Myopia multi-parameter prediction method based on deep learning

Abstract

The invention discloses a myopia multi-parameter prediction method based on deep learning, and the method comprises the following steps: S1, obtaining a myopia screening data set based on deep neural network training, and carrying out the preprocessing of the data; s2, performing re-marshalling on the data; s3, establishing a myopia multi-parameter prediction model; s4, training and verifying the network model through a loss function; and S5, performing myopia multi-parameter prediction by using the model. The method has the beneficial effects that by establishing the multi-layer sensor with residual connection, myopia prediction is carried out on a plurality of parameters such as an eye axis, an equivalent spherical lens and an axis ratio at any moment in the future based on any two times of visual inspection data, so that the overall situation of myopia progress is reflected; meanwhile, the age to be predicted serves as input to be used for specifying the future time interval corresponding to the model prediction result, myopia progress prediction at any moment in the future is achieved, and the method has great significance in monitoring the myopia progress and guiding myopia prevention and control.

Description

Technical Field

[0001] The present invention relates to the technical field of myopia prediction, and in particular to a myopia multi-parameter prediction method based on deep learning. Background Art

[0002] Myopia, as an irreversible refractive error, seriously affects patients' daily lives and significantly increases their risk of suffering from various visual disorders. Accurate prediction of the future progression of myopia can help take appropriate intervention measures as early as possible to slow down the progression of myopia, which is of great significance for preventing the occurrence of high myopia or related complications. At present, many studies have been devoted to establishing prediction models for the progression of myopia, such as myopia prediction based on traditional machine learning methods such as random forests. However, due to the large number of influencing factors and individual differences in myopia progression, it is extremely challenging to predict the long-term progression of myopia in the future based on conventional optometry examination-related parameters. In recent years, with the development of deep learning technology and the accumulation of myopia screening data, myopia prediction based on deep learning is expected to achieve better performance.

[0003] At present, the more typical myopia prediction methods mainly include: myopia prediction based on traditional machine learning and myopia prediction methods based on deep learning. Typical traditional machine learning methods include using random forest algorithms to predict equivalent spherical power, see the literature "Lin, Haotian, et al. Prediction of myopia development among Chinese school-aged children using refraction data from electronic medical records: a retrospective, multicentre machine learning study. PLoS Medicine, 15.11 (2018): e1002674". This type of method has clear design logic and strong interpretability of the algorithm, but the effect is often not good enough when processing large-scale data, and the examination of equivalent spherical lenses is easily affected by the correction method, the state of the eyes during the examination, the detection equipment, etc. The single use of equivalent spherical lens parameters is not enough to fully predict the progression of myopia. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a myopia multi-parameter prediction method based on deep learning.

[0005] The purpose of the present invention is achieved by the following technical solution: A myopia multi-parameter prediction method based on deep learning, comprising the following steps:

[0006] S1: Obtain a myopia screening dataset based on deep neural network training and preprocess the data;

[0007] S2: reorganize the data;

[0008] S3: Establish a multi-parameter prediction model for myopia;

[0009] S4: Train and verify the network model through the loss function;

[0010] S5: Use the model for multi-parameter prediction of myopia.

[0011] Preferably, in step S1, the myopia screening data includes age, gender, spherical lens, cylindrical lens, eye axis, corneal curvature, and naked eye visual acuity.

[0012] Preferably, in step S1, preprocessing is to convert the spherical lens and the cylindrical lens to obtain equivalent spherical lens parameters, and to convert the axial line and the corneal curvature to obtain the axial ratio parameters. After preprocessing, the parameters of each data include age, gender, equivalent spherical lens, axial line, axial ratio, and naked eye visual acuity.

[0013] Preferably, the equivalent spherical lens parameter = spherical lens + 0.5*cylindrical lens.

[0014] Preferably, the axial ratio parameter is the eye axis divided by the average value of the corneal curvature K1 and the corneal curvature K2.

[0015] Preferably, in step S2, data with the same number of detections are grouped together.

[0016] Preferably, in step S3, the model includes an input layer, a hidden layer and an output layer, the hidden layer is composed of several fully connected layers and nonlinear mapping layers, and a forward residual connection is added between two fully connected layers.

[0017] Preferably, in step S3, the model uses the data of the two examinations before and after preprocessing in step S1 and the age to be predicted as input after further calculation, wherein the calculated data is composed of the annual growth rate of the myopia parameter obtained by subtracting the first examination data and the two examination data and dividing by the time interval between the two examinations, and the age to be predicted is used to specify the future time corresponding to the model prediction result.

[0018] Preferably, step S4 further includes the following steps:

[0019] S41: Divide the data set in step S2 into a training set and a validation set;

[0020] S42: Use one of the groups as a validation set and all the other groups as training sets, and use the training set data to train the network model;

[0021] S43: Repeat step S42 to obtain a training model corresponding to each group of verification sets, and complete model training;

[0022] S44: Test each validation set in turn using its corresponding training model to obtain myopia prediction results for all data.

