CN111128318A - Optical biological parameter prediction method and device - Google Patents
Optical biological parameter prediction method and device Download PDFInfo
- Publication number
- CN111128318A CN111128318A CN201911319885.0A CN201911319885A CN111128318A CN 111128318 A CN111128318 A CN 111128318A CN 201911319885 A CN201911319885 A CN 201911319885A CN 111128318 A CN111128318 A CN 111128318A
- Authority
- CN
- China
- Prior art keywords
- information
- optical
- data
- model
- questionnaire
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
- G16H10/20—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for electronic clinical trials or questionnaires
Abstract
The invention provides an optical biological parameter prediction method and equipment, wherein the method comprises the following steps: acquiring optical biometric data, expected time and questionnaire survey data of a user, wherein the optical biometric data comprises the age of the user, time information and optical biometric parameters during optical biometric measurement, and the questionnaire survey data comprises a plurality of auxiliary information capable of influencing eyes; predicting an optical biometric parameter of the user at the desired time from the optical biometric data and questionnaire survey data using a prediction model.
Description
Technical Field
The invention relates to the field of artificial intelligent medical data analysis, in particular to an optical biological parameter prediction method and device.
Background
Myopia is one of the eye disorders and one of the most serious public health problems worldwide. Particularly, in recent years, with the popularization of various intelligent devices, the incidence of myopia in good-haired people such as children and teenagers is increasing, the myopia seriously harms physical and mental health of the children and the teenagers, and the learning and life of the children and the teenagers are greatly influenced.
At present, mechanisms and personnel in the field can accurately detect optical biological parameters of a human body, such as axial length, corneal thickness, anterior chamber depth and the like, by various accurate measurement means, and further comprehensively judge the current state of the eyeball.
In recent years, more and more research is being put into predicting visual power or diopter. In the prior art, most of the chinese patent with application number 2018100240709 uses the past vision information to directly predict the future vision or uses the past diopter to directly predict the future diopter. Although the methods have a certain prediction effect on the development trend of the eye state, the information such as the visual power value and the diopter is the development result of the eye state, and the factors causing the change of the eye state of a human body are many, and the same poor vision can be caused by the change of different parts of the eye, so the accuracy of the result obtained by the scheme of directly predicting the visual power value or the diopter is not high enough, and the change condition of the detailed parts of the eye cannot be reflected.
Disclosure of Invention
In view of the above, the present invention provides a method for training an optical biological parameter prediction model, comprising:
acquiring sample data and label information thereof, wherein the sample data comprises optical biometric data and questionnaire survey data, the optical biometric data comprises time difference of two optical biometrics of the same measured person, time information and optical biometric parameters during the previous optical biometric measurement, and age, the questionnaire survey data comprises a plurality of auxiliary information capable of influencing eyes, and the label information comprises the optical biometric parameters during the next optical biometric measurement;
and training a prediction model by using a plurality of sample data and label information thereof so as to enable the prediction model to obtain predicted optical biological parameters according to the sample data, and optimizing self parameters according to the difference between the predicted optical biological parameters and the optical biological parameters in the label information.
Optionally, the obtaining of the sample data and the tag information thereof includes:
acquiring any two optical biological measurement record data of the same measured person from the plurality of optical biological measurement record data according to the identity information of the measured person;
and determining the time difference according to the optical biological measurement time information in the two optical biological measurement record data, and acquiring the former optical biological measurement time and the optical biological parameter corresponding to the latter optical biological measurement time.
Optionally, the obtaining of the sample data and the tag information thereof includes:
obtaining a questionnaire filled by the testee according to the identity information of the testee, wherein the questionnaire comprises a plurality of questionnaire information capable of influencing eyes;
and converting the question-answer information into a numerical value which can be read by a prediction model.
Optionally, before acquiring the sample data and the tag information thereof, the method includes:
obtaining a questionnaire filled by the testee according to the identity information of the testee, wherein the questionnaire comprises a plurality of questionnaire information capable of influencing eyes;
determining whether conflicts exist among the question answering information or not according to the question answering information and a preset question answering rule;
and eliminating the questionnaires with conflicts.
