CN110189318A - Pulmonary nodule detection method and system with semantic feature score - Google Patents
Pulmonary nodule detection method and system with semantic feature score Download PDFInfo
- Publication number
- CN110189318A CN110189318A CN201910465182.2A CN201910465182A CN110189318A CN 110189318 A CN110189318 A CN 110189318A CN 201910465182 A CN201910465182 A CN 201910465182A CN 110189318 A CN110189318 A CN 110189318A
- Authority
- CN
- China
- Prior art keywords
- network
- semantic feature
- lung neoplasm
- score
- interest
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 230000002685 pulmonary effect Effects 0.000 title claims abstract description 19
- 208000020816 lung neoplasm Diseases 0.000 claims abstract description 117
- 238000000034 method Methods 0.000 claims abstract description 13
- 238000010276 construction Methods 0.000 claims abstract description 8
- 238000004891 communication Methods 0.000 claims description 15
- 210000004072 lung Anatomy 0.000 claims description 15
- 230000000505 pernicious effect Effects 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 8
- 238000009434 installation Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 206010058467 Lung neoplasm malignant Diseases 0.000 description 4
- 201000005202 lung cancer Diseases 0.000 description 4
- 230000002308 calcification Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 206010028980 Neoplasm Diseases 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 201000011510 cancer Diseases 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 208000019693 Lung disease Diseases 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000007850 degeneration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000013100 final test Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
Classifications
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The present invention relates to a kind of pulmonary nodule detection method and system with semantic feature score.It specifically includes that construction feature extractor network, region suggest network, the first semantic feature network, sorter network and the second semantic feature network, network, the first semantic feature network, sorter network and the second semantic feature network, which are trained, is suggested to the feature extractor network, region；Chest CT image to be evaluated is inputted into trained feature extractor network, and obtains the classification of Lung neoplasm position and the Lung neoplasm by sorter network, the semantic feature score of Lung neoplasm is obtained by the second semantic feature network.The above method can be handled chest CT image by corresponding network, and the classification of Lung neoplasm position and the Lung neoplasm is obtained by sorter network, and the semantic feature score of Lung neoplasm is obtained by the second semantic feature network.The semantic feature score of Lung neoplasm can be obtained automatically.
Description
Technical field
The present invention relates to the Lung neoplasm detection technique fields of CT image, more particularly to the lung knot for having semantic feature score
Save detection method and system.
Background technique
Lung cancer is the No.1 cancer in all cancers.In all malignant tumours, have 13% the sick hair rate of height and
19.5% high mortality.The five year survival rate of advanced lung cancer only has 16%, but if can be made a definite diagnosis in early days, five Nian Shengcun
Rate can be improved to 70%.Lung cancer early stage shows as nodule form, and pulmonary nodule is intrapulmonary in irregular and subcircular disease
Stove region.In these focal areas, there is 30% to show as the positive, therefore Lung neoplasm detects the discovery to lung cancer early stage to Guan Chong
It wants.
With the development of big data era, deep learning slowly becomes a kind of trend in the application of medical field.Its
In the target detection based on deep learning have become Lung neoplasm positioning with classification main means.
It, can be according to texture, fineness, spicule sign, circularity, boundary, sign of lobulation, interior when checking lung CT image due to doctor
The characterization semantic feature such as portion's structure and calcification to position Lung neoplasm and classify, and the target detection technique based on deep learning
Only Lung neoplasm is positioned and classified, in the case where no semantic feature score, such testing result can not make doctor
Convince.
Summary of the invention
Based on this, a kind of pulmonary nodule detection method with semantic feature score is provided.It aims to solve the problem that in the prior art only
Lung neoplasm is positioned and is classified, the technical issues of without semantic feature score.
