CN107507197A - A kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks - Google Patents
A kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks Download PDFInfo
- Publication number
- CN107507197A CN107507197A CN201710712015.4A CN201710712015A CN107507197A CN 107507197 A CN107507197 A CN 107507197A CN 201710712015 A CN201710712015 A CN 201710712015A CN 107507197 A CN107507197 A CN 107507197A
- Authority
- CN
- China
- Prior art keywords
- image
- neural networks
- convolutional neural
- pulmonary parenchyma
- lung
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- 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
-
- 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
- G06V2201/031—Recognition of patterns in medical or anatomical images of internal organs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The present invention proposes a kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks, the present invention is pre-processed using clustering algorithm to lung CT image, obtain the pulmonary parenchyma region of CT images and the data set in non-pulmonary parenchyma region, the data set of known lung CT image is divided into training set and checking collects, using the data set of unknown lung CT image as test set;Convolutional neural networks model is established, convolutional neural networks model is trained using training set and checking collection data, the convolutional neural networks model after being trained;By in the convolutional neural networks model after test set input training, obtain CT image pulmonary parenchymas region, realize and pulmonary parenchyma extracted region function is carried out to the lung CT image of unknown patient, and lower basis is built to next stage automatic searching lung cancer region, contribute to the lung cancer of next stage to extract, classify.
Description
Technical field
The invention belongs to technical field of medical image processing, and in particular to one kind is based on clustering algorithm and convolutional neural networks
Pulmonary parenchyma extracting method.
Background technology
In histology, lung tissue is divided into pulmonary parenchyma and interstitial lung two parts, and pulmonary parenchyma is the bronchial each fraction of intrapulmonary
Branch and its eventually a large amount of alveolar structures at end, interstitial lung is connective tissue and blood vessel, lymphatic vessel, nerve etc..In order to assess and study
Pulmonary volume, lung cancer, often doctor need the first step to know pulmonary parenchyma situation, the accurate segmentation of pulmonary parenchyma is as further study of lung
The tissue and lesion of interior each organ dysfunction, are played a very important role to doctor.
During the research and application of image, people are often only interested in some parts in image.These portions
Divide and be frequently referred to target, they generally correspond to specific region with unique properties in image.In order to distinguish and analyze target, need
These region disconnectings are extracted, further utilized on this basis.Image segmentation is exactly to divide the image into their own characteristics
Region and extract the technology and process of interesting target.Classical image partition method such as threshold method, region increase, edge
Detection, cluster and nerual network technique etc..
Threshold segmentation is most ancient cutting techniques, and most simple and practical.In many cases, objective area in image
Its gray value has differences between background area in other words different zones, can now enter the homogeneity of gray scale as foundation
Row segmentation, it is generally affected by noise larger.
Edge detection algorithm is relatively adapted to the segmentation of edge gray value transition simple image more less than more significant and noise.
It is more complicated and image compared with very noisy be present for edge, then face the contradiction of noise immunity and accuracy of detection.If improve inspection
Precision is surveyed, then pseudo-edge caused by noise can cause irrational profile:If improving noise immunity, profile missing inspection and position can be produced
Put deviation.
Cluster analysis is an important branch of non-supervisory pattern-recognition in pattern-recognition.According to the internal junction of data acquisition system
Structure is divided into different classifications so that the feature of sample sample point that is as similar as possible, and belonging to a different category in same class
Difference it is as big as possible.
The basic thought of dividing method based on neutral net is to obtain linear decision letter by training multi-layer perception (MLP)
Number, is then classified to reach the purpose of segmentation with decision function to pixel.This method needs substantial amounts of training data.God
The connection of flood tide through network be present, be readily incorporated spatial information, can preferably solve noise and problem of non-uniform in image.Choosing
It is this method subject matter to be solved to select which kind of network structure.
In summary, the dividing method having pointed out at present is mostly directed to particular problem, and has his own strong points, not a kind of logical
Use standard.Split for pulmonary parenchyma, the present invention proposes a kind of convolutional neural networks using cluster generation data training, non-threshold
The automatic segmentation pulmonary parenchyma region of value, it is efficiently and accurate.
