Children ophthalmic disease hospital guide method and system based on large model and block chain

Abstract

The invention provides a child ophthalmic disease hospital guide method and system based on a large model and a block chain, and is applied to the technical field of data processing. The method comprises the following steps: processing eye information of a target child and question and answer information of the target child to generate a target child ophthalmology label; preprocessing the training data set to generate a training data set with target feature data; processing a preset eye early warning model based on the training data set with the target feature data to generate a target eye early warning model; processing the target child ophthalmology label based on the target eye early warning model to generate a target eye recognition result; processing the identity information of the target child to generate consultation information of the target child; processing the target eye recognition result and the consultation information of the target child to generate block chain consultation information; and processing the block chain consultation information and the consultation information of the target child to generate eye report information of the target child.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for guiding children's ophthalmic diseases based on a large model and blockchain. Background Art

[0002] At present, in order to solve the problems of long consultation time and low patient satisfaction, hospitals have gradually added guidance systems, but the existing guidance systems for children's ophthalmological diseases still have the following problems:

[0003] 1. There is only a general medical and fundus image-based consultation guidance system, but no knowledge base on children's ophthalmological diseases and examinations. Therefore, the continuous dialogue capability is not flexible enough, lacks strong logical reasoning ability, and the output results are not accurate enough;

[0004] 2. Lack of integration with the field of data security, failure to include transparent management and immutability of data, and inability to ensure data integrity and security.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of this application is to provide a method and system for guiding children's ophthalmic diseases based on a large model and blockchain, which at least overcomes the problems of the prior art to a certain extent, and comprehensively obtains various types of information and related data models of target children, from eye, question and answer, identity, consultation demand information to training data sets and preset models, and the data is diverse and comprehensive. According to the eye and question and answer information, the children's ophthalmology label is obtained through calculation and analysis, and the training data set is preprocessed. After multiple links such as building a mapping table, grouping, extracting features, dividing data sets and training prediction, the target feature data training set is obtained, and the preset model is trained with it. After data processing, architecture parameter setting and training evaluation optimization, the target eye warning model is obtained. Then, the model is used to process the label to obtain the recognition result, and the identity information is processed to generate the consultation information, covering a variety of information generation steps, and then the results and the consultation information are de-noised, fused and encrypted into blockchain consultation information. Finally, the eye report information is generated according to this and the consultation demand, containing rich physiological characteristics, risk levels and other content, and the format is diverse and easy to view, providing all-round and accurate services for children's ophthalmology consultation and health management.

[0007] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present invention.

[0008] According to one aspect of the present application, a method for guiding children's ophthalmic diseases based on a large model and blockchain is provided, including: obtaining eye information of a target child, question and answer information of a target child, identity information of a target child, consultation information of a target child, a training data set, and a preset eye warning model, wherein the training data is used to characterize a question and answer corpus of other children, an eye change data set, and consultation results; processing the eye information of the target child and the question and answer information of the target child to generate an ophthalmic label for the target child; preprocessing the training data set to generate a training data set with target feature data, wherein the target The feature data is used to characterize the risk factors that affect eye features; the preset eye warning model is processed based on the training data set with the target feature data to generate a target eye warning model; the target child's ophthalmic label is processed based on the target eye warning model to generate a target eye recognition result; the identity information of the target child is processed to generate consulting information of the target child; the target eye recognition result and the consulting information of the target child are processed to generate blockchain consulting information; the blockchain consulting information and the consulting information of the target child are processed to generate eye report information of the target child.

[0009] Another aspect of the present application is a pediatric ophthalmic disease guidance device based on a large model and blockchain, characterized in that the device includes: an acquisition module for acquiring the target child's eye information, the target child's question and answer information, the target child's identity information, the target child's consultation information, a training data set and a preset eye warning model, wherein the training data is used to characterize other children's question and answer corpora, eye change data sets and consultation results; a processing module for processing the target child's eye information and the target child's question and answer information to generate a target child's ophthalmic label; preprocessing the training data set to generate a training data set with target feature data set, wherein the target feature data is used to characterize the risk factors affecting eye features; based on the training data set with the target feature data, the preset eye warning model is processed to generate a target eye warning model; based on the target eye warning model, the ophthalmic label of the target child is processed to generate a target eye recognition result; the identity information of the target child is processed to generate consulting information of the target child; the target eye recognition result and the consulting information of the target child are processed to generate blockchain consulting information; the blockchain consulting information and the consulting information of the target child are processed to generate eye report information of the target child.

[0010] According to another aspect of the present application, an electronic device is characterized in that it includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to implement the above-mentioned pediatric ophthalmic disease guidance method based on a big model and blockchain by executing the executable instructions.

[0011] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a second processor, the above-mentioned method for guiding children's ophthalmic diseases based on a large model and blockchain is implemented.

[0012] The application provides a method and system for guiding children's ophthalmic diseases based on a large model and blockchain. The server comprehensively obtains various information and related data models of the target children, from eye, question and answer, identity, consultation purpose information to training data sets and preset models. The data is diverse and comprehensive. Based on the eye and question and answer information, the children's ophthalmology label is obtained through calculation and analysis, and the training data set is preprocessed. After multiple links such as building a mapping table, grouping, extracting features, dividing the data set and training prediction, the target feature data training set is obtained, and the preset model is trained with it. After data processing, architecture parameter setting and training evaluation optimization, the target eye warning model is obtained. Then the model is used to process the label to obtain the recognition result, and the identity information is processed to generate the consultation information, covering a variety of information generation steps, and then the results and the consultation information are noise-reduced, fused and encrypted into blockchain consultation information. Finally, the eye report information is generated according to this and the consultation needs, containing rich physiological characteristics, risk levels and other content, and the format is diverse and easy to view, providing all-round and accurate services for children's ophthalmology guidance and health management.

[0013] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flowchart of a method for guiding children's ophthalmic diseases based on a big model and blockchain provided in one embodiment of the present application is shown;

[0015] Figure 2 A structural schematic diagram of a pediatric ophthalmic disease diagnosis guidance device based on a large model and blockchain provided in one embodiment of the present application is shown. DETAILED DESCRIPTION

[0016] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0017] Combine the following Figure 1To describe the pediatric ophthalmological disease guidance method based on a large model and blockchain according to an exemplary embodiment of the present application. It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited in this regard. On the contrary, the embodiments of the present application can be applied to any applicable scenario.

[0018] In one embodiment, the present application also proposes a method and system for guiding pediatric ophthalmic diseases based on a big model and blockchain. Figure 1 The following schematically shows a flow chart of a method for guiding children's ophthalmic diseases based on a large model and blockchain according to an embodiment of the present application. Figure 1 As shown, the method is applied to a server, comprising:

[0019] S101, obtaining the target child's eye information, the target child's question and answer information, the target child's identity information, the target child's consultation information, a training data set, and a preset eye warning model.

[0020] In one embodiment, the training data is used to characterize other children's question and answer corpora, eye change datasets, and consultation results.

[0021] The eye information of the target children includes the following: vision test data, for example, using a standard eye chart to test the target children's vision and record their naked eye vision, such as 0.8 for the right eye and 0.6 for the left eye. At the same time, it also includes data on corrected vision. If the child wears glasses, the vision after wearing appropriate glasses is recorded, which helps to evaluate the correction effect of the glasses and whether there are further changes in vision. For young children or children who cannot cooperate with the eye chart examination, some special vision testing methods will be used, such as visual evoked potential (VEP) examination or preferential gaze method. The vision-related data obtained by these methods can provide a more comprehensive understanding of the visual function status of the target children, especially for the early detection of some potential vision problems, such as amblyopia.

[0022] Intraocular pressure measurement results: Use an intraocular pressure meter to measure the intraocular pressure of the target child. The normal intraocular pressure range is generally between 10-21 mmHg. Assuming that the measured intraocular pressure value of the target child is 16 mmHg, this data is within the normal range, but it still needs to be combined with other eye information for comprehensive analysis. Changes in intraocular pressure are related to a variety of eye diseases, such as glaucoma, so regular monitoring of intraocular pressure is crucial for children's eye health. If the intraocular pressure value is high or low, further examinations are required, such as 24-hour intraocular pressure monitoring, gonioscopy, etc., to rule out or confirm the presence of eye diseases and determine whether appropriate treatment measures are needed.

[0023] The eye structure examination report obtains detailed information about the eye structure of the target child through eye ultrasound, optical coherence tomography (OCT) and other examination methods. For example, OCT examination can clearly show the structure of each layer of the retina and the condition of the optic disc, including retinal thickness, macular morphology, optic disc thickness, etc. Suppose the examination report shows that the retinal thickness of the target child is within the normal range, but there are slight pigment changes in the macular area. This finding suggests the need for further observation or related functional examinations, such as visual sensitivity tests in the macular area. Eye ultrasound examination can help observe the morphology of the eyeball, the clarity of the vitreous body, the position and state of the lens, and whether there are space-occupying lesions in the fundus. If small dot-like echoes are found in the vitreous body, it indicates the risk of vitreous hemorrhage and the fundus needs to be dilated for examination.

[0024] The target children's question and answer information includes the following: Daily eye habits inquiry, asking the target children how long they use electronic devices (such as mobile phones, tablets, TVs, etc.) every day. For example, the child answered that they use electronic devices for an average of 2-3 hours a day, mainly for watching cartoons and playing some simple games. Knowing this information helps to assess the potential impact of electronic device use on children's eye health, because long-term use of electronic devices leads to the occurrence or aggravation of eye problems such as myopia. Ask about children's reading habits, including reading distance, reading posture, etc. If the child indicates that he often reads books or uses electronic devices at a close distance (less than 33 cm) and has an incorrect posture when reading (such as lying or lying down), these are factors that increase the risk of myopia. The system can give corresponding eye suggestions based on this information, such as maintaining a correct reading posture and resting the eyes every once in a while.

