CN109993735A - Image partition method based on concatenated convolutional - Google Patents
Image partition method based on concatenated convolutional Download PDFInfo
- Publication number
- CN109993735A CN109993735A CN201910248901.5A CN201910248901A CN109993735A CN 109993735 A CN109993735 A CN 109993735A CN 201910248901 A CN201910248901 A CN 201910248901A CN 109993735 A CN109993735 A CN 109993735A
- Authority
- CN
- China
- Prior art keywords
- medical image
- image
- pixel
- convolutional
- data set
- 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
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- 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
- 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
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30048—Heart; Cardiac
-
- 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
- G06T2207/30096—Tumor; Lesion
Abstract
The invention discloses a kind of image partition methods based on concatenated convolutional, are related to technical field of image segmentation.The image partition method based on concatenated convolutional is the following steps are included: step S1, acquire the medical images of multiple lesion regions;Step S2, the lesion region of collected medical image is successively carried out to artificial edge mark, to obtain label information;Step S3, label information is standardized pretreatment, to obtain two-dimentional data set;Step S4, the multilayer two-dimension convolutional neural networks based on concatenated convolutional are established, and multilayer two-dimension convolutional neural networks are trained using two-dimentional data set, to obtain neural network model;Step S5, it inputs the medical image of patient to be split and is standardized pretreatment, to obtain two-dimentional data set to be processed;Step S6, two-dimentional data set to be processed is input to the neural network model, the medical image of the patient to be split is divided automatically, to obtain the lesion region of the patient.
Description
Technical field
The present invention relates to technical field of image segmentation, more particularly to a kind of image partition method based on concatenated convolutional.
Background technique
The great attention for being constantly subjected to people of medical image segmentation, and partitioning algorithm is the key that influence segmentation effect
Factor, therefore scholars constantly study partitioning algorithm the characteristics of for medical image, emerges in large numbers many excellent calculations
Method, such as the dividing method based on boundary, the dividing method based on region, and corresponding age complementary medicine diagnosis person into
Row diagnoses and produces positive effect.
But as people's health consciousness gradually increases the continuous development with medical level, accuracy and efficiency to seeing a doctor
More stringent requirements are proposed.It is desirable to calculate it is machine-readable enter diagnosing patient CT (Computed Tomography, electricity
Sub- computed tomography), MRI (Magnetic Resonance, magnetic resonance imaging) image, be automatically and accurately diagnosed to be patient
The case where.More stringent requirements are proposed for this accuracy to image segmentation, and the very big bottleneck of image segmentation is the spy of image
Sign is extracted.The shape of the characteristics of having its own in medical image, organ can change, and space also occurs that overlapping, and phase
Gray value between adjacent organ is often relatively.It is at present, limited based on traditional image segmentation algorithm ability in feature extraction,
It is difficult accurately to be partitioned into target area, therefore does not just can guarantee the accuracy diagnosed automatically.And the appearance of deep learning, just
It can make up for it the defect of conventional segmentation methods, be preferably suitable for automatic diagnosis.
Deep learning has more hidden layers, uses the output of bottom as high-rise input.It is unsupervised from top to bottom
The process of habit, it can learn useful feature automatically, and be advanced features by rudimentary character representation;It is to have prison from top to bottom
The learning process superintended and directed optimizes the parameter of whole network by the data of tape label, adjusts, and has whole network more preferable
Feature learning ability.The study and expression structure of this feature have very strong robustness to deformation, the translation of image.In this way
Just solve the problems, such as that medical image features are not easy study, we can be in conjunction with traditional sorting algorithm, will not same district
Regional partition comes out, and marks interested region, is finally completed the segmentation of image.
In 2014, Long et al. proposed the side FCN (fully convolutional network, full convolutional network)
Method, devises a kind of input picture for arbitrary size, the frame of training convolutional network complete end to end, and realization divides pixel-by-pixel
Class has established the basic framework that image, semantic segmentation problem is solved using depth network.VGG (Visual is utilized in FCN method
Geometry Group Network, neural network) 16 networks, VGG16 network have 16 convolutional layers, 5 maximum pond layers,
3 full articulamentums and 1 softmax (normalization exponential function) layer.3 full articulamentums are converted to convolutional layer by FCN, and are moved
Except softmax layers, the network for being originally used for full figure classification is converted to the network for generating image segmentation.The segmentation of this method is accurate
Degree is 62.2%.