A kind of pulmonary nodule detection method with semantic feature score, which comprises
Construction feature extractor network, region suggest that network, the first semantic feature network, sorter network and second are semantic special
Network is levied,
Wherein,
The feature extractor network, for the chest CT image of Lung neoplasm will to be had as the defeated of feature extractor network
Enter, and extracts the characteristic pattern of the chest CT image；
Network is suggested in the region, for suggesting the input of network using characteristic pattern as region, and obtains area-of-interest；
The first semantic feature network for using characteristic pattern as the input of the first semantic feature network, and obtains institute
State the semantic feature score of area-of-interest；
The sorter network, for using the area-of-interest as the input of sorter network, and to area-of-interest
Position is finely adjusted, and obtains the classification of Lung neoplasm position and the Lung neoplasm；
The second semantic feature network, for using the semantic feature score of area-of-interest as the second semantic feature net
The input of network, and the semantic feature score of area-of-interest is finely adjusted, obtain the semantic feature score of Lung neoplasm；
Network, the first semantic feature network, sorter network and second are semantic to be suggested to the feature extractor network, region
Character network is trained；
Chest CT image to be evaluated is inputted into trained feature extractor network, and Lung neoplasm is obtained by sorter network
The classification of position and the Lung neoplasm is obtained the semantic feature score of Lung neoplasm by the second semantic feature network.
The above method can be handled chest CT image by corresponding network, and obtain Lung neoplasm position by sorter network
And the classification of the Lung neoplasm, the semantic feature score of Lung neoplasm is obtained by the second semantic feature network.Lung can be obtained automatically
The semantic feature score of tubercle.
The region suggests being provided with 3 anchor boxes that area ratio is 1:4 in network in one of the embodiments,.
In one of the embodiments, the sorter network for Lung neoplasm it is good it is pernicious classify, described second is semantic
Character network obtains the score of one group of semantic feature.
The position of the Lung neoplasm is position coordinates, the lung of Lung neoplasm in CT image in one of the embodiments,
Tubercle is classified as the good pernicious classification of the Lung neoplasm, and the semantic feature of the Lung neoplasm is scored at for the Lung neoplasm
One group of semantic feature score.
A kind of Lung neoplasm detection system with semantic feature score, the system comprises:
The characteristic pattern of CT image obtains module, and the characteristic pattern of the CT image obtains module and is used for construction feature extractor net
Network using the chest CT image with Lung neoplasm as the input of the network, and extracts the characteristic pattern of CT image；
Area-of-interest obtains module, and the area-of-interest obtains module and suggests network for constructing region, by feature
Scheme the input for suggesting network as region, and obtains area-of-interest；
First semantic feature score obtains module, and it is semantic for constructing first that the first semantic feature score obtains module
Character network using characteristic pattern as the input of the first semantic feature network, and obtains the semantic feature score of area-of-interest；
Lung neoplasm position and classification obtain module, and the Lung neoplasm position and classification obtain module for constructing classification net
Network, and using area-of-interest as the input of sorter network, the position of area-of-interest is finely adjusted, Lung neoplasm position is obtained
With the classification of the Lung neoplasm；
Second semantic feature score obtains module, and it is semantic for constructing second that the second semantic feature score obtains module
Character network, and using area-of-interest as the input of the second semantic feature network, to the semantic feature score of area-of-interest
It is finely adjusted, obtains the semantic feature score of Lung neoplasm.
The region suggests being provided with 3 anchor boxes that area ratio is 1:4 in network in one of the embodiments,.
In one of the embodiments, the sorter network for Lung neoplasm it is good it is pernicious classify, the semantic feature
Network obtains the score of one group of semantic feature.
The position of the Lung neoplasm is position coordinates, the lung of Lung neoplasm in CT image in one of the embodiments,
Tubercle is classified as the good pernicious classification of the Lung neoplasm, and the semantic feature of the Lung neoplasm is scored at for the Lung neoplasm
One group of semantic feature score.
A kind of computer storage medium is stored with an at least executable instruction, the executable finger in the storage medium
Enabling makes processor execute the corresponding operation of pulmonary nodule detection method.
A kind of computer installation, comprising: processor, memory, communication interface and communication bus, the processor, storage
Device and communication interface complete mutual communication by the communication bus, and the memory is for storing at least one executable finger
It enables, the executable instruction makes the processor execute the corresponding operation of pulmonary nodule detection method.
Detailed description of the invention
Fig. 1 is the flow chart of the pulmonary nodule detection method of the embodiment of the present invention.
Fig. 2 is the operational flow diagram for each network that the pulmonary nodule detection method of the embodiment of the present invention is applied to.
Fig. 3 is that the schematic diagram of 12 anchor boxes in a pixel of network phase is suggested in the region of the embodiment of the present invention.
