An auxiliary segmentation method and system for peripheral vascular images

Abstract

The present invention relates to the technical field of peripheral blood vessel image segmentation, and specifically relates to a method and system for assisting in the segmentation of peripheral blood vessel images. The present invention obtains a lesion intensity index according to the gray-scale distribution and edge distribution within a preset neighborhood of lesion pixel points; obtains a shape feature factor according to the contour features of the lesion region; obtains a structural feature factor according to the gray-scale distribution of the lesion region; obtains a range feature factor according to the number of pixel points in the lesion region; obtains an extended weight coefficient according to the differences between the above feature factors and the lesion intensity index; obtains the extended sampling priority according to the distribution of the extended weight coefficients of the lesion regions in each lesion blood vessel image; and further obtains an extended training model; and segments the peripheral blood vessel images according to the extended training model. The present invention makes the extended peripheral blood vessel samples balanced and rich, increases the generalization ability of the deep learning segmentation network, and improves the accuracy of the assisted segmentation.

Description

Technical Field

[0001] The present invention relates to the technical field of peripheral blood vessel image segmentation, and particularly to a method and system for assisting in the segmentation of peripheral blood vessel images. Background Art

[0002] Peripheral blood vessels refer to the blood vessels in the human limbs, head and neck, trunk, etc., excluding the cardiovascular and cerebrovascular vessels. Assisting in the segmentation of the images of human peripheral blood vessels has important clinical and research significance. It can help doctors and researchers accurately locate the position and morphology of blood vessel structures, thereby providing certain support for diagnosing diseases, formulating treatment plans, and studying changes in blood vessel anatomical structures, saving a large amount of time and labor costs.

[0003] In the prior art, a deep learning segmentation network is often used to assist in the segmentation of peripheral blood vessel images. Since various types of lesions may occur in human peripheral blood vessels, a large number of lesion samples are required to train the deep learning segmentation network during the assistance of the segmentation, so it is necessary to expand the peripheral blood vessel images. However, the expansion of peripheral blood vessel images in the prior art is usually random sampling, without considering problems such as different types, structures, and distributions of peripheral blood vessel lesions, resulting in sample imbalance and insufficient richness in the expanded peripheral blood vessels, reducing the generalization ability of the deep learning segmentation network and making it difficult to segment various different types of peripheral blood vessel images well. Summary of the Invention

[0004] In order to solve the technical problem that the prior art usually expands peripheral vascular images by random sampling without considering the different types, structures, and distributions of peripheral vascular lesions, resulting in sample imbalance and insufficient richness of the expanded peripheral blood vessels, reducing the generalization ability of the deep learning segmentation network, and making it difficult to segment various types of peripheral vascular images well, the purpose of the present invention is to provide a peripheral vascular image-assisted segmentation method and system, and the technical scheme adopted is as follows: a peripheral vascular image-assisted segmentation method, the method comprising: collecting all peripheral vascular images; taking the peripheral vascular image with lesions as the lesion vascular image; selecting any lesion pixel point in any lesion vascular image as a reference lesion pixel point; obtaining the lesion intensity index of the reference lesion pixel point according to the grayscale distribution in a preset neighborhood of the reference lesion pixel point and the gradient distribution of the peripheral vascular edge pixel point; and obtaining the lesion intensity index of the reference lesion pixel point according to each lesion vessel. The lesion pixel distribution of the image is used to obtain the lesion area in each lesion blood vessel image; one lesion area is selected as the reference lesion area; according to the contour characteristics of the reference lesion area, the shape characteristic factor of the reference lesion area is obtained; according to the grayscale distribution in the reference lesion area, the structural characteristic factor of the reference lesion area is obtained; according to the number characteristics of lesion pixels in the reference lesion area, the range characteristic factor of the reference lesion area is obtained; according to the shape characteristic factor difference, structural characteristic factor difference, range characteristic factor difference of the reference lesion area compared with other lesion areas and the lesion intensity index of each lesion pixel, the expansion weight coefficient of the reference lesion area is obtained; according to the expansion weight coefficient distribution characteristics of all lesion areas in each lesion blood vessel image, the expansion sampling priority of each lesion blood vessel image is obtained; based on the expansion sampling priority, an expanded training model is obtained; and the peripheral blood vessel image is segmented according to the expanded training model.

[0005] Furthermore, the method for obtaining the lesion strength index includes: obtaining the lesion strength index according to a lesion strength index calculation formula, and the lesion strength index calculation formula is as follows: In the formula, represents the lesion intensity index of the reference lesion pixel; Represents the number of pixels within a preset neighborhood of a reference lesion pixel; Indicates the first pixel in the preset neighborhood of the reference lesion pixel. The gray value of each pixel; Represents the grayscale mean of the pixels within the preset neighborhood of the reference lesion pixel; Represents the grayscale mean of the pixel points in the lesion vessel image where the reference lesion pixel point is located; Represents the number of peripheral blood vessel edge pixels within a preset neighborhood of a reference lesion pixel; Represents the first The gradient direction angle of a peripheral blood vessel edge pixel point; It represents the average value of the gradient direction angles of the peripheral blood vessel edge pixel points within the preset neighborhood of the reference lesion pixel point.

