A method for cerebral vessel segmentation of non-angiographic images

Abstract

This invention discloses a method for segmenting cerebral blood vessels in non-angiographic images, comprising the following steps: S1, inputting the original image A, i.e., the original non-angiographic image; S2, filtering out air, outputting image B; S3, dissecting the skull, outputting image C; S4, performing preliminary ablation of the cerebral cortex, outputting image D; S5, iteratively ablating the cerebral cortex, outputting image E, i.e., the final cerebral blood vessel image. This invention can dissect the skull and iteratively ablate the cerebral cortex without affecting the vascular components, thereby obtaining the purest and most complete cerebral blood vessel components possible.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, and relates to skull dissection and vascular enhancement during the segmentation process, specifically to a method for cerebral vascular segmentation of non-angiography images. Background Technology

[0002] For angiographic images (MRA, CTA, and DSA), the pixel values ​​of vascular tissue are significantly larger, and threshold segmentation methods can be directly used to obtain images of cerebral blood vessels. However, capturing these types of images requires special technical means, especially for CTA and DSA images, which require the injection of contrast agents into the patient to achieve imaging.

[0003] For cerebral vascular segmentation of non-angiographic images, images with clearly defined vascular features, such as T1W-TFE sequences, are generally selected. In these images, the principal component pixel values ​​of blood vessels are relatively large, but the pixel values ​​of small vessels and vessel edges show a gradient decreasing trend, even becoming indistinguishable from peripheral tissues. It is difficult to determine a suitable segmentation threshold for these images; a low threshold results in segmented blood vessels mixed with other tissue structures, primarily the outer layer of skull muscle tissue and the cerebral cortex (e.g.,...). Figure 1 When the threshold is too high, the segmented vascular structure will be incomplete (e.g., Figure 2 ). Summary of the Invention

[0004] For cerebral vessel segmentation in non-angiographic images, thresholding based solely on pixel dimensions is insufficient to obtain pure and complete cerebral vessels, with the main interfering tissues being the skull and cerebral cortex. To address these issues, this invention discloses a method for cerebral vessel segmentation in non-angiographic images.

[0005] The specific plan is as follows:

[0006] A method for segmenting cerebral blood vessels in non-angiographic images, characterized by comprising the following steps:

[0007] S1. Input the original image A, i.e., the original non-angiography image;

[0008] S2, air filtering and processing, output image B;

[0009] S3. Remove and process the skull, output image C;

[0010] S4. Initial ablation of the cerebral cortex is processed, and image D is output.

[0011] S5. Iterative ablation of the cerebral cortex is performed to output image E, which is the final cerebral vascular image.

[0012] Furthermore, step S2 is used to remove impurities such as air from outside the skull to avoid affecting the skull dissection; specifically, an air threshold of V is set.air Threshold segmentation fills the data with V0=0, and the principle of air filtering is:

[0013]

[0014] Where p is the three-dimensional coordinate (x, y, z) of the image pixel, A(p) is the pixel value of the original image A at coordinate p, and B(p) is the pixel value of the image B at coordinate p; after filtering out air, all pixels above the scalp skin are set to V0. Further, step S3 is used to remove the scalp tegmentum and most of the muscle tissue below the scalp; this is done from three dimensions of the image (cross-section, sagittal plane, and coronal plane), where the distance from the scalp skin to the cerebral cortex skin in the cross-section includes d. x1 d x2 d y1 d y2 ; where d x1 and d x2 d represents the distance from the left and right sides of the scalp to the left and right sides of the cerebral cortex. y1 and d y2 V represents the distance from the anterior and posterior sides of the scalp to the anterior and posterior sides of the cerebral cortex, respectively; the typical pixel value of the gap between the cerebral cortex and the skull in the original non-angiographic image is V. gap The specific process is as follows:

[0015] a. Scan line by line from the left boundary to the right boundary of the skull cross-section until a pixel B(p) > V0 is encountered. This point is marked as p1, which is near the skull skin. Starting from point p1, continue scanning to the right for a length of d, where d > d0. x1 Until pixel B(p) is encountered in V gap Within the range of ±10, this point is denoted as p2, which is near the cerebral cortex; the area between point p1 and point p2 is the skull. Setting the pixels in this range to V0 will allow the left side of the skull to be peeled off.