Fig. 4 is the area-of-interest schematic diagram of the sorting phase of the embodiment of the present invention.
Fig. 5 is the final detection result schematic diagram of the embodiment of the present invention.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing to the present invention
Specific embodiment be described in detail.Many details are explained in the following description in order to fully understand this hair
It is bright.But the invention can be embodied in many other ways as described herein, those skilled in the art can be not
Similar improvement is done in the case where violating intension of the present invention, therefore the present invention is not limited by the specific embodiments disclosed below.
It should be noted that it can directly on the other element when element is referred to as " being fixed on " another element
Or there may also be elements placed in the middle.When an element is considered as " connection " another element, it, which can be, is directly connected to
To another element or it may be simultaneously present centering elements.
Unless otherwise defined, all technical and scientific terms used herein and belong to technical field of the invention
The normally understood meaning of technical staff is identical.Term as used herein in the specification of the present invention is intended merely to description tool
The purpose of the embodiment of body, it is not intended that in the limitation present invention.Term " and or " used herein includes one or more phases
Any and all combinations of the listed item of pass.
As depicted in figs. 1 and 2, described the present invention provides a kind of pulmonary nodule detection method with semantic feature score
Method includes:
Network, the first semantic feature network, sorter network and second are suggested in S100, construction feature extractor network, region
Semantic feature network,
Wherein,
The feature extractor network, for the chest CT image of Lung neoplasm will to be had as the defeated of feature extractor network
Enter, and extracts the characteristic pattern of the chest CT image；
Network is suggested in the region, for suggesting the input of network using characteristic pattern as region, and obtains area-of-interest；
The first semantic feature network for using characteristic pattern as the input of the first semantic feature network, and obtains institute
State the semantic feature score of area-of-interest；
The sorter network, for using the area-of-interest as the input of sorter network, and to area-of-interest
Position is finely adjusted, and obtains the classification of Lung neoplasm position and the Lung neoplasm；
The second semantic feature network, for using the semantic feature score of area-of-interest as the second semantic feature net
The input of network, and the semantic feature score of area-of-interest is finely adjusted, obtain the semantic feature score of Lung neoplasm；
S200, network, the first semantic feature network, sorter network and the are suggested to the feature extractor network, region
Two semantic feature networks are trained；
S300, chest CT image to be evaluated is inputted into trained feature extractor network, and is obtained by sorter network
The classification of Lung neoplasm position and the Lung neoplasm is obtained the semantic feature score of Lung neoplasm by the second semantic feature network.
The above method can be handled chest CT image by corresponding network, and obtain Lung neoplasm position by sorter network
And the classification of the Lung neoplasm, the semantic feature score of Lung neoplasm is obtained by the second semantic feature network.Lung can be obtained automatically
The semantic feature score of tubercle.
It should be noted that using LICD/IDR as the data set of test and training in this method.LICD/IDR be by
Univ Michigan-Ann Arbor USA, University of Iowa, Chicago University, University of California, Cornell University's Weill Medical College are common
Create lung images database alliance (Lung Image Database Consortium, LIDC).In order to further study
Pulmonary disease, United States Institute of Health created image data base resource planning (Image Database in 2004
Resource Initiative, IDRI).LIDC/IDRI image database is established within 2011, it is the maximum lung in the world today
Nodule database.
LIDC/IDRI data set only has the score of 8 tubercle semantic features to the tubercle of > 3mm, in addition to " internal structure "
Score range 1-4 point and the score range of " calcification " 1-6 exceptionally, " texture ", " fineness ", " sign of lobulation ", " boundary ",
The score range of " circularity " and " spicule sign " is at 1-5 points.
The above method of the invention is described in detail below by way of specific example.
As depicted in figs. 1 and 2, step 1, construction feature extractor network, the feature extractor network will be for that will have
Input of the chest CT image of Lung neoplasm as the network, and extract the characteristic pattern of chest CT image.Training this feature extractor
Network mainly determines the size of network layer and the parameter of network layer.
It should be noted that the feature extractor network can be Inception Resnet V2 structural network.
The feature extractor network of Inception Resnet V2 be by residual error network (Residual Network) and
Inception combination of network forms.With the increase of network depth, model will appear the problem of gradient is disappeared with network degeneration,
Residual error network can be very good to solve the problems, such as this by short link.Similarly with the increase of network depth, convolution kernel vector
It certainly will increase, deconvolution parameter amount can also be increase accordingly, and the small convolution of Inception Web vector graphic 1x1 replaces original convolution
Core, to achieve the purpose that accelerate training speed.