[0006] Furthermore, the method for obtaining the lesion area includes: taking the pixel points corresponding to the diseased peripheral blood vessel area in each lesion blood vessel image as lesion pixel points; using the connected component labeling algorithm to obtain all the connected components formed by all the lesion pixel points in each lesion blood vessel image, and taking all the connected components as all the lesion areas in each lesion blood vessel image.

[0007] Furthermore, the method for obtaining the shape feature factor includes: obtaining all the edge pixel points and the centroid pixel point of the reference lesion area; obtaining the connection lines between the centroid pixel point and all the edge pixel points as the shape feature line segments; taking the centroid pixel point as the center, evenly dividing the reference lesion area along a preset angle to obtain all the lesion modules; selecting the longest shape feature line segment and the shortest shape feature line segment in each lesion module, taking the angle formed by every two adjacent longest shape feature line segments in all the lesion modules as the first shape feature angle, and taking the angle formed by every two adjacent shortest shape feature line segments in all the lesion modules as the second shape feature angle; obtaining the shape feature factor of the reference lesion area according to the first shape feature angle and the second shape feature angle, and the calculation formula is as follows: In the formula, represents the shape feature factor of the reference lesion area; represents the number of the first shape feature angles; represents the number of the second shape feature angles, and ; represents the th first shape feature angle and the th first shape feature angle of the first-order difference between the angles; represents the th first shape feature angle and the th first shape feature angle of the first-order difference between the angles; represents the absolute value function.

[0008] Furthermore, the method for obtaining the structural feature factor includes: obtaining the structural feature factor according to the structural feature factor calculation formula, and the structural feature factor calculation formula is as follows: In the formula, represents the structural feature factor of the reference lesion area; represents the number of shape feature line segments within the reference lesion area; represents the The gray mean value of the pixel points in the strip-shaped feature line segment; Denote the gray mean value of the pixel points in the th strip-shaped feature line segment in the reference lesion area; Denote the th strip-shaped feature line segment in the reference lesion area; Denote the th strip-shaped feature line segment in the reference lesion area; the gray value of the th pixel point in the th strip-shaped feature line segment in the reference lesion area; the gray value of the th pixel point in the

[0009] Further, the method for obtaining the range feature factor includes: obtaining the range feature factor according to the range feature factor calculation formula, and the range feature factor calculation formula is as follows: In the formula, Denote the range feature factor of the reference lesion area; Denote the number of pixel points in the reference lesion area; Denote the number of pixel points in all lesion areas in the lesion blood vessel image where the reference lesion area is located; Denote the number of pixel points in the lesion blood vessel image where the reference lesion area is located.

[0010] Further, the method for obtaining the expansion weight coefficient includes: obtaining the expansion weight coefficient according to the expansion weight coefficient calculation formula, and the expansion weight coefficient calculation formula is as follows: In the formula, Denote the expansion weight coefficient of the reference lesion area; Denote the number of pixel points in the reference lesion area; Denote the lesion intensity index of the th pixel point in the reference lesion area; Denote the number of other lesion areas in the lesion blood vessel image where the reference lesion area is located; Denote the shape feature factor of the reference lesion area; Denote the th shape feature factor of other lesion areas in the lesion blood vessel image where the reference lesion area is located; Denote the structure feature factor of the reference lesion area; Denote the th structure feature factor of other lesion areas in the lesion blood vessel image where the reference lesion area is located; Denote the range feature factor of the reference lesion area; Indicates the range feature factor of the th other lesion area in the lesion vessel image where the reference lesion area is located; Indicates the absolute value function.

[0011] Further, the method for obtaining the augmented sampling priority includes: taking any lesion vessel image as the target image; accumulating and summing the lesion intensity indicators of all lesion areas in the target image to obtain the target weight coefficient of the target image; accumulating and summing the lesion intensity indicators of all lesion areas in all lesion vessel images to obtain the total image weight coefficient; taking the ratio between the target weight coefficient and the total image weight coefficient as the augmented sampling priority of the target image; changing the target image to obtain the augmented sampling priority of each lesion vessel image.

[0012] Further, obtaining the augmented training model based on the augmented sampling priority includes: taking the peripheral vessel image without lesion areas as the normal vessel image; enhancing and augmenting all the collected lesion vessel images according to the augmented sampling priority of all the lesion vessel images until the number of the augmented lesion vessel images reaches a preset first quantity, and taking the augmented lesion vessel images and the normal vessel image as the augmented training model.

[0013] A peripheral vessel image assisted segmentation system, the system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the above-mentioned peripheral vessel image assisted segmentation method are implemented.

