CN109816667A - A kind of Lung neoplasm dividing method based on CT images - Google Patents
A kind of Lung neoplasm dividing method based on CT images Download PDFInfo
- Publication number
- CN109816667A CN109816667A CN201910052966.2A CN201910052966A CN109816667A CN 109816667 A CN109816667 A CN 109816667A CN 201910052966 A CN201910052966 A CN 201910052966A CN 109816667 A CN109816667 A CN 109816667A
- Authority
- CN
- China
- Prior art keywords
- lung neoplasm
- lung
- dividing method
- segmentation
- type
- 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
Abstract
The present invention provides a kind of Lung neoplasm dividing method based on CT images, belongs to computer-aided medical science field.The present invention uses the method for classifier to carry out automatic discrimination to Lung neoplasm type first, and determination belongs to after certain type of Lung neoplasm, designs corresponding dividing method for the characteristics of difference type Lung neoplasm, realizes and divide to the fast accurate of different type Lung neoplasm.The present invention provides solution the problems such as the automatic accurate segmentation of Lung neoplasm, Lung neoplasm segmentation precision to realize.
Description
Technical field
The present invention relates to computer-aided medical science fields, are related to a kind of Lung neoplasm dividing method based on CT images.
Background technique
The disease incidence of lung cancer is high, the death rate is high, 5 years survival rates are low, is the number one killer of global Cancer death.Based on CT shadow
The computer-aided diagnosis technology of picture is also applied to more and more in the diagnosis of lung cancer.And it is crucial in computer-aided diagnosis
It is that and processing analysis is carried out to Lung neoplasm, first have to be partitioned into complete Lung neoplasm region, extract the image in the region later
Feature is learned, Lung neoplasm is analyzed and processed using the characteristic information extracted, and then achievees the purpose that diagnose lung cancer.
Lung neoplasm can be divided into different types, such as angiosynizesis, lung wall adhesion, frosted glass etc. according to certain standard.And
Different types of Lung neoplasm shows different features in CT images, is difficult to find the dividing method of unified accurate stable to it
It is split, so only using single dividing method in computer-aided diagnosis and being split to Lung neoplasm is can exist
Very big problem.Therefore the mode only combined using a variety of dividing methods, is split for different types of Lung neoplasm,
It accurately could be completely partitioned into Lung neoplasm region, more accurate characteristic information is extracted, obtain more accurate diagnostic result.?
The type for also needing to judge Lung neoplasm before this, can just find out corresponding dividing method.Clinically mainly pass through doctor at present
It is raw to carry out artificial judgment, and we achieve the purpose that Fast Segmentation during segmentation, by the side for establishing classifier
Method carries out a rapidly judgement to Lung neoplasm type, provides the judging result of an auxiliary, can also finally be carried out by doctor
Confirmation modification.Critically important a part is exactly the design of partitioning algorithm in this, and the requirement for partitioning algorithm is exactly quick
Accurate stable.Speed soon will obtain segmentation result fastly first, while have good robustness, can be very steadily defeated
Segmentation result out.
Deficiency is still had for the research of Lung neoplasm dividing method both at home and abroad at present, still has biggish improvement and promotion empty
Between.If existing most research designs single dividing method both for certain type of Lung neoplasm, there are dividing methods
It is unstable, the problems such as segmentation precision is not high；The research for carrying out System Partition simultaneously for polymorphic type Lung neoplasm is seldom, therefore can not
Diagnostic analysis systematically is carried out to lung cancer.
In view of the above-mentioned problems, the present invention devises a new experimental program to carry out accurately different types of Lung neoplasm
Quickly segmentation.
Summary of the invention
The purpose of the present invention is aiming at the shortcomings in the prior art, propose it is a kind of based on CT images systematically to Lung neoplasm into
The method that row is quick and precisely divided, this method can carry out the segmentation of system for different types of Lung neoplasm, improve Lung neoplasm
Segmentation precision and robustness.
To achieve the above object, the technical solution adopted by the present invention the following steps are included:
The initial segmentation of one, Lung neoplasm: will be first to CT during sorter model is established with final Classification and Identification
Lung neoplasm region in image carries out initial segmentation, carries out feature extraction to initial segmentation result, the feature extracted is for dividing
The foundation and Classification and Identification process of class device model.
The foundation of two, Lung neoplasm sorter models: enough various types Lung neoplasm image datas are compiled, to lung knot
It saves region and carries out initial segmentation, extract characteristic, establish sorter model for training.
The Classification and Identification of three, Lung neoplasm types: initial segmentation is carried out using same method to the region to be detected, is extracted
Same feature carries out judgement type with trained disaggregated model before, gives label.
The corresponding dividing method of the different Lung neoplasm types of four, design: it according to the Lung neoplasm type determined before, designs
The corresponding Lung neoplasm partitioning algorithm of each type.
Detailed description of the invention
Fig. 1 is the flow chart of the Lung neoplasm dividing method based on CT images；
Table 2 is the control of Lung neoplasm classification；
Table 3 is segmentation result comparison.
Specific embodiment
It elaborates with reference to the accompanying drawings of the specification to a specific embodiment of the invention.
The specific implementation process is as follows:
The initial segmentation of one, Lung neoplasm: firstly, using the initial segmentation method of threshold value or cluster to the lung in CT images
Knuckle areas carries out primary segmentation, obtains initial segmentation region.This partial content mainly will be applied to multi-categorizer model and build
In vertical and subsequent Classification and Identification process.In view of the object for establishing model and the object for Classification and Identification are necessary
Be it is unified, the process of Classification and Identification entire in this way is just meaningful, can just obtain correct classification recognition result.Classifying more
During device is established, the Lung neoplasm CT images data established for sorter model are partitioned into just using initial segmentation method
Beginning region, this is as the object for establishing model；Equally, it carries out during precisely dividing being also first to precisely being divided
Region carry out initial segmentation, obtain initial segmentation region, as the input of classifier, export the judgement type for Lung neoplasm.
On the basis of Lung neoplasm initial segmentation, be extracted including grey level histogram feature, morphological feature, textural characteristics,
80 kinds of 3-D quantitative features including Laplacian feature and wavelet character describe Lung neoplasm, as shown in table 1.
Wherein, wavelet character described in table 1 is the ash extracted based on 8 high frequency low frequency components on three directions of small echo
Spend histogram feature and textural characteristics.Such as, XLLHIt is that low-pass filtering is carried out along the direction x and y to image X, carries out high pass in the z-direction
Filtering.
The foundation of two, Lung neoplasm sorter models: firstly, Lung neoplasm to be divided into the pulmonary nodule being sticked on blood vessel, is sticked to
Other further types of Lung neoplasms such as pulmonary nodule, frosted glass type on lung wall compile enough above-mentioned several types lungs
Tubercle image data determines the information such as the Lung neoplasm position in every group of data, using initial segmentation method above-mentioned to every
Group data are partitioned into initial Lung neoplasm region, extract the characteristic in initial segmentation region, according to given contingency table shown in table 2
Label.After having handled all data, the characteristic of multiple types Lung neoplasm is obtained, with these characteristics training multi-categorizer mould
Type.
Table 2
| Digital label | Lung neoplasm type |
| 1 | Ground glass |
| 2 | Angiosynizesis |
| 3 | Lung wall adhesion |
| n | More polymorphic type |
The Classification and Identification of three, Lung neoplasm types: using the multi-categorizer model being established above come to being split
Lung neoplasm region carries out classification judgement, determines its type.Firstly, carrying out initial segmentation to the region to be divided, initially divided
Region is cut, the characteristic in initial segmentation region is then extracted, is input in classifier, carries out classification judgement output Lung neoplasm
Type.
The corresponding dividing method of the different Lung neoplasm types of four, design: corresponding to a variety of different types of Lung neoplasms designs
A variety of dividing methods.It is directed to different characteristic of a variety of different type tubercles in CT images, such as isolatism and adhesive type
Will appear as knuckle areas in image, whether there is or not adhesions with other regions, therefore both tubercles will obtain after initial segmentation
It is different as a result, the characteristic information extracted is also discrepant, so first carrying out classification judgement to Lung neoplasm according to this.
Design corresponding dividing method finally, for each type of Lung neoplasm, as adhesive type Lung neoplasm will design can remove it is viscous
The even partitioning algorithm in region.Targetedly dividing method is designed with this, improves the segmentation precision of each type tubercle, while
Just improve whole segmentation precision.
Finally, by inventive algorithm and document [1] " the sequences segmentation method of angiosynizesis type Lung neoplasm image " before, text
Offer [2] " dividing based on rarefaction representation and the ground glass type Lung neoplasm of random walk " and document [3] " A
SegmentationFramework ofPulmonaryNodules in Lung CTImages " the partitioning algorithm result mentioned
It compares, because the index that they mention is different, we calculate the evaluation index that respective document proposes to carry out method pair
Than comparing result is as shown in table 3.The index calculated in document 1 is probability edge index (PRI), it is to examine practical segmentation knot
The parameter of the consistency of attribute symbiosis between fruit and reference result.Assuming that (xi,yi) it is a pixel pair in original image S, if
SaMiddle label is ai,aj), then in SbMiddle label is bi,bj) also should be identical.Shown in calculation formula such as formula (1).
Wherein, M is the number of pixels in original image S；I is discriminant function, and whether main function is to judge pixel to having
There is same tag.PRI value is bigger, illustrates that practical segmentation result and the attribute symbiosis consistency of doctor's manual segmentation result are better,
Show that segmentation result is better.
The index calculated in document 2 and 3 is OVERLAP, it measures the weight between the segmentation result of algorithm and goldstandard
Folded area ratio, shown in calculation formula such as formula (2).
Wherein, SAAnd SGTIt is not algorithm segmentation result and goldstandard.|SA∩SGT| it is both comprising SAIt again include SGTPixel
Number.|SA∪SGT| it indicates to include SAOr SGTOr the number of pixels comprising the two, in the ideal case, a segmentation result is more quasi-
Really, OVERLAP value is closer to 1.
Comparison is as it can be seen that dividing method used in the present invention all increases to the accuracy that different type Lung neoplasm is divided.
Table 1
Table 2
| Digital label | Lung neoplasm type |
| 1 | Ground glass |
| 2 | Angiosynizesis |
| 3 | Lung wall adhesion |
| n | More polymorphic type |
Table 3
Claims (3)
1. a kind of Lung neoplasm dividing method based on CT images, which is characterized in that detailed process is as follows:
The initial segmentation of one, Lung neoplasm；
The foundation of two, Lung neoplasm sorter models；
The Classification and Identification of three, Lung neoplasm types；
The corresponding dividing method of the different Lung neoplasm types of four, design is simultaneously applied.
2. the Lung neoplasm dividing method according to claim 1 based on CT images, it is characterised in that according to Lung neoplasm type
Difference, corresponding dividing method is designed, to improve Lung neoplasm segmentation precision, it is first determined type, according to a unification
Standard classify to Lung neoplasm；Then model is established, enough various types Lung neoplasm image datas are compiled, to lung
Knuckle areas carries out initial segmentation, extracts characteristic, establishes sorter model for training.
3. the Lung neoplasm dividing method according to claim 1 based on CT images, which is characterized in that in cutting procedure first
The judgement of Lung neoplasm type is carried out using disaggregated model: initial segmentation being carried out using same method to the region to be detected, is mentioned
Same feature is taken, judges type with classifier, gives label.Then the region is carried out according to label corresponding dividing method
Segmentation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910052966.2A CN109816667A (en) | 2019-01-21 | 2019-01-21 | A kind of Lung neoplasm dividing method based on CT images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910052966.2A CN109816667A (en) | 2019-01-21 | 2019-01-21 | A kind of Lung neoplasm dividing method based on CT images |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109816667A true CN109816667A (en) | 2019-05-28 |
Family
ID=66604664
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910052966.2A Pending CN109816667A (en) | 2019-01-21 | 2019-01-21 | A kind of Lung neoplasm dividing method based on CT images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109816667A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110910348A (en) * | 2019-10-22 | 2020-03-24 | 上海联影智能医疗科技有限公司 | Method, device, equipment and storage medium for classifying positions of pulmonary nodules |
| US11605164B2 (en) | 2019-10-22 | 2023-03-14 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for lung nodule evaluation |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102324109A (en) * | 2011-09-26 | 2012-01-18 | 上海理工大学 | Method for three-dimensionally segmenting insubstantial pulmonary nodule based on fuzzy membership model |
| CN104751178A (en) * | 2015-03-31 | 2015-07-01 | 上海理工大学 | Pulmonary nodule detection device and method based on shape template matching and combining classifier |
| CN106650830A (en) * | 2017-01-06 | 2017-05-10 | 西北工业大学 | Deep model and shallow model decision fusion-based pulmonary nodule CT image automatic classification method |
| CN107481215A (en) * | 2016-06-07 | 2017-12-15 | 宁夏医科大学 | CT Lung neoplasm dividing methods based on spatial distribution |
| CN108010013A (en) * | 2017-11-03 | 2018-05-08 | 天津大学 | A kind of lung CT image pulmonary nodule detection methods |
| CN108549912A (en) * | 2018-04-19 | 2018-09-18 | 北京工业大学 | A kind of medical image pulmonary nodule detection method based on machine learning |
| CN108897984A (en) * | 2018-05-07 | 2018-11-27 | 上海理工大学 | Based on correlation analysis between CT images group feature and lung cancer gene expression |
| CN109214397A (en) * | 2018-09-14 | 2019-01-15 | 郑州大学 | The dividing method of Lung neoplasm in a kind of lung CT image |
-
2019
- 2019-01-21 CN CN201910052966.2A patent/CN109816667A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102324109A (en) * | 2011-09-26 | 2012-01-18 | 上海理工大学 | Method for three-dimensionally segmenting insubstantial pulmonary nodule based on fuzzy membership model |
| CN104751178A (en) * | 2015-03-31 | 2015-07-01 | 上海理工大学 | Pulmonary nodule detection device and method based on shape template matching and combining classifier |
| CN107481215A (en) * | 2016-06-07 | 2017-12-15 | 宁夏医科大学 | CT Lung neoplasm dividing methods based on spatial distribution |
| CN106650830A (en) * | 2017-01-06 | 2017-05-10 | 西北工业大学 | Deep model and shallow model decision fusion-based pulmonary nodule CT image automatic classification method |
| CN108010013A (en) * | 2017-11-03 | 2018-05-08 | 天津大学 | A kind of lung CT image pulmonary nodule detection methods |
| CN108549912A (en) * | 2018-04-19 | 2018-09-18 | 北京工业大学 | A kind of medical image pulmonary nodule detection method based on machine learning |
| CN108897984A (en) * | 2018-05-07 | 2018-11-27 | 上海理工大学 | Based on correlation analysis between CT images group feature and lung cancer gene expression |
| CN109214397A (en) * | 2018-09-14 | 2019-01-15 | 郑州大学 | The dividing method of Lung neoplasm in a kind of lung CT image |
Non-Patent Citations (4)
| Title |
|---|
| SUDIPTA MUKHOPADHYAY: "A Segmentation Framework of Pulmonary Nodules in Lung CT Images", 《J DIGIT IMAGING》 * |
| 张伟 等: "血管粘连型肺结节图像的序列分割方法", 《计算机工程与设计》 * |
| 李祥霞 等: "基于稀疏表示和随机游走的磨玻璃型肺结节分割", 《自动化学报》 * |
| 龚敬 等: "ＣＴ 图像肺结节计算机辅助检测技术研究进展", 《北京生物医学工程》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110910348A (en) * | 2019-10-22 | 2020-03-24 | 上海联影智能医疗科技有限公司 | Method, device, equipment and storage medium for classifying positions of pulmonary nodules |
| US11605164B2 (en) | 2019-10-22 | 2023-03-14 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for lung nodule evaluation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106780460B (en) | A kind of Lung neoplasm automatic checkout system for chest CT images | |
| CN108052880B (en) | Virtual and real lane line detection method for traffic monitoring scene | |
| CN105740945B (en) | A kind of people counting method based on video analysis | |
| Nguyen et al. | Structure and context in prostatic gland segmentation and classification | |
| US10445619B2 (en) | Automatic glandular and tubule detection in histological grading of breast cancer | |
| Tulsani et al. | Segmentation using morphological watershed transformation for counting blood cells | |
| CN108921083B (en) | Illegal mobile vendor identification method based on deep learning target detection | |
| EP2847738B1 (en) | Method and apparatus for image scoring and analysis | |
| CN110059697A (en) | A kind of Lung neoplasm automatic division method based on deep learning | |
| CN103020639A (en) | Method for automatically identifying and counting white blood cells | |
| Varma et al. | Astute Segmentation and Classification of leucocytes in blood microscopic smear images using titivated K-means clustering and robust SVM techniques | |
| CN101826209B (en) | Canny model-based method for segmenting three-dimensional medical image | |
| CN110120042A (en) | A kind of crop map based on SLIC super-pixel and automatic threshold segmentation is as pest and disease damage method for extracting region | |
| CN108052904B (en) | Method and device for acquiring lane line | |
| CN104217213B (en) | A kind of medical image multistage sorting technique based on symmetric theory | |
| CN110188763B (en) | Image significance detection method based on improved graph model | |
| CN107730499A (en) | A kind of leucocyte classification method based on nu SVMs | |
| Wang et al. | Identifying neutrophils in H&E staining histology tissue images | |
| CN109816667A (en) | A kind of Lung neoplasm dividing method based on CT images | |
| Khotanlou et al. | Segmentation of multiple sclerosis lesions in brain MR images using spatially constrained possibilistic fuzzy C-means classification | |
| Cheng et al. | Image segmentation technology and its application in digital image processing | |
| CN108664969A (en) | Landmark identification method based on condition random field | |
| CN110866547A (en) | Automatic classification system and method for traditional Chinese medicine decoction pieces based on multiple features and random forest | |
| CN109543498A (en) | A kind of method for detecting lane lines based on multitask network | |
| CN102298780B (en) | Method for detecting shadow of color image |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190528 |