CN109961423A - A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium - Google Patents
A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium Download PDFInfo
- Publication number
- CN109961423A CN109961423A CN201910119541.9A CN201910119541A CN109961423A CN 109961423 A CN109961423 A CN 109961423A CN 201910119541 A CN201910119541 A CN 201910119541A CN 109961423 A CN109961423 A CN 109961423A
- Authority
- CN
- China
- Prior art keywords
- lung neoplasm
- data
- lung
- training data
- disaggregated model
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 89
- 230000002685 pulmonary effect Effects 0.000 title claims abstract description 20
- 208000020816 lung neoplasm Diseases 0.000 claims abstract description 443
- 210000004072 lung Anatomy 0.000 claims abstract description 78
- 238000000034 method Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims description 260
- 238000012545 processing Methods 0.000 claims description 27
- 238000012216 screening Methods 0.000 claims description 19
- 238000004891 communication Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 10
- 230000002708 enhancing effect Effects 0.000 claims description 8
- 238000012360 testing method Methods 0.000 abstract description 12
- 238000010586 diagram Methods 0.000 description 8
- 235000013399 edible fruits Nutrition 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 206010058467 Lung neoplasm malignant Diseases 0.000 description 2
- 238000002591 computed tomography Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 201000005202 lung cancer Diseases 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 206010020718 hyperplasia Diseases 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 210000005265 lung cell Anatomy 0.000 description 1
- 238000004064 recycling Methods 0.000 description 1
- 230000011664 signaling Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
- G06T2207/30064—Lung nodule
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
- G06V2201/032—Recognition of patterns in medical or anatomical images of protuberances, polyps nodules, etc.
Abstract
The embodiment of the invention discloses a kind of pulmonary nodule detection method based on disaggregated model, server and storage mediums, this method comprises: obtaining lung CT image data, and Lung neoplasm candidate item to be detected and the corresponding destination image data of the Lung neoplasm candidate item to be detected are determined based on the lung CT image data and Lung neoplasm detection model;The destination image data is handled using Lung neoplasm disaggregated model, obtains the corresponding classification results of the Lung neoplasm candidate item to be detected;The corresponding object detection results of the Lung neoplasm candidate item to be detected are determined based on the classification results;Wherein, the Lung neoplasm disaggregated model includes the first disaggregated model and the second disaggregated model, and first disaggregated model is different with the network structure of second disaggregated model;The classification results include the first classification results and the second classification results.Using the embodiment of the present invention, the classification accuracy of Lung neoplasm can effectively improve, the false positive of Lung neoplasm in testing result is effectively reduced.
Description
Technical field
The present invention relates to field of computer technology more particularly to a kind of pulmonary nodule detection methods based on disaggregated model, clothes
Business device and storage medium.
Background technique
It is increasingly worse due to environment, the probability that people suffer from Lung neoplasm is considerably increased, leads to more and more people all
Lung neoplasm is suffered from.Lung cells hyperplasia or foreign matter can all lead to the generation of Lung neoplasm, and Lung neoplasm is cause lung cancer one
Key factor.Extracing Lung neoplasm as early as possible is the key that Lung neoplasm is effectively prevent to be converted into lung cancer.Currently, existing Lung neoplasm detection
Mode can quick detection goes out Lung neoplasm directly from lung's computed tomography (Computed Tomography, CT) figure,
But usually along with a large amount of false positive Lung neoplasm in testing result.A large amount of false positive Lung neoplasm can be brought to the diagnosis of doctor
Severe jamming, to increase the probability of mistaken diagnosis.Therefore, it is to need to be solved that the false positive of Lung neoplasm in testing result, which how is effectively reduced,
Certainly the problem of.
Summary of the invention
The embodiment of the present invention provides a kind of pulmonary nodule detection method based on disaggregated model, server and storage medium, can
To effectively improve the classification accuracy of Lung neoplasm, the false positive of Lung neoplasm in testing result is effectively reduced.
In a first aspect, the embodiment of the invention provides a kind of pulmonary nodule detection methods based on disaggregated model, comprising:
Lung CT image data are obtained, and are determined based on the lung CT image data and Lung neoplasm detection model to be detected
Lung neoplasm candidate item and the corresponding destination image data of the Lung neoplasm candidate item to be detected;
The destination image data is handled using Lung neoplasm disaggregated model, it is candidate to obtain the Lung neoplasm to be detected
The corresponding classification results of item;
The corresponding object detection results of the Lung neoplasm candidate item to be detected are determined based on the classification results;
Wherein, the Lung neoplasm disaggregated model includes the first disaggregated model and the second disaggregated model, the first classification mould
Type is different with the network structure of second disaggregated model;The classification results include the first classification results and the second classification knot
Fruit;It is described that the destination image data is handled using Lung neoplasm disaggregated model, it is candidate to obtain the Lung neoplasm to be detected
The corresponding classification results of item, comprising:
The destination image data is inputted in first disaggregated model and is handled, the Lung neoplasm to be detected is obtained
Corresponding first classification results of candidate item;
The destination image data is inputted in second disaggregated model and is handled, the Lung neoplasm to be detected is obtained
Corresponding second classification results of candidate item.
In one embodiment, it is true lung that first classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the first probability value, and it is true lung that second classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the second probability value;
It is described to determine the corresponding object detection results of the Lung neoplasm candidate item to be detected based on the classification results, packet
It includes:
Obtain the most probable value in first probability value and second probability value;
The most probable value is determined as the corresponding object detection results of the Lung neoplasm candidate item to be detected.
In one embodiment, it is true lung that first classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the first probability value, and it is true lung that second classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the second probability value;The Lung neoplasm disaggregated model further includes third disaggregated model, the third disaggregated model
It is different with the size of data of input data of first disaggregated model;The classification results further include third classification results;
It is described that the destination image data is handled using Lung neoplasm disaggregated model, obtain the Lung neoplasm to be detected
The corresponding classification results of candidate item, further includes:
The destination image data is inputted in the third disaggregated model and is handled, the Lung neoplasm to be detected is obtained
The corresponding third classification results of candidate item, the third classification results, which are used to indicate the Lung neoplasm candidate item to be detected, is
The probability of true Lung neoplasm is third probability value;
It is described to determine the corresponding object detection results of the Lung neoplasm candidate item to be detected based on the classification results, packet
It includes:
Obtain the most probable value in first probability value, second probability value and the third probability value;
The most probable value is determined as the corresponding object detection results of the Lung neoplasm candidate item to be detected.
In one embodiment, described to determine lung to be detected based on the lung CT image data and Lung neoplasm detection model
Nodule candidates and the corresponding destination image data of the Lung neoplasm candidate item to be detected, comprising:
The lung CT image data are detected using the Lung neoplasm detection model, obtain initial detecting as a result,
The initial detecting result includes multiple Lung neoplasm candidate items, each Lung neoplasm candidate item in the multiple Lung neoplasm candidate item
It is the probability and corresponding tubercle coordinate of true Lung neoplasm;
It will be that the probability of true Lung neoplasm is more than or equal to the lung knot of probability threshold value in the multiple Lung neoplasm candidate item
Section candidate item is determined as Lung neoplasm candidate item to be detected;
It is determined from the lung CT image data based on the corresponding tubercle coordinate of the Lung neoplasm candidate item to be detected
The corresponding destination image data of the Lung neoplasm candidate item to be detected, the size of data of the destination image data and the lung knot
The size of data for saving the input data of disaggregated model is identical.
In one embodiment, the method also includes:
Training data is obtained, the training data includes the corresponding original data of multiple Lung neoplasm candidate items, described
Each of multiple Lung neoplasm candidate items Lung neoplasm candidate item is that the probability of true Lung neoplasm is all larger than or is equal to probability threshold
Value;
Data cutout processing is carried out to the training data, obtains the corresponding target training number of the Lung neoplasm disaggregated model
According to the size of data of the target training data is identical as the size of data of input data of the Lung neoplasm disaggregated model;
The Lung neoplasm disaggregated model is trained using the target training data, the Lung neoplasm after being trained point
Class model.
