CN106780453A - A kind of method realized based on depth trust network to brain tumor segmentation - Google Patents
A kind of method realized based on depth trust network to brain tumor segmentation Download PDFInfo
- Publication number
- CN106780453A CN106780453A CN201611115044.4A CN201611115044A CN106780453A CN 106780453 A CN106780453 A CN 106780453A CN 201611115044 A CN201611115044 A CN 201611115044A CN 106780453 A CN106780453 A CN 106780453A
- Authority
- CN
- China
- Prior art keywords
- image
- segmentation
- brain tumor
- trust network
- oedema
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The present invention realizes the segmentation to brain tumor by the method based on depth trust network, is beneficial to the auxiliary diagnosis to patient's brain tumor disease.The method processes original image first with sef-adapting filter, histogram equalization and luminance transformation, reduces the noise of original image and strengthens the contrast of image.Then data set is generated to the image zooming-out image block for treating.Then classification is carried out to oedema, necrosis, tumor region in complete brain tumor using depth trust network and realizes just segmentation, recycle fuzzy C-means clustering segmentation to the further Accurate Segmentation of image, the result for obtaining is the segmentation result of complete brain tumor.
Description
Technical field
The invention belongs to the identification of computer Medical Images Classification and the category of segmentation, more specifically, it is related to a kind of base
The dividing method to brain tumor is realized in depth trust network, automatic segmentation is realized to brain tumor for medical assistance.
Background technology
Today's society, brain tumor more threatens the health of people.Statistics display patient with brain tumors number is increasing.
The characteristics of there is complex structure and important function due to brain, various informative, the limit that clinically brain tumor and its tumour sample become
The early diagnosis and therapy of brain tumor disease is made.With the progress of the technology of computer-aided medical diagnosis, successfully
For mammary gland and the auxiliary diagnosis of pulmonary lesion.If the Accurate Segmentation of medical image can be realized using computer, will be favourable
Yu doctor timely diagnoses and proposes corresponding therapeutic scheme.MRI (MRI) has "dead" infringement, bone free pseudo-
Shadow, can many-sided multi-parameter imaging.It is particularly suitable for clinical brain lesionses inspection, is that doctor timely diagnoses and proposes corresponding
Therapeutic scheme provides condition.MRI has turned into clinical Main Diagnosis means.But current clinical brain tumor
Analysis is also rested in qualitatively analysis mostly, and the method for quantitative analysis brain tumor is also rare at present.Artificial segmentation takes very much
Thing and because people's difference may produce unpredictable mistake.The accurately and effectively segmentation of brain tumor is realized, is computer medical science
The very important research topic of process field.
Clinical brain tumor segmentation is based on artificial segmentation, but artificial segmentation is more rough and excessively bothersome, so with
Computer realizes that automatic segmentation becomes very meaningful as the mode of medical aided diagnosis.But medical image is different from
The specific characteristic of other images, is used alone any traditional partitioning algorithm, is all extremely difficult to preferable effect.Correlation is learned
The development of section and the proposition of some new theories and method are filled with new blood for Medical Image Segmentation Techniques, researchers by this
A little New technical uses have obtained preferable segmentation effect in practice.Wherein most of segmentations are with image basic handling, feature
Extraction and the technology of feature selecting, image classification and segmentation.More conventional image partition method is related to fuzzy set reason at present
The method of opinion, based on statistical method etc.；Classifying identification method then linear differentiation, k nearest neighbor, neutral net etc..While this
A little methods all have certain limitation and most downright bad and water for being used for single lesion segmentation, have ignored in complete brain tumor
The segmentation in swollen region.Turn into the Hot subject of computer realm with machine learning, its simulation mankind learning behavior is obtaining
The characteristic of new knowledge and skills is used to process high dimensional data, such as Medical Images Classification identification by more.Depth trust network,
Convolutional neural networks and automatic stacking coding are all the research directions of hot topic in machine learning.Depth trust network earliest by
Hinton 2006 propose, be on the basis of RBM (limited Boltzmann machine), by being continuously increased hidden layer number, while
RBM is used in the top layer of network, and bayesian belief network is used in other layers, so as to obtain DBN (depth trust network).
DBN belongs to the category of pattern-recognition, is generally used for the classification of data, and it is passed in the training process of input data by forward direction
Broadcast and constantly optimize network parameter with the iteration of backpropagation, then according to the different pixel of true value image or image block type
Pixel or image block to respective image correspondence position add different labels and realize classification, and disaggregated model is mapped as into two-value
Image obtains just segmentation result.Result of the DBN output results better than traditional neural network algorithm obtained using pre-training, herein
On the basis of obtain accurate brain tumor segmentation result using fuzzy C-means clustering (F C M) method.
The content of the invention
It is an object of the invention to design a kind of side for realizing splitting complete brain tumor based on depth trust network
Method.In cutting procedure, many classification (including water is carried out for complete brain tumor by depth trust network using view data
Swollen, necrosis and tumour) while mapping the bianry image for forming just segmentation.
Realization preferably splits purpose, it is necessary to carry out the operation such as pre-processing to image, mainly including herein below：First
According to the image-forming principle of MRI, the noise in removal MRI is strengthened the contrast and details of image using luminance transformation, recycled
Histogram equalization reduction because device parameter is different and caused by brightness of image, the difference of contrast, obtain substantially intensity profile
Uniform image.Then the true value image to pretreated image zooming-out image block and complete brain tumor generates input data
Collection, then many classification are carried out to tumour, oedema, necrotic zone by depth trust network, finally carried out using fuzzy C-means clustering
Accurate Segmentation.
Know-why is as shown in figure 1, particular technique flow is as follows：
Step one：Image is processed first with sef-adapting filter, there is influence classification in image to eliminate
Noise, and strengthen the contrast of tumour, oedema and necrotic zone in brain image by histogram equalization and luminance transformation.
Step 2：The image processed using step one carries out the extraction of image block, and generates corresponding input data (instruction
Practice and test data set).
Step 3：The potential feature of data in step 2 is extracted using unsupervised depth trust network, according to
The training of feature adds label to respective data blocks and many classification (1- oedema, 2- necrosis, 3- tumours) is realized to complete brain tumor, obtains
To many disaggregated models.
Step 4：Using the step 3 disaggregated model first segmentation result of mapping generation, according to F C M to the image after just segmentation
Split, the result of segmentation is the complete brain tumor with oedema, necrosis and tumor region.
Brief description of the drawings
Fig. 1 is that the present invention is based on technical scheme figure of the depth trust network realization to the dividing method of brain tumor.
Specific embodiment
Specific embodiment of the invention is described below in conjunction with the accompanying drawings, so as to those skilled in the art preferably
Understand the present invention.Requiring particular attention is that, in the following description, may desalination and ignore it is relevant with the present invention
Know the content introduction of function and design.
In the present embodiment, the present invention mainly includes following link to brain tumor dividing method：
1. image preprocessing, 2. image block extract and the generation of data set, 3. many classification of complete brain tumor, 4. fuzzy C-mean algorithm
Cluster segmentation
Image preprocessing is using filtering, histogram equalization and luminance transformation, image zooming-out image block generation input data
Collection, the use depth trust network of innovation is classified to complete brain tumor according to oedema, necrosis and tumor region, will be classified
Model mapping generation bianry image obtains just segmentation result, and its step is as follows：1 according to true value to the image block that is extracted in image
Setting type.Type with image block central point is the type of the block, to oedema, necrosis and tumor region in complete brain tumor
Setting type, is designated as 1- oedema, 2- necrosis, 3- tumours respectively.In 2 depth trust network training process, using true value corresponding diagram
As the type of block sets type to the corresponding image block of MRI.3 using forward-propagating and the successive ignition of backpropagation, optimization
Network parameter, then according to the different pixel of true value image or image block type to the pixel of respective image correspondence position or
Image block adds different labels and realizes classification, generates the disaggregated model of complete brain tumor.Disaggregated model in 3 is mapped life by 4
Into bianry image, just segmentation result is obtained.In last fuzzy C-means clustering cutting procedure, corresponding cluster is divided an image into,
To existing cluster, degree of membership square M, the M=[M of data pixels are randomly choosedij],M(0).By M(k)Calculate center vector R(k)=
[Rj], wherein k is iterations；Calculate the subordinated-degree matrix M after updating(k+1)=[Mij (k+1)].When | | M(k+1)-M(k)| |, wherein
| | * | | calculates the distance between central point and pixel, meets the corresponding error rate requirements of FCM and then terminates, and otherwise makes M(k)=
[Mij (k+1)], continue iteration.Using brain tumor and the difference of the normal structure pixel degree of membership of surrounding, obtain with oedema, bad
The segmentation result of the complete brain tumor of dead and tumour.
The present invention is a kind of to realize having the characteristics that the dividing method of brain tumor based on depth trust network：
The present invention proposes that one kind carries out dividing method to complete brain tumor, can accurately to the segmentation of complete brain tumor.
The use depth trust network of innovation effectively can be split to the complete brain tumor with oedema, necrosis and tumour.Brain
The segmentation result of tumour can aid in doctor to the diagnosis of patient's state of an illness and propose therapeutic scheme.
Although being described to illustrative specific embodiment of the invention above, in order to the technology people of this technology neck
Member understands the present invention, it should be apparent that the invention is not restricted to the scope of specific embodiment, to the ordinary skill of the art
For personnel, as long as various change is in appended claim restriction and the spirit and scope of the present invention for determining, these changes
Change is it will be apparent that all utilize the innovation and creation of present inventive concept in the row of protection.
Claims (2)
1. the present invention is a kind of method realized based on depth trust network to brain tumor segmentation, mainly including herein below：It is first
First image is pre-processed, then many classification (including oedema, necrosis is carried out for complete brain tumor using depth trust network
And tumour) while map the bianry image for forming just segmentation, then realize accurate point with fuzzy C-means clustering (FCM) method
Cut.
Technical scheme is as follows：
Step one：Image is processed first with sef-adapting filter, to eliminate the noise classified in the presence of influence in image,
And strengthen the contrast of tumour, oedema and necrotic zone in brain image by histogram equalization, luminance transformation, be beneficial to
Improve the degree of accuracy of classification and segmentation.
Step 2：Using step one process image carry out the extraction of image block, and generate corresponding input data (training and
Test data set).
Step 3：The potential feature of data in step 2 is extracted using unsupervised depth trust network, according to feature
Training type is set to respective data blocks, realize, to many classification of complete brain tumor (1- oedema, 2- necrosis, 3- tumours), obtaining
To many disaggregated models.
Step 4：Using the step 3 disaggregated model first segmentation result of mapping generation, the image after just segmentation is divided according to FCM
Cut, the result of segmentation is oedema, necrosis and tumor region.
2. it is according to claim 1 it is a kind of based on depth trust network realize brain tumor segmentation method.It is characterized in that
During using depth trust network, according to the gray feature of image block, many classification (water to complete brain tumor are realized
Swollen, necrosis and tumor region) and segmentation, only tumor region therein is classified and split rather than single.
The main characteristic of the invention lies in that the process split to complete brain tumor carries out some improvement, to improve the standard of segmentation
True property.Mainly include：(1) image is pre-processed using adaptive-filtering, histogram equalization and luminance transformation, (2) are right
Pretreatment image zooming-out image block generation input data set, (3) using depth trust network to complete brain tumor according to oedema,
Necrosis and tumor region are classified, and disaggregated model mapping generation bianry image is obtained into just segmentation result.(4) using fuzzy
C mean clusters dividing method is further split to first segmentation result, obtains accurate segmentation result.
Complete brain tumor is classified using depth trust network, its step is as follows：
1. type is set to the image block extracted in image according to true value.Type with image block central point is the class of the block
Type, to oedema, necrosis and tumor region setting type in complete brain tumor, is designated as 1- oedema, 2- necrosis, 3- tumours respectively.
2. in depth trust network training process, the corresponding image block of image is set using the type of true value correspondence image block
Type.
3., using forward-propagating and the successive ignition of backpropagation, network parameter is optimized, then according to the picture that true value image is different
Vegetarian refreshments or image block type are added different labels and realize classification generation to the pixel or image block of respective image correspondence position
The disaggregated model of complete brain tumor.
4., by the disaggregated model mapping generation bianry image in 3, just segmentation result is obtained.
In last fuzzy C-means clustering cutting procedure, corresponding cluster is divided an image into, to existing cluster, random selection
The degree of membership square M of data pixels.
M=[Mij],M(0)。
By M(k)Calculate center vector R(k)=[Rj], wherein k is iterations；Calculate the subordinated-degree matrix M after updating(k+1)=
[Mij (k+1)].When | | M(k+1)-M(k)| |, wherein | | * | | calculates the distance between central point and pixel, meet FCM wrong accordingly
Rate requirement by mistake then terminates, and otherwise makes M(k)=[Mij (k+1)], continue iteration.Normal structure pixel using brain tumor and surrounding is subordinate to
The difference of category degree, obtains last segmentation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611115044.4A CN106780453A (en) | 2016-12-07 | 2016-12-07 | A kind of method realized based on depth trust network to brain tumor segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611115044.4A CN106780453A (en) | 2016-12-07 | 2016-12-07 | A kind of method realized based on depth trust network to brain tumor segmentation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106780453A true CN106780453A (en) | 2017-05-31 |
Family
ID=58874780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611115044.4A Pending CN106780453A (en) | 2016-12-07 | 2016-12-07 | A kind of method realized based on depth trust network to brain tumor segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106780453A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292884A (en) * | 2017-08-07 | 2017-10-24 | 北京深睿博联科技有限责任公司 | The method and device of oedema and hemotoncus in a kind of identification MRI image |
| CN107845098A (en) * | 2017-11-14 | 2018-03-27 | 南京理工大学 | Liver cancer image full-automatic partition method based on random forest and fuzzy clustering |
| CN108629772A (en) * | 2018-05-08 | 2018-10-09 | 上海商汤智能科技有限公司 | Image processing method and device, computer equipment and computer storage media |
| CN109102512A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of MRI brain tumor image partition method based on DBN neural network |
| CN110706225A (en) * | 2019-10-14 | 2020-01-17 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Tumor identification system based on artificial intelligence |
| CN111445456A (en) * | 2020-03-26 | 2020-07-24 | 北京推想科技有限公司 | Classification model, network model training method and device, and identification method and device |
| CN112686845A (en) * | 2020-12-23 | 2021-04-20 | 合肥联宝信息技术有限公司 | Image processing method and device and computer readable medium |
| CN113240964A (en) * | 2021-05-13 | 2021-08-10 | 广西英腾教育科技股份有限公司 | Cardiopulmonary resuscitation teaching machine |
| US11227387B2 (en) | 2020-05-18 | 2022-01-18 | Prince Mohammad Bin Fahd University | Multi-stage brain tumor image processing method and system |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104834943A (en) * | 2015-05-25 | 2015-08-12 | 电子科技大学 | Brain tumor classification method based on deep learning |
-
2016
- 2016-12-07 CN CN201611115044.4A patent/CN106780453A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104834943A (en) * | 2015-05-25 | 2015-08-12 | 电子科技大学 | Brain tumor classification method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| PARMIDA MORADI BIRGANI 等: "MRI Segmentation Using Fuzzy C-means Clustering Algorithm Basis Neural Network", 《2008 3RD INTERNATIONAL CONFERENCE ON INFORMATION AND COMMUNICATION TECHNOLOGIES：FROM THEORY TO APPLICATIONS》 * |
| 杨佳玲 等: "基于深度信念网络的肺结节良恶性分类", 《科学技术与工程》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292884B (en) * | 2017-08-07 | 2020-09-29 | 杭州深睿博联科技有限公司 | Method and device for identifying edema and hematoma in MRI (magnetic resonance imaging) image |
| CN107292884A (en) * | 2017-08-07 | 2017-10-24 | 北京深睿博联科技有限责任公司 | The method and device of oedema and hemotoncus in a kind of identification MRI image |
| CN107845098A (en) * | 2017-11-14 | 2018-03-27 | 南京理工大学 | Liver cancer image full-automatic partition method based on random forest and fuzzy clustering |
| CN108629772A (en) * | 2018-05-08 | 2018-10-09 | 上海商汤智能科技有限公司 | Image processing method and device, computer equipment and computer storage media |
| CN108629772B (en) * | 2018-05-08 | 2023-10-03 | 上海商汤智能科技有限公司 | Image processing method and device, computer equipment and computer storage medium |
| CN109102512A (en) * | 2018-08-06 | 2018-12-28 | 西安电子科技大学 | A kind of MRI brain tumor image partition method based on DBN neural network |
| CN109102512B (en) * | 2018-08-06 | 2021-03-09 | 西安电子科技大学 | DBN neural network-based MRI brain tumor image segmentation method |
| CN110706225B (en) * | 2019-10-14 | 2020-09-04 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Tumor identification system based on artificial intelligence |
| CN110706225A (en) * | 2019-10-14 | 2020-01-17 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Tumor identification system based on artificial intelligence |
| CN111445456A (en) * | 2020-03-26 | 2020-07-24 | 北京推想科技有限公司 | Classification model, network model training method and device, and identification method and device |
| US11227387B2 (en) | 2020-05-18 | 2022-01-18 | Prince Mohammad Bin Fahd University | Multi-stage brain tumor image processing method and system |
| CN112686845A (en) * | 2020-12-23 | 2021-04-20 | 合肥联宝信息技术有限公司 | Image processing method and device and computer readable medium |
| CN112686845B (en) * | 2020-12-23 | 2022-04-15 | 合肥联宝信息技术有限公司 | Image processing method and device and computer readable medium |
| CN113240964A (en) * | 2021-05-13 | 2021-08-10 | 广西英腾教育科技股份有限公司 | Cardiopulmonary resuscitation teaching machine |
| CN113240964B (en) * | 2021-05-13 | 2023-03-31 | 广西英腾教育科技股份有限公司 | Cardiopulmonary resuscitation teaching machine |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106780453A (en) | A kind of method realized based on depth trust network to brain tumor segmentation | |
| CN111476292B (en) | Small sample element learning training method for medical image classification processing artificial intelligence | |
| CN108492297B (en) | MRI brain tumor positioning and intratumoral segmentation method based on deep cascade convolution network | |
| CN108364006B (en) | Medical image classification device based on multi-mode deep learning and construction method thereof | |
| Tian et al. | Multi-path convolutional neural network in fundus segmentation of blood vessels | |
| Zhang et al. | Review of breast cancer pathologigcal image processing | |
| CN112101451B (en) | Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block | |
| Zhong et al. | Boosting‐based cascaded convolutional neural networks for the segmentation of CT organs‐at‐risk in nasopharyngeal carcinoma | |
| Dong et al. | Multi-view secondary input collaborative deep learning for lung nodule 3D segmentation | |
| CN108549912A (en) | A kind of medical image pulmonary nodule detection method based on machine learning | |
| CN107330883A (en) | A kind of medical image lesion region positioning and sorting technique | |
| CN111767952B (en) | Interpretable lung nodule benign and malignant classification method | |
| CN110097921B (en) | Visualized quantitative method and system for glioma internal gene heterogeneity based on image omics | |
| CN113344864A (en) | Ultrasonic thyroid nodule benign and malignant prediction method based on deep learning | |
| CN112990344B (en) | Multi-view classification method for pulmonary nodules | |
| Ye et al. | Medical image diagnosis of prostate tumor based on PSP-Net+ VGG16 deep learning network | |
| CN113160120A (en) | Liver blood vessel segmentation method and system based on multi-mode fusion and deep learning | |
| Tang et al. | Detection of pulmonary nodules based on a multiscale feature 3D U-Net convolutional neural network of transfer learning | |
| Anter et al. | Deep federated machine learning-based optimization methods for liver tumor diagnosis: A review | |
| CN109919216B (en) | Counterlearning method for computer-aided diagnosis of prostate cancer | |
| Huang et al. | Breast cancer diagnosis based on hybrid SqueezeNet and improved chef-based optimizer | |
| Tang et al. | NSCGCN: A novel deep GCN model to diagnosis COVID-19 | |
| CN114387282A (en) | Accurate automatic segmentation method and system for medical image organs | |
| Liu et al. | Automated classification of cervical Lymph-Node-Level from ultrasound using depthwise separable convolutional swin transformer | |
| Zhang et al. | Classification of benign and malignant pulmonary nodules based on deep learning |
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: 20170531 |
|
| RJ01 | Rejection of invention patent application after publication |