[0014] The present invention has the following beneficial effects: The present invention collects all peripheral blood vessel images for subsequent segmentation of peripheral blood vessel images using a deep learning segmentation network. Since there are differences in gray scale, structure, and the edges of peripheral blood vessels between the lesion area and other normal areas, the gray scale distribution within the preset neighborhood of lesion pixels and the gradient distribution of peripheral blood vessel edge pixels are used to obtain the lesion intensity index of the lesion pixels. Since different types of peripheral blood vessel lesions will also cause differences in the shape, structure, and scope of peripheral blood vessels, the shape feature factor, structure feature factor, and scope feature factor of each lesion area are analyzed. To improve the segmentation ability of the deep learning segmentation network, it is necessary to focus on expanding the lesion areas with relatively high overall lesion intensity indices and relatively low occurrence frequencies. Therefore, based on the differences in shape feature factors, structure feature factors, and scope feature factors of the reference lesion area compared to other lesion areas, as well as the lesion intensity index of each lesion pixel, the expansion weight coefficient of the reference lesion area is obtained. Since the importance of the lesion areas in each lesion blood vessel image is different when expanding the training samples, the priority of expansion sampling for each lesion blood vessel image is analyzed. An expanded training model is obtained based on the expansion sampling priority. The peripheral blood vessel images are segmented according to the expanded training model. The present invention takes into account the characteristics of different peripheral blood vessel lesions, resulting in balanced and rich expanded peripheral blood vessel samples, increasing the generalization ability of the deep learning segmentation network and improving the accuracy of auxiliary segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is a flowchart of a method for auxiliary segmentation of peripheral blood vessel images provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following, in combination with the accompanying drawings and preferred embodiments, details the specific implementation manners, structures, features, and effects of a method and system for auxiliary segmentation of peripheral blood vessel images proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this invention belongs.

[0019] The following specifically describes the specific solutions of a method and system for assisting in the segmentation of peripheral blood vessel images provided by the present invention in conjunction with the accompanying drawings.

[0020] Please refer to Figure 1 , which shows a method for assisting in the segmentation of peripheral blood vessel images provided by an embodiment of the present invention. The method includes: Step S1: Collect all peripheral blood vessel images; and use the peripheral blood vessel images with lesions as lesion blood vessel images.

[0021] The embodiments of the present invention are mainly applied to the scenario of expanding the sample of peripheral blood vessel images. In order to expand the peripheral blood vessel images, first, the peripheral blood vessel images are obtained, and the peripheral blood vessel images with lesions are sampled and expanded. Since a deep learning segmentation network is used to segment the peripheral blood vessel images in the embodiments of the present invention, and the deep learning segmentation network requires a large number of training samples, that is, peripheral blood vessel images containing lesion regions, the peripheral blood vessel images with lesions are used as lesion blood vessel images, and then the lesion blood vessel images are enhanced and expanded to ensure that there are enough training samples for the deep learning segmentation network.

[0022] In an embodiment of the present invention, a U-Net deep learning model is used to assist in the segmentation of peripheral blood vessel images. It should be noted that in other embodiments of the present invention, other models such as convolutional neural networks (CNNs) can also be used for the same operation, and both the U-Net deep learning model and the convolutional neural network are well-known technical means to those skilled in the art, and will not be limited and described in detail herein.

[0023] In an embodiment of the present invention, since there are significant differences in the gray values of pixel points in the lesion regions of peripheral blood vessels compared to those in other normal regions, and the area of the lesion regions in the peripheral blood vessel images is small, the pixel points in the lesion regions are manually labeled with abnormal labels, and the peripheral blood vessel images with abnormal labels are used as lesion blood vessel images.

[0024] Step S2: Arbitrarily select a diseased pixel point in any one of the diseased blood vessel images as a reference diseased pixel point; obtain the diseased intensity index of the reference diseased pixel point according to the gray distribution within the preset neighborhood of the reference diseased pixel point and the gradient distribution of the peripheral blood vessel edge pixel points; obtain the diseased area in each diseased blood vessel image according to the distribution of the diseased pixel points in each diseased blood vessel image; arbitrarily select a diseased area as a reference diseased area; obtain the shape feature factor of the reference diseased area according to the contour feature of the reference diseased area; obtain the structural feature factor of the reference diseased area according to the gray distribution within the reference diseased area; obtain the range feature factor of the reference diseased area according to the quantity feature of the diseased pixel points within the reference diseased area.