In one embodiment, the target training data includes positive sample training data and negative sample training data, institute
It states and the Lung neoplasm disaggregated model is trained using the target training data, the Lung neoplasm classification mould after being trained
Type, comprising:
The positive sample training data is determined from the target training data;
Data enhancing processing is carried out to the positive sample training data, obtains first object positive sample training data;
Classified using the first object positive sample training data and the negative sample training data to the Lung neoplasm
Model is trained, and obtains the first Lung neoplasm base categories model;
Based on the first object positive sample training data, the negative sample training data and the first Lung neoplasm base
Plinth disaggregated model, the Lung neoplasm disaggregated model after being trained;
Wherein, the ratio data between the first object positive sample training data and the negative sample training data is the
One ratio.
In one embodiment, described to be based on the first object positive sample training data, the negative sample training data
And the first Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained, comprising:
Data screening is carried out to the negative sample training data, obtains first object negative sample training data;
Using the first object positive sample training data and the first object negative sample training data to described
One Lung neoplasm base categories model is trained, and obtains the second Lung neoplasm base categories model;
Based on the first object positive sample training data, the first object negative sample training data and described second
Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained;
Wherein, the data between the first object positive sample training data and the first object negative sample training data
Ratio is the second ratio.
In one embodiment, described to be based on the first object positive sample training data, the first object negative sample
Training data and the second Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained, comprising:
Data screening is carried out to the first object positive sample training data, obtains the second target positive sample training data;
Data screening is carried out to the first object negative sample training data, obtains the second target negative sample training data;
Using the second target positive sample training data and the second target negative sample training data to described
Two Lung neoplasm base categories models are trained, the Lung neoplasm disaggregated model after being trained;
Wherein, the data between the second target positive sample training data and the second target negative sample training data
Ratio is second ratio.
Second aspect, the Lung neoplasm detection device based on disaggregated model that the embodiment of the invention provides a kind of, comprising:
Acquiring unit, for obtaining lung CT image data;
Processing unit is used for:
Lung neoplasm candidate item to be detected and described is determined based on the lung CT image data and Lung neoplasm detection model
The corresponding destination image data of Lung neoplasm candidate item to be detected;
The destination image data is handled using Lung neoplasm disaggregated model, it is candidate to obtain the Lung neoplasm to be detected
The corresponding classification results of item;
The corresponding object detection results of the Lung neoplasm candidate item to be detected are determined based on the classification results;
Wherein, the Lung neoplasm disaggregated model includes the first disaggregated model and the second disaggregated model, the first classification mould
Type is different with the network structure of second disaggregated model;The classification results include the first classification results and the second classification knot
Fruit;The processing unit, is specifically used for:
The destination image data is inputted in first disaggregated model and is handled, the Lung neoplasm to be detected is obtained
Corresponding first classification results of candidate item;
The destination image data is inputted in second disaggregated model and is handled, the Lung neoplasm to be detected is obtained
Corresponding second classification results of candidate item.
The third aspect, the embodiment of the invention provides a kind of server, including processor, communication interface and memory, institutes
Processor, the communication interface and the memory to be stated to be connected with each other, wherein the memory is used to store computer program,
The computer program includes program instruction, and the processor is configured for calling described program instruction, executes above-mentioned first
The described in any item pulmonary nodule detection methods based on disaggregated model of aspect.
Fourth aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage
Media storage has computer program, and the computer program includes program instruction, and described program instructs when being executed by a processor
The processor is set to execute the described in any item pulmonary nodule detection methods based on disaggregated model of above-mentioned first aspect.
In the embodiment of the present invention, determine that Lung neoplasm to be detected is waited based on lung CT image data and Lung neoplasm detection model
After option and the corresponding destination image data of Lung neoplasm candidate item to be detected, recycle Lung neoplasm disaggregated model to target image
Data are handled to obtain classification results, and determine the corresponding target detection of Lung neoplasm candidate item to be detected based on classification results
As a result, so as to effectively improve the classification accuracy of Lung neoplasm, the false positive of Lung neoplasm in testing result is effectively reduced.
Detailed description of the invention
Technical solution in order to illustrate the embodiments of the present invention more clearly, below will be to needed in embodiment description
Attached drawing is briefly described, it should be apparent that, drawings in the following description are some embodiments of the invention, general for this field
For logical technical staff, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of flow diagram for pulmonary nodule detection method that first embodiment of the invention provides;
Fig. 2 is a kind of flow diagram for pulmonary nodule detection method that second embodiment of the invention provides;
Fig. 3 is a kind of structural schematic diagram of Lung neoplasm detection device provided in an embodiment of the present invention;
Fig. 4 is a kind of structural schematic diagram of server provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description.
Referring to Figure 1, Fig. 1 is a kind of pulmonary nodule detection method based on disaggregated model that first embodiment of the invention provides
Flow diagram.Specifically, as shown in Figure 1, the pulmonary nodule detection method may comprise steps of:
S101, server obtain lung CT image data, and detect mould based on the lung CT image data and Lung neoplasm
Type determines Lung neoplasm candidate item to be detected and the corresponding destination image data of the Lung neoplasm candidate item to be detected.
In the embodiment of the present invention, lung CT image data are three-dimensional data, and server can be obtained directly from CT equipment
Take the lung CT image data.After server gets lung CT image data, Lung neoplasm detection model, the Lung neoplasm are obtained
Detection model can be stored in server local, also can store in other cloud servers.The server by utilizing lung knot
Section detection model detects the lung CT image data, obtains initial detecting result.The initial detecting result includes determining
Each Lung neoplasm candidate item is the general of true Lung neoplasm in multiple Lung neoplasm candidate items out, multiple Lung neoplasm candidate item
Tubercle coordinate corresponding to each Lung neoplasm candidate item in rate and multiple Lung neoplasm candidate item.
Further, server will be the probability of true Lung neoplasm in multiple Lung neoplasm candidate item more than or equal to general
The Lung neoplasm candidate item of rate threshold value is determined as Lung neoplasm candidate item to be detected, which is, for example, 85%, the lung to be detected
Nodule candidates can be one or more.Server be then based on the corresponding tubercle coordinate of Lung neoplasm candidate item to be detected from
The corresponding destination image data of Lung neoplasm candidate item to be detected, the number of the destination image data are determined in lung CT image data
It is identical as the size of data of the input data of Lung neoplasm disaggregated model according to size.Specifically, server is waited with Lung neoplasm to be detected
Point centered on the corresponding tubercle coordinate of option, data intercept size is the number of target data size from lung CT image data
According to as the corresponding destination image data of Lung neoplasm candidate item to be detected;The data of the input data of Lung neoplasm disaggregated model are big
Small is also the target data size.
It should be noted that server can will be the probability of true Lung neoplasm in multiple Lung neoplasm candidate item less than general
The Lung neoplasm candidate item of rate threshold value is determined directly as false positive Lung neoplasm.
S102, the server by utilizing Lung neoplasm disaggregated model handle the destination image data, obtain described
The corresponding classification results of Lung neoplasm candidate item to be detected.
In the embodiment of the present invention, server obtains Lung neoplasm disaggregated model, which can be stored in
Server local also can store in other cloud servers.Lung neoplasm disaggregated model includes the first disaggregated model and second
Disaggregated model, first disaggregated model are different with the network structure of second disaggregated model.In one embodiment, this first point
The network structure of class model can be to be obtained with the basic network struction of network VGG-16, may include 12 layers of convolution.This
The network structure of two disaggregated models can be to be obtained with the basic network struction of network DualPathNetwork (DPN) -68.
Wherein, the size of data of the input data of first disaggregated model is the first size of data, second disaggregated model
Input data size of data be the second size of data;First size of data and second size of data can be identical, and
It is identical as above-mentioned target data size, e.g. 48*48*48.The classification results include the first classification results and the second classification knot
Fruit.Server by utilizing Lung neoplasm disaggregated model handles destination image data, and it is corresponding to obtain Lung neoplasm candidate item to be detected
Classification results, specifically include: server will the destination image data input the first disaggregated model in handle, obtain to be checked
Survey corresponding first classification results of Lung neoplasm candidate item;It is true that first classification results, which are used to indicate Lung neoplasm candidate item to be detected,
The probability of real Lung neoplasm is the first probability value.The destination image data is inputted in the second disaggregated model and is handled by server,
Obtain corresponding second classification results of Lung neoplasm candidate item to be detected;Second classification results are used to indicate Lung neoplasm to be detected and wait
Option is that the probability of true Lung neoplasm is the second probability value.
