CN109300548A - A kind of optimization method and system for predicting diopter adjusted value in SMILE refractive surgery - Google Patents
A kind of optimization method and system for predicting diopter adjusted value in SMILE refractive surgery Download PDFInfo
- Publication number
- CN109300548A CN109300548A CN201810890395.5A CN201810890395A CN109300548A CN 109300548 A CN109300548 A CN 109300548A CN 201810890395 A CN201810890395 A CN 201810890395A CN 109300548 A CN109300548 A CN 109300548A
- Authority
- CN
- China
- Prior art keywords
- value
- nomogram
- prediction
- training sample
- postoperative
- 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.)
- Granted
Links
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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- Artificial Intelligence (AREA)
- Pathology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses the optimization methods and system of diopter adjusted value in a kind of prediction SMILE refractive surgery.Wherein, this method comprises the following steps: using preferred machine learning algorithm, using the preoperative parameter of training sample and its postoperative optometry degree as training attribute, training generates prediction model;Ideal value is set as using postoperative optometry degree as objective attribute target attribute for new case to be predicted, cooperates with the preoperative parameter of the new case to be predicted as input, applied forecasting model generates Nomogram predicted value.This method can effectively promote the predictablity rate of existing Nomogram predicted value, can reduce dependence of the preoperative solution formulation process to expert using this method, reduce the professional threshold of preoperative solution formulation, further promote the rate of precision of preoperative solution formulation.
Description
Technical field
The present invention relates to it is a kind of prediction SMILE (full femtosecond laser) refractive surgery in diopter adjusted value optimization method,
More particularly to a kind of optimization method based on diopter adjusted value in machine learning method prediction SMILE refractive surgery, it relates to simultaneously
And the system for realizing the optimization method.
Background technique
The problem of myopia has been increasingly becoming a globalization, investigation display, the whole world has more than 1,500,000,000 people and suffers from
Myopia, the adolescent myopia illness rate between China, 15-25 years old age are 75% or more.Treatment myopia, long sight and astigmatism etc.
Ametropic cornea refractive surgery (excimer laser surgery, femtosecond laser operation) is this century newest common ophthalmological operation
One of.SMILE refractive surgery be most it is newly developed it is a kind of treat ametropic surgical technic, effect and safety are more previous
Surgical technic have very big promotion.
Refractive surgery guarantee safety, validity on the basis of, in order to reach optimal vision correction effect, still have
Some problems demands solve, such as the evaluation of personalized operation decision, the operation designing of precision, surgical effect and operation
Effect prediction etc..Relatively uniform operation screening, operation designing and therapeutic scheme is not suitable for all people's eye individual;Meanwhile
Corneal surface shape parameter is numerous and jumbled, and experience levels different to the sensibility of laser and clinician are different in addition, is also extremely difficult to
The standard of decision, consistency.However, utilizing machine learning techniques pair with the continuous expansion of current refractive surgery data volume
Surgical data carries out forecast analysis, starts to be possibly realized to obtain personalized, precision operation plan.This will assist doctor
Careful design operation plan is obviously improved the postoperative curative effect of patient.Currently, SMILE refractive surgery is based primarily upon doctor's priori hand
Art experience comprehensively considers the factors such as diopter of correction, cylindrical mirror degree, age and carries out important surgical parameters setting, wherein Nomogram value
It sets vital for the curative effect of refractive surgery.
Currently, existing research proposition analyzes the Nomogram value in refractive surgery scheme using machine learning techniques
Prediction.For example, Multi-regression method utilizes multiple linear regression analysis method, study in LASIK (Laser Assisted
In Situ Keratomileusis, Laser in Situ Keratomileusis) in operation plan formulation process, Nomogram value
With the linear relationship of preoperative parameters.But the principle and lasik surgery of SMILE operation are different, it is demonstrated experimentally that
In SMILE operation, linear relationship is not present between Nomogram value and preoperative parameter, existing method is not suitable for the operation side SMILE
The formulation of case.
It is confirmed by research and experiment, the reinforcing classifier and neural network algorithm in machine learning are in SMILE operation
The prediction of Nomogram value has higher accuracy, has reached clinical application acceptable standard.However, there are still many problems still
It is to be optimized, to further increase the accuracy of Nomogram value prediction in SMILE operation.
Summary of the invention
In view of the deficiencies of the prior art, primary technical problem to be solved by this invention is to provide a kind of prediction SMILE
The optimization method of diopter adjusted value in refractive surgery.
Another technical problem to be solved by this invention provides diopter adjusted value in a kind of prediction SMILE refractive surgery
Optimization system.
For achieving the above object, the present invention uses following technical solutions:
According to a first aspect of the embodiments of the present invention, diopter adjusted value in a kind of prediction SMILE refractive surgery is provided
Optimization method includes the following steps:
Using preferred machine learning algorithm, using the preoperative parameter of training sample and its postoperative optometry degree as training attribute,
Training generates prediction model;
Ideal value is set as using postoperative optometry degree as objective attribute target attribute for new case to be predicted, cooperates with this to be predicted new
The preoperative parameter of case generates Nomogram predicted value as input, applied forecasting model.
Wherein more preferably, the optimization method for predicting diopter adjusted value in SMILE refractive surgery can also include such as
Lower step:
Based on the Nomogram predicted value that preferred machine learning algorithm generates, deflection forecast model is constructed to Nomogram value
Prediction deviation predicted, and correct Nomogram predicted value.
Wherein more preferably, the Nomogram predicted value generated based on preferred machine learning algorithm, constructs deflection forecast
Model predicts the prediction deviation of Nomogram value, and corrects Nomogram predicted value, includes the following steps:
Using preferred machine learning algorithm, the attribute based on training sample, the prediction model of building prediction Nomogram value;
Training sample applied forecasting model is predicted, Nomogram predicted value is obtained, and calculates Nomogram prediction
Deviation between value and the Nomogram value of doctor's setting;
Using preferred machine learning algorithm, it is based on training sample attribute and Nomogram predicted value, constructs prediction deviation
Deflection forecast model;
For new case to be predicted, successively it is raw to generate Nomogram predicted value, deflection forecast model for applied forecasting model
The Nomogram predicted value is corrected at its prediction deviation, and with the prediction deviation, obtains revised Nomogram prediction
Value.