[0023] Preferably, the loss function is calculated as:

[0024]

[0025] Among them, Y i is the vector composed of the real eye axis, equivalent spherical lens and axis ratio, Y' i is a vector consisting of the predicted axial length, equivalent spherical lens and axial ratio, and M is the number of samples.

[0026] The present invention has the following advantages: the present invention establishes a multi-layer perceptron with residual connections, and predicts myopia of multiple parameters such as the axial length, equivalent spherical lens, axial ratio ratio, etc. at any time in the future based on any two previous and subsequent eye examination data, thereby reflecting the overall situation of myopia progression. At the same time, the age to be predicted is used as an input to specify the future time interval corresponding to the model prediction result, thereby realizing the prediction of myopia progression at any time in the future, which is of great significance for monitoring myopia progression and guiding myopia prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the structure of the multi-parameter prediction model. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0029] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0031] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0032] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, or the positions or positional relationships in which the product of the invention is usually placed when in use, or the positions or positional relationships commonly understood by those skilled in the art, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0033] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0034] In this embodiment, if Figure 1 As shown, a myopia multi-parameter prediction method based on deep learning includes the following steps:

[0035] S1: Obtain a myopia screening dataset based on deep neural network training and preprocess the data;

[0036] S2: reorganize the data;

[0037] S3: Establish a multi-parameter prediction model for myopia;

[0038] S4: Train and verify the network model through the loss function;

[0039] S5: Use the model to predict myopia with multiple parameters. Specifically, based on the network model trained in step S4, the first examination data and the annual growth rate calculated based on the two examination data are input into the network in combination with the age to be predicted, and the model output is the multi-parameter prediction result of myopia corresponding to the given age to be tested. By establishing a multi-layer perceptron with residual connections, the myopia prediction of multiple parameters such as the axial length, equivalent spherical lens, axial ratio ratio, etc. at any time in the future is predicted based on any two previous and subsequent optometry examination data, thereby reflecting the overall situation of myopia progression. At the same time, the age to be predicted is used as an input to specify the future time interval corresponding to the model prediction result, so as to realize the prediction of myopia progression at any time in the future, which is of great significance for monitoring myopia progression and guiding myopia prevention and control.

[0040] Further, in step S1, the myopia screening data includes age, gender, spherical lens, cylindrical lens, eye axis, corneal curvature, and naked eye vision. Further, in step S1, the preprocessing is to convert the spherical lens and cylindrical lens to obtain equivalent spherical lens parameters, and convert the eye axis and corneal curvature to obtain the axial ratio parameter. After preprocessing, the parameters of each data include age, gender, equivalent spherical lens, eye axis, axial ratio, and naked eye vision. In this embodiment, the data is timed every six months, and each data contains at least two inspection records, wherein the equivalent spherical lens parameter = spherical lens + 0.5* cylindrical lens, and the axial ratio parameter is the eye axis divided by the average value of corneal curvature K1 and corneal curvature K2, that is, the axial ratio ratio = eye axis / (0.5*corneal curvature K1+0.5*corneal curvature K2).

[0041] In this embodiment, in step S2, data with the same number of tests are grouped together. Specifically, since in deep learning, data with different numbers of tests cannot be put into the same batch to train the network model, data with the same number of tests are grouped together here to ensure that the number of inspection records in the same group of data is consistent, which is convenient for subsequent model training. In this embodiment, all data are continuously screened for 1.5 to 4.5 years, with half a year as the minimum time interval, so that each has 3-9 inspection records, and the test data with the same number of records are divided into a group in turn, that is, divided into 7 groups, so that the number of inspection records for each patient in the same group of data is consistent.

[0042] Furthermore, in step S3, the model includes an input layer, a hidden layer and an output layer, the hidden layer is composed of several fully connected layers and nonlinear mapping layers, and a forward residual connection is added between two fully connected layers. Specifically, a 77-layer perceptron (MLP) network model is established, including an input layer, a hidden layer and an output layer; wherein the hidden layer is composed of 75 fully connected layers and a nonlinear mapping layer, the number of neurons in the fully connected layer is 512, and the nonlinear mapping layer uses the LeaklyReLU function, and then a forward residual connection is added between each two consecutive fully connected layers in the hidden layer, and finally a multilayer perceptron network model containing residual connections is formed, as shown in FIG. Figure 1 As shown. Further, the model takes the two inspection data before and after the preprocessing of step S1 and the age to be predicted as input after further calculation, wherein the calculated data is composed of the annual growth rate of myopia parameters obtained by subtracting the first inspection data and the two inspection data and then dividing by the time interval between the two inspections, and the age to be predicted is used to specify the future time corresponding to the model prediction result, and the model finally outputs myopia multi-parameters including the eye axis, equivalent spherical lens, and axis rate ratio prediction values ​​corresponding to the age to be predicted. Specifically, the network model takes any two inspection data and the age to be predicted as model input, wherein any two inspection data are further converted to obtain input data, including primary inspection age, primary inspection equivalent spherical lens, primary inspection eye axis, primary inspection axis rate ratio, primary inspection naked eye vision, equivalent spherical lens change rate, eye axis annual change rate, axis rate ratio annual change rate, naked eye vision annual change rate, gender, and age to be predicted, a total of 11 input parameters; the model output is the eye axis, equivalent spherical lens, and axis rate ratio prediction values ​​corresponding to the age to be predicted, a total of 3 output parameters. Therefore, the number of neurons in the input layer of the network is 11, and the number of neurons in the output layer is 3.