In one embodiment, first size of data and second size of data can also be different.First data are big
Small is, for example, 48*48*48, which is, for example, 64*64*64.Above-mentioned target data size includes that the first data are big
Small and the second size of data, the destination image data include first object image data and the second destination image data.Server
Determine that Lung neoplasm to be detected is candidate from lung CT image data based on the corresponding tubercle coordinate of Lung neoplasm candidate item to be detected
The corresponding destination image data of item, specifically include: server is centered on the corresponding tubercle coordinate of Lung neoplasm candidate item to be detected
Point, data intercept size is the data of the first size of data from lung CT image data, corresponding as the first disaggregated model
First object image data;And the point centered on the corresponding tubercle coordinate of Lung neoplasm candidate item to be detected, from lung CT image number
It is the data of the second size of data according to middle data intercept size, as corresponding second destination image data of the second disaggregated model.
Further, which is inputted in the first disaggregated model and is handled by server, obtains lung knot to be detected
Save corresponding first classification results of candidate item;And second destination image data is inputted in the second disaggregated model and is handled,
Obtain corresponding second classification results of Lung neoplasm candidate item to be detected.
S103, the server determine that the corresponding target of the Lung neoplasm candidate item to be detected is examined based on the classification results
Survey result.
In the embodiment of the present invention, server determines lung knot to be detected based on above-mentioned first classification results and the second classification results
Save the corresponding object detection results of candidate item.Specifically, server obtains in above-mentioned first probability value and the second probability value
Most probable value, and the most probable value is determined as the corresponding object detection results of Lung neoplasm candidate item to be detected.The target
Testing result is used to indicate the final probability that Lung neoplasm candidate item to be detected is true Lung neoplasm.Using aforesaid way, detecting
After model carries out initial detecting for Lung neoplasm candidate item, being directed to respectively using the first disaggregated model and the second disaggregated model is
The Lung neoplasm candidate item to be detected that the probability of true Lung neoplasm is more than or equal to probability threshold value carries out further classification processing;And
Based on the classification results that the first disaggregated model and the second disaggregated model obtain, determine that detection Lung neoplasm candidate item is true lung knot
The final probability of section;So as to combine detection model, the first disaggregated model and the second disaggregated model to carry out Lung neoplasm inspection simultaneously
It surveys, can effectively improve the classification accuracy of Lung neoplasm.
In one embodiment, server is based on above-mentioned classification results and determines the corresponding mesh of Lung neoplasm candidate item to be detected
After marking testing result, judge that Lung neoplasm candidate item to be detected is the final of true Lung neoplasm indicated by the object detection results
Whether probability is more than or equal to default value, which is, for example, 95%;If so, determining that the Lung neoplasm to be detected is waited
Option is true Lung neoplasm;It is on the contrary, it is determined that the Lung neoplasm candidate item to be detected is false positive Lung neoplasm.Using aforesaid way,
The false positive of Lung neoplasm in testing result can be effectively reduced.
In one embodiment, above-mentioned Lung neoplasm disaggregated model further includes third disaggregated model, the third disaggregated model
Network structure can be to be obtained with the basic network struction of network VGG-16, may include 12 layers of convolution.Third classification mould
The size of data of the input data of type is third size of data, which is, for example, 32*32*32.Third classification mould
Type is different with the size of data of the input data of above-mentioned first disaggregated model, that is to say the third size of data and above-mentioned first number
According to of different sizes.Specifically, which is less than above-mentioned first size of data.Above-mentioned target data size further includes
Three size of data, the destination image data further include third destination image data.Server is based on Lung neoplasm candidate item to be detected
Corresponding tubercle coordinate determines the corresponding destination image data of Lung neoplasm candidate item to be detected from lung CT image data, also
Include: server point centered on the corresponding tubercle coordinate of Lung neoplasm candidate item to be detected, is intercepted from lung CT image data
Size of data is the data of third size of data, as the corresponding third destination image data of third disaggregated model.
Wherein, above-mentioned classification results further include third classification results.Server by utilizing Lung neoplasm disaggregated model is to target figure
As data are handled, the corresponding classification results of Lung neoplasm candidate item to be detected are obtained, further includes: server is by third target figure
As being handled in data input third disaggregated model, the corresponding third classification results of Lung neoplasm candidate item to be detected are obtained, are somebody's turn to do
It is third probability value that third classification results, which are used to indicate the probability that Lung neoplasm candidate item to be detected is true Lung neoplasm,.Server base
The mode of the corresponding object detection results of Lung neoplasm candidate item to be detected is determined in classification results are as follows: obtains above-mentioned first probability
Most probable value in value, the second probability value and third probability value, and the most probable value is determined as Lung neoplasm to be detected
The corresponding object detection results of candidate item.
It should be noted that third disaggregated model is different with the size of data of the input data of above-mentioned second disaggregated model,
It that is to say that the third size of data is different with above-mentioned second size of data.Specifically, third size of data is less than above-mentioned second number
According to size.The third disaggregated model is big using different input datas from first disaggregated model and second disaggregated model
It is small, the case where major tubercle and lesser tubercle can be combined, so as to expand the applicable application scenarios of this programme.
In the embodiment of the present invention, determine that Lung neoplasm to be detected is waited based on lung CT image data and Lung neoplasm detection model
After option and the corresponding destination image data of Lung neoplasm candidate item to be detected, recycle Lung neoplasm disaggregated model to target image
Data are handled to obtain classification results, and determine the corresponding target detection of Lung neoplasm candidate item to be detected based on classification results
As a result, so as to effectively improve the classification accuracy of Lung neoplasm, the false positive of Lung neoplasm in testing result is effectively reduced.
Fig. 2 is referred to, Fig. 2 is a kind of pulmonary nodule detection method based on disaggregated model that second embodiment of the invention provides
Flow diagram.Specifically, as shown in Fig. 2, the pulmonary nodule detection method may comprise steps of:
S201, server obtain training data.
In the embodiment of the present invention, server obtains initial CT image data, and initial to this using Lung neoplasm detection model
CT image data is detected, and initial detecting result is obtained.The initial detecting result includes the multiple Lung neoplasms candidate determined
Each Lung neoplasm candidate item is the probability and multiple Lung neoplasm of true Lung neoplasm in item, multiple Lung neoplasm candidate item
Tubercle coordinate corresponding to each Lung neoplasm candidate item in candidate item.Further, it is candidate to be based on target Lung neoplasm for server
Tubercle coordinate obtain the corresponding original data of target Lung neoplasm candidate item from initial CT image data, and by mesh
The corresponding original data of mark Lung neoplasm candidate item is determined as training data.Wherein, target Lung neoplasm candidate item is multiple
It is that the probability of true Lung neoplasm is more than or equal to the Lung neoplasm candidate item of probability threshold value, the target lung knot in Lung neoplasm candidate item
Section candidate item can be multiple.
S202, the server carry out data cutout processing to the training data, and it is corresponding to obtain Lung neoplasm disaggregated model
Target training data.
In the embodiment of the present invention, the data of the input data of the size of data and Lung neoplasm disaggregated model of target training data
Size is identical.Specifically, server point centered on the corresponding tubercle coordinate of target Lung neoplasm candidate item is cut from training data
Take the data that size of data is target data size as the corresponding target training data of Lung neoplasm disaggregated model;Lung neoplasm classification
The size of data of the input data of model is also the target data size.Wherein, Lung neoplasm disaggregated model specifically includes first point
Class model and the second disaggregated model.The size of data of the input data of first disaggregated model is the first size of data, this second
The size of data of the input data of disaggregated model is the second size of data.Above-mentioned target data size include the first size of data and
Second size of data, the target training data include first object training data and the second target training data.Server is with mesh
Point centered on the corresponding tubercle coordinate of mark Lung neoplasm candidate item, data intercept size is the first size of data from training data
Data, as the corresponding first object training data of the first disaggregated model;And it is sat with the corresponding tubercle of target Lung neoplasm candidate item
It is designated as central point, data intercept size is the data of the second size of data from training data, corresponding as the second disaggregated model
The second target training data.
