Blood vessel segmentation method, device and equipment and storage medium

Abstract

The invention provides a blood vessel segmentation method, apparatus and device, and a storage medium. The method comprises the steps of obtaining a blood vessel wall image of a to-be-segmented blood vessel; slicing the blood vessel wall image to obtain a cross section slice image of the blood vessel; and segmenting the cross section slice image based on a pre-trained image segmentation model, and segmenting a blood vessel wall from the cross section slice image. According to the invention, the blood vessel is automatically analyzed through the full convolutional neural network, and the efficiency and accuracy of interpretation of the magnetic resonance blood vessel wall image are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing, and in particular to a blood vessel segmentation method, device, equipment and storage medium. Background Art

[0002] Magnetic resonance vessel wall imaging (VWI) is currently the only non-invasive intracranial vessel wall imaging that can directly visualize intracranial vessel wall lesions, which is helpful for the identification of stroke etiology, prognosis judgment, and treatment monitoring. Clinically, the interpretation of this image currently requires imaging experts to conduct qualitative observations and quantitative measurements based on the morphology of vessel wall lesions. Convolutional neural networks are a recently emerged deep learning technology that can be used for automatic image segmentation. However, current vessel wall image segmentation technology still requires manual outlining to mark abnormal locations, a process that consumes a lot of manpower and costs. Summary of the invention

[0003] In view of this, the present disclosure proposes a blood vessel segmentation method, apparatus, device and storage medium, which improve the efficiency and accuracy of blood vessel image segmentation.

[0004] According to a first aspect of the present disclosure, a blood vessel segmentation method is provided, comprising: Acquire a blood vessel wall image of the blood vessel to be segmented; Slicing the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; The cross-sectional slice image is segmented based on a pre-trained image segmentation model, and a blood vessel wall is segmented from the cross-sectional slice image.

[0005] In a possible implementation, when the slicing process is performed on the blood vessel wall image, the process includes: acquiring a blood vessel segment determined from the blood vessel wall image; The blood vessel segment is straightened to reconstruct continuous cross-sectional slices of a preset thickness as the cross-sectional slice image.

[0006] In a possible implementation, when the cross-sectional slice image is segmented, it is based on a pre-trained FCN model.

[0007] In a possible implementation, the cross-sectional slice image is segmented based on a pre-trained image segmentation model, and when the blood vessel wall is segmented from the cross-sectional slice image, at least one of the segmented background and lumen is also included; Wherein, the vessel wall, the background and the lumen are characterized by different channels.

[0008] In a possible implementation, the slicing process is performed on the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel, including: Selecting the blood vessel segment according to the region of interest; The selected blood vessel segment is sliced ​​to generate continuous cross-sectional slice images.

[0009] In a possible implementation, segmenting a cross-sectional slice image based on a pre-trained FCN model includes: Use adaptive histogram equalization to process cross-sectional slice images; Convolving the processed cross-sectional slice image with the filter to generate a plurality of feature images corresponding to the number of filters; The multiple feature images are upsampled to generate multiple types of segmented images, and the multiple types of segmented images are fused to obtain a final segmentation result.

[0010] According to a second aspect of the present disclosure, there is provided a device for segmenting a blood vessel, comprising: An acquisition module, used for acquiring a blood vessel wall image of a blood vessel to be segmented; A processing module, used for performing slice processing on the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; A segmentation module, used to segment the cross-sectional slice image based on a pre-trained image segmentation model, and segment the blood vessel wall from the cross-sectional slice image; The fusion module is used to upsample the multiple feature images to generate multiple types of segmented images, and fuse the multiple types of segmented images to obtain a final segmentation result.

[0011] According to a third aspect of the present disclosure, there is provided a blood vessel segmentation device, comprising: a processor; a memory for storing instructions executable by the processor; Wherein, the processor is configured to implement any of the above methods when executing the executable instructions.

[0012] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, characterized in that when the computer program instructions are executed by a processor, any one of the methods described above is implemented.

[0013] In the present disclosure, a blood vessel segmentation method includes: obtaining a blood vessel wall image of a blood vessel to be segmented; slicing the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; segmenting the cross-sectional slice image based on a pre-trained image segmentation model, and segmenting the blood vessel wall from the cross-sectional slice image. The present disclosure performs automated analysis of blood vessels through a fully convolutional neural network, thereby improving the efficiency and accuracy of interpreting magnetic resonance blood vessel wall images.

[0014] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0016] Figure 1 A flow chart showing a blood vessel segmentation method according to an embodiment of the present disclosure; Figure 2 A structural diagram of an FCN network architecture according to an embodiment of the present disclosure is shown; Figure 3 An example of segmenting an image according to an embodiment of the present disclosure is shown; Figure 4 A schematic block diagram of a blood vessel segmentation device according to an embodiment of the present disclosure is shown; Figure 5 A schematic block diagram of a blood vessel segmentation device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0017] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0018] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0019] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0020] <Method Example> Figure 1 FIG. 5 is a flow chart showing a blood vessel segmentation method according to an embodiment of the present disclosure. Figure 1 As shown, the method includes steps S100-S120.

[0021] S100, acquiring a blood vessel wall image of a blood vessel to be segmented; S110, performing slicing processing on the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; S120 segments the cross-sectional slice image based on a pre-trained image segmentation model, and segments the blood vessel wall from the cross-sectional image.

[0022] It should be noted here that the vascular wall image is a 3D image of the whole brain obtained based on magnetic resonance vascular wall imaging (VWI, vessel wall imaging) technology.

[0023] In one possible implementation, intracranial vascular wall images are acquired on a 3-Tesla whole-body MR system equipped with a 32-channel head coil. The imaging parameters include: 3D sagittal direction; TR = 900 milliseconds; TE = 15 milliseconds; field of view = 70×170 mm 2 ; Number of slices = 240; Spatial resolution = 0.53 mm.

[0024] Further, a vascular segment is selected according to the region of interest; (for example, in a specific embodiment of the present application, the plaques of ICAD lesions are mainly present in the terminal segment of the internal carotid artery, the M1 segment of the middle cerebral artery, the vertebral artery and the basilar artery, that is, at least one of the above vascular segments is selected and sliced.) Specifically, the selection of a blood vessel segment includes: specifying the start and end points of the blood vessel segment, then tracking the blood vessel centerline, and manually drawing the blood vessel centerline; Based on the manually drawn vascular centerline in the above vascular centerline tracking process, the selected vascular segment is straightened through 3D curved multi-planar reconstruction (MPR) to reconstruct cross-sectional slices of preset thickness. The cross-sectional slice images obtained in the vascular segment are continuous.

[0025] Preferably, in a specific embodiment of the present application, the preset thickness is 0.5 mm, the cross-sectional slices are sampled to 0.1 mm in-plane resolution, and the cross-sectional slices are exported in DICOM format.

[0026] According to the cross-sectional slice images obtained in the above steps, the blood vessel wall is segmented from the cross-sectional slice images based on the fully convolutional neural network (FCN) architecture, such as Figure 2 ,The FCN network architecture consists of an encoding path and a decoding path.

[0027] Specifically, the encoding path includes 3 convolutional units, and adjacent convolutional units are connected through a maximum pooling layer. The encoding path can downsample the image size to half its size.

[0028] Each convolutional unit consists of a convolutional layer, a batch normalization layer, and a parametric rectified linear unit activation layer.

[0029] In one possible implementation, a cross-sectional slice image of a set input size W (128×128) is first processed using contrast-limited adaptive histogram equalization (CLAHE), corresponding to (12.8 mm 2 ) field of view, first divide the input cross-sectional slice image into 8×8 grids of equal size and non-overlapping, where each grid size is 16×16; histogram equalization is performed on each divided grid, and then bilinear interpolation is used to eliminate the mutations between blocks for each grid after CLAHE; they are combined to eliminate artificially induced boundaries and improve local contrast.

[0030] Among them, using contrast-limited adaptive histogram processing to equalize grayscale values ​​of images belongs to a conventional image processing technology in the technical field of this application and will not be described in detail in this application.

[0031] The contrast-enhanced cross-sectional slice image is then sent to the first convolutional unit, which uses 32 learnable filters of size K=3 to convolve with the input image to generate 32 feature maps of size W (128×128).

[0032] It should be pointed out that the number and size of filters and the corresponding number of feature maps generated can be adjusted according to actual conditions. The number is not specifically limited here. The following batch normalization, PReLU activation and convolutional layers also have the same filter size and number of feature maps.

[0033] According to the given equation:

[0034] Calculate the output image size of the convolutional layer, where W is the input image size; O is the convolutional layer output image size; K is the filter size; P is the padding; and S is the stride.

[0035] Setting the stride S=1 and padding P=1, the size of the output feature map O is the same as the size of the input image W.

[0036] It should be noted here that in order to ensure that the size of the output image is the same as the size of the input image, the set stride and padding parameters can be adjusted accordingly according to actual conditions, and are not specifically limited in this application.

[0037] Furthermore, the PReLU excitation function is used to convert the features of 32 cross-sectional slice images of size W into nonlinear features.

[0038] The cross-sectional slice image is then downsampled to W / 2 (64×64) through a maximum pooling layer, and then the steps of the first convolution unit are repeated using the second convolution unit to further increase the number of feature maps while reducing the image size, resulting in 64 feature images of size W / 2 (64×64), thereby achieving hierarchical extraction of low-order and high-order convolution features.

[0039] Similarly, the above steps are repeated for the third convolution unit to obtain 128 feature images of size W / 4 (32×32).

[0040] Finally, in the encoding stage, 256 feature images of size W / 8 (16×16) are generated for the subsequent decoding process.

[0041] The features derived from the above encoding path also need to be decoded by the decoding path. Similarly, the same three convolutional units as the encoding path are used. For each convolutional unit of the decoding path, it is connected to the corresponding unit in the encoding path through a skip connection, which helps the decoding path to better recover spatial detailed information by reusing the feature map.

[0042] Specifically, the 256 feature images of size W / 8 (16×16) obtained in the above encoding path are upsampled to reduce the number of feature images while enlarging the image size, thereby obtaining 128 feature images of size W / 4 (32×32).

[0043] It should be pointed out that in the above steps, a transposed convolution layer is used instead of direct upsampling to achieve a transformation in the opposite direction of the normal convolution; After the above steps, 32 feature images of size W (128×128) are finally generated. The generated 32 feature images are convolved again to obtain 3 multi-class segmentation images of size W (128×128).

[0044] The final generated multi-class segmentation result has the same dimensions as the input cross-sectional slice image, where the categories include background, lumen, and vessel wall; After the above process, given the input two-dimensional cross-sectional slice image, a multi-class image in the form of a matrix of size (128×128×3) is output, which is characterized by the background channel, lumen channel and vascular wall channel respectively; the final activation function is set to softmax to ensure that the sum of the three channels is 1 in all pixels.

[0045] The multi-classification segmentation results are stacked and fused into the final effective segmentation result, such as Figure 3The following are examples of input images, multi-class segmentation images, and final valid segmentation images.

[0046] The FCN network architecture model is trained using sample data to obtain a blood vessel segmentation model that can automatically segment blood vessel wall images and complete unsupervised data labeling.

[0047] Specifically, first, more vascular branch slices are included in the culture sample to enable the program to better learn the vascular wall morphology; and adjacent cross-sectional slices from the same vascular segment are used as samples to provide additional regularization for the segmentation results, wherein the adjacent range of the same vascular segment is set to 0.6mm-1mm. Preferably, slices within the range of 0.8mm are selected for subsequent calculations; Secondly, the dice coefficient was used to evaluate the overlap rate between the automatic segmentation and manual results. The higher the overlap rate, the closer the automatic segmentation results were to the manual results.

[0048] Among them, the dice coefficient Dice is defined as:

[0049] Wherein, A represents the pixel block statistics of the blood vessel wall segmented by the automatic segmentation image, and M represents the pixel block statistics of the blood vessel wall segmented by the manual segmentation image; Furthermore, a soft Dice coefficient is defined:

[0050] Where A is the probability map, A and M are reshaped into one-dimensional vectors, and the range of Dice is [0,1], where 1 indicates the maximum similarity between the automatic segmentation result and the manual result.

[0051] Dice is calculated separately for the lumen and the vessel wall with equal importance, and the results are not dominated by high-voxel results. In addition, in order to accommodate multi-class results, the loss function L is defined as:

[0052] It should be noted that, because the background channel is the corresponding channel of the lumen channel and the vascular wall channel, the background channel is not calculated when calculating the loss function.

[0053] The present disclosure provides a blood vessel segmentation method, comprising: obtaining a blood vessel wall image of a blood vessel to be segmented; slicing the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; segmenting the cross-sectional slice image based on a pre-trained image segmentation model, and segmenting the blood vessel wall from the cross-sectional slice image. The present disclosure performs automated analysis of blood vessels through a fully convolutional neural network, thereby improving the efficiency and accuracy of interpreting magnetic resonance blood vessel wall images.

[0054] <Device Example> Figure 4 FIG. 2 is a schematic block diagram of a blood vessel segmentation apparatus according to an embodiment of the present disclosure. Figure 4 As shown, the device comprises: An acquisition module 100 is used to acquire a blood vessel wall image of a blood vessel to be segmented; The processing module 101 is used to slice the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; The segmentation module 102 is used to segment the cross-sectional slice image based on a pre-trained image segmentation model, and segment the blood vessel wall from the cross-sectional slice image.

[0055] <Equipment Embodiment> Figure 5 FIG. 2 is a schematic block diagram of a blood vessel segmentation device according to an embodiment of the present disclosure. Figure 5 As shown, the blood vessel segmentation device includes: a processor 201 and a memory 202 for storing executable instructions of the processor 201. The processor 201 is configured to implement any of the above-mentioned blood vessel segmentation methods when executing the executable instructions.

[0056] It should be noted that The memory 202 is a computer-readable storage medium that can be used to store software programs, computer executable programs, and various modules.

[0057] The input device 203 may be used to receive input numbers or signals. The signals may be key signals related to user settings and function control of the device / terminal / server. The output device 204 may include display devices such as display screens.

[0058] <Storage Medium Embodiment> According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is further provided, on which computer program instructions are stored. When the computer program instructions are executed by the processor 201, any of the above-mentioned blood vessel segmentation methods is implemented.

[0059] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A blood vessel segmentation method, characterized in that: include: Acquire a blood vessel wall image of the blood vessel to be segmented; Slicing the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; The cross-sectional slice image is segmented based on a pre-trained image segmentation model, and a blood vessel wall is segmented from the cross-sectional slice image.

2. The method according to claim 1, characterized in that When the vascular wall image is sliced, it includes: acquiring a blood vessel segment determined from the blood vessel wall image; The blood vessel segment is straightened to reconstruct continuous cross-sectional slices of a preset thickness as the cross-sectional slice image.

3. The method according to claim 1, characterized in that When segmenting cross-sectional slice images, it is based on the pre-trained FCN model.

4. The method according to claim 1, characterized in that Segmenting the cross-sectional slice image based on a pre-trained image segmentation model, and segmenting the blood vessel wall from the cross-sectional slice image, and also segmenting at least one of the background and the lumen; Wherein, the vessel wall, the background and the lumen are characterized by different channels.

5. The method according to claim 1, characterized in that The slicing process is performed on the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel, comprising: Selecting the blood vessel segment according to the region of interest; The selected blood vessel segment is sliced ​​to generate continuous cross-sectional slice images.

6. The method according to claim 3, characterized in that When segmenting cross-sectional slice images based on the pre-trained FCN model, it includes: Use adaptive histogram equalization to process cross-sectional slice images; Convolving the processed cross-sectional slice image with the filter to generate a plurality of feature images corresponding to the number of filters; The multiple feature images are upsampled to generate multiple types of segmented images, and the multiple types of segmented images are fused to obtain a final segmentation result.

7. A blood vessel segmentation device, characterized in that: include: Acquisition module, processing module, segmentation module; The acquisition module is configured to acquire a blood vessel wall image of the blood vessel to be segmented; The processing module is configured to perform slicing processing on the blood vessel wall image to obtain a cross-sectional slice image of the blood vessel; The segmentation module is configured to segment the cross-sectional slice image based on a pre-trained image segmentation model, and segment the blood vessel wall from the cross-sectional slice image.

8. The device according to claim 7, characterized in that The segmentation module also includes: a fusion module; The fusion module is configured to upsample the multiple feature images to generate multiple types of segmented images, and fuse the multiple types of segmented images to obtain a final segmentation result.

9. A blood vessel segmentation device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the method described in any one of claims 1 to 6 when executing the executable instructions.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images