Network and the first semantic feature network are suggested in step 2, building region, using characteristic pattern as the input of two networks,
Network and the first semantic feature network are suggested in training region, obtain the semantic feature score of area-of-interest and area-of-interest.
Network is suggested in the region and the first semantic feature network includes multiple convolutional layers and full articulamentum.For example, region
It is recommended that network includes 1 convolutional layer and 2 layers of full articulamentum, 2 layers of full articulamentum, which do the differentiation of prospect respectively and extract, feels emerging
Interesting region.First semantic feature network can use 2 convolutional layers and 2 full articulamentums.It should be noted that in convolution process
In, the padding of pixel filling is set as ' SAME ', the size of input picture will not be changed after convolution.
The region is suggested in network, since Lung neoplasm detection is that wisp detection increases on the basis of existing technology
Area ratio is 3 anchor boxes of 1:4.
As shown in Fig. 2, suggesting, using characteristic pattern as the input of the network, existing by the convolution kernel of 1x1 in network in region
Each pixel on characteristic pattern is slided, and a series of sliding window is generated.To each sliding window, according to area ratio
For 1:4,1:2,1:1,2:1,12 anchor boxes (Anchor) of combination producing of length-width ratio 1:2,1:1,2:1, each pixel
Size is 1x1, and area 1, area ratio is the ratio between anchor box area and pixel point areas, and specific schematic diagram is as shown in Figure 3.Then again
Choose positive negative sample according to the IOU (Intersection over Union) with true frame, IOU>0.7 is positive sample, IOU<
0.3 is negative sample, and then region suggests that network is adjusted by the position that loss function carries out coordinate to anchor box, obtains region of interest
Domain, meanwhile, the first semantic feature network gives a mark to semantic feature by loss function, and the semantic feature for obtaining area-of-interest obtains
Point.
It should be noted that area-of-interest is the output that network is suggested in region, that is, it there may be the region of Lung neoplasm,
As shown in Figure 4.
The size of 12 anchor boxes in 11 pixels of table
Different from 9 anchor boxes in Faster RCNN in the prior art, invention increases 3 that area ratio is 1:4
Anchor box is more suitable for the detection of small size Lung neoplasm.
Step 3, building sorter network and the second semantic feature network, using area-of-interest as the input of sorter network,
Using the semantic feature score of area-of-interest as the input of the second semantic feature network, training sorter network and the second semantic spy
Network is levied, the semantic feature score of position and area-of-interest to area-of-interest is finely adjusted, final to obtain Lung neoplasm position
It sets, the semantic feature score of the tubercle and the classification of the tubercle.
In the sorter network and the second semantic feature network, including multiple convolutional layers and full articulamentum.For example, classification net
Network applies 1 Ge Juan base and 1 full articulamentum.Second convolutional layer of semantic feature network application 1 and 3 full articulamentums.
It should be noted that sorter network for Lung neoplasm it is good it is pernicious classify, semantic feature network obtain 8 semantemes
The score of feature.
Specifically, in step 2, the position of area-of-interest and the semanteme of area-of-interest are determined by loss function
Feature score, and determine fore/background.Specific loss function has the loss of fore/background, the loss of area-of-interest position and lung
The loss three parts of 8 semantic feature scores of tubercle form.In step 3, area-of-interest is finely tuned by loss function
The semantic feature score of position and area-of-interest, and determination is good pernicious.Specific loss function has the loss of Lung neoplasm classification,
The loss three parts of the loss of Lung neoplasm position and 8 semantic feature scores of Lung neoplasm form.The calculating that step 2 and step 3 are used
Formula specifically includes following formula:
Wherein, i is the index of anchor box.In step 2, λ1∑iLcls(pi, pi *) be fore/background loss, wherein λ1=1,
piRepresentative model is predicted as the probability of fore/background, pi *For the true fore/background of tubercle, value is 0 or 1, pi *=1 represents
The tubercle is prospect, pi *=0 tubercle is background.In step 3, λ1∑iLcls(pi, pi *) it is the loss of Lung neoplasm classification, wherein
λ1=1, piRepresentative model is predicted as the probability of positive nodule, pi *For the true classification of tubercle, value is 0 or 1, pi *=1 generation
The table tubercle is pernicious, pi *=0 tubercle is benign.