Wherein more preferably, the optimization method for predicting diopter adjusted value in SMILE refractive surgery can also include such as
Lower step:
Based on the Nomogram value that training sample attribute and doctor are set, postoperative optometry degree prediction model is constructed;It will
Nomogram initial prediction and its k nearest neighbor Nomogram value input postoperative optometry degree prediction model, and selection wherein causes best
The Nomogram value of postoperative optometry degree is as final optimization pass result.
Wherein more preferably, the Nomogram value set based on training sample attribute and doctor, constructs postoperative optometry degree
Prediction model;Nomogram initial prediction and its k nearest neighbor Nomogram value are inputted into postoperative optometry degree prediction model, select it
In cause the Nomogram value of best postoperative optometry degree as final optimization pass result, include the following steps:
Using preferred machine learning algorithm, the attribute based on training sample, the prediction model of building prediction Nomogram value;
Using preferred machine learning algorithm, the Nomogram value of attribute and doctor's setting based on training sample, building is in advance
Survey the postoperative optometry degree prediction model of postoperative optometry degree;
For new case to be predicted, applied forecasting model generates its Nomogram initial prediction, and sets step-length, generates
Its k nearest neighbor set;
Respectively by the Nomogram initial prediction of new case to be predicted and the k nearest neighbor set of generation, the new disease is cooperateed with
Other attributes of example input postoperative optometry degree prediction model, generate corresponding postoperative optometry degree, and selection causes postoperative optometry degree exhausted
To the smallest Nomogram value of value as final predicted value.
Wherein more preferably, the optimization method for predicting diopter adjusted value in SMILE refractive surgery, uses preferred machine
Learning algorithm is trained training sample, before generating prediction model, can also include the following steps:
The initial training sample of acquisition is expanded, increases the quantity of minority's training sample in initial training sample, obtains
To training sample.
Wherein more preferably, the initial training sample of described pair of acquisition expands, and increases minority's instruction in initial training sample
The quantity for practicing sample, obtains training sample, includes the following steps:
S111 calculates in initial training sample each attribute to the information gain of Nomogram value, determines invariable attribute and can
Become attribute;
S112, for any one minority's sample, it is nearest that the Euclidean distance based on global property screens minority's sample
Adjacent similar sample;
S113 generates new minority's training sample based on minority's sample, wherein new minority's training sample
Invariable attribute value is identical as minority's sample, and the variable attribute value of new minority's training sample is minority's sample
The random median of this sample respective attributes similar with the arest neighbors;
S114 repeats step S112~S113, until training sample is reached based on the distributed number of different Nomogram values
It is balanced.
According to a second aspect of the embodiments of the present invention, diopter adjusted value in a kind of prediction SMILE refractive surgery is provided
The optimization system of optimization method, including processor and memory;Being stored on the memory can be used transports on the processor
Capable computer program realizes following steps when the computer program is executed by the processor:
Using preferred machine learning algorithm, using the preoperative parameter of training sample and its postoperative optometry degree as training attribute,
Training generates prediction model;
Ideal value is set as using postoperative optometry degree as objective attribute target attribute for new case to be predicted, cooperates with this to be predicted new
The preoperative parameter of case generates Nomogram predicted value as input, applied forecasting model.
Wherein more preferably, following steps can also be realized when the computer program is executed by the processor:
Based on the Nomogram predicted value that preferred machine learning algorithm generates, deflection forecast model is constructed to Nomogram value
Prediction deviation predicted, and correct Nomogram predicted value.
Wherein more preferably, when the computer program is executed by the processor, following steps can also be realized:
Based on the Nomogram value that training sample attribute and doctor are set, postoperative optometry degree prediction model is constructed;It will
Nomogram initial prediction and its k nearest neighbor Nomogram value input postoperative optometry degree prediction model, and selection wherein causes best
The Nomogram value of postoperative optometry degree is as final optimization pass result.
The optimization method of diopter adjusted value in prediction SMILE refractive surgery provided by the present invention, is based on machine existing
In device study prediction SMILE operation on the basis of the method for Nomogram value, by the distribution of balance training sample data, reduces and predict
Deviation makes full use of the postoperative degree constrained optimization prediction model of optometry in three months, proposes four kinds of prioritization schemes: sample balance optimizing,
Deviation optimization, k nearest neighbor optimization, objective attribute target attribute optimization;It efficiently solves since training sample is distributed uneven, prediction deviation is big etc.
The not accurate problem of prediction that factor causes, effectively improves the precision of existing machine learning prediction model.
Detailed description of the invention
Fig. 1 is the flow chart of the optimization method of diopter adjusted value in prediction SMILE refractive surgery provided by the present invention;
Fig. 2 is to expand the initial training sample of acquisition in embodiment provided by the present invention, obtain training sample
Flow chart;
Fig. 3 is in embodiment provided by the present invention, using showing for multi-layered perception neural networks model data propagated forward
It is intended to;
Fig. 4 is in embodiment provided by the present invention, using showing for multi-layered perception neural networks model error backpropagation
It is intended to;
Fig. 5 is to optimize the flow chart optimized to Nomogram value using objective attribute target attribute in embodiment provided by the present invention;
Fig. 6 is to be optimized using deviation to the modified flow chart of Nomogram value in embodiment provided by the present invention;
Fig. 7 is to optimize the flow chart optimized to Nomogram value using k nearest neighbor in embodiment provided by the present invention;
Fig. 8 is the configuration diagram of overall plan in embodiment provided by the present invention;
Fig. 9 is that the structure of the optimization system of diopter adjusted value in prediction SMILE refractive surgery provided by the present invention is shown
It is intended to.
Specific embodiment
Detailed specific description is carried out to technology contents of the invention in the following with reference to the drawings and specific embodiments.
In currently available training operation case, the growing number of different Nomogram (diopter adjusted value) values
Often different, this is usually determined by the actual distribution situation of case, for example, high degree Nomogram case quantity compared with minuent
Several case quantity is often deficient.However, the distribution imbalance of training data, which normally results in learning model, tends to fitting greatly
The case of many classifications causes the predictablity rate of minority's case not high.Therefore, carrying out distribution equilibrium optimization to training data will have
Conducive to the existing prediction result of promotion.In addition to this, the prediction result of machine learning algorithm unavoidably with the actual set of doctor
There are certain deviation, design optimization algorithm reduces this deviation and will be helpful to be fitted doctor's experience to greatest extent value, realizes accurate
Prediction effect.In addition, postoperative three months optometry degree are important surgical effect evaluation index, in existing research only as
The screening index of training cases is only screened ideal case and is trained to prediction model, to prediction model study to ideal
Operation case;In this way, training cases quantity (undesirable operation case will not be used as model training) is not only greatly reduced,
Easily cause prediction model poor fitting, but can not constrained forecast model make it avoid generating the prediction for leading to undesirable postoperative effect
Value.Therefore, this important surgical effect evaluation index of postoperative three months optometry degree how is made full use of, prediction model is carried out about
Beam and optimization will bring further promotion for the prediction precision of machine learning model.