In one embodiment, Lung neoplasm disaggregated model further includes third disaggregated model, the input of the third disaggregated model
The size of data of data is third size of data.Above-mentioned target data size further includes third size of data, target training number
According to further including third target training data.Server point centered on the corresponding tubercle coordinate of target Lung neoplasm candidate item, from instruction
Practice the data that data intercept size in data is third size of data, as the corresponding third target training number of third disaggregated model
According to.
In one embodiment, third size of data, can be with target Lung neoplasm candidate item pair less than the first size of data
Point centered on the tubercle coordinate answered, data intercept size is the number of third size of data directly from the first object training data
According to as the corresponding third target training data of third disaggregated model.When the first size of data is identical as the second size of data,
It can be directly using first object training data as the corresponding second target training data of the second disaggregated model.
Target training data described in S203, the server by utilizing is trained the Lung neoplasm disaggregated model, obtains
Lung neoplasm disaggregated model after training.
In the embodiment of the present invention, which includes positive sample training data and negative sample training data.It can be with
It is the data note by the intersection in target training data between true Lung neoplasm less than the first preset percentage (such as 20%)
Be negative sample training data;It can be and the intersection in target training data between true Lung neoplasm is greater than the second default percentage
The data of ratio are denoted as positive sample training data;Positive sample training data can also be determined directly according to the markup information in data
Out.
Specifically, server determines positive sample training data from the target training data first, and to the positive sample
Training data carries out data enhancing processing, obtains first object positive sample training data, so that first object positive sample training
Ratio data between data and the negative sample training data reaches the first ratio, which is, for example, 1:3.Server is right
Lung neoplasm disaggregated model is trained using the first object positive sample training data and the negative sample training data afterwards, is obtained
To the first Lung neoplasm base categories model;Finally based on the first object positive sample training data, the negative sample training data with
And the first Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained.
Wherein, due to the usual pole of ratio in target training data between positive sample training data and negative sample training data
Degree is uneven, therefore needs to carry out data enhancing processing to the positive sample training data in target training data, so that data enhance
Ratio data between the first object positive sample training data obtained after processing and negative sample training data reaches the first ratio.
Carrying out data enhancing processing to positive sample training data includes: to overturn to three reference axis in positive sample training data
(flip), and/or to the direction of three reference axis the enhancing behaviour such as (such as being rotated by 90 °, 180 degree, 270 degree) data is rotated
Make.In one embodiment, positive sample training data can be enhanced into presupposition multiple (such as 16 times), so that positive sample trains number
It can achieve the first ratio according to the ratio between negative sample training data.For example, data enhance 16 times=3 (3 coordinates
The overturning of axis)+3 (rotation of 3 angles)+3*3 (overturning is superimposed with rotation)+initial data.
In one embodiment, server be based on the first object positive sample training data, the negative sample training data with
And the first Lung neoplasm base categories model, the mode of the Lung neoplasm disaggregated model after being trained are as follows: server is first to this
Negative sample training data carries out data screening, obtains first object negative sample training data, so that the first object positive sample is instructed
The ratio data practiced between data and the first object negative sample training data reaches the second ratio, which is, for example, 1:
1.Wherein, data screening, which may is that it will is true Lung neoplasm in the negative sample training data, to be carried out to the negative sample training data
Probability corresponding less than the target Lung neoplasm candidate item of the first numerical value data removal, obtain first object negative sample training number
According to.Server then utilize the first object positive sample training data and the first object negative sample training data to this first
Lung neoplasm base categories model is trained, and obtains the second Lung neoplasm base categories model;Finally it is based on the positive sample of the first object
This training data, the first object negative sample training data and the second Lung neoplasm base categories model, after being trained
Lung neoplasm disaggregated model.Wherein, the server by utilizing first object positive sample training data and first object negative sample instruction
Before white silk data are trained the first Lung neoplasm base categories model, the first Lung neoplasm base categories mould can be first adjusted
The learning rate (learning rate) of type.
In one embodiment, server is based on the first object positive sample training data, the first object negative sample is instructed
Practice data and the second Lung neoplasm base categories model, the mode of the Lung neoplasm disaggregated model after being trained are as follows: server
Data screening is carried out to the first object positive sample training data first, obtains the second target positive sample training data;And to this
First object negative sample training data carries out data screening, obtains the second target negative sample training data.The positive sample of second target
Ratio data between this training data and the second target negative sample training data remains second ratio.Wherein, to this
First object positive sample training data carries out data screening may is that be true lung in the first object positive sample training data
The probability of tubercle is less than the corresponding data removal of target Lung neoplasm candidate item of second value, obtains the training of the second target positive sample
Data.Carrying out data screening to the first object negative sample training data may is that the first object negative sample training data
In be that the probability of true Lung neoplasm is less than the target Lung neoplasm candidate item corresponding data removal of third value, obtain the second target
Negative sample training data.Then server utilizes the second target positive sample training data and the second target negative sample training
Data are trained the second Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained.Wherein, server
Using the second target positive sample training data and the second target negative sample training data to the second Lung neoplasm basis point
Before class model is trained, the learning rate of the second Lung neoplasm base categories model can be first adjusted.
It should be noted that server trained after Lung neoplasm disaggregated model after, determined using sample data
The classification accuracy of Lung neoplasm disaggregated model after training.If detecting that the classification of the Lung neoplasm disaggregated model after training is accurate
Rate is more than or equal to default accuracy rate threshold value, then deconditioning.The default accuracy rate threshold value is, for example, 99%.If detected
The classification accuracy of Lung neoplasm disaggregated model after training is less than default accuracy rate threshold value, then continues to screen the second target positive sample
Training data and the second target negative sample training data, constantly diminution training data, adjust the learning rate of Lung neoplasm disaggregated model;
And continue to train Lung neoplasm disaggregated model using the positive and negative sample training data after screening, the mould until Lung neoplasm after training is classified
Until the classification accuracy of type reaches default accuracy rate threshold value.Using aforesaid way, Lung neoplasm disaggregated model can be continuously improved
Classification accuracy.
In the embodiment of the present invention, server by utilizing target training data is trained Lung neoplasm disaggregated model, is instructed
Lung neoplasm disaggregated model after white silk, specifically includes: being trained, is obtained to the first disaggregated model using first object training data
The first disaggregated model after training;The second disaggregated model is trained using the second target training data, after being trained
Second disaggregated model;Third disaggregated model is trained using third target training data, the third classification after being trained
Model.Specific training method, which can refer to, to be described above, and details are not described herein.
S204, the server obtain lung CT image data, and are examined based on the lung CT image data and Lung neoplasm
It surveys model and determines Lung neoplasm candidate item to be detected and the corresponding destination image data of the Lung neoplasm candidate item to be detected.
Described in S205, the server by utilizing training after Lung neoplasm disaggregated model to the destination image data at
Reason, obtains the corresponding classification results of the Lung neoplasm candidate item to be detected.
S206, the server determine that the corresponding target of the Lung neoplasm candidate item to be detected is examined based on the classification results
Survey result.
In the embodiment of the present invention, the specific implementation of step S204 to step S206 are referred to real shown in above-mentioned Fig. 1
Apply in example that step S101 is to the associated description of step S103, details are not described herein again.
In the embodiment of the present invention, data cutout is carried out to training data first and handles to obtain target training data, utilizes mesh
Mark training data is trained Lung neoplasm disaggregated model, and the Lung neoplasm disaggregated model after being trained is then based on lung CT
Image data and Lung neoplasm detection model determine that Lung neoplasm candidate item to be detected and Lung neoplasm candidate item to be detected are corresponding
After destination image data, the Lung neoplasm disaggregated model after recycling training handles destination image data to obtain classification knot
Fruit, and the corresponding object detection results of Lung neoplasm candidate item to be detected are determined based on classification results, so as to effectively improve
The classification accuracy of Lung neoplasm, the false positive of Lung neoplasm in testing result is effectively reduced.
Fig. 3 is referred to, Fig. 3 is a kind of knot of Lung neoplasm detection device based on disaggregated model provided in an embodiment of the present invention
Structure schematic diagram.The Lung neoplasm detection device of the embodiment of the present invention includes the unit for executing above-mentioned pulmonary nodule detection method.Tool
Body, the Lung neoplasm detection device 300 of the embodiment of the present invention may include: that acquiring unit 301, processing unit 302 and training are single
Member 303.Wherein:
The acquiring unit 301, for obtaining lung CT image data;
The processing unit 302, is used for:
Lung neoplasm candidate item to be detected and described is determined based on the lung CT image data and Lung neoplasm detection model
The corresponding destination image data of Lung neoplasm candidate item to be detected;
The destination image data is handled using Lung neoplasm disaggregated model, it is candidate to obtain the Lung neoplasm to be detected
The corresponding classification results of item;
The corresponding object detection results of the Lung neoplasm candidate item to be detected are determined based on the classification results;
Wherein, the Lung neoplasm disaggregated model includes the first disaggregated model and the second disaggregated model, the first classification mould
Type is different with the network structure of second disaggregated model;The classification results include the first classification results and the second classification knot
Fruit;The processing unit 302, is specifically used for:
The destination image data is inputted in first disaggregated model and is handled, the Lung neoplasm to be detected is obtained
Corresponding first classification results of candidate item;
The destination image data is inputted in second disaggregated model and is handled, the Lung neoplasm to be detected is obtained
Corresponding second classification results of candidate item.
In one embodiment, it is true lung that first classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the first probability value, and it is true lung that second classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the second probability value;The processing unit 302, is specifically used for:
Obtain the most probable value in first probability value and second probability value;
The most probable value is determined as the corresponding object detection results of the Lung neoplasm candidate item to be detected.
In one embodiment, the Lung neoplasm disaggregated model further includes third disaggregated model, the third disaggregated model
It is different with the size of data of input data of first disaggregated model;The classification results further include third classification results;
The processing unit 302 is also used to input the destination image data in the third disaggregated model
Reason, obtains the corresponding third classification results of the Lung neoplasm candidate item to be detected, the third classification results are used to indicate
The Lung neoplasm candidate item to be detected is that the probability of true Lung neoplasm is third probability value;
The processing unit 302, is specifically used for:
Obtain the most probable value in first probability value, second probability value and the third probability value;
The most probable value is determined as the corresponding object detection results of the Lung neoplasm candidate item to be detected.
In one embodiment, the processing unit 302, is specifically used for:
The lung CT image data are detected using the Lung neoplasm detection model, obtain initial detecting as a result,
The initial detecting result includes multiple Lung neoplasm candidate items, each Lung neoplasm candidate item in the multiple Lung neoplasm candidate item
It is the probability and corresponding tubercle coordinate of true Lung neoplasm;
It will be that the probability of true Lung neoplasm is more than or equal to the lung knot of probability threshold value in the multiple Lung neoplasm candidate item
Section candidate item is determined as Lung neoplasm candidate item to be detected;
It is determined from the lung CT image data based on the corresponding tubercle coordinate of the Lung neoplasm candidate item to be detected
The corresponding destination image data of the Lung neoplasm candidate item to be detected, the size of data of the destination image data and the lung knot
The size of data for saving the input data of disaggregated model is identical.
In one embodiment, the acquiring unit 301, is also used to obtain training data, and the training data includes more
The corresponding original data of a Lung neoplasm candidate item, each of the multiple Lung neoplasm candidate item Lung neoplasm candidate item are
The probability of true Lung neoplasm is all larger than or is equal to probability threshold value;
The processing unit 302 is also used to carry out data cutout processing to the training data, obtains the Lung neoplasm point
The corresponding target training data of class model, the input of the size of data of the target training data and the Lung neoplasm disaggregated model
The size of data of data is identical;
The training unit 303, for being trained using the target training data to the Lung neoplasm disaggregated model,
Lung neoplasm disaggregated model after being trained.
In one embodiment, the target training data includes positive sample training data and negative sample training data, institute
Training unit 303 is stated, is specifically used for:
The positive sample training data is determined from the target training data;
Data enhancing processing is carried out to the positive sample training data, obtains first object positive sample training data;
Classified using the first object positive sample training data and the negative sample training data to the Lung neoplasm
Model is trained, and obtains the first Lung neoplasm base categories model;
Based on the first object positive sample training data, the negative sample training data and the first Lung neoplasm base
Plinth disaggregated model, the Lung neoplasm disaggregated model after being trained;
Wherein, the ratio data between the first object positive sample training data and the negative sample training data is the
One ratio.
In one embodiment, the training unit 303, is specifically used for:
Data screening is carried out to the negative sample training data, obtains first object negative sample training data;
Using the first object positive sample training data and the first object negative sample training data to described
One Lung neoplasm base categories model is trained, and obtains the second Lung neoplasm base categories model;
Based on the first object positive sample training data, the first object negative sample training data and described second
Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained;
Wherein, the data between the first object positive sample training data and the first object negative sample training data
Ratio is the second ratio.
In one embodiment, the training unit 303, is specifically used for:
Data screening is carried out to the first object positive sample training data, obtains the second target positive sample training data;
Data screening is carried out to the first object negative sample training data, obtains the second target negative sample training data;
Using the second target positive sample training data and the second target negative sample training data to described
Two Lung neoplasm base categories models are trained, the Lung neoplasm disaggregated model after being trained;
Wherein, the data between the second target positive sample training data and the second target negative sample training data
Ratio is second ratio.
Specifically, the Lung neoplasm detection device 300 can be realized in above-mentioned Fig. 1 or embodiment illustrated in fig. 2 by said units
Pulmonary nodule detection method in some or all of step.It should be understood that the embodiment of the present invention is the device of corresponding method embodiment
Embodiment, the description to embodiment of the method, is also applied for the embodiment of the present invention.
In the embodiment of the present invention, determine that Lung neoplasm to be detected is waited based on lung CT image data and Lung neoplasm detection model
After option and the corresponding destination image data of Lung neoplasm candidate item to be detected, recycle Lung neoplasm disaggregated model to target image
Data are handled to obtain classification results, and determine the corresponding target detection of Lung neoplasm candidate item to be detected based on classification results
As a result, so as to effectively improve the classification accuracy of Lung neoplasm, the false positive of Lung neoplasm in testing result is effectively reduced.
Fig. 4 is referred to, Fig. 4 is a kind of structural schematic diagram of server provided in an embodiment of the present invention.The server is used for
Execute above-mentioned method.As shown in figure 4, the server 400 in the present embodiment may include: 401 He of one or more processors
Memory 402.Optionally, which may also include one or more communication interfaces 403.Above-mentioned processor 401, communication interface
403 and memory 402 can be connected by bus 404, or can connect, be carried out in Fig. 4 with bus mode by other means
It illustrates.
Wherein, the processor 401 can be central processing unit (Central Processing Unit, CPU), should
Processor can also be other general processors, digital signal processor (Digital Signal Processor, DSP), specially
With integrated circuit (Application Specific Integrated Circuit, ASIC), ready-made programmable gate array
(Field-Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor are patrolled
Collect device, discrete hardware components etc..General processor can be microprocessor or the processor be also possible to it is any conventional
Processor etc..
The communication interface 403 can be used for receiving and sending messages or the interaction of signaling and the reception of signal and transmitting, communication connect
Mouth 403 may include receiver and transmitter, for being communicated with other equipment.The memory 402 can mainly include storage
Program area and storage data area, wherein storing program area can storage program needed for storage program area, at least one function
(such as text store function, position store function etc.);Storage data area, which can be stored, uses created number according to server
It according to (such as image data, lteral data) etc., and may include application memory program etc..In addition, memory 402 may include height
Fast random access memory, can also include nonvolatile memory, a for example, at least disk memory, flush memory device,
Or other volatile solid-state parts.
The memory 402 is also used to store program instruction.The processor 401 can call above-mentioned memory 402 to deposit
The program instruction of storage realizes pulmonary nodule detection method as shown in the embodiment of the invention.
Wherein, processor 401 can be used for calling described program instruction execution following steps: by the communication interface 403
Lung CT image data are obtained, and determine that Lung neoplasm to be detected is waited based on the lung CT image data and Lung neoplasm detection model
Option and the corresponding destination image data of the Lung neoplasm candidate item to be detected;Using Lung neoplasm disaggregated model to the target
Image data is handled, and the corresponding classification results of the Lung neoplasm candidate item to be detected are obtained;It is true based on the classification results
Determine the corresponding object detection results of the Lung neoplasm candidate item to be detected.
Wherein, the Lung neoplasm disaggregated model includes the first disaggregated model and the second disaggregated model, the first classification mould
Type is different with the network structure of second disaggregated model;The classification results include the first classification results and the second classification knot
Fruit;Processor 401 call described program instruction execution described in using Lung neoplasm disaggregated model to the destination image data into
Row processing specifically executes following steps: by the target when obtaining the corresponding classification results of the Lung neoplasm candidate item to be detected
Image data is inputted in first disaggregated model and is handled, and obtains the Lung neoplasm candidate item to be detected corresponding described the
One classification results;The destination image data is inputted in second disaggregated model and is handled, the lung to be detected is obtained
Corresponding second classification results of nodule candidates.
In one embodiment, it is true lung that first classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the first probability value, and it is true lung that second classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of tubercle is the second probability value;Processor 401 call described program instruction execution described in based on the classification results it is true
When the corresponding object detection results of the fixed Lung neoplasm candidate item to be detected, specifically executes following steps: it is general to obtain described first
Most probable value in rate value and second probability value;The most probable value is determined as the Lung neoplasm to be detected to wait
The corresponding object detection results of option.
In one embodiment, the Lung neoplasm disaggregated model further includes third disaggregated model, the third disaggregated model
It is different with the size of data of input data of first disaggregated model;The classification results further include third classification results;Place
Manage device 401 call described program instruction execution described in using Lung neoplasm disaggregated model to the destination image data at
Reason is also used to execute following steps: by the target figure when obtaining the corresponding classification results of the Lung neoplasm candidate item to be detected
It is handled as data input in the third disaggregated model, obtains the corresponding third of the Lung neoplasm candidate item to be detected
Classification results, it is that the third classification results, which are used to indicate the probability that the Lung neoplasm candidate item to be detected is true Lung neoplasm,
Three probability values;Processor 401 based on the classification results determines the lung to be detected described in described program instruction execution calling
When the corresponding object detection results of nodule candidates, following steps are specifically executed: obtaining first probability value, described second general
Most probable value in rate value and the third probability value;The most probable value is determined as the Lung neoplasm to be detected to wait
The corresponding object detection results of option.
In one embodiment, processor 401 is calling described in described program instruction execution based on the lung CT image
Data and Lung neoplasm detection model determine Lung neoplasm candidate item to be detected and the corresponding mesh of the Lung neoplasm candidate item to be detected
When logo image data, following steps are specifically executed: the lung CT image data being carried out using the Lung neoplasm detection model
Detection obtains initial detecting as a result, the initial detecting result includes multiple Lung neoplasm candidate items, the multiple Lung neoplasm candidate
Each Lung neoplasm candidate item is the probability and corresponding tubercle coordinate of true Lung neoplasm in;By the multiple Lung neoplasm
Be in candidate item true Lung neoplasm probability be more than or equal to probability threshold value Lung neoplasm candidate item be determined as lung knot to be detected
Save candidate item;It is determined from the lung CT image data based on the corresponding tubercle coordinate of the Lung neoplasm candidate item to be detected
The corresponding destination image data of the Lung neoplasm candidate item to be detected, the size of data of the destination image data and the lung knot
The size of data for saving the input data of disaggregated model is identical.
In one embodiment, processor 401 can also call described program instruction execution following steps: by the communication
Interface 403 obtains training data, and the training data includes the corresponding original data of multiple Lung neoplasm candidate items, described more
Each of a Lung neoplasm candidate item Lung neoplasm candidate item is that the probability of true Lung neoplasm is all larger than or is equal to probability threshold value;
Data cutout processing is carried out to the training data, obtains the corresponding target training data of the Lung neoplasm disaggregated model, it is described
The size of data of target training data is identical as the size of data of input data of the Lung neoplasm disaggregated model;Utilize the mesh
Mark training data is trained the Lung neoplasm disaggregated model, the Lung neoplasm disaggregated model after being trained.
In one embodiment, the target training data includes positive sample training data and negative sample training data, place
Reason device 401 using the target training data is carried out the Lung neoplasm disaggregated model described in described program instruction execution calling
Training when Lung neoplasm disaggregated model after being trained, specifically executes following steps: determining from the target training data
The positive sample training data;Data enhancing processing is carried out to the positive sample training data, obtains first object positive sample instruction
Practice data;Using the first object positive sample training data and the negative sample training data to Lung neoplasm classification mould
Type is trained, and obtains the first Lung neoplasm base categories model;Based on the first object positive sample training data, the negative sample
This training data and the first Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained;Wherein, described
Ratio data between first object positive sample training data and the negative sample training data is the first ratio.
In one embodiment, processor 401 call described program instruction execution described in based on the first object just
Sample training data, the negative sample training data and the first Lung neoplasm base categories model, the lung after being trained
When tubercle disaggregated model, following steps are specifically executed: data screening being carried out to the negative sample training data, obtains first object
Negative sample training data;Utilize the first object positive sample training data and the first object negative sample training data pair
The first Lung neoplasm base categories model is trained, and obtains the second Lung neoplasm base categories model;Based on first mesh
Positive sample training data, the first object negative sample training data and the second Lung neoplasm base categories model are marked, is obtained
Lung neoplasm disaggregated model after to training;Wherein, the first object positive sample training data and the first object negative sample
Ratio data between training data is the second ratio.
In one embodiment, processor 401 call described program instruction execution described in based on the first object just
Sample training data, the first object negative sample training data and the second Lung neoplasm base categories model, are instructed
When Lung neoplasm disaggregated model after white silk, following steps are specifically executed: data are carried out to the first object positive sample training data
Screening, obtains the second target positive sample training data;Data screening is carried out to the first object negative sample training data, is obtained
Second target negative sample training data;It is instructed using the second target positive sample training data and the second target negative sample
Practice data to be trained the second Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained;Wherein, institute
Stating the ratio data between the second target positive sample training data and the second target negative sample training data is described second
Ratio.
In the specific implementation, processor 401 described in the embodiment of the present invention etc. can be performed it is above-mentioned shown in fig. 1 or fig. 2
The implementation of each unit described in Fig. 3 of the embodiment of the present invention also can be performed in implementation described in embodiment of the method,
It does not repeat herein.
In the embodiment of the present invention, determine that Lung neoplasm to be detected is waited based on lung CT image data and Lung neoplasm detection model
After option and the corresponding destination image data of Lung neoplasm candidate item to be detected, recycle Lung neoplasm disaggregated model to target image
Data are handled to obtain classification results, and determine the corresponding target detection of Lung neoplasm candidate item to be detected based on classification results
As a result, so as to effectively improve the classification accuracy of Lung neoplasm, the false positive of Lung neoplasm in testing result is effectively reduced.
The embodiment of the invention also provides a kind of computer readable storage medium, the computer-readable recording medium storage
There is computer program, lung described in embodiment corresponding to Fig. 1 or Fig. 2 can be realized when the computer program is executed by processor
Step some or all of in nodule detection methods can also realize the Lung neoplasm detection device of embodiment illustrated in fig. 3 of the present invention
Function can also realize the function of the server of embodiment illustrated in fig. 4 of the present invention, not repeat herein.
The computer readable storage medium can be Lung neoplasm detection device or server described in previous embodiment
Internal storage unit, such as the hard disk or memory of Lung neoplasm detection device or server.The computer-readable storage medium
Matter is also possible to the External memory equipment of the Lung neoplasm detection device or server, for example, the Lung neoplasm detection device or
The plug-in type hard disk being equipped on person's server, intelligent memory card (Smart Media Card, SMC), secure digital (Secure
Digital, SD) card, flash card (Flash Card) etc..
The embodiment of the invention also provides a kind of computer program products comprising instruction, when it runs on computers
When, so that step some or all of in the computer execution above method.
In this application, term "and/or", only a kind of incidence relation for describing affiliated partner, indicates may exist
Three kinds of relationships, for example, A and/or B, can indicate: individualism A exists simultaneously A and B, these three situations of individualism B.Separately
Outside, character "/" herein typicallys represent the relationship that forward-backward correlation object is a kind of "or".
In the various embodiments of the application, magnitude of the sequence numbers of the above procedures are not meant to the elder generation of execution sequence
Afterwards, the execution sequence of each process should be determined by its function and internal logic, the implementation process structure without coping with the embodiment of the present invention
At any restriction.
The above, some embodiments only of the invention, but scope of protection of the present invention is not limited thereto, and it is any
Those familiar with the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or replace
It changes, these modifications or substitutions should be covered by the protection scope of the present invention.
Claims (10)
1. a kind of pulmonary nodule detection method based on disaggregated model, which is characterized in that the described method includes:
Lung CT image data are obtained, and determine lung knot to be detected based on the lung CT image data and Lung neoplasm detection model
Save candidate item and the corresponding destination image data of the Lung neoplasm candidate item to be detected;
The destination image data is handled using Lung neoplasm disaggregated model, obtains the Lung neoplasm candidate item pair to be detected
The classification results answered;
The corresponding object detection results of the Lung neoplasm candidate item to be detected are determined based on the classification results;
Wherein, the Lung neoplasm disaggregated model include the first disaggregated model and the second disaggregated model, first disaggregated model and
The network structure of second disaggregated model is different;The classification results include the first classification results and the second classification results;Institute
It states and the destination image data is handled using Lung neoplasm disaggregated model, it is corresponding to obtain the Lung neoplasm candidate item to be detected
Classification results, comprising:
The destination image data is inputted in first disaggregated model and is handled, it is candidate to obtain the Lung neoplasm to be detected
Corresponding first classification results of item;
The destination image data is inputted in second disaggregated model and is handled, it is candidate to obtain the Lung neoplasm to be detected
Corresponding second classification results of item.
2. the method according to claim 1, wherein first classification results are used to indicate the lung to be detected
Nodule candidates are that the probability of true Lung neoplasm is the first probability value, and second classification results are used to indicate the lung to be detected
Nodule candidates are that the probability of true Lung neoplasm is the second probability value;
It is described to determine the corresponding object detection results of the Lung neoplasm candidate item to be detected based on the classification results, comprising:
Obtain the most probable value in first probability value and second probability value;
The most probable value is determined as the corresponding object detection results of the Lung neoplasm candidate item to be detected.
3. the method according to claim 1, wherein first classification results are used to indicate the lung to be detected
Nodule candidates are that the probability of true Lung neoplasm is the first probability value, and second classification results are used to indicate the lung to be detected
Nodule candidates are that the probability of true Lung neoplasm is the second probability value;The Lung neoplasm disaggregated model further includes third classification mould
The size of data of type, the third disaggregated model and the input data of first disaggregated model is different;The classification results are also
Including third classification results;
It is described that the destination image data is handled using Lung neoplasm disaggregated model, it is candidate to obtain the Lung neoplasm to be detected
The corresponding classification results of item, further includes:
The destination image data is inputted in the third disaggregated model and is handled, it is candidate to obtain the Lung neoplasm to be detected
The corresponding third classification results of item, it is true that the third classification results, which are used to indicate the Lung neoplasm candidate item to be detected,
The probability of Lung neoplasm is third probability value;
It is described to determine the corresponding object detection results of the Lung neoplasm candidate item to be detected based on the classification results, comprising:
Obtain the most probable value in first probability value, second probability value and the third probability value;
The most probable value is determined as the corresponding object detection results of the Lung neoplasm candidate item to be detected.
4. according to the method in any one of claims 1 to 3, which is characterized in that described to be based on the lung CT image number
Lung neoplasm candidate item to be detected and the corresponding target of the Lung neoplasm candidate item to be detected are determined according to Lung neoplasm detection model
Image data, comprising:
The lung CT image data are detected using the Lung neoplasm detection model, obtain initial detecting as a result, described
Initial detecting result includes multiple Lung neoplasm candidate items, each Lung neoplasm candidate item is true in the multiple Lung neoplasm candidate item
The probability of real Lung neoplasm and corresponding tubercle coordinate;
It will be that the probability of true Lung neoplasm is more than or equal to the Lung neoplasm time of probability threshold value in the multiple Lung neoplasm candidate item
Option is determined as Lung neoplasm candidate item to be detected;
It is determined from the lung CT image data based on the corresponding tubercle coordinate of the Lung neoplasm candidate item to be detected described
The corresponding destination image data of Lung neoplasm candidate item to be detected, the size of data of the destination image data and the Lung neoplasm point
The size of data of the input data of class model is identical.
5. according to the method in any one of claims 1 to 3, which is characterized in that the method also includes:
Training data is obtained, the training data includes the corresponding original data of multiple Lung neoplasm candidate items, the multiple
Each of Lung neoplasm candidate item Lung neoplasm candidate item is that the probability of true Lung neoplasm is all larger than or is equal to probability threshold value;
Data cutout processing is carried out to the training data, obtains the corresponding target training data of the Lung neoplasm disaggregated model,
The size of data of the target training data is identical as the size of data of input data of the Lung neoplasm disaggregated model;
The Lung neoplasm disaggregated model is trained using the target training data, the Lung neoplasm classification mould after being trained
Type.
6. according to the method described in claim 5, it is characterized in that, the target training data include positive sample training data and
Negative sample training data, it is described that the Lung neoplasm disaggregated model is trained using the target training data, it is trained
Lung neoplasm disaggregated model afterwards, comprising:
The positive sample training data is determined from the target training data;
Data enhancing processing is carried out to the positive sample training data, obtains first object positive sample training data;
Using the first object positive sample training data and the negative sample training data to the Lung neoplasm disaggregated model
It is trained, obtains the first Lung neoplasm base categories model;
Based on the first object positive sample training data, the negative sample training data and first Lung neoplasm basis point
Class model, the Lung neoplasm disaggregated model after being trained;
Wherein, the ratio data between the first object positive sample training data and the negative sample training data is the first ratio
Example.
7. according to the method described in claim 6, it is characterized in that, it is described based on the first object positive sample training data,
The negative sample training data and the first Lung neoplasm base categories model, the Lung neoplasm disaggregated model after being trained,
Include:
Data screening is carried out to the negative sample training data, obtains first object negative sample training data;
Using the first object positive sample training data and the first object negative sample training data to first lung
Tubercle base categories model is trained, and obtains the second Lung neoplasm base categories model;
Based on the first object positive sample training data, the first object negative sample training data and the second lung knot
Save base categories model, the Lung neoplasm disaggregated model after being trained;
Wherein, the ratio data between the first object positive sample training data and the first object negative sample training data
For the second ratio.
8. the method according to the description of claim 7 is characterized in that it is described based on the first object positive sample training data,
The first object negative sample training data and the second Lung neoplasm base categories model, the Lung neoplasm point after being trained
Class model, comprising:
Data screening is carried out to the first object positive sample training data, obtains the second target positive sample training data;
Data screening is carried out to the first object negative sample training data, obtains the second target negative sample training data;
Using the second target positive sample training data and the second target negative sample training data to second lung
Tubercle base categories model is trained, the Lung neoplasm disaggregated model after being trained;
Wherein, the ratio data between the second target positive sample training data and the second target negative sample training data
For second ratio.
