CN117084812A - Method for constructing mandibular lead decision system based on deep neural network - Google Patents
Method for constructing mandibular lead decision system based on deep neural network Download PDFInfo
- Publication number
- CN117084812A CN117084812A CN202311059518.8A CN202311059518A CN117084812A CN 117084812 A CN117084812 A CN 117084812A CN 202311059518 A CN202311059518 A CN 202311059518A CN 117084812 A CN117084812 A CN 117084812A
- Authority
- CN
- China
- Prior art keywords
- mandibular
- decision
- neural network
- deep neural
- lead
- 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
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 238000005259 measurement Methods 0.000 claims abstract description 53
- 210000003625 skull Anatomy 0.000 claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000003062 neural network model Methods 0.000 claims abstract description 37
- 238000012795 verification Methods 0.000 claims abstract description 22
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000010276 construction Methods 0.000 claims abstract description 5
- 125000000250 methylamino group Chemical group [H]N(*)C([H])([H])[H] 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 9
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000011161 development Methods 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 3
- 230000009653 mandibular growth Effects 0.000 claims description 3
- 238000012937 correction Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000003759 clinical diagnosis Methods 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 4
- 210000003781 tooth socket Anatomy 0.000 description 4
- 210000000988 bone and bone Anatomy 0.000 description 3
- 230000012010 growth Effects 0.000 description 3
- 238000007689 inspection Methods 0.000 description 3
- 206010061274 Malocclusion Diseases 0.000 description 2
- 210000002455 dental arch Anatomy 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 230000000415 inactivating effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000000399 orthopedic effect Effects 0.000 description 1
- 230000002294 pubertal effect Effects 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61C—DENTISTRY; APPARATUS OR METHODS FOR ORAL OR DENTAL HYGIENE
- A61C7/00—Orthodontics, i.e. obtaining or maintaining the desired position of teeth, e.g. by straightening, evening, regulating, separating, or by correcting malocclusions
- A61C7/002—Orthodontic computer assisted systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Dentistry (AREA)
- Epidemiology (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Dental Tools And Instruments Or Auxiliary Dental Instruments (AREA)
Abstract
The application provides a method for constructing a mandibular leading decision system based on a deep neural network, which comprises the steps of acquiring measured values of a plurality of measurement indexes on a skull side position slice according to decision thinking of an orthodontist as input characteristics of mandibular leading decision-related deep neural network construction; constructing a data set by using a considerable amount of skull side position slices containing different decision suggestions and corresponding input features thereof, and dividing the data set into a training set, a verification set and a test set according to a proportion; constructing a mandibular leading decision system based on three mandibular leading decision suggestions of a deep neural network by using a Pytorch framework and adopting a mode of adjusting a weighting value by a back propagation algorithm, and respectively training and verifying the deep neural network model by a training set and a verification set; and detecting the data samples in the test set by using the deep neural network model, and outputting mandibular leading decision advice basically consistent with the orthodontist. The application can solve the problem of subjective and tedious in the existing mandibular leading decision.
Description
Technical Field
The application relates to the technical field of medical informatization, in particular to a method for constructing a mandibular lead decision system based on a deep neural network.
Background
Class II malocclusion is a deformity in which the mesial-distal relationship between the upper and lower jawbone and the dental arch is not adjusted, the lower jaw and the dental arch are in the mesial-distal position, and molar is in the mesial-distal relationship. The symptoms of malocclusion, such as anterior maxillary process, posterior mandibular shrinkage, lip inclination of the upper anterior teeth, and excessive short lower face are typically represented.
The patient with adolescent bone type II deformity has stronger growth and development potential, and can obtain obvious effect of improving mandibular retroversion through mandibular leading orthopedics. The traditional mandibular lead decision process at present is: the doctor carries out manual fixed point on the head side position sheet of the patient, acquires key index values to comprehensively analyze the bone type and the growth stage of the patient so as to judge the necessity of mandibular lead. The inventor of the present application has found that the conventional method has the following disadvantages: the first and decision analysis systems are numerous, the key indexes of the adopted references are different, the key indexes considered by a single analysis system are dozens, and the workload of clinicians is high; secondly, the diagnostic head shadow measurement reference indexes of each doctor have great difference, the diagnostic process has great subjectivity, and the decision standard of the system consensus is difficult to form.
