Cerebrovascular disease plaque medical image segmentation method
By calculating the credibility of candidate seed points and the growth effect of growth areas, the selection and growth criteria of seed points are improved, and the problem of inaccurate segmentation of cerebrovascular plaques in the prior art is solved, better segmentation effect is achieved, and more accurate disease diagnosis and treatment are supported.
Patent Information
- Application Number
- CN202510567614.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-15
AI Technical Summary
In the prior art, due to inappropriate selection of seed points, the cerebrovascular plaque area is undersegment or oversegment, which affects the diagnosis and treatment of the disease.
By calculating the credibility of candidate seed points, selecting the maximum value of grayscale and gradient values as the target seed points, and combining the growth effect and internal and external differences of the growth area as the growth criteria and termination conditions, the edges and noise points of the plaque area are excluded to achieve more accurate regional segmentation.
It effectively avoids undersegment and oversegment, improves the accuracy of segmentation of plaque areas, and supports more accurate disease diagnosis and treatment.
Smart Images
Figure CN120495319A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image processing, and in particular to a medical image segmentation method for cerebrovascular disease plaques. Background Art
[0002] Atherosclerotic plaques are yellowish-white, porridge-like masses embedded in the inner layer of arterial walls. Plaque formation is a key manifestation of intracranial vascular disease. In the medical industry, magnetic resonance imaging (MRI) technology can be used to obtain medical images of the vessel wall. Image segmentation technology can then be used to segment plaques within the vessel wall, allowing for clear visualization of the size and distribution of each plaque and assessment of its impact on the vessel wall. Image segmentation technology can provide accurate and reliable data support for long-term observation of intracranial arteriosclerosis, thereby assisting in the diagnosis and treatment of the disease.
[0003] Among them, the region growing algorithm, as an image segmentation technique, has the advantages of simple principle and easy implementation, and is widely used in image segmentation in various fields. The region growing algorithm is an image segmentation technique based on the similarity between pixels. Because the grayscale value of plaque in the blood vessel wall is greater than the grayscale value of normal blood vessel wall, the region growing algorithm can be used to segment the plaque from the blood vessel wall. The region growing algorithm mainly includes the following steps: selecting some pixels from the image as seed points; merging neighboring pixels with high similarity to the seed points into a growth region according to a growth criterion, where the growth criterion is the grayscale value difference between the pixels; repeating the previous step with the merged pixel as the new seed point, stopping the growth after reaching a preset number of iterations, and finally obtaining each segmented region.
[0004] However, the seed points selected by the region growing algorithm in existing technologies may be located at the edges of the plaque or at noise points. Seed points located at the edges of the plaque are restricted by the growth criteria and can only grow toward the center of the plaque, resulting in low growth efficiency. Growth stops after reaching a preset number of iterations, leading to under-segmentation of the plaque. On the other hand, when region growing seed points located at noise points, segmented regions that do not belong to the plaque are generated, resulting in over-segmentation of the plaque. Therefore, existing technologies suffer from poor plaque segmentation, hindering disease diagnosis and treatment. Summary of the Invention
[0005] The present invention provides a medical image segmentation method for cerebrovascular disease plaques, aiming to solve the technical problem in the prior art of under-segmentation or over-segmentation of plaque areas due to inappropriate selection of seed points.
[0006] A cerebrovascular disease plaque medical image segmentation method of the present invention comprises the following steps:
[0007] Acquire images of blood vessel walls in medical images of cerebrovascular disease plaques;
[0008] The vascular wall image is segmented using the region growing algorithm to obtain the plaque regions;
[0009] The region growing algorithm includes target seed points; target seed points are candidate seed points with a credibility greater than a threshold; candidate seed points are pixels corresponding to the maximum grayscale value in the vascular wall image; the credibility of each candidate seed point is positively correlated with the ratio of the grayscale value of the corresponding pixel to the maximum grayscale value of all pixels in the vascular wall image, the ratio of the mean grayscale value of each neighboring pixel of the corresponding pixel to the mean grayscale value of all pixels in the vascular wall image, and the ratio of the maximum gradient value of the grayscale values of all pixels in the vascular wall image to the maximum gradient value of each neighboring pixel of the corresponding pixel.