9. a kind of server, which is characterized in that including processor, communication interface and memory, the processor, the communication are connect
Mouth and the memory are connected with each other, wherein for the memory for storing computer program, the computer program includes journey
Sequence instruction, the processor are configured for calling described program instruction, execute as described in any item of the claim 1 to 8
Pulmonary nodule detection method based on disaggregated model.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer journey
Sequence, the computer program include program instruction, and described program instruction executes the processor such as
Pulmonary nodule detection method described in any item of the claim 1 to 8 based on disaggregated model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910119541.9A CN109961423A (en) | 2019-02-15 | 2019-02-15 | A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910119541.9A CN109961423A (en) | 2019-02-15 | 2019-02-15 | A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109961423A true CN109961423A (en) | 2019-07-02 |
Family
ID=67023830
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910119541.9A Pending CN109961423A (en) | 2019-02-15 | 2019-02-15 | A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109961423A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781805A (en) * | 2019-10-23 | 2020-02-11 | 上海极链网络科技有限公司 | Target object detection method, device, computing equipment and medium |
CN111340827A (en) * | 2020-05-18 | 2020-06-26 | 天津精诊医疗科技有限公司 | Lung CT image data processing and analyzing method and system |
CN111461220A (en) * | 2020-04-01 | 2020-07-28 | 腾讯科技(深圳)有限公司 | Image analysis method, image analysis device, and image analysis system |
CN111709371A (en) * | 2020-06-17 | 2020-09-25 | 腾讯科技(深圳)有限公司 | Artificial intelligence based classification method, device, server and storage medium |
CN111932564A (en) * | 2020-09-24 | 2020-11-13 | 平安科技(深圳)有限公司 | Picture identification method and device, electronic equipment and computer readable storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106940816A (en) * | 2017-03-22 | 2017-07-11 | 杭州健培科技有限公司 | Connect the CT image Lung neoplasm detecting systems of convolutional neural networks entirely based on 3D |
CN107274451A (en) * | 2017-05-17 | 2017-10-20 | 北京工业大学 | Isolator detecting method and device based on shared convolutional neural networks |
CN108133476A (en) * | 2017-12-26 | 2018-06-08 | 安徽科大讯飞医疗信息技术有限公司 | A kind of Lung neoplasm automatic testing method and system |
CN108230323A (en) * | 2018-01-30 | 2018-06-29 | 浙江大学 | A kind of Lung neoplasm false positive screening technique based on convolutional neural networks |
CN108830188A (en) * | 2018-05-30 | 2018-11-16 | 西安理工大学 | Vehicle checking method based on deep learning |
CN108986085A (en) * | 2018-06-28 | 2018-12-11 | 深圳视见医疗科技有限公司 | CT image pulmonary nodule detection method, device, equipment and readable storage medium storing program for executing |
CN108986067A (en) * | 2018-05-25 | 2018-12-11 | 上海交通大学 | Pulmonary nodule detection method based on cross-module state |
CN109003260A (en) * | 2018-06-28 | 2018-12-14 | 深圳视见医疗科技有限公司 | CT image pulmonary nodule detection method, device, equipment and readable storage medium storing program for executing |
CN109300530A (en) * | 2018-08-08 | 2019-02-01 | 北京肿瘤医院 | The recognition methods of pathological picture and device |
-
2019
- 2019-02-15 CN CN201910119541.9A patent/CN109961423A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106940816A (en) * | 2017-03-22 | 2017-07-11 | 杭州健培科技有限公司 | Connect the CT image Lung neoplasm detecting systems of convolutional neural networks entirely based on 3D |
CN107274451A (en) * | 2017-05-17 | 2017-10-20 | 北京工业大学 | Isolator detecting method and device based on shared convolutional neural networks |
CN108133476A (en) * | 2017-12-26 | 2018-06-08 | 安徽科大讯飞医疗信息技术有限公司 | A kind of Lung neoplasm automatic testing method and system |
CN108230323A (en) * | 2018-01-30 | 2018-06-29 | 浙江大学 | A kind of Lung neoplasm false positive screening technique based on convolutional neural networks |
CN108986067A (en) * | 2018-05-25 | 2018-12-11 | 上海交通大学 | Pulmonary nodule detection method based on cross-module state |
CN108830188A (en) * | 2018-05-30 | 2018-11-16 | 西安理工大学 | Vehicle checking method based on deep learning |
CN108986085A (en) * | 2018-06-28 | 2018-12-11 | 深圳视见医疗科技有限公司 | CT image pulmonary nodule detection method, device, equipment and readable storage medium storing program for executing |
CN109003260A (en) * | 2018-06-28 | 2018-12-14 | 深圳视见医疗科技有限公司 | CT image pulmonary nodule detection method, device, equipment and readable storage medium storing program for executing |
CN109300530A (en) * | 2018-08-08 | 2019-02-01 | 北京肿瘤医院 | The recognition methods of pathological picture and device |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781805A (en) * | 2019-10-23 | 2020-02-11 | 上海极链网络科技有限公司 | Target object detection method, device, computing equipment and medium |
CN111461220A (en) * | 2020-04-01 | 2020-07-28 | 腾讯科技(深圳)有限公司 | Image analysis method, image analysis device, and image analysis system |
CN111461220B (en) * | 2020-04-01 | 2022-11-01 | 腾讯科技(深圳)有限公司 | Image analysis method, image analysis device, and image analysis system |
CN111340827A (en) * | 2020-05-18 | 2020-06-26 | 天津精诊医疗科技有限公司 | Lung CT image data processing and analyzing method and system |
CN111709371A (en) * | 2020-06-17 | 2020-09-25 | 腾讯科技(深圳)有限公司 | Artificial intelligence based classification method, device, server and storage medium |
CN111709371B (en) * | 2020-06-17 | 2023-12-22 | 腾讯科技(深圳)有限公司 | Classification method, device, server and storage medium based on artificial intelligence |
CN111932564A (en) * | 2020-09-24 | 2020-11-13 | 平安科技(深圳)有限公司 | Picture identification method and device, electronic equipment and computer readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109961423A (en) | A kind of pulmonary nodule detection method based on disaggregated model, server and storage medium | |
CN110807429B (en) | Construction safety detection method and system based on tiny-YOLOv3 | |
CN109815156A (en) | Displaying test method, device, equipment and the storage medium of visual element in the page | |
US9349076B1 (en) | Template-based target object detection in an image | |
CN108520229A (en) | Image detecting method, device, electronic equipment and computer-readable medium | |
CN111428726A (en) | Panorama segmentation method, system, equipment and storage medium based on graph neural network | |
CN108985199A (en) | Detection method, device and the storage medium of commodity loading or unloading operation | |
CN109271762B (en) | User authentication method and device based on slider verification code | |
CN109039503A (en) | A kind of frequency spectrum sensing method, device, equipment and computer readable storage medium | |
US20220207266A1 (en) | Methods, devices, electronic apparatuses and storage media of image processing | |
US20130135446A1 (en) | Street view creating system and method thereof | |
CN110910334B (en) | Instance segmentation method, image processing device and computer readable storage medium | |
CN110162454A (en) | Game running method and device, storage medium and electronic device | |
CN110070533A (en) | A kind of evaluating method of object detection results, device, equipment and storage medium | |
CN110335313A (en) | Audio collecting device localization method and device, method for distinguishing speek person and system | |
CN109389105A (en) | A kind of iris detection and viewpoint classification method based on multitask | |
CN110210480A (en) | Character recognition method, device, electronic equipment and computer readable storage medium | |
CN110415237A (en) | Skin blemishes detection method, detection device, terminal device and readable storage medium storing program for executing | |
CN114926726A (en) | Unmanned ship sensing method based on multitask network and related equipment | |
CN110298302A (en) | A kind of human body target detection method and relevant device | |
CN109961103A (en) | The training method of Feature Selection Model, the extracting method of characteristics of image and device | |
AU2021204584A1 (en) | Methods, apparatuses, devices and storage media for detecting correlated objects involved in image | |
CN112464785A (en) | Target detection method and device, computer equipment and storage medium | |
CN112215271A (en) | Anti-occlusion target detection method and device based on multi-head attention mechanism | |
CN115546824B (en) | Taboo picture identification method, apparatus and storage medium |
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 |