The content of the invention
In view of the shortcomings of the prior art, the present invention proposes that a kind of pulmonary parenchyma based on clustering algorithm and convolutional neural networks carries
Take method.
A kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks, comprises the following steps:
Step 1:Lung CT image is pre-processed using clustering algorithm, obtains the pulmonary parenchyma region of CT images and non-lung
The data set of parenchyma section, the data set of known lung CT image is divided into training set and checking collects, by unknown lung
The data set of CT images is as test set;
Step 1.1:Lung CT image is standardized, the lung CT image after standardization is split, point
It is segmented into the image fritter that size is A;
Step 1.2:Use CT value average value and CT value minimum value of the Kmeans algorithms respectively to size for A image fritter
Clustered, it is low-density tissue and the class of high density tissue two to be clustered;
Step 1.3:By to the poly- of the CT value minimum values of the cluster result of the CT value average values of image fritter and image fritter
Class result carries out cross inspection, removes the background area of CT images;
Step 1.4:Extract the pulmonary parenchyma region in the cluster result of the CT value average values of image fritter and image fritter
The common factor in the pulmonary parenchyma region of the cluster result of CT value minimum values;
Step 1.5:The common factor in the pulmonary parenchyma region to being obtained in step 1.4 does largest connected gymnastics and made, and obtains CT images
Pulmonary parenchyma region and non-pulmonary parenchyma region data set, the data set of known lung CT image is divided into and training set and tested
Card collection, using the data set of lung CT image to be divided as test set.
Step 2:Convolutional neural networks model is established, convolutional neural networks model is entered using training set and checking collection data
Row training, the convolutional neural networks model after being trained;
Step 2.1:The image fritter that the size that training set and checking are concentrated is A is extended, is expanded to size
For B image fritter;
Step 2.2:Convolutional neural networks model is established, the image fritter after extension is inputted into convolutional neural networks model,
The weight and deviation of each layer of training convolutional neural networks model;
Step 2.3:Checking is collected into input convolutional neural networks model to be classified, is lost by run time and checking collects
Classification accuracy determine optimize training parameter, the convolutional neural networks model after being trained.
Step 3:By in the convolutional neural networks model after test set input training, CT image pulmonary parenchymas region is obtained.
The convolutional neural networks model structure is:First layer is image fritter input layer, and the second layer is convolutional layer, the 3rd
Layer is maximum pond layer, and the 4th layer is full articulamentum;
The convolutional layer includes convolutional layer ReLU layers and Norm layers;
The full articulamentum includes full articulamentum ReLU layers, forgets layer, full articulamentum grader and Softmax functions at random
Layer.
The criteria for classifying of the size A is:The size divided is to include the lung in CT images in A image fritter
Portion organizes, and the image fritter automatic sliced time of each CT images is within 50MS.
The figure that the size of the image fritter that the size of pulmonary parenchyma is A and non-pulmonary parenchyma is A is concentrated in the training set and checking
As the quantity of fritter respectively accounts for 50%.
It is described by the cluster knot of the cluster result of the CT value average values of image fritter and the CT value minimum values of image fritter
Fruit carries out cross inspection, and the detailed process for removing the background area of CT images is as follows:
Examine whether four radial directions of the image fritter of each low-density tissue have the image fritter of high density tissue,
If in the presence of the image fritter of the low-density tissue is doubtful pulmonary parenchyma region, and otherwise the image fritter of the low-density tissue is
Background area.
The image fritter that the size that training set and checking are concentrated is A is extended, and it is B's to be expanded to size
The detailed process of image fritter is as follows:
By size to expand to image fritter of the size for B on the position centered on A image fritter in former CT images.
Described training parameter includes:Learning rate, convolution kernel size, convolution kernel number, Norm layers standardization port number
Mesh, full connection first layer output number, Dropout layers forgetting rate, pond channel type, Batch numbers, Epochs numerical value.
Beneficial effects of the present invention:
The present invention proposes a kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks, and the present invention passes through nothing
Supervised learning algorithm, i.e. clustering algorithm, to the automatic division pulmonary parenchyma region of CT picture of patient and non-pulmonary parenchyma region, reach automatic
Generate the training set needed for training convolutional neural networks, checking collects function;Pass through supervised learning algorithm, i.e. convolutional Neural net
Network, for pulmonary parenchyma territorial classification, optimal convolutional neural networks model is designed, reach the high pulmonary parenchyma territorial classification of accuracy rate
Effect;Realize and pulmonary parenchyma extracted region function is carried out to the lung CT image of unknown patient, and to next stage automatic searching
Lower basis is built in lung cancer region, contributes to the lung cancer of next stage to extract, classify.
Brief description of the drawings
Fig. 1 is the pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks in the specific embodiment of the invention
Flow chart;
Fig. 2 is the process schematic split in the specific embodiment of the invention to the lung CT image after standardization;
Wherein, (a) is the lung CT image that cut size is 64*64;
(b) it is lung CT image that cut size is 32*32;
(c) it is lung CT image that cut size is 16*16;
(d) it is lung CT image that cut size is 8*8;
(e) it is lung CT image that cut size is 4*4;
(f) it is lung CT image that cut size is 2*2;
Fig. 3 is that the process in the pulmonary parenchyma region and non-pulmonary parenchyma region that obtain CT images in the specific embodiment of the invention is shown
It is intended to;
Wherein, (a) is to use what Kmeans algorithms were clustered to size for the CT value average values of A image fritter respectively
As a result;
(b) it is the low-density class cross assay of CT value average values;
(c) it is to use the result that Kmeans algorithms are clustered to size for the CT value minimum values of A image fritter respectively;
(d) it is the low-density class cross assay of CT value minimum values;
(e) it is the pulmonary parenchyma region in the cluster result of the CT value average values of extraction image fritter and the CT values of image fritter
The schematic diagram of the common factor in the pulmonary parenchyma region of the cluster result of minimum value;
(f) schematic diagram in the pulmonary parenchyma region for CT images and non-pulmonary parenchyma region;
Fig. 4 be in the specific embodiment of the invention by the image fritter that size is A be extended for size be B image it is small
The schematic diagram of block;
Wherein, (a) is the image fritter that size is A;
(b) it is image fritter that size is B;
Fig. 5 is convolutional neural networks model schematic in the specific embodiment of the invention;
Wherein, (a) is image fritter input layer;
(b) it is convolutional layer;
(c) it is maximum pond layer;
(d) it is full articulamentum;
Fig. 6 is to the schematic diagram for the pulmonary parenchyma region progress three-dimensional modeling being partitioned into the specific embodiment of the invention;
Wherein, (a) is the classification three-dimensional reconstruction result of patients with chronic obstructive pulmonary diseases;
(b) it is the classification three-dimensional reconstruction result of the patients with lung cancer under common CT scan image datas;
(c) the classification three-dimensional reconstruction result of the patients with lung cancer under scanning pattern is imaged for the general tumour of PET/CT equipment.
Embodiment
The specific embodiment of the invention is described in detail below in conjunction with the accompanying drawings.
A kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks, as shown in figure 1, including following step
Suddenly:
Step 1:Lung CT image is pre-processed using clustering algorithm, obtains the pulmonary parenchyma region of CT images and non-lung
The data set of parenchyma section, the data set of known lung CT image is divided into training set and checking collects, by unknown lung
The data set of CT images is as test set.
Step 1.1:Lung CT image is standardized, the lung CT image after standardization is split, point
It is segmented into the image fritter that size is A.
In present embodiment, the size A criteria for classifying is:The size divided is to include CT in A image fritter
Lung tissue in image, and the image fritter automatic sliced time of each CT images is within 50MS.
The process schematic that lung CT image after standardization is split is as shown in Fig. 2 to the lung CT after standardization
The elapsed time and segmentation feature that image is split are as shown in table 1.
The elapsed time and segmentation feature that table 1 is split to the lung CT image after standardization
In present embodiment, the assessment with reference to the characteristics of the elapsed time in table 1 and segmentation, it can be seen that with fritter
Size reduction, time loss exponentially increase, and the difference of feature is increasing between each fritter, and therefore, the present invention is last
Select optimal sizes of the 8*8 as image fritter A.
Step 1.2:Use CT value average value and CT value minimum value of the Kmeans algorithms respectively to size for A image fritter
Clustered, it is low-density tissue and the class of high density tissue two to be clustered.
In present embodiment, Kmeans algorithms are used to gather respectively to size for the CT value average values of A image fritter
Shown in the result of class such as Fig. 3 (a), Kmeans algorithms are used to gather respectively to size for the CT value minimum values of A image fritter
Shown in the result of class such as Fig. 3 (c).
Step 1.3:By to the poly- of the CT value minimum values of the cluster result of the CT value average values of image fritter and image fritter
Class result carries out cross inspection, removes the background area of CT images.
In present embodiment, it is described by the CT values of the cluster result of the CT value average values of image fritter and image fritter most
The cluster result of small value carries out cross inspection, and the detailed process for removing the background area of CT images is as follows:
Examine whether four radial directions of the image fritter of each low-density tissue have the image fritter of high density tissue,
If in the presence of the image fritter of the low-density tissue is doubtful pulmonary parenchyma region, and otherwise the image fritter of the low-density tissue is
Background area.
The low-density class cross assay of CT value average values, as shown in Fig. 3 (b), the low-density class ten of CT value minimum values
Word assay, as shown in Fig. 3 (d).
Step 1.4:Extract the pulmonary parenchyma region in the cluster result of the CT value average values of image fritter and image fritter
The common factor in the pulmonary parenchyma region of the cluster result of CT value minimum values, as shown in Fig. 3 (e).
Step 1.5:The common factor in the pulmonary parenchyma region to being obtained in step 1.4 does largest connected gymnastics and made, and obtains CT images
Pulmonary parenchyma region and non-pulmonary parenchyma region data set, as shown in Fig. 3 (f), by the data set of known lung CT image draw
It is divided into training set and checking collects, using the data set of lung CT image to be divided as test set.
In present embodiment, the image fritter and non-pulmonary parenchyma that the size of pulmonary parenchyma is A are concentrated in the training set and checking
Size be A the quantity of image fritter respectively account for 50%.The data set of known lung CT image is divided into training set and tested
The ratio setting of card collection is 7: 1.
Found from Fig. 3, use Kmeans algorithms to be clustered respectively to size for the CT value minimum values of A image fritter
Result pulmonary parenchyma region be more than use Kmeans algorithms to gather respectively to size for the CT value average values of A image fritter
The pulmonary parenchyma region of the result of class, i.e. CT values minimum value clustering method beyond real pulmonary parenchyma edge, and remove in human body
In terms of following ambient noise, CT value minimum value clustering method effects are better than CT value average value clustering methods.Therefore the present invention takes
The intersection area of the two, you can, again can be by CT values most to retain the accurate pulmonary parenchyma region of CT value average value clustering methods
Small value clustering method control ambient noise.After above step, still there is the ambient noise below few body not go
Remove, so as to be operated using connected component, be completely removed background, obtain more accurate pulmonary parenchyma region.
Step 2:Convolutional neural networks model is established, convolutional neural networks model is entered using training set and checking collection data
Row training, the convolutional neural networks model after being trained.
Step 2.1:The image fritter that the size that training set and checking are concentrated is A is extended, is expanded to size
For B image fritter.
In present embodiment, as shown in figure 4, by size on the position centered on A image fritter in former CT images
The image fritter that size is B is expanded to, can avoid rolling up undersized fritter because convolution kernel size is big in convolution process
Product DeGrain.
In present embodiment, size B is 32*32.
Step 2.2:Convolutional neural networks model is established, the image fritter after extension is inputted into convolutional neural networks model,
The weight and deviation of each layer of training convolutional neural networks model.
In present embodiment, convolutional neural networks model structure use simplification after AlexNet structures as shown in figure 5,
Including:First layer is image fritter input layer, and the second layer is convolutional layer, and third layer is maximum pond layer, and the 4th layer is to connect entirely
Connect layer.
The convolutional layer includes convolutional layer ReLU layers and Norm layers;
The full articulamentum includes full articulamentum ReLU layers, forgets layer, full articulamentum grader and Softmax functions at random
Layer.