[0010] In the above scheme, the credibility of the candidate seed point is obtained by calculating the difference between the grayscale value and gradient value of the candidate seed point or its neighborhood pixel points and all pixel points, and the candidate seed point with a credibility greater than the threshold is used as the target seed point, excluding the cases at the edge of the patch area and the noise point, making the patch area segmentation more accurate.
[0011] Preferably, the credibility C of the candidate seed point k k for:
[0012]
[0013] Where A k is the gray value of candidate seed point k, A max is the maximum grayscale value of all pixels in the vascular wall image, is the mean grayscale value of each neighboring pixel point of the candidate seed point k, is the mean gray value of all pixels in the vascular wall image, B max is the maximum gradient value of the grayscale values of all pixels in the blood vessel wall image, B k,max is the maximum gradient value of each neighborhood pixel of the candidate seed point k, and sigmoid() is the normalization function.
[0014] Preferably, the growth criterion of the region growing algorithm is the growth effect of the growing region; the growth effect is negatively correlated with the absolute value of the difference between the ratio of the credibility mean of all pixels in the growing region after the previous iteration and the current iteration and 1, and is negatively correlated with the absolute value of the difference between the homogeneity of the gray level co-occurrence matrix of the growing region after the current iteration and the homogeneity mean of the gray level co-occurrence matrix of the growing region in the same direction of all iterations before the current iteration.
[0015] In the above scheme, the growth effect of the growing area is used as the growth criterion, which can adapt to the segmentation conditions of different patch areas and achieve a good segmentation effect for patch areas with gradual changes.
[0016] Preferably, the growth effect D of the growth region m after the nth iteration is m,n for:
[0017]
[0018] Where, is the mean credibility of all pixels in the growth area m after the nth iteration, is the mean credibility of all pixels in the growth area m after the n-1th iteration, E m,n is the homogeneity of the gray-level co-occurrence matrix of the growth region m after the nth iteration, is the homogeneity mean of the gray-level co-occurrence matrix of the growth region m in the first n-1 iterations, σ(E m,n-1 ) is the standard deviation of the homogeneity of the gray-level co-occurrence matrix of the growth region m in the first n-1 iterations, exp() is an exponential function with the natural constant e as the base, and the homogeneity characterizes the similarity between the elements in the gray-level co-occurrence matrix.
[0019] Preferably, the termination condition of the region growing algorithm is that the difference between the inside and outside of the corresponding growing region reaches a maximum value; wherein, the difference between the inside and outside of the growing region m after the nth iteration F m,n for:
[0020]
[0021] Where, E m,n is the homogeneity of the gray-level co-occurrence matrix of the growth region m after the nth iteration, E m,n ′ is the homogeneity of the gray level co-occurrence matrix of the external area of each growth area after the nth iteration, A m,n is the mean gray value of all pixels in the growth area m after the nth iteration, A m,n ′ is the mean grayscale value of all pixels in the outer area of each growth area after the nth iteration, D m,n is the growth effect of the growth region m after the nth iteration.
[0022] In the above scheme, the homogeneity difference, gray value difference and growth effect between the corresponding area and its external area are used to jointly characterize the internal and external difference, and used as the termination condition, which can more accurately reflect the termination condition and achieve better segmentation effect on the patch area.
[0023] Preferably, the internal and external differences of the corresponding growth area reach a maximum value, which includes: establishing a coordinate system with the number of iterations of the corresponding growth area as the horizontal axis and the internal and external differences after iteration as the vertical axis, and taking the internal and external differences of the data points in the coordinate system when the slope of the curve formed by the data points is 0 as the maximum value.
[0024] Preferably, the threshold is obtained using the ostu algorithm.
[0025] Among the above schemes, the ostu algorithm has the advantages of no need for human intervention, strong adaptability and high computational efficiency.
[0026] Preferably, the candidate seed points are obtained by dividing the blood vessel wall image into multiple square windows with a side length of 2m+1 pixels, and taking the pixel point corresponding to the maximum grayscale value in each square window as the candidate seed point, where m is an integer greater than or equal to 1.
[0027] In the above solution, a plurality of evenly distributed candidate seed points can be obtained in the vascular wall image, thereby preventing some small plaque areas from being missed and finding various plaque areas in the vascular wall image.
[0028] Preferably, the vascular wall image is separated from the cerebrovascular disease plaque medical image by TOF-MRA technology.
[0029] Preferably, the medical image of cerebrovascular disease plaque is an image that has been filtered and denoised, wherein the filtering method is one of Gaussian filtering, median filtering and mean filtering.
[0030] In the above scheme, filtering and denoising are performed to avoid the influence of noise on image segmentation in subsequent steps.
[0031] The beneficial effects are:
[0032] The present invention's solution uses the maximum grayscale value in the vessel wall image to determine candidate seed points. Candidate seed points with a confidence level greater than a threshold are then selected as target seed points for region segmentation. Because the target seed point is determined by the grayscale and gradient differences between the candidate seed point and all pixels, candidate seed points located at the edges of plaque regions and at noise points are excluded, thus avoiding under-segmentation or over-segmentation during region segmentation. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 This is a flowchart of the steps of the cerebrovascular disease plaque medical image segmentation method according to an embodiment of the present invention;
[0034] Figure 2 Flowchart of the steps for obtaining target seed points according to an embodiment of the present invention. DETAILED DESCRIPTION
[0035] The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, but should not be construed as limiting the present invention.
[0036] like Figure 1 As shown, the present invention provides a method for segmenting medical images of cerebrovascular disease plaques, comprising the following steps:
[0037] S1. Acquire a blood vessel wall image in a medical image of a cerebrovascular disease plaque.
[0038] In this step, the medical image of cerebrovascular disease plaque is usually an HRMR image, namely a high-resolution magnetic resonance (MRI) image, which is a high-resolution image obtained using magnetic resonance imaging technology. HRMR images are acquired using magnetic resonance imaging equipment.
[0039] Vibration from the acquisition equipment and the magnetic field generated by alternating current can introduce noise into the acquired HRMR images, leading to inaccurate subsequent image processing results. Therefore, to improve the accuracy of image processing results, HRMR images must be denoised. For example, Gaussian filtering, median filtering, or mean filtering can be used to denoise HRMR images.
[0040] Since plaques only exist in the vascular wall area, when segmenting the HRMR image, the interference of the brain tissue background must be eliminated, that is, the vascular wall area needs to be extracted to obtain a vascular wall image containing the plaque. In this step, the vascular wall image is separated from the HRMR image using TOF-MRA technology. The full name of TOF-MRA technology is time-of-flight magnetic resonance angiography technology. TOF-MRA technology separates the vascular wall image from the HRMR image based on the signal difference between blood flow and static tissue. The subsequent image segmentation processing is all performed on the vascular wall image, which can eliminate the interference of the brain tissue background.
[0041] S2. Segment the vascular wall image using a region growing algorithm to obtain plaque regions.
[0042] Region growing algorithms are currently available. Their basic idea is to group pixels with similar properties together to form a growing region. Starting with a set of seed points, the algorithm compares the grayscale values of these seed points with their neighboring pixels, merging neighboring pixels with high similarity into the growing region where the seed points are located. Region growing then continues using these newly merged pixels as new seed points until a termination condition is met. Specifically, the prior art region growing algorithm includes the following steps:
[0043] S21. Select seed points: The selection of seed points is based on image features, such as grayscale values. Since the grayscale values of the plaque area are greater than those of the normal blood vessel wall area, grayscale values greater than the set value can be used as seed points for region growing.
[0044] S22. Determine the growth criterion: The growth criterion defines the conditions under which pixels are merged into the growth region. It is typically based on the grayscale difference between pixels. For example, if the grayscale difference between a seed point and its neighboring pixels is less than a threshold, the neighboring pixels are merged into the growth region where the seed point is located.
[0045] S23, region growing: Starting from the seed point, merge the neighboring pixels that meet the conditions into the growing region according to the growth criterion, and update the seed point. Repeat this step and iterate continuously.
[0046] S24. Termination Condition: The growth process of the growing region requires a termination condition, which is usually when pixels that do not meet the growth criteria can be merged or when a preset number of iterations is reached. When the growing region stops growing, the corresponding segmented regions, i.e., patch regions, are obtained.
[0047] However, in step S21 of the prior art, the selected seed point may be located at the edge of the plaque or at a noise point. Since the grayscale values of the pixels on the inner and outer sides of the plaque differ greatly, the seed point located at the edge can only grow unidirectionally toward the center of the plaque, with low growth efficiency, and stops growing when the growing area reaches a preset number of iterations, which will cause under-segmentation in the growing area. If the seed point is located at a noise point, when the seed point is used for region growth, a segmented area that does not belong to the plaque will be obtained, resulting in over-segmentation. Therefore, in the prior art, due to improper selection of seed points, the plaque segmentation effect is poor, which in turn affects the diagnosis and treatment of the disease.
[0048] The present invention improves the principle of selecting seed points, and replaces the seed points in the prior art with target seed points. The target seed point is a candidate seed point whose credibility is greater than a threshold. The candidate seed point is the pixel point corresponding to the maximum grayscale value in the vascular wall image. This is because the grayscale value of the pixel point in the plaque area is greater than the grayscale value of the pixel point in the normal vascular wall. Therefore, the pixel point corresponding to the maximum grayscale value in the vascular wall image is used as the candidate seed point, and the candidate seed point must include the pixel point in the plaque area. In order to avoid the situation where the candidate seed point is located at the edge of the plaque area, noise point, and normal vascular wall area, the credibility of each candidate seed point is compared with the threshold to exclude them. Therefore, if Figure 2 As shown, obtaining the target seed point also includes the following steps:
[0049] S211 : Acquire candidate seed points in the blood vessel wall image.
[0050] In one embodiment, the candidate seed points are obtained by dividing the blood vessel wall image into multiple square windows with a side length of 2m+1 pixels, and taking the pixel point corresponding to the maximum grayscale value in each square window as the candidate seed point, where m is an integer greater than or equal to 1.
[0051] In this embodiment, each square window can correspond to a candidate seed point, and then multiple evenly distributed candidate seed points can be obtained in the blood vessel wall image, so that the pixel points in each plaque can be used as candidate seed points, avoiding the omission of some small plaques.
[0052] S212: Calculate the credibility of each candidate seed point.
[0053] The credibility of each candidate seed point is positively correlated with the ratio of the grayscale value of the corresponding pixel to the maximum grayscale value of all pixels in the vascular wall image, the ratio of the mean grayscale value of each neighborhood pixel of the corresponding pixel to the mean grayscale value of all pixels in the vascular wall image, and the ratio of the maximum gradient value of the grayscale values of all pixels in the vascular wall image to the maximum gradient value of each neighborhood pixel of the corresponding pixel.
[0054] This is because the grayscale values of pixels within plaques are greater than those of pixels in normal blood vessel walls. Therefore, the closer the grayscale value of the candidate seed point is to the maximum grayscale value of all pixels in the blood vessel wall image, and the larger the ratio between the two, the higher the confidence that the candidate seed point is located in the plaque region. The smaller the ratio, the higher the confidence that the candidate seed point is located in the normal blood vessel wall region, and the lower the confidence that it is located in the plaque region. Therefore, the confidence of a candidate seed point is positively correlated with the ratio between the two.
[0055] The larger the ratio of the mean grayscale value of the candidate seed point's neighborhood to the mean grayscale value of all pixels in the vessel wall image, the higher the confidence that the candidate seed point is located in the plaque region. The smaller the ratio, the higher the confidence that the candidate seed point is located in the normal vessel wall region, and the lower the confidence that it is located in the plaque region. Therefore, the confidence of the candidate seed point is positively correlated with the ratio.
[0056] The smaller the ratio of the maximum gradient of the grayscale values of all pixels in the vascular wall image to the maximum gradient of each neighboring pixel, the larger the maximum gradient of each neighboring pixel. This indicates that the grayscale value variation between the corresponding pixel and its neighboring pixels is more dramatic, and the more likely the seed point is located at the edge of a plaque region or at a noise point. The smaller the ratio, the less reliable the candidate seed point, and thus, candidate seed points located at the edge of a plaque region or at a noise point can be excluded. It should be noted that although the image is filtered and denoised in step S1, due to the limitations of the filtering and denoising algorithm, noise cannot be completely removed. Therefore, step S212 can further eliminate the influence of noise on the selection of seed points.
[0057] In one embodiment, the credibility C of the candidate seed point k k for:
[0058]
[0059] Where A k is the gray value of candidate seed point k, A max is the maximum grayscale value of all pixels in the vascular wall image, is the mean grayscale value of each neighboring pixel point of the candidate seed point k, is the mean gray value of all pixels in the vascular wall image, B max is the maximum gradient value of the grayscale values of all pixels in the blood vessel wall image, B k,nax is the maximum gradient value of each neighborhood pixel of the candidate seed point k, and sigmoid() is the normalization function.
[0060] In the above formula, When the candidate seed point is located at the normal blood vessel wall, its value is negative, which makes the credibility of the candidate seed point lower and makes it easier to exclude the candidate seed point located at the normal blood vessel wall.
[0061] In this embodiment, by comparing the grayscale value of the candidate seed point with the maximum grayscale value of all pixels, the average grayscale value of the neighborhood pixels of the candidate seed point and all pixels, and the maximum gradient value of the grayscale values of all pixels and the candidate seed point, the possibility that the candidate seed point is located in the patch area can be comprehensively reflected, and the situation where the candidate seed point is located at the edge of the patch area and the noise point can be excluded, thereby preventing the growth area from being under-segmented and over-segmented during growth.
[0062] S213: Obtain a threshold value according to the credibility of each candidate seed point.
[0063] In one embodiment, the threshold is obtained using the OSTU algorithm. Specifically, the credibility of each candidate seed point is input into the OSTU algorithm to obtain a corresponding threshold. The OSTU algorithm is an automatic threshold selection method that calculates the inter-class variance under different thresholds and finds the threshold that maximizes the inter-class variance, thereby classifying each candidate seed point into a target seed point and a non-target seed point.
[0064] Of course, in other embodiments, the threshold value may also be set manually based on experience.
[0065] S214: Obtain target seed points.
[0066] Compare the credibility of each candidate seed point with the threshold, and select the candidate seed point with a credibility higher than the threshold as the target seed point. Then, region growth is performed around the target seed point.
[0067] In step S211 to step S214, the target seed point is obtained by calculating the credibility of the candidate seed point. Since the target seed point excludes the points at the edge of the patch area and the noise points, under-segmentation and over-segmentation of the segmented area can be prevented.
[0068] In step S22 of the prior art, the growth criterion is the grayscale difference between the seed point and its neighboring pixels. By comparing this difference with a threshold, a decision is made as to whether to merge the neighboring pixels into the growth region where the seed point is located. However, because the grayscale values and distribution of pixels within each patch vary, using a single threshold cannot adapt to the segmentation of different patch regions. Furthermore, patch regions may have gradients along their edges, so using a threshold for region segmentation does not achieve a good segmentation effect.
[0069] The present invention improves the growth criterion and uses the growth effect of the growing region as the growth criterion. The growth effect is negatively correlated with the absolute value of the difference between the ratio of the mean credibility values of all pixels in the growing region after the previous iteration and the current iteration and 1, and is negatively correlated with the absolute value of the difference between the homogeneity of the gray-level co-occurrence matrix of the growing region after the current iteration and the homogeneity mean of the gray-level co-occurrence matrix of the growing region in the same direction for all iterations before the current iteration.
[0070] This is because the grayscale values of pixels within the same patch area have a small difference. When the difference in the credibility of all pixels in the growth area after the previous iteration and the current iteration is small, that is, the smaller the absolute value of the difference between the ratio of the two and 1, it indicates that the growth area is within the patch area, and the growth effect of the growth area is good. On the contrary, when the difference between the two is large, it indicates that the growth area after the current iteration is likely to be outside the patch area, and the growth effect of the growth area is poor. The calculation of the credibility of any pixel is the same as the calculation of the credibility of the candidate seed point in step S212, and will not be repeated here.
[0071] Among them, the grayscale co-occurrence matrix belongs to the prior art, which is a statistical method for image texture analysis. The grayscale co-occurrence matrix in the present invention is obtained based on the grayscale value of each pixel point in the corresponding growth area. And each growth area will obtain a corresponding grayscale co-occurrence matrix in its different directions. In order to be comparable, the grayscale co-occurrence matrix of the present invention is obtained in the same direction of the growth area. Each element in the grayscale co-occurrence matrix represents the frequency of occurrence of grayscale values corresponding to two pixel points in the corresponding growth area in a certain direction. The grayscale co-occurrence matrix in each direction characterizes the distribution of texture features of the corresponding growth area in that direction. The homogeneity of the grayscale co-occurrence matrix characterizes the similarity between the elements in the grayscale co-occurrence matrix, and the greater the homogeneity, the more uniform the texture of the image, and vice versa, the more complex the texture of the image. The calculation of homogeneity is a prior art and will not be repeated here.
[0072] If the difference between the homogeneity of the gray level co-occurrence matrix of the growth region after the current iteration and the homogeneity mean of the gray level co-occurrence matrix of the growth region in the same direction after each previous iteration is small, it means that the growth region is located in the patch area, and the growth effect of the growth region is good. On the contrary, if the difference between the two is large, it means that the growth region after the current iteration is likely to be located outside the patch area, and the growth effect of the growth region is poor. Therefore, step S22 also includes the following steps:
[0073] S221. Obtain the growth effect of each growth area.
[0074] In one embodiment, the growth effect D of the growth region m after the nth iteration is m,n for:
[0075]
[0076] Where, is the mean credibility of all pixels in the growth area m after the nth iteration, is the mean credibility of all pixels in the growth area m after the n-1th iteration, E m,n is the homogeneity of the gray-level co-occurrence matrix of the growth region m after the nth iteration, is the homogeneity mean of the gray-level co-occurrence matrix of the growth region m in the first n-1 iterations, σ(E m,n-1 ) is the standard deviation of the homogeneity of the gray-level co-occurrence matrix of the growth region m in the first n-1 iterations, and exp() is an exponential function with the natural constant e as the base.
[0077] In this embodiment, the growth effect of the growth area of the current iteration is reflected by comparing the differences in pixel points in the growth area of the previous iteration and the current iteration. Therefore, it can adapt to the segmentation of different patch areas, and can also achieve good segmentation effects when there is a gradient in the patch area.
[0078] S222. Set the judgment conditions of the growth criterion.
[0079] When growing a region, set an effect threshold. When the growth effect of the growing region is greater than the effect threshold, the pixels added in this iteration are retained and the next growth is performed. When the growth effect of the growing region is less than or equal to the effect threshold, the pixels added in this iteration are removed. For example, the effect threshold can be 0.5, and of course the effect threshold can be set as needed.
[0080] In step S24 of the prior art, the termination condition is that pixels that do not meet the growth criterion can be merged or a preset number of iterations has been reached. However, when the growth criterion is used as the termination condition, since the termination condition is determined solely based on the growth criterion and does not consider the overall distribution of pixels within the growth region, inaccurate growth region segmentation may occur. When the preset number of iterations is used as the termination condition, due to the different sizes of the patch regions, the preset number of iterations may not be sufficient to segment all patch regions.
[0081] Therefore, the present invention improves the termination condition of the region growing algorithm. The termination condition of the region growing algorithm is that the internal and external differences of the corresponding growing region reach a maximum value. The internal and external differences are positively correlated with the ratio of the homogeneity of the grayscale co-occurrence matrix of the growing region after the current iteration and the external region of each growing region, and are positively correlated with the ratio of the grayscale value mean of all pixels of the growing region after the current iteration and the external region of each growing region, and are positively correlated with the growth effect of the growing region after the current iteration. This is because as the number of iterations increases, the growing region gradually grows to overlap with the corresponding patch region, and at this time the internal and external differences of the growing region also reach a maximum value. Therefore, step S24 also includes the following steps:
[0082] S241. Obtain the internal and external differences of each growth area.
[0083] In one embodiment, the difference between the inside and outside of the growing region m after the nth iteration is F m,n for:
[0084]
[0085] Where, E m,n is the homogeneity of the gray-level co-occurrence matrix of the growth region m after the nth iteration, E m,n ′ is the homogeneity of the gray level co-occurrence matrix of the external area of each growth area after the nth iteration, A m,n is the mean gray value of all pixels in the growth area m after the nth iteration, A m,n ′ is the mean grayscale value of all pixels in the outer area of each growth area after the nth iteration, D m,n is the growth effect of the growth region m after the nth iteration.
[0086] In this step, the internal and external difference is characterized by the homogeneity difference, gray value difference and growth effect between the corresponding growing area and its external area, which can more accurately reflect the termination condition and achieve better segmentation effect on the patch area.
[0087] S242. When the difference between the inside and outside of each growth region reaches a maximum value, the growth of the corresponding growth region is terminated.
[0088] In one embodiment, a coordinate system is established with the number of iterations corresponding to the growth region as the horizontal axis and the internal-external difference after iteration as the vertical axis. The maximum internal-external difference of a data point is taken when the slope of the curve formed by the data points in the coordinate system is 0. This is because as the number of iterations increases, the internal-external difference of the growth region gradually increases and then stabilizes, resulting in a data point on the curve with a slope of 0. The internal-external difference of this data point is taken as the maximum internal-external difference of the corresponding growth region, and the growth region corresponding to the number of iterations for this data point is taken as the final segmentation region, i.e., the patch region.
[0089] The present invention's medical image segmentation method for cerebrovascular plaques first selects candidate sink seed points from the vascular wall image. The credibility of each candidate seed point is then determined by comparing the grayscale and gradient values of the corresponding candidate seed point or its neighboring pixels with all pixels. Candidate seed points with a credibility greater than a threshold are selected as target seed points for region growing. Because target seed points located at the edges of the plaque region and in noise areas are excluded, under-segmentation or over-segmentation during region growing can be avoided, resulting in better segmentation and facilitating disease diagnosis and treatment.
[0090] Moreover, the growth effect of the growing area is used as the growth criterion, which can adapt to the segmentation of different patch areas and achieve good segmentation effect even when there is a gradient in the patch area.
[0091] Furthermore, the homogeneity difference, gray value difference and growth effect between the growing area and its external area are used to characterize the internal and external difference, and used as the termination condition, which can achieve a better segmentation effect on the patch area.
[0092] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.
Claims
1. A method for segmenting medical images of cerebrovascular disease plaques, characterized in that: The steps include: Acquire images of blood vessel walls in medical images of cerebrovascular disease plaques; The vascular wall image is segmented using the region growing algorithm to obtain the plaque regions; The region growing algorithm includes target seed points; target seed points are candidate seed points with a credibility greater than a threshold; candidate seed points are pixels corresponding to the maximum grayscale value in the vascular wall image; the credibility of each candidate seed point is positively correlated with the ratio of the grayscale value of the corresponding pixel to the maximum grayscale value of all pixels in the vascular wall image, the ratio of the mean grayscale value of each neighboring pixel of the corresponding pixel to the mean grayscale value of all pixels in the vascular wall image, and the ratio of the maximum gradient value of the grayscale values of all pixels in the vascular wall image to the maximum gradient value of each neighboring pixel of the corresponding pixel.
2. The cerebrovascular disease plaque medical image segmentation method according to claim 1, characterized in that: The credibility C of the candidate seed point k k for: Where A k is the gray value of candidate seed point k, A max is the maximum grayscale value of all pixels in the vascular wall image, is the mean grayscale value of each neighboring pixel point of the candidate seed point k, is the mean gray value of all pixels in the vascular wall image, B max is the maximum gradient value of the grayscale values of all pixels in the blood vessel wall image, B k，max is the maximum gradient value of each neighborhood pixel of the candidate seed point k, and sigmoid() is the normalization function.
3. The cerebrovascular disease plaque medical image segmentation method according to claim 1, characterized in that: The growth criterion of the region growing algorithm is the growth effect of the growing region; the growth effect is negatively correlated with the absolute value of the difference between the ratio of the mean credibility of all pixels in the growing region after the previous iteration and the current iteration and 1, and is negatively correlated with the absolute value of the difference between the homogeneity of the gray level co-occurrence matrix of the growing region after the current iteration and the homogeneity mean of the gray level co-occurrence matrix of the growing region in the same direction of all iterations before the current iteration.
4. The cerebrovascular disease plaque medical image segmentation method according to claim 3, characterized in that: The growth effect D of the growth region m after the nth iteration m，n for: Where, is the mean credibility of all pixels in the growth area m after the nth iteration, is the mean credibility of all pixels in the growth area m after the n-1th iteration, E m，n is the homogeneity of the gray-level co-occurrence matrix of the growth region m after the nth iteration, is the homogeneity mean of the gray-level co-occurrence matrix of the growth region m in the first n-1 iterations, σ(E m，n-1 ) is the standard deviation of the homogeneity of the gray-level co-occurrence matrix of the growth region m in the first n-1 iterations, exp() is an exponential function with the natural constant e as the base, and the homogeneity characterizes the similarity between the elements in the gray-level co-occurrence matrix.
5. The cerebrovascular disease plaque medical image segmentation method according to claim 4, characterized in that: The termination condition of the region growing algorithm is that the difference between the inside and outside of the corresponding growing region reaches the maximum value; among them, the difference between the inside and outside of the growing region m after the nth iteration is F m，n for: Where, E m，n is the homogeneity of the gray-level co-occurrence matrix of the growth region m after the nth iteration, E m，n ′ is the homogeneity of the gray level co-occurrence matrix of the external area of each growth area after the nth iteration, A m，n is the mean gray value of all pixels in the growth area m after the nth iteration, A m，n ′ is the mean grayscale value of all pixels in the outer area of each growth area after the nth iteration, D m，n is the growth effect of the growth region m after the nth iteration.
6. The cerebrovascular disease plaque medical image segmentation method according to claim 5, characterized in that: The internal and external differences of the corresponding growth area reach a maximum value, which includes: establishing a coordinate system with the number of iterations of the corresponding growth area as the horizontal axis and the internal and external differences after iteration as the vertical axis, and taking the internal and external differences of the data points in the coordinate system when the slope of the curve formed by the data points is 0 as the maximum value.
7. The cerebrovascular disease plaque medical image segmentation method according to claim 1, characterized in that: The threshold is obtained using the ostu algorithm.
8. The cerebrovascular disease plaque medical image segmentation method according to claim 1, characterized in that: The candidate seed points are obtained by dividing the blood vessel wall image into multiple square windows with a side length of 2m+1 pixels, and taking the pixel point corresponding to the maximum grayscale value in each square window as the candidate seed point, where m is an integer greater than or equal to 1.
9. The cerebrovascular disease plaque medical image segmentation method according to claim 1, characterized in that: TOF-MRA technology is used to separate vascular wall images from medical images of cerebrovascular disease plaques.
10. The cerebrovascular disease plaque medical image segmentation method according to claim 1, characterized in that: The medical image of the cerebrovascular disease plaque is an image that has been filtered and denoised, wherein the filtering method is one of Gaussian filtering, median filtering and mean filtering.