementioned embodiments, the target machine learning model can be determined based on the model validation data displayed by multiple candidate machine learning models shown in the candidate machine learning model selection interface and the target group. Based on this, in one embodiment, obtaining the data to be predicted for the target object based on the target machine learning model includes:

[0205] In response to the triggering of the selection control corresponding to the target machine learning model, the prediction data type corresponding to the target machine learning model is displayed so as to determine the data to be predicted for the target object based on the prediction data type;

[0206] The data to be predicted is input into the target machine learning model to obtain the refractive error prediction results, including:

[0207] In response to the upload command that inputs the data to be predicted into the target machine learning model, the refractive error prediction results are displayed.

[0208] For example Figure 9As shown, the target machine learning model is candidate machine learning model M1. In response to the triggering of the selection control corresponding to the candidate machine learning model, the prediction data upload interface 910 corresponding to candidate machine learning model M1 is displayed. The prediction data upload interface 910 includes a prediction data upload module 911 and an upload control 912. The prediction data upload module 911 allows the upload of data to be predicted. Below the prediction data upload module 911, it is indicated that the data to be uploaded is target gene data. After uploading the target gene data, the upload control 912 can be triggered to upload the target gene data to candidate machine learning model M1 for refractive error prediction.

[0209] Candidate machine learning model M1 can display refractive error prediction results after performing refractive error prediction, for example... Figure 10 As shown. The refractive error prediction result display interface 1010 can display the refractive error prediction results. The refractive error prediction result display interface 1010 includes a target object refractive error prediction result display area 1011 and a target population refractive error distribution display area 1012. The target object refractive error prediction result display area 1011 displays the refractive error prediction results obtained by the candidate machine learning model M1 based on the target gene data; the target population refractive error distribution display area 1012 displays the distribution of the target candidate machine learning model M1's refractive error predictions for the target population.

[0210] Visualizing the results of refractive error prediction can make it easier to obtain the prediction results and improve prediction efficiency.

[0211] Description of apparatus and devices according to embodiments of this disclosure

[0212] It is understood that although the steps in the above flowcharts are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated in this embodiment, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the above flowcharts may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.

[0213] It should be noted that in the various specific embodiments of this disclosure, when processing is required based on data related to the characteristics of the target object, such as target object attribute information or a set of attribute information, the permission or consent of the target object will be obtained first. Furthermore, the collection, use, and processing of this data will comply with the relevant laws, regulations, and standards of the relevant regions. In addition, when this application embodiment needs to obtain target object attribute information, separate permission or consent from the target object will be obtained through pop-up windows or redirection to a confirmation page. Only after obtaining the target object's separate permission or consent will the necessary target object-related data for the normal operation of this application embodiment be obtained.

[0214] Figure 11 A schematic diagram of the structure of the refractive error prediction device 1100 provided in this embodiment of the disclosure. The refractive error prediction device 1100 includes:

[0215] The determining unit 1110 is used to determine the target machine learning model from multiple candidate machine learning models based on the target group of the target object. The multiple candidate machine learning models are pre-trained based on multiple training sample data corresponding to multiple groups. Each training sample data includes sample gene data, sample environmental data, and refractive error risk label.

[0216] The acquisition unit 1120 is used to acquire the data to be predicted of the target object based on the target machine learning model;

[0217] The prediction unit 1130 is used to input the data to be predicted into the target machine learning model to obtain the refractive error prediction result.

[0218] Optionally, in one embodiment, the determining unit 1110 is specifically used for:

[0219] Obtain multiple validation sample data corresponding to the target group of the target object;

[0220] Multiple candidate machine learning models are validated based on multiple validation sample data, resulting in multiple validation scores;

[0221] The target machine learning model is determined from multiple candidate machine learning models based on multiple validation scores.

[0222] Optionally, in one implementation, the training process for multiple candidate machine learning models includes:

[0223] Acquire multiple training sample data corresponding to multiple groups. Each training sample data includes sample genetic data, sample environmental data, and refractive error risk labels.

[0224] Multiple first-preset machine learning models are trained based on multiple sample gene data and corresponding multiple refractive error risk labels to obtain multiple first candidate machine learning models and the prediction data obtained by each first candidate machine learning model for each sample gene data.

[0225] The second preset machine learning model is trained based on multiple prediction data obtained from the gene data of each sample by multiple first candidate machine learning models, the corresponding sample environment data, and the refractive error risk label, to obtain the second candidate machine learning model.

[0226] Multiple first-candidate machine learning models and second-candidate machine learning models are identified as multiple candidate machine learning models.

[0227] Optionally, in one implementation, a second preset machine learning model is trained based on multiple prediction data obtained from multiple first candidate machine learning models for each sample's gene data, corresponding sample environment data, and refractive error risk labels to obtain a second candidate machine learning model, including:

[0228] Using multiple prediction data and corresponding sample environment data as input features, and inputting the input features into a second preset machine learning model, risk prediction results are obtained;

[0229] The parameters of the second preset machine learning model are updated based on the loss between the risk prediction results and the refractive error risk label.