The characteristics of subjective and tedious clinical diagnosis and treatment process of traditional mandibular leading decision making and centralized medical treatment time of children orthodontic patients are faced, and higher requirements are put forward on clinical efficiency of doctors. Therefore, an efficient and accurate pubertal mandibular pre-lead decision system based on deep neural network formation is becoming an urgent need for aiding clinical diagnosis.
Disclosure of Invention
Aiming at the technical problem of subjectively complicacy in the conventional mandibular leading decision clinical diagnosis and treatment process, the application provides a method for constructing a mandibular leading decision system based on a deep neural network.
In order to solve the technical problems, the application adopts the following technical scheme:
a method for constructing a mandibular lead decision system based on a deep neural network comprises the following steps:
construction of input features: acquiring a plurality of skull side position slice measurement indexes highly related to mandibular lead decision according to decision thinking of orthodontics specialist, arranging and combining the plurality of measurement indexes into a plurality of random index sets containing part or all of the measurement indexes, wherein each random index set is an input characteristic;
formation of data sets: acquiring mandibular leading decision advice of an orthodontist as a data tag by using a considerable number of skull side pieces, and simultaneously acquiring measurement values of a plurality of measurement indexes of each skull side piece; constructing a data set corresponding to each input feature, wherein the data set comprises measured values of all measurement indexes under the input feature and corresponding data labels thereof, and dividing all the data sets into a training set, a verification set and a test set in proportion;
building a mandibular lead decision system based on a deep neural network: constructing a mandibular leading decision system based on a mandibular leading decision system by using a Pytorch framework and adopting a mode of adjusting a weighting value by a back propagation algorithm, and respectively training and verifying the mandibular leading decision system by using a training set and a verification set in a data set corresponding to each input characteristic to obtain a mandibular leading decision-making recommended mandibular leading neural network model;
verification of mandibular lead decision system: the method comprises the steps of detecting data samples in a test set by using a deep neural network model capable of outputting mandibular lead decision suggestions, comparing and checking consistency of results of the data samples in the test set with the decision results of an orthodontist group, and judging the input characteristic with highest decision efficiency according to the decision results of the orthodontist group to be used as the input characteristic of a final decision system, so that a mandibular lead decision system based on the deep neural network is formed.
Further, in the step of constructing the input feature, the step of obtaining a plurality of skull side position slice measurement indexes highly related to the mandibular lead decision according to decision thinking of an orthodontist specifically includes: according to the selected typical skull side position slice of the mandibular lead successful case, tens of head shadow measurement data are measured, and according to decision thinking of orthodontist, ten measurement indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH with highest correlation degree with mandibular growth and development are screened.
Further, in the step of forming the data set, using a considerable number of skull side pieces, obtaining mandibular lead decision advice of an orthodontist as a data tag specifically includes: after the original piece of each skull side piece and the key head shadow measurement data are arranged, a mandibular leading decision is made by an orthodontics doctor group by adopting a double-blind method, three types of decision suggestions of recommended leading, non-recommended leading and trial leading are formed, and the three types of decision suggestions are used as data labels; aiming at the same skull side piece, the first decision opinion is integrated into a data set, the first decision opinion is not integrated, the evaluation is rearranged, the second evaluation is integrated into the data set, the second decision opinion is not integrated, and finally, the decision result data set of the orthodontic specialist doctor group is formed.
Further, in the step of forming the data set, using a considerable number of skull side pieces, and simultaneously obtaining the measured values of the plurality of measurement indexes of each skull side piece specifically includes: selecting skull side pieces with different sexes, different age stages and different jawbone types, and screening out a considerable amount of skull side pieces which meet the quality standard and meet the definition condition as an original file of a collected data set; the selected skull side position slice is subjected to the existing automatic fixed point identification method to obtain the measured value of ten measurement indexes, namely CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH.
Further, in the step of constructing the mandibular lead decision system based on the deep neural network, the deep neural network model includes a first full-connection layer, a first batch normalization layer, a first Mish activation function, a second full-connection layer, a second batch normalization layer, a second Mish activation function, a third full-connection layer and a Dropout layer which are sequentially connected from an input end to an output end.
Further, in the step of building the mandibular lead decision system based on the deep neural network, training and verifying the deep neural network model respectively by using a training set and a verification set in a data set corresponding to each input feature, and obtaining the deep neural network model capable of outputting mandibular lead decision suggestions specifically includes: training a deep neural network model by using a skull lateral position sheet in a training set, deep learning the deep neural network model to extract measured value characteristics of ten measured indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH, obtaining a plurality of training models through training for a plurality of times, predicting verification set data by using each training model, recording accuracy of each training model, and taking the training model with highest accuracy as the deep neural network model capable of outputting mandibular lead decision advice.
Compared with the prior art, the method for automatically making mandibular lead decision based on the deep neural network has the following advantages:
1. the automation of mandibular leading decision is realized for the first time, the standard formation of a booster clinical early correction decision system is realized, an automatic decision model for early correction of children is established for the first time, and the mandibular leading decision clinical diagnosis and treatment process is objective and simple.
2. The intelligent and integrated service improves the clinical efficiency: the application combines the imaging measurement result with the correction decision, and aims to provide high-efficiency, accurate, intelligent and integrated service, thereby reducing the diagnosis and treatment burden of clinicians and improving the auxiliary diagnosis and treatment efficiency.
3. The core decision system is a deep neural network model constructed by the existing Pytorch, the model accuracy of the current optimal input characteristics exceeds 97%, the training set and the verification set are good, meanwhile, the test set also shows good performance in 10 times of inspection and verification, and the model has universality.
Drawings
Fig. 1 is a schematic flow chart of a method for constructing a mandibular lead decision system based on a deep neural network.
Fig. 2 is a schematic diagram of the source of the cranium lateral-slice measurement index CVM of the input feature provided in the present application.
Fig. 3 is a schematic diagram showing the source of the skull side position slice measurement index ANB, SNB, SNA, wits of the input feature provided by the application.
FIG. 4 is a schematic diagram of the source of the measuring indexes SN-MP, Y-Axis, OP-FH of the skull side position slice of the input characteristic part provided by the application.
FIG. 5 is a schematic diagram of the source of the input feature of the present application, namely the skull side-lobe measurement index S-Go/N-Me, over jet.
Fig. 6 is a schematic diagram of a deep neural network model provided by the application.
Fig. 7 is a block diagram of a deep neural network model provided by the present application.
Detailed Description
The application is further described with reference to the following detailed drawings in order to make the technical means, the creation characteristics, the achievement of the purpose and the effect of the implementation of the application easy to understand.
Referring to fig. 1, the application provides a method for constructing a mandibular lead decision system based on a deep neural network, comprising the following steps:
construction of input features: acquiring a plurality of skull side position slice measurement indexes highly related to mandibular lead decision according to decision thinking of orthodontics specialist, arranging and combining the plurality of measurement indexes into a plurality of random index sets containing part or all of the measurement indexes, wherein each random index set is an input characteristic;
formation of data sets: acquiring mandibular leading decision advice of an orthodontist as a data tag by using a considerable number of skull side pieces, and simultaneously acquiring measured values of a plurality of measuring indexes of each skull side piece, namely key head shadow measuring data; constructing a data set corresponding to each input feature, wherein the data set comprises measured values of all measurement indexes under the input feature and corresponding data labels thereof, namely, combining the acquired key head shadow measurement data with decision suggestions (data labels) to form a group of data, forming a data set for judging mandibular lead decision by using a considerable amount of key head shadow measurement data and decision suggestion data containing different measurement indexes, forming a decision part of the data set by decision suggestions, and dividing all the data sets into a training set, a verification set and a test set according to proportion;
building a mandibular lead decision system based on a deep neural network: constructing a mandibular leading decision system based on a depth neural network by using the existing Pytorch framework and adopting a mode of adjusting a weighting value by a back propagation algorithm, and respectively training and verifying the depth neural network model by using a training set and a verification set in a data set corresponding to each input characteristic to obtain the depth neural network model capable of outputting mandibular leading decision suggestions;
verification of mandibular lead decision system: the method comprises the steps of detecting data samples in a test set by using a deep neural network model capable of outputting mandibular lead decision suggestions, comparing and checking consistency of results of the data samples in the test set with the decision results of an orthodontist group, and judging the input characteristic with highest decision efficiency according to the decision results of the orthodontist group to be used as the input characteristic of a final decision system, so that a mandibular lead decision system based on the deep neural network is formed.
As a specific embodiment, in the step of constructing the input feature, the acquiring, according to decision thinking of the orthodontist, a plurality of skull side position slice measurement indexes highly related to mandibular lead decision specifically includes: according to the selected typical skull side position slice of the mandibular lead successful case, tens of head shadow measurement data are measured, and according to decision thinking of orthodontist, ten measurement indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH with highest correlation degree with mandibular growth and development are screened. As a preferred embodiment, ten measurement metrics are combined in a rank-wise arrangement into a total of 638 random metrics sets of 5-10 measurement metrics, each random metric set being an input feature.
As a specific embodiment, in the step of forming the data set, using a considerable number of skull side position slices, obtaining the mandibular anterior decision advice of the orthodontist as the data tag specifically includes: after the original piece of each skull side piece and the key head shadow measurement data are arranged, a double-blind method is adopted to carry out mandibular lead decision by an orthodontist group (for example, the orthodontist group consists of three orthodontists with orthodontics clinical experiences of more than ten years and more than 100 cases with leading function correction completion cases), three types of decision suggestions of recommended lead, non-recommended lead and trial lead are formed, and the three types of decision suggestions are used as data labels; aiming at the same skull side piece, the first decision opinion is integrated into a data set, the first decision opinion is not integrated, the evaluation is rearranged, the second evaluation is integrated into the data set, the second decision opinion is not integrated, and finally, the decision result data set of the orthodontic specialist doctor group is formed. The key head shadow measurement data refer to measurement values corresponding to a plurality of measurement indexes in the construction of the input features.
As a specific embodiment, in the step of forming the dataset, using a considerable number of skull side pieces, and simultaneously obtaining the measured values of a plurality of measurement indexes of each skull side piece specifically includes: selecting skull side pieces with different sexes, different age stages and different jawbone types, screening out a considerable amount of skull side pieces which meet quality standards and meet definition conditions as original files of a collected data set, wherein the quality standards and the definition conditions of the skull side pieces, namely, screening requirements, comprise clear soft and hard tissues, basically overlapping left and right earplug images, and enabling an eye and ear plane to be approximately parallel to the bottom edge of a film; the selected skull side position slice is subjected to the existing automatic fixed point identification method to obtain the measured value of ten measurement indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH, and the measured value is used as the candidate measurement of the input characteristics constructed by the mandibular leading decision-making related deep neural networkAnd (5) an index. The specific meanings of the corresponding measurement values of the ten measurement indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH are well known to those skilled in the art, and referring to FIGS. 2 to 5, the CVM can be evaluated on a skull side position plate, and is a common method for judging the bone age and maturity of a patient, and the method is divided into six periods CS 1-CS 6, wherein CS 1-CS 5 can be used for mandibular guide correction; ANB is the angle formed by the connecting line from the upper tooth socket seat point (A) to the nose root point (N) and the connecting line from the nose root point to the lower tooth socket seat point (B), and is the difference between the SNA angle and the SNB angle, and the reference range of the normal value of the ANB is 2.7+/-2.0 degrees; wits refers to the direction of function from the upper and lower socket points A, B, respectivelyThe plane (OP) is perpendicular, the two feet are respectively AO and BO, then the distance between the two points is measured to reflect the mutual position relation of the front parts of the upper and lower jawbones, and the normal value reference range is 0.0 plus or minus 2.0mm; overjet, covering, refers to cusp stagger +.>When the upper jaw front tooth cover passes through the horizontal distance of the lower jaw front tooth, the normal range is 2.0 plus or minus 1.0mm; SN-MP refers to the angulation of the anterior cranio-planar (SN) and Mandibular Plane (MP), with a normal reference range of 30.4+ -5.6 °; SNB is an angle formed by a center (S) of a butterfly saddle, a nose root point (N) and a lower tooth socket point (B), reflects the position relation of a lower jaw relative to a cranium, and when the angle is overlarge, the lower jaw is in a forward protruding state, otherwise, the lower jaw is in a backward shrinking state, and the normal value reference range is 80.1+/-3.9 degrees; SNA: the angle is formed by a butterfly saddle center, a nose root point and an upper tooth socket seat point. Reflecting the front-back position relation of the upper jaw relative to the cranium, wherein the normal value reference range is 82+/-4 degrees; Y-Axis: the lower anterior angle of the intersection of the connecting line of the center of the butterfly saddle and the vertex of the chin and the plane of the orbit ear also reflects the sudden shrinkage of the chin, and the normal value reference range is 80.1+/-3.9 degrees; S-Go/N-Me: S-Go is high at the back and N-Me is high at the front, the ratio of the two is used for evaluating the facial height and the growth type, and the normal value reference range is 66.0+/-4.0%; OP-FH: />The plane angle is the included angle between the occipital plane and the orbital plane, and represents the inclination of the occipital plane, and the normal value reference range is 9.3+/-1.0 degrees.
As a specific embodiment, please refer to fig. 6 and 7, in the step of building the mandibular lead decision system based on the deep neural network, the deep neural network model includes a first full connection layer (Linear 1), a first batch normalization layer (Batch Normalization 1), a first mix Activation function (mix Activation 1), a second full connection layer (Linear 2), a second batch normalization layer (Batch Normalization 2), a second mix Activation function (mix Activation 2), a third full connection layer (Linear 3) and a Dropout layer sequentially connected from an input end to an output end, where the first full connection layer and the second full connection layer are used to map the learned "distributed feature representation" to the sample marking space; the batch normalization is a method for enabling the neural network to train faster and more stably, calculates the mean value and variance of each mini-batch, and pulls the mean value and variance back to the standard normal distribution with the mean value of 0 and the variance of 1; the Mish activation function is used for increasing the nonlinear capability of the network; the third full-connection layer is a classification layer, the output dimension is 3, and the probability (not normalized) of three categories is obtained; the Dropout layer is located after the third fully-connected layer and is used for randomly inactivating a part of neurons when training the neural network model so as to prevent the neural network model from being overfitted. The deep neural network model also adopts a back propagation algorithm to adjust the weighted value, and prevents overfitting through a verification set, so that errors are reduced. The parameters of each connection layer and the batch normalization layer are set as follows:
Linear1：weight<25×25>，bias<25>；
Batch Normalization1：weight<25>，bias<25>，running_mean<25>，running_var<25>；
Linear2：weight<25×11>，bias<25>；
Batch Normalization2：weight<25>，bias<25>，running_mean<25>，running_var<25>；
Linear3：weight<3×25>，bias<3>。
wherein weight represents the weight; bias represents bias; running_mean represents the mean value of the training phase; running_var represents the variance of the training phase.
As a specific embodiment, in the step of building the mandibular lead decision system based on the deep neural network, training and verifying the deep neural network model by using a training set and a verification set in a data set corresponding to each input feature, where the obtaining the deep neural network model capable of outputting mandibular lead decision advice specifically includes: training a deep neural network model by using a skull lateral position sheet in a training set, extracting measured value characteristics of ten measurement indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH by deep learning of the deep neural network model, obtaining a plurality of training models by training for a plurality of times, predicting verification set data by using each training model, recording accuracy of each training model, taking the training model with highest accuracy as the deep neural network model capable of outputting mandibular leading decision advice, and taking parameters corresponding to the training model with highest accuracy as parameters for adjusting the model.
As a specific embodiment, the deep neural network model capable of outputting mandibular lead decision advice is used for detecting data samples in the test set, and comparing and checking the consistency of the results (Kappa value) with the decision results of the data samples in the test set by the orthodontist group; through inspection, the accuracy of the deep neural network model of the input characteristics of the final decision system exceeds 97%, which indicates that the accuracy of the deep neural network model screened by the method is high, and the network model has universality.
Compared with the prior art, the method for automatically making mandibular lead decision based on the deep neural network has the following advantages:
1. the automation of mandibular leading decision is realized for the first time, the standard formation of a booster clinical early correction decision system is realized, an automatic decision model for early correction of children is established for the first time, and the mandibular leading decision clinical diagnosis and treatment process is objective and simple.
2. The intelligent and integrated service improves the clinical efficiency: the application combines the imaging measurement result with the correction decision, and aims to provide high-efficiency, accurate, intelligent and integrated service, thereby reducing the diagnosis and treatment burden of clinicians and improving the auxiliary diagnosis and treatment efficiency.
3. The core decision system is a deep neural network model constructed by the existing Pytorch, the model accuracy of the current optimal input characteristics exceeds 97%, the training set and the verification set are good, meanwhile, the test set also shows good performance in 10 times of inspection and verification, and the model has universality.
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present application and not for limiting the same, and although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present application, which is intended to be covered by the scope of the claims of the present application.
Claims (6)
1. The method for constructing the mandibular lead decision system based on the deep neural network is characterized by comprising the following steps of:
construction of input features: acquiring a plurality of skull side position slice measurement indexes highly related to mandibular lead decision according to decision thinking of orthodontics specialist, arranging and combining the plurality of measurement indexes into a plurality of random index sets containing part or all of the measurement indexes, wherein each random index set is an input characteristic;
formation of data sets: acquiring mandibular leading decision advice of an orthodontist as a data tag by using a considerable number of skull side pieces, and simultaneously acquiring measurement values of a plurality of measurement indexes of each skull side piece; constructing a data set corresponding to each input feature, wherein the data set comprises measured values of all measurement indexes under the input feature and corresponding data labels thereof, and dividing all the data sets into a training set, a verification set and a test set in proportion;
building a mandibular lead decision system based on a deep neural network: constructing a mandibular leading decision system based on a mandibular leading decision system by using a Pytorch framework and adopting a mode of adjusting a weighting value by a back propagation algorithm, and respectively training and verifying the mandibular leading decision system by using a training set and a verification set in a data set corresponding to each input characteristic to obtain a mandibular leading decision-making recommended mandibular leading neural network model;
verification of mandibular lead decision system: the method comprises the steps of detecting data samples in a test set by using a deep neural network model capable of outputting mandibular lead decision suggestions, comparing and checking consistency of results of the data samples in the test set with the decision results of an orthodontist group, and judging the input characteristic with highest decision efficiency according to the decision results of the orthodontist group to be used as the input characteristic of a final decision system, so that a mandibular lead decision system based on the deep neural network is formed.
2. The method for constructing a mandibular lead decision system based on a deep neural network according to claim 1, wherein in the step of constructing the input features, acquiring a plurality of skull side slice measurement indexes highly related to the mandibular lead decision according to decision thinking of an orthodontist specifically includes: according to the selected typical skull side position slice of the mandibular lead successful case, tens of head shadow measurement data are measured, and according to decision thinking of orthodontist, ten measurement indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH with highest correlation degree with mandibular growth and development are screened.
3. The method for constructing a mandibular lead decision system based on a deep neural network according to claim 1, wherein in the step of forming the dataset, a considerable number of skull side pieces are used, and the acquiring mandibular lead decision advice of an orthodontist as a data tag specifically comprises: after the original piece of each skull side piece and the key head shadow measurement data are arranged, a mandibular leading decision is made by an orthodontics doctor group by adopting a double-blind method, three types of decision suggestions of recommended leading, non-recommended leading and trial leading are formed, and the three types of decision suggestions are used as data labels; aiming at the same skull side piece, the first decision opinion is integrated into a data set, the first decision opinion is not integrated, the evaluation is rearranged, the second evaluation is integrated into the data set, the second decision opinion is not integrated, and finally, the decision result data set of the orthodontic specialist doctor group is formed.
4. The method for constructing a mandibular lead decision system based on a deep neural network according to claim 1, wherein in the step of forming the dataset, a considerable number of skull side pieces are used, and simultaneously acquiring the measured values of a plurality of measurement indexes of each skull side piece specifically comprises: selecting skull side pieces with different sexes, different age stages and different jawbone types, and screening out a considerable amount of skull side pieces which meet the quality standard and meet the definition condition as an original file of a collected data set; the selected skull side position slice is subjected to the existing automatic fixed point identification method to obtain the measured value of ten measurement indexes, namely CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH.
5. The method for constructing a mandibular lead decision system based on a deep neural network according to claim 1, wherein in the step of constructing the mandibular lead decision system based on a deep neural network, the deep neural network model includes a first fully connected layer, a first batch normalization layer, a first mix activation function, a second fully connected layer, a second batch normalization layer, a second mix activation function, a third fully connected layer, and a Dropout layer, which are sequentially connected from an input end to an output end.