[0025] Description of eye discomfort symptoms. When the target children are asked whether they have eye discomfort, the children describe that their eyes sometimes feel dry, especially after watching TV or reading for a long time. This symptom is related to excessive use of the eyes, insufficient tear secretion, or environmental factors (such as dry air). The system can further ask about the frequency and duration of the symptoms and whether they are accompanied by other symptoms (such as blurred vision, eye pain, etc.) in order to more accurately determine the cause. If the child mentions that the eyes occasionally have short-term blurred vision, especially when looking at nearby objects suddenly after looking at distant objects, this is a manifestation of problems with the eye's accommodation function, such as accommodation spasm. Based on these question and answer information, the system can recommend further accommodation function tests, such as accommodation amplitude, accommodation flexibility, etc., to clarify the consultation and take appropriate treatment measures, such as visual training.

[0026] The target child's identity information includes the following: basic personal information, the target child's name is Xiao Ming, the gender is male, and the date of birth is September 11, 2016. This basic information is very important for establishing a child's personal profile. At the same time, in the subsequent data analysis and report generation, it can be classified and analyzed according to factors such as age and gender, such as studying the characteristics and trends of eye diseases in children of different age groups or genders. The contact information and home address information of the child's parents are also recorded, which helps to understand the living environment of the child. Certain environmental factors (such as whether there are pollution sources nearby) have an impact on eye health. Although this is not direct eye health information, it can be used as one of the reference factors for comprehensive evaluation. School and learning situation, understand the name of Xiao Ming's school and grade information. The school's learning environment (such as classroom lighting, whether the height of desks and chairs is appropriate, and the position of sitting in class) has an impact on children's eye health. If Xiao Ming's school classroom is not well lit, it will increase his risk of eye fatigue during learning. Based on this situation, the system can recommend the school or parents to take corresponding improvement measures, such as adding lighting equipment. Ask Xiao Ming about his study schedule, including the class time and homework amount every day. Heavy study tasks and long-term close-up use of the eyes lead to the occurrence or development of myopia. For example, if Xiao Ming spends more than 2 hours a day doing homework and attends various extracurricular tutoring classes after school, the long-term close-up use of the eyes is more serious. The system can remind parents to pay attention to their children's eye use and arrange study and rest time reasonably, such as doing outdoor activities every once in a while, to protect their children's eye health.

[0027] The target children's consultation demand information includes the following: The main purpose of parents bringing Xiao Ming to the consultation is to assist the doctor in consulting Xiao Ming's recent vision loss through system analysis and suggestions. The parents have taken Xiao Ming to some other places for some preliminary examinations, but the results are unclear. They hope to use the system to obtain more comprehensive analysis and consultation directions. For example, the system can provide some disease consultation hypotheses such as refractive error, congenital cataract, optic neuritis, etc. based on Xiao Ming's eye information, question and answer information, and comparison with other similar cases, and recommend further examination items such as mydriasis optometry, anterior segment photography, fundus examination, etc. to clarify the consultation. Eye health guidance, another purpose of consultation is that parents are concerned about Xiao Ming's eye health, and hope that the system can provide personalized eye advice based on Xiao Ming's specific situation. For example, although Xiao Ming currently has no obvious eye disease, his parents want to know how to prevent the occurrence of myopia, especially considering that Xiao Ming has a heavy learning task and likes to use electronic devices. The system can formulate an eye use plan suitable for Xiao Ming based on his eye use habits, age and other factors, such as daily outdoor activity time recommendations, reasonable time limits for using electronic devices, correct reading and writing posture guidance, etc., to help Xiao Ming maintain good eye health.

[0028] The training data set contains many other children's question and answer records in similar eye consultation scenarios. For example, some children will say "I always see double, especially at night", and the doctor will then ask "How long has this situation lasted?" "Has your vision declined significantly?" "Have you suffered any trauma?" and other questions, as well as subsequent consultation results and treatment recommendations are all recorded in the question and answer corpus. These rich question and answer records can provide a reference for the situation analysis of the target children. By comparing the symptom descriptions and consultation results of different children, it is helpful to discover potential disease patterns and laws. The language expressions in the question and answer corpus are diverse, covering the language characteristics of children of different ages. For young children, the language expression is relatively simple and direct, such as "eye pain"; while older children use more complex words to describe symptoms, such as "I feel a foreign body sensation in my eyes, and my vision seems a little blurry." By analyzing these different language expressions, we can better understand children's eye problems and improve the system's ability to recognize and process various symptom descriptions.

[0029] The eye change dataset records the changes in eye parameters of other children at different time points. For example, for a group of children with myopia, the dataset contains the change data of their visual acuity, intraocular pressure, axial length and other parameters from the first discovery of myopia to the subsequent regular review. By analyzing these data, we can understand the development trend of myopia, such as the relationship between the growth rate of axial length and the deepening of myopia, so as to provide a reference for the prevention and control of myopia in target children. In addition, the eye change dataset also includes the eye changes of some special cases, such as the changes in eye structure and function of children with congenital eye diseases during their growth and development. These data are of great significance for studying the natural course of children's eye diseases, predicting the direction of disease development, and evaluating the treatment effect. For example, for children with congenital cataracts, recording the degree of lens opacity before and after surgery, the effect of postoperative visual rehabilitation training and other eye change data can provide experience for the treatment of other similar cases.

[0030] The consultation result records the final conclusions of other children after a series of examinations and consultations. For example, after a detailed examination, a child is consulted for strabismus. The record will detail the type of strabismus (such as intermittent exotropia, partial accommodative esotropia, etc.), degree (such as the degree of strabismus when looking at near and far, the degree of strabismus with and without glasses, whether there is stereoscopic vision when looking at near and far), and recommended treatment plans (such as glasses correction, surgical treatment, visual training, etc.). These consultation results can be used as reference cases for consultation and treatment of target children. Especially when the target children have similar symptoms, the system can quickly provide consultation and treatment suggestions based on previous consultation results to improve consultation efficiency. At the same time, the consultation results also contain some information about the prognosis of the disease, such as the expected recovery of vision after treatment of a certain eye disease, whether it is easy to relapse, etc. This is very important for parents to understand the development of their children's disease and the expected treatment effect, and it also helps doctors to consider various factors when formulating a treatment plan and choose the most appropriate treatment plan.

[0031] The preset eye warning model uses a convolutional neural network (CNN) architecture. CNN has advantages in processing image data (such as eye structure images, fundus images, etc.). It can automatically extract features in images, such as retinal vascular morphology, optic nerve head morphology, etc. These features are of great significance for consultation and early warning of eye diseases. For example, in the early consultation of glaucoma, the CNN model can detect potential signs of lesions in advance by analyzing the degree of depression of the optic nerve head in the fundus image, the thickness of the retinal nerve fiber layer, and other features, thereby issuing a warning signal.

[0032] The model also includes a recurrent neural network (RNN) or long short-term memory network (LSTM) part to process time series data, such as the target child's vision change history, eye axis change, fundus change, etc. RNN and LSTM can capture the time dependency in the data, better understand the trend of eye parameters over time, and thus more accurately predict the risk of eye diseases. For example, if a child's optic disc thickness shows a gradual downward trend over a period of time, the model can predict whether the optic disc thickness will continue to decline and reach a dangerous level based on historical data, and remind doctors and parents to take corresponding measures in advance. The model is trained with a large amount of eye-related data. The training data includes multi-source data such as eye information, question and answer information, and examination results of children of different ages, genders, and eye health conditions. During the training process, the model continuously adjusts internal parameters to minimize the error between the predicted results and the actual consultation results. For example, for a child who is consulted for myopia, the model makes predictions based on the input eye information (such as axial length, corneal curvature, etc.) and question and answer information (such as daily eye habits, etc.). If the prediction result is inconsistent with the actual consultation, the model will adjust the parameters according to the error back propagation algorithm to improve the accuracy of the prediction.

[0033] In order to improve the performance of the model, a variety of optimization techniques were used, such as data enhancement technology, which increased the diversity of training data and prevented the model from overfitting by rotating, flipping, and scaling the original data. Regularization techniques, such as L1 and L2 regularization, were also used to limit the size of model parameters and avoid the model being too complex and resulting in a decrease in generalization ability. At the same time, in terms of model evaluation, methods such as cross-validation were used to divide the training data into multiple subsets, and one of the subsets was used as the validation set and the other subsets were used as the training set in turn to evaluate the performance of the model on different subsets to ensure that the model has good stability and generalization ability.

[0034] S102, processing the eye information of the target child and the question-and-answer information of the target child to generate an ophthalmological label for the target child.

[0035] In one implementation, the target child's question and answer information is processed to generate the target child's physiological characteristic information and the target child's basic information; the target child's eye information is processed to generate the target child's eye characteristic information. Suppose we have a target child named Xiao Ming. First, we process Xiao Ming's question and answer information. For example, by asking Xiao Ming's parents, we know that Xiao Ming's date of birth is September 11, 2016 (age information), his gender is male (gender information), he watches the electronic screen for an average of 3 hours a day (eye time information), and he mainly watches the electronic screen in the living room at home (eye location information), with no family history, allergy history, systemic medical history, and surgical history. This information constitutes Xiao Ming's physiological characteristic information and basic information. Then, Xiao Ming is examined in the eyes, and it is found that Xiao Ming's naked eye vision is 0.8 for the right eye and 0.6 for the left eye (eye information). Intraocular pressure measurement shows that the intraocular pressure of the right eye is 15mmHg and the intraocular pressure of the left eye is 14mmHg. By processing these eye information, Xiao Ming's eye feature information is generated.