When solving the problems, such as that training sample distribution is uneven and lacks, SMOTE (Synthetic Minority Over-
Sampling Technique) algorithm as a kind of more effective sample expands algorithm, pass through simulation minority's sample ex hoc genus anne
Each attribute value of other nearest samples generates completely new minority's sample.In the embodiment provided by originally issuing a statement, to guarantee
Newly generated minority's sample meets doctor and rule is arranged, and will only be based on to the lesser sample attribute of Nomogram value influence power
The fine tuning of SMOTE method.
In addition, neural network algorithm as a kind of machine learning method, is also widely used for building classifier in recent years
Prediction model.Neural network is a kind of nonlinear model of deep layer, is made of multilayered nonlinear unit, in enough training datas
Under support, arbitrary function can be theoretically approached, and the learning parameter of its network number of plies, the number of nodes of each layer and network all may be used
With according to circumstances specific setting, there is very strong flexibility and ability to express, can be used for solving numerous classification and regression problem.
Neural network algorithm is trained by reverse propagated error, by constantly comparing the prediction result of network and actual value
Compared with error being traveled in each layer of network from back to front, adjusts weight by gradient descent method to optimize the property of network
Can, until output error is less than the standard value of setting.It, will be using neural network algorithm as preferred in invention method
Machine learning algorithm carry out prediction model building, however, the optimization method proposed is suitable for other machines learning classification simultaneously
And regression forecasting algorithm, it is not limited in optimizing neural network algorithm.
In conclusion the current existing machine learning method predicted Nomogram value in SMILE operation plan
Shang Youke optimizes space.The distribution of balance training sample data reduces prediction deviation, postoperative three months optometry degree is made full use of to constrain
Optimal prediction model will effectively promote the existing precision based on Nomogram value in machine learning prediction SMILE operation plan.
As shown in Figure 1, the optimization method provided by the present invention for predicting diopter adjusted value in SMILE refractive surgery, packet
It includes following steps: using preferred machine learning algorithm, belong to using the preoperative parameter of training sample and its postoperative optometry degree as training
Property, training generates prediction model;Ideal value is set as using postoperative optometry degree as objective attribute target attribute for new case to be predicted, is assisted
For preoperative parameter with the new case to be predicted as input, applied forecasting model generates Nomogram predicted value.Can also include
Following steps: the Nomogram predicted value generated based on preferred machine learning algorithm constructs deflection forecast model to Nomogram
The prediction deviation of value is predicted, and corrects Nomogram predicted value.It can also include the following steps: to be based on training sample attribute
And the Nomogram value of doctor's setting, postoperative optometry degree prediction model is constructed, by Nomogram initial prediction and its k nearest neighbor
Nomogram value inputs postoperative optometry degree prediction model, selects the Nomogram value for wherein leading to best postoperative optometry degree as most
Whole optimum results;Wherein, Nomogram initial prediction can be the Nomogram predicted value of prediction model generation, be also possible to
Deflection forecast model generate repair after Nomogram predicted value.Detailed specific description is done to this process below.
Attribute and its art in embodiment provided by the present invention, using preferred machine learning algorithm based on training sample
Posteriority luminosity is trained training sample, before generating prediction model, can also include the following steps:
S11 expands the initial training sample of acquisition, increases the number of minority's training sample in initial training sample
Amount, obtains the training sample of the distributed number equilibrium of Nomogram value.
According to the diopter of postoperative predetermined time, screening and feature normalization are carried out to SMILE operation case historical data
Deng pretreatment, initial training sample is obtained.Wherein, carrying out the pretreated methods such as screening and feature normalization to historical data can
Think existing arbitrary processing method, it is not limited here.The initial training sample can be existing for SMILE dioptric hand
Diopter adjusted value predicts any training sample used in art.
In based on machine learning algorithm building SMILE refractive surgery before Nomogram value prediction model, to initial training
Sample carries out the processing of sample balance optimizing, i.e., is expanded using SMOTE algorithm initial training sample, increases minority's training sample
This quantity (using Nomogram difference value as tag along sort), reaches training sample based on the distributed number of different Nomogram values
To equilibrium.As shown in Fig. 2, specifically comprising the following steps:
S111 calculates in initial training sample each attribute to the information gain of Nomogram value, determines invariable attribute and can
Become attribute.
It calculates each attribute in initial training sample and the biggish category of information gain value is selected to the information gain of Nomogram value
Property, i.e., the attribute being affected to Nomogram value, as invariable attribute.It, can be in embodiment provided by the present invention
Be arranged gain threshold, information gain be greater than gain threshold attribute be invariable attribute, comprising: the age, gender, eye not, diopter of correction,
Cylindrical mirror degree.Remaining attribute is as variable attribute, comprising: preoperative uncorrected visual acuity, is most preferably corrected defects of vision, corneal diameter, light at astigmatism axis
School district, corneal curvature, corneal central thickness etc..
Wherein, it is as follows to calculate formula used in information gain:
Entropy(SA) be node (attribute) A entropy, Gain (SA, A) and it is information gain at node A.In formula
Each variable meaning is as follows: NAIndicate the sample set on node A,Expression belongs to classification C in whole samples of node Ai
Sample size, node A shares M attribute, NAmIndicate the sample size in node A selection attribute m,Expression is selecting
Belong to classification C in the sample of attribute miSample size.
In based on preferred machine learning algorithm building SMILE refractive surgery before Nomogram value prediction model, use
SMOTE algorithm expands training sample, increases minority's training samples number (using Nomogram difference value as contingency table
Label), so that training sample is reached balanced based on the distributed number of different Nomogram values, it is accurate with the prediction of Optimization Learning model
Degree can effectively solve due to the uneven caused not accurate problem of prediction of training sample distribution.
S112, for any minority's sample D (D belongs to minority sample set S), the Euclidean distance based on global property screens it
The similar sample B of arest neighbors;That is D is identical as the Nomogram value of B, and other attribute values are most like;
S113 generates new minority's training sample D based on Dnew, wherein DnewInvariable attribute value it is identical as D, Dnew's
Variable attribute value is the random median (numeric type) of D and B respective attributes;The respective attributes value for assuming sample D and B is a
And b, then DnewAttribute value be (1-r) × a+r × b, wherein r ∈ [0,1] be random number;
S114 repeats step S112~S113, until training sample is reached based on the distributed number of different Nomogram values
It is balanced.In embodiment provided by the present invention, using Nomogram difference value as tag along sort.
S12 is belonged to using preferred machine learning algorithm using the preoperative parameter of training sample and its postoperative optometry degree as training
Property, training generates prediction model.
Training sample is trained using preferred machine learning algorithm, in embodiment provided by the present invention, preferably
Machine learning algorithm can be the classifier algorithm in existing arbitrary machine learning.Such as: Softmax regression algorithm,
REPTree decision Tree algorithms, Xgboost algorithm (Chen, T.&Guestrin, C.XGBoost:a scalable tree
Boosting system.KDD ' 16, August 13-17,2016, San Francisco, CA, USA), multi-layered perceptron neural net
Network algorithm etc..
In embodiment provided by the present invention, predict in SMILE refractive surgery in the optimization process of diopter adjusted value,
Constructed respectively using multi-layered perception neural networks algorithm as preferred machine learning algorithm for required all kinds of prediction models into
Row explanation.Multi-layered perception neural networks model is trained using training sample, generates prediction model.
As a kind of machine learning method of classics, multi-layered perception neural networks algorithm answering in continuous type numerical prediction
Preferable effect is obtained with middle.Wherein, Multilayer Perceptron algorithm is a kind of construction BP network model
Multiple data sets of input can be mapped on the data set of single output by algorithm.It is wrapped in multi-layered perception neural networks
Contain multilayer node, is equipped with weight in the connection of the node of adjacent layer, the destination of study is the side distribution for these connections
Correct weight.Back-propagation algorithm is compared by constantly exporting network with desired output, and error propagation is returned
Upper one layer, the weight on side is adjusted from bottom to top, and is repeated continuously, until output error is less than the standard formulated.
In embodiment provided by the present invention, preoperative parameter using training sample based on training sample and its postoperative test
Luminosity is trained multi-layered perception neural networks model, generates the prediction model of Nomogram, and the algorithm used is
Multilayer Perceptron, specifically comprises the following steps:
Place is normalized as training attribute in S121, preoperative parameter and its postoperative optometry degree to training sample respectively
Reason generates normalization characteristic value, is input to normalizing characteristic value as input data in multi-layered perception neural networks model, at random
Distribute the weight in multi-layered perception neural networks on each side.
The characteristic value (preoperative parameter and its postoperative optometry degree including training sample) in training sample is extracted, and by feature
Value normalizes in [0,1] section.Wherein, the preoperative parameter of training sample includes diopter of correction, cylindrical mirror degree, naked eye optometry degree, year
Multiple possible correlative factors for influencing diopter adjusted value such as age.The information of characteristic value can directly be obtained by preoperative planning information
It takes.In embodiment provided by the present invention, each category feature value is normalized can be acquired using the calculating of following formula:
Normalizing characteristic value=(former characteristic value-feature minimum value)/(profile maxima-feature minimum value).
Profile maxima is the maximum value of value in same class characteristic value, and feature minimum value is value in same class characteristic value
The smallest value.Such as in age categories, value range is 0~100, and wherein the profile maxima of age categories is 100, feature
Minimum value is 0.
S122 calculates the output of next layer of neuron according to the data of the weight on each side and input layer in sequence, and
The output of output layer neuron is obtained as a result, i.e. propagated forward.
As shown in figure 3, the specific implementation step of the propagated forward used are as follows: according to top-down sequence, by upper one layer
The output of neuron is denoted as x, and the weight on side is denoted as W between adjacent layer, and bias is denoted as b.It is opened using following formula from input layer
Begin the output h for constantly calculating next layer of neuronw,b(x), until obtaining the output result of output layer neuron:
Wherein, f (z) is neuron activation functions,I indicates i-th of nerve of same layer neuron
Member.
S123 calculates the overall error of output node, and these errors is propagated back to network, such as Fig. 4 with back-propagation algorithm
It is shown, to calculate gradient;
S124, using all weights in gradient descent algorithm adjustment network, to reduce the error of output layer;Wherein,
Gradient descent algorithm is this field conventional treatment method, is just repeated no more herein.
S125 repeats S122~S124, until output layer error is less than given standard error, generation prediction model.
In embodiment provided by the present invention, Multilayer Perceptron algorithm is modeled, the parameter being related to
Setting such as the following table 1:
Table is arranged in 1 Multilayer Perceptron algorithm parameter of table
S13 is set as ideal value 0 using postoperative optometry degree as objective attribute target attribute for new case to be predicted, to constrain
Prediction model generates the Nomogram predicted value for leading to ideal surgical effect, cooperates with the preoperative parameter conduct of the new case to be predicted
Input, applied forecasting model generate Nomogram predicted value.
The program using postoperative optometry degree as target variable, take this can by undesirable operation case (postoperative optometry diopter is exhausted
Value is greater than and 0.5) safely introduces training sample, enriches the type and quantity (such as the postoperative optometry of certain case of training sample
Degree is -0.25, then it is believed that the target of the operation plan is that the aftertreatment effect of patient is adjusted to -0.25);Meanwhile
During prediction, it is set as 0 degree of ideal value using postoperative optometry degree as objective attribute target attribute, energy operative constraint prediction model avoids giving birth to
At the Nomogram predicted value for leading to undesirable surgical effect, postoperative effect is promoted.I.e. using objective attribute target attribute optimization pair
Nomogram predicted value optimizes, as shown in Figure 5.
In embodiment provided by the present invention, predict to obtain using the prediction model that preferred machine learning algorithm generates
After Nomogram predicted value, it can also include the following steps:
S14 constructs deflection forecast model pair based on the Nomogram predicted value that preferred machine learning algorithm generates
The prediction deviation of Nomogram value is predicted, and corrects Nomogram predicted value.