[0230] Optionally, in one implementation, acquiring multiple training sample data corresponding to multiple groups includes:

[0231] Acquire multiple candidate gene data corresponding to multiple populations, and screen the multiple candidate gene data based on preset quality assessment conditions to determine multiple sample gene data;

[0232] Obtain the sample environment data and refractive error risk label of the sample object corresponding to each sample gene data;

[0233] Multiple training sample data were constructed based on multiple sample genetic data, corresponding multiple sample environmental data, and multiple refractive error risk labels.

[0234] Optionally, in one implementation, after screening multiple candidate gene data based on preset quality assessment conditions and determining multiple sample gene data, the method further includes:

[0235] Genome-wide association analysis was performed on gene data from multiple samples to obtain the results of refractive error-related loci analysis.

[0236] Genetic filtering was performed on gene data from multiple samples based on the results of refractive error-related site analysis and a predetermined threshold.

[0237] Optionally, in one implementation, each training sample data includes sample genetic data, sample environmental data, sample object behavior data, and refractive error risk labels;

[0238] The second candidate machine learning model is trained based on multiple prediction data obtained from the gene data of each sample by multiple first candidate machine learning models, the corresponding sample environment data, and the refractive error risk label, to obtain the second candidate machine learning model, including:

[0239] The second preset machine learning model is trained based on multiple prediction data obtained from the gene data of each sample by multiple first candidate machine learning models, the corresponding sample environment data, sample object behavior data, and refractive error risk labels, to obtain the second candidate machine learning model.

[0240] Optionally, in one embodiment, the refractive error prediction device 1100 further includes:

[0241] A receiving unit (not shown) is used to receive a prediction request for refractive error prediction of a target object and to obtain the target group of the target object based on the prediction request.

[0242] A sending unit (not shown) is used to send the refractive error prediction result to the initiator of the prediction request.

[0243] Reference Figure 12 , Figure 12 To implement the refractive error prediction method of this embodiment, the terminal 140 includes a partial structural block diagram, comprising: a radio frequency (RF) circuit 1210, a memory 1215, an input unit 1230, a display unit 1240, a sensor 1250, an audio circuit 1260, a wireless fidelity (WiFi) module 1270, a processor 1280, and a power supply 1290, etc. Those skilled in the art will understand that... Figure 12 The terminal 140 structure shown does not constitute a limitation on a mobile phone or computer, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0244] The RF circuit 1210 can be used to receive and transmit signals during information transmission or calls. In particular, it receives downlink information from the base station and processes it with the processor 1280; in addition, it transmits uplink data to the base station.

[0245] The memory 1215 can be used to store software programs and modules. The processor 1280 executes various functional applications of the terminal and the identification of lane line change points by running the software programs and modules stored in the memory 1215.

[0246] The input unit 1230 can be used to receive input numeric or character information, and to generate key signal inputs related to the terminal's settings and function control. Specifically, the input unit 1230 may include a touch panel 1231 and other input devices 1232.

[0247] The display unit 1240 can be used to display input or provided information, as well as various menus of the terminal. The display unit 1240 may include a display panel 1241.

[0248] Audio circuitry 1260, speaker 1261, and microphone 1262 provide an audio interface.

[0249] In this embodiment, the processor 1280 included in the terminal 140 can execute the refractive error prediction method of the previous embodiment.

[0250] The terminal 140 in this disclosure includes, but is not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, and aircraft. This invention can be applied to various scenarios, including but not limited to disease diagnosis and vision correction.

[0251] Figure 13 This is a partial structural block diagram of a server 110 for implementing the refractive error prediction method of this disclosure. The server 110 can vary significantly due to different configurations or performance, and may include one or more central processing units (CPUs) 1322 (e.g., one or more processors) and storage devices 1332, and one or more storage media 1330 (e.g., one or more mass storage devices) for storing application programs 1342 or data 1344. The storage devices 1332 and storage media 1330 can be temporary or persistent storage. The program stored in the storage media 1330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server 110. Furthermore, the CPU 1322 may be configured to communicate with the storage media 1330 and execute the series of instruction operations in the storage media 1330 on the server 110.

[0252] Server 110 may also include one or more power supplies 1326, one or more wired or wireless network interfaces 1350, one or more input / output interfaces 1358, and / or one or more operating systems 1341, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0253] The central processing unit 1322 in server 110 can be used to execute the refractive error prediction method of the present disclosure embodiments.

[0254] This disclosure also provides a computer-readable storage medium for storing program code for executing the refractive error prediction methods of the foregoing embodiments.

[0255] This disclosure also provides a computer program product comprising a computer program. A processor of a computer device reads and executes the computer program, causing the computer device to perform the refractive error prediction method described above.

[0256] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in this disclosure and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “including,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatuses.

[0257] It should be understood that in this disclosure, "at least one item" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer 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 represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0258] It should be understood that in the description of the embodiments disclosed herein, "multiple" means two or more, "greater than", "less than", "exceeding" etc. are understood to exclude the number itself, and "above", "below", "within" etc. are understood to include the number itself.