[0036] The physiological characteristic information, basic information and eye characteristic information of the target child are processed, and the processed data are embedded into the vector database and assigned a unique identifier to generate the child eye data matching value. The child eye data matching value is processed, and the target child ophthalmology label is generated according to the marked eye muscle disease, refractive disease, lens disease, fundus disease, glaucoma, ocular surface disease and ocular trauma and the disease classification under different pediatric ophthalmology departments. Xiao Ming's physiological characteristic information and eye characteristic information are processed comprehensively. For example, suppose we have an algorithm that considers the weight of the influence of age, eye use time and eye use place on vision and intraocular pressure. If the age is young, the eye use time is long and the light in the eye use place is poor, it will have a greater negative impact on eye health, which will in turn affect the eye data matching value. Finally, the child eye data matching value is processed to generate Xiao Ming's ophthalmology label. For example, if Xiao Ming's eye data matching value shows that he has a higher risk of myopia, then his ophthalmology label is "child with high risk of myopia".

[0037] The method includes a calculation formula for obtaining a matching value of the child's eye data, and the calculation formula is:

[0038]

[0039] Among them, a and b are vectors related to the eye data of the target child. a represents the eye feature vector of a target child, which contains the child’s eye data information such as vision level, intraocular pressure value, axial length, etc. Similarly, b is the eye feature vector of another target child or the same target child at different time points.

[0040] Assume that a is a vector representing the eye physiological characteristics of Xiao Ming, and b is a standard vector representing the eye physiological characteristics of normal children of the same age.

[0041] For example, a=(a1, a2, a3, a4), where a1 represents the value corresponding to Xiao Ming's age (the value corresponding to 8 years old), a2 represents the value corresponding to the length of time Xiao Ming uses his eyes (the value corresponding to 3 hours), a3 represents the value corresponding to the lighting conditions of the place where Xiao Ming uses his eyes (assuming the value corresponding to the lighting conditions in the living room), and a4 represents the value corresponding to Xiao Ming's vision (such as the combined value corresponding to 0.8 for the right eye and 0.6 for the left eye).

[0042] b=(b1, b2, b3, b4) is the standard vector of normal children of the same age group. For example, b1 is the value corresponding to the normal age of this age group, b2 is the value corresponding to the normal length of time of eye use (assuming the value corresponding to 1-2 hours), b3 is the value corresponding to the lighting conditions in the normal eye use place (assuming the value corresponding to good lighting), and b4 is the value corresponding to normal vision (such as the value corresponding to 1.0).

[0043] First, calculate ab, which is a1×b1+a2×b2+a3×b3+a4×b4. This step is to calculate the sum of the products of the corresponding elements of the two vectors.

[0044] Then calculate |a| and |b|,

[0045] Finally, The result is the similarity between Xiao Ming's eye data and the eye data of normal children of the same age. If the result is close to 1, it means that Xiao Ming's eye data is similar to that of normal children; if the result is close to 0, it means that the difference is large and there are eye health problems.

[0046] S103, preprocessing the training data set to generate a training data set with target feature data.

[0047] In one implementation, a preset mapping table is obtained, wherein the preset mapping table is used to characterize the historical eye characteristics and related physiological characteristics and basic information of children in the target age group. Suppose we want to build a preset mapping table for children aged 3-10 years old. First, collect a large amount of historical eye examination data of children aged 3-10 years old, which comes from ophthalmology clinic records in different regions and hospitals. For example, eye examination data of 1,000 children are collected, including optometry results (such as visual acuity, myopia, amblyopia, astigmatism, etc.), intraocular pressure measurement values, eye structure examination (such as corneal curvature, axial length, lens opacity, etc.), fundus examination results (such as retinal vascular morphology, optic nerve head morphology, etc.) and related medical history (such as whether there is a family hereditary eye disease, whether it is premature, whether there is a history of eye trauma, etc.). These data are grouped according to age, such as 3-4 years old as a group, 5-6 years old as a group, 7-8 years old as a group, and 9-10 years old as a group. For each age group, calculate the statistical indicators such as the mean, standard deviation, and distribution range of each eye feature. For example, for the 3-4 years old age group, the average visual acuity is about 0.6, the average intraocular pressure is about 13 mmHg, the average axial length is about 21 mm, etc. These statistical indicators constitute the basic content of the preset mapping table, which can characterize the typical eye feature information of children in each target age group.

[0048] When a new child comes for ophthalmological consultation, the system first queries the preset mapping table based on the child's age. For example, when a 5-year-old child comes for consultation, the system queries the preset mapping table to obtain the average vision, intraocular pressure and other eye characteristics of children in the 5-6 age group. The child's actual eye examination data is then compared with the data in the preset mapping table to preliminarily determine whether the child's eye characteristics are within the normal range. If the child's vision is significantly lower than the average vision of the 5-6 age group, the system will regard it as a potential abnormality and focus on it in subsequent consultations.

[0049] The training data set is grouped based on the preset mapping table to generate a grouped training data set, wherein the grouped training data set includes eye image information and data information of children of different age groups. The training data set also comes from children's eye-related data collected from multiple channels, including the hospital's ophthalmology medical record database, experimental data from ophthalmology research institutions, etc. Assume that the training data set contains eye data of 5,000 children, which covers children of different ages, genders, and regions. The data content includes detailed eye examination images (such as fundus images, eye ultrasound images, etc.), examination reports (including values ​​such as visual acuity, intraocular pressure, and axial length), and related clinical consultation results (such as whether the child suffers from myopia, strabismus, amblyopia, glaucoma, cataracts, and other eye diseases). For example, for a child with myopia, the data set will include his vision examination records at different times, retinal changes shown in fundus images, data on the growth of axial length over time, and consultation conclusions for diagnosis of myopia. The training data set is grouped using a preset mapping table. According to the age grouping range set in the preset mapping table, the data of 5,000 children in the training data set are classified into corresponding age groups. For example, the data of children aged 3-4 are grouped into one group, and those aged 5-6 are grouped into one group, etc. For each age group, the eye image information is further screened. For example, in the 3-4 year old age group, clear and representative fundus images, eye ultrasound images, etc. are selected. These images will serve as an important data source for subsequent feature extraction and model training. Finally, a training data set with grouped eye image information of children of different age groups is obtained. Each group is representative of the eye features of children of that age group, laying the foundation for subsequent precise analysis and model training.

[0050] Different features are extracted from the grouped training data sets to generate the original feature library. For the grouped training data sets, a variety of advanced technologies are used for feature extraction. Taking fundus images as an example, using convolutional neural network (CNN) technology, the convolution layer of CNN can automatically extract local features in the image, such as the morphological characteristics of retinal blood vessels (blood vessel thickness, curvature, branching, etc.), the shape and color characteristics of the optic nerve head, etc. For example, through multiple convolution operations of the convolution layer, subtle changes in retinal blood vessels can be highlighted, which are related to certain eye diseases (such as retinopathy of prematurity, familial exudative chorioretinopathy, etc.). At the same time, the feature extraction algorithm in deep learning is used to extract features from the numerical data in the eye examination report (such as visual acuity, intraocular pressure, axial length, optic disc thickness, thickness of each layer of the retina in the macular area, etc.). For example, the visual acuity data is correlated with age to extract the characteristic patterns of visual acuity changes in children of different ages; the intraocular pressure data is dynamically analyzed to extract the characteristics of intraocular pressure fluctuations, which is of great significance for the consultation of diseases such as congenital glaucoma and congenital cataracts.

[0051] Through the above-mentioned feature extraction method, the images and numerical data in each grouped training data set are converted into a series of representative feature vectors. For example, for a set of fundus images, the feature vectors corresponding to each image are obtained after feature extraction. These feature vectors contain the feature information of key structures such as retinal blood vessels and optic discs in the image. The feature vectors of all groups are summarized to form the original feature library. Each record in the original feature library represents the feature information of a child's eye data. This feature information covers many aspects such as eye structure, function, and age-related changes, providing a rich data foundation for subsequent model training and analysis.

[0052] The original feature library is divided into various feature data sets to generate training sets and validation sets. The original feature library is usually divided into training sets and validation sets by random division. The common division ratio is 70%-80% of the data as the training set and 20%-30% of the data as the validation set. For example, there are 10,000 feature records in the original feature library. According to the ratio of 70%:30%, 7,000 records are divided into the training set and 3,000 records are divided into the validation set. In the division process, it is necessary to ensure that the training set and the validation set have a certain representativeness and balance in terms of different age groups, different genders, and different disease types. The training set should not be overly concentrated on the children's data of a certain age group or a certain disease type, otherwise it will lead to deviations in model training. For example, when dividing, it is necessary to ensure that the children's data of each age group have a reasonable proportion in the training set and the validation set, and at the same time, the children's data with different eye diseases (such as myopia, hyperopia, strabismus, cataracts, optic neuritis, etc.) are evenly distributed according to the incidence rate.

[0053] The training set is mainly used to train the model. The model continuously adjusts its own parameters by learning the relationship between the feature data in the training set and the corresponding consultation results to improve the accuracy of consultation on eye diseases. For example, during the training process, the model learns the differences between the axial length characteristics and retinal imaging characteristics of myopic children and normal children, so that it can predict myopia based on new children's eye data. The validation set is used to evaluate and verify the performance of the model during the model training process. At each stage of model training (such as after each round of training), the data of the validation set is used to input the model to obtain the model's prediction results, and then the prediction results are compared with the actual consultation results in the validation set to calculate evaluation indicators such as accuracy, recall rate, and F1 value. If the performance indicators of the model on the validation set do not improve or even decrease, it can be judged that the model has overfitting or other problems, and the model parameters or training methods need to be adjusted.

[0054] Based on the classifier, each validation set divided by the original feature library is predicted to generate a prediction result. You can choose classifiers such as support vector machine (SVM), random forest (RandomForest), neural network (NeuralNetwork) for prediction. Taking the random forest classifier as an example, it is an integrated learning model composed of multiple decision trees. When building a random forest classifier, first determine the number of decision trees, for example, set it to 100 decision trees. When each decision tree is constructed, a part of the features (such as randomly selecting 60% of the features) is randomly selected from the feature data of the original feature library for node splitting, which can increase the diversity of the model and avoid overfitting. For the neural network classifier, a multi-layer perceptron (MLP) structure can be constructed. For example, the number of nodes in the input layer is determined according to the feature dimension of the original feature library. Assuming that the original feature has 50 dimensions, the input layer has 50 nodes; the hidden layer is set to 2-3 layers, and the number of nodes in each layer is adjusted according to experience and experimental results, such as 30 nodes in the first hidden layer and 20 nodes in the second hidden layer; the number of nodes in the output layer is determined according to the number of eye disease types to be predicted. For example, to predict myopia, amblyopia, and strabismus, the output layer has 3 nodes, representing the prediction probabilities of the three diseases respectively.

[0055] Input the feature data in the validation set into the selected classifier for prediction. For example, for a validation set containing 3,000 feature records, input the feature vector of each record into 100 decision trees of the random forest classifier in turn. Each decision tree makes a judgment based on the feature value in the feature vector, and each decision tree will output a prediction result (such as whether the child has a certain eye disease). Then, according to the voting mechanism of the random forest, the prediction results of the 100 decision trees are combined to obtain the final prediction result. For example, if 80 decision trees predict that the child has myopia and 20 decision trees predict that the child does not have myopia, then the final prediction result is that the child has myopia. For the neural network classifier, the feature vector is calculated through the input layer and the hidden layer, and the predicted probability of each eye disease is obtained in the output layer, and the disease with the highest probability is selected as the prediction result. Through the prediction of the validation set, the prediction result of each sample can be obtained, and these prediction results will be used for subsequent comparison with the actual consultation results to evaluate the performance of the classifier.

[0056] Use the preset algorithm to divide each training set in the original feature library for training and generate the prediction results of the validation set. You can choose the gradient descent algorithm (GradientDescent) and its variants (such as stochastic gradient descent, Adagrad, Adadelta, Adam, etc.) for training. Taking the Adam algorithm as an example, it combines the advantages of the momentum method and the adaptive learning rate. During the training process, the Adam algorithm calculates the gradient of the loss function (such as the cross entropy loss function, which is used to measure the difference between the model prediction result and the actual consultation result) to the model parameters based on the training set data, and then updates the model parameters according to the gradient and momentum terms. The momentum term can accelerate the convergence speed of the model, so that the model can move towards the optimal solution faster during the training process; the adaptive learning rate can automatically adjust the learning rate according to the update of the parameters, using a larger learning rate to accelerate convergence in the early stage of training, and using a smaller learning rate in the later stage to avoid overshoot, thereby improving the stability and accuracy of the model. For deep learning models (such as neural networks), the back propagation algorithm (BackPropagation) can also be used to calculate the gradient of the loss function to each layer of parameters, and then update the parameters according to the preset algorithm (such as the Adam algorithm). The back-propagation algorithm propagates the error signal layer by layer through reverse calculation from the output layer to the input layer, so that the gradient of the parameters of each layer can be accurately calculated, providing an effective parameter update basis for model training.

[0057] During the training process, the data of the training set are input into the model in batches. For example, 7000 training set data are divided into 70 batches, each batch has 100 data. For each batch of data, the model performs forward propagation to calculate the prediction results, and then calculates the loss function value between the prediction results and the actual consultation results. Then, the gradient of the loss function to the model parameters is calculated by the back propagation algorithm, and finally the model parameters are updated according to the gradient using the preset algorithm (such as the Adam algorithm). After multiple iterations of training (such as 1000 iterations), the model gradually learns the relationship between the feature data in the training set and the consultation results, and improves the prediction ability of eye diseases. During the training process, the validation set is regularly used to evaluate the model performance and generate validation set class prediction results. For example, every 100 training times, the validation set data is used to make a prediction, and the validation set class prediction results are obtained (similar to the process of predicting the validation set based on the classifier), and the evaluation indicators (such as accuracy, recall, etc.) are calculated. By observing the changes in the validation set class prediction results and evaluation indicators, the progress of the model training can be understood, and whether the model has problems such as overfitting or underfitting can be judged. If the accuracy of the validation set no longer improves or even decreases, it means that the model is overfitting and corresponding measures (such as adding regularization terms, adjusting model complexity, etc.) need to be taken to improve the model.

[0058] The prediction results and the validation set prediction results are processed to generate target feature data, which are used to characterize the feature information reflecting the eye abnormalities of children. The prediction results of the validation set based on the classifier and the prediction results of the validation set generated during the training process are compared and analyzed. For example, the prediction accuracy, recall rate and other indicators of each eye disease in the two prediction results are compared. If it is found that the recall rate of myopia in the prediction results based on the classifier is high, but the accuracy rate is low, while the accuracy rate of myopia in the validation set prediction results during the training process is high, but the recall rate is low, it is necessary to further analyze the reasons. Through comparative analysis, the features that have an important impact on the consultation of children's eye abnormalities are screened out. For example, it is found that the features such as axial length, corneal curvature, and retinal vascular morphology are closely related to the prediction of eye diseases in both prediction results. These features are the key features that affect children's eye abnormalities. At the same time, for some features that are unstable in the prediction results or contribute less to the prediction (such as some less relevant medical history information or less significant eye structure features), they can be considered to be excluded to simplify the model and improve the accuracy and efficiency of the model.

[0059] After feature screening, the final target feature data is determined. For example, features such as axial length, corneal curvature, and retinal vascular morphology are finally determined as target feature data. These feature data can more effectively characterize the feature information that affects children's eye abnormalities. These target feature data are sorted and standardized. For example, the axial length is uniformly expressed in certain units (such as millimeters), and the retinal vascular morphology can be expressed by quantitative indicators (such as vascular tortuosity index, vascular branch density, etc.). The generated target feature data can be stored in a specific data structure, such as a database table or a data file, so that it can be called and analyzed in the children's ophthalmology disease consultation system later. For example, when consulting new children's eye data, the system can directly extract the target feature data, use the model trained based on these features to provide fast and accurate consultation, improve the efficiency and accuracy of the consultation, and provide strong support for the early detection and treatment of children's ophthalmology diseases. At the same time, these target feature data can also be used for further research and analysis, such as studying the changing laws of eye abnormalities in children of different ages, exploring the pathogenesis of eye diseases, etc., to provide a data basis and theoretical basis for the development of children's ophthalmology.

[0060] S104, processing the preset eye warning model based on the training data set with target feature data to generate a target eye warning model.

[0061] In one embodiment, after a series of previous steps, we have determined the target feature data, such as axial length, corneal curvature, retinal vascular morphology (expressed as quantitative indicators such as vascular curvature index, vascular branching density), visual field (expressed as quantitative indicators such as average visual acuity, average defect and square root of loss variance), etc. Suppose that we have screened out data records related to these target features from a large amount of children's eye data collected previously. For axial length, the data set contains accurate axial length measurements of children of different ages to the millimeter level; corneal curvature is recorded as a specific curvature value; retinal vascular morphology data is quantified into corresponding index and density values ​​through professional image analysis software. These target feature data are integrated with the corresponding basic information of the children (such as age, gender, family history, etc.) and clinical consultation results (whether there is an eye disease and the specific type of disease) to form a training data set with target feature data. For example, a data entry is: [age 6 years old, male, no family history of eye disease, axial length 22.5mm, corneal curvature 43D, retinal vascular tortuosity index 0.8, vascular branching density 12, average visual acuity 26.2dB, average defect 0dB, square root of loss variance 1.5dB, consultation result is normal vision]. Such a data set covers a wealth of information and provides a solid data foundation for training accurate eye warning models.

[0062] Preprocess the training data set with target feature data. First, check the completeness and accuracy of the data and handle missing values. For a small amount of missing axial length data, the mean filling method can be used, that is, the missing values ​​are filled according to the average value of the axial length of children of the same age; for data entries with more missing values ​​or missing key information, consider deleting them from the data set. Then perform data standardization. For example, for numerical features such as axial length and corneal curvature, they can be normalized to a specific interval, such as [0,1] or [-1,1]. This helps to improve the efficiency and accuracy of model training and avoid the model's excessive attention or neglect of certain features due to large differences in the numerical ranges of different features.

[0063] Select a suitable preset eye warning model architecture, such as a combination of convolutional neural network (CNN) and recurrent neural network (RNN). The CNN part is mainly used to process feature data related to eye images, such as the extraction of features such as retinal vascular morphology in fundus images. Its convolution layer can effectively capture local features in the image, and the pooling layer can reduce the data dimension and reduce the amount of calculation while retaining important feature information. The RNN part is suitable for processing data related to time series, such as the change data of children's vision at different stages and the change trend of axial length with age. This combined architecture can make full use of the characteristics of different types of data and improve the model's ability to warn of eye disease risks. Set initial parameters for the preset eye warning model. For example, in the CNN part, determine the size and number of convolution kernels and initialize them to random values. Assume that the convolution kernel size is set to 3x3 and the number is 32, which is used to extract preliminary features of fundus images. In the RNN part, determine the number of hidden layers and nodes, such as setting 2 hidden layers with 128 nodes per layer to process the complex relationship of time series data. At the same time, set training-related parameters such as learning rate. The initial learning rate can be set to 0.001. As the training progresses, the learning rate is gradually reduced according to the learning rate adjustment strategy (such as exponential decay) to ensure that the model can converge more stably in the later stage of training.

[0064] The training data set with target feature data is divided into a training set and a validation set according to a certain ratio, for example, 80% of the data is used as a training set and 20% of the data is used as a validation set. During the training process, the samples in the training set are input into the preset eye warning model in sequence. For image feature data (such as fundus image-related features), high-level feature representations are extracted through convolution, pooling and other operations in the CNN part; for numerical feature data (such as axial length, corneal curvature, etc.) and categorical feature data (such as gender, family history, etc.), the corresponding encoding conversion is performed and then fused with the features extracted by CNN, and then input into the RNN part for time series analysis and model learning. The model makes predictions based on the input data and calculates the loss function value between the prediction results and the actual consultation results. The cross entropy loss function is used to measure the error of classification tasks (such as predicting whether there is an eye disease and the specific type of disease). For regression tasks (such as predicting the trend of vision changes, the growth trend of axial length, etc.), the mean square error loss function can be used. Then, the gradient of the loss function to the model parameters is calculated through the back-propagation algorithm, and the model parameters are updated according to the preset optimization algorithm (such as the Adam optimization algorithm), so that the model gradually learns the characteristic patterns and laws in the data and improves the accuracy of the prediction.

[0065] Perform multiple iterations of training, for example, 1000 epochs. After each epoch, use the validation set to evaluate the model. Calculate the model's accuracy, recall, F1 value and other evaluation indicators on the validation set to measure the model's performance. If it is found that the model's performance indicators on the validation set no longer improve or even decrease, it means that the model is overfitting. At this point, some measures can be taken to prevent overfitting, such as adding regularization terms (such as L1 or L2 regularization), stopping training early (stopping training when the validation set performance does not improve for several consecutive epochs), and using data enhancement techniques (such as rotating and flipping fundus images to increase data diversity) to ensure that the model has good generalization ability and can accurately warn of new children's eye data.

[0066] When the performance indicators of the model on the training set and the validation set meet the expected requirements or after multiple iterations of training, the model training is considered complete. For example, the prediction accuracy of the model on the validation set for common eye diseases (such as myopia, amblyopia, strabismus, etc.) reaches more than 90%, the recall rate reaches more than 85%, the F1 value reaches more than 88%, and the performance indicators remain stable for several consecutive epochs. At this time, the obtained model is the target eye warning model. The target eye warning model can be applied to actual clinical consultation and screening work in pediatric ophthalmology. For example, in routine physical examinations at school, by collecting children's eye images (such as fundus images) and basic information (such as age, gender, family history, etc.), and inputting them into the target eye warning model, the model can quickly predict whether the child has an eye disease risk and give corresponding warning prompts. This helps to detect potential eye problems early, take timely intervention measures, and prevent further development of the disease. Compared with traditional consultation methods, this model can process a large amount of complex data and quickly give relatively accurate warning results, which improves the efficiency and accuracy of children's eye disease consultation and provides more effective protection for children's eye health. At the same time, as more data accumulates, the target eye warning model can be continuously optimized and updated to further improve its performance and reliability.

[0067] S105, processing the target child ophthalmology label based on the target eye warning model to generate a target eye recognition result.

[0068] In one implementation, the target child ophthalmology label is processed based on the target eye warning model to generate a target approximation, and the target approximation is processed to generate a target eye recognition result.

[0069] The target eye warning model includes a calculation formula for obtaining the target approximation value, which is:

[0070] Where N is the number of samples, yi is the true category label of the i-th sample, θ j is the angle between the feature vector of the i-th sample and the j-th category weight vector, s is a learnable scale factor used to control the discrimination between different categories, and m is the additive margin, a constant greater than 0, used to add an additional margin to the angle of the true category.

[0071] Suppose we want to analyze the eye data of a new child. First, the eye data of this child is input as a sample into the target eye warning model. The model calculates the corresponding angle based on its internal category weight vector and the eye feature vector of this child. Then, the target approximation is calculated using the above formula. Based on the obtained target approximation, further processing is performed to obtain the target eye recognition result. For example, if the target approximation shows that the eye data of this child is very close to the myopia category, then the target eye recognition result is that this child has a higher risk of myopia, and then corresponding preventive or treatment measures can be taken, such as recommending regular vision checks and controlling eye use time.

[0072] In another implementation, assume that we have a small data set with a sample size N=5, that is, there are eye data samples of 5 children.

[0073] The true category label y i , assuming that the strabismus category labels of the five children in the eye muscle disease department are:

[0074] y1=1 (representing intermittent exotropia), y2=2 (representing partially accommodative esotropia), y3=1 (representing vertical dissociative strabismus), y4=3 (representing constant exotropia), y5=2 (representing constant esotropia).

[0075] Angle θ ij , assuming that we calculate the angle θ between each sample and the weight vector of each category ij (For simplicity, we assume that there are only three categories: intermittent exotropia, partially accommodative esotropia, and vertical dissociative strabismus):

[0076] For the first sample (i=1):

[0077] θ 11 =0.3 (the angle with the intermittent exotropia category weight vector), θ 12 =0.6 (the angle with the weight vector of the partial accommodative esotropia category), θ 13 =0.9 (the angle with the vertical dissociative strabismus category weight vector).

[0078] For the second sample (i=2):

[0079] θ21 =0.7;θ 22 =0.4;θ 23 =0.8;

[0080] For the third sample (i=3):

[0081] θ 31 =0.2;θ 32 =0.7;θ 33 =0.8;

[0082] For the fourth sample (i=4):

[0083] θ 41 =0.8;θ 42 =0.9;θ 43 =0.3;

[0084] For the fifth sample (i=5):

[0085] θ 51 =0.7;θ 52 =0.3;θ 53 =0.8;

[0086] Scaling factor s and additive margin m, assuming that the scaling factor s = 2 and the additive margin m = 0.1.

[0087] Calculate the molecular part

[0088] For the first sample (i=1), y1=1 (intermittent exotropia):

[0089]

[0090] For the second sample (i=2), y2=2 (partial accommodative esotropia):

[0091]

[0092] And so on, the molecular parts of all samples are calculated.

[0093] Calculate the denominator

[0094] For the first sample (i=1): the numerator has been calculated to be 5.53,

[0095] in the denominator Since y1 = 1 (intermittent exotropia), we need to calculate the cases of j = 2 (partial accommodative esotropia) and j = 3 (vertical dissociative strabismus):

[0096] For j = 2:

[0097] For j=3:

[0098] Denominator = 5.53 + (5.21 + 3.47) = 14.21.

[0099] For the second sample (i=2):

[0100] The numerator is 5.16, and the denominator is Since y2 = 2 (partial accommodative esotropia), we need to calculate the cases of j = 1 (intermittent exotropia) and j = 3 (vertical dissociative strabismus):

[0101] For j = 1:

[0102] For j=3:

[0103] Denominator = 5.16 + (4.62 + 4.03) = 13.81.

[0104] And so on, calculate the denominator part of all samples.

[0105] Calculate the logarithmic part

[0106] For the first sample:

[0107] For the second sample:

[0108] And so on, the logarithmic part of all samples is calculated.

[0109] Calculating the loss function

[0110] Substitute the values ​​of the logarithmic part calculated above into:

[0111] (calculate after adding up the logarithmic parts of all five samples). Through this example, we can see how to calculate the loss function related to the target approximation based on the given formula and assumed data, and then use it to optimize the model parameters in model training to improve the accuracy of the target eye warning model.

[0112] S106, processing the identity information of the target child to generate consulting information of the target child.

[0113] In one embodiment, the identity information of the target child is processed to generate the identity type information of the target child, wherein the identity type information of the target child is used to characterize the consultation duration, number of consultations and historical consultation records of the target child. Assume that the target child Xiao Ming, his identity information is as follows: his name is Xiao Ming, he is 8 years old, male, his home address is located in a street in a certain district of a certain city, and his guardian's contact information is a mobile phone number 138xxxx5678. Xiao Ming first came for consultation on May 1, 2023, and this is his third time to come for consultation. The system records the duration of each consultation, the first consultation duration is 30 minutes, the second consultation duration is 45 minutes, and the current consultation duration is 60 minutes. Based on this information, the system calculates the average consultation duration as (30+45+60) / 3=45 minutes. The number of consultations is 3 times. Thus, Xiao Ming's identity type information is generated, for example, it can be expressed as [average consultation duration 45 minutes, number of consultations 3 times]. This identity type information will be used for subsequent analysis of the development trend of Xiao Ming's eye problems and a comprehensive judgment of his eye use. If Xiao Ming's average consultation time is long and the number of consultations is high, it suggests that his eye problems are more complicated or his parents are very concerned about his eye health. Based on this information, the system can ask more in-depth questions about related symptoms and medical history in subsequent Q&A interactions, or provide parents with more detailed eye health knowledge and precautions. For example, the system can automatically push some articles about the prevention and care of common eye diseases in children to parents, or give priority to experienced experts when recommending registration departments and doctors.

[0114] In the analysis of the eye condition of the target child Xiao Ming, corneal curvature is an important eye feature parameter. Assume that at the first consultation (March 01, 2023), the corneal curvature measurement value of Xiao Ming's right eye was 43.5D (diopter), and that of the left eye was 44.0D. By the second consultation (September 15, 2023), the corneal curvature of the right eye became 43.8D, and that of the left eye became 44.2D. The measurement results of this consultation (November 01, 2023) showed that the corneal curvature of the right eye was 44.0D, and that of the left eye was 44.3D. When the system processes these corneal curvature data, it will incorporate them into the historical eye feature information sequence. For example, corneal curvature changes: [(2023-03-01, right eye 43.5D, left eye 44.0D), (2023-09-15, right eye 43.8D, left eye 44.2D), (2023-11-01, right eye 44.0D, left eye 44.3D)].

[0115] When generating eye influencing factors and weight information, the rate of change of corneal curvature can be used as an influencing factor. For example, the rate of change of corneal curvature of the right eye = (44.0-43.5) / (2023-11-01-2023-03-01) = 0.083 (assuming that the time interval is in months). Assume that after clinical research and data analysis, the weight of the impact of the rate of change of corneal curvature on eye health is set to 0.1. When generating target feature data, the eigenvalue corresponding to the rate of change of corneal curvature is 0.0083 (0.083×0.1). At this time, the target feature data can be expressed as: [vision-related eigenvalue -0.04 (0.1×0.4), intraocular pressure-related eigenvalue 0.05, eye axis-related eigenvalue 0.08, family history eigenvalue 0, eye habit eigenvalue unknown, corneal curvature-related eigenvalue 0.0083].