6. The method for constructing a mandibular lead decision system based on a deep neural network according to claim 1, wherein in the step of constructing the mandibular lead decision system based on a deep neural network, training and verifying the deep neural network model by using a training set and a verification set in a data set corresponding to each input feature, respectively, and obtaining the deep neural network model capable of outputting mandibular lead decision suggestions specifically includes: training a deep neural network model by using a skull lateral position sheet in a training set, deep learning the deep neural network model to extract measured value characteristics of ten measured indexes of CVM, ANB, wits, overjet, SN-MP, SNB, SNA, Y-Axis, S-Go/N-Me and OP-FH, obtaining a plurality of training models through training for a plurality of times, predicting verification set data by using each training model, recording accuracy of each training model, and taking the training model with highest accuracy as the deep neural network model capable of outputting mandibular lead decision advice.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311059518.8A CN117084812A (en) | 2023-08-22 | 2023-08-22 | Method for constructing mandibular lead decision system based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311059518.8A CN117084812A (en) | 2023-08-22 | 2023-08-22 | Method for constructing mandibular lead decision system based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117084812A true CN117084812A (en) | 2023-11-21 |
Family
ID=88773063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311059518.8A Pending CN117084812A (en) | 2023-08-22 | 2023-08-22 | Method for constructing mandibular lead decision system based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117084812A (en) |
-
2023
- 2023-08-22 CN CN202311059518.8A patent/CN117084812A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6034852B2 (en) | (Semi) automatically designing dental implants | |
| RU2745879C2 (en) | System and method of performing orthodontic diagnostic analysis of a patient | |
| JP5362577B2 (en) | System and method for facilitating automated tooth measurement and diagnosis | |
| JP2009501036A5 (en) | ||
| EP4239538A2 (en) | Method for automated generation of orthodontic treatment final setups | |
| CN111709959B (en) | Digital intelligent diagnosis method for orthodontic treatment | |
| CN109767841A (en) | A kind of scale model search method and device based on cranium jaw face three-dimensional configuration database | |
| CN112869894B (en) | Design method, preparation system and preparation method of shell-shaped tooth appliance | |
| Gerken et al. | Objective computerised assessment of residual ridge resorption in the human maxilla and maxillary sinus pneumatisation | |
| Bāgante et al. | Cleft-related nose deformation evaluation and measurement methods. Literature review | |
| CN113130071A (en) | Method and system for evaluating tooth invisible correction difficulty and optimizing tooth moving path | |
| CN117084812A (en) | Method for constructing mandibular lead decision system based on deep neural network | |
| Lee et al. | Commercial artificial intelligence lateral cephalometric analysis: part 1--the possibility of replacing manual landmarking with artificial intelligence service. | |
| US11488305B2 (en) | Segmentation device | |
| RU2523627C1 (en) | Method for selecting therapeutic options for anterior occlusion | |
| Somvasoontra et al. | Associations among the anterior maxillary dental arch form, alveolar bone thickness, and the sagittal root position of the maxillary central incisors in relation to immediate implant placement: A cone-beam computed tomography analysis | |
| CN114463328B (en) | Automatic orthodontic difficulty coefficient evaluation method | |
| Lee et al. | Commercial artificial intelligence lateral cephalometric analysis: part 2--effects of human examiners on artificial intelligence performance, a pilot study. | |
| Al Shahrani | 3D geometric morphometric analysis of tooth shape in hypodontia | |
| CN115170531A (en) | Method and system for processing mandibular impacted wisdom tooth image | |
| CN115887031B (en) | Tooth arrangement method | |
| Bomvana | Treatment difficulty assessment of postgraduate orthodontic cases at a university clinic | |
| Bleyer | A Predictive Model of" Favorable" Versus" Unfavorable" Growth For Orthodontic Treatment Planning | |
| CN116421341A (en) | Orthognathic surgery planning method, orthognathic surgery planning equipment, orthognathic surgery planning storage medium and orthognathic surgery navigation system | |
| Larkin et al. | Accuracy of artificial intelligence‐assisted growth prediction in skeletal Class I preadolescent patients using serial lateral cephalograms for a 2‐year growth interval |
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 |