A method for semi-automatic extraction and display of blood vessel classification

Abstract

The present invention belongs to the field of digital medical technology, and discloses a method for semi-automatic extraction and display of vascular grading: S1, image enhancement and denoising are performed on the original CTA image in sequence; then the bone, skin, soft tissue and muscle models are reconstructed; S2, the hierarchical main vascular network is automatically extracted in the original CTA image, and the vascular branch information is recorded on the nodes of the vascular binary tree; S3, the perforating blood vessels are obtained on the original CTA image, and the perforating blood vessel paths are placed in the sub-nodes corresponding to the upper blood vessels in the main network of the blood vessels, and finally a complete vascular grading network including the perforating blood vessels is formed; S4, a vascular network projection capable of distinguishing the depth is formed on the skin model obtained in S1. Compared with the traditional method, the method of the present invention can more intuitively and accurately obtain the subordinate information of the perforating blood vessels and the depth information of the blood vessels from the skin, thereby helping doctors to better plan surgical plans.

Description

Technical Field

[0001] The invention belongs to the field of digital medical technology, and in particular relates to a method and system for semi-automatic extraction and display of blood vessel classification. Background Art

[0002] Currently, during the anterolateral thigh free flap transplantation, Doppler ultrasound is generally used to evaluate and locate the blood vessels before the operation. Based on experience, the doctor deflects the ultrasound probe to the direction consistent with the blood flow direction and keeps the angle with the sound beam less than 60° to explore the blood vessels point by point. Then, the direction of the blood vessels is roughly estimated based on the information obtained at each non-continuous point. Finally, the positioning marks are drawn on the body surface and recorded. This method has the following disadvantages:

[0003] (1) The subordinate relationship of the perforating vessels is difficult to determine. During surgery, the main trunk and descending branch of the lateral femoral artery must be freed in retrograde fashion, the fascia lata must be preserved, and the vessels must be ligated after freeing the vascular pedicle to the required length. Unclear knowledge of the subordinate relationship of the vessels will increase the difficulty of the surgery.

[0004] (2) Although the direction of the perforating vessels can be roughly determined, it is difficult to intuitively obtain information about the depth of the vessels from the skin and the overall morphology of the entire vascular network. Perforating vessels have the characteristics of unstable locations and many variations, which makes it difficult to plan and determine the surgical plan. Summary of the invention

[0005] The purpose of the present invention is to provide a method for semi-automatic extraction and display of blood vessel classification to solve the problems existing in the prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for semi-automatic extraction and display of blood vessel classification specifically comprises the following steps:

[0008] S1, performing image enhancement processing and denoising processing on the original CTA image in sequence; then reconstructing the bone model, reconstructing the skin model, and reconstructing the soft tissue and muscle model based on the denoising results;

[0009] S2, automatically extracting hierarchical trunk vascular networks in the original CTA images and recording vascular branch information at the nodes of the vascular binary tree;

[0010] S3, select a layer with better imaging on the original CTA image, and select the path feature points of the perforating vessels, which include the number of layers of the original CTA image and the X and Y coordinates of each layer. Connect these feature points to obtain the perforating vessels, and put the perforating vessel paths into the sub-nodes of the corresponding upper-layer vessels in the main network of the vessels obtained in S2, and finally form a complete vascular hierarchical network including the perforating vessels;

[0011] S4, for the area that needs to be processed in the vascular grading network, uniformly sample the vascular model along the vascular centerline to obtain a series of sampling points, orthogonally project the sampling points at the main camera's line of sight, use the ray method to obtain the position of the projection point on the skin, calculate the distance between the sampling point and its projection point on the skin, and draw the projection points of the vascular sampling on the skin one by one. According to the distance, uniformly interpolate the specified color in the RGB color space to obtain different colors, connect the projection points one by one with the color corresponding to the depth of the projection point, and finally form a vascular network projection that can distinguish the depth on the skin model obtained in S1.

[0012] Furthermore, in S1, the original CTA image is sequentially subjected to image enhancement processing using an improved adaptive histogram equalization algorithm and denoising processing using a downsampling filter.

[0013] Furthermore, in S1, the process of performing image enhancement processing on the original CTA image using the improved AHE algorithm is as follows:

[0014] S11, dividing the original CTA image into blocks and calculating the cdf value of each block;

[0015] S12, for non-boundary pixels in the original CTA image, find four blocks adjacent to the pixel, and record the cdf values ​​of the four blocks as: ful(D), fur(D), fbl(D), fbr(D); Δx and Δy are the ratios of the distance from the current pixel to the geometric center of the upper left corner of the block to the length and width of the block respectively; then perform bilinear interpolation to obtain the cdf value of the current pixel;

[0016] S13, for the pixel points located at the vertices in the boundary area of ​​the original CTA image, the cdf of the block in which it is located is used as its own cdf; for the pixel points located at non-vertices in the boundary area of ​​the original CTA image, the cdf mapping values ​​of the two adjacent blocks in which it is located are used for single linear interpolation to obtain its own cdf, thereby completing the image enhancement processing of the improved AHE algorithm.

[0017] Furthermore, in S12, the bilinear interpolation formula is as follows:

[0018] f(D)=(1-Δx)(1-Δy)ful(D)+Δxfbl(D)+Δyful(D)+ΔxΔyfbr(D).

[0019] Furthermore, in S1, a voxel-based three-dimensional surface reconstruction algorithm is used to reconstruct the bone model, a marching cube algorithm is used to reconstruct the skin model, and volume reconstruction is used to reconstruct the soft tissue and muscle models.

[0020] Furthermore, S2 specifically includes the following sub-steps:

[0021] S21, selecting a seed point in the original CTA image;

[0022] S22, performing a growth operation based on the seed point to obtain a main network of blood vessels, specifically including the following operations:

[0023] S221, based on the obtained seed points, the eight-neighborhood growing algorithm is used layer by layer on the original CTA image to screen out the connected domains whose Hu values ​​match the blood vessels; then, each screened connected domain is subjected to Hough transform to determine whether the region is a circle, and if so, the coordinate position and radius of the circle center are calculated; otherwise, it is determined to be a noise point and the connected domain is deleted;

[0024] S222, sequentially traverse all layers of the results obtained in S221, for the current layer:

[0025] S2221, if a circular connected domain in a layer above the current layer has no corresponding circular connected domain in the current layer, the blood vessel branch where the circular connected domain is located ends in the current layer, and other circular connected regions in the current layer continue to be processed;

[0026] S2222, if the number of circular connected domains corresponding to a circular connected domain in a layer above the current layer in the current layer is 1, then the blood vessel where the corresponding circular connected domain in the current layer is located continues to grow toward the layer below the current layer;

[0027] S2223, if a circular connected domain in a layer above the current layer corresponds to multiple circular connected domains in the current layer, it is considered that the blood vessel has branches, and the branch information is recorded on the node of the blood vessel binary tree, and then the CTA image is grown along each blood vessel branch to the next layer of the current layer;

[0028] The above process is repeated until all layers of the original CTA image are processed, or each vascular branch grows until the number of circular connected domains corresponding to the current layer is 0, the growth operation ends, and the main network of blood vessels is obtained.

[0029] Furthermore, the following operation is performed after S2: the vascular fault in the vascular trunk network obtained by S2 is used as a new seed point to repeat the growth operation in S2 to obtain a branch network corresponding to the breakpoint, and the branch network is added to the vascular trunk network obtained by S2 to obtain an updated vascular trunk network.

[0030] Compared with the prior art, the present invention uses the patient's CTA data as a basis before surgery, and uses voxel reconstruction, surface rendering reconstruction, volume reconstruction and other reconstruction methods for different parts to reconstruct the patient's leg bones, skin, and soft tissues, and then extracts the main blood vessels in a semi-automatic way, and then reconstructs and completes the microvascular network including perforating blood vessels, and then expresses the relationship between the blood vessels and the surrounding skin in a three-dimensional way, and uses a projection algorithm to project the vascular network to the surface of the skin layer of the model, and then samples along the center line of the blood vessel, and uses the main camera matrix as the transformation matrix to transform the coordinates of the sampling points, calculates the depth information from the blood vessel to the skin, and finally uses different colors to represent different depths, and draws the vascular network on the skin model, so that the doctor can intuitively see the subordinate relationship of the blood vessels, the overall morphology of the vascular network and the depth change of the blood vessels from the skin along the blood vessel. Compared with the traditional method of using Doppler exploration, the method of the present invention can more intuitively and accurately obtain the subordinate information of the perforating blood vessels and the depth information of the blood vessels from the skin, thereby helping doctors to better plan surgical plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Improved sampling interpolation in AHE algorithm;

[0032] Figure 2 This is a schematic diagram of the binary tree structure of blood vessels;

[0033] Figure 3 This is a screenshot of the obtained vascular network;

[0034] Figure 4 Screenshot of the vascular network projected onto the surface of the skin model based on the different depths of the blood vessels from the skin. DETAILED DESCRIPTION

[0035] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] Technical terms involved in this invention:

[0037] CTA data: CTA angiography data.

[0038] cdf: cumulative distribution function.

[0039] Hu value: Hounsfield Units, a unit used to describe tissue density in computed tomography imaging.

[0040] The method for semi-automatic extraction and display of blood vessel classification provided by the present invention specifically comprises the following steps:

[0041] S1, the original CTA images are enhanced using the improved adaptive histogram equalization (AHE) algorithm and denoised using downsampling filtering; then, for the denoising results, the bone model is reconstructed using a voxel-based 3D surface reconstruction algorithm, the skin model is reconstructed using a marching cube algorithm, and the soft tissue and muscle models are reconstructed using volume reconstruction.

[0042] The bone model, skin model, soft tissue and muscle model obtained in this step can be displayed independently or fused in the subsequent research of the present invention. The skin model is used in S4 to obtain the vascular network projection, and the bone model, soft tissue and muscle model obtained are used in the subsequent research of the method of the present invention.

[0043] As a preferred implementation of the present invention, in S1, the process of performing image enhancement processing on the original CTA image using the improved AHE algorithm is as follows:

[0044] S11, dividing the original CTA image into blocks, and calculating the cdf value of each block.

[0045] S12, for non-boundary pixels in the original CTA image (boundary see Figure 1 The red and green areas in the middle), find the four blocks adjacent to the pixel (see Figure 1 The square with a solid black border in the middle right half of the figure is the block), and the cdf values ​​of these four blocks are recorded as: ful(D), fur(D), fbl(D), fbr(D); Δx and Δy are the distance from the current pixel to the geometric center of the upper left corner of the block ( Figure 1 The ratio of the distance from the black pixel in the upper left corner of the left half of the image to the length and width of the block; then bilinear interpolation is performed to obtain the cdf value of the current pixel. The bilinear interpolation formula is as follows:

[0046] f(D)=(1-Δx)(1-Δy)ful(D)+Δxfbl(D)+Δyful(D)+ΔxΔyfbr(D)

[0047] (Note: The influence of the cdf of the four blocks on the mapping of the blue pixel here is determined by the distance from the geometric center of the four blocks to the pixel. The closer the distance, the greater the influence, and vice versa.)

[0048] S13, for the pixel points located at the vertices in the boundary area of ​​the original CTA image (pixels in the red area), the cdf of the block in which it is located is used as its own cdf; for the pixel points located at non-vertices in the boundary area of ​​the original CTA image (pixels in the green area), the cdf mapping values ​​of the two adjacent blocks in which it is located are used for single linear interpolation to obtain its own cdf, thereby completing the image enhancement processing of the improved AHE algorithm.

[0049] In the above improved AHE algorithm, interpolation is used to accelerate histogram equalization, which can optimize the block effect of the image after AHE equalization.

[0050] S2, automatically extracting the hierarchical trunk vascular network in the original CTA image, specifically including the following sub-steps:

[0051] S21, selecting a seed point in the original CTA image;

[0052] S22, performing a growth operation based on the seed points to obtain a main network of blood vessels. Specifically, the following operations are included:

[0053] S221, based on the obtained seed points, the eight-neighborhood growing algorithm is used layer by layer on the original CTA image to screen out the connected domains whose Hu values ​​match the blood vessels; then, each screened connected domain is subjected to Hough transform to determine whether the area is a circle, and if so, its center coordinate position and radius are calculated; otherwise, it is determined to be a noise point and the connected domain is deleted.

[0054] S222, sequentially traverse all layers of the results obtained in S221, for the current layer:

[0055] S2221, if a circular connected domain in a layer above the current layer has no corresponding circular connected domain in the current layer, the blood vessel branch where the circular connected domain is located ends in the current layer, and other circular connected regions in the current layer continue to be processed.

[0056] S2222: If the number of circular connected domains corresponding to a circular connected domain in a layer above the current layer in the current layer is 1, the blood vessel where the corresponding circular connected domain in the current layer is located continues to grow toward the layer below the current layer.

[0057] S2223, if a circular connected domain in the previous layer of the current layer corresponds to multiple circular connected domains in the current layer, it is considered that the blood vessel has branched, and the branch information is recorded in the node of the blood vessel binary tree (see the binary tree diagram Figure 2 ), and then grow along each vascular branch to the next layer of the current layer of the CTA image. The branch information includes the starting layer index of this segment of blood vessels, the ending layer index of this segment of blood vessels, the coordinate list of the blood vessel centerline, the blood vessel radius list, etc.

[0058] The above process is repeated until all layers of the original CTA image are processed, or each vascular branch grows until the number of circular connected domains corresponding to the current layer is 0, the growth operation ends, and the main network of blood vessels is obtained.

[0059] As a preferred embodiment of the present invention, the following operations are performed after S2: the CTA raw data may contain noise points and the blood vessel development cannot be displayed continuously due to unclear shooting. Therefore, after the automatic extraction of the trunk network is completed, it is necessary to use tools to quickly manually correct the erroneous blood vessels caused by noise points and faults. The use of traditional reconstruction software requires drawing blood vessels layer by layer, which takes a long time. Therefore, in the present invention, a semi-automatic method combining manual operation with an algorithm is used to repair the erroneous blood vessels caused by noise points and faults. The specific steps are: the operator observes the blood vessel fault in the trunk network of the blood vessel obtained by S2, determines the blood vessel position at the fault, and uses this position as a new seed point to repeat the growth operation in S2, thereby obtaining a branch network corresponding to the breakpoint, and adding the branch network to the trunk network of the blood vessel obtained by S2 to obtain an updated trunk network of the blood vessel.

[0060] S3. Since the perforating vessels are thin, the original CTA data is easily affected by the layer thickness and angiography, resulting in many layers not being able to be coherently developed or being poorly developed. It is difficult to automatically reconstruct such vessels using the growth algorithm. Even if they are barely reconstructed, they cannot form continuous and smooth vessels. Therefore, the present invention adopts the binary tree structure that has been extracted and automatically segmented in the above steps, manually selects the layers with better imaging on the original CTA image, and selects the path feature points of the perforating vessels. These feature points include the number of layers of the original CTA image, the X and Y coordinates of each layer, connect these feature points to obtain the perforating vessels, and put the perforating vessel paths into the child nodes corresponding to the upper layer vessels in the main network of the vessels, and finally form a complete vascular hierarchical network including the perforating vessels (see the extraction effect). Figure 3 ).

[0061] S4, for the area that needs to be processed in the vascular hierarchical network, uniformly sample the vascular model along the vascular centerline to obtain a series of sampling points, orthogonally project the sampling points at the main camera's line of sight, use the ray method to obtain the position of the projection point on the skin, calculate the distance between the sampling point and its projection point on the skin, and draw the projection points of the vascular sampling on the skin one by one. According to the distance, uniformly interpolate a specified color (such as red) in the RGB color space to obtain different colors, connect the projection points one by one with the color corresponding to the depth of the projection point, and finally form a vascular network projection that can distinguish the depth on the skin model obtained in S1 (see the screenshot of the effect). Figure 4 ).

[0062] In summary, compared with the traditional method of using Doppler exploration, the method of the present invention can more intuitively and accurately obtain the subordinate information of the perforating blood vessels and the depth information of the blood vessels from the skin, thereby helping doctors to better plan surgical plans.

Claims

1. A method for semi-automatic extraction and display of blood vessel classification, characterized in that: The specific steps include: S1, performing image enhancement processing and denoising processing on the original CTA image in sequence; then reconstructing the bone model, reconstructing the skin model, and reconstructing the soft tissue and muscle model based on the denoising results; S2, automatically extracting the hierarchical trunk vascular network in the original CTA image, and recording the vascular branch information at the nodes of the vascular binary tree; specifically including the following sub-steps: S21, selecting a seed point in the original CTA image; S22, performing a growth operation based on the seed point to obtain a main network of blood vessels, specifically including the following operations: S221, based on the obtained seed points, the eight-neighborhood growing algorithm is used to screen out the connected domains whose Hu values ​​match the blood vessels on the original CTA image layer by layer; then, each screened connected domain is subjected to Hough transform to determine whether the region is a circle, and if so, the coordinate position and radius of the circle center are calculated; otherwise, it is determined to be a noise point and the connected domain is deleted; S222, sequentially traverse all layers of the results obtained in S221, for the current layer: S2221, if a circular connected domain in a layer above the current layer has no corresponding circular connected domain in the current layer, the blood vessel branch where the circular connected domain is located ends in the current layer, and other circular connected regions in the current layer continue to be processed; S2222, if the number of circular connected domains corresponding to a circular connected domain in a layer above the current layer in the current layer is 1, then the blood vessel where the corresponding circular connected domain in the current layer is located continues to grow toward the layer below the current layer; S2223, if a circular connected domain in a layer above the current layer corresponds to multiple circular connected domains in the current layer, it is considered that the blood vessel has branches, and the branch information is recorded on the node of the blood vessel binary tree, and then the CTA image is grown along each blood vessel branch to the next layer of the current layer; Repeat the above process until all layers of the original CTA image are processed, or each vascular branch grows to the point where the number of circular connected domains corresponding to the current layer is 0, the growth operation ends, and the main network of blood vessels is obtained; S3, select a layer with clear imaging on the original CTA image, and select the path feature points of the perforating vessels, which include the number of layers of the original CTA image and the X and Y coordinates of each layer. Connect these feature points to obtain the perforating vessels, and put the perforating vessel paths into the sub-nodes corresponding to the upper-layer vessels in the main network of the vessels obtained in S2, and finally form a complete vascular hierarchical network including the perforating vessels; S4, for the area that needs to be processed in the vascular grading network, uniformly sample the vascular model along the vascular centerline to obtain a series of sampling points, orthogonally project the sampling points at the main camera's line of sight, use the ray method to obtain the position of the projection point on the skin, calculate the distance between the sampling point and its projection point on the skin, and draw the projection points of the vascular sampling on the skin one by one. According to the distance, uniformly interpolate the specified color in the RGB color space to obtain different colors, connect the projection points one by one with the color corresponding to the depth of the projection point, and finally form a vascular network projection that can distinguish the depth on the skin model obtained in S1.

2. The method for semi-automatic extraction and display of blood vessel classification according to claim 1, characterized in that: In S1, the original CTA image is enhanced by using an improved adaptive histogram equalization algorithm, and denoised by using a downsampling filter.

3. The method for semi-automatic extraction and display of blood vessel classification according to claim 2, characterized in that: In S1, the process of image enhancement processing using the improved AHE algorithm on the original CTA image is as follows: S11, dividing the original CTA image into blocks and calculating the cdf value of each block; S12, for non-boundary pixels in the original CTA image, find four blocks adjacent to the pixel, and record the cdf values ​​of the four blocks as: ful(D), fur(D), fbl(D), fbr(D); Δx and Δy are the ratios of the distance from the current pixel to the geometric center of the upper left corner of the block to the length and width of the block respectively; then perform bilinear interpolation to obtain the cdf value of the current pixel; S13, for the pixel point located at the vertex in the boundary area of ​​the original CTA image, the cdf of the block where it is located is used as its own cdf; for the pixel point located at the non-vertex in the boundary area of ​​the original CTA image, the cdf mapping values ​​of the two adjacent blocks where it is located are used for single linear interpolation to obtain its own cdf, thereby completing the image enhancement processing of the improved AHE algorithm.

4. The method for semi-automatic extraction and display of blood vessel classification according to claim 3, characterized in that: In S12, the bilinear interpolation formula is as follows: f(D)=(1-Δx)(1-Δy)ful(D)+Δxfbl(D)+Δyful(D)+ΔxΔyfbr(D).

5. The method for semi-automatic extraction and display of blood vessel classification according to claim 1, characterized in that: In S1, the bone model was reconstructed using a voxel-based 3D surface reconstruction algorithm, the skin model was reconstructed using a marching cubes algorithm, and the soft tissue and muscle models were reconstructed using volume reconstruction.

6. The method for semi-automatic extraction and display of blood vessel classification according to claim 1, characterized in that: After S2, the following operations are performed: the vascular fault in the vascular trunk network obtained in S2 is used as a new seed point to repeat the growth operation in S2 to obtain a branch network corresponding to the breakpoint, and the branch network is added to the vascular trunk network obtained in S2 to obtain an updated vascular trunk network.

Images