[0043] In this embodiment, step S4 further includes the following steps:

[0044] S41: Divide the data set in step S2 into a training set and a validation set;

[0045] S42: Use one of the groups as a validation set and all the other groups as training sets, and use the training set data to train the network model;

[0046] S43: Repeat step S42 to obtain a training model corresponding to each group of verification sets, and complete model training;

[0047] S44: Test each validation set in turn using its corresponding training model to obtain myopia prediction results for all data. Further, the loss function is calculated as:

[0048]

[0049] Among them, Y iis the vector composed of the real eye axis, equivalent spherical lens and axis ratio, Y' i is the vector composed of the predicted eye axis, equivalent spherical lens and axis rate ratio, and M is the number of samples. Specifically, the loss function, optimizer, learning rate, and learning iteration number are set, wherein the optimizer adopts the Adam optimizer, the initial learning rate is 0.0001, and the learning rate after training after each iteration of 100 rounds is reduced by 10 times. After 300 rounds of iterative training, it is considered that the neural network model parameters have been trained. Therefore, the data is divided into 7 groups, so the cross-validation mode is adopted in the embodiment to divide the training set and the test set, and the model is trained and verified, that is, firstly, the first group is used as the verification set, the 2nd to 7th groups are used as the training set, and the network model is trained using the training set data, and the above process is repeated, and each group is used as the verification set in turn, and the remaining all groups are used as the training set to train the network model, and 7 corresponding training models for each group as the verification set are obtained, and the whole training process of the model is completed; after the training is completed, the test phase is carried out, and each group of verification sets is tested in turn using the corresponding training model, and the myopia multi-parameter prediction results of all data are finally obtained.

[0050] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A myopia multi-parameter prediction method based on deep learning, characterized by: The following steps are involved: S1: Obtain a myopia screening dataset based on deep neural network training and preprocess the data; S2: reorganize the data; S3: Establish a multi-parameter prediction model for myopia; S4: Train and verify the network model through the loss function; S5: Use the model for multi-parameter prediction of myopia.

2. The myopia multi-parameter prediction method based on deep learning according to claim 1, characterized in that: In step S1, the myopia screening data includes age, gender, spherical lens, cylindrical lens, eye axis, corneal curvature, and naked eye visual acuity.

3. The myopia multi-parameter prediction method based on deep learning according to claim 2, characterized in that: In step S1, the preprocessing is to convert the spherical lens and the cylindrical lens to obtain the equivalent spherical lens parameters, and to convert the eye axis and the corneal curvature to obtain the axial ratio parameters. After preprocessing, the parameters of each data include age, gender, equivalent spherical lens, eye axis, axial ratio, and naked eye visual acuity.

4. The myopia multi-parameter prediction method based on deep learning according to claim 3, characterized in that: Equivalent spherical lens parameters = spherical lens + 0.5*cylindrical lens.

5. The myopia multi-parameter prediction method based on deep learning according to claim 3, characterized in that: The axial ratio parameter is the axial length divided by the average of the corneal curvature K1 and the corneal curvature K2.

6. The myopia multi-parameter prediction method based on deep learning according to claim 1, characterized in that: In step S2, data with the same number of detections are grouped together.

7. The myopia multi-parameter prediction method based on deep learning according to claim 6, characterized in that: In step S3, the model includes an input layer, a hidden layer and an output layer, the hidden layer is composed of a plurality of fully connected layers and a nonlinear mapping layer, and a forward residual connection is added between two of the fully connected layers.

8. The myopia multi-parameter prediction method based on deep learning according to claim 7, characterized in that: In step S3, the model uses the data of the two examinations before and after preprocessing in step S1 and the age to be predicted after further calculation as input, wherein the calculated data is composed of the annual growth rate of the myopia parameter obtained by subtracting the data of the first examination and the two examinations and dividing it by the time interval between the two examinations, and the age to be predicted is used to specify the future time corresponding to the model prediction result.

9. The myopia multi-parameter prediction method based on deep learning according to claim 8, characterized in that: The step S4 further includes the following steps: S41: Divide the data set in step S2 into a training set and a validation set; S42: Use one of the groups as a validation set and all the other groups as training sets, and use the training set data to train the network model; S43: Repeat step S42 to obtain a training model corresponding to each group of verification sets, and complete model training; S44: Test each validation set in turn using its corresponding training model to obtain myopia prediction results for all data.

10. The myopia multi-parameter prediction method based on deep learning according to claim 9, characterized in that: The loss function is calculated as: Among them, Y i is the vector composed of the real eye axis, equivalent spherical lens and axis ratio, Y' i is a vector consisting of the predicted axial length, equivalent spherical lens and axial ratio, and M is the number of samples.

Images