In present embodiment, class categories quantity is considered to greatest extent and only includes 2 classes, rather than the one of AlexNet
Thousand kinds of classifications, therefore the convolutional neural networks model only remains a convolution nuclear volume in AlexNet 5 convolutional layers and is
6 convolutional layer, and ReLU active coatings and Norm normalization layers are added, stochastic gradient descent can be accelerated and prevent over-fitting;Pond
Change layer and have selected maximum pond rather than average value pond, network complexity can be reduced;It is special that full articulamentum includes 120 distributions
Sign, and then adds ReLU active coatings and Dropout forgets layer, stochastic gradient can be accelerated to restrain, and prevent over-fitting;Quan Lian
The second layer for connecing layer is a 2 classification grader, act as classify pulmonary parenchyma region and non-pulmonary parenchyma region;Finally use
For Softmax functions as output layer, its probability distribution exported is approximate to represent output distribution.
Step 2.3:Checking is collected into input convolutional neural networks model to be classified, is lost by run time and checking collects
Classification accuracy determine optimize training parameter, the convolutional neural networks model after being trained.
In present embodiment, training parameter includes:Learning rate, convolution kernel size, convolution kernel number, the standardization of Norm layers
Number of active lanes, full connection first layer output number, Dropout layers forgetting rate, pond channel type, Batch numbers, Epochs numbers
Value.
In present embodiment, convolutional neural networks model optimization parameter and experimental result are as shown in table 2:
The convolutional neural networks model optimization parameter of table 2 and experimental result
It can be seen from table 2, the first row is as reference standard, it is seen that the size of convolution kernel increases to 10 from 5 × 5 in convolutional layer
× 10, checking accuracy rate is not lifted, and simply corresponding time loss improves 27%;When Epochs value is arranged to 50 to 80
When, checking accuracy rate is not lifted and time loss sharply rises to three times;When learning rate is arranged to 0.0001, time loss
To be double, but its rate of accuracy reached can receive to sacrifice time cost within the specific limits to lift accuracy rate to 99.17%;
Make our accuracys close to 100% when learning rate is set to 0.00001, but the 85% of time cost.To sum up institute
State, present invention determine that optimized parameter as shown in last column in table 2.
The training parameter of optimization is:Learning rate is 0.0001, convolution kernel size is 5*5, convolution kernel number is 6,
Norm layers standardization number of active lanes is 3, full connection first layer output number is 120, Dropout layer forgetting rates be 0.5 (50%),
Pond layer Max types, Batch numbers are 128, Epochs numerical value is 50.
Step 3:By in the convolutional neural networks model after test set input training, CT image pulmonary parenchymas region is obtained.
In present embodiment, in the convolutional neural networks model after test set input is trained, and the lung to being partitioned into is real
Matter region carries out three-dimensional modeling, tests the accuracy rate and versatility of the convolutional neural networks model.
In convolutional neural networks model after test set input is trained, and the pulmonary parenchyma region to being partitioned into carries out three-dimensional
Modeling, the goldstandard for carrying out pulmonary parenchyma region to lung CT image to be divided according to doctor delimited, then existed according to test set
Classification results in neutral net, the progress of the image of goldstandard and classification results is overlapping, calculate accuracy rate and (wear this similitude system
Number), sensitivity, specificity etc., as shown in fig. 6, wherein, classification three-dimensional reconstruction result such as Fig. 6 of patients with chronic obstructive pulmonary diseases
(a) shown in, shown in classification three-dimensional reconstruction result such as Fig. 6 (b) of the patients with lung cancer under common CT scan image datas, PET/CT is set
Shown in classification three-dimensional reconstruction result such as Fig. 6 (c) of patients with lung cancer under standby general tumour imaging scanning pattern.
Claims (9)
1. a kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks, it is characterised in that comprise the following steps:
Step 1:Lung CT image is pre-processed using clustering algorithm, obtains the pulmonary parenchyma region of CT images and non-pulmonary parenchyma
The data set in region, the data set of known lung CT image is divided into training set and checking collects, by unknown lung CT figure
The data set of picture is as test set;
Step 2:Convolutional neural networks model is established, convolutional neural networks model is instructed using training set and checking collection data
Practice, the convolutional neural networks model after being trained;
Step 3:By in the convolutional neural networks model after test set input training, CT image pulmonary parenchymas region is obtained.
2. the pulmonary parenchyma extracting method according to claim 1 based on clustering algorithm and convolutional neural networks, its feature exist
In the step 1 comprises the following steps:
Step 1.1:Lung CT image is standardized, the lung CT image after standardization is split, is divided into
Size is A image fritter;
Step 1.2:Kmeans algorithms are used to be carried out respectively to size for the CT values average value and CT values minimum value of A image fritter
Cluster, it is low-density tissue and the class of high density tissue two to be clustered;
Step 1.3:By to the cluster knot of the cluster result of the CT value average values of image fritter and the CT value minimum values of image fritter
Fruit carries out cross inspection, removes the background area of CT images;
Step 1.4:Extract the pulmonary parenchyma region in the cluster result of the CT value average values of image fritter and the CT values of image fritter
The common factor in the pulmonary parenchyma region of the cluster result of minimum value;
Step 1.5:The common factor in the pulmonary parenchyma region to being obtained in step 1.4 does largest connected gymnastics and made, and obtains the lung of CT images
Parenchyma section and the data set in non-pulmonary parenchyma region, training set and checking are divided into by the data set of known lung CT image
Collection, using the data set of lung CT image to be divided as test set.
3. the pulmonary parenchyma extracting method according to claim 1 based on clustering algorithm and convolutional neural networks, its feature exist
In the step 2 comprises the following steps:
Step 2.1:The image fritter that the size that training set and checking are concentrated is A is extended, it is B's to be expanded to size
Image fritter;
Step 2.2:Convolutional neural networks model is established, the image fritter after extension is inputted into convolutional neural networks model, training
The weight and deviation of each layer of convolutional neural networks model;
Step 2.3:Checking is collected into input convolutional neural networks model to be classified, point of collection is lost and verified by run time
Class accuracy rate determines the training parameter optimized, the convolutional neural networks model after being trained.
4. the pulmonary parenchyma extracting method according to claim 1 based on clustering algorithm and convolutional neural networks, its feature exist
In the convolutional neural networks model structure is:First layer is image fritter input layer, and the second layer is convolutional layer, and third layer is
Maximum pond layer, the 4th layer is full articulamentum;
The convolutional layer includes convolutional layer ReLU layers and Norm layers;
The full articulamentum includes full articulamentum ReLU layers, forgets layer, full articulamentum grader and Softmax function layers at random.
5. the pulmonary parenchyma extracting method according to claim 2 based on clustering algorithm and convolutional neural networks, its feature exist
In the criteria for classifying of the size A is:The size divided is to include lung's group in CT images in A image fritter
Knit, and the image fritter automatic sliced time of each CT images is within 50MS.
6. the pulmonary parenchyma extracting method according to claim 2 based on clustering algorithm and convolutional neural networks, its feature exist
The image fritter that the size of the image fritter that the size of pulmonary parenchyma is A and non-pulmonary parenchyma is A is concentrated in, the training set and checking
Quantity respectively account for 50%.
7. the pulmonary parenchyma extracting method according to claim 2 based on clustering algorithm and convolutional neural networks, its feature exist
In described by the cluster result of the CT value average values of image fritter and the progress of the cluster result of the CT value minimum values of image fritter
Cross is examined, and the detailed process for removing the background area of CT images is as follows:
Examine whether four radial directions of the image fritter of each low-density tissue have the image fritter of high density tissue, if depositing
Then the image fritter of the low-density tissue is doubtful pulmonary parenchyma region, and otherwise the image fritter of the low-density tissue is background
Region.
8. the pulmonary parenchyma extracting method according to claim 3 based on clustering algorithm and convolutional neural networks, its feature exist
In the image fritter that the size for concentrating training set and checking is A is extended, and is expanded to the image that size is B
The detailed process of fritter is as follows:
By size to expand to image fritter of the size for B on the position centered on A image fritter in former CT images.
9. the pulmonary parenchyma extracting method according to claim 3 based on clustering algorithm and convolutional neural networks, its feature exist
In described training parameter includes:It is learning rate, convolution kernel size, convolution kernel number, Norm layers standardization number of active lanes, complete
Connect first layer output number, Dropout layers forgetting rate, pond channel type, Batch numbers, Epochs numerical value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710712015.4A CN107507197B (en) | 2017-08-18 | 2017-08-18 | Lung parenchyma extraction method based on clustering algorithm and convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710712015.4A CN107507197B (en) | 2017-08-18 | 2017-08-18 | Lung parenchyma extraction method based on clustering algorithm and convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107507197A true CN107507197A (en) | 2017-12-22 |
CN107507197B CN107507197B (en) | 2020-11-24 |
Family
ID=60691358
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710712015.4A Active CN107507197B (en) | 2017-08-18 | 2017-08-18 | Lung parenchyma extraction method based on clustering algorithm and convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107507197B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108009594A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of image-recognizing method based on change packet convolution |
CN109065165A (en) * | 2018-07-25 | 2018-12-21 | 东北大学 | A kind of chronic obstructive pulmonary disease prediction technique based on reconstruction air flue tree Image |
CN109087296A (en) * | 2018-08-07 | 2018-12-25 | 东北大学 | A method of extracting human region in CT image |
CN109300122A (en) * | 2018-09-14 | 2019-02-01 | 沈阳东软医疗***有限公司 | Image procossing and Threshold, device and equipment |
CN109508650A (en) * | 2018-10-23 | 2019-03-22 | 浙江农林大学 | A kind of wood recognition method based on transfer learning |
CN109636802A (en) * | 2019-01-18 | 2019-04-16 | 天津工业大学 | Pulmonary parenchyma based on depth convolutional neural networks is through CT image partition method |
CN110796652A (en) * | 2019-10-30 | 2020-02-14 | 上海联影智能医疗科技有限公司 | Image processing method, computer device, and storage medium |
CN111882560A (en) * | 2020-06-16 | 2020-11-03 | 北京工业大学 | Lung parenchymal CT image segmentation method based on weighted full-convolution neural network |
CN113409306A (en) * | 2021-07-15 | 2021-09-17 | 推想医疗科技股份有限公司 | Detection device, training method, training device, equipment and medium |
CN113706541A (en) * | 2020-05-20 | 2021-11-26 | 青岛海信医疗设备股份有限公司 | Image processing method and device |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106372390A (en) * | 2016-08-25 | 2017-02-01 | 姹ゅ钩 | Deep convolutional neural network-based lung cancer preventing self-service health cloud service system |
CN106651830A (en) * | 2016-09-28 | 2017-05-10 | 华南理工大学 | Image quality test method based on parallel convolutional neural network |
CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
-
2017
- 2017-08-18 CN CN201710712015.4A patent/CN107507197B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106372390A (en) * | 2016-08-25 | 2017-02-01 | 姹ゅ钩 | Deep convolutional neural network-based lung cancer preventing self-service health cloud service system |
CN106651830A (en) * | 2016-09-28 | 2017-05-10 | 华南理工大学 | Image quality test method based on parallel convolutional neural network |
CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
Non-Patent Citations (1)
Title |
---|
娜孜古丽•库尔曼别克等: "基于CT图像肺实质分割的方法", 《科技创新与应用》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108009594A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of image-recognizing method based on change packet convolution |
CN108009594B (en) * | 2017-12-25 | 2018-11-13 | 北京航空航天大学 | A kind of image-recognizing method based on change grouping convolution |
CN109065165A (en) * | 2018-07-25 | 2018-12-21 | 东北大学 | A kind of chronic obstructive pulmonary disease prediction technique based on reconstruction air flue tree Image |
CN109065165B (en) * | 2018-07-25 | 2021-08-17 | 东北大学 | Chronic obstructive pulmonary disease prediction method based on reconstructed airway tree image |
CN109087296A (en) * | 2018-08-07 | 2018-12-25 | 东北大学 | A method of extracting human region in CT image |
CN109087296B (en) * | 2018-08-07 | 2021-08-10 | 东北大学 | Method for extracting human body region in CT image |
CN109300122A (en) * | 2018-09-14 | 2019-02-01 | 沈阳东软医疗***有限公司 | Image procossing and Threshold, device and equipment |
US11478163B2 (en) | 2018-09-14 | 2022-10-25 | Neusoft Medical Systems Co., Ltd. | Image processing and emphysema threshold determination |
CN109508650A (en) * | 2018-10-23 | 2019-03-22 | 浙江农林大学 | A kind of wood recognition method based on transfer learning |
CN109636802A (en) * | 2019-01-18 | 2019-04-16 | 天津工业大学 | Pulmonary parenchyma based on depth convolutional neural networks is through CT image partition method |
CN110796652A (en) * | 2019-10-30 | 2020-02-14 | 上海联影智能医疗科技有限公司 | Image processing method, computer device, and storage medium |
CN113706541A (en) * | 2020-05-20 | 2021-11-26 | 青岛海信医疗设备股份有限公司 | Image processing method and device |
CN113706541B (en) * | 2020-05-20 | 2024-04-19 | 青岛海信医疗设备股份有限公司 | Image processing method and device |
CN111882560A (en) * | 2020-06-16 | 2020-11-03 | 北京工业大学 | Lung parenchymal CT image segmentation method based on weighted full-convolution neural network |
CN111882560B (en) * | 2020-06-16 | 2024-05-31 | 北京工业大学 | Lung parenchyma CT image segmentation method based on weighted full convolution neural network |
CN113409306A (en) * | 2021-07-15 | 2021-09-17 | 推想医疗科技股份有限公司 | Detection device, training method, training device, equipment and medium |
Also Published As
Publication number | Publication date |
---|---|
CN107507197B (en) | 2020-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107507197A (en) | A kind of pulmonary parenchyma extracting method based on clustering algorithm and convolutional neural networks | |
CN104751178B (en) | Lung neoplasm detection means and method based on shape template matching combining classification device | |
CN106815481B (en) | Lifetime prediction method and device based on image omics | |
CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
CN107103187A (en) | The method and system of Lung neoplasm detection classification and management based on deep learning | |
Jafarzadeh Ghoushchi et al. | An extended approach to the diagnosis of tumour location in breast cancer using deep learning | |
CN101667289B (en) | Retinal image segmentation method based on NSCT feature extraction and supervised classification | |
CN107368671A (en) | System and method are supported in benign gastritis pathological diagnosis based on big data deep learning | |
CN110135271A (en) | A kind of cell sorting method and device | |
CN104102839B (en) | A kind of Alzheimer disease cortex automatic classification method based on multiple dimensioned grid surface shape facility | |
CN112950614B (en) | Breast cancer pathological image classification method based on multi-scale cavity convolution | |
CN107133638A (en) | Multi-parameter MRI prostate cancer CAD method and system based on two graders | |
CN109461163A (en) | A kind of edge detection extraction algorithm for magnetic resonance standard water mould | |
CN109087296A (en) | A method of extracting human region in CT image | |
Mohammed et al. | An efficient approach to diagnose brain tumors through deep CNN | |
CN110188592B (en) | Urine formed component cell image classification model construction method and classification method | |
Syed et al. | Detection of tumor in MRI images using artificial neural networks | |
Bhat et al. | Convolutional neural network approach for the classification and recognition of lung nodules | |
Lin et al. | Hyperparameter Optimization of Deep Learning Networks for Classification of Breast Histopathology Images. | |
Wiharto et al. | Classification of blast cell type on acute myeloid leukemia (AML) based on image morphology of white blood cells | |
Benbakreti et al. | Impact of the data augmentation on the detection of brain tumor from MRI images based on CNN and pretrained models | |
Peng et al. | Blood vessels segmentation by using cdnet | |
Sari et al. | Cholesterol level detection through eye image using fractal and decision tree | |
Abdulrahman et al. | Classification of retina diseases from OCT using genetic programming | |
Singh et al. | Detection of Brain Tumors Through the Application of Deep Learning and Machine Learning Models |
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 |