[0025] The peripheral blood vessels of the human body may undergo pathological changes due to various reasons, such as deformation, distortion, blockage, etc. of the peripheral blood vessels. The segmentation effects of diseased blood vessel images containing diseased areas of different degrees are different. In order to achieve better segmentation of all types of peripheral blood vessel diseases, it is necessary to provide sufficient training samples for each type of peripheral blood vessel disease in the U-Net deep learning model. In order to describe different types of peripheral blood vessel diseases, it is necessary to calculate the intensity of the pixel points in the diseased area of each diseased blood vessel image. Since there are significant differences between the gray values of the diseased area and those of other normal areas, and different types of peripheral blood vessel diseases will cause significant differences in the corresponding peripheral blood vessel structures, that is, different types of peripheral blood vessel diseases will cause differences in the peripheral blood vessel edges. Therefore, in the embodiments of the present invention, the diseased intensity index of the reference diseased pixel point is obtained according to the gray distribution within the preset neighborhood of the reference diseased pixel point and the gradient distribution of the peripheral blood vessel edge pixel points.

[0026] Preferably, in an embodiment of the present invention, the method for obtaining the diseased area includes: marking the peripheral blood vessel areas with diseases in each diseased blood vessel image according to the prior art, and taking the pixel points corresponding to the peripheral blood vessel areas with diseases in each diseased blood vessel image as diseased pixel points; using the connected component labeling algorithm to obtain all the connected components formed by all the diseased pixel points in each diseased blood vessel image, and taking all the connected components as all the diseased areas in each diseased blood vessel image.

[0027] Preferably, in an embodiment of the present invention, the method for obtaining the diseased intensity index includes: obtaining the diseased intensity index according to the diseased intensity index calculation formula, and the diseased intensity index calculation formula is as follows: In the formula, represents the diseased intensity index of the reference diseased pixel point; represents the number of pixel points within the preset neighborhood of the reference diseased pixel point; represents the th pixel point within the preset neighborhood of the reference diseased pixel point; Represents the average gray value of the pixels within the preset neighborhood of the reference lesion pixel point; Represents the average gray value of the pixels in the lesion blood vessel image where the reference lesion pixel point is located; Represents the number of peripheral blood vessel edge pixels within the preset neighborhood of the reference lesion pixel point; Represents the gradient direction angle of the nth peripheral blood vessel edge pixel within the preset neighborhood of the reference lesion pixel point, where the gradient direction angle of the peripheral blood vessel edge pixel can be obtained by the prior art and will not be elaborated here; Represents the average gradient direction angle of the peripheral blood vessel edge pixels within the preset neighborhood of the reference lesion pixel point.

[0028] In the calculation formula of the lesion intensity index, the difference between the gray value of each pixel within the preset neighborhood of the reference pixel point and the overall gray value of the preset neighborhood The larger the difference, the more uneven the blood flow around the reference pixel point. At this time, the area where the reference pixel point is located is more likely to have a lesion, that is, the lesion intensity index of the reference pixel point is larger; the difference between the gray value of each pixel within the preset neighborhood of the reference pixel point and the overall gray value of the lesion blood vessel image The larger the difference, the more significant the gray value of the pixels in the preset neighborhood is in the lesion blood vessel image, that is, the greater the difference in blood flow between the preset neighborhood where the reference pixel point is located and other areas, that is, the area where the reference pixel point is located is more likely to have a lesion, that is, the lesion intensity index of the reference pixel point is larger; The larger the difference, the greater the change in the gradient direction of the peripheral blood vessel edge pixels, that is, the more likely the peripheral blood vessel where the reference pixel point is located has a structural change. At this time, the lesion intensity index of the peripheral blood vessel area corresponding to the reference pixel point is larger.

[0029] In an embodiment of the present invention, a rectangular area with centered on the reference pixel point is set as the preset neighborhood of the reference pixel point. It should be noted that in other embodiments of the present invention, the preset neighborhood can be set by itself and is not limited here.

[0030] Due to the fact that in reality, different types of peripheral blood vessel lesions will cause differences in the shape, structure, and range of the peripheral blood vessels. In order to ensure the stability of the sample quantity of various types of peripheral blood vessel lesions when expanding the training samples of the deep learning segmentation network in the future, in the embodiments of the present invention, the shape characteristics, structural characteristics, and range characteristics of each lesion area are analyzed.

[0031] Since the lesion areas of the peripheral blood vessels usually have large shape differences and may present elliptical, linear, branched, etc., in an embodiment of the present invention, an arbitrary lesion area is selected as the reference lesion area, and according to the contour characteristics of the reference lesion area, the shape characteristic factor of the reference lesion area is obtained.

[0032] Preferably, in an embodiment of the present invention, the method for obtaining the shape feature factor includes: obtaining all edge pixel points and the centroid pixel point of the reference lesion area.

[0033] Obtain the connection lines between the centroid pixel point and all edge pixel points as the shape feature line segments.

[0034] Taking the centroid pixel point as the center, evenly divide the reference lesion area along a preset angle to obtain all lesion modules. In an embodiment of the present invention, the preset angle is set to 30°, that is, taking the centroid pixel point as the center, rotate the longest shape feature line segment in the reference lesion area, and take the reference lesion area passed by the longest shape feature line segment every time it rotates 30° as each lesion module. Thus, 12 lesion modules can be obtained for the reference lesion area. It should be noted that in other embodiments of the present invention, the preset angle can be set by itself and is not limited herein.

