A 3D microvascular structure segmentation method and system based on degraded micro-CT images
Through the Unnet network combining data augmentation, feature extraction and attention modules, the weighted loss function is used to process degraded micro CT images, which solves the problem of insufficient accuracy and generalization ability of microvascular structure segmentation in the prior art, and achieves more accurate and continuous vascular segmentation effect.
Patent Information
- Application Number
- CN202410717240.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-04
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-06-04
AI Technical Summary
When processing degraded microCT images, it is difficult to accurately extract the characteristics and details of microvascular structures, resulting in poor segmentation results and weak generalization capabilities, especially in the face of vascular structures, noise and artifacts of multiple scales. Not effective when disturbed.
The Unet network is used to perform 3D microvascular structure segmentation, and the combination of data augmentation, feature extraction, maximum pooling layer, attention module and weighted loss function (Focal loss and Dice loss) can realize the feature extraction and segmentation of images to generate accurate binary values. Segmented images and perform 3D structure reconstruction.
Improve the accuracy and edge continuity of segmentation results, enhance the generalization ability of the network, reduce the impact of artifacts, and ensure the segmentation effect in complex edge areas, especially in the case of uneven sample. .
Smart Images

Figure CN118587234B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the field of image processing technology, and in particular to a 3D microvascular structure segmentation method and system based on degraded micro-CT images. Background Art
[0002] Traditional CT image vascular segmentation methods include threshold segmentation, edge detection, watershed algorithm, etc. Among them, the threshold segmentation method is simple to calculate and has the advantage of high efficiency, but it only considers the characteristics of the pixel gray value itself, so it is easily disturbed by artifacts; the edge detection method can quickly and accurately locate the segmentation edge, but it cannot guarantee the continuity of the edge, and often has poor results in high-detail areas with a large number of broken edges; the segmentation edge continuity of the watershed algorithm is good, but it is easy to over-segment the background as the foreground target. When the image data contains vascular structures of various scales and the CT image is degraded by noise, artifacts, etc., the existing traditional methods often have poor generalization ability and are difficult to accurately extract the characteristics and details of the microvascular structure, so it is difficult to obtain good segmentation results. Summary of the invention
[0003] To this end, an embodiment of the present invention provides a 3D microvascular structure segmentation method and system based on degraded micro-CT images to solve the technical problems of poor generalization ability of the prior art, difficulty in accurately extracting the features and details of the microvascular structure, and poor segmentation results.
[0004] In order to achieve the above purpose, the embodiment of the present invention provides the following technical solutions:
[0005] According to a first aspect of an embodiment of the present invention, a 3D microvascular structure segmentation method based on a degraded micro-CT image is