[0116] When generating eye judgment information, if the rate of change of corneal curvature is fast, it may indicate that the corneal morphology is changing, which may be related to the development of myopia or other eye problems. Combined with other factors, such as decreased vision, increased axial length, astigmatism, etc., the system can make a more comprehensive judgment on Xiao Ming's eye condition. For example, if the rate of change of corneal curvature is accelerated, the vision is significantly reduced, the axial length grows rapidly, and the difference between K1 and K2 values, the system will be more certain that Xiao Ming has excessive eye use or abnormal eye development, and then give more targeted suggestions in the consultation information, such as further examination of corneal topography and observation of details of corneal morphological changes, so as to timely detect potential problems and take corresponding measures, such as adjusting eye habits, wearing appropriate glasses, or performing orthokeratology lens trial evaluation (if relevant conditions are met). At the same time, in the generated eye report information, the relevant data and changes of corneal curvature will also be listed in detail for reference by doctors and parents, so as to have a more comprehensive understanding of Xiao Ming's eye physiological characteristics and potential risks.

[0117] The target child's identity type information is processed to generate the target child's historical eye feature information. The system obtains Xiao Ming's historical eye feature information from his previous consultation records and examination reports. For example, at the first consultation (March 01, 2023), Xiao Ming's naked eye vision test results were 1.0 for the right eye and 0.8 for the left eye, intraocular pressure measurement values ​​were 15mmHg for the right eye and 14mmHg for the left eye, and the axial length was 23.8mm for the right eye and 24.2mm for the left eye. At the second consultation (September 15, 2023), the naked eye vision was 0.8 for the right eye and 0.7 for the left eye, the intraocular pressure was 16mmHg for the right eye and 15mmHg for the left eye, and the axial length was 24.1mm for the right eye and 24.4mm for the left eye. In this consultation (November 1, 2023), the naked eye vision is 0.6 for the right eye and 0.6 for the left eye, the intraocular pressure is 17 mmHg for the right eye and 16 mmHg for the left eye, and the axial length is 24.5 mm for the right eye and 24.7 mm for the left eye. The eye feature data at different time points are integrated to form a historical eye feature information sequence of Xiao Ming. For example, it can be expressed as: naked eye vision change: [(2023-03-01, right eye 1.0, left eye 0.8), (2023-09-15, right eye 0.8, left eye 0.7), (2023-11-01, right eye 0.6, left eye 0.6)]. Changes in intraocular pressure: [(2023-03-01, right eye 15mmHg, left eye 14mmHg), (2023-09-15, right eye 16mmHg, left eye 15mmHg), (2023-12-01, right eye 17mmHg, left eye 16mmHg)]. Changes in axial length: [(2023-03-01, right eye 23.8mm, left eye 24.2mm), (2023-09-15, right eye 24.1mm, left eye 24.4mm), (2023-11-01, right eye 24.5mm, left eye 24.7mm)].

[0118] By analyzing these historical eye characteristics, we can observe that Xiao Ming's vision is gradually declining, the intraocular pressure shows signs of slowly rising, and the axial length of the eye is gradually increasing. These changing trends are very important for judging the direction of Xiao Ming's eye problems. For example, if the axial length grows too fast, it indicates that myopia is developing rapidly, and timely intervention measures need to be taken, such as adjusting eye habits, increasing outdoor activity time, or considering wearing orthokeratology lenses.

[0119] The historical eye feature information of the target children is processed to generate eye influencing factors and weight information corresponding to the eye influencing factors. The system extracts eye influencing factors from the historical eye feature information according to preset rules and algorithms. For example, the rate of visual acuity decline, the rate of intraocular pressure change, the rate of axial length growth, etc. can all be used as eye influencing factors. When calculating the rate of visual acuity decline, the difference between the current visual acuity and the first visual acuity is divided by the time interval, such as the rate of visual acuity decline of the right eye = (0.6-1.0) / (2023-11-01-2023-03-01) = -0.05 (assuming that the time interval is in months). The calculation method of the intraocular pressure change rate and the axial length growth rate are similar. At the same time, the corresponding weight is determined for each influencing factor. The determination of weights can be based on a large amount of clinical data and research results. For example, after analyzing the eye data of many children, it is found that the rate of visual acuity decline has a higher weight on eye health, assuming it is set to 0.4; the weight of the intraocular pressure change rate is 0.2; and the weight of the axial length growth rate is 0.3.

[0120] These eye-affecting factors and weight information can more accurately quantify the severity and development trend of Xiao Ming's eye problems. For example, if Xiao Ming's vision declines at a faster rate and its corresponding weight is higher, the system will pay more attention to this factor. When generating eye judgment information and consultation information, it will focus on content related to vision protection, such as asking Xiao Ming whether there are problems with his eye habits (such as long-term close-up use of the eyes, incorrect eye posture, etc.), and provide targeted eye suggestions (such as regular rest for the eyes, maintaining correct reading and writing posture, etc.).

[0121] Based on the preset eye impact feature set, the eye impact factors and the weight information corresponding to the eye impact factors are processed to generate target feature data. The preset eye impact feature set contains a series of feature categories related to eye health, such as vision-related features, intraocular pressure-related features, eye axis-related features, family history features (assuming that Xiao Ming's family has no history of genetic eye diseases, the feature value is 0), eye usage habits features (the initial assumption is unknown, and then obtained through questions and answers), etc. The system matches and integrates the eye impact factors and weight information with the preset eye impact feature set. For example, for the impact factor of the rate of vision decline, it is mapped to the vision-related feature category and multiplied by its weight of 0.4. Assuming that the intraocular pressure change rate and the axial length growth rate are 0.02 and 0.03 respectively after calculation and weight allocation, the generated target feature data can be expressed as: [vision-related feature value -0.04, intraocular pressure-related feature value 0.02, axial length-related feature value 0.03, family history feature value 0 (assuming that Xiao Ming's family has no history of genetic eye disease), eye habit feature value unknown].

[0122] The target feature data will serve as an important basis for subsequent analysis and judgment. For example, when comparing and analyzing the eye data of other children, or when inputting into the eye warning model for disease risk prediction, these target feature data can provide more comprehensive and accurate information. If Xiao Ming's target feature data is compared with the normal eye feature data range of children of the same age, and it is found that the vision-related feature values ​​are beyond the normal range, the system can further analyze the reasons and prompt parents in the consultation information to pay attention to Xiao Ming's vision problems and recommend a more detailed eye examination.

[0123] The target feature data is processed to generate the target child's eye judgment information. The system analyzes the target feature data and the preset eye judgment rules. For example, if the vision-related feature value is poor and the axial length growth rate is fast, the system determines that Xiao Ming has excessive eye use or bad eye habits. At the same time, considering that Xiao Ming is 8 years old, children of this age are in the stage of learning to use their eyes more, and the impact of eye habits on eye health is more critical. The system will also combine the previously obtained question and answer information (if any), such as whether Xiao Ming often watches TV and plays video games for a long time. Assume that in the previous question and answer, Xiao Ming said that he would play video games for 2-3 hours every day after school, which further confirms the nature of excessive eye use. Combining these factors, eye judgment information is generated, such as "Xiao Ming has excessive eye use, mainly because he has used electronic devices for too long, and the current decline in vision and the trend of changes in axial length suggest that he needs to adjust his eye habits in a timely manner." Eye judgment information will directly affect the subsequent generation of consulting information and the provision of suggestions. For example, the system will emphasize the importance of reducing the time spent using electronic devices in the consultation information, and recommend that Xiao Ming watch TV and play games within 1 hour a day. At the same time, it will recommend that parents take Xiao Ming for regular vision examinations and pay close attention to changes in vision so that problems can be discovered and measures can be taken in a timely manner.

[0124] The target children's eye judgment information is processed to generate consulting information for the target children. Based on the eye judgment information and all the information collected previously, the system generates detailed consulting information. For example, the consulting information will begin with a summary of Xiao Ming's current eye condition, such as "Xiao Ming, 8 years old, currently has a visual acuity of 0.4 in the right eye and 0.6 in the left eye. His visual acuity is on a downward trend, and his intraocular pressure and axial length have also changed accordingly. After analysis, there is a problem of excessive eye use." Then provide specific suggestions, such as "It is recommended that you reduce the time you use electronic devices to less than 1 hour a day. Increase outdoor activity time, at least 2 hours a day. Pay attention to eye hygiene and maintain correct reading and writing posture. Take Xiao Ming to the hospital regularly for eye examinations such as naked eye vision, intraocular pressure, axial length, optometry, corrected vision and corneal curvature. It is recommended to recheck every 3 months." Finally, the reply information also includes some intervention measures to control axial growth, such as "Choose OK lenses or defocus soft lenses or defocus frame glasses based on Xiao Ming's corneal curvature and whether he can try contact lenses."

[0125] The system will pass the generated consulting information to parents in a variety of ways, such as displaying it on the system interface, sending text messages or emails, etc. Parents can understand Xiao Ming's eye condition based on the consulting information and take corresponding measures to improve Xiao Ming's eye habits and protect his eye health. At the same time, the consulting information also provides a basis for parents to further communicate with doctors. Parents can ask doctors more targeted questions based on this information in subsequent consultations or consultations, thereby improving the efficiency and quality of medical services.

[0126] S107, processing the target eye recognition result and the consultation information of the target child to generate blockchain information.

[0127] In one embodiment, the target eye recognition result and the target child's consulting information are subjected to noise reduction processing to generate the target eye recognition result after noise reduction and the target child's consulting information after noise reduction. Original data situation: Assume that the eye recognition result of the target child Xiao Ming contains vision measurement data, intraocular pressure measurement data, and fundus image analysis results. There are some fluctuations in the vision measurement data. For example, the results of multiple measurements of the right eye's vision are 0.8, 0.7, 0.9, and 0.7, respectively. The intraocular pressure measurement values ​​also have similar small range fluctuations, such as the right eye intraocular pressure measurement values ​​of 16mmHg, 15.5mmHg, 16.2mmHg, and 15.8mmHg. There are some subtle noise points in the fundus image analysis results caused by the equipment or shooting environment, which affect the accurate judgment of features such as retinal vascular morphology and optic nerve head morphology. Noise reduction method and process: For vision measurement data, the system uses mean filtering to reduce noise. Calculate the average of multiple measurement values, i.e. (0.8+0.7+0.9+0.7) / 4=0.775, and use it as the right eye visual acuity value after noise reduction (keep two decimal places). For intraocular pressure measurement data, the same mean filter is used, (16+15.5+16.2+15.8) / 4=15.875 mmHg (keep three decimal places). For fundus images, use image filtering algorithms, such as Gaussian filtering, to remove noise caused by equipment or environment, so that features such as retinal vascular morphology and optic nerve head morphology are clearer. After noise reduction processing, more stable and accurate visual acuity and intraocular pressure values, as well as clearer fundus image analysis results, are obtained as the target eye recognition results after noise reduction.