[0016]

[0017] Where C(p) is the pixel value of image C at coordinate p;

[0018] b. Following step a, scan line by line from the right boundary to the left boundary of the skull cross-section, where the scan length d > d. x2 Then the skull on the right side of the head can be dissected;

[0019] c. Following step a, scan column by column from the front boundary to the rear boundary of the skull transverse section, where the scan length d > d. y1 Then the skull can be dissected on the front side of the head;

[0020] d. Following step a, scan column by column from the posterior boundary to the front boundary of the skull cross-section, where the scan length d > d y2 Then the posterior skull can be dissected; at this point, the transverse skull dissection is complete.

[0021] e. Complete the sagittal plane skull dissection in sequence according to steps a to d;

[0022] f. Follow steps a to d to complete the dissection of the skull in the coronal plane; at this point, all skull dissection is complete, and image C is obtained.

[0023] Furthermore, step S4 is used to eliminate the more easily segmented cerebral cortex, reducing the number of subsequent ablation iterations; specifically, it involves determining an initial threshold V for a cerebral cortex. init Then the output image D of this step is:

[0024]

[0025] Where D(p) is the pixel value of image D at coordinate p.

[0026] Further, step S5 specifically involves: determining a maximum threshold V for the cerebral cortex. max Then the width of the pixel value to be ablated is W = V. max -V init Let the gradient of the ablation threshold be g, then the total number of iterations is N = W / g; the process of a single iteration is as follows:

[0027] First determine the threshold range [v] low ,v upper ]:

[0028]

[0029] Where i is the iteration order, and 0 <i≤N;

[0030] Then perform binary segmentation:

[0031]

[0032] R i For the binary image of each iteration process, find R. i All connected components R i (j), calculate the number of pixels m in each connected component. j Since the target principal component for binary segmentation is the cerebral cortex, the connected components with a larger number of pixels represent the cerebral cortex, while those with a smaller number represent approximate vascular components. A pixel count threshold M is defined, and the set of connected components corresponding to the cerebral cortex is as follows:

[0033] S i ={R i(j)|m j >M}

[0034] Where j is the index of the connected component;

[0035] Iterate this process N times to obtain the set S = {S} of all connected regions of the cerebral cortex. i}, then the final output image E is:

[0036]

[0037] Where E(p) is the pixel value of image E at coordinate p.

[0038] The beneficial effects of this invention are as follows:

[0039] This invention can peel off the skull and iteratively ablate the cerebral cortex without affecting the vascular components, so as to obtain the purest and most complete cerebral vascular components. Attached Figure Description

[0040] Figure 1 This is a segmentation map with a lower threshold.

[0041] Figure 2 This is a segmentation map with a higher threshold.

[0042] Figure 3 This is a flowchart of the present invention.

[0043] Figure 4 These are example images involved in each step of the present invention, wherein... Figure 4 .1 represents the original image A (non-angiographic image); Figure 4 Image .2 is image B (after filtering air); Figure 4 Image C (after skull dissection) is shown in figure 3.

[0044] Figure 4 Image 4 is D (after initial ablation of the cerebral cortex); Figure 4 .5 represents image E (cerebral vascular image).

[0045] Figure 5 This is a schematic diagram illustrating the principle of skull dissection in the cross-section of the skull according to the present invention. Detailed Implementation

[0046] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the present invention.

[0047] like Figure 3As shown, this invention provides a method for segmenting brain vessels in non-angiographic images. The method involves dissecting the skull without affecting the vascular components and iteratively ablating the cerebral cortex to obtain the purest and most complete brain vascular components possible. The method includes the following steps:

[0048] S1. Input the original image A, which is the original non-angiographic image (T1W-TFE sequence image);

[0049] S2, air filtering and processing, output image B;

[0050] S3. Remove and process the skull, output image C;

[0051] S4. Initial ablation of the cerebral cortex is processed, and image D is output.

[0052] S5. Iterative ablation of the cerebral cortex is performed to output image E, which is the final cerebral vascular image.

[0053] In this embodiment, step S2 is:

[0054] The primary purpose of air filtration is to remove impurities such as air from outside the skull, preventing any impact on skull dissection. An air threshold value of V is used. air Threshold segmentation fills the data with V0=0, and the principle of air filtering is:

[0055]

[0056] Where p is the three-dimensional coordinate (x, y, z) of the image pixel, A(p) is the pixel value of the original image A at coordinate p, and B(p) is the pixel value of the image B at coordinate p, and so on. After air filtering, all pixels above the scalp skin are set to V0.

[0057] In this embodiment, step S3 is:

[0058] Skull dissection is a key step in this method, combining pixel processing and morphological processing. It primarily involves removing the suprasquadron and most of the muscle tissue beneath the skull. Skull dissection is performed in three dimensions of the image (transverse, sagittal, and coronal planes), as follows: Figure 5 This is a schematic diagram illustrating the principle of transverse skull dissection. Clinical anatomy tells us that the distance d from the epidermis to the cerebral cortex in a transverse section of the skull is... x1 d x2 d y1 d y2 The typical pixel value of the gap between the cerebral cortex and the skull in T1W-TFE sequence images is V. gap Perform craniotomy as follows:

[0059] a. Scan line by line from the left boundary to the right boundary of the skull cross-section until a pixel B(p) > V0 is encountered. This point is denoted as p1, and it is located near the skull skin. Starting from point p1, continue scanning to the right for a length of d, where d > d0. x1 Until pixel B(p) is encountered in V gap Within a range of ±10, this point is denoted as p2, which is located near the cerebral cortex. The area between points p1 and p2 represents the skull. Setting the pixels in this interval to V0 allows for the dissection of the left side of the skull.

[0060]

[0061] b. Similarly, following step a, scan line by line from the right boundary to the left boundary of the skull cross-section, where the scan length d > d. x2 Then the dissection of the right side of the skull can be completed.

[0062] c. Similarly, following step a, scan column by column from the front boundary to the rear boundary of the skull transverse section. At this time, the scan length d > d y1 Then the skull can be dissected from the front of the head.

[0063] d. Similarly, following step a, scan column by column from the posterior boundary to the front boundary of the skull cross-section, where the scan length d > d y2 Then the posterior cranial bone dissection can be completed. At this point, the transverse section of the skull has been dissected. e. Follow steps a to d to complete the sagittal section of the skull dissection.

[0064] f. Follow steps a through d sequentially to complete the coronal plane dissection of the skull. At this point, all skull dissection is complete, resulting in image C.

[0065] In this embodiment, step S4 is:

[0066] Initial ablation of the cerebral cortex is essentially threshold segmentation, aimed only at eliminating the more easily segmented areas of the cortex to reduce the number of subsequent ablation iterations. An initial threshold V for a given cerebral cortex is determined. init Then the output image D of this step is:

[0067]

[0068] In this embodiment, step S5 is:

[0069] After initial ablation of the cerebral cortex, the minimum effective value of the image is V. init The remaining cerebral cortex pixels are similar to those of blood vessels, requiring morphological differentiation and removal. A maximum threshold V for the cerebral cortex needs to be determined. max Then the width of the pixel value to be ablated is W = V. max -V initLet the gradient of the ablation threshold be g, then the total number of iterations is N = W / g. A single iteration is as follows:

[0070] First determine the threshold range [v] low ,v upper ]:

[0071]

[0072] Where i is the iteration order, and 0 <i≤N。

[0073] Then perform binary segmentation:

[0074]

[0075] R i For the binary image of each iteration process, find R. i All connected components R i (j), calculate the number of pixels m in each connected component. j Since the target principal component for binary segmentation is the cerebral cortex, the connected components with a larger number of pixels represent the cerebral cortex, while those with a smaller number represent approximate vascular components. Define a pixel count threshold M, and the set of connected components corresponding to the cerebral cortex is as follows:

[0076] S i ={R i (j)|m j >M}

[0077] Where j is the index of the connected component.

[0078] Iterate this process N times to obtain the set S = {S} of all connected regions of the cerebral cortex. i}, then the final output image E is:

[0079]

[0080] like Figure 4 As shown, Figure 4 .1 represents the original T1W-TFE sequence image as input. Figure 4 .2 is the image after air filtering. Because the pixel value of air is small and sparse, the difference in display effect between the two before and after processing is not significant, but it can reduce the influence of air on subsequent skull dissection. Figure 4 Image .3 shows the skull after it has been completely removed, and the blood vessels on the surface of the brain have been preserved intact. Figure 4 .4 is an image of the initial ablation of the cerebral cortex, mainly to reduce the time required for subsequent ablation iterations; Figure 4 .5 represents the cerebral vascular image output after iterative ablation of the cerebral cortex, compared with... Figure 1 and Figure 2 In comparison, it is evident that the cerebral blood vessels output by this method are pure and structurally intact.

[0081] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered within the scope of protection of this invention.

Claims

1. A method for segmenting cerebral blood vessels in non-angiographic images, characterized in that, Includes the following steps: S1. Input the original image That is, the original non-angiographic image; S2. Air filtration and processing, output image. ; S3. Skull dissection and processing, output image. ; S4. Preliminary ablation of the cerebral cortex, image output. ; S5. Iterative ablation of the cerebral cortex for image output. That is, the final cerebral vascular image; Step S2 is used to remove air impurities from outside the skull to avoid affecting skull dissection; specifically, an air threshold is set as follows: Threshold segmentation fills the data as The principle of air filtration is as follows: in, Three-dimensional coordinates of image pixels , Original image In coordinates The pixel values ​​below, For image In coordinates The pixel values ​​below; after filtering the air, all pixels above the skull skin are set to [value missing]. ; Step S3 is used to remove the scalp tectum and most of the muscle tissue beneath the skull; this is performed from three dimensions of the image, including the transverse, sagittal, and coronal planes. In the transverse image, the distance from the scalp epidermis to the cerebral cortex epidermis includes... , , , ,in, and The distances from the left and right sides of the scalp to the left and right sides of the cerebral cortex are respectively. The distances from the anterior and posterior sides of the scalp to the anterior and posterior sides of the cerebral cortex are respectively; the typical pixel value of the gap between the cerebral cortex and the skull in the original non-angiographic image is... The specific process is as follows: a. Scan line by line from the left boundary to the right boundary of the skull cross-section until a pixel is encountered. Mark this point as It is located near the skin of the scalp; from the point Begin, continue scanning to the right, the scan length is... ,at this time until a pixel is encountered exist Within the range, this point is denoted as It is located near the cerebral cortex; point Time The space between the pixels represents the skull; therefore, the pixels in this space are set to... Then the left side of the skull can be peeled off; in, For image In coordinates The pixel value below; b. Following step a, scan line by line from the right boundary to the left boundary of the skull cross-section. The scan length at this point is... Then the skull on the right side of the head can be dissected; c. Following step a, scan column by column from the front boundary to the rear boundary of the skull transverse section. The scan length at this point is... Then the skull can be dissected on the front side of the head; d. Following step a, scan column by column from the posterior boundary to the front boundary of the skull cross-section. The scan length at this point is... Then the posterior skull can be dissected; at this point, the transverse skull dissection is complete. e. Complete the sagittal plane skull dissection in sequence according to steps a~d; f. Follow steps a-d sequentially to complete the coronal plane dissection of the skull; at this point, all skull dissection is complete, and the image is obtained. ; Step S4 is used to eliminate the more easily segmented cerebral cortex, reducing the number of subsequent ablation iterations; specifically, it involves determining an initial threshold for a cerebral cortex. The output image of this step for: in, For image In coordinates The pixel value below; Step S5 specifically involves: determining a threshold value for the maximum value of the cerebral cortex. The width of the pixel value that needs to be ablated is... - Let the gradient of the iterative ablation threshold be... The total number of iterations is The single iteration process is as follows: First determine the threshold range : in, For the iteration order, and ; Then perform binary segmentation: For the binary image of each iteration process, find All connected components Calculate the number of pixels in each connected component. Since the target principal component for binary segmentation is the cerebral cortex, the component with a larger number of connected pixels represents the cerebral cortex, while the component with a smaller number represents the approximate vascular component; a pixel count threshold is defined. The set of connected components corresponding to the cerebral cortex is: in, The index of the connected component; Iterate the process repeatedly Next, obtain the set of all connected regions of the cerebral cortex. The final output image for: in, For image In coordinates The pixel values ​​below.

Images