[0035] Select the longest shape feature line segment and the shortest shape feature line segment in each lesion module, take the angle formed by every two adjacent longest shape feature line segments of all lesion modules as the first shape feature angle, and take the angle formed by every two adjacent shortest shape feature line segments of all lesion modules as the second shape feature angle; according to the above process, in an embodiment of the present invention, the number of the first shape feature angles is the same as that of the second shape feature angles, and each has 11.

[0036] Obtain the shape feature factor of the reference lesion area according to the first shape feature angle and the second shape feature angle, and the calculation formula is as follows: In the formula, represents the shape feature factor of the reference lesion area; represents the number of the first shape feature angles; represents the number of the second shape feature angles, and ; represents the th first shape feature angle and the th first shape feature angle, and the first-order difference between their angles; represents the th first shape feature angle and the th first shape feature angle, and the first-order difference between their angles; represents the absolute value function.

[0037] In the calculation formula of the shape feature factor, the included angle formed by the adjacent longest shape feature line segments of different lesion modules can reflect the contour feature distribution of the edge of the reference lesion area far from the centroid pixel point, and the included angle formed by the adjacent shortest shape feature line segments can reflect the contour feature distribution of the reference lesion area close to the centroid pixel point; the average of the sum of the first-order differences of the angles of the included angles formed by all adjacent longest shape feature line segments, that is, the first shape feature included angle, and the first-order differences of the angles of the included angles formed by the adjacent shortest shape feature line segments, that is, the second shape feature included angle, is obtained to obtain the shape feature factor of the reference lesion area.

[0038] In the actual situation, the pathological changes of peripheral blood vessels may be affected by cysts, calcifications, inflammations, etc. At this time, the tissue composition density of peripheral blood vessels is different, which is reflected in the gray-scale distribution of the lesion area in the peripheral blood vessel image. Therefore, in the embodiments of the present invention, according to the gray-scale distribution in the reference lesion area, the structural feature factor of the reference lesion area is obtained.

[0039] Preferably, in an embodiment of the present invention, the method for obtaining the structural feature factor includes: obtaining the structural feature factor according to the calculation formula of the structural feature factor, and the calculation formula of the structural feature factor is as follows: In the formula, represents the structural feature factor of the reference lesion area; represents the number of shape feature line segments in the reference lesion area; represents the th shape feature line segment in the reference lesion area; represents the th shape feature line segment in the reference lesion area; represents the th shape feature line segment in the reference lesion area; represents the th shape feature line segment in the reference lesion area; th pixel point in the represents the th shape feature line segment in the reference lesion area; th pixel point in the represents the absolute value function.

[0040] In the calculation formula of the structural feature factor, the average gray-scale difference between two adjacent shape feature line segments in the reference lesion area is averaged. The larger the average value , the greater the overall gray-scale change of the reference lesion area; the average of the gray-scale differences between every two adjacent pixel points in each shape feature line segment in the reference lesion area is averaged , where the larger the mean value, the greater the gray-scale change of the centroid pixel point to the edge of the reference lesion area; therefore, is used as the structural feature factor of the reference lesion area.

[0041] Since there may be significant differences in the areas of different types of peripheral vascular lesion areas, in the embodiments of the present invention, according to the number feature of the lesion pixel points in the reference lesion area, the range feature factor of the reference lesion area is obtained.

[0042] Preferably, in one embodiment of the present invention, the method for obtaining the range feature factor includes: obtaining the range feature factor according to the range feature factor calculation formula, and the range feature factor calculation formula is as follows: In the formula, represents the range feature factor of the reference lesion area; represents the number of pixel points in the reference lesion area; represents the number of pixel points in all lesion areas in the lesion blood vessel image where the reference lesion area is located; represents the number of pixel points in the lesion blood vessel image where the reference lesion area is located.

[0043] In the range feature factor calculation formula, calculate the ratio of the number of pixel points in the reference lesion area to all lesion areas in the peripheral blood vessel image , and the ratio of the number of pixel points in the reference lesion area to the number of all pixel points in the peripheral blood vessel image . The larger the ratio, the more prominent the range of the reference lesion area in all lesion areas. Therefore, is used as the range feature factor of the reference lesion area.

[0044] Step S3: Obtain the expansion weight coefficient of the reference lesion area according to the shape feature factor difference, structural feature factor difference, range feature factor difference of the reference lesion area compared with other lesion areas, and the lesion intensity index of each lesion pixel point; obtain the expansion sampling priority of each lesion blood vessel image according to the distribution feature of the expansion weight coefficients of all lesion areas in each lesion blood vessel image.

[0045] Obtain the shape feature factor, structure feature factor, range feature factor of each lesion area and the lesion intensity index of each pixel point within the lesion area according to the above process; in order to improve the segmentation ability of the deep learning segmentation network, when expanding the samples, it is necessary to focus on expanding the lesion areas with relatively high overall lesion intensity indexes and relatively low occurrence frequencies, so as to enhance the balance and richness of the final samples in the deep learning segmentation network. Therefore, according to the differences in shape feature factors, structure feature factors, range feature factors between the reference lesion area and other lesion areas, and the lesion intensity index of each lesion pixel point, obtain the expansion weight coefficient of the reference lesion area.

[0046] Preferably, in one embodiment of the present invention, the method for obtaining the expansion weight coefficient includes: obtaining the expansion weight coefficient according to the expansion weight coefficient calculation formula, and the expansion weight coefficient calculation formula is as follows: In the formula, represents the expansion weight coefficient of the reference lesion area; represents the number of pixel points within the reference lesion area; represents the th pixel point within the reference lesion area; represents the number of other lesion areas in the lesion blood vessel image where the reference lesion area is located; represents the shape feature factor of the reference lesion area; represents the th shape feature factor of other lesion areas in the lesion blood vessel image where the reference lesion area is located; represents the structure feature factor of the reference lesion area; represents the th structure feature factor of other lesion areas in the lesion blood vessel image where the reference lesion area is located; represents the range feature factor of the reference lesion area; represents the th range feature factor of other lesion areas in the lesion blood vessel image where the reference lesion area is located; represents the absolute value function.

[0047] In the expansion weight coefficient calculation formula, calculate the total lesion intensity index of the reference lesion area, where the larger the total lesion intensity index, the more serious the lesion in the reference area. At this time, more attention should be paid to the reference lesion area during the expansion of the training samples, that is, the higher the weight coefficient of the reference lesion area during expansion; calculate the differences in shape feature factors, structure feature factors, range feature factors between the reference lesion area and all other lesion areas and take the average , where the greater the mean difference is, the more unique the lesion in the reference lesion area is, and the lower the occurrence frequency of the peripheral vascular lesion corresponding to the reference lesion area is. At this time, when expanding the training samples, it is more necessary to consider the uniqueness of the reference lesion area, that is, the higher the weight coefficient of the reference lesion area during expansion.

[0048] Since the importance of the lesion areas in each lesion vascular image is different when expanding the training samples, it is necessary to analyze the expansion weight coefficients of all lesion areas in each lesion vascular image during sample expansion. That is, in the embodiments of the present invention, according to the distribution characteristics of the expansion weight coefficients of all lesion areas in each lesion vascular image, the expansion sampling priority degree of each lesion vascular image is obtained, so as to pay attention to more important lesion vascular images in the follow-up.

[0049] Preferably, in an embodiment of the present invention, the method for obtaining the expansion sampling priority degree includes: taking any lesion vascular image as the target image; accumulating and summing the lesion intensity indicators of all lesion areas in the target image to obtain the target weight coefficient of the target image; accumulating and summing the lesion intensity indicators of all lesion areas in all lesion vascular images to obtain the total image weight coefficient; taking the ratio between the target weight coefficient and the total image weight coefficient as the expansion sampling priority degree of the target image; changing the target image to obtain the expansion sampling priority degree of each lesion vascular image.

[0050] In an embodiment of the present invention, the calculation formula for the expansion sampling priority degree includes: In the formula, represents the expansion sampling priority degree of the target image; represents the number of lesion areas in the target image; represents the th expansion weight coefficient of the lesion area in the target image; represents the number of lesion areas in all lesion vascular images; represents the th expansion weight coefficient of the lesion area in all lesion vascular images.

[0051] In the calculation formula for the expansion sampling priority degree, the greater the proportion of the sum of the expansion weight coefficients of all lesion areas in the target image to the sum of the expansion weight coefficients of the lesion areas in all lesion vascular images, the more important the target image is, and the higher the priority of being extracted during the expansion of the training samples. At this time, more sampling should be performed on the target image during the enhanced expansion of the training samples, that is, the greater the expansion sampling priority degree.

[0052] Step S4: Obtain the expanded training model based on the expansion sampling priority degree; segment the peripheral vascular image according to the expanded training model.

[0053] Preferably, in one embodiment of the present invention, obtaining an augmented training model based on the augmented sampling priority includes: using the peripheral blood vessel image without the lesion area as the normal blood vessel image; calculating the ratio between the augmented sampling priority of each diseased blood vessel image and the augmented sampling priorities of all diseased blood vessel images as the first ratio, and performing sampling augmentation on all diseased blood vessel images so that the ratio between various types of newly added training samples conforms to the first ratio. When the number of augmented diseased blood vessel images reaches a preset first quantity, using the augmented diseased blood vessel images and the normal blood vessel images as all the augmented peripheral blood vessel images. In one embodiment of the present invention, the preset first quantity is set to 25% of all the augmented peripheral blood vessel images. It should be noted that in other embodiments of the present invention, the preset first quantity can be set by oneself and is not limited herein.

[0054] Using all the augmented peripheral blood vessel images to train the U-Net model, and obtaining a model with better segmentation effect for various types of diseased peripheral blood vessels as the augmented training model. Using the augmented training model to perform segmentation processing on the peripheral blood vessel image, maximizing the problem of incomplete segmentation results caused by the presence of vascular lesions, improving the accuracy of auxiliary segmentation, and completing the auxiliary segmentation of the peripheral blood vessel image.

[0055] Thus, the segmentation of the peripheral blood vessel image is completed.

[0056] In summary, collect all peripheral blood vessel images; use the diseased peripheral blood vessel images as the diseased blood vessel images; select an arbitrary diseased pixel point in any one of the diseased blood vessel images as the reference diseased pixel point; obtain the lesion intensity index of the reference diseased pixel point according to the gray distribution within the preset neighborhood of the reference diseased pixel point and the gradient distribution of the peripheral blood vessel edge pixel points; obtain the lesion area in each diseased blood vessel image according to the distribution of the diseased pixel points in each diseased blood vessel image; select an arbitrary lesion area as the reference lesion area; obtain the shape feature factor of the reference lesion area according to the contour feature of the reference lesion area; obtain the structural feature factor of the reference lesion area according to the gray distribution within the reference lesion area; obtain the range feature factor of the reference lesion area according to the number feature of the diseased pixel points within the reference lesion area; obtain the augmented weight coefficient of the reference lesion area according to the differences in the shape feature factor, structural feature factor, range feature factor between the reference lesion area and other lesion areas, and the lesion intensity index of each diseased pixel point; obtain the augmented sampling priority of each diseased blood vessel image according to the distribution feature of the augmented weight coefficients of all the lesion areas in each diseased blood vessel image; obtain the augmented training model based on the augmented sampling priority; perform segmentation on the peripheral blood vessel image according to the augmented training model.

[0057] An embodiment of the present invention further provides a peripheral vascular image assisted segmentation system, which includes a memory, a processor, and a computer program, where the memory is used to store the corresponding computer program, the processor is used to run the corresponding computer program, and when the computer program runs in the processor, it can implement the methods described in steps S1-S4.

[0058] It should be noted that: the above sequence of embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the particular order or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0059] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. The key point of each embodiment is to illustrate the differences from other embodiments.

Claims

1. A peripheral vascular image-assisted segmentation method, characterized in that: The method comprises: collecting all peripheral vascular images; taking the peripheral vascular images with lesions as lesion vascular images; selecting a lesion pixel point in any lesion vascular image as a reference lesion pixel point; obtaining a lesion intensity index of the reference lesion pixel point according to the grayscale distribution in a preset neighborhood of the reference lesion pixel point and the gradient distribution of the peripheral vascular edge pixel points; obtaining the lesion area in each lesion vascular image according to the lesion pixel distribution of each lesion vascular image; selecting a lesion area as a reference lesion area; obtaining a shape feature factor of the reference lesion area according to the contour feature of the reference lesion area; obtaining a structural feature factor of the reference lesion area according to the grayscale distribution in the reference lesion area; obtaining a structural feature factor of the reference lesion area according to the lesion in the reference lesion area According to the number of pixel points, the range characteristic factor of the reference lesion area is obtained; according to the difference in shape characteristic factors, structural characteristic factors, range characteristic factors and the lesion intensity index of each lesion pixel point between the reference lesion area and other lesion areas, the expansion weight coefficient of the reference lesion area is obtained; according to the distribution characteristics of the expansion weight coefficients of all lesion areas in each lesion vessel image, the expansion sampling priority of each lesion vessel image is obtained; based on the expansion sampling priority, an expanded training model is obtained; according to the expanded training model, the peripheral vascular image is segmented; the method for obtaining the lesion intensity index includes: obtaining the lesion intensity index according to the lesion intensity index calculation formula, and the lesion intensity index calculation formula is as follows: In the formula, represents the lesion intensity index of the reference lesion pixel; Represents the number of pixels within a preset neighborhood of a reference lesion pixel; Indicates the first pixel in the preset neighborhood of the reference lesion pixel. The gray value of each pixel; Represents the grayscale mean of the pixels within the preset neighborhood of the reference lesion pixel; Represents the grayscale mean of the pixel points in the lesion vessel image where the reference lesion pixel point is located; Represents the number of peripheral blood vessel edge pixels within a preset neighborhood of a reference lesion pixel; Represents the first The gradient direction angle of the peripheral blood vessel edge pixel point; Represents the mean value of the gradient direction angle of the peripheral blood vessel edge pixel points within the preset neighborhood of the reference lesion pixel point.

2. The peripheral vascular image-assisted segmentation method according to claim 1, characterized in that: The method for acquiring the lesion area includes: taking the pixel points corresponding to the peripheral blood vessel area where the lesion occurs in each lesion blood vessel image as the lesion pixel points; using a connected domain labeling algorithm to obtain all connected domains formed by all lesion pixel points in each lesion blood vessel image, and taking all the connected domains as all lesion areas in each lesion blood vessel image.

3. The peripheral vascular image-assisted segmentation method according to claim 1, characterized in that: The method for obtaining the shape characteristic factor includes: obtaining all edge pixel points and centroid pixel points of the reference lesion area; obtaining the connection line between the centroid pixel point and all edge pixel points as a shape characteristic line segment; taking the centroid pixel point as the center, equally dividing the reference lesion area along a preset angle to obtain all lesion modules; selecting the longest shape characteristic line segment and the shortest shape characteristic line segment in each lesion module, and taking the angle formed by each adjacent two longest shape characteristic line segments of all lesion modules as the first shape characteristic angle, and taking the angle formed by each adjacent two shortest shape characteristic line segments of all lesion modules as the second shape characteristic angle; according to the first shape characteristic angle and the second shape characteristic angle, the shape characteristic factor of the reference lesion area is obtained, and the calculation formula is as follows: In the formula, represents the shape characteristic factor of the reference lesion area; represents the number of the first shape feature angle; represents the number of the second shape feature angle, and ; Indicates The angle between the first shape feature angle and the The first order difference between the angles of the first shape feature angles; Indicates The angle between the first shape feature angle and the The first order difference between the angles of the first shape feature angles; represents the absolute value function.

4. The peripheral vascular image-assisted segmentation method according to claim 3, characterized in that: The method for obtaining the structural characteristic factor includes: obtaining the structural characteristic factor according to a structural characteristic factor calculation formula, and the structural characteristic factor calculation formula is as follows: In the formula, represents the structural characteristic factor of the reference lesion area; Indicates the number of shape feature line segments in the reference lesion area; Indicates the first The grayscale mean of the pixels in the bar-shaped feature segment; Indicates the first The grayscale mean of the pixels in the bar-shaped feature segment; Indicates the first The number of pixels in the bar shape feature segment; Indicates the first The first of the shape feature segments The gray value of each pixel; Indicates the first The first of the shape feature segments The gray value of each pixel; represents the absolute value function.

5. The peripheral vascular image-assisted segmentation method according to claim 1, characterized in that: The method for obtaining the range characteristic factor includes: obtaining the range characteristic factor according to a range characteristic factor calculation formula, and the range characteristic factor calculation formula is as follows: In the formula, represents the range characteristic factor of the reference lesion area; represents the number of pixels in the reference lesion area; Indicates the number of pixels in all lesion areas in the lesion vessel image where the reference lesion area is located; Represents the number of pixels in the lesion vessel image where the reference lesion area is located.

6. The peripheral vascular image-assisted segmentation method according to claim 1, characterized in that: The method for obtaining the expanded weight coefficient includes: obtaining the expanded weight coefficient according to an expanded weight coefficient calculation formula, and the expanded weight coefficient calculation formula is as follows: In the formula, represents the expansion weight coefficient of the reference lesion area; represents the number of pixels in the reference lesion area; Indicates the first The lesion intensity index of each pixel; Indicates the number of other lesion areas in the lesion vessel image where the reference lesion area is located; represents the shape characteristic factor of the reference lesion area; Indicates the first Shape characteristic factors of other lesion areas; represents the structural characteristic factor of the reference lesion area; Indicates the first structural characteristic factors of other lesion areas; represents the range characteristic factor of the reference lesion area; Indicates the first Other range characteristics of the lesion area; represents the absolute value function.

7. The peripheral vascular image-assisted segmentation method according to claim 1, characterized in that: The method for obtaining the extended sampling priority includes: taking any diseased blood vessel image as a target image; accumulating and summing the lesion intensity indices of all lesion areas in the target image to obtain a target weight coefficient of the target image; accumulating and summing the lesion intensity indices of all lesion areas in all lesion blood vessel images to obtain a total image weight coefficient; taking the ratio between the target weight coefficient and the total image weight coefficient as the extended sampling priority of the target image; and changing the target image to obtain the extended sampling priority of each lesion blood vessel image.

8. The peripheral vascular image-assisted segmentation method according to claim 1, characterized in that: The expanded training model is obtained based on the expansion sampling priority, including: taking the peripheral vascular image that does not include the lesion area as the normal vascular image; enhancing and expanding all the acquired lesion vascular images according to the expansion sampling priority of all the lesion vascular images, until the expanded lesion vascular images reach a preset first number, and taking the expanded lesion vascular images and the normal vascular images as the expanded training model.

9. A peripheral vascular image-assisted segmentation system, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of a peripheral vascular image-assisted segmentation method as described in any one of claims 1 to 8 are implemented.

Images