A CT image-based blood vessel segmentation method, a segmentation system, a computing device and a storage medium

Abstract

The application discloses a blood vessel segmentation method and system based on CT images, a computing device and a storage medium, and comprises the following steps: marking the blood vessels to be segmented on a CT image; centering on each type of marked blood vessel, cutting a blood vessel marked image of the same size and containing the blood vessel, and cutting a CT blood vessel image corresponding to the CT image; performing the same data enhancement on the blood vessel marked image and the CT blood vessel image to make multiple groups of image sets; wherein the data enhancement comprises one of translation, rotation and brightness enhancement; inputting the image sets into a deep learning network for training to obtain a blood vessel segmentation model; cutting the CT image to be segmented to the same size and inputting the CT image to the blood vessel segmentation model to identify the blood vessel class, and putting the prediction result back into the CT image to be segmented to complete the blood vessel segmentation. The application can complete all segmentation at one time for the segmentation task of multiple blood vessels in any human body part, is high in efficiency and short in time consumption.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a blood vessel segmentation method based on CT images, a segmentation system, a computing device and a storage medium. BACKGROUND

[0002] In radiotherapy, it is necessary to accurately delineate the organs at risk around the lesion in medical images. In order to improve the work efficiency of doctors and ensure the stability of the delineation results of organs at risk, reliable automatic delineation technology of organs at risk is needed.

[0003] In the human thoracic cavity, the blood vessels are mostly small, long and tortuous arteriovenous vessels. It is difficult to accurately determine the boundary between different blood vessels connected to each other. In addition, in a horizontal plane image, the area ratio of the blood vessels is small, and there is a great data imbalance problem for the segmentation and positioning of each type of blood vessel.

[0004] At present, there are several methods for segmenting blood vessels in CT images using deep learning;

[0005] First, different types of blood vessels are segmented individually. Although 2D network segmentation is relatively accurate, for multi-class blood vessel segmentation, data processing and network training need to be done for each class, which increases a lot of redundant work, and for different types of blood vessels connected to each other, it is difficult to effectively determine the boundary between them, which can easily lead to false positives or errors.

[0006] Second, multiple blood vessels are considered as the same class for 3D network model segmentation, and then according to the structure of the human body, the different types of blood vessels are subdivided. It is difficult to separate the overall segmentation result according to the structural characteristics of the blood vessels, and forced boundary separation based on rules can easily result in a large number of misjudgments at the junction of the blood vessels.

[0007] Third, multiple blood vessels are segmented in multiple categories. Due to the narrow and long characteristics of the blood vessels, there is a serious data imbalance problem, and the segmentation network model is difficult to converge.

[0008] As can be seen from the above, in the process of simultaneously segmenting multiple types of blood vessels, the existing deep learning blood vessel segmentation method has many difficulties. SUMMARY

[0009] In view of the above problems in the prior art, the present application provides a blood vessel segmentation method based on CT images, a segmentation system, a computing device and a storage medium, which realizes classification and segmentation of multiple types of blood vessels based on 2D multi-classification network deep learning.

[0010] The first object of the present application is to provide a blood vessel segmentation method based on CT images, comprising:

[0011] Collecting a CT image of a preset region, marking a blood vessel to be segmented on the CT image;

[0012] Centering on each type of marked blood vessel, cutting a blood vessel marked image of the same size and containing the blood vessel, and cutting a CT blood vessel image on the CT image with the same cutting frame;

[0013] Performing the same data enhancement on the blood vessel marked image and the CT blood vessel image to make a plurality of image sets; wherein the data enhancement includes one of translation, rotation and brightness enhancement;

[0014] Constructing a deep learning network;

[0015] Inputting the image set into the deep learning network for training to obtain a blood vessel segmentation model;

[0016] Cutting the CT image to be segmented to the same size and inputting into the blood vessel segmentation model to identify the blood vessel class, and putting the prediction result back into the CT image to be segmented to complete the blood vessel segmentation.

[0017] As a further improvement of the application, the body part is divided into three regions, namely head and neck to heart, heart to pelvic cavity and below the pelvic cavity;

[0018] The blood vessel marking method is:

[0019] Taking n-class blood vessel segmentation as an example: the background region is marked as 0, and different classes of blood vessels are marked as 1-n.

[0020] As a further improvement of the application, taking the size of the blood vessel marked image and the CT blood vessel image as C*R as an example:

[0021] The translation is: the CT blood vessel image and the CT blood vessel image are randomly translated by a pixels along the x-axis direction, -C / 2≤a≤C / 2; and randomly translated by b pixels along the y-axis direction, -R / 2≤b≤R / 2; after translation, cutting is performed to obtain another set of CT blood vessel images and CT blood vessel images;

[0022] The rotation is: rotating the CT blood vessel image and the CT blood vessel image by an angle θ around the center of the image, -90°≤θ≤90°; after rotation, cutting is performed to obtain another set of CT blood vessel images and CT blood vessel images;

[0023] The brightness enhancement is: histogram equalization is performed on the CT blood vessel image and the CT blood vessel image to enhance the brightness.

[0024] As a further improvement of the application, the deep learning network is a 2D multi-class deep learning segmentation network;

[0025] The 2D multi-class deep learning segmentation network takes the average of the loss values of each type of blood vessel except the background as the final loss value.

[0026] A second object of the present application is to provide a CT image-based blood vessel segmentation system, comprising:

[0027] A marking module is configured to collect CT images of a preset region, and mark blood vessels to be segmented on the CT images.

[0028] A cropping module is configured to crop a blood vessel marking image of the same size and containing each type of marked blood vessel as the center, and crop a CT blood vessel image on the CT image with the same cropping frame.

[0029] A data enhancement module is configured to perform the same data enhancement on the blood vessel marking image and the CT blood vessel image to produce multiple sets of images.

[0030] A model construction module is configured to construct a deep learning network.

[0031] A model training module is configured to input the image set into the deep learning network for training to obtain a blood vessel segmentation model.

[0032] A segmentation module is configured to crop the CT image to be segmented to the same size, input the CT image to be segmented into the blood vessel segmentation model, identify the blood vessel class, and put the prediction result back into the CT image to be segmented to complete the blood vessel segmentation.

[0033] As a further improvement of the present application, in the marking module:

[0034] The body part is divided into three regions, namely the head and neck to the heart, the heart to the pelvic cavity, and below the pelvic cavity.

[0035] The blood vessel marking method is as follows: taking n-class blood vessel segmentation as an example: the background region is marked as 0, and different types of blood vessels are marked as 1-n.

[0036] As a further improvement of the present application, in the data enhancement module:

[0037] Taking the size of the blood vessel marking image and the CT blood vessel image as an example:

[0038] The translation is: randomly translate the CT blood vessel image and the CT blood vessel image along the x-axis direction by a pixels, -C / 2≤a≤C / 2; randomly translate the CT blood vessel image and the CT blood vessel image along the y-axis direction by b pixels, -R / 2≤b≤R / 2; after translation, do cropping to obtain another set of CT blood vessel image and CT blood vessel image.

[0039] The rotation is that the CT blood vessel image and the CT blood vessel image are rotated around the image center by an angle theta, -90 DEG <= theta <= 90 DEG, and after rotation, cutting is performed to obtain another group of CT blood vessel images and CT blood vessel images.

[0040] The brightness enhancement is that the CT blood vessel image and the CT blood vessel image are subjected to histogram equalization to enhance brightness.

[0041] As a further improvement of the application, the deep learning network is a 2D multi-class deep learning segmentation network.

[0042] The 2D multi-class deep learning segmentation network takes the average of the loss values of each class of blood vessels except the background as the final loss value.

[0043] A third object of the application is to provide a computing device comprising a memory, a processor and computer instructions stored on the memory and executable on the processor, the processor implementing the steps of the above-mentioned blood vessel segmentation method when executing the instructions.

[0044] A fourth object of the application is to provide a storage medium storing computer instructions, the computer instructions implementing the steps of the above-mentioned blood vessel segmentation method when executed by a processor.

[0045] Compared with the prior art, the application has the following advantages:

[0046] 1. The application can complete all segmentation at one time for the segmentation task of multiple blood vessels in any human body part, with high efficiency and short time consumption.

[0047] 2. The application adopts a 2D multi-class deep learning segmentation network based on multi-class loss average, which has high segmentation result accuracy and strong robustness.

[0048] 3. The application adopts a multi-class blood vessel joint segmentation method, which can promote the segmentation learning among different classes of blood vessels, and is helpful to accurately and effectively separate the boundaries of different classes of blood vessels connected to each other.

[0049] 4. The application can be further applied to the segmentation of other multiple classes of organs in CT images. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A flowchart of the blood vessel segmentation method based on CT images disclosed by an embodiment of the application is shown in the figure;

[0051] Figure 2 A 5-class blood vessel stereoscopic structure diagram disclosed by an embodiment of the application is shown in the figure;

[0052] Figure 3A test image cropping schematic diagram disclosed for an embodiment of the present application;

[0053] Figure 4 A 5-class blood vessel segmentation result diagram disclosed for an embodiment of the present application; wherein a is a left / right internal jugular vein segmentation result, b is a left / right brachiocephalic vein segmentation result, and c is a superior vena cava segmentation result;

[0054] Figure 5 A framework diagram of a blood vessel segmentation system based on CT images disclosed for an embodiment of the present application. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0056] The present application will be further described in detail below in connection with the drawings:

[0057] As shown in Figure 1 , the present application provides a blood vessel segmentation method based on CT images, comprising:

[0058] Step 1, coarse positioning of body regions: divide the body into three regions, i.e., from below the head and neck to near the heart, from below the heart to the pelvic cavity, and from below the pelvic cavity, and then classify and segment the blood vessels in each region.

[0059] Step 2, collect CT images of the preset regions, and mark the blood vessels to be segmented on the CT images; wherein,

[0060] Taking n-class blood vessel segmentation as an example: mark the background region as 0, and mark different classes of blood vessels as 1-n.

[0061] Step 3, take each class of marked blood vessels as the center, crop a blood vessel marked image of the same size containing the blood vessel, and the size of the blood vessel marked image is C*R; and crop a CT blood vessel image on the CT image with the same cropping frame, and it is necessary to ensure that the cropping position does not exceed the boundary of the original image during the cropping process.

[0062] Step 4, perform the same data augmentation on the blood vessel marked image and the CT blood vessel image to produce multiple groups of image sets; wherein the data augmentation includes one or more of translation, rotation and brightness enhancement;

[0063] Translation mode: take the center of mass of each blood vessel labeled image of each type of blood vessel as the center, make a C*R rectangular frame, randomly translate the rectangular frame along the x-axis direction by a pixels (-C / 2≤a≤C / 2), and randomly translate the rectangular frame along the y-axis direction by b pixels (-R / 2≤b≤R / 2), and then crop according to the rectangular frame coordinates; the corresponding CT image is operated in the same way. Each label and CT image is randomly translated, and the cropped position is required to be ensured not to exceed the original image boundary.

[0064] Rotation mode: take each blood vessel labeled image as the center, and make a random rotation angle θ (-90°≤θ≤90°), and then take each blood vessel labeled image of each type of blood vessel as the center, and make a cropping operation.

[0065] Brightness enhancement mode: histogram equalization is performed on the CT original image to enhance the brightness and highlight the blood vessel information, and then the translation and rotation enhancement of the above are repeated on the brightness enhanced image.

[0066] The enhancement multiple is n times (n∈R), and when the proportion difference between the types of blood vessels is large, more enhancement can be performed on the type with small proportion to reduce the imbalance of the data.

[0067] Step 5, constructing a deep learning network; wherein,

[0068] The deep learning network uses a 2D multi-class deep learning segmentation network, and in any 2D multi-class segmentation network, the dice loss is used as the loss function, and in order to reduce the data imbalance between the target organ and the background, the average of the loss values of each type of blood vessel except the background is used as the final Loss value in the network training.

[0069] Step 6, inputting the image set into the deep learning network for training to obtain a blood vessel segmentation model;

[0070] Step 7, performing the same size cropping on the CT image to be segmented, and inputting the CT image to be segmented into the blood vessel segmentation model, identifying the blood vessel type, and putting the prediction result back into the CT image to be segmented, to complete the blood vessel segmentation.

[0071] As shown in Figure 5 The present application provides a blood vessel segmentation system based on CT image, comprising:

[0072] The positioning module is used for rough positioning of the body region: the body is divided into three regions, which are below the head and neck to the vicinity of the heart, below the heart to the pelvic cavity, and below the pelvic cavity, and then the blood vessels in each region are classified and segmented.

[0073] The marking module is used for collecting the CT image of the preset region, and marking the blood vessels to be segmented on the CT image; wherein, taking n-class blood vessel segmentation as an example: the background region is marked as 0, and different types of blood vessels are marked as 1-n.

[0074] a cropping module, configured to crop a same size of a blood vessel marked image centered on each type of marked blood vessel, the size of the blood vessel marked image being C*R, and crop a CT blood vessel image on the CT image with the same cropping frame, and ensure that the cropping position does not exceed the original image boundary during the cropping.

[0075] a data enhancement module, configured to perform the same data enhancement on the blood vessel marked image and the CT blood vessel image to make a plurality of image sets; wherein the data enhancement includes one or more of translation, rotation and brightness enhancement.

[0076] a translation mode: taking the center of the centroid of each blood vessel marked image of each type of blood vessel as the center, a C*R rectangular frame is made, the rectangular frame is randomly translated by a pixels along the x-axis direction (-C / 2≤a≤C / 2), and b pixels along the y-axis direction (-R / 2≤b≤R / 2), and then cropped according to the rectangular frame coordinates; the same operation is performed on the CT image. Each label and CT image is randomly translated, and it is required to ensure that the cropping position does not exceed the original image boundary.

[0077] a rotation mode: taking each blood vessel marked image as the center, a random rotation angle θ (-90°≤θ≤90°) is made, and then taking each blood vessel marked image of each type of blood vessel as the center, a cropping operation is performed.

[0078] a brightness enhancement mode: histogram equalization is performed on the CT original image to enhance brightness and highlight blood vessel information, and then the translation and rotation enhancement is repeated on the brightness enhanced image.

[0079] the enhancement multiple is n times (n∈R), and when the proportion difference between the categories is large, more enhancement can be performed on the category with small proportion to reduce the imbalance of the data.

[0080] a model construction module, configured to construct a deep learning network; wherein,

[0081] the deep learning network is first used in a 2D multi-class deep learning segmentation network, and in any 2D multi-class segmentation network, the dice loss is used as the loss function, and to reduce the data imbalance between the target organ and the background, the average of the loss values of each type of blood vessel except the background is used as the final Loss value in the network training.

[0082] a model training module, configured to input the image set into the deep learning network for training to obtain a blood vessel segmentation model.

[0083] a segmentation module, configured to crop a to-be-segmented CT image to the same size, input the to-be-segmented CT image into the blood vessel segmentation model, identify the blood vessel category, and put the prediction result back into the to-be-segmented CT image to complete the blood vessel segmentation.

[0084] Embodiment:

[0085] The application provides a blood vessel segmentation method and system based on CT images, comprising:

[0086] Step 1. Taking 5 types of venous vessels between the head and neck and the heart as an example, including the left internal jugular vein, the right internal jugular vein, the left brachiocephalic vein, the right brachiocephalic vein and the superior vena cava, the structural diagram is as shown in Figure 2 .

[0087] Step 2. Labeling the 5 types of blood vessels to be segmented on the CT image; wherein the background (except the area outside the blood vessels) is marked as 0, and the 5 types of blood vessels are marked as 1-5 in turn;

[0088] Step 3. Taking each type of labeled blood vessel as the center, cutting a blood vessel label image of the same size containing the blood vessel, and cutting the image with a size of 128*200; at the same time, cutting the original CT image at the same coordinate position;

[0089] Step 4. Data enhancement: using translation to enhance, taking the centroid of each label image of each type of blood vessel as the center, making a 128*200 rectangular frame, randomly translating the rectangular frame along the x and y axes by a and b respectively, the value range is [-20, 20], and then cutting according to the frame. Due to the certain proportional difference between each type, in order to reduce the imbalance of the data, the left and right internal jugular veins are enhanced by 2 times, and the left and right brachiocephalic veins and the superior vena cava are enhanced by 3 times.

[0090] Step 5. Deep learning network construction: building an arbitrary 2D multi-classification segmentation network, taking dice loss as the loss function, and taking the average of the loss values of each type of blood vessel except the background as the final Loss value.

[0091] Step 6. Network model training: inputting the prepared label and the corresponding CT image into the 2D multi-classification network for training to obtain a blood vessel segmentation model;

[0092] Step 7. Segmentation prediction: cutting the test image, selecting the centroid of the largest connected region of each CT image as the reference, taking 128 pixels to the left, 100 pixels above and below, and cutting the image to 128*200; then taking 128 pixels to the right, 100 pixels above and below, and cutting the image to 128*200, as shown in Figure 3 ; calling the blood vessel segmentation model to realize prediction on the cut image, and putting the prediction results back into the original image size by left-right splicing, to complete the blood vessel segmentation, as shown in Figure 4 .

[0093] The present application provides a kind of computing device, including memory, processor and computer instruction stored on memory and can be run on processor, when processor executes instruction, the steps of the above-mentioned blood vessel segmentation method are realized;Wherein,

[0094] The technical scheme of the computing device belongs to the same concept with the technical scheme of the above-mentioned segmentation method, and the details not described in detail in the technical scheme of the computing device can be referred to the description of the technical scheme of the above-mentioned segmentation method.

[0095] The computing device can be any type of stationary or mobile computing device, including mobile computer or mobile computing device (for example, tablet computer, personal digital assistant, laptop computer, notebook computer, netbook, etc.), mobile phone (for example, smart phone), wearable computing device (for example, smart watch, smart glasses, etc.) or other types of mobile devices, or stationary computing device such as desktop computer or PC;The computing device can also be a mobile or stationary server.

[0096] The computer instruction includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc.

[0097] The present application provides a kind of storage medium, which stores computer instruction, and the steps of the above-mentioned blood vessel segmentation method are realized when computer instruction is executed by processor;Wherein,

[0098] The technical scheme of the storage medium belongs to the same concept with the technical scheme of the above-mentioned segmentation method, and the details not described in detail in the technical scheme of the storage medium can be referred to the description of the technical scheme of the above-mentioned segmentation method.

[0099] The storage medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0100] It should be noted that, for each of the above method embodiments, in order to facilitate description, it is expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other order or simultaneously.Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily necessary for the present application.

[0101] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0102] The preferred embodiments of the application disclosed above are only used to illustrate the present application. Alternative embodiments do not describe all the details and limit the application to the specific embodiments. Obviously, many modifications and variations can be made in light of the contents of the specification. The specification selects and describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited by the claims and their full scope and equivalents.

Claims

1. A method for blood vessel segmentation based on CT images, characterized in that, The method comprises the following steps: Collecting a CT image of a preset region of a body part, and marking a blood vessel to be segmented on the CT image; wherein the body part is divided into three regions, namely the head and neck to the heart, the heart to the pelvic cavity, and below the pelvic cavity; the blood vessel marking method is as follows: taking n-class blood vessel segmentation as an example: the background region is marked as 0, and different classes of blood vessels are marked as 1-n; Centering on each class of marked blood vessels, a blood vessel marked image of the same size containing the blood vessel is cropped, and a CT blood vessel image is cropped on the CT image with the same cropping frame; The same data enhancement is performed on the blood vessel marked image and the CT blood vessel image to produce multiple image sets; wherein the data enhancement includes one of translation, rotation and brightness enhancement; A deep learning network is constructed; wherein the deep learning network is a 2D multi-class deep learning segmentation network; the 2D multi-class deep learning segmentation network takes dice loss as the loss function, and the average of the loss values of each class of blood vessels except the background is taken as the final Loss value; The image set is input into the deep learning network for training to obtain a blood vessel segmentation model; The CT image to be segmented is cropped to the same size and input into the blood vessel segmentation model to identify the blood vessel class, and the prediction result is put back into the CT image to be segmented to complete the blood vessel segmentation.

2. The blood vessel segmentation method of claim 1, wherein, Taking the size of the blood vessel marked image and the CT blood vessel image as C*R as an example: The translation is as follows: the CT blood vessel image and the blood vessel marked image are randomly translated by a pixels along the x-axis direction, -C / 2≤a≤C / 2; and randomly translated by b pixels along the y-axis direction, -R / 2≤b≤R / 2; after translation, cropping is performed to obtain another set of CT blood vessel images and blood vessel marked images; The rotation is as follows: the CT blood vessel image and the blood vessel marked image are rotated by an angle θ around the center of the image, -90°≤θ≤90°; after rotation, cropping is performed to obtain another set of CT blood vessel images and blood vessel marked images; The brightness enhancement is as follows: the CT blood vessel image and the blood vessel marked image are histogram equalized for brightness enhancement.

3. A CT image-based blood vessel segmentation system, characterized by, The method comprises the following steps: The marking module is used for collecting a CT image of a preset region of a body part, and marking a blood vessel to be segmented on the CT image; wherein the body part is divided into three regions, namely the head and neck to the heart, the heart to the pelvic cavity, and below the pelvic cavity; the blood vessel marking method is as follows: taking n-class blood vessel segmentation as an example: the background region is marked as 0, and different classes of blood vessels are marked as 1-n; The cropping module is used for centering on each class of marked blood vessels, cropping a blood vessel marked image of the same size containing the blood vessel, and cropping a CT blood vessel image on the CT image with the same cropping frame; The data enhancement module is used for performing the same data enhancement on the blood vessel marked image and the CT blood vessel image to produce multiple image sets; wherein the data enhancement includes one of translation, rotation and brightness enhancement; The model construction module is configured to construct a deep learning network, wherein the deep learning network is a 2D multi-class deep learning segmentation network; the 2D multi-class deep learning segmentation network takes a dice loss as a loss function, and takes an average of loss values of each class of blood vessels except background as a final loss value. The model training module is configured to input the image set into the deep learning network for training to obtain a blood vessel segmentation model. The segmentation module is configured to crop a to-be-segmented CT image to the same size, input the to-be-segmented CT image into the blood vessel segmentation model, identify a blood vessel class, and put a prediction result back into the to-be-segmented CT image to complete blood vessel segmentation.

4. The blood vessel segmentation system of claim 3, wherein, In the data enhancement module: Taking the size of the blood vessel marked image and the CT blood vessel image as C*R for example: The translation is that the CT blood vessel image and the blood vessel marked image are randomly translated by a pixels along the x-axis direction, -C / 2≤a≤C / 2, and randomly translated by b pixels along the y-axis direction, -R / 2≤b≤R / 2; after translation, cutting is performed to obtain another set of CT blood vessel images and blood vessel marked images; The rotation is that the CT blood vessel image and the blood vessel marked image are rotated by an angle θ around the image center, -90°≤θ≤90°; after rotation, cutting is performed to obtain another set of CT blood vessel images and blood vessel marked images; The brightness enhancement is that the CT blood vessel image and the blood vessel marked image are subjected to histogram equalization for brightness enhancement.

5. A computing device comprising a memory, a processor, and computer instructions stored on the memory and executable on the processor, wherein, The processor executes the instructions to implement the steps of the blood vessel segmentation method in any one of claims 1-2.

6. A storage medium storing computer instructions, characterized in that, The computer instructions are executed by the processor to implement the steps of the blood vessel segmentation method in any one of claims 1-2.

Images