Ultrasonic blood vessel image segmentation method

Abstract

The invention provides an ultrasonic blood vessel image segmentation method under topological continuity constraint. By introducing the blood vessel topological continuity prior, a blood vessel fine segmentation result with higher topological fidelity is obtained through the steps of background removal, quasi flow field estimation, blood vessel coarse segmentation and blood vessel fine topological repair in sequence, and the blood vessel segmentation performance can be remarkably improved. According to the invention, a more continuous three-dimensional fine structure of the blood vessel can be reconstructed in a short time, and accurate measurement of the inner diameter of the blood vessel is realized. The method is beneficial to promoting accurate evaluation and quantification of vascular morphological characteristics, and lays a foundation for development of computer-aided diagnosis and prognosis evaluation means.

Description

Technical Field

[0001] The present invention relates to the field of biomedical image processing, and in particular to an ultrasonic vascular image segmentation method. Background Art

[0002] The morphological characteristics of microvessels are associated with many neurovascular diseases, especially in the early stages of the disease. In order to quantify these morphological characteristics, a segmentation algorithm is needed to identify the vascular region from the vascular image. Therefore, the automated vascular segmentation algorithm can assist doctors in evaluating the morphological characteristics of microvessels, which is of great significance for computer-aided diagnosis and prognosis assessment.

[0003] In recent years, ultrafast Doppler ultrasound imaging has developed rapidly and is very suitable for imaging microvessels. Compared with traditional B-mode ultrasound and Doppler ultrasound, ultrafast Doppler ultrasound imaging significantly improves the signal-to-noise ratio and the sensitivity of small blood vessel detection. However, it is still not easy to segment blood vessels in ultrafast Doppler ultrasound imaging. First, the brightness of blood vessels and background at different locations in ultrasound images is not uniform, which means that a simple global thresholding strategy is not feasible; second, the boundaries of blood vessels in ultrasound images are often fuzzy, which makes it difficult for the algorithm to determine the exact location of the blood vessel boundary and measure the inner diameter of the blood vessel; finally, small blood vessels may be insufficiently perfused in a short imaging time, resulting in vascular rupture in the ultrasound imaging results. Therefore, there are still many challenges in accurate and reliable ultrasound vascular image segmentation, which hinders the quantitative evaluation of vascular morphological characteristics. It is imperative to develop accurate, reliable, and high-topological fidelity blood vessel segmentation algorithms. Summary of the invention

[0004] In order to solve the common processing problems in ultrasonic vascular images, such as uneven brightness, blurred vascular boundaries, and easy breakage of small blood vessels, the present invention proposes an ultrasonic vascular image segmentation method under topological continuity constraints. By introducing the vascular topological continuity prior, the present invention sequentially obtains a more accurate and reliable vascular fine segmentation result with higher topological fidelity through the steps of background removal, quasi-flow field estimation, vascular coarse segmentation, and vascular fine topological repair. The present invention can reconstruct a more continuous vascular three-dimensional fine structure in a short time and accurately estimate the inner diameter of the blood vessel.

[0005] The technical solution of the present invention is as follows: The ultrasonic vascular image segmentation algorithm described in the present invention includes the steps of background removal, quasi-flow field estimation, rough vascular segmentation, and fine vascular topology restoration, and is used to identify vascular regions from ultrasonic vascular images and reconstruct the three-dimensional fine structure of the vessels.

[0006] In the background removal step, the original ultrasound B-mode image I0 is preprocessed using a typical background removal algorithm to obtain a preprocessed ultrasound vascular image I. When removing the background, conventional algorithms such as the averaging method, principal component analysis method, robust principal component analysis method, and rolling ball method can be used to estimate the background of the ultrasound vascular image, and then the background is subtracted from the original image I0 to obtain the preprocessed ultrasound vascular image I. The background removal step can effectively highlight the foreground of the blood vessels and remove interference to facilitate subsequent processing.

[0007] In the quasi-flow field estimation step, the local axial direction and local radial direction of the blood vessel at each pixel are obtained by weighted principal component analysis for the preprocessed ultrasonic vascular image I, forming a quasi-flow field F. Considering that the orientation of the blood vessel can be obtained from the relative position relationship of the brighter pixels in I, the present invention considers using weighted principal component analysis to obtain the principal component that best matches the arrangement of the bright pixels as the estimation result of the axial direction of the blood vessel. The specific steps are: first, pixels near pixel p whose distance does not exceed the preset parameter r are taken as neighborhood pixels, and the grayscale values ​​corresponding to the neighborhood pixels are used as weights to perform weighted principal component analysis, and the first principal component obtained is used as the local axial estimation result of point p, and the second principal component is used as the local radial estimation result of point p.

[0008] In the blood vessel rough segmentation step, the preprocessed ultrasonic blood vessel image I is combined with the estimated quasi-flow field F, and the rough segmentation result R is obtained using the local thresholding method with topological constraints. Considering that the brightness of the ultrasonic blood vessel image I is not uniform, the present invention uses the local thresholding method for rough segmentation; in addition, considering that the blood vessel boundary is often fuzzy, the present invention also uses the axial direction in the quasi-flow field F to set the local threshold to ensure that the blood vessel is not too wide or too thin. The specific steps are: based on the estimation results of the local axial and radial directions of each pixel p, the local grayscale distribution Ω(p) of the pixel p along the radial direction is calculated; further identify the maximum value I in the local grayscale distribution of the pixel p max (p) and the minimum value I min (p), and according to the preset proportionality coefficient θ, calculate the local threshold T(p) = θ×I max (p) + (1–θ)×I min (p), and then the coarse segmentation result R = {p; I (p) ≥ T (p)} is obtained. The proportional coefficient θ can be adjusted according to the point spread function of the imaging system, and can generally be set near 0.5 to ensure that the half-maximum full width of the blood vessel width and the axial brightness change are consistent.

[0009] In the blood vessel fine topology repair step, the rough segmentation result R is repaired using the topological continuity constraint to obtain a more continuous fine blood vessel segmentation result R *. Since there may be broken blood vessels in the rough segmentation results of ultrasonic blood vessels, considering the continuity of the real blood vessel structure, the present invention introduces topological continuity prior to repair broken blood vessels. The specific steps are: first remove the interference area whose area does not exceed the preset threshold σ; then use the remaining broken blood vessel areas as nodes to construct a graph G, and the weight of each edge (i, j) in the graph G is determined by the generalized distance between area i and area j. The generalized distance is defined by the user. Preferably, the generalized distance is defined as the shortest straight-line distance between the pixels contained in the two areas, that is, min {‖p1-p2‖; p1∈area i, p2∈area j}. Blood vessel repair hopes to select edges with smaller weights (the two blood vessel areas connected by these edges are closer and are more likely to be blood vessels that were originally connected) and repair them. To achieve this goal, some edges are retained in G and the remaining edges are deleted, so that all nodes are connected and the weights of the edges are as small as possible. Preferably, the minimum spanning tree of the graph or the shortest path tree starting from a certain node can be obtained, and only the edges on the tree are left in G. Finally, based on each edge (i, j) retained in the graph, the image pixels between the vascular region i and the vascular region j are also added to R as vascular pixels using the regional interpolation method, and finally the locally repaired vascular fine structure R is obtained. * .

[0010] The specific steps of the regional interpolation method used in the vascular fine topology repair step are as follows: First, the axial skeletons of the vascular regions i and j are obtained by using the skeletonization algorithm, that is, B i and B j ; Then find B i and B j Between the two neighboring ends E i and E j , and calculate the vascular area in E i and E j The radial lengths at i Department and E j The local width at E is then connected by a straight line segment i and E j , the width of the straight line segment is from E i The local width at E changes slowly to j The local width at , thereby connecting vascular region i and vascular region j.

[0011] The present invention has the following advantages: By introducing topological continuity prior, the present invention can effectively deal with common problems in ultrasonic vascular image processing, such as uneven brightness, blurred vascular boundaries, and easy breakage of small blood vessels, and achieve more accurate and fine blood vessel segmentation with higher topological fidelity. The rough blood vessel segmentation of the present invention takes into account the local topological structure of the blood vessels through flow field information, and the fine blood vessel topological repair takes into account the global topological continuity constraints of the blood vessels; the combination of local topology and global topology can significantly improve the performance of blood vessel segmentation, making the subsequent evaluation and quantification of blood vessel morphological characteristics more accurate and reliable, laying the foundation for the future development of ultrasound-based computer-aided diagnosis and prognosis evaluation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a flow chart of the ultrasonic vascular image segmentation algorithm of the present invention; Figure 2 This is a comparison diagram of ultrasonic vascular segmentation results, from left to right are respectively an input ultrasonic vascular image, a vascular segmentation result using the traditional local Otsu method, and a vascular segmentation result obtained using the algorithm of the present invention. DETAILED DESCRIPTION

[0013] The present invention is further described below by means of specific examples to provide a better understanding of the present invention, but the present invention is not limited thereto.

[0014] like Figure 1 As shown, the present invention comprises the following steps:

[0015] S1, background removal step: removing the background of the input original ultrasonic vascular image I0 to obtain the preprocessed ultrasonic vascular image I; when removing the background, conventional algorithms such as averaging method, principal component analysis method, robust principal component analysis method, rolling ball method, etc. can be used to estimate the background of the ultrasonic vascular image, and then the background is subtracted from the original image I0 to obtain the preprocessed ultrasonic vascular image I.

[0016] S2, quasi-flow field estimation step: The local axial direction and local radial direction of the blood vessel at each pixel are obtained by using local weighted principal component analysis on the preprocessed ultrasonic blood vessel image I to form a quasi-flow field F. The preset parameter r can be adjusted with reference to the width of the blood vessel and can be set to 5 to 30 pixels.

[0017] S3, blood vessel rough segmentation step: The preprocessed ultrasound blood vessel image I and the estimated quasi-flow field F are subjected to a topologically constrained local thresholding method to obtain a rough segmentation result R. The specific steps are as follows: S31, calculating the local grayscale distribution Ω(p) of pixel p along the radial direction, where the radial search range needs to be guaranteed to exceed the width of the blood vessel, which can be set to 5-30 pixels; S32, obtain the maximum value I of the grayscale distribution according to the local grayscale distribution max(p) and minimum value I min (p), the local threshold T(p) is obtained by weighting it in proportion θ, and binarization is performed based on T(p), that is, the set of pixels whose brightness exceeds the local threshold {p; I(p) ≥ T(p)} is taken as the coarse segmentation result R; θ can be set near 1 / 2.

[0018] S4: The blood vessel fine topology repair step repairs the rough segmentation result to obtain a more continuous fine blood vessel segmentation result R * The specific steps are as follows: S41, discarding interference regions in R whose area does not exceed a preset threshold σ. σ can be adjusted according to the actual size of the blood vessel and the noise level, and can be selected to be a number within 20; S42, constructing a graph G using the remaining blood vessel regions as nodes, wherein the edges of the graph G are determined by the generalized distance between the two connected regions; preferably, the generalized distance is defined as the shortest straight-line distance between the pixels contained in the two regions, i.e., min {|p1-p2|; p1∈region i, p2∈region j}; S43, select a part of the edges of the graph G and keep them and discard the rest of the edges so that all nodes are connected, and optionally, solve the minimum spanning tree or the shortest path tree and keep the edges in the tree; the minimum spanning tree can be solved by Kruskal algorithm or Prim algorithm, and the shortest path tree can be solved by Bellman-Ford or Dijkstra algorithm; S44. Use regional interpolation to repair the edge retained in graph G, that is, draw a straight line between the two regions connected by this edge to connect the proximal ends of the two regions, and add the pixels on the straight line to the blood vessel segmentation result, and finally obtain a more continuous and fine blood vessel segmentation result R * .

[0019] Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the art can use the above disclosed methods and technical contents to make many possible changes and modifications to the technical solution of the present invention without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of protection of the technical solution of the present invention.

Claims

1. A method for segmenting an ultrasonic vascular image, characterized in that: The blood vessel segmentation and repair with topological continuity constraints consists of the following steps: (1) The original ultrasound B-mode image I0 is preprocessed using a typical background removal algorithm to obtain a preprocessed ultrasound vascular image I; (2) Using the quasi-flow field estimation method on I, the local axial direction and local radial direction of the blood vessel at each pixel are obtained to form the quasi-flow field F; (3) According to the radial direction of each pixel in the quasi-flow field, a rough segmentation result R is obtained using a blood vessel rough segmentation method; specifically, the local grayscale distribution Ω of each pixel along the radial direction is extracted, and the threshold T for distinguishing blood vessels from the background is calculated. I is binarized using T to obtain the blood vessel rough segmentation result R; (4) Perform fine topological repair on the rough segmentation result R of the blood vessels to obtain the fine structure R of the blood vessels * .

2. The quasi-flow field estimation method according to claim 1, characterized in that: For the ultrasonic vascular image I, the local axial and radial directions are estimated for each pixel p; the specific steps are: first, pixels near the pixel p whose distance does not exceed the preset parameter r are taken as neighborhood pixels, and the grayscale values ​​corresponding to the neighborhood pixels are used as weights to perform weighted principal component analysis, and the first principal component of the result is taken as the local axial estimation result of point p, and the second principal component is taken as the local radial estimation result of point p.

3. The blood vessel rough segmentation method according to claim 1, characterized in that: The specific steps are: based on the estimation results of the local axial and radial directions of each pixel p, calculate the local grayscale distribution Ω(p) of the pixel p along the radial direction; further identify the maximum value I in the local grayscale distribution of the pixel p max (p) and the minimum value I min (p), and according to the preset proportionality coefficient θ, calculate the local threshold T(p)=θ×I max (p)+ (1–θ)×I min (p), and then the coarse segmentation result R = {p; I(p) ≥ T(p)} is obtained.

4. The method for repairing blood vessel fine topology according to claim 1, characterized in that: The coarse blood vessel segmentation result R is subjected to fine blood vessel repair using continuity constraints; the specific steps are as follows: first, the interference area in R whose area does not exceed the preset threshold σ is discarded; then, the broken blood vessel areas that are retained are used as nodes to construct a graph G, and the weight of each edge (i, j) in the graph G is determined by the generalized distance between the blood vessel area i and the blood vessel area j; wherein the generalized distance is user-defined, and preferably, the generalized distance is defined as the shortest straight-line distance between the pixels contained in the two areas; then, a part of the edges in the graph G are selected to be retained, and the remaining edges are deleted, so that all nodes are connected, and preferably, the edges in the shortest path tree or the minimum spanning tree are used as the retained edges; finally, based on each edge (i, j) retained in the graph, the image pixels between the blood vessel area i and the blood vessel area j are also added to R as blood vessel pixels using the regional interpolation method, and finally the fine blood vessel structure R after local repair is obtained * .

5. The regional interpolation method according to claim 4, characterized in that: The specific steps are as follows: First, the skeletonization algorithm is used to obtain the axial skeletons of the vascular regions i and j, that is, B i and B j ; Then find B i and B j Between the two neighboring ends E i and E j , and calculate the vascular area in E i and E j The radial lengths at i Department and E j The local width at E is then connected by a straight line segment i and E j , the width of the straight line segment is from E i The local width at E changes slowly to j The local width at , thereby connecting vascular region i and vascular region j.

Images