Based on the Nomogram predicted value that preferred machine learning algorithm generates, learning model is constructed to Nomogram value
Prediction deviation is predicted, and passes through this drift correction Nomogram predicted value.Nomogram is predicted using deviation optimization
Value optimizes, as shown in fig. 6, specifically comprising the following steps:
S141, using preferred machine learning algorithm, the attribute based on training sample, the prediction of building prediction Nomogram value
Model M1;Wherein construct prediction model M1Process training sample is instructed using preferred machine learning algorithm with step S12
The process for practicing generation prediction model is identical, just repeats no more herein.
S142, to training sample applied forecasting model M1It is predicted, obtains Nomogram predicted valueAnd calculate its with
The Nomogram value n of doctor's setting0Between deviation nr, it is denoted as
S143 is based on training sample attribute and Nomogram predicted value using preferred machine learning algorithmBuilding prediction
Deviation nrDeflection forecast model M2;Wherein, deflection forecast model M is constructed2Process and step S12 in use preferred engineering
It is mutually similar to practise the process that algorithm is trained generation prediction model to training sample, just repeats no more herein.
S144, for new case to be predicted, successively applied forecasting model M1Generate its Nomogram predicted valueDeviation
Prediction model M2Generate its prediction deviationIt is used in combinationAmendmentRevised Nomogram predicted value is obtained, is denoted as
The optimization of deviation that this method uses, be on the basis of existing optimization machine learning algorithm prediction Nomogram value,
The difference (deviation) between the Nomogram predicted value and the Nomogram value of doctor's setting is calculated, learning model is further constructed
(deflection forecast model) predicts the deviation, and the drift correction Nomogram predicted value as obtained by prediction.The program
The prediction deviation of existing machine learning algorithm can be effectively reduced, the prediction precision of learning model is further promoted.
In embodiment provided by the present invention, predict to obtain using the prediction model that preferred machine learning algorithm generates
After Nomogram predicted value, it can also include the following steps:
S15 constructs postoperative optometry degree prediction model based on the Nomogram value that training sample attribute and doctor are set, will
Nomogram initial prediction and its k nearest neighbor Nomogram value input postoperative optometry degree prediction model, and selection wherein causes best
The Nomogram value of postoperative optometry degree is as final optimization pass result;Wherein, Nomogram initial prediction can be prediction model
The Nomogram predicted value of generation is also possible to the revised Nomogram predicted value of deflection forecast model generation.
Using this important feedback information of postoperative optometry degree, the Nomogram set based on training sample attribute and doctor
Value, constructs postoperative optometry degree prediction model, and Nomogram initial prediction and its k nearest neighbor Nomogram value are inputted postoperative optometry
Prediction model is spent, selects the Nomogram value for wherein leading to best postoperative optometry degree as final optimization pass result.As shown in fig. 7,
Specifically comprise the following steps:
S151, using preferred machine learning algorithm, the attribute based on training sample, the prediction of building prediction Nomogram value
Model M1;
S152, using preferred machine learning algorithm, the Nomogram value of attribute and doctor's setting based on training sample, structure
Build the postoperative optometry degree prediction model M of predicting surgical posteriority luminosity3;Wherein, postoperative optometry degree is postoperative three months diopter of correction
SDAfter.The Nomogram value of attribute and doctor's setting based on training sample, constructs the postoperative optometry degree of predicting surgical posteriority luminosity
Prediction model M3Process and step S12 in using preferred machine learning algorithm generation prediction model is trained to training sample
Process it is mutually similar, just repeat no more herein.
S153, for new case to be predicted, applied forecasting model M1Generate its Nomogram initial predictionAnd it sets
Step-length generates its k nearest neighbor set.
It is step-length with 0.05 in embodiment provided by the present invention, generates its k nearest neighbor setWherein
Preferably, K=2 is set, i.e.,I.e. the Nomogram value as alternative shares 3:
Nomogram initial prediction
S154, respectively by the Nomogram initial prediction of new case to be predictedAnd its k nearest neighbor set N, cooperate with the case
Other attributes (preoperative parameter) input postoperative optometry degree prediction model M3, generate corresponding postoperative optometry degree SDAfter, selection causes
Postoperative optometry degree absolute value | SDAfter| the smallest Nomogram value (Or its some neighbour) it is used as final predicted value.
Using this important feedback information of postoperative optometry degree, the Nomogram set based on training sample attribute and doctor
Value, constructs postoperative optometry degree prediction model, and Nomogram initial prediction and its k nearest neighbor (Nomogram value) are inputted postoperative test
Luminosity prediction model selects the Nomogram value for wherein leading to best postoperative optometry degree as final optimization pass result.Program benefit
It uses postoperative optometry degree as feedback information, Nomogram predicted value is finely adjusted and takes excellent, effectively increases former machine learning mould
The prediction precision of type.
It is correct in SMILE operation to illustrate to obtain Nomogram value below by a case by the method for the invention
Property:
Patient, male, 23 years old, preoperative right eye uncorrected visual acuity 0.04, left eye uncorrected visual acuity 0.05, preoperative diopter was right eye :-
7.00DS-1.25DC*5, left eye -6.25DS-2.00DC*180.Corneal thickness: 586 μm of right eye, 583 μm of left eye.Average cornea
Curvature right eye 44.2D, left eye 44.3D calculate hand using the Nomogram prediction model of related mechanism optimization in this patent in art
Nomogram value, right eye: 0.35D, left eye: 0.25D are inputted in art.3 months uncorrected visual acuity right eyes 1.2 of postoperative patients, left eye
1.2.Post-operative refractive degree right eye 0.00DS, left eye 0.00DS;Postoperative 6 months uncorrected visual acuity right eyes 1.2, left eye 1.2. post-operative refractive
Spend right eye -0.50DC*135, left eye+0.25DS;Patient's vision and diopter are fine.
In embodiment provided by the present invention, step S13, step S14 and step S15 use neural network algorithm as
It is preferred that machine learning algorithm constructs required prediction model respectively, optimization method proposed by the invention is suitable for other machines simultaneously
Device learning classification and regression forecasting algorithm, are not limited in optimizing neural network algorithm.Since this patent is intended to protect
Prioritization scheme does not limit its optimizable machine learning method.