Optionally, the predictive model is a decision tree model.
Optionally, training a prediction model by using a plurality of sample data and label information thereof, including:
determining a preset value combination of the hyperparameters of the decision tree model, wherein the hyperparameters comprise the number of decision trees and the number of the most used characteristics of each decision tree;
and training by utilizing the plurality of sample data and label information thereof under the condition that the decision model uses each combination respectively to obtain a model with optimal performance.
Optionally, the plurality of sample data is divided into a plurality of shares, one of which is configured as a validation set for validating performance of the predictive model, and the other of which is configured as a training set for training the predictive model.
Optionally, the sample data further includes gender information, and the auxiliary information includes parent vision information of the testee and eye condition information of the testee.
The invention also provides an optical biological parameter prediction method, which comprises the following steps:
acquiring optical biometric data, expected time and questionnaire survey data of a user, wherein the optical biometric data comprises the age of the user, time information and optical biometric parameters during optical biometric measurement, and the questionnaire survey data comprises a plurality of auxiliary information capable of influencing eyes;
predicting an optical biometric parameter of the user at the desired time from the optical biometric data and questionnaire survey data using a prediction model.
Optionally, the obtaining questionnaire survey data includes:
acquiring a questionnaire filled by a user, wherein the questionnaire comprises a plurality of questioning and answering information capable of influencing eyes;
and converting the question-answer information into a numerical value which can be read by a prediction model.
Optionally, the predictive model is a decision tree model.
Optionally, the optical biological parameters include ocular axial length, corneal thickness, anterior chamber depth.
Correspondingly, the invention also provides an optical biological parameter prediction model training device, which comprises at least one processor and a memory which is in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the optical biometric prediction model training method described above.
Accordingly, the present invention also provides an optical biological parameter prediction device comprising at least one processor and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the optical biometric parameter prediction method described above.
According to the training method and the training device of the optical biological parameter prediction model provided by the invention, a large number of sample data comprising the time difference of two optical biological measurements of the measured person, the time information and the optical biological parameters during the previous optical biological measurement, the age and the questionnaire survey data and the label information comprising the optical biological parameters during the next optical biological measurement are used for training the prediction model, so that the trained prediction model can predict the optical biological parameters of the measured person at a future time point according to the current optical biological measurement data and the questionnaire survey data of the measured person, and has stronger robustness and higher accuracy.
According to the optical biological parameter prediction method and the device provided by the invention, time information, optical biological parameters and ages of a user during optical biological measurement, questionnaire survey data and expected time are used as input, the input data are identified by using a prediction model, and the optical biological parameters of the user at the expected time are output, so that the accurate prediction of the future optical biological parameters is realized, and the optical biological parameter prediction method and the device have higher reference value for subsequent treatment or prevention of eye diseases.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a schematic diagram of the contents of a questionnaire in an embodiment of the invention;
fig. 2 is a flowchart of training an eye state prediction model in an embodiment of the present invention.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it should be understood that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In addition, the technical features involved in the different embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The embodiment of the invention provides a method for training an optical biological parameter prediction model, wherein the prediction model is a machine learning model, such as a logistic regression model, a random forest model, an SVM model and the like are feasible.
Firstly, sample data and label information thereof are obtained. The sample data includes optical biometric data and questionnaire survey data, wherein the optical biometric data includes time difference between two optical biometrics of the same subject, time information and optical biometric parameters at the time of the previous optical biometric measurement, and age. The optical biometric time means a time when a subject (user or patient) receives a medical eye examination. The age is the age of the subject at the time of the previous optical biometric measurement, and the value can be calculated from the birth date of the subject and the time information at the time of the previous optical biometric measurement.
The optical biological parameter may be ocular axial length, corneal thickness, anterior chamber depth, and the like.
By way of example, the optical biometric data in a piece of sample data includes Gap, which has a value of timen-timem，timemIs the time when the measured person A receives one optical biological measurementnIs the tested person A at timemThe next time to receive eye examination. The sample data also includes timemAnd at timemEye axis length information AXL detected and recorded in real timemAnd the measured person A is at timemAge at agem. All of the above times are feasible to the nearest year, month and day.
The questionnaire survey data comprises a plurality of auxiliary information which can affect the eyes, and specific optional contents are various, such as genetic information, life habit information, eye abnormality information, information of the environment where the equipment is located, identification information of whether the equipment for correcting vision is worn, and the like, and the information is expressed in a form that can be read by a machine learning model.
The label information includes the optical biometric parameters at the time of the last optical biometric measurement, such as the time of the person A in the above examplenEye axis length information AXL detected and recorded in real timen。
As a preferred embodiment, the sample data further includes gender information of the subject, which may be part of the optical biometric data or as an ancillary information. The auxiliary information at least comprises the vision information of the parents of the testee and the eye condition information of the testee. Experiments prove that the preferred information used in the embodiment is information with strong influence on the eyes of the teenagers, and the prediction accuracy of the trained model can be improved by using the information as a part of training data.
After the plurality of sample data and the label information are obtained, the initial prediction model is trained, so that the prediction model predicts the optical biological parameters according to the sample data, and optimizes the parameters according to the difference between the predicted optical biological parameters and the optical biological parameters in the label information.
The prediction model is configured to predict an optical biometric parameter at a later point in time based on the input optical biometric data and questionnaire survey data. For example, the model reads the sample data about the subject A, and applies the sample data to the subject AThe content is processed by feature extraction and the like, and one eye axis length information AXL is outputn', indicates the time of the person A analyzed by the modelnEye axis length information of time. Then comparing AXLn' and AXLnDue to AXLnIs the actual value, AXLnThe model calculates an estimated value according to parameters of the model, a certain difference exists between the model and the model, particularly, the difference is possibly large at the initial training stage, the training target is to reduce the difference, and therefore, after the model reads a large amount of sample data and label information to predict, compare and correct the parameters, the difference can be continuously reduced, and the expected performance is achieved. Since this is a regression problem, the evaluation indicators that can be used in training are the mean absolute error and the R-square.
As another alternative, the model may also be configured as a variation value of the optical biological parameter, which represents a variation of the optical biological parameter at a later time point. For example, the model reads the sample data about the measured person a, performs feature extraction processing on the content of the sample data, and outputs an eye axis length information Δ AXL representing the time of the measured person a analyzed by the modelnThe change in length of the eye axis. Then comparing the delta AXL with the AXLn-AXLmTo optimize its own parameters.
It should be noted that the time information and the age are different concepts, and the time information is a time point and should be accurate to at least the year. The change of optical biological parameters of people in different times can be different, which is probably caused by time background related factors. For example, for two subjects in different times, even if the age, the biological parameter measured at the previous time, the time difference between the two measurements, and other characteristics are the same, the biological parameters may be different between the two measurements only if the measurement time is different. Therefore, the purpose of adding the characteristic of measuring time in the training is to enable the prediction model to take the time factor into consideration, so that the accuracy of the prediction result is improved.
According to the optical biological parameter prediction model training scheme provided by the invention, a large amount of sample data comprising the time difference of two optical biological measurements of the measured person, the time information and the optical biological parameters at the time of the previous optical biological measurement, the age and the questionnaire survey data and the label information comprising the optical biological parameters at the time of the next optical biological measurement are used for training the prediction model, so that the trained prediction model can predict the optical biological parameters of the measured person at a future time point according to the current optical biological measurement data and the questionnaire survey data of the measured person, and has stronger robustness and higher accuracy.
In a preferred embodiment, a decision tree model is used as the prediction model. The decision tree model adopts a supervised machine learning prediction algorithm, and can be used for classification prediction and regression prediction. There are a variety of alternative decision tree based integrated machine learning models, such as random forest, XGBoost, lightGBM, Adaboost, GBDT, etc. The prediction models based on the decision tree have the advantages of simplicity, intuition, wide application range and the like. In this embodiment, a random forest regression algorithm is specifically adopted.
The regression tree is a tree structure including three nodes, a root node, an internal node and a leaf node. Each non-leaf node represents a feature, each branch of the non-leaf node including an output of the feature over a range of values, each leaf node storing a category.
The regression tree divides the input space by adopting a heuristic method, the whole feature space is traversed, the optimal segmentation feature and the optimal segmentation point are found, the input space is divided into two parts, and then the operation is repeated. And the optimal slicing characteristics and the optimal slicing points are determined by minimizing a slicing error, i.e., a least square of a true value and a predicted value of the divided area.
The process of prediction using decision trees is: and starting from the root node, testing the corresponding characteristics of the sample to be predicted, selecting an output branch according to the characteristic value of the sample until reaching the leaf node, and taking the value stored in the leaf node as a decision result.
In a random forest, different training sets are generated by a bootstrap sampling method, different feature sets are randomly selected, so that a plurality of different regression trees are generated, and the average value of the predicted values of all decision trees is the predicted value of the random forest.
Random forest regression is a machine learning method for an integration class of regression tasks. It learns to construct multiple decision trees during training and takes the average of the regression predictions of the individual trees as the output of the model.
The decision tree model includes two hyper-parameters, which are the number of decision trees and the number of features most used per decision tree, respectively. To improve the efficiency of model training, a plurality of selectable values are first provided for each hyper-parameter, and then a grid of sets of hyper-parameter values is built to represent all possible combinations, each grid point representing one set of hyper-parameter values. The preset selectable value of the hyper-parameter may be an empirical value or a reference value obtained beforehand by other training data.
Therefore, in the model training process, the default value combination of the hyper-parameters of the decision tree model is determined firstly, and then under the condition that the decision tree model uses each combination, a plurality of sample data and label information thereof are used for training to obtain the model with the optimal performance.
Further, the plurality of sample data is divided into a plurality of shares, wherein one share is configured as a verification set and used for verifying the performance of the prediction model, and the other shares are configured as a training set and used for training the prediction model.
For example, the training set (sample data and label information) is divided into 5 parts, and 5 rounds of training are performed for each set of super-parameter values. 4 parts of data are used as a training set for each round of training, and model parameters are learned; and the other data is used as a verification set for evaluating the model, so that the robustness of the model is enhanced by a five-fold cross-validation method.
Because the model uses the decision tree model, the influence of different factors on the vision of the teenagers can be analyzed and presented according to the characteristic importance of the model, wherein the factors comprise all factors in the optometry data and questionnaire survey data, and therefore, instructive opinions are provided for the protection of the vision of the teenagers.
In a specific embodiment, the influence information shown in the following table can be obtained by using a decision tree model:
| feature(s) | Importance of | Feature(s) | Importance of |
| gap | 0.2835 | Poor eyesight when wearing glasses | 0.0084 |
| Length of eye axis | 0.1259 | City | 0.0079 |
| Age (age) | 0.1213 | Rural area | 0.0075 |
| Year of measurement | 0.0732 | Squinting | 0.0074 |
| Eyesight of father and father | 0.0324 | Soreness and distension of the eyes | 0.0072 |
| Eyesight of mother | 0.0273 | |
0.0071 |
| Eye for working day | 0.0221 | Oblique eye | 0.0069 |
| Outdoor weekend | 0.0216 | |
0.0068 |
| Working outdoor | 0.0206 | |
0.0068 |
| Eye for weekend | 0.0197 | |
0.0061 |
| Distance between eyes and book | 0.0111 | Red blood silk | 0.0056 |
| Antenatal/cesarean section | 0.0108 | Side view | 0.0056 |
| Sitting posture | 0.0107 | Preventive measures | 0.0055 |
| Head distortion | 0.0105 | Blinking eye | 0.0044 |
| |
0.0103 | |
0.0036 |
| |
0.0101 | |
0.0034 |
| Desk lamp lighting | 0.0101 | Lacrimation | 0.0032 |
| |
0.0096 | |
0.0031 |
| Distance between finger and pen point | 0.0095 | |
0.0027 |
| Fluorescent lamp lighting | 0.0092 | History of oxygen inhalation | 0.0022 |
| |
0.0089 | Mountain village | 0.0007 |
| Frown | 0.0086 | Pasturing area | 0.0004 |
| Preterm/term | 0.0084 |
Wherein the first and third columns are related to factors involved in optical biological measurement data or questionnaire survey data, the second and fourth columns are the influence degree of the factors on the change of the length information of the eye axis, and the larger the numerical value is, the higher the influence degree is. Where gap is the time difference and "eye length" refers to the length of the eye at the time of the previous optical biometric measurement. As can be seen from the above table, factors having a large influence on the length of the axial line of the subject's eye include age, current year, vision condition of parents, and eye condition of the subject.
With respect to the collection and generation of the sample data and the label information thereof, the present embodiment provides the following preferred scheme for data cleaning and processing. Specifically, a large amount of optical biometric record data may be acquired first, such as those available through medical institutions, and such records typically include the identity information, such as the subject's ID (unique number) or name, and its optical biometric parameters. If the same testee has more than two optometry records, the optometry records can be used for generating the sample data.
And acquiring any two optical biological measurement record data of the same measured person in a large amount of optical biological measurement record data according to the identity information of the measured person. And determining the time difference according to the optical biological measurement time information in the two optical biological measurement record data, and acquiring the former optical biological measurement time as the time information, acquiring the optical biological parameter corresponding to the former optical biological measurement time and acquiring the optical biological parameter corresponding to the latter optical biological measurement time.
Any two optical biological measurement records of different time of the same tested person number can be converted into a sample, wherein the record of the previous time simulates the current record of the sample; later in time recordings simulate "future" recordings. If some person has n complete optometry records, it can generate corresponding record
And (4) sampling. In addition, each sample data used by the invention only needs the related content of one eye actually, and the optical biological measurement record usually contains the information of the two eyes of the tested person, so that the data can be split when the sample data is collected, and more sample data can be obtained by expansion. Specifically, if a certain optical biometric record contains the eye axis length data of the left eye and the right eye at the same time, the record can be split into two records, and the left-eye optical biometric parameter data and the right-eye optical biometric parameter data are used respectively, so that two sample data can be obtained.
Similarly, a large number of questionnaires filled by the testees can be acquired, wherein identity information such as the ID (unique number) or name of the testee is also recorded, and complete sample data and tag information thereof for each testee can be obtained by matching the identity information.
The questionnaire filled by the testee comprises a plurality of question and answer information which can influence the eyes, for example, the question of 'whether the father of the testee is short-sighted' or 'not' can be two options of 'yes' or 'no', and also can comprise a plurality of options of 'high short-sightedness', 'medium short-sightedness', 'low short-sightedness', 'far-sightedness', 'no short-sightedness', and the question and answer information is formed through option marks of 'ABCD' or '1234'.
Other question and answer information such as "average cumulative outdoor activity time of subject monday to friday", "subject has the following symptoms" and the like are also expressed in the option manner. As shown in fig. 1, the questionnaire employed in the present embodiment has 13 questions, and these questions may be multiple choices or single choices. It should be noted that the question answering shown in fig. 1 is only for explaining the auxiliary information of the present invention, and not for explaining the content of the auxiliary information.
After the questionnaire filled by the testee is acquired, the question and answer information in the questionnaire is converted into a numerical value which can be read by the prediction model. The numerical values may include two classes, representing continuity features and classification features in the machine learning model, respectively, and may further include null values without numerical values. The types of these values may be floating point, used to represent continuity features; the discrete type is suitable for single-choice questions with only two options and is converted into binary continuity characteristics; and the binary type is suitable for representing the continuity characteristics of the multiple choice questions.
In a preferred embodiment, a data cleansing process may also be performed in order to obtain accurate sample data. For the optical biometric data, some contents such as the age (or birthday) of the subject, the optical biometric date, the optical biometric parameters may be set as indispensable contents, and the contents may be excluded when they are absent from the acquired record.
Some or all questions can be set as the necessary questions for the questionnaire, and the necessary questions can be removed if the information of the necessary questions is lacked in the question-answering records. In addition, whether conflicts exist among the question answering information can be determined according to the question answering information and preset question answering rules, and therefore the questionnaires with the conflicts can be removed. By way of example, question 7 in fig. 1 is "is the child the following symptoms? ", this question has a total of 9 options, including 8 different symptoms and the" none above "option. If a record answers with both "none above" and any other option selected, the record is cleared. According to the optimal data cleaning scheme, questionnaire data with conflict information are removed, so that the accuracy of sample data is improved, the influence of unreasonable training data on the model learning process is avoided, and the training efficiency and performance of the prediction model are improved.
After the performance of the prediction model reaches the prediction level through training, the model can be used for prediction. The embodiment of the invention provides an optical biological parameter prediction method, wherein the model can be a machine learning model such as logistic regression, random forest, SVM and the like, and the model can be trained by using the training method without limitation. The prediction method comprises the following steps:
optical biometric data of a user, including the user's age, time information at the time of optical biometric measurement, and optical biometric parameters, expected time, and questionnaire data, including a plurality of auxiliary information capable of constituting an influence on the eyes, are acquired. In this embodiment, the optical biometric parameter and the time information refer to information obtained by the user once receiving the optical biometric parameter measurement, the age is the age of the user once receiving the optical biometric parameter, the user can provide the birth date, and then the system calculates the age of the user according to the time information. The expected time is a time point desired to be predicted, and is a future time point, such as three years, five years, or ten years after the current time point.
The optical biological parameter may be ocular axial length, corneal thickness, anterior chamber depth, and the like.
The optical biometric parameters of the user at the desired time are predicted from the optical biometric data and questionnaire survey data using a prediction model. The trained predictive model infers the optical biological parameter of the user at a future time point according to the characteristics of the input information.
As an example, the optometric data includes time information at the time of optical biometrics of the user AmAnd eye length information AXLmAge, agem(ii) a Number of questionnairesThe data is model readable data converted from the questionnaire of FIG. 1; the expected time is n, meaning timemAfter n years, it is recorded as timen. The prediction model outputs the time of the user A according to the datanThe eye axis length information AXLnOr relative to AXLmChange value of Δ AXL (AXL)m+ΔAXL＝AXLn)。
It should be noted that the time information and the age are different concepts, and the time information is a time point and should be accurate to at least the year. The purpose of adding the characteristic of the measuring time is to enable a prediction model to take the time factor into consideration, so that the accuracy of a prediction result is improved.
According to the optical biological parameter prediction scheme provided by the invention, time information, optical biological parameters, age, questionnaire survey data and expected time of a user in the light examination are used as input, the input data is identified by using a prediction model, and the optical biological parameters of the user in the expected time are output, so that the accurate prediction of the future optical biological parameters is realized, and the optical biological parameter prediction scheme has a high reference value for subsequent treatment or prevention of eye diseases.
When the optical biological parameters of the two eyes of the user are respectively predicted by the scheme, the optical biological parameters input in the two recognition processes may be different (the left eye and the right eye respectively), and other data are the same. In addition, the method does not fix the size of the expected time, the user can provide any numerical value according to the requirement, and if the change curve of the optical biological parameter is required to be predicted, only a plurality of expected times need to be given, and multiple deductions are made.
Regarding the questionnaire survey data during prediction, in practical application, the user may fill in the questionnaire first, and then the question-answer information is converted into a numerical value that can be read by the prediction model, which may specifically refer to data conversion of a training scheme, and details are not repeated here.
On the basis of predicting the length of the eye axis, a linear regression model can be established based on the linear relation between the change of the length of the eye axis and the change of diopter, and the change of diopter is predicted, so that the change curve of diopter is predicted.
Linear regression refers to a regression analysis that models the relationship between multiple variables using the least squares function of a linear regression equation, assuming that there is a linear correlation between two or more variables. In the present application, the change of the axial length of the eye is an independent variable, the change of the diopter is a dependent variable, and a linear function of the change of the diopter and the change of the axial length of the eye can be obtained by using linear regression, that is, the change of the diopter can be predicted.
In order to train the diopter prediction model synchronously, diopter information corresponding to the biometric parameter may also be added as label information at the time of training. Training the diopter prediction model by using the training sample data and the label information thereof, so that the diopter prediction model outputs diopter information according to the input eye axis length; or the diopter prediction model outputs the diopter information change according to the input change information of the eye axis length.
FIG. 2 shows a flow chart for training an eye state prediction model by first obtaining a large number of questionnaires and optical biometric records for data cleansing, i.e., rejecting unreasonable, conflicting, missing data; then, performing record matching according to the identity information of the testee to obtain the optical biological measurement record and questionnaire of each testee; then, generating a sample to obtain sample data which can be read by the prediction model and label information thereof; then training an eye axis length prediction model; and then training a diopter prediction model by using the eye axis length information output by the eye axis length prediction model.
When in prediction, the future ocular axis length is predicted by the ocular axis length prediction model, and then the ocular axis length is input into the diopter prediction model to obtain the future diopter; or predicting the future axial length change value (the future axial length-the current axial length) by using the axial length prediction model, and then inputting the future axial length change value into the diopter prediction model to obtain the future diopter.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications therefrom are within the scope of the invention.
Claims (14)
1. An optical biological parameter prediction model training method is characterized by comprising the following steps:
acquiring sample data and label information thereof, wherein the sample data comprises optical biometric data and questionnaire survey data, the optical biometric data comprises time difference of two optical biometrics of the same measured person, time information and optical biometric parameters during the previous optical biometric measurement, and age, the questionnaire survey data comprises a plurality of auxiliary information capable of influencing eyes, and the label information comprises the optical biometric parameters during the next optical biometric measurement;
and training a prediction model by using a plurality of sample data and label information thereof so as to enable the prediction model to obtain predicted optical biological parameters according to the sample data, and optimizing self parameters according to the difference between the predicted optical biological parameters and the optical biological parameters in the label information.
2. The method of claim 1, wherein the obtaining sample data and tag information thereof comprises:
acquiring any two optical biological measurement record data of the same measured person from the plurality of optical biological measurement record data according to the identity information of the measured person;
and determining the time difference according to the optical biological measurement time information in the two optical biological measurement record data, and acquiring the former optical biological measurement time and the optical biological parameter corresponding to the latter optical biological measurement time.
3. The method according to claim 1 or 2, wherein the acquiring sample data and tag information thereof comprises:
obtaining a questionnaire filled by the testee according to the identity information of the testee, wherein the questionnaire comprises a plurality of questionnaire information capable of influencing eyes;
and converting the question-answer information into a numerical value which can be read by a prediction model.
4. The method of claim 1, prior to obtaining the sample data and its tag information, comprising:
obtaining a questionnaire filled by the testee according to the identity information of the testee, wherein the questionnaire comprises a plurality of questionnaire information capable of influencing eyes;
determining whether conflicts exist among the question answering information or not according to the question answering information and a preset question answering rule;
and eliminating the questionnaires with conflicts.
5. The method of claim 1, wherein the predictive model is a decision tree model.
6. The method of claim 5, wherein training a predictive model using a plurality of said sample data and its label information comprises:
determining a preset value combination of the hyperparameters of the decision tree model, wherein the hyperparameters comprise the number of decision trees and the number of the most used characteristics of each decision tree;
and training by utilizing the plurality of sample data and label information thereof under the condition that the decision model uses each combination respectively to obtain a model with optimal performance.
7. The method according to claim 5 or 6, wherein the plurality of sample data is divided into a plurality of shares, one of which is configured as a validation set for validating the performance of the predictive model, and the other of which is configured as a training set for training the predictive model.
8. The method of claim 1, wherein the sample data further comprises gender information, and the auxiliary information comprises parent vision information of the testee and eye condition information of the testee.
9. An optical biological parameter prediction method, comprising:
acquiring optical biometric data, expected time and questionnaire survey data of a user, wherein the optical biometric data comprises the age of the user, time information and optical biometric parameters during optical biometric measurement, and the questionnaire survey data comprises a plurality of auxiliary information capable of influencing eyes;
predicting an optical biometric parameter of the user at the desired time from the optical biometric data and questionnaire survey data using a prediction model.
10. The method of claim 9, wherein the obtaining questionnaire survey data comprises:
acquiring a questionnaire filled by a user, wherein the questionnaire comprises a plurality of questioning and answering information capable of influencing eyes;
and converting the question-answer information into a numerical value which can be read by a prediction model.
11. The method of claim 9, wherein the predictive model is a decision tree model.
12. The method of any one of claims 1-11, wherein the optical biological parameters include axial length of the eye, corneal thickness, anterior chamber depth.
13. An optical biological parameter predictive model training apparatus comprising at least one processor and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the optical biometric parameter prediction model training method of any one of claims 1-8.
14. An optical biometric prediction device comprising at least one processor and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the optical biometric parameter prediction method of any one of claims 9-12.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911319885.0A CN111128318A (en) | 2019-12-19 | 2019-12-19 | Optical biological parameter prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911319885.0A CN111128318A (en) | 2019-12-19 | 2019-12-19 | Optical biological parameter prediction method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111128318A true CN111128318A (en) | 2020-05-08 |
Family
ID=70500308
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911319885.0A Pending CN111128318A (en) | 2019-12-19 | 2019-12-19 | Optical biological parameter prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111128318A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040054358A1 (en) * | 2002-03-28 | 2004-03-18 | Cox Ian G. | System and method for predictive ophthalmic correction |
| CN107358036A (en) * | 2017-06-30 | 2017-11-17 | 北京机器之声科技有限公司 | A kind of child myopia Risk Forecast Method, apparatus and system |
| CN108364687A (en) * | 2018-01-10 | 2018-08-03 | 北京郁金香伙伴科技有限公司 | Eyeball trend prediction method and prediction model construction method and equipment |
-
2019
- 2019-12-19 CN CN201911319885.0A patent/CN111128318A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040054358A1 (en) * | 2002-03-28 | 2004-03-18 | Cox Ian G. | System and method for predictive ophthalmic correction |
| CN107358036A (en) * | 2017-06-30 | 2017-11-17 | 北京机器之声科技有限公司 | A kind of child myopia Risk Forecast Method, apparatus and system |
| CN108364687A (en) * | 2018-01-10 | 2018-08-03 | 北京郁金香伙伴科技有限公司 | Eyeball trend prediction method and prediction model construction method and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20200315518A1 (en) | Apparatus for processing data for predicting dementia through machine learning, method thereof, and recording medium storing the same | |
| Cao et al. | Evaluating the performance of various machine learning algorithms to detect subclinical keratoconus | |
| KR20170061222A (en) | The method for prediction health data value through generation of health data pattern and the apparatus thereof | |
| CN108780663A (en) | Digital personalized medicine platform and system | |
| JP2007531559A (en) | System and method for analyzing wavefront aberrations | |
| Mengoudi et al. | Augmenting dementia cognitive assessment with instruction-less eye-tracking tests | |
| CN112700858B (en) | Early warning method and device for myopia of children and teenagers | |
| CN108335757B (en) | Method for predicting diopter adjustment value in SMILE operation | |
| Lavric et al. | Detecting keratoconus from corneal imaging data using machine learning | |
| KR102320580B1 (en) | Myopia prediction method and system using deep learning | |
| CN113870239A (en) | Vision detection method and device, electronic equipment and storage medium | |
| David | Machine learning for the prelude diagnosis of dementia. | |
| Biondi et al. | Eye-Movement behavior identification for AD diagnosis | |
| CN114360728A (en) | Prediction model for mild cognitive dysfunction of diabetes and construction method of nomogram | |
| Castro-Luna et al. | Diagnosis of subclinical keratoconus based on machine learning techniques | |
| CN116563932A (en) | Eye image recognition method and related equipment based on multitask learning | |
| Musto et al. | A machine learning approach for predicting deterioration in Alzheimer’s disease | |
| Pham et al. | Multimodal deep learning model of predicting future visual field for glaucoma patients | |
| Asirvatham et al. | Hybrid deep learning network to classify eye diseases | |
| Kamran et al. | A comprehensive set of novel residual blocks for deep learning architectures for diagnosis of retinal diseases from optical coherence tomography images | |
| CN108109696B (en) | Data processing method and device | |
| CN111128318A (en) | Optical biological parameter prediction method and device | |
| CN107157498B (en) | Voice fatigue degree detection method for mental fatigue | |
| Finch | Recursive partitioning in the presence of multilevel data | |
| CN115223232A (en) | Eye health comprehensive management system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |