A method, apparatus and device for segmenting a CTA image
By preprocessing CTA images and using a segmentation network, the problem of difficult segmentation of the true and false lumens of the aorta was solved, achieving high-precision automated segmentation.
Patent Information
- Application Number
- CN202211565601.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-07
- Publication Date
- 2025-12-09
- Estimated Expiration
- 2042-12-07
AI Technical Summary
In CTA images, the contrast between the true lumen and the false lumen of the aorta is low and the boundaries are weak, making accurate segmentation difficult.
By preprocessing the CTA image, the target region image is extracted and downsampled, and then input into the first segmentation network for aortic region segmentation. After extracting the center line, a straightened blood vessel image is generated and input into the second segmentation network for true and false lumen segmentation. Finally, the result is restored to the original image.
It improves the segmentation accuracy of the true and false lumens of the aorta, restores the combined morphology of the circular lumen of the aorta, makes the dissection features more obvious, and achieves automated high-precision segmentation.
Smart Images
Figure CN115880314B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of medical image processing, in particular to a method, device and equipment for segmenting a CTA image. BACKGROUND
[0002] A CTA (CT angiography) image is an image obtained by performing CT scanning on a human body after injecting contrast agent into the blood vessels. The blood vessels in the CTA image are clearly developed, and the arteries and main branches can be clearly displayed, which can help doctors understand the blood supply of a patient.
[0003] Aortic dissection refers to the blood in the aorta entering the aortic media from the aortic intimal tear, separating the media, and expanding along the long axis of the aorta to form a true lumen and a false lumen of the aortic wall.
[0004] Due to the complexity of the aortic anatomical structure and the problem of the contrast agent, the contrast between the true lumen and the false lumen in the CTA image is low and the boundary is weak and difficult to identify, so the true lumen and the false lumen in the aorta cannot be accurately segmented. SUMMARY
[0005] Therefore, the embodiments of the present application provide a method, device and equipment for segmenting a CTA image to accurately segment the true lumen and the false lumen of the aorta in the CTA image.
[0006] To solve the above problems, the technical scheme provided by the embodiments of the present application is as follows:
[0007] A method for segmenting a CTA (CT angiography) image, the method comprising:
[0008] obtaining a CTA image in which an aortic dissection exists;
[0009] extracting a target region image from the CTA image;
[0010] down-sampling the target region image to generate a first target image, inputting the first target image into a first segmentation network, identifying an aorta region in the first target image, and up-sampling the first target image corresponding to the down-sampling to generate a second target image;
[0011] extracting a center line of the aorta region in the second target image;
[0012] sampling a cross-sectional image perpendicular to the center line from the second target image at each sampling point on the center line, and stacking each cross-sectional image to generate a blood vessel straightening image;
[0013] input the blood vessel straightening image into a second segmentation network, identify a true lumen region and a false lumen region in the blood vessel straightening image;
[0014] restore the true lumen region and the false lumen region in the blood vessel straightening image to the second target image to obtain the true lumen region and the false lumen region in the second target image.
[0015] In a possible implementation, the extracting a target region image from the CTA image comprises:
[0016] extracting a region with a largest volume and a pixel value greater than a target threshold value from the CTA image as a skeleton region;
[0017] obtaining a circumscribed cuboid of the skeleton region, determining a center point of each slice image in the circumscribed cuboid, and extracting an image group in a preset edge length range with the center point as a center in each slice image to form a target region image.
[0018] In a possible implementation, the method further comprises:
[0019] adjusting a pixel value of the target region image to a preset range.
[0020] In a possible implementation, the method further comprises:
[0021] adjusting a pixel interval of the CTA image to a preset value, the preset value being determined according to an average value of pixel intervals of different CTA images.
[0022] In a possible implementation, the extracting a center line of the aortic region in the second target image comprises:
[0023] extracting a skeleton line of the aortic region in the second target image by using a morphological thinning algorithm;
[0024] determining a center line of the aortic region from the skeleton line.
[0025] In a possible implementation, the determining a center line of the aortic region from the skeleton line comprises:
[0026] dividing points in the skeleton line into end points, connection points and branch points, the end points having one neighborhood point, the connection points having two neighborhood points, and the branch points having three or more neighborhood points;
[0027] determining two end points farthest away as a starting point and a terminal point;
[0028] Starting from the starting point, each point along the skeleton line is judged, when the judgment point is a bifurcation point, any neighborhood point of the bifurcation point is selected to continue the judgment, when the judgment point is an end point and is not a termination point, the bifurcation point with unselected neighborhood points is returned to continue the judgment from any unselected neighborhood point, when the judgment point is an end point and is a termination point, the starting point, the termination point and the points between the starting point and the termination point are determined as the center line of the aortic region.
[0029] In a possible implementation, the first segmentation network is a first three-dimensional Unet network, and the second segmentation network is a second three-dimensional Unet network.
[0030] An apparatus for segmenting a CT angiography (CTA) image, the apparatus comprising:
[0031] An acquisition unit configured to acquire a CTA image in which an aortic dissection exists;
[0032] A first extraction unit configured to extract a target region image from the CTA image;
[0033] A first identification unit configured to down-sample the target region image to generate a first target image, input the first target image into a first segmentation network, identify an aortic region in the first target image, and up-sample the first target image corresponding to the down-sampling to generate a second target image;
[0034] A second extraction unit configured to extract a center line of the aortic region in the second target image;
[0035] A sampling unit configured to sample, at each sampling point on the center line, a cross-sectional image perpendicular to the center line from the second target image, and stack each of the cross-sectional images to generate a vessel straightening image;
[0036] A second identification unit configured to input the vessel straightening image into a second segmentation network, identify a true lumen region and a false lumen region in the vessel straightening image;
[0037] A restoration unit configured to restore the true lumen region and the false lumen region in the vessel straightening image to the second target image to obtain the true lumen region and the false lumen region in the second target image.
[0038] An apparatus for segmenting a CT angiography (CTA) image, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for segmenting a CTA image as described above when executing the computer program.
[0039] A computer readable storage medium, the computer readable storage medium has stored instructions, when the instructions run on a terminal device, make the terminal device execute the method for segmenting the CT angiography CTA image as described above.
[0040] Therefore, the embodiments of the present application have the following beneficial effects:
[0041] The embodiments of the present application first preprocess the acquired three-dimensional CTA image to obtain a target region image. Then, the target region image is down-sampled to obtain a low-resolution first target image. The first target image is input into a first segmentation network, i.e., a coarse segmentation network, to obtain a segmentation prediction result of the down-sampled aortic region. The segmentation prediction result of the down-sampled aortic region is up-sampled to the original resolution of the target region image to generate a second target image, and the aortic region and the corresponding center line are extracted therefrom. On the basis of the aortic region segmentation, a two-dimensional cross-sectional image perpendicular to the center line at each sampling point on the center line is obtained along the aortic center line, stacked together to form a three-dimensional vessel straightening image, and input into a second segmentation network, i.e., a fine segmentation network, for true and false lumen segmentation. Finally, the true and false lumen segmentation result of the vessel straightening image is restored to the second target image to obtain the true and false lumen segmentation result. The vessel straightening image of the aorta is obtained, the combined shape of the aortic circular lumen is restored, the features of the dissection are more obvious, and the segmentation accuracy of the true and false lumens of the aorta is higher on the basis of automation. BRIEF DESCRIPTION OF DRAWINGS
[0042] Figure 1 A schematic diagram of an exemplary application scenario provided by the embodiments of the present application;
[0043] Figure 2 A flowchart of a method for segmenting a CTA image provided by the embodiments of the present application;
[0044] Figure 3 A schematic diagram of a CTA image in the embodiments of the present application;
[0045] Figure 4 A schematic diagram of a cross-sectional image in the embodiments of the present application;
[0046] Figure 5 A schematic diagram of a vessel straightening image in the embodiments of the present application;
[0047] Figure 6 A schematic diagram of adjusting the pixel value of a target region image in the embodiments of the present application;
[0048] Figure 7 A schematic diagram of extracting the center line of the aortic region in the embodiments of the present application;
[0049] Figure 8A schematic diagram of a device for segmenting a CTA image is provided in the embodiments of the present application. DETAILED DESCRIPTION
[0050] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the embodiments of the present application will be further described in detail below with reference to the drawings and specific embodiments.
[0051] In order to facilitate understanding and explaining the technical solutions provided by the embodiments of the present application, the background art of the present application will be described first.
[0052] CTA (CT angiography, CT angiography) image is an image obtained by CT scanning after injecting contrast agent into the blood vessels of the human body. The blood vessels in the CTA image are clearly developed, and the arteries and main branches can be clearly displayed, which can help doctors understand the blood supply of the patient.
[0053] Aortic dissection refers to the blood in the aorta entering the aortic media from the aortic intimal tear, causing the media to separate, and expanding along the long axis of the aorta to form a true lumen and a false lumen separation form of the aortic wall.
[0054] Due to the complexity of the aortic anatomical structure, it presents a curved state in the human body, and the aorta in the CTA image sequence is not completely regular circular, which brings difficulties to the true and false lumen segmentation. Moreover, due to the problem of contrast agent, the contrast of true and false lumen in the CTA image is low and the boundary is weak and difficult to identify, so the true and false lumens in the aorta cannot be accurately segmented.
[0055] Based on this, the embodiments of the present application provide a method, device and equipment for segmenting a CTA image. The embodiments of the present application first preprocess the acquired three-dimensional CTA image to obtain a target region image. Then, the target region image is down-sampled to obtain a low-resolution first target image. The first target image is input into a first segmentation network, i.e. a coarse segmentation network, to obtain a segmentation prediction result of the down-sampled aortic region. The segmentation prediction result of the down-sampled aortic region is up-sampled to the original resolution of the target region image to generate a second target image, and the aortic region therein is extracted and the corresponding center line is extracted. On the basis of the aortic region segmentation, a two-dimensional cross-sectional image perpendicular to the center line at each sampling point on the center line is obtained along the aortic center line, stacked together to form a three-dimensional vascular straightening image, and input into a second segmentation network, i.e. a fine segmentation network, for true and false lumen segmentation. Finally, the true and false lumen segmentation result of the vascular straightening image is restored to the second target image to obtain the true and false lumen segmentation result. The vascular straightening image of the aorta is obtained, the combined form of the aortic circular lumen is restored, the dissection has more obvious features, and the segmentation accuracy of the true and false lumens of the aorta is higher on the basis of automation.
[0056] For the convenience of understanding the method for segmenting the CTA image provided in the embodiments of the present application, the following will be described in conjunction with the example of the scenario shown in Figure 1 Referring to Figure 1 The figure is a schematic diagram of an exemplary application scenario provided by the embodiments of the present application.
[0057] In the embodiments of the present application, first, the CTA image of the aortic dissection is acquired. The region of interest is extracted from the CTA image to obtain the target region image with the original resolution. Then, the target region image is down-sampled and input into the first segmentation network, i.e., the coarse segmentation network, to obtain the segmentation prediction result of the down-sampled aortic region, and then up-sampled to the original resolution to extract the aortic region. Along the centerline direction of the aorta, the two-dimensional cross-sectional images perpendicular to the centerline at each sampling point on the centerline are acquired and stacked together to form a three-dimensional straightened vessel image, which is input into the second segmentation network, i.e., the fine segmentation network, for true and false lumen segmentation. Finally, the true and false lumen segmentation result of the straightened vessel image is restored to the image with the original resolution to obtain the true and false lumen segmentation result.
[0058] Those skilled in the art can understand that Figure 1 The framework diagram shown is only one example in which the embodiments of the present application can be implemented. The scope of application of the embodiments of the present application is not limited by any aspect of the framework.
[0059] For the convenience of understanding the present application, the following will describe a method for segmenting a CTA image provided by the embodiments of the present application in conjunction with the accompanying drawings.
[0060] Referring to Figure 2 The figure is a flowchart of a method for segmenting a CTA image provided by the embodiments of the present application, as shown in Figure 2 The method can include S201-S207:
[0061] S201: Acquire the CTA image in which the aortic dissection exists.
[0062] The CTA image is a three-dimensional CTA image, i.e., the CTA image includes multiple slice images, and the size of the CTA image can be M*N*L, where M, N are the length and width of each slice image, for example, M is 512, N is 512, and L is the number of slice images in the CTA image. Referring to Figure 3 , Figure 3 The figure is a schematic diagram of the CTA image of the aortic dissection provided by the embodiments of the present application. Figure 3 Four slice images of the CTA image are shown in
[0063] In this step, the CTA image acquired has an aortic dissection, indicating that the patient has a blood vessel false lumen cavity region in the vascular cavity region. Further, the true lumen region and the false lumen region need to be further determined.
[0064] S202: Extract a target region image from the CTA image.
[0065] The aorta is generally located in the middle region of the CTA image. In order to reduce the influence of other regions of the CTA image on subsequent identification, the region of interest can be extracted from the CTA image first, that is, a three-dimensional target region image is extracted. Since the aorta is contained in the sternum skeleton, the skeleton has a higher influence on the density and is easier to extract, so the skeleton region in the CTA image can be used to extract the target region image.
[0066] In a possible implementation, the specific implementation of S202 of extracting the target region image from the CTA image can include:
[0067] A1: Extract a region with the largest volume and a pixel value greater than a target threshold from the CTA image to determine the skeleton region.
[0068] The target threshold (such as 200) is set, and the image threshold segmentation is performed by using an image processing tool to extract an image region with a pixel value greater than the set target threshold in the CTA image. The region will contain the skeleton, high-density blood vessels, and the like. The extracted region is sorted in descending order of volume, and the region with the largest volume is extracted as the skeleton region. The target threshold can be set according to actual conditions, and the value of the target threshold is not limited in the embodiments of the present application.
[0069] A2: Obtain an outer rectangle of the skeleton region, determine a center point of each slice image in the outer rectangle, and extract an image group in a preset edge length range with the center point as the center in each slice image to form a target region image.
[0070] Then find the outer rectangle of the skeleton region, set the initial vertex coordinates of the outer rectangle as (X0, Y0, Z0), and the three edge lengths of the corresponding axes as (a, b, c). Then based on the outer rectangle coordinates, the center point coordinates (Xc, Yc) of each slice image in the outer rectangle can be calculated.
[0071] Xc=X0+a / 2, Yc=Y0+b / 2.
[0072] A target region image is extracted from the center point in each slice image within a preset edge length range. Assuming that the preset edge length of the x-axis is d and the preset edge length of the y-axis is e, for example, d is M / 2=256 and e is N / 2=256. Then, the target region image can be extracted in the area [Xc-d / 2:Xc+d / 2,Yc-e / 2:Yc+e / 2] in each slice image. For example, the image in the range [Xc-128:Xc+128,Yc-128:Yc+128] is extracted from the center point in each slice image to form a three-dimensional target region image. The size of the target region image can be (M / 2)*(M / 2)*L. The preset edge length can be set according to actual conditions, and the present embodiment does not limit the value of the preset edge length.
[0073] In the present embodiment, the obtained target region image includes the aorta region and removes the non-central region in the CTA image that does not need to be focused on. The subsequent identification is more accurate using the target region image.
[0074] S203: downsampling the target region image to generate a first target image, inputting the first target image into a first segmentation network, identifying the aorta region in the first target image, and upsampling the first target image corresponding to the downsampling to generate a second target image.
[0075] The target region image is downsampled to a fixed preset size to generate a first target image. The sizes of the axes of the target region image can be different, and then the downsampling can be performed respectively. For example, if the downsampling rate is 2, the size of the first target image is (M / 4)*(N / 4)*(L / 2). The first target image is input into the first segmentation network for aorta segmentation. The first segmentation network can be understood as a coarse segmentation network. The first segmentation network can obtain the segmentation prediction result of whether each pixel of the first target image is an aorta region.
[0076] In a possible implementation, the first segmentation network is a first three-dimensional Unet network. In actual application, the size of the first target image is a multiple of 2 raised to the power of Q, where Q is the number of down-sampling of the UNet network, and 2 is the size of the pooling kernel. After Q times of down-sampling, the size of the feature map of the first target image should reach about 4. For example, the first three-dimensional Unet network has 5 times of down-sampling, that is, the down-sampling kernel is [2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], and [2, 2, 2]. The segmentation prediction result of whether each pixel of the first target image is in the aortic region is obtained through the first three-dimensional Unet network, and then the aortic region in the first target image can be obtained. Then, the first target image is up-sampled corresponding to the down-sampling to generate a second target image. The size of the second target image is restored to (M / 2)*(M / 2)*L, and at this time, the segmentation prediction result of whether each pixel in the second target image is in the aortic region can be obtained, and then the aortic region in the second target image can be obtained.
[0077] In actual application, a threshold T1 can be set, and for the segmentation prediction result x of the second target image, a pixel greater than the threshold T1 is set to 1, representing that the pixel belongs to the aortic region, and the rest is set to 0, representing that the pixel does not belong to the aortic region, to obtain the mask value of the aortic region of the second target image.
[0078]
[0079] All pixels with f(x) being 1 are connected to obtain a plurality of connected regions, and the largest connected region is found, that is, the aortic region in the second target image.
[0080] In a possible implementation, the 3D Unet network in the first segmentation network can also be replaced with other segmentation networks.
[0081] S204: Extract the center line of the aortic region in the second target image.
[0082] After the aortic region in the second target image is obtained, the center line of the aortic region can be extracted, and the center line can represent the overall trend of the aorta.
[0083] The specific implementation of extracting the center line of the aortic region will be described in detail in subsequent embodiments, and will not be described here.
[0084] S205: Sample a cross-sectional image perpendicular to the center line from the second target image at each sampling point on the center line, and stack the cross-sectional images to generate a blood vessel straightening image.
[0085] Because the aorta is curved within the human body, it is not perfectly circular in CTA image sequences, making the segmentation of the true and false lumens difficult. Therefore, in this embodiment, cross-sectional images perpendicular to the centerline are sampled from the second target image at various sampling points along the centerline. These cross-sectional images are then stacked together to generate a straightened blood vessel image. Simultaneously, the correspondence between the straightened blood vessel image and the second target image is recorded, that is, the pixel positions in the second target image from which the straightened blood vessel image was sampled are recorded.
[0086] For example, setting the sampling interval to 1mm results in K sampling points. Each sampling point corresponds to a 96*96 pixel 2D cross-sectional image, and stacking them together yields a blood vessel straightening image of size 96*96*K. (See also...) Figure 4 The image shown is a schematic diagram of a cross-sectional image. Because the cross-sectional image is obtained by sampling perpendicular to the aortic centerline, the aorta appears as a relatively regular circle in the image, making the subsequent segmentation of the true lumen and false lumen more accurate. See also... Figure 5 The image shown is a schematic diagram of a straightened blood vessel.
[0087] S206: Input the straightened blood vessel image into the second segmentation network to identify the true lumen region and the false lumen region in the straightened blood vessel image.
[0088] The image of the straightened blood vessel is input into a second segmentation network for true and false lumen segmentation. This second segmentation network can be understood as a fine-grained segmentation network. It predicts whether each pixel in the straightened blood vessel image represents a true or false lumen region.
[0089] In one possible implementation, the second segmentation network is a second 3D UNet network. In practical applications, the size of the straightened blood vessel image should be a multiple of 2 raised to the power of Q, where Q is the number of downsampling operations performed by the UNet network, and 2 is the size of the pooling kernel. After Q downsampling operations, the feature map size of the straightened blood vessel image should reach approximately 4. For example, if the input image size is 96*96*K, the second 3D UNet network performs downsampling operations of [4,4,5], i.e., the downsampling kernels are [2,2,2], [2,2,2], [2,2,2], [2,2,2], [1,1,2]. The segmentation prediction result obtained by the second 3D UNet network, determining whether each pixel in the straightened blood vessel image is a true cavity, is x. true The segmentation prediction result for whether it is a false cavity is x false This allows us to obtain the true lumen region and the false lumen region in the straightened blood vessel image.
[0090] In practical applications, a threshold T2 can be set for the segmentation prediction result x to determine whether it is a true cavity. trueThe pixels greater than the threshold T2 are set to 1, representing that the pixels belong to the true lumen region, and the rest are set to 0, representing that the pixels do not belong to other regions, to obtain the mask value of the true lumen region in the blood vessel straightening image.
[0091]
[0092] Similarly, the threshold T3 can be set to predict the segmentation result x of whether it is a false lumen. false The pixels greater than the threshold T3 are set to 2, representing that the pixels belong to the false lumen region, and the rest are set to 0, representing that the pixels do not belong to other regions, to obtain the mask value of the false lumen region in the blood vessel straightening image.
[0093]
[0094] The mask values of the true lumen and the false lumen are merged to obtain the true lumen region and the false lumen region in the blood vessel straightening image.
[0095] In a possible implementation, the 3D Unet network in the second segmentation network can also be replaced with other segmentation networks.
[0096] S207: Restoring the true lumen region and the false lumen region in the blood vessel straightening image to the second target image to obtain the true lumen region and the false lumen region in the second target image.
[0097] After obtaining the true and false lumen segmentation result of the blood vessel straightening image, the result can also be restored to the second target image. Since the corresponding relationship between the pixels of the blood vessel straightening image and the second target image has been recorded, the true and false lumen segmentation result of the second target image can be obtained through reverse mapping as an initial result. Because the forward mapping will have some repeated pixels, the image obtained after reverse mapping will have holes, and these holes can be filled through the nearest neighbor interpolation method to obtain the true lumen region and the false lumen region in the second target image.
[0098] Based on the description of S201-S207, the embodiments of the present application first pre-process the acquired three-dimensional CTA image to obtain a target region image. Then, the target region image is down-sampled to obtain a low-resolution first target image. The first target image is input into a first segmentation network, i.e., a coarse segmentation network, to obtain a segmentation prediction result of the down-sampled aortic region. The segmentation prediction result of the down-sampled aortic region is up-sampled to the original resolution of the target region image to generate a second target image, and the aortic region therein is extracted and the corresponding center line is extracted. On the basis of the aortic region segmentation, a two-dimensional section image perpendicular to the center line at each sampling point on the center line is obtained along the aortic center line, stacked together to form a three-dimensional straightened vessel image, and input into a second segmentation network, i.e., a fine segmentation network, for true and false lumen segmentation. Finally, the true and false lumen segmentation result of the straightened vessel image is restored to the second target image to obtain the true and false lumen segmentation result. The straightened vessel image of the aorta is obtained, the combined shape of the aortic circular lumen is restored, the features of the dissection are more obvious, and the segmentation accuracy of the true and false lumens of the aorta is higher on the basis of automation.
[0099] Since the pixel value (i.e., CT value) in the CTA image has a wide range of values, the pixel value can be adjusted to a preset range to make the subsequent segmentation of the target object more accurate. Therefore, in a possible implementation manner, after S202, it can further include: adjusting the pixel value of the target region image to a preset range.
[0100] In a specific implementation, the pixel value can be adjusted to the preset range by adjusting the window width and window level. The minimum value of the window width range is window_min and the maximum value is window_max by setting the window width and window level, and the pixel value p of each pixel is adjusted by using the following formula ′
[0101] ′
[0102]
[0103] ′ ′
[0104] Based on the above formula, the pixel value of the target region image can be adjusted to the range of 0-q, for example, window_min=0, window_max=1000, q=255, through the above formula, the pixel value of the pixel point whose pixel value of the target region image exceeds 1000 is converted to 255, and the pixel value of the pixel point whose pixel value of the target region image is less than 0 is converted to 0. The pixel value of the pixel point whose pixel value of the target region image is in the range of 0-1000 is converted to the range of 0-255 in proportion.
[0105] wherein, the preset range can be set according to actual situation, and the value of the preset range is not limited in the embodiment of the application. Referring to Figure 6 It is shown that the process of adjusting the pixel value of the target region image.
[0106] In the embodiment of the application, adjusting the pixel value of the target region image to a smaller range can make the subsequent aortic region segmentation and true and false cavity region segmentation more accurate.
[0107] In a possible implementation, after S201, it can further include:
[0108] The pixel spacing of the CTA image is adjusted to a preset value, and the preset value is determined according to the average value of the pixel spacing of different CTA images.
[0109] Since the acquisition devices of CTA images are different, the pixel spacing spacing of different CTA images also has a large difference. In order to make the recognition effect of the first segmentation network and the second segmentation network more accurate, it is necessary to unify the pixel spacing of different CTA images, that is, it is necessary to adjust the pixel spacing of the CTA image to a preset value.
[0110] In actual application, the pixel spacing spacing(ΔX i , ΔY i , ΔZ i ) of all original CTA images can be obtained, and the average value is calculated as follows:
[0111]
[0112] wherein, i is the i-th CTA image, and n is the total number of CTA images.
[0113] Taking as the target pixel spacing, i.e., the preset value, a cubic spline interpolation algorithm (other interpolation methods can also be selected) is used for image transformation, so that the pixel spacing of the transformed image remains the same, i.e., equal to
[0114] In a possible implementation, the specific implementation of S204 extracting the centerline of the aorta region in the second target image can include:
[0115] B1: extracting the skeleton line of the aorta region in the second target image by using a morphological thinning algorithm.
[0116] The skeleton line can be extracted from the aorta region by using the morphological thinning algorithm, and the skeleton line includes the centerline of the aorta region and some other branch lines. Referring to FIG. 2B, the black lines are the skeleton line, and the centerline can be extracted from the skeleton line. Figure 7
[0117] B2: determining the centerline of the aorta region from the skeleton line.
[0118] In a possible implementation, the specific implementation of B2 determining the centerline of the aorta region from the skeleton line can include:
[0119] C1: dividing the points in the skeleton line into end points, connection points, and branch points, the end points having one neighborhood point, the connection points having two neighborhood points, and the branch points having three or more neighborhood points.
[0120] Each point on the skeleton line is classified. When the point has only one neighborhood point, the point is defined as an end point; when the point has two neighborhood points, the point is defined as a connection point; and when the point has three or more neighborhood points, the point is defined as a branch point. For example, in FIG. 2C, points A, B, C, and D are end points, points E and F are branch points, and the other points in the skeleton line are connection points. Figure 7
[0121] C2: determining the two end points farthest away as the start point and the end point.
[0122] The two end points farthest away along the skeleton line are determined as the start point and the end point of the aorta region. For example, point A is the start point, and point B is the end point.
[0123] C3: starting from the start point, judging each point along the skeleton line, when the judged point is a branch point, selecting any neighborhood point of the branch point to continue the judgment, when the judged point is an end point and is not the end point, returning to the branch point with unselected neighborhood points to continue the judgment from any unselected neighborhood point, and when the judged point is an end point and is the end point, determining the start point, the end point, and the points between the start point and the end point as the centerline of the aorta region.
[0124] Starting from the starting point, each point on the skeleton line is judged. When a branch point is reached, any neighborhood point of the branch point is selected to continue the judgment. For example, starting from point A, when point E is judged, the path to point C is selected for judgment. When the judgment point is an end point and is not a termination point, return to the branch point with unselected neighborhood points to continue the judgment from any unselected neighborhood point. For example, when point C is judged as an end point and is not a termination point, return to point E to reselect the path to point F to continue the judgment. By analogy, until a point is found as an end point and a termination point, the starting point, the termination point and the points between the starting point and the termination point on the skeleton line can be determined as the center line of the aortic region. For example, when point B is found as a termination point, points A, B and points on the skeleton line between points A and B are determined as the center line of the aortic region.
[0125] In the embodiment of the present application, the center line of the aortic region can be extracted, which represents the direction of the core blood vessel of the aorta. By accurately extracting the center line of the aortic region, the true and false lumen regions in the aortic region can be accurately segmented.
[0126] Based on the method for segmenting the CTA image provided in the above method embodiment, the present application further provides a device for segmenting the CTA image, which will be described below in combination with the drawings.
[0127] Referring to Figure 8 , the figure is a structural schematic diagram of a device for segmenting a CTA image provided in an embodiment of the present application. As Figure 8 , the device for segmenting the CTA image comprises:
[0128] The acquisition unit 801 is configured to acquire a CTA image in which an aortic dissection exists.
[0129] The first extraction unit 802 is configured to extract a target region image from the CTA image.
[0130] The first identification unit 803 is configured to down-sample the target region image to generate a first target image, input the first target image into a first segmentation network, identify an aortic region in the first target image, and up-sample the first target image corresponding to the down-sampling to generate a second target image.
[0131] The second extraction unit 804 is configured to extract a center line of the aortic region in the second target image.
[0132] The sampling unit 805 is configured to sample a cross-sectional image perpendicular to the center line from the second target image at each sampling point on the center line, and stack each cross-sectional image to generate a blood vessel straightening image.
[0133] The second identification unit 806 is configured to input the blood vessel straightening image into a second segmentation network, identify a true lumen region and a false lumen region in the blood vessel straightening image.
[0134] The restoring unit 807 is configured to restore the true lumen region and the false lumen region in the blood vessel straightening image into the second target image, to obtain the true lumen region and the false lumen region in the second target image.
[0135] In a possible implementation, the first extraction unit comprises:
[0136] The first extraction sub-unit is configured to extract a region with a largest volume and a pixel value greater than a target threshold value from the CTA image as a skeleton region.
[0137] The second extraction sub-unit is configured to obtain a circumscribed cuboid of the skeleton region, determine a center point of each slice image in the circumscribed cuboid, and extract a target region image composed of images within a preset edge length range with the center point as a center in each slice image.
[0138] In a possible implementation, the device further comprises:
[0139] The first adjusting unit is configured to adjust a pixel value of the target region image to a preset range.
[0140] In a possible implementation, the device further comprises:
[0141] The second adjusting unit is configured to adjust a pixel interval of the CTA image to a preset value, and the preset value is determined according to an average value of pixel intervals of different CTA images.
[0142] In a possible implementation, the second extraction unit comprises:
[0143] The third extraction sub-unit is configured to extract a skeleton line of the aortic region in the second target image by using a morphological thinning algorithm.
[0144] The first determination sub-unit is configured to determine a center line of the aortic region from the skeleton line.
[0145] In a possible implementation, the first determination sub-unit comprises:
[0146] The division sub-unit is configured to divide points in the skeleton line into end points, connection points, and branch points, the end points have one neighborhood point, the connection points have two neighborhood points, and the branch points have three or more neighborhood points.
[0147] The second determination sub-unit is configured to determine two end points farthest away as a starting point and an ending point.
[0148] a judging subunit, configured to judge each point along the skeleton line from the starting point, when the judging point is a bifurcation point, select any neighborhood point of the bifurcation point to continue judging, when the judging point is an end point and is not a termination point, return to the bifurcation point with unselected neighborhood points to continue judging from any unselected neighborhood point, and when the judging point is an end point and is a termination point, determine the starting point, the termination point and the points between the starting point and the termination point as the center line of the aortic region.
[0149] In a possible implementation, the first segmentation network is a first three-dimensional Unet network, and the second segmentation network is a second three-dimensional Unet network.
[0150] In addition, the embodiment of the present application further provides a device for segmenting a CTA image, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the method for segmenting a CTA image according to any one of the above is implemented.
[0151] In addition, the embodiment of the present application further provides a computer readable storage medium, wherein instructions are stored in the computer readable storage medium, and when the instructions run on a terminal device, the terminal device executes the method for segmenting a CTA image according to any one of the above.
[0152] In this way, the embodiment of the present application first pre-processes the acquired three-dimensional CTA image to obtain a target region image. Then, the target region image is down-sampled to obtain a low-resolution first target image. The first target image is input into a first segmentation network, i.e., a coarse segmentation network, to obtain a segmentation prediction result of the down-sampled aortic region. The segmentation prediction result of the down-sampled aortic region is up-sampled to the original resolution of the target region image to generate a second target image, the aortic region in the second target image is extracted, and a corresponding center line is extracted. On the basis of the aortic region segmentation, a two-dimensional cross-section image perpendicular to the center line at each sampling point on the center line is acquired along the aortic center line, stacked together to form a three-dimensional straightened vessel image, and input into a second segmentation network, i.e., a fine segmentation network, for true and false lumen segmentation. Finally, the true and false lumen segmentation result of the straightened vessel image is restored to the second target image to obtain the true and false lumen segmentation result. The straightened vessel image of the aorta is acquired, the combined shape of the aortic circular lumen is restored, the characteristics of the dissection are more obvious, and on the basis of automation, the segmentation accuracy of the true and false lumens of the aorta is higher.
[0153] It should be noted that the various embodiments described herein are described with reference to a plurality of distinct embodiments; however, persons of ordinary skill in the art will understand that each of these embodiments represents but a single exemplification of the various embodiments disclosed herein. Accordingly, any particular stated implementation of an embodiment should not be treated as a limitation to the embodiments described herein. Additionally, it is contemplated that individuals skilled in the art will be able to devise their own implementation of embodiments combining aspects of the various embodiments described herein without the need for undue experimentation.
[0154] It should be noted that in the present application, "at least one" means one or more, and "multiple" means two or more. "And / or" is used to describe the relationship between the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent: only A, only B, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c, can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.
[0155] It should also be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.
[0156] The steps of the method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module can be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0157] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method of segmenting a CT angiography, CTA, image, characterized by, The method comprises: obtaining a CTA image in which an aortic dissection exists; extracting a target region image from the CTA image; down-sampling the target region image to generate a first target image, inputting the first target image into a first segmentation network, identifying an aortic region in the first target image, and up-sampling the first target image corresponding to the down-sampling to generate a second target image; extracting a center line of the aortic region in the second target image; sampling a cross-section image perpendicular to the center line from the second target image at each sampling point on the center line to generate a blood vessel straightening image by stacking each cross-section image; inputting the blood vessel straightening image into a second segmentation network to identify a true lumen region and a false lumen region in the blood vessel straightening image; restoring the true lumen region and the false lumen region in the blood vessel straightening image to the second target image to obtain the true lumen region and the false lumen region in the second target image; the extraction of the center line of the aortic region in the second target image comprises: extracting a skeleton line of the aortic region in the second target image by using a morphological thinning algorithm; determining the center line of the aortic region from the skeleton line; the determination of the center line of the aortic region from the skeleton line comprises: dividing points in the skeleton line into end points, connection points and branch points, the end points having one neighborhood point, the connection points having two neighborhood points, and the branch points having three or more neighborhood points; determining the two most distant end points as a starting point and a terminal point; starting from the starting point, judging each point along the skeleton line, when the judged point is a branch point, selecting any neighborhood point of the branch point to continue the judgment, when the judged point is an end point and is not the terminal point, returning to a branch point having an unselected neighborhood point, and continuing the judgment from any unselected neighborhood point, and when the judged point is an end point and is the terminal point, determining the starting point, the terminal point and points between the starting point and the terminal point as the center line of the aortic region.
2. The method of claim 1, wherein, the extraction of the target region image from the CTA image comprises: extracting a region with a pixel value greater than a target threshold and a largest volume from the CTA image as a skeleton region; obtaining a circumscribed cuboid of the skeleton region, determining a center point of each slice image in the circumscribed cuboid, and extracting an image group within a preset edge length range centered on the center point in each slice image to form a target region image.
3. The method of claim 2, wherein, The method further comprises: adjusting the pixel value of the target region image to a preset range.
4. The method according to any one of claims 1 to 3, characterized in that, The method further comprises: adjusting the pixel interval of the CTA image to a preset value, the preset value being determined according to the average value of the pixel intervals of different CTA images.
5. The method of claim 1, wherein, The first segmentation network is a first three-dimensional Unet network, and the second segmentation network is a second three-dimensional Unet network.
6. An apparatus for segmenting a CT angiography, CTA, image, the apparatus comprising: a processor configured to: receive a CTA image; and segment the CTA image using a neural network trained to segment CTA images. The device comprises: an obtaining unit configured to obtain a CTA image in which an aortic dissection exists; a first extracting unit configured to extract a target region image from the CTA image; The first identification unit is configured to down-sample the target region image to generate a first target image, input the first target image into a first segmentation network, identify an aorta region in the first target image, and up-sample the first target image corresponding to the down-sampling to generate a second target image. The second extraction unit is configured to extract a center line of the aorta region in the second target image. The sampling unit is configured to sample a cross-section image perpendicular to the center line from the second target image at each sampling point on the center line, and stack each cross-section image to generate a blood vessel straightening image. The second identification unit is configured to input the blood vessel straightening image into a second segmentation network, identify a true lumen region and a false lumen region in the blood vessel straightening image. The restoring unit is configured to restore the true lumen region and the false lumen region in the blood vessel straightening image to the second target image to obtain the true lumen region and the false lumen region in the second target image. The second extraction unit comprises: The third extraction sub-unit is configured to extract a skeleton line of the aorta region in the second target image by using a morphological thinning algorithm. The first determination sub-unit is configured to determine the center line of the aorta region from the skeleton line. The first determination sub-unit comprises: The division sub-unit is configured to divide points in the skeleton line into end points, connection points, and branch points, the end points having one neighborhood point, the connection points having two neighborhood points, and the branch points having three or more neighborhood points. The second determination sub-unit is configured to determine two end points farthest away as a start point and an end point. The judgment sub-unit is configured to start from the start point to judge each point along the skeleton line, when the judged point is a branch point, select any neighborhood point of the branch point to continue the judgment, when the judged point is an end point and is not the end point, return to a branch point having an unselected neighborhood point, and continue the judgment from any unselected neighborhood point, and when the judged point is an end point and is the end point, determine the start point, the end point, and points between the start point and the end point as the center line of the aorta region.
7. A device for segmenting CT angiography (CTA) images, characterized in that, The computer readable storage medium stores instructions, and when the instructions run on the terminal device, the terminal device performs the method for segmenting the CT angiography (CTA) image. The computer readable storage medium stores instructions, and when the instructions run on the terminal device, the terminal device performs the method for segmenting the CT angiography (CTA) image.
8. A computer-readable storage medium, characterized in that,
Citation Information
Patent Citations
Dissection artery segmentation method and device
CN110796670A
Septotomy Catheter For Aortic Dissection
US20160228181A1