Step S11, step S13, step S14 and tetra- kinds of prioritization schemes of step S15 are mutually indepedent, in practical applications, can
To select any one of therein or multinomial scheme of simultaneous selection, to the existing Nomogram value prediction model based on machine learning into
Row optimization.
For the prioritization scheme that step S11, step S14, step S151 are provided, the training sample of use is ideal operation
Case, i.e., postoperative optometry diopter absolute value are less than or equal to 0.5;For step S12, step S152 provide prioritization scheme,
The training sample used for overall situation operation case, that is, may include the undesirable operation that postoperative optometry diopter absolute value is greater than 0.5
Case.
The optimization method of diopter adjusted value, overall scheme framework in prediction SMILE refractive surgery provided by the present invention
Figure is as shown in Figure 8;It is pre- that the mechanism such as equilibrium data distribution, drift correction and the constraint of postoperative optometry degree are applied to optimization machine learning
The problem of surveying cornea refractive surgery Nomogram value, and targetedly designed, including model algorithm design, specifically
Parameter setting and the planning of overall flow step.The results showed this method can effectively promote the pre- of existing machine learning model
Accuracy rate is surveyed, the sample result of prediction has reached clinical application standard.Preoperative solution formulation process can be reduced using this method
Dependence to expert reduces the professional threshold of preoperative solution formulation, further promotes the rate of precision of preoperative solution formulation, is doctor
Design SMILE refractive surgery scheme provides strong Assistance And Instruction.
On the other hand, the present invention also provides a kind of optimization systems of diopter adjusted value in prediction SMILE refractive surgery.Such as
Shown in Fig. 9, which includes processor 92 and the memory 91 for being stored with 42 executable instruction of processor;
Wherein, processor 92 can be general processor, such as central processing unit (CPU), can also be at digital signal
Device (DSP), specific integrated circuit (ASIC) are managed, or is arranged to implement the integrated electricity of one or more of the embodiment of the present invention
Road.
Wherein, memory 91 are transferred to CPU for storing program code, and by the program code.Memory 91 can wrap
Include volatile memory, such as random access memory (RAM);Memory 91 also may include nonvolatile memory, such as
Read-only memory, flash memory, hard disk or solid state hard disk;Memory 91 can also include the group of the memory of mentioned kind
It closes.
Specifically, the optimization system of diopter adjusted value in SMILE refractive surgery is predicted provided by the embodiment of the present invention,
Including processor 92 and memory 91;The computer program run on processor 92 can be used by being stored on memory 91, work as meter
Calculation machine program realizes following steps when being executed by processor 92:
S21 is belonged to using preferred machine learning algorithm using the preoperative parameter of training sample and its postoperative optometry degree as training
Property, training generates prediction model;
S22 is set as ideal value using postoperative optometry degree as objective attribute target attribute for new case to be predicted, and collaboration should be to pre-
The preoperative parameter of new case is surveyed as input, applied forecasting model generates Nomogram predicted value.
Wherein, following steps can also be realized when computer program is executed by processor 92;
S23 constructs deflection forecast model pair based on the Nomogram predicted value that preferred machine learning algorithm generates
The prediction deviation of Nomogram value is predicted, and corrects Nomogram predicted value.
When the Nomogram predicted value generated based on preferred machine learning algorithm, deflection forecast model is constructed to Nomogram
The prediction deviation of value predicted, and when correcting Nomogram predicted value, is realized such as when computer program is executed by processor 92
Lower step;
S231, using preferred machine learning algorithm, the attribute based on training sample, the prediction of building prediction Nomogram value
Model;
S232 predicts training sample applied forecasting model, obtains Nomogram predicted value, and calculate Nomogram
Deviation between predicted value and the Nomogram value of doctor's setting;
S233 is based on training sample attribute and Nomogram predicted value using preferred machine learning algorithm, and building prediction is inclined
The deflection forecast model of difference;
S234, for new case to be predicted, successively applied forecasting model generates Nomogram predicted value, deflection forecast mould
Type generates its prediction deviation, and corrects the Nomogram predicted value with the prediction deviation, obtains revised Nomogram
Predicted value.
Wherein, following steps can also be realized when computer program is executed by processor 92;
S24 constructs postoperative optometry degree prediction model based on the Nomogram value that training sample attribute and doctor are set;It will
Nomogram initial prediction and its k nearest neighbor Nomogram value input postoperative optometry degree prediction model, and selection wherein causes best
The Nomogram value of postoperative optometry degree is as final optimization pass result.
Wherein, when the Nomogram value based on training sample attribute and doctor's setting, postoperative optometry degree prediction model is constructed;
Nomogram initial prediction and its k nearest neighbor Nomogram value are inputted into postoperative optometry degree prediction model, selection wherein causes most
It is realized when the Nomogram value of good postoperative optometry degree is as final optimization pass result, when computer program is executed by processor 92 as follows
Step;
S241, using preferred machine learning algorithm, the attribute based on training sample, the prediction of building prediction Nomogram value
Model;
S242, using preferred machine learning algorithm, the Nomogram value of attribute and doctor's setting based on training sample, structure
Build the postoperative optometry degree prediction model of predicting surgical posteriority luminosity;
S243, for new case to be predicted, applied forecasting model generates its Nomogram initial prediction, and sets step
It is long, generate its k nearest neighbor set;
S244, respectively by the Nomogram initial prediction of new case to be predicted and the k nearest neighbor set of generation, described in collaboration
Other attributes of new case input postoperative optometry degree prediction model, generate corresponding postoperative optometry degree, and selection leads to postoperative optometry
The smallest Nomogram value of absolute value is spent as final predicted value.
Wherein, when use attribute and its postoperative optometry degree of the preferred machine learning algorithm based on training sample are to training sample
It is trained, before generating prediction model, computer program can also realize following steps by the execution of processor 92;
S20 expands the initial training sample of acquisition, increases the number of minority's training sample in initial training sample
Amount, obtains training sample.
Wherein, expand when to the initial training sample of acquisition, increase minority's training sample in initial training sample
Quantity, when obtaining training sample, computer program is executed by processor 92 and realizes following steps;
S201 calculates in initial training sample each attribute to the information gain of Nomogram value, determines invariable attribute and can
Become attribute;
S202, for any one minority's sample, it is nearest that the Euclidean distance based on global property screens minority's sample
Adjacent similar sample;
S203 generates new minority's training sample based on minority's sample, wherein new minority's training sample
Invariable attribute value is identical as minority's sample, and the variable attribute value of new minority's training sample is minority's sample
The random median of this sample respective attributes similar with the arest neighbors;
S204 repeats step S202~S203, until training sample is reached based on the distributed number of different Nomogram values
It is balanced.
Above to it is provided by the present invention prediction SMILE refractive surgery in diopter adjusted value optimization method and system into
Detailed description is gone.For those of ordinary skill in the art, to it under the premise of without departing substantially from true spirit
Any obvious change done, the infringement for all weighing composition to the invention patent, will undertake corresponding legal liabilities.
Claims (10)
1. the optimization method of diopter adjusted value in a kind of prediction SMILE refractive surgery, it is characterised in that include the following steps:
Using preferred machine learning algorithm, using the preoperative parameter of training sample and its postoperative optometry degree as training attribute, training
Generate prediction model;
Ideal value is set as using postoperative optometry degree as objective attribute target attribute for new case to be predicted, cooperates with the new case to be predicted
Preoperative parameter as input, applied forecasting model generate Nomogram predicted value.
2. predicting the optimization method of diopter adjusted value in SMILE refractive surgery as described in claim 1, it is characterised in that also
It may include steps of:
Based on the Nomogram predicted value that preferred machine learning algorithm generates, deflection forecast model is constructed to the pre- of Nomogram value
It surveys deviation to be predicted, and corrects Nomogram predicted value.
3. predicting the optimization method of diopter adjusted value in SMILE refractive surgery as claimed in claim 2, it is characterised in that institute
The Nomogram predicted value generated based on preferred machine learning algorithm is stated, prediction of the deflection forecast model to Nomogram value is constructed
Deviation is predicted, and corrects Nomogram predicted value, is included the following steps:
Using preferred machine learning algorithm, the attribute based on training sample, the prediction model of building prediction Nomogram value;
Training sample applied forecasting model is predicted, Nomogram predicted value is obtained, and calculate Nomogram predicted value with
Deviation between the Nomogram value of doctor's setting;
Using preferred machine learning algorithm, it is based on training sample attribute and Nomogram predicted value, constructs the deviation of prediction deviation
Prediction model;
For new case to be predicted, successively applied forecasting model generates Nomogram predicted value, deflection forecast model generates it
Prediction deviation, and the Nomogram predicted value is corrected with the prediction deviation, obtain revised Nomogram predicted value.
4. predicting the optimization method of diopter adjusted value in SMILE refractive surgery as described in claim 1, it is characterised in that also
It may include steps of:
Based on the Nomogram value that training sample attribute and doctor are set, postoperative optometry degree prediction model is constructed;By Nomogram
Initial prediction and its k nearest neighbor Nomogram value input postoperative optometry degree prediction model, and selection wherein leads to best postoperative optometry
The Nomogram value of degree is as final optimization pass result.
5. predicting the optimization method of diopter adjusted value in SMILE refractive surgery as claimed in claim 4, it is characterised in that institute
The Nomogram value based on training sample attribute and doctor's setting is stated, postoperative optometry degree prediction model is constructed;It will be at the beginning of Nomogram
Beginning predicted value and its k nearest neighbor Nomogram value input postoperative optometry degree prediction model, and selection wherein leads to best postoperative optometry degree
Nomogram value as final optimization pass result, include the following steps:
Using preferred machine learning algorithm, the attribute based on training sample, the prediction model of building prediction Nomogram value;
Using preferred machine learning algorithm, the Nomogram value of attribute and doctor's setting based on training sample constructs predicting surgical
The postoperative optometry degree prediction model of posteriority luminosity;
For new case to be predicted, applied forecasting model generates its Nomogram initial prediction, and sets step-length, generates its K
Neighbour's set;
Respectively by the Nomogram initial prediction of new case to be predicted and the k nearest neighbor set of generation, cooperate with the new case's
Other attributes input postoperative optometry degree prediction model, generate corresponding postoperative optometry degree, and selection leads to postoperative optometry degree absolute value
The smallest Nomogram value is as final predicted value.
6. the optimization method of diopter adjusted value, feature in the prediction SMILE refractive surgery as described in claim 1,2 or 4
It is that the attribute using preferred machine learning algorithm based on training sample and its postoperative optometry degree are trained training sample, it is raw
Before prediction model, it can also include the following steps:
The initial training sample of acquisition is expanded, increases the quantity of minority's training sample in initial training sample, is instructed
Practice sample.
7. predicting the optimization method of diopter adjusted value in SMILE refractive surgery as claimed in claim 6, it is characterised in that institute
It states and the initial training sample of acquisition is expanded, increase the quantity of minority's training sample in initial training sample, trained
Sample includes the following steps:
S111 calculates each attribute in initial training sample and determines invariable attribute and variable category to the information gain of Nomogram value
Property;
S112, for any one minority's sample, it is same that the Euclidean distance based on global property screens minority's sample arest neighbors
Class sample;
S113 generates new minority's training sample based on minority's sample, wherein new minority's training sample it is constant
Attribute value is identical as minority's sample, the variable attribute value of new minority's training sample be minority's sample with
The random median of the similar sample respective attributes of arest neighbors;
S114 repeats step S112~S113, until training sample reaches balanced based on the distributed number of different Nomogram values.
8. the optimization system of diopter adjusted value in a kind of prediction SMILE refractive surgery, it is characterised in that including processor and deposit
Reservoir;The available computer program run on the processor is stored on the memory, when the computer program quilt
The processor realizes following steps when executing:
Using preferred machine learning algorithm, using the preoperative parameter of training sample and its postoperative optometry degree as training attribute, training
Generate prediction model;
Ideal value is set as using postoperative optometry degree as objective attribute target attribute for new case to be predicted, cooperates with the new case to be predicted
Preoperative parameter as input, applied forecasting model generate Nomogram predicted value.
9. predicting the optimization system of diopter adjusted value in SMILE refractive surgery as claimed in claim 8, it is characterised in that when
The computer program can also realize following steps when being executed by the processor:
Based on the Nomogram predicted value that preferred machine learning algorithm generates, deflection forecast model is constructed to the pre- of Nomogram value
It surveys deviation to be predicted, and corrects Nomogram predicted value.
10. predicting the optimization system of diopter adjusted value in SMILE refractive surgery as claimed in claim 8, it is characterised in that
When the computer program is executed by the processor, following steps can also be realized:
Based on the Nomogram value that training sample attribute and doctor are set, postoperative optometry degree prediction model is constructed;By Nomogram
Initial prediction and its k nearest neighbor Nomogram value input postoperative optometry degree prediction model, and selection wherein leads to best postoperative optometry
The Nomogram value of degree is as final optimization pass result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810890395.5A CN109300548B (en) | 2018-08-07 | 2018-08-07 | Optimization method and system for predicting diopter adjustment value in SMILE refractive surgery |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810890395.5A CN109300548B (en) | 2018-08-07 | 2018-08-07 | Optimization method and system for predicting diopter adjustment value in SMILE refractive surgery |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109300548A true CN109300548A (en) | 2019-02-01 |
CN109300548B CN109300548B (en) | 2022-08-19 |
Family
ID=65172781
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810890395.5A Active CN109300548B (en) | 2018-08-07 | 2018-08-07 | Optimization method and system for predicting diopter adjustment value in SMILE refractive surgery |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109300548B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110211686A (en) * | 2019-06-11 | 2019-09-06 | 复旦大学附属眼耳鼻喉科医院 | A kind of high myopia cataract intraocular lens precisely select system |
CN111414972A (en) * | 2020-03-30 | 2020-07-14 | 王雁 | XGboost-based automatic eye refractive correction multi-source data classification method |
WO2022014246A1 (en) * | 2020-07-14 | 2022-01-20 | Sony Group Corporation | Device, computer program and method for predicting post-surgical performance of a patient |
CN114639460A (en) * | 2022-05-16 | 2022-06-17 | 天津医科大学眼科医院 | Cycloplegic demand prediction and paralysis post-diopter refractive state prediction method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140288540A1 (en) * | 2012-09-14 | 2014-09-25 | Carl Zeiss Meditec Ag | Eye surgery refraction correction |
CN108335757A (en) * | 2018-02-05 | 2018-07-27 | 王雁 | A method of diopter adjusted value in prediction SMILE operations |
-
2018
- 2018-08-07 CN CN201810890395.5A patent/CN109300548B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140288540A1 (en) * | 2012-09-14 | 2014-09-25 | Carl Zeiss Meditec Ag | Eye surgery refraction correction |
CN108335757A (en) * | 2018-02-05 | 2018-07-27 | 王雁 | A method of diopter adjusted value in prediction SMILE operations |
Non-Patent Citations (2)
Title |
---|
刘文陆: "基于最大熵理论的眼屈光激光矫正预测技术", 《武汉理工大学学报》 * |
曹艳: "基于均衡分类的脑卒中风险预测模型", 《计算机工程与设计》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110211686A (en) * | 2019-06-11 | 2019-09-06 | 复旦大学附属眼耳鼻喉科医院 | A kind of high myopia cataract intraocular lens precisely select system |
CN111414972A (en) * | 2020-03-30 | 2020-07-14 | 王雁 | XGboost-based automatic eye refractive correction multi-source data classification method |
CN111414972B (en) * | 2020-03-30 | 2023-09-05 | 王雁 | XGBoost-based eye refraction correction multisource data automatic classification method |
WO2022014246A1 (en) * | 2020-07-14 | 2022-01-20 | Sony Group Corporation | Device, computer program and method for predicting post-surgical performance of a patient |
CN114639460A (en) * | 2022-05-16 | 2022-06-17 | 天津医科大学眼科医院 | Cycloplegic demand prediction and paralysis post-diopter refractive state prediction method |
Also Published As
Publication number | Publication date |
---|---|
CN109300548B (en) | 2022-08-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109300548A (en) | A kind of optimization method and system for predicting diopter adjusted value in SMILE refractive surgery | |
Gharehchopogh et al. | Slime mould algorithm: A comprehensive survey of its variants and applications | |
CN108538389B (en) | Method and system for predicting diopter adjustment value in SMILE refractive surgery | |
CN108335757B (en) | Method for predicting diopter adjustment value in SMILE operation | |
Ogink et al. | Predicting discharge placement after elective surgery for lumbar spinal stenosis using machine learning methods | |
Ahmad et al. | Genetic algorithm-artificial neural network (GA-ANN) hybrid intelligence for cancer diagnosis | |
Sakhuja et al. | Genetic algorithm based fuzzy time series tourism demand forecast model | |
CN110009160A (en) | A kind of power price prediction technique based on improved deepness belief network | |
Cheng et al. | Adaptive combination forecasting model for China’s logistics freight volume based on an improved PSO-BP neural network | |
CN103324981A (en) | Chemoradiotherapy standardized quality control quantitative method based on neural network | |
Tembusai et al. | K-nearest neighbor with k-fold cross validation and analytic hierarchy process on data classification | |
Vavilova et al. | Application of genetic algorithm for adjusting the structure of multilayered neural network for prediction of investment processes | |
CN108888407A (en) | A method of based on diopter adjusted value in the prediction SMILE operation of Softmax regression algorithm | |
Jaszkiewicz et al. | The light beam search–outranking based interactive procedure for multiple-objective mathematical programming | |
Kampouridis et al. | Evolving directional changes trading strategies with a new event-based indicator | |
Haque et al. | Parameter and Hyperparameter Optimisation of Deep Neural Network Model for Personalised Predictions of Asthma | |
Valdés-Mas et al. | Machine learning for predicting astigmatism in patients with keratoconus after intracorneal ring implantation | |
CN110378464A (en) | The management method and device of the configuration parameter of artificial intelligence platform | |
Ghandar et al. | An experimental study of multi-objective evolutionary algorithms for balancing interpretability and accuracy in fuzzy rulebase classifiers for financial prediction | |
Tóth-Laufer et al. | Risk level calculation for body physical exercise with different fuzzy based methods | |
CN110322055A (en) | A kind of method and system improving data risk model scoring stability | |
Shamshirband et al. | Adaptive neuro-fuzzy determination of the effect of experimental parameters on vehicle agent speed relative to vehicle intruder | |
Srikamdee et al. | Sugarcane yield and quality forecasting models: adaptive ES vs. deep learning | |
Zhu et al. | From numeric to granular models: A quest for error and performance analysis | |
Gomez-Ruiz et al. | Improving the consistency of AHP matrices using a multi-layer perceptron-based model |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |