CN114299018A - CT image blood vessel segmentation method based on centerline topological communication and multi-view information - Google Patents
CT image blood vessel segmentation method based on centerline topological communication and multi-view information Download PDFInfo
- Publication number
- CN114299018A CN114299018A CN202111633527.4A CN202111633527A CN114299018A CN 114299018 A CN114299018 A CN 114299018A CN 202111633527 A CN202111633527 A CN 202111633527A CN 114299018 A CN114299018 A CN 114299018A
- Authority
- CN
- China
- Prior art keywords
- blood vessel
- image
- segmentation
- result
- centerline
- 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
- 210000004204 blood vessel Anatomy 0.000 title claims abstract description 41
- 230000011218 segmentation Effects 0.000 title claims abstract description 37
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000004891 communication Methods 0.000 title claims abstract description 6
- 210000001367 artery Anatomy 0.000 claims abstract description 22
- 210000003462 vein Anatomy 0.000 claims abstract description 21
- 230000000877 morphologic effect Effects 0.000 claims abstract description 16
- 210000004072 lung Anatomy 0.000 claims abstract description 12
- 239000002245 particle Substances 0.000 claims abstract description 10
- 230000010339 dilation Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000002685 pulmonary effect Effects 0.000 claims description 6
- 238000000926 separation method Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000012986 modification Methods 0.000 claims description 3
- 230000004048 modification Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 2
- 230000000007 visual effect Effects 0.000 claims description 2
- 230000002792 vascular Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 210000001147 pulmonary artery Anatomy 0.000 description 1
- 210000003492 pulmonary vein Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Abstract
The invention relates to a CT image blood vessel segmentation method based on centerline topological communication and multi-view information, which comprises the following steps: step S1: acquiring an original CT image and preprocessing the original CT image; step S2, dividing the preprocessed CT image into a plurality of image blocks of the region of interest containing artery and vein vessel labels; s3, sending the image block into a multi-scale deep supervision network based on U-Net for segmentation, and obtaining a preliminary result from the arteriovenous segmentation result output by the network through a multi-view voting strategy; and step S5, obtaining derived blood vessel particles by adopting morphological dilation, and correcting the artery and the vein by utilizing the lung blood vessel morphological information learned by the lung blood vessel network model to obtain a final segmentation result. The method has high accuracy and strong generalization capability, and can obtain accurate arteriovenous segmentation results in different types of CT images.
Description
Technical Field
The invention relates to the field of image processing, in particular to a CT image blood vessel segmentation method based on centerline topological communication and multi-view information.
Background
In recent years, research on arteriovenous segmentation of lung CT images mainly focuses on the field of supervised learning methods. With the development of deep learning methods, in arteriovenous feature analysis, complex feature hierarchies can be automatically learned directly from input data, and a plurality of deep learning models are applied to pulmonary artery and vein segmentation, such as CNN, FCN, GCN, U-Net and the like. However, the current methods still have the following limitations: (1) the method is easily affected by two kinds of arteriovenous fractures, one is the fracture of the artery or vein caused by insufficient connectivity, and the other is the fracture of the artery or vein caused by misclassification, so that the blood vessel segmentation effect is poor; (2) arteries and veins have many structures that are interlaced with each other, resulting in adhesions in the segmentation result.
Disclosure of Invention
In view of the above, the present invention provides a method for segmenting a blood vessel of a CT image based on centerline topological connection and multi-view information,
in order to achieve the purpose, the invention adopts the following technical scheme:
a CT image vessel segmentation method based on centerline topological connection and multi-view information comprises the following steps:
step S1: acquiring an original CT image and preprocessing the original CT image;
step S2, dividing the preprocessed CT image into a plurality of image blocks of the region of interest containing the blood vessel labels;
step S3, sending the image block into a multi-scale deep supervision network based on U-Net for segmentation, and obtaining a preliminary result by a multi-view voting strategy for the arteriovenous segmentation result output by the network;
step S4, correcting the wrong part in the preliminary result by using the blood vessel topological connection information learned from the center line model;
and step S5, obtaining derived blood vessel particles by adopting morphological dilation, and correcting the artery and the vein by utilizing the lung blood vessel morphological information learned by the lung blood vessel network model to obtain a final segmentation result.
Further, the pretreatment specifically comprises: and normalizing each pixel value to be between 0 and 1, and taking a block by adopting a sliding window with a preset size.
Further, the step S2 is specifically: the CT image is divided into image patches of the region of interest containing arteriovenous vessel labels from three views, including the sagittal, coronal, and cross-sectional planes.
Further, the multi-scale deep supervision network based on U-Net specifically includes: using 4 coding layers like U-Net and 4 decoding layers as a backbone, the encoder and decoder are connected to a multi-scale feature fusion block (MSFF) in each basic block, which consists essentially of two 3 × 3 convolutional layers, one BN layer and 2 × 2 pooling layers with a step number of 2; after each convolution operation is carried out, activation is carried out through a Relu function, and then an encoder and a decoder are connected through a jump connection; in the up-sampling stage, information between final output and side output is gradually gathered by utilizing deep supervision, and relatively effective characteristics in a hidden layer are obtained; finally, the output of the network is classified by a sigmiod function.
Further, the multi-view policy specifically includes: the blocks with N visual angles are tested through the N networks, and the obtained results are subjected to a voting strategy to obtain a fused result; and the voting strategy is to convert the results of the N networks into N channels by one _ hot coding, and finally, the index corresponding to the maximum value of the element is taken out through argmax and is output and stored.
Further, the derived vascular particle is obtained by using morphological dilation, and the modified expression is as follows:
C(La,Lv)+A(La,Lv)＝Lcor (1)
wherein C (L)a,Lv) Is a preliminary result of centerline segmentation, A (L)a,Lv) Is the preliminary result of arteriovenous separation, Lcor(La,Lv) For the corrected result, LaIs an artery, LvIs a vein.
Furthermore, the pulmonary vascular morphology information segmented by the pulmonary vascular network model is used for modifying the artery and the vein to obtain a segmentation result, and the modification expression is as follows:
Lcor(La)∩P(Lav)＝Rrep(La) (2)
Lcor(Lv)∩P(Lav)＝Rrep(Lv) (3)
Rrep(La)+Rrep(Lv)＝Rrep(La,Lv) (4)
in the above formula, Lcor(La) And are each Lcor(Lv) Using the centerline corrected arterial and venous results, P (L)av) Is the result of the separation of the lung whole blood vessel, Rrep(La,Lv) And (5) modifying the form information.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention uses a multi-view strategy, improves the accuracy of the artery and the vein, increases the continuity and the integrity between the blood vessels, and improves the sensitivity of the blood vessel segmentation;
2. according to the invention, the accuracy and the connectivity of the artery and vein are further improved through a center line correction algorithm, the influence of pathological changes on the segmentation of the blood vessel is reduced, and the problem of adhesion of the artery and the vein is solved through distance transformation.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a diagram of a multi-scale deep supervision network structure based on U-Net according to an embodiment of the present invention;
FIG. 3 illustrates a multi-view strategy in an embodiment of the invention;
FIG. 4 is a centerline strategy in an embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
Referring to fig. 1, the present invention provides a method for segmenting a blood vessel of a CT image based on centerline topological connection and multi-view information, comprising the following steps:
step S1: acquiring an original CT image, carrying out normalization operation on each pixel value to be between [0 and 1], and adopting a sliding window with the size of 64x64x32 to take blocks;
step S2, dividing the CT image into image blocks of the region of interest containing arteriovenous vessel labels from three views (sagittal plane, coronal plane and cross section);
step S3, sending the image block into a multi-scale deep supervision network based on U-Net for segmentation, and obtaining a preliminary result by a multi-view voting strategy for the arteriovenous segmentation result output by the network;
referring to fig. 2, in this embodiment, the multi-scale deep supervision network based on U-Net specifically includes: the overall network utilizes 4 coding layers like U-Net and 4 decoding layers as the backbone. The encoder and decoder are connected to a multi-scale feature fusion block (MSFF) in each basic block, which is mainly composed of two 3 × 3 convolutional layers, one BN layer, and 2 × 2 pooled layers with a step number of 2. After each convolution operation is performed, activation is performed by the Relu function, and then the encoder and decoder are connected by a skip connection. In the up-sampling stage, information between final output and side output is gradually gathered by using deep supervision, and relatively effective characteristics in a hidden layer are obtained. Finally, the output of the net is sorted by a sigmiod function of size 1x1x 1.
In this embodiment, referring to fig. 3, the multi-view policy is to test the three view slices through the three proposed networks, and pass the obtained results through a voting policy to obtain a fused result. And the voting strategy is to convert the results of the three networks into three channels by one _ hot coding, and finally, the index corresponding to the maximum value of the element is taken out through argmax and is output and stored.
And step S4, correcting the preliminary result of arteriovenous segmentation by adopting the segmented central line result, solving the problems of arteriovenous segmentation error and arteriovenous segmentation, wherein the integral arteriovenous blood vessel is slightly larger than a whole blood vessel after morphological expansion, so the original blood vessel shape can be recovered by keeping the intersection of the whole blood vessel and the arteriovenous, and the central line strategy is as shown in figure 4.
And step S5, obtaining derived blood vessel particles by adopting morphological dilation, and correcting the artery and the vein by utilizing the lung blood vessel morphological information learned by the lung blood vessel network model to obtain a final segmentation result.
In this embodiment, when the network learns the centerline features, the network pays attention to the skeleton and topology connectivity information of the vessel tree, and the arteriovenous segmentation network pays attention to the accuracy and detail information of each vessel particle. As the central line is more attention to the skeleton and the connectivity thereof, better effect can be obtained in the aspect of the connectivity of the artery and the vein, and the correction expression is as follows:
C(La,Lv)+A(La,Lv)＝Lcor (1)
wherein C (L)a,Lv) Is a preliminary result of centerline segmentation, A (L)a,Lv) Is the preliminary result of arteriovenous separation, Lcor(La,Lv) For the corrected result, LaIs an artery and has a value of 1, LvIs venous and has a value of 2. 1) We modify LvIs 3, the results are added. 2) The result of a value of 2 is modified to 1 and the result of a value of 6 is modified to 2.
Because the topology of the vascular particles on the center line is repaired, a large amount of the vascular particles which should exist cannot be supplemented. Therefore, the derived vascular particles are obtained by morphological expansion treatment on the arteriovenous separately and are repaired. However, the morphological dilation can cause the loss of the original artery and vein morphological information, so the pulmonary vascular morphological information segmented by the pulmonary vascular network model is used for correcting the artery and vein to obtain a segmentation result, and the correction expression is as follows:
Lcor(La)∩P(Lav)＝Rrep(La) (2)
Lcor(Lv)∩P(Lav)＝Rrep(Lv) (3)
Rrep(La)+Rrep(Lv)＝Rrep(La,Lv) (4)
in the above formula, Lcor(La) And are each Lcor(Lv) Using the centerline corrected arterial and venous results, P (L)av) Is the result of the separation of the lung whole blood vessel, Rrep(La,Lv) And (5) modifying the form information. And finally, solving the adhesion problem through distance conversion post-processing.
The distance transformation is to calculate the distance between the pixel point with the label value of 3 and the artery and the vein by the distance transformation, wherein the distance between the pixel point with the label value of 3 and the artery is converted into 1 when the pixel point is close to the artery, and the distance between the pixel point with the label value of 3 and the vein is converted into 2 when the pixel point is close to the vein.
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (7)
1. A CT image blood vessel segmentation method based on centerline topological connection and multi-view information is characterized by comprising the following steps:
step S1: acquiring an original CT image and preprocessing the original CT image;
step S2, dividing the preprocessed CT image into a plurality of image blocks of the region of interest containing the blood vessel labels;
step S3, sending the image block into a multi-scale deep supervision network based on U-Net for segmentation, and obtaining a preliminary result by a multi-view voting strategy for the arteriovenous segmentation result output by the network;
step S4, correcting the wrong part in the preliminary result by using the blood vessel topological connection information learned from the center line model;
and step S5, obtaining derived blood vessel particles by adopting morphological dilation, and correcting the artery and the vein by utilizing the lung blood vessel morphological information learned by the lung blood vessel network model to obtain a final segmentation result.
2. The CT image vessel segmentation method based on centerline topological connection and multi-view information as claimed in claim 1, wherein the preprocessing specifically comprises: and normalizing each pixel value to be between 0 and 1, and taking a block by adopting a sliding window with a preset size.
3. The method for segmenting the blood vessel of the CT image based on the topological connection of the central line and the multi-view information according to claim 1, wherein the step S2 is specifically as follows: the CT image is divided into image patches of the region of interest containing arteriovenous vessel labels from three views, including the sagittal, coronal, and cross-sectional planes.
4. The CT image vessel segmentation method based on centerline topological connection and multi-view information according to claim 1, wherein the U-Net based multi-scale deep supervision network specifically comprises: using 4 coding layers like U-Net and 4 decoding layers as a backbone, the encoder and decoder are connected to a multi-scale feature fusion block (MSFF) in each basic block, which consists essentially of two 3 × 3 convolutional layers, one BN layer and 2 × 2 pooling layers with a step number of 2; after each convolution operation is carried out, activation is carried out through a Relu function, and then an encoder and a decoder are connected through a jump connection; in the up-sampling stage, information between final output and side output is gradually gathered by utilizing deep supervision, and relatively effective characteristics in a hidden layer are obtained; finally, the output of the network is classified by a sigmiod function.
5. The method for segmenting the blood vessel of the CT image based on the topological communication of the central line and the multi-view information according to claim 1, wherein the multi-view strategy specifically comprises: the blocks with N visual angles are tested through the N networks, and the obtained results are subjected to a voting strategy to obtain a fused result; and the voting strategy is to convert the results of the N networks into N channels by one _ hot coding, and finally, the index corresponding to the maximum value of the element is taken out through argmax and is output and stored.
6. The method for segmenting the blood vessel of the CT image based on the topological connection of the central line and the multi-view information according to claim 1, wherein the derived blood vessel particles are obtained by using morphological dilation, and the modified expression is as follows:
C(La,Lv)+A(La,Lv)＝Lcor (1)
wherein C (L)a,Lv) Is a preliminary result of centerline segmentation, A (L)a,Lv) Is the preliminary result of arteriovenous separation, Lcor(La,Lv) For the corrected result, LaIs an artery, LvIs a vein.
7. The method for segmenting the blood vessel of the CT image based on the topological communication of the central line and the multi-view information as claimed in claim 5, wherein the pulmonary blood vessel morphological information segmented by the pulmonary blood vessel network model is used for modifying the artery and the vein to obtain the segmentation result, and the modification expression is as follows:
Lcor(La)∩P(Lav)＝Rrep(La) (2)
Lcor(Lv)∩P(Lav)＝Rrep(Lv) (3)
Rrep(La)+Rrep(Lv)＝Rrep(La,Lv) (4)
in the above formula, Lcor(La) And are each Lcor(Lv) Using the centerline corrected arterial and venous results, P (L)av) Is the result of the separation of the lung whole blood vessel, Rrep(La,Lv) And (5) modifying the form information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111633527.4A CN114299018A (en) | 2021-12-29 | 2021-12-29 | CT image blood vessel segmentation method based on centerline topological communication and multi-view information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111633527.4A CN114299018A (en) | 2021-12-29 | 2021-12-29 | CT image blood vessel segmentation method based on centerline topological communication and multi-view information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114299018A true CN114299018A (en) | 2022-04-08 |
Family
ID=80971825
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111633527.4A Pending CN114299018A (en) | 2021-12-29 | 2021-12-29 | CT image blood vessel segmentation method based on centerline topological communication and multi-view information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114299018A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091573A (en) * | 2019-12-20 | 2020-05-01 | 广州柏视医疗科技有限公司 | CT image pulmonary vessel segmentation method and system based on deep learning |
| CN112716446A (en) * | 2020-12-28 | 2021-04-30 | 深圳硅基智能科技有限公司 | Method and system for measuring pathological change characteristics of hypertensive retinopathy |
| CN112733953A (en) * | 2021-01-19 | 2021-04-30 | 福州大学 | Lung CT image arteriovenous vessel separation method based on Non-local CNN-GCN and topological subgraph |
| WO2021104056A1 (en) * | 2019-11-27 | 2021-06-03 | 中国科学院深圳先进技术研究院 | Automatic tumor segmentation system and method, and electronic device |
| US20210390338A1 (en) * | 2020-06-15 | 2021-12-16 | Dalian University Of Technology | Deep network lung texture recogniton method combined with multi-scale attention |
-
2021
- 2021-12-29 CN CN202111633527.4A patent/CN114299018A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021104056A1 (en) * | 2019-11-27 | 2021-06-03 | 中国科学院深圳先进技术研究院 | Automatic tumor segmentation system and method, and electronic device |
| CN111091573A (en) * | 2019-12-20 | 2020-05-01 | 广州柏视医疗科技有限公司 | CT image pulmonary vessel segmentation method and system based on deep learning |
| US20210390338A1 (en) * | 2020-06-15 | 2021-12-16 | Dalian University Of Technology | Deep network lung texture recogniton method combined with multi-scale attention |
| CN112716446A (en) * | 2020-12-28 | 2021-04-30 | 深圳硅基智能科技有限公司 | Method and system for measuring pathological change characteristics of hypertensive retinopathy |
| CN112733953A (en) * | 2021-01-19 | 2021-04-30 | 福州大学 | Lung CT image arteriovenous vessel separation method based on Non-local CNN-GCN and topological subgraph |
Non-Patent Citations (1)
| Title |
|---|
| 潘林: "" Learning multi-view and centerline topology connectivity information for pulmonary artery-vein separation"", 《COMPUTERS IN BIOLOGY AND MEDICINE》, 12 March 2023 (2023-03-12) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020104006A4 (en) | Radar target recognition method based on feature pyramid lightweight convolutional neural network | |
| CN109886273B (en) | CMR image segmentation and classification system | |
| CN111914644B (en) | Dual-mode cooperation based weak supervision time sequence action positioning method and system | |
| CN109816661B (en) | Tooth CT image segmentation method based on deep learning | |
| CN111127482B (en) | CT image lung and trachea segmentation method and system based on deep learning | |
| CN111091573B (en) | CT image pulmonary vessel segmentation method and system based on deep learning | |
| CN110298844B (en) | X-ray radiography image blood vessel segmentation and identification method and device | |
| CN110992351B (en) | sMRI image classification method and device based on multi-input convolution neural network | |
| CN111110228B (en) | Electrocardiosignal R wave detection method and device | |
| CN108073918B (en) | Method for extracting blood vessel arteriovenous cross compression characteristics of fundus retina | |
| CN112001928B (en) | Retina blood vessel segmentation method and system | |
| CN110942466B (en) | Cerebral artery segmentation method and device based on deep learning technology | |
| CN111932554A (en) | Pulmonary blood vessel segmentation method, device and storage medium | |
| CN112884788B (en) | Cup optic disk segmentation method and imaging method based on rich context network | |
| CN110751636A (en) | Fundus image retinal arteriosclerosis detection method based on improved coding and decoding network | |
| CN112651969B (en) | Trachea tree hierarchical extraction method combining multi-information fusion network and regional growth | |
| CN111161287A (en) | Retinal vessel segmentation method based on symmetric bidirectional cascade network deep learning | |
| CN111340094A (en) | Capsule endoscope image auxiliary classification system and classification method based on deep learning | |
| CN110584654A (en) | Multi-mode convolutional neural network-based electrocardiosignal classification method | |
| CN111956208A (en) | ECG signal classification method based on ultra-lightweight convolutional neural network | |
| CN115640507A (en) | Abnormal data screening method based on electrocardio-heart sound joint analysis | |
| CN112733953B (en) | Lung CT image arteriovenous vessel separation method based on Non-local CNN-GCN and topological subgraph | |
| CN112330662B (en) | Medical image segmentation system and method based on multi-level neural network | |
| CN111899272B (en) | Fundus image blood vessel segmentation method based on coupling neural network and line connector | |
| CN114299018A (en) | CT image blood vessel segmentation method based on centerline topological communication and multi-view information |
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 |