Chen et al. the end of FCN frame increase full connection CRF (conditional random field, condition with
Airport), DeepLab model is proposed, up-samples to obtain coarse segmentation using output result of the bilinear interpolation to FCN first
As a result, being a joint structure CRF model with each pixel in the result figure and solving, finally to segmentation result Stepwise Refinement.
The segmentation accuracy of this method is 71.6%.
CRF is modeled as Recognition with Recurrent Neural Network (recurrent neural network, abbreviation RNN) by Zheng et al..It should
Network can directly use BP (back propagation, back-propagation algorithm) algorithm to be trained, do not needed end to end
CNN and CRF model is respectively trained offline.The segmentation accuracy of this method is 72.0%.
Noh et al. improves the FCN network architecture, passes through study one and the full symmetric deconvolution of FCN network
On the one hand network can detecte the object instance of different stage scale in image, can only detect single ruler so as to avoid FCN
The drawbacks of spending semantic objects;On the other hand, by the combination of uncoiling lamination and anti-pond layer, in the pixel classifications figure of output more
Object detail is reflected well, has obtained the segmentation effect of high quality.The segmentation accuracy of this method is 72.5%.
Chen et al. is basic framework with FCN and introduces attention model, and attention model is applied to the scale of feature
Dimension, it may be assumed that change of scale is carried out to input picture first, the image after variation is used as to the input of FCN parallel, and for each
The corresponding Pixel-level weight figure of the image study of scale is used for FCN classification annotation pixel-by-pixel after the weighting of these weight figures.This method
Segmentation accuracy be 75.7%.
In the practical application of medical image segmentation, using the images such as the CT of patient or MRI as input, by analysis and
After processing, the result of segmentation is showed, and further according to segmentation as a result, according to the result of segmentation, integrative medicine circle
Knowledge, auxiliary related doctor quickly, efficient the case where being diagnosed to be patient, issue corresponding treatment scheme in time.
Image, semantic dividing method is established on the machine learning algorithm basis for having supervision end to end, is needed a large amount of
Pixel-level marks sample, and according to the research of existing literature, the time-consuming that Pixel-level mark is carried out to image is object in uncalibrated image
15 times of position time-consuming, therefore, the image of Pixel-level mark are difficult to largely obtain, and apply model which limits end-to-end method
It encloses.
Summary of the invention
The main purpose of the present invention is to provide a kind of image partition methods based on concatenated convolutional, it is intended to can be high-precision
Degree, automatic segmentation dilatancy myocardial region.
To achieve the above object, the present invention provides a kind of image partition method based on concatenated convolutional, comprising the following steps:
Step S1, the medical image of multiple lesion regions is acquired;
Step S2, the lesion region of collected medical image is successively carried out to artificial edge mark, to be marked
Sign information;
Step S3, the label information is standardized pretreatment, to obtain two-dimentional data set;
Step S4, the multilayer two-dimension convolutional neural networks based on concatenated convolutional are established, and utilize the two-dimentional data set pair
The multilayer two-dimension convolutional neural networks are trained, to obtain neural network model;
Step S5, it inputs the medical image of patient to be split and is standardized pretreatment, to obtain to be processed two
Dimension data collection;
Step S6, the two-dimentional data set to be processed is input to the neural network model, by the patient to be split
Medical image divided automatically, to obtain the lesion region of the patient.
Preferably, the medical image of the lesion region is the myocardial region medical image for suffering from dilatancy cardiac muscle.
Preferably, the step S3 the following steps are included:
Step S31, in the label information, selection includes the myocardial region medical image of dilatancy cardiac muscle;
Step S32: the myocardial region medical image in step S31 is subjected to resampling, to reach default resolution
The myocardial region medical image of rate;
Step S33, place is normalized in the myocardial region medical image for reaching default resolution ratio in step S32
Reason;
Step S34, the myocardial region medical image after normalized is cut to obtain pre-set dimension, and will cut out
Myocardial region medical image after cutting changes into two-dimentional data set.
Preferably, the step S33 further include:
The pixel value of myocardial region medical image after the normalized is between 0~255.
Preferably, the step S34 further include:
The myocardial region medical image cut after size be two-dimensional convolution neural network reception size.
Preferably, the step S4 the following steps are included:
Step S41, the two-dimentional data set is inputted in the multilayer two-dimension convolutional neural networks;
Step S42, according to the quantity of the two-dimentional data set, using five folding cross-validation methods to the two-dimentional data set into
Row cross validation when each cross validation, uses the two-dimentional data set of preset quantity as training sample, and remainder is as surveying
Sample sheet;And initialize the weight parameter of neuron in the convolutional layer and warp lamination of the multilayer two-dimension convolutional neural networks;
Step S43, the two-dimentional data set of a patient is inputted in the two-dimensional convolution neural network;
Step S44, the multilayer two-dimension convolutional neural networks are trained by propagated forward algorithm;Pass through normalization
Exponential function classifier exports the probability distribution of the pixel of myocardial region medical image, and presets a fixed threshold: if described
The probability value of some pixel of myocardial region medical image is greater than the fixed threshold, then is determined as the pixel for prediction just
The pixel really divided;If the probability value of some pixel of the myocardial region medical image is less than the fixed threshold, sentence
The fixed pixel is the pixel of prediction error segmentation;
Step S45, pass through the probability distribution and mark of the pixel of Jie Kade similarity algorithm calculating myocardium region medical image
Sign the error of information;
Step S46, optimize and update the convolutional layer and warp of the multilayer two-dimension convolutional neural networks based on the error
The weight parameter of neuron in lamination;
Step S47, step S43 to step S46 is repeated, until training loss and test loss no longer reduce.
Preferably, the Jie Kade similarity algorithm equation are as follows:
Wherein, P is the probability distribution of the pixel of myocardial region medical image, and T is label information, and PT is myocardial region medicine
The probability distribution of the pixel of image and the product of label information, ‖ P ‖2It is the L2- norm of P, ‖ T ‖2It is the L2- norm of T, ‖ PT ‖2It is
The L2- norm of PT.
Preferably, the step S46 further include:
Optimized by Adam majorized function neural in the convolutional layer and warp lamination of the multilayer two-dimension convolutional neural networks
The weight parameter of member.
Beneficial effect of the present invention has:
1, traditional-handwork or semi-automatic method heavy workload are directed to, marks the problems such as of low quality, error is big, the present invention can
The medical image of lesion region is divided automatically with realizing, and higher cutting accuracy can be obtained;
2, the present invention can be good at splitting lesion region, and the generalization ability of this method is very strong, be suitable for big
The different patient of most lesion regions.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow diagrams of the image partition method of concatenated convolutional;
Fig. 2 is the flow diagram of the embodiment of the present invention;
Fig. 3 is that the present invention is based on the network structures in the image partition method of concatenated convolutional.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
The following further describes the present invention with reference to the drawings.
As shown in Figure 1 and Figure 2, the embodiment of the present invention provides a kind of image partition method based on concatenated convolutional, including following
Step:
Step S1, the medical image of multiple lesion regions is acquired.
Specifically, the medical image of the lesion region is the myocardial region medical image for suffering from dilatancy cardiac muscle.
In a particular embodiment, 1155 cases for suffering from dilatancy cardiac muscle are acquired, the medicine figure of its myocardial sites MRI is acquired
Picture.
Step S2, the lesion myocardial region of collected medical image is successively subjected to artificial edge mark, with
To label information.
In a particular embodiment, by experienced dilatancy cardiac muscle doctor, edge mark is successively carried out for lesion region,
To obtain 1155 label informations.
Step S3, the label information is standardized pretreatment, to obtain two-dimentional data set:
Step S31, in the label information, selection includes the myocardial region medical image of dilatancy cardiac muscle;
Step S32: the myocardial region medical image in step S31 is subjected to resampling, to reach default resolution
The myocardial region medical image of rate;In a particular embodiment, myocardial region medical image is resurveyed to 1.0*1.0*
1.0mm3Resolution ratio.The resolution ratio can be configured according to actual use situation.
Step S33, the myocardial region medical image for being up to default resolution ratio is normalized;The normalizing
Change the pixel value of treated myocardial region medical image between 0~255.
Step S34, the myocardial region medical image after normalized is cut to obtain pre-set dimension, and will cut out
Myocardial region medical image after cutting changes into two-dimentional data set.
Specifically, the size after the myocardial region medical image is cut is the reception ruler of two-dimensional convolution neural network
It is very little.
Step S4, the multilayer two-dimension convolutional neural networks based on concatenated convolutional are established, and utilize the two-dimentional data set pair
The multilayer two-dimension convolutional neural networks are trained, to obtain neural network model.As shown in figure 3, the two-dimensional convolution is neural
The structure of network includes down sample module and coder-decoder module.Down sample module is made of seven layers of convolutional layer, and this seven layers
The convolution kernel of convolutional layer is in the same size, for extracting deeper, abstract semantic feature.Coder-decoder module is by four layers
Convolutional layer and four layers of warp lamination are constituted, and in encoder layer, each layer of input is added as next layer with the output of this layer
Input, in decoder layer, each layer of warp lamination all merges the convolutional layer feature of corresponding encoder layer, to improve feature
Reuse rate, avoids Character losing:
Step S41, the two-dimentional data set is inputted in the multilayer two-dimension convolutional neural networks;It is carried out with label information
It is training sample training two-dimensional convolution neural network that two-dimentional data set is obtained after standardization pretreatment.
Step S42, according to the quantity of the two-dimentional data set, using five folding cross-validation methods to the two-dimentional data set into
Row cross validation when each cross validation, uses the two-dimentional data set of preset quantity as training sample, and remainder is as surveying
Sample sheet;And initialize the weight parameter of neuron in the convolutional layer and warp lamination of the multilayer two-dimension convolutional neural networks.
In a particular embodiment, using 80% two-dimentional data set in 1155 patients as training sample, residue 20%
Two-dimentional data set as test sample carry out cross validation.
Step S43, the two-dimentional data set of a patient is inputted in the two-dimensional convolution neural network;
Step S44, the multilayer two-dimension convolutional neural networks are trained by propagated forward algorithm;Pass through normalization
Exponential function classifier exports the probability distribution of the pixel of myocardial region medical image, and presets a fixed threshold: if described
The probability value of some pixel of myocardial region medical image is greater than the fixed threshold, then is determined as the pixel for prediction just
The pixel really divided;If the probability value of some pixel of the myocardial region medical image is less than the fixed threshold, sentence
The fixed pixel is the pixel of prediction error segmentation;
Specifically, multilayer two-dimension convolutional neural networks are trained by propagated forward, input the 2-D data of patient
Collection is exported by each layer network of multilayer two-dimension convolutional neural networks, and the output on upper layer is input to next layer again, and repeating should
To the last one layer of process, and fixed threshold m=0.3 is set, finally export dilatancy cardiac muscle figure.The fixed threshold can be according to reality
Border situation is configured.
Step S45, pass through the probability distribution and mark of the pixel of Jie Kade similarity algorithm calculating myocardium region medical image
Sign the error of information;The Jie Kade similarity algorithm equation are as follows:
Wherein, P is the probability distribution of the pixel of myocardial region medical image, and T is label information, and PT is myocardial region medicine
The probability distribution of the pixel of image and the product of label information, ‖ P ‖2It is the L2- norm of P, ‖ T ‖2It is the L2- norm of T, ‖ PT ‖2It is
The L2- norm of PT.
Step S46, optimize and update the convolutional layer and warp of the multilayer two-dimension convolutional neural networks based on the error
The weight parameter of neuron in lamination;Specifically, the multilayer two-dimension convolutional neural networks are optimized by Adam majorized function
The weight parameter of neuron in convolutional layer and warp lamination.In other embodiments, it is possible to use other excellent in the prior art
Change function.
Step S47, step S43 to step S46 is repeated, until training loss and test loss no longer reduce.
Step S5, it inputs the medical image of patient to be split and is standardized pretreatment, to obtain to be processed two
Dimension data collection.For case to be diagnosed, the medical image with the same mode in position is acquired, at identical pretreatment
Reason.
Step S6, the two-dimentional data set to be processed is input to the neural network model, by the patient to be split
Medical image divided automatically, to obtain the lesion region of the patient.For the case to be diagnosed, the disease is inputted
The lesion region for suffering from dilatancy cardiac muscle of the patient can be obtained in the entire MRI data of people, the model obtained by above-mentioned training.
The present invention is based on the multiple dimensioned networks of concatenated convolutional, and combine U-Net (U-shaped convolutional neural networks) and ResNet
One kind that the advantages of (Residual Neural Network, residual error neural network) reconfigures is divided for dilatancy cardiac muscle
Method.The multilayer two-dimension convolutional neural networks first pass around the down sample module of seven layers of convolutional layer composition, which can mention
Get deeper, higher semantic feature.Then the coding formed using four layers of convolutional layer and four layers of warp lamination
Device-decoder module merges each layer of input and the output phase adduction of this layer in the part feature of corresponding encoder layer
As next layer of input, this improves feature reuse rates, avoid the loss of feature.
As it can be seen from table 1 U-Net network is compared, using identical Jie Kade similarity as loss function, the present invention
The neural network model of image partition method based on concatenated convolutional has all reached preferable as a result, the present invention in four indexs
Average DSC (Differential Scanning Calorimeter, differential scanning calorimetry) value of neural network model reach
0.80, the effect more than U-Net illustrates that the image partition method of the invention based on concatenated convolutional is greatly improved segmentation
Performance.Wherein, AUC is Area Under Curve, model evaluation index;F-Measure is that accuracy and recall rate weighting are adjusted
With average, for classification of assessment model quality.
The dilatancy cardiac muscle segmentation result comparison sheet of table 1U-Net and multilayer two-dimension convolutional neural networks
It should be understood that the above is only a preferred embodiment of the present invention, the scope of the patents of the invention cannot be therefore limited,
It is all to utilize equivalent structure or equivalent flow shift made by description of the invention and accompanying drawing content, it is applied directly or indirectly in
Other related technical areas are included within the scope of the present invention.
Claims (8)
1. a kind of image partition method based on concatenated convolutional, which comprises the following steps:
Step S1, the medical image of multiple lesion regions is acquired;
Step S2, the lesion region of collected medical image is successively carried out to artificial edge mark, to obtain label letter
Breath;
Step S3, the label information is standardized pretreatment, to obtain two-dimentional data set;
Step S4, the multilayer two-dimension convolutional neural networks based on concatenated convolutional are established, and using the two-dimentional data set to described
Multilayer two-dimension convolutional neural networks are trained, to obtain neural network model;
Step S5, it inputs the medical image of patient to be split and is standardized pretreatment, to obtain two-dimemsional number to be processed
According to collection;
Step S6, the two-dimentional data set to be processed is input to the neural network model, by the doctor of the patient to be split
It learns image information to be divided automatically, to obtain the lesion region of the patient.
2. the image partition method according to claim 1 based on concatenated convolutional, which is characterized in that the lesion region
Medical image is the myocardial region medical image for suffering from dilatancy cardiac muscle.
3. the image partition method according to claim 2 based on concatenated convolutional, which is characterized in that the step S3 includes
Following steps:
Step S31, in the label information, selection includes the myocardial region medical image of dilatancy cardiac muscle;
Step S32: the myocardial region medical image in step S31 is subjected to resampling, to reach default resolution ratio
Myocardial region medical image;
Step S33, the myocardial region medical image for reaching default resolution ratio in step S32 is normalized;
Step S34, the myocardial region medical image after normalized is cut to obtain pre-set dimension, and will be after cutting
Myocardial region medical image change into two-dimentional data set.
4. the image partition method according to claim 3 based on concatenated convolutional, which is characterized in that the step S33 is also
Include:
The pixel value of myocardial region medical image after the normalized is between 0~255.
5. the image partition method according to claim 3 based on concatenated convolutional, which is characterized in that the step S34 is also
Include:
The myocardial region medical image cut after size be two-dimensional convolution neural network reception size.
6. the image partition method according to claim 1 based on concatenated convolutional, which is characterized in that the step S4 includes
Following steps:
Step S41, the two-dimentional data set is inputted in the multilayer two-dimension convolutional neural networks;
Step S42, according to the quantity of the two-dimentional data set, the two-dimentional data set is handed over using five folding cross-validation methods
Fork verifying, when each cross validation, uses the two-dimentional data set of preset quantity as training sample, remainder is as test specimens
This;And initialize the weight parameter of neuron in the convolutional layer and warp lamination of the multilayer two-dimension convolutional neural networks;
Step S43, the two-dimentional data set of a patient is inputted in the two-dimensional convolution neural network;
Step S44, the multilayer two-dimension convolutional neural networks are trained by propagated forward algorithm;By normalizing index
Function category device exports the probability distribution of the pixel of myocardial region medical image, and presets a fixed threshold: if the cardiac muscle
The probability value of some pixel of region medical image is greater than the fixed threshold, then is determined as the pixel for correct point of prediction
The pixel cut;If the probability value of some pixel of the myocardial region medical image is less than the fixed threshold, determining should
Pixel is the pixel of prediction error segmentation;
Step S45, believed by the probability distribution of the pixel of Jie Kade similarity algorithm calculating myocardium region medical image and label
The error of breath;
Step S46, optimize and update the convolutional layer and warp lamination of the multilayer two-dimension convolutional neural networks based on the error
The weight parameter of middle neuron;
Step S47, step S43 to step S46 is repeated, until training loss and test loss no longer reduce.
7. the image partition method according to claim 6 based on concatenated convolutional, which is characterized in that the Jie Kade is similar
Spend algorithm equations are as follows:
Wherein, P is the probability distribution of the pixel of myocardial region medical image, and T is label information, and PT is myocardial region medical image
Pixel probability distribution and label information product, | | P | |2It is the L2- norm of P, | | T | |2It is the L2- norm of T, | | PT | |2
It is the L2- norm of PT.
8. the image partition method according to claim 6 based on concatenated convolutional, which is characterized in that the step S46 is also
Include:
Optimize neuron in the convolutional layer and warp lamination of the multilayer two-dimension convolutional neural networks by Adam majorized function
Weight parameter.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910248901.5A CN109993735A (en) | 2019-03-29 | 2019-03-29 | Image partition method based on concatenated convolutional |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910248901.5A CN109993735A (en) | 2019-03-29 | 2019-03-29 | Image partition method based on concatenated convolutional |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109993735A true CN109993735A (en) | 2019-07-09 |
Family
ID=67131716
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910248901.5A Pending CN109993735A (en) | 2019-03-29 | 2019-03-29 | Image partition method based on concatenated convolutional |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109993735A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111047606A (en) * | 2019-12-05 | 2020-04-21 | 北京航空航天大学 | Pathological full-section image segmentation algorithm based on cascade thought |
CN111128345A (en) * | 2019-11-19 | 2020-05-08 | 上海联影智能医疗科技有限公司 | Medical image acquisition method, medical scanning device and computer storage medium |
CN111340828A (en) * | 2020-01-10 | 2020-06-26 | 南京航空航天大学 | Brain glioma segmentation based on cascaded convolutional neural networks |
CN111445440A (en) * | 2020-02-20 | 2020-07-24 | 上海联影智能医疗科技有限公司 | Medical image analysis method, equipment and storage medium |
CN111784628A (en) * | 2020-05-11 | 2020-10-16 | 北京工业大学 | End-to-end colorectal polyp image segmentation method based on effective learning |
CN111862123A (en) * | 2020-07-29 | 2020-10-30 | 南通大学 | CT abdominal artery blood vessel classification identification method based on deep learning |
CN111932501A (en) * | 2020-07-13 | 2020-11-13 | 太仓中科信息技术研究院 | Seal ring surface defect detection method based on semantic segmentation |
CN112183448A (en) * | 2020-10-15 | 2021-01-05 | 中国农业大学 | Hulled soybean image segmentation method based on three-level classification and multi-scale FCN |
ES2813777A1 (en) * | 2019-09-23 | 2021-03-24 | Quibim S L | METHOD AND SYSTEM FOR THE AUTOMATIC SEGMENTATION OF HYPERINTENSITIES OF WHITE SUBSTANCE IN BRAIN MAGNETIC RESONANCE IMAGES (Machine-translation by Google Translate, not legally binding) |
CN112669313A (en) * | 2021-01-15 | 2021-04-16 | 济南浪潮高新科技投资发展有限公司 | Metal surface defect positioning and classifying method |
WO2021104378A1 (en) * | 2019-11-29 | 2021-06-03 | Shanghai United Imaging Intelligence Co., Ltd. | System and method for reconstructing magnetic resonance images |
CN112907608A (en) * | 2021-03-17 | 2021-06-04 | 北京智象云中医科技有限公司 | Traditional Chinese medicine infrared image human body segmentation algorithm based on deep learning |
CN116721019A (en) * | 2023-08-10 | 2023-09-08 | 成都信息工程大学 | Multi-camera video image stitching method based on deep learning |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106934397A (en) * | 2017-03-13 | 2017-07-07 | 北京市商汤科技开发有限公司 | Image processing method, device and electronic equipment |
CN108053417A (en) * | 2018-01-30 | 2018-05-18 | 浙江大学 | A kind of lung segmenting device of the 3DU-Net networks based on mixing coarse segmentation feature |
CN108062754A (en) * | 2018-01-19 | 2018-05-22 | 深圳大学 | Segmentation, recognition methods and device based on dense network image |
CN109087258A (en) * | 2018-07-27 | 2018-12-25 | 中山大学 | A kind of image rain removing method and device based on deep learning |
CN109389584A (en) * | 2018-09-17 | 2019-02-26 | 成都信息工程大学 | Multiple dimensioned rhinopharyngeal neoplasm dividing method based on CNN |
-
2019
- 2019-03-29 CN CN201910248901.5A patent/CN109993735A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106934397A (en) * | 2017-03-13 | 2017-07-07 | 北京市商汤科技开发有限公司 | Image processing method, device and electronic equipment |
CN108062754A (en) * | 2018-01-19 | 2018-05-22 | 深圳大学 | Segmentation, recognition methods and device based on dense network image |
CN108053417A (en) * | 2018-01-30 | 2018-05-18 | 浙江大学 | A kind of lung segmenting device of the 3DU-Net networks based on mixing coarse segmentation feature |
CN109087258A (en) * | 2018-07-27 | 2018-12-25 | 中山大学 | A kind of image rain removing method and device based on deep learning |
CN109389584A (en) * | 2018-09-17 | 2019-02-26 | 成都信息工程大学 | Multiple dimensioned rhinopharyngeal neoplasm dividing method based on CNN |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
ES2813777A1 (en) * | 2019-09-23 | 2021-03-24 | Quibim S L | METHOD AND SYSTEM FOR THE AUTOMATIC SEGMENTATION OF HYPERINTENSITIES OF WHITE SUBSTANCE IN BRAIN MAGNETIC RESONANCE IMAGES (Machine-translation by Google Translate, not legally binding) |
WO2021058843A1 (en) * | 2019-09-23 | 2021-04-01 | Quibim, S.L. | Method and system for the automatic segmentation of white matter hyperintensities in brain magnetic resonance images |
CN111128345A (en) * | 2019-11-19 | 2020-05-08 | 上海联影智能医疗科技有限公司 | Medical image acquisition method, medical scanning device and computer storage medium |
CN111128345B (en) * | 2019-11-19 | 2024-03-22 | 上海联影智能医疗科技有限公司 | Medical image acquisition method, medical scanning device and computer storage medium |
WO2021104378A1 (en) * | 2019-11-29 | 2021-06-03 | Shanghai United Imaging Intelligence Co., Ltd. | System and method for reconstructing magnetic resonance images |
US11348291B2 (en) | 2019-11-29 | 2022-05-31 | Shanghai United Imaging Intelligence Co., Ltd. | System and method for reconstructing magnetic resonance images |
CN111047606A (en) * | 2019-12-05 | 2020-04-21 | 北京航空航天大学 | Pathological full-section image segmentation algorithm based on cascade thought |
CN111047606B (en) * | 2019-12-05 | 2022-10-04 | 北京航空航天大学 | Pathological full-section image segmentation algorithm based on cascade thought |
CN111340828A (en) * | 2020-01-10 | 2020-06-26 | 南京航空航天大学 | Brain glioma segmentation based on cascaded convolutional neural networks |
CN111445440A (en) * | 2020-02-20 | 2020-07-24 | 上海联影智能医疗科技有限公司 | Medical image analysis method, equipment and storage medium |
CN111445440B (en) * | 2020-02-20 | 2023-10-31 | 上海联影智能医疗科技有限公司 | Medical image analysis method, device and storage medium |
CN111784628A (en) * | 2020-05-11 | 2020-10-16 | 北京工业大学 | End-to-end colorectal polyp image segmentation method based on effective learning |
CN111784628B (en) * | 2020-05-11 | 2024-03-29 | 北京工业大学 | End-to-end colorectal polyp image segmentation method based on effective learning |
CN111932501A (en) * | 2020-07-13 | 2020-11-13 | 太仓中科信息技术研究院 | Seal ring surface defect detection method based on semantic segmentation |
CN111862123B (en) * | 2020-07-29 | 2024-01-23 | 南通大学 | Deep learning-based CT abdominal artery blood vessel hierarchical recognition method |
CN111862123A (en) * | 2020-07-29 | 2020-10-30 | 南通大学 | CT abdominal artery blood vessel classification identification method based on deep learning |
CN112183448B (en) * | 2020-10-15 | 2023-05-12 | 中国农业大学 | Method for dividing pod-removed soybean image based on three-level classification and multi-scale FCN |
CN112183448A (en) * | 2020-10-15 | 2021-01-05 | 中国农业大学 | Hulled soybean image segmentation method based on three-level classification and multi-scale FCN |
CN112669313A (en) * | 2021-01-15 | 2021-04-16 | 济南浪潮高新科技投资发展有限公司 | Metal surface defect positioning and classifying method |
CN112907608A (en) * | 2021-03-17 | 2021-06-04 | 北京智象云中医科技有限公司 | Traditional Chinese medicine infrared image human body segmentation algorithm based on deep learning |
CN116721019B (en) * | 2023-08-10 | 2023-10-10 | 成都信息工程大学 | Multi-camera video image stitching method based on deep learning |
CN116721019A (en) * | 2023-08-10 | 2023-09-08 | 成都信息工程大学 | Multi-camera video image stitching method based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109993735A (en) | Image partition method based on concatenated convolutional | |
CN110930416B (en) | MRI image prostate segmentation method based on U-shaped network | |
CN111145170B (en) | Medical image segmentation method based on deep learning | |
CN107610087B (en) | Tongue coating automatic segmentation method based on deep learning | |
CN109376636B (en) | Capsule network-based eye fundus retina image classification method | |
CN105469100B (en) | Skin biopsy image pathological characteristics recognition methods based on deep learning | |
CN108460726A (en) | A kind of magnetic resonance image super-resolution reconstruction method based on enhancing recurrence residual error network | |
CN107437092A (en) | The sorting algorithm of retina OCT image based on Three dimensional convolution neutral net | |
CN104346617B (en) | A kind of cell detection method based on sliding window and depth structure extraction feature | |
CN107909566A (en) | A kind of image-recognizing method of the cutaneum carcinoma melanoma based on deep learning | |
CN111627019A (en) | Liver tumor segmentation method and system based on convolutional neural network | |
CN108334899A (en) | Quantify the bone age assessment method of information integration based on hand bone X-ray bone and joint | |
CN110930421A (en) | Segmentation method for CBCT (Cone Beam computed tomography) tooth image | |
CN109166102A (en) | It is a kind of based on critical region candidate fight network image turn image interpretation method | |
CN109544518A (en) | A kind of method and its system applied to the assessment of skeletal maturation degree | |
CN111915594A (en) | End-to-end neural network-based breast cancer focus segmentation method | |
CN110298303A (en) | A kind of crowd recognition method based on the long pan of memory network in short-term path learning | |
CN111916206B (en) | CT image auxiliary diagnosis system based on cascade connection | |
CN109165658A (en) | A kind of strong negative sample underwater target detection method based on Faster-RCNN | |
CN110459303B (en) | Medical image abnormity detection device based on depth migration | |
CN114398979A (en) | Ultrasonic image thyroid nodule classification method based on feature decoupling | |
CN109671060A (en) | Area of computer aided breast lump detection method based on selective search and CNN | |
Cao et al. | Research for face image super-resolution reconstruction based on wavelet transform and SRGAN | |
CN116864103A (en) | Myopenia diagnosis method based on multi-modal contrast learning | |
CN111383759A (en) | Automatic pneumonia diagnosis system |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190709 |
|
RJ01 | Rejection of invention patent application after publication |