[0128] Example of noise reduction of target child consultation information, original information: In the consultation information of the target child Xiao Ming, there are some vague or inaccurate statements when parents describe Xiao Ming's eye habits. For example, parents say that Xiao Ming "often" watches TV, but do not specify the specific duration, which varies from 1 to 3 hours a day; for the description of Xiao Ming's eye discomfort symptoms, some vague words are used, such as "sometimes the eyes feel uncomfortable", but it is not specified whether it is pain, dryness or other feelings, as well as the frequency and time pattern of the discomfort symptoms. Noise reduction method and process: The system clarifies the information by further asking parents. Regarding the issue of the length of time watching TV, the system will ask "How long do you watch TV every day when you say you often watch TV?" The parent answered "About 2 hours a day." For the symptoms of eye discomfort, the system will ask "Do you say that your eyes are uncomfortable, are they painful, itchy, dry or other feelings? How often does this discomfort occur? How long does it last each time?" The parent answered "It is a dry feeling, which occurs about 2-3 times a week, and lasts for more than ten minutes each time." Through this interactive method, more accurate and detailed information can be obtained, and vague and inaccurate statements can be transformed into clear and specific information as the target child consultation information after noise reduction.

[0129] The target eye recognition results after noise reduction are processed to generate clinical feature information. Taking Xiao Ming's eye recognition results after noise reduction as an example, the visual acuity value is 0.78 for the right eye (after noise reduction) and 0.65 for the left eye (assuming that the original data is obtained after similar processing), the intraocular pressure value is 15.88 mmHg for the right eye and 14.25 mmHg for the left eye (assuming that the left eye data has also been processed), and the fundus image shows that the retinal vascular morphology is normal but the edge of the optic nerve head is slightly blurred. Based on these data, the system can generate clinical feature information. For example, decreased vision (the right eye's visual acuity is lower than the lower limit of the normal visual acuity range for the same age group, and the normal visual acuity of an 8-year-old child is 1.0 and above), normal intraocular pressure (the intraocular pressure value is within the normal range, and the normal intraocular pressure range is 10-21 mmHg), and the optic nerve head morphology is suspected to be slightly abnormal (based on fundus image analysis). These clinical feature information can provide important basis for subsequent consultation and analysis.

[0130] The clinical feature information and the target child consultation information after noise reduction are fused to generate eye consultation information. The system fuses the generated clinical feature information (such as decreased vision, normal intraocular pressure, and suspected slight abnormality in the optic nerve head morphology) with the target child consultation information after noise reduction (such as watching TV for about 2 hours a day, dry eyes 2-3 times a week, each time lasting more than ten minutes). The eye consultation information is expressed as: "Xiao Ming, 8 years old (assumed age), the current eye condition shows decreased vision (right eye 0.78, left eye 0.65), normal intraocular pressure (right eye 15.88mmHg, left eye 14.25mmHg), and fundus images indicate suspected slight abnormality in the optic nerve head morphology. His daily eye habits are to watch TV for about 2 hours a day, with symptoms of dry eyes, which occur 2-3 times a week, each lasting more than ten minutes. Comprehensive analysis shows that decreased vision is related to eye habits, but abnormal optic nerve head morphology needs further examination and confirmation. It is recommended to reduce the time of watching TV, increase outdoor activities, closely observe the symptoms of dry eyes, and further examine the fundus and optic nerve function if necessary."

[0131] The eye consultation information is encrypted to generate blockchain information. The system uses an asymmetric encryption algorithm (such as the RSA algorithm) to encrypt the eye consultation information. First, a public key and a private key are generated, the public key is made public, and the private key is kept confidential. The eye consultation information is converted into a digital format (such as binary data), and then encrypted using the public key. The encrypted information becomes a string of seemingly disordered ciphertext, such as "010101101010111010100011..." (for example only). The encrypted ciphertext is used as blockchain information. Only authorized personnel with private keys (such as doctors, professional researchers, etc.) can use the private key to decrypt the ciphertext and obtain the original eye consultation information, thereby ensuring the security and privacy of the information during storage and transmission on the blockchain.

[0132] S108, processing the blockchain consultation information and the consultation purpose information of the target child to generate eye report information of the target child.

[0133] In one implementation, the blockchain consultation information is processed to generate eye physiological characteristic data. Assume that the blockchain consultation information contains data records of multiple eye examinations of the target child Xiao Ming. For example, in terms of vision examination, the vision of the right eye at different time points is 0.8, 0.7, and 0.6, and the vision of the left eye is 0.9, 0.8, and 0.7; the intraocular pressure measurement values ​​of the right eye are 15mmHg, 16mmHg, and 15.5mmHg, respectively, and the left eye is 14mmHg, 15mmHg, and 14.5mmHg; the axial length measurement results of the right eye are 21.5mm, 22mm, and 22.3mm, and the left eye is 21mm, 21.8mm, and 22mm. In addition, there is fundus image data. After being processed by image analysis technology, information such as retinal vascular morphology (such as quantitative indicators such as vascular thickness and curvature) and optic nerve head morphology (such as quantitative values ​​such as area and cup-disc ratio) is obtained. The system organizes and analyzes these data to extract key eye physiological characteristic data. For example, the average visual acuity is calculated as (0.8+0.7+0.6) / 3=0.700 (retain three decimal places), and the average visual acuity of the left eye is (0.9+0.8+0.7) / 3=0.800. The average intraocular pressure is (15+16+15.5) / 3=15.5 mmHg for the right eye, and (14+15+14.5) / 3=14.5 mmHg for the left eye. The average axial length is (21.5+22+22.3) / 3=21.933 mm for the right eye, and (21+21.8+22) / 3=21.6 mm for the left eye. At the same time, the quantitative indicators of retinal vascular morphology (such as the average vessel thickness of 0.3 mm, the curvature index of 0.7, etc.) and the quantitative values ​​of the optic nerve head morphology (such as the area of ​​2.5 mm 2 , cup-to-disc ratio is 0.3, etc.) These data constitute Xiao Ming’s eye physiological characteristics data.

[0134] Based on the consultation purpose information of the target child, the ocular physiological characteristic data is processed to generate ocular risk level information. The analysis direction is determined according to the consultation purpose. It is assumed that Xiao Ming's consultation purpose this time is to assess the risk of myopia, because his parents found that Xiao Ming's vision has declined significantly recently and his eye habits are not good (he uses electronic devices for a long time every day). According to this consultation purpose, the system focuses on ocular physiological characteristic data related to myopia, such as the trend of vision change and the growth of axial length. From the previously calculated ocular physiological characteristic data, Xiao Ming's right eye vision shows a downward trend, and the axial length of the right eye is gradually increasing, and the growth rate is relatively fast (such as calculating the change rate of axial length, the right eye is (22.3-21.5) / (third inspection time-first inspection time), assuming that the time interval is 1 year, the annual growth rate of axial length is 0.8mm / year, which exceeds the normal growth rate range of the same age group, and the normal range is 0.12-0.4mm / year). According to the preset myopia risk assessment standard, considering the trend of vision decline, the growth rate of axial length and other factors, the system assesses Xiao Ming's eye risk level as high risk. For example, if vision declines significantly and the axial length increases too fast, the system determines that the myopia risk level is high. If Xiao Ming's vision declines slowly and the axial length increases within the normal range, he may be assessed as low risk or medium risk.

[0135] The eye risk level information is processed to generate eye classification information. Since Xiao Ming is assessed to be at high risk of myopia, the system classifies his eyes as myopia tendency. If there are other eye problems, such as abnormally high intraocular pressure and obvious changes in the morphology of the optic nerve head, it may be classified as suspected glaucoma, etc. For different classifications, the system will have corresponding labels and processing methods. For example, for the myopia tendency category, further attention will be paid to factors related to the development of myopia, such as eye use environment, family history of myopia, etc. (assuming there is no history of myopia in the family), and relevant information will be highlighted in subsequent reports.

[0136] The eye classification information is processed to generate the eye report information of the target child. The eye report information will first contain basic information, such as "Name: Xiao Ming, Age: 8 years old, Gender: Male". Then the eye physiological characteristics data are listed in detail, such as "Average visual acuity of the right eye: 0.700, average visual acuity of the left eye: 0.800; average intraocular pressure of the right eye: 15.5mmHg, average intraocular pressure of the left eye: 14.5mmHg; average axial length of the right eye: 21.933mm, average axial length of the left eye: 21.6mm; retinal vascular morphology: average vascular thickness 0.3mm, curvature index 0.7; optic nerve head morphology: area 2.5mm 2, cup-to-disc ratio 0.3". Then the eye risk level and eye classification are explained, "Eye risk level: high (myopia tendency)". The report will also give suggestions, such as "In view of Xiao Ming's current eye condition, the risk of myopia is relatively high. It is recommended to reduce the time of using electronic devices to less than 1 hour a day; increase outdoor activities, at least 2 hours a day; check vision, intraocular pressure and axial length every 3 months, and pay close attention to eye changes; maintain correct reading and writing posture, and pay attention to eye hygiene". If there is other relevant information, such as family medical history, allergy history, etc. (assuming there is no history of allergies and no history of genetic eye diseases in the family), it will also be reflected in the report, "Family medical history: no history of genetic eye diseases; history of allergies: none". Finally, the report may also contain some popular science knowledge, such as "Myopia is a common eye problem in children, which is mainly related to genetic and environmental factors. Early intervention and good eye habits can help prevent the occurrence and development of myopia".

[0137] The ocular report information can be presented in an electronic document (such as PDF format) or in a specific medical information system interface. The report is clearly laid out and easy to read, with clear titles and categories for each piece of data and information. For example, physiological characteristic data is presented in a table, risk levels and ocular classifications are displayed in prominent fonts, and the recommendations are listed in a bulleted list, making it easier for parents and doctors to view and understand. If necessary, a printed version of the report can also be generated for parents to use in consultation or daily records.

[0138] In this application, the server obtains various information about the target children and related data sets and models, including eye information (vision, intraocular pressure, eye structure, etc.), question and answer information (eye habits, discomfort symptoms, etc.), identity information (basic and school learning situation), consultation demand information, as well as training data sets (question and answer corpus, eye change data set, consultation results) and preset eye warning model (combined with CNN and RNN architecture). Then the information is processed in series, such as generating children's ophthalmology labels based on the eyes and question and answer information, which are obtained through complex calculations and analysis; pre-processing the training data set to obtain a training set with target feature data, involving multiple links such as preset mapping table construction, data grouping, feature extraction, data set division, and prediction training; using the training set to train the preset eye warning model to generate the target eye warning model requires data processing, model architecture selection and parameter setting, training evaluation optimization and other processes.

[0139] Then, based on the target eye warning model, the target child ophthalmology label is processed to obtain the target eye recognition result, and then the identity information is processed to generate consulting information, including the steps of generating identity type information, historical eye feature information, eye influencing factors and weight information, target feature data, eye judgment information, etc. After that, the target eye recognition result and consulting information are denoised, fused, and encrypted to generate blockchain information information. Finally, the eye report information is generated based on the blockchain information information and consulting purpose information, including eye physiological feature data, risk level information, classification information and detailed report content (basic information, data, suggestions, relevant medical history, popular science knowledge, etc.), and the report format is diverse and easy to view and understand.

[0140] In one embodiment, if Figure 2 As shown, the present application also provides a pediatric ophthalmic disease diagnosis device based on a large model and blockchain, including:

[0141] An acquisition module 201 is used to acquire the target child's eye information, the target child's question and answer information, the target child's identity information, the target child's consultation purpose information, a training data set, and a preset eye warning model, wherein the training data is used to characterize other children's question and answer corpora, eye change data sets, and consultation results;

[0142] The processing module 202 is used to process the eye information of the target child and the question and answer information of the target child to generate an ophthalmic label for the target child; pre-process the training data set to generate a training data set with target feature data, wherein the target feature data is used to characterize risk factors affecting eye features; process the preset eye warning model based on the training data set with target feature data to generate a target eye warning model; process the ophthalmic label of the target child based on the target eye warning model to generate a target eye recognition result; process the identity information of the target child to generate consulting information for the target child; process the target eye recognition result and the consulting information of the target child to generate blockchain information; process the blockchain consulting information and the consulting purpose information of the target child to generate an eye report information for the target child.

[0143] Each embodiment in this application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for evaluating the children's ophthalmic disease guidance method based on a large model and blockchain, an electronic device, an electronic device, and a readable storage medium embodiment, since it is basically similar to the above-mentioned children's ophthalmic disease guidance method based on a large model and blockchain, the description is relatively simple, and the relevant parts can be referred to the partial description of the above-mentioned children's ophthalmic disease guidance method based on a large model and blockchain.

Claims

1. A method for guiding children's ophthalmic diseases based on a large model and blockchain, characterized in that: include: Obtaining the target child's eye information, the target child's question and answer information, the target child's identity information, the target child's consultation purpose information, a training data set, and a preset eye warning model, wherein the training data is used to characterize other children's question and answer corpora, eye change data sets, and consultation results; Processing the eye information of the target child and the question-and-answer information of the target child to generate an ophthalmology label for the target child; Preprocessing the training data set to generate a training data set with target feature data, wherein the target feature data is used to characterize risk factors affecting eye features; Processing the preset eye warning model based on the training data set with target feature data to generate a target eye warning model; Processing the target child ophthalmology label based on the target eye warning model to generate a target eye recognition result; Processing the identity information of the target child to generate consulting information of the target child; Processing the target eye recognition result and the consultation information of the target child to generate blockchain consultation information; The blockchain consultation information and the consultation information of the target child are processed to generate eye report information of the target child.

2. The method according to claim 1, characterized in that The eye information of the target child and the question-and-answer information of the target child are processed to generate an ophthalmological label of the target child, including: Processing the question-and-answer information of the target child to generate physiological characteristic information and basic information of the target child, wherein the physiological characteristic information of the target child includes family history, birth history, allergy history, and systemic medical history and surgical history other than ophthalmology, and the basic information of the target child includes the name of the target child, the date of birth of the target child, the weight of the target child, the target child's eye use time information, and the target child's eye use location information; Processing the eye information of the target child to generate eye feature information of the target child; Processing the physiological characteristic information of the target child, the basic information of the target child and the eye characteristic information of the target child, embedding the processed data into a vector database and assigning a unique identifier, and generating a child eye data matching value; The matching values ​​of the children's eye data are processed to generate target children's ophthalmology labels according to the marked eye muscle diseases, refractive diseases, lens diseases, fundus diseases, glaucoma, ocular surface diseases, ocular trauma and other children's ophthalmology departments and the classification of diseases under different children's ophthalmology departments; The method includes a calculation formula for obtaining a matching value of the child's eye data, and the calculation formula is: Among them, a and b are vectors related to the target child's eye data.

3. The method according to claim 1, characterized in that Preprocessing the training data set to generate a training data set with target feature data includes: Obtaining a preset mapping table, wherein the preset mapping table is used to characterize historical eye characteristics and related physiological characteristics and basic information of children in a target age group; Performing grouping processing on the training data set based on the preset mapping table to generate a grouped training data set, wherein the grouped training data set includes eye image information and data information of children of different age groups; Perform different feature extraction on the grouped training data set to generate an original feature library; Dividing the original feature library into various feature data sets to generate a training set and a validation set; Based on the classifier, each validation set divided by the original feature library is predicted to generate a prediction result; Use the preset algorithm to divide each training set in the original feature library for training and generate the prediction results of the validation set class; The prediction results and the validation set prediction results are processed to generate target feature data, where the target feature data is used to characterize feature information reflecting eye abnormalities in children.

4. The method of claim 1, processing the target child ophthalmology label based on the target eye warning model to generate a target eye recognition result, comprising: Processing the target pediatric ophthalmology label based on the target eye warning model to generate a target approximation value; Processing the target approximation value to generate a target eye recognition result; The target eye warning model includes a calculation formula for obtaining a target approximate value, and the calculation formula is: Where N is the number of samples, y i is the true category label of the i-th sample, θ ij is the angle between the feature vector of the i-th sample and the j-th category weight vector, s is a learnable scale factor used to control the discrimination between different categories, and m is the additive margin, a constant greater than 0, used to add an additional margin to the angle of the true category.

5. The method according to claim 4, characterized in that The identity information of the target child is processed to generate consulting information of the target child, including: Processing the identity information of the target child to generate identity type information of the target child, wherein the identity type information of the target child is used to characterize the consultation duration, number of consultations and historical consultation records of the target child; Processing the identity type information of the target child to generate historical eye feature information of the target child; Processing the historical eye feature information of the target child to generate an eye influencing factor and weight information corresponding to the eye influencing factor; Processing the eye impact factor and the weight information corresponding to the eye impact factor based on a preset eye impact feature set to generate target feature data; Processing the target feature data to generate eye judgment information of the target child; The eye judgment information of the target child is processed to generate consulting information of the target child.

6. The method according to claim 5, characterized in that The target eye recognition result and the consultation information of the target child are processed to generate blockchain consultation information, including: Performing noise reduction processing on the target eye recognition result and the target child's consultation information to generate a noise-reduced target eye recognition result and a noise-reduced target child's consultation information; Processing the denoised target eye recognition result to generate clinical feature information; fusing the clinical feature information and the noise-reduced target child consultation information to generate eye consultation information; The eye consultation information is encrypted to generate blockchain information.

7. The method of claim 6, processing the blockchain consultation information and the consultation information of the target child to generate eye report information of the target child, comprising: Processing the blockchain consultation information to generate eye features and basic information data of the target child; Processing the eye characteristics and basic information data of the target child based on the consultation purpose information of the target child to generate eye risk level information; Processing the eye risk level information to generate eye classification information; The eye classification information is processed to generate eye report information of the target child.

8. A pediatric ophthalmological disease diagnosis device based on a large model and blockchain, characterized in that: The device includes: An acquisition module, used to acquire the target child's eye information, the target child's question and answer information, the target child's identity information, the target child's consultation information, a training data set, and a preset eye warning model, wherein the training data is used to characterize other children's question and answer corpora, eye change data sets, and consultation results; A processing module is used to process the eye information of the target child and the question and answer information of the target child to generate an ophthalmic label of the target child; pre-process the training data set to generate a training data set with target feature data, wherein the target feature data is used to characterize risk factors affecting eye features; based on the training data set with target feature data, the preset eye warning model is processed to generate a target eye warning model; based on the target eye warning model, the ophthalmic label of the target child is processed to generate a target eye recognition result; the identity information of the target child is processed to generate consulting information of the target child; the target eye recognition result and the consulting information of the target child are processed to generate blockchain consulting information; the blockchain consulting information and the consulting information of the target child are processed to generate eye report information of the target child.

9. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions for the first processor; The first processor is configured to execute the pediatric ophthalmic disease diagnosis guidance method based on a big model and blockchain of any one of claims 1 to 7 by executing executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, it implements the children's ophthalmic disease diagnosis guidance method based on a big model and blockchain according to any one of claims 1 to 7.

Images