In step 2,It is the loss of area-of-interest position, wherein λ2=2,It is a vector, represents the coordinate of two points in the prediction block upper left corner and the lower right corner, ti *It represents true
The coordinate of two points in the real frame upper left corner and the lower right corner.In step 3,It is the loss of Lung neoplasm position,
Middle λ2=2,It is a vector, represents the coordinate of two points in the prediction block upper left corner and the lower right corner, ti *
Represent the coordinate of two points in the true frame upper left corner and the lower right corner.
It is the loss of 8 semantic feature scores of Lung neoplasm, wherein λ3=1, ciIt is the 8 of model prediction
A semantic feature score, ci *For the true semantic feature score of Lung neoplasm.
Lcls(pi, pi *)=- log [pipi *+(1-pi)(1-pi *)] (2),
Wherein, L in step 2cls(pi, pi *) lost for the logarithm of fore/background.L in step 3cls(pi, pi *) it is lung knot
Save good pernicious logarithm loss.
Lbbox(ti, ti *)=R (ti-ti *) (3),
Lcha(ci, ci *)=R (ci-ci *) (4),
Wherein, R is smooth L1 function.Compared to smoothL2 function, smoothL1 function can effectively prevent gradient
The appearance of explosion issues has more robustness.Specific formula for calculation are as follows:
In step 2, in the loss of lung area-of-interest positionIn step 3, in the loss of Lung neoplasm positionIn semantic feature loss
In the present embodiment, the appearance of overfitting problem in order to prevent, can by former lung CT figure carry out 90 degree, 180 degree, 270
Flip horizontal and the dropout strategy of degree.The problem of gradient is widely varied in order to prevent has used ladder in the present embodiment
It spends threshold method (clip gradient), specific training parameter is as shown in table 2.
2 parameter setting of table
| Parameter | Setting |
| mini-batch | 1 |
| Dropout probability value | 0.3 |
| The value of Momentum optimizer | 0.9 |
| The learning rate of 1-30000 step | 0.0003 |
| The learning rate of 30001-60000 step | 0.00003 |
| The learning rate of 60001-90000 step | 0.000003 |
| Grads threshold | 10 |
It should be noted that above-mentioned each network can be trained under Tensorflow frame, processor Intel
(R) Core (TM) i7-6850K CPU 3.60GHz*12, inside saves as 11GB, and GPU video card is 1080Ti.Under this configuration surroundings,
The time that training primary (90000 step) is spent is about 9 hours, and the time of each step consumption is 0.6 second.
In order to quantify to verify the validity of published method of the present invention, used accuracy rate, sensibility, specificity and AUC this
Four indexs measure the detection effect of model.Wherein:
Accuracy rate (accuracy rate, AR): the sample number that model is correctly classified accounts for the ratio of total number of samples.
Sensibility (true positive rate, TPR): the positive sample number (true positives) that model identifies accounts for all sun
The ratio of property sample number (summations of true positives and false negative).
Specific (true negative rate, TNR): the negative sample number (true negative) that model identifies accounts for all yin
The ratio of property sample number (summation of false positive and true negative).
In 8 semantic feature scores, the mean absolute error and mark of score are marked using model prediction score and doctor
Quasi- difference is used as evaluation criterion, calculation formula are as follows:
Wherein xiFor model prediction score,The average value of score is marked for doctor.
Wherein xiFor model prediction score,The average value of score is marked for doctor.
The comparison of 38 semantic feature scores of Lung neoplasm of table
Analytical table 3 it can be concluded that, the error between method disclosed by the invention and the error and doctor of doctor is fairly close.
The testing result of the good pernicious classification of Lung neoplasm in the different models of table 4
Analytical table 4 it can be concluded that, method disclosed by the invention improve Faster R-CNN region suggest network, improve
Sensibility and accuracy rate.
In the present embodiment, final testing result is as shown in Figure 5.The circular image that the lower left corner of picture in Fig. 5 outlines
For Lung neoplasm.The semantic feature score of the Lung neoplasm are as follows:
Fineness: 5, internal structure: 1, calcification: 6, circularity: 4, boundary: 4, sign of lobulation: 1, spicule sign: 1, texture: 5.