[0259] In the several embodiments provided in this disclosure, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0260] The units described as separate components may or may not be physically separate. 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 the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0261] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0262] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0263] It should also be understood that the various implementation methods provided in this disclosure can be combined arbitrarily to achieve different technical effects.

[0264] The above is a detailed description of the embodiments of this disclosure. However, this disclosure is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this disclosure. All such equivalent modifications or substitutions are included within the scope defined by the claims of this disclosure.

Claims

1. A method for predicting refractive errors, characterized in that, include: The target machine learning model is determined from multiple candidate machine learning models based on the target group of the target object. The multiple candidate machine learning models are pre-trained based on multiple training sample data corresponding to multiple groups. Each training sample data includes sample gene data, sample environmental data, and refractive error risk label. Based on the target machine learning model, obtain the data to be predicted for the target object; The data to be predicted is input into the target machine learning model to obtain the refractive error prediction result.

2. The method according to claim 1, characterized in that, The target machine learning model is determined from multiple candidate machine learning models based on the target population of the target object, including: Obtain multiple validation sample data corresponding to the target group of the target object; The multiple candidate machine learning models are validated based on multiple validation sample data to obtain multiple validation scores; Based on the multiple validation scores, the target machine learning model is determined from the multiple candidate machine learning models.

3. The method according to claim 1, characterized in that, The training process for the multiple candidate machine learning models includes: Acquire the multiple training sample data corresponding to the multiple groups, wherein each training sample data includes the sample gene data, the sample environmental data, and the refractive error risk label; Multiple first preset machine learning models are trained based on multiple sample gene data and corresponding multiple refractive error risk labels to obtain multiple first candidate machine learning models and prediction data obtained by each first candidate machine learning model for each sample gene data. The second preset machine learning model is trained based on the multiple prediction data obtained by the first candidate machine learning models for each sample gene data, the corresponding sample environment data, and the refractive error risk label to obtain the second candidate machine learning model. The plurality of first candidate machine learning models and the second candidate machine learning models are determined as the plurality of candidate machine learning models.

4. The method according to claim 3, characterized in that, The second preset machine learning model is trained based on multiple predicted data obtained from multiple first candidate machine learning models for each sample gene data, the corresponding sample environment data, and the refractive error risk label to obtain a second candidate machine learning model, including: Using multiple predicted data and corresponding sample environment data as input features, and inputting the input features into the second preset machine learning model, risk prediction results are obtained; The parameters of the second preset machine learning model are updated based on the loss between the risk prediction result and the refractive error risk label.

5. The method according to claim 2, characterized in that, The step of obtaining the multiple training sample data corresponding to the multiple groups includes: Acquire multiple candidate gene data corresponding to the multiple populations, and screen the multiple candidate gene data based on preset quality assessment conditions to determine multiple sample gene data; Obtain the sample environment data and the refractive error risk label of the sample object corresponding to each of the sample gene data; The multiple training sample data are constructed based on multiple sample gene data, multiple corresponding sample environmental data, and multiple refractive error risk labels.

6. The method according to claim 5, characterized in that, After screening the multiple candidate gene data based on preset quality assessment conditions to determine multiple sample gene data, the method further includes: Genome-wide association analysis was performed on gene data from multiple samples to obtain the results of refractive error-related loci analysis. Genetic filtering was performed on the gene data of multiple samples based on the results of refractive error-related site analysis and a predetermined threshold.

7. The method according to claim 3, characterized in that, Each training sample data includes the sample gene data, the sample environment data, the sample object behavior data, and the refractive error risk label; The second preset machine learning model is trained based on multiple predicted data obtained from multiple first candidate machine learning models for each sample gene data, the corresponding sample environment data, and the refractive error risk label to obtain a second candidate machine learning model, including: The second preset machine learning model is trained based on the multiple prediction data obtained by the first candidate machine learning model for each sample gene data, the corresponding sample environment data, the sample object behavior data, and the refractive error risk label, to obtain the second candidate machine learning model.

8. The method according to claim 1, characterized in that, Before the target machine learning model is determined from multiple candidate machine learning models based on the target population of the target object, the process also includes: Receive a prediction request for refractive error prediction of a target object, and obtain the target group of the target object based on the prediction request; After inputting the data to be predicted into the target machine learning model to obtain the refractive error prediction result, the method further includes: The refractive error prediction result is sent to the party that initiated the prediction request.

9. A refractive error prediction device, characterized in that, include: The determining unit is used to determine the target machine learning model from multiple candidate machine learning models based on the target group of the target object. The multiple candidate machine learning models are pre-trained based on multiple training sample data corresponding to multiple groups. Each training sample data includes sample gene data, sample environmental data, and refractive error risk label. The acquisition unit is used to acquire the data to be predicted of the target object based on the target machine learning model; The prediction unit is used to input the data to be predicted into the target machine learning model to obtain the refractive error prediction result.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the refractive error prediction method according to any one of claims 1 to 8.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the refractive error prediction method according to any one of claims 1 to 8.

12. A computer program product comprising a computer program that is read and executed by a processor of a computer device, causing the computer device to perform the refractive error prediction method according to any one of claims 1 to 8.

Images