The present invention also provides a kind of Lung neoplasm detection system with semantic feature score, the system comprises:
The characteristic pattern of CT image obtains module, and the characteristic pattern of the CT image obtains module and is used for construction feature extractor net
Network using the chest CT image with Lung neoplasm as the input of the network, and extracts the characteristic pattern of CT image；
Area-of-interest obtains module, and the area-of-interest obtains module and suggests network for constructing region, by feature
Scheme the input for suggesting network as region, and obtains area-of-interest；
First semantic feature score obtains module, and it is semantic for constructing first that the first semantic feature score obtains module
Character network using characteristic pattern as the input of the first semantic feature network, and obtains the semantic feature score of area-of-interest；
Lung neoplasm position and classification obtain module, and the Lung neoplasm position and classification obtain module for constructing classification net
Network, and using area-of-interest as the input of sorter network, the position of area-of-interest is finely adjusted, Lung neoplasm position is obtained
With the classification of the Lung neoplasm；
Second semantic feature score obtains module, and it is semantic for constructing second that the second semantic feature score obtains module
Character network, and using area-of-interest as the input of the second semantic feature network, to the semantic feature score of area-of-interest
It is finely adjusted, obtains the semantic feature score of Lung neoplasm.
In the present embodiment, the region suggests being provided with 3 anchor boxes that area ratio is 1:4 in network.
In the present embodiment, the sorter network for Lung neoplasm it is good it is pernicious classify, the semantic feature network obtains
The score of one group of semantic feature.
In the present embodiment, the position of the Lung neoplasm is point of the position coordinates of Lung neoplasm in CT image, the Lung neoplasm
Class is the good pernicious classification of the Lung neoplasm, and the semantic feature of the Lung neoplasm is scored at one group of semanteme for the Lung neoplasm
Feature score.
The present invention also provides a kind of computer storage medium, at least one executable finger is stored in the storage medium
It enables, the executable instruction makes processor execute the corresponding operation of pulmonary nodule detection method.
The present invention also provides a kind of computer installations, comprising: processor, memory, communication interface and communication bus, institute
It states processor, memory and communication interface and completes mutual communication by the communication bus, the memory is for storing
An at least executable instruction, the executable instruction make the processor execute the corresponding behaviour of pulmonary nodule detection method
Make.
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection of the invention
Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (10)
1. a kind of pulmonary nodule detection method with semantic feature score, which is characterized in that the described method includes:
Network, the first semantic feature network, sorter network and the second semantic feature net are suggested in construction feature extractor network, region
Network,
Wherein,
The feature extractor network, for input of the chest CT image of Lung neoplasm as feature extractor network will to be had,
And extract the characteristic pattern of the chest CT image；
Network is suggested in the region, for suggesting the input of network using characteristic pattern as region, and obtains area-of-interest；
The first semantic feature network for using characteristic pattern as the input of the first semantic feature network, and obtains the sense
The semantic feature score in interest region；
The sorter network, for using the area-of-interest as the input of sorter network, and to the position of area-of-interest
It is finely adjusted, obtains the classification of Lung neoplasm position and the Lung neoplasm；
The second semantic feature network, for using the semantic feature score of area-of-interest as the second semantic feature network
Input, and the semantic feature score of area-of-interest is finely adjusted, obtain the semantic feature score of Lung neoplasm；
Network, the first semantic feature network, sorter network and the second semantic feature are suggested to the feature extractor network, region
Network is trained；
Chest CT image to be evaluated is inputted into trained feature extractor network, and Lung neoplasm position is obtained by sorter network
And the classification of the Lung neoplasm, the semantic feature score of Lung neoplasm is obtained by the second semantic feature network.
2. the pulmonary nodule detection method according to claim 1 with semantic feature score, which is characterized in that the region
It is recommended that being provided with 3 anchor boxes that area ratio is 1:4 in network.
3. the pulmonary nodule detection method according to claim 1 with semantic feature score, which is characterized in that the classification
Network for Lung neoplasm it is good it is pernicious classify, the second semantic feature network obtains the score of one group of semantic feature.
4. the pulmonary nodule detection method according to claim 1 with semantic feature score, which is characterized in that the lung knot
The position of section is that the position coordinates of Lung neoplasm in CT image, the Lung neoplasm are classified as the good pernicious classification of the Lung neoplasm,
The semantic feature of the Lung neoplasm is scored at one group of semantic feature score for the Lung neoplasm.
5. a kind of Lung neoplasm detection system with semantic feature score, which is characterized in that the system comprises:
The characteristic pattern of CT image obtains module, and the characteristic pattern of the CT image obtains module and is used for construction feature extractor network,
Using the chest CT image with Lung neoplasm as the input of the network, and extract the characteristic pattern of CT image；
Area-of-interest obtains module, and the area-of-interest obtains module and suggests network for constructing region, characteristic pattern is made
Suggest the input of network for region, and obtains area-of-interest；
First semantic feature score obtains module, and the first semantic feature score obtains module for constructing the first semantic feature
Network using characteristic pattern as the input of the first semantic feature network, and obtains the semantic feature score of area-of-interest；
Lung neoplasm position and classification obtain module, and the Lung neoplasm position and classification obtain module for constructing sorter network, and
Using area-of-interest as the input of sorter network, the position of area-of-interest is finely adjusted, obtains Lung neoplasm position and institute
State the classification of Lung neoplasm；
Second semantic feature score obtains module, and the second semantic feature score obtains module for constructing the second semantic feature
Network, and using area-of-interest as the input of the second semantic feature network, the semantic feature score of area-of-interest is carried out
Fine tuning, obtains the semantic feature score of Lung neoplasm.
6. the Lung neoplasm detection system according to claim 5 with semantic feature score, which is characterized in that the region
It is recommended that being provided with 3 anchor boxes that area ratio is 1:4 in network.
7. the Lung neoplasm detection system according to claim 5 with semantic feature score, which is characterized in that the classification
Network for Lung neoplasm it is good it is pernicious classify, the semantic feature network obtains the score of one group of semantic feature.
8. the Lung neoplasm detection system according to claim 5 with semantic feature score, which is characterized in that the lung knot
The position of section is that the position coordinates of Lung neoplasm in CT image, the Lung neoplasm are classified as the good pernicious classification of the Lung neoplasm,
The semantic feature of the Lung neoplasm is scored at one group of semantic feature score for the Lung neoplasm.
9. a kind of computer storage medium, an at least executable instruction, the executable instruction are stored in the storage medium
Processor is set to execute the corresponding operation of pulmonary nodule detection method as described in any one of Claims 1-4.
10. a kind of computer installation, comprising: processor, memory, communication interface and communication bus, the processor, memory
Mutual communication is completed by the communication bus with communication interface, the memory is for storing at least one executable finger
It enables, the executable instruction makes the processor execute the pulmonary nodule detection method as described in any one of Claims 1-4
Corresponding operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910465182.2A CN110189318B (en) | 2019-05-30 | 2019-05-30 | Pulmonary nodule detection method and system with semantic feature score |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910465182.2A CN110189318B (en) | 2019-05-30 | 2019-05-30 | Pulmonary nodule detection method and system with semantic feature score |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110189318A true CN110189318A (en) | 2019-08-30 |
| CN110189318B CN110189318B (en) | 2022-12-06 |
Family
ID=67719124
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910465182.2A Expired - Fee Related CN110189318B (en) | 2019-05-30 | 2019-05-30 | Pulmonary nodule detection method and system with semantic feature score |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110189318B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110738643A (en) * | 2019-10-08 | 2020-01-31 | 上海联影智能医疗科技有限公司 | Method for analyzing cerebral hemorrhage, computer device and storage medium |
| CN111598853A (en) * | 2020-04-30 | 2020-08-28 | 科大讯飞股份有限公司 | Pneumonia-oriented CT image scoring method, device and equipment |
| CN113139928A (en) * | 2020-01-16 | 2021-07-20 | 中移(上海)信息通信科技有限公司 | Training method of pulmonary nodule detection model and pulmonary nodule detection method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2003024184A2 (en) * | 2001-09-14 | 2003-03-27 | Cornell Research Foundation, Inc. | System, method and apparatus for small pulmonary nodule computer aided diagnosis from computed tomography scans |
| CN107492095A (en) * | 2017-08-02 | 2017-12-19 | 西安电子科技大学 | Medical image pulmonary nodule detection method based on deep learning |
| CN109087317A (en) * | 2018-11-13 | 2018-12-25 | 中国科学院大学 | A kind of Lung neoplasm image partition method |
-
2019
- 2019-05-30 CN CN201910465182.2A patent/CN110189318B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2003024184A2 (en) * | 2001-09-14 | 2003-03-27 | Cornell Research Foundation, Inc. | System, method and apparatus for small pulmonary nodule computer aided diagnosis from computed tomography scans |
| CN107492095A (en) * | 2017-08-02 | 2017-12-19 | 西安电子科技大学 | Medical image pulmonary nodule detection method based on deep learning |
| CN109087317A (en) * | 2018-11-13 | 2018-12-25 | 中国科学院大学 | A kind of Lung neoplasm image partition method |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110738643A (en) * | 2019-10-08 | 2020-01-31 | 上海联影智能医疗科技有限公司 | Method for analyzing cerebral hemorrhage, computer device and storage medium |
| CN110738643B (en) * | 2019-10-08 | 2023-07-28 | 上海联影智能医疗科技有限公司 | Analysis method for cerebral hemorrhage, computer device and storage medium |
| CN113139928A (en) * | 2020-01-16 | 2021-07-20 | 中移(上海)信息通信科技有限公司 | Training method of pulmonary nodule detection model and pulmonary nodule detection method |
| CN113139928B (en) * | 2020-01-16 | 2024-02-23 | 中移(上海)信息通信科技有限公司 | Training method of lung nodule detection model and lung nodule detection method |
| CN111598853A (en) * | 2020-04-30 | 2020-08-28 | 科大讯飞股份有限公司 | Pneumonia-oriented CT image scoring method, device and equipment |
| CN111598853B (en) * | 2020-04-30 | 2024-02-13 | 讯飞医疗科技股份有限公司 | CT image scoring method, device and equipment for pneumonia |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110189318B (en) | 2022-12-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ramakrishnan et al. | Habitat-matterport 3d dataset (hm3d): 1000 large-scale 3d environments for embodied ai | |
| CN111754596B (en) | Editing model generation method, device, equipment and medium for editing face image | |
| CN104572804B (en) | A kind of method and its system of video object retrieval | |
| Sheng et al. | Improving video temporal consistency via broad learning system | |
| CN104424634B (en) | Object tracking method and device | |
| US7995841B2 (en) | Hybrid graph model for unsupervised object segmentation | |
| CN110189318A (en) | Pulmonary nodule detection method and system with semantic feature score | |
| Cao et al. | Ancient mural restoration based on a modified generative adversarial network | |
| AU2019227478A1 (en) | An improved content aware fill that leverages the boundaries of underlying objects | |
| CN106204118A (en) | A kind of method and apparatus found for commercial circle | |
| CN111008654A (en) | Method and system for identifying rooms in house type graph | |
| CN109711416A (en) | Target identification method, device, computer equipment and storage medium | |
| WO2023103600A1 (en) | Expression generation method and apparatus, device, medium and computer program product | |
| US11182905B2 (en) | Algorithmic approach to finding correspondence between graphical elements | |
| Zhou et al. | Deep learning-based instance segmentation for indoor fire load recognition | |
| Yu et al. | Robust point cloud normal estimation via neighborhood reconstruction | |
| Rutter et al. | Automated object tracing for biomedical image segmentation using a deep convolutional neural network | |
| CN111177811A (en) | Automatic fire point location layout method applied to cloud platform | |
| CN106934398A (en) | Image de-noising method based on super-pixel cluster and rarefaction representation | |
| Zhou et al. | Computerized image analysis: texture‐field orientation method for pectoral muscle identification on MLO‐view mammograms | |
| CN113284237A (en) | Three-dimensional reconstruction method, system, electronic equipment and storage medium | |
| CN103837135A (en) | Workpiece detecting method and system | |
| CN115953330B (en) | Texture optimization method, device, equipment and storage medium for virtual scene image | |
| Pang et al. | Salient object detection via effective background prior and novel graph | |
| CN110176021A (en) | In conjunction with the level set image segmentation method and system of the conspicuousness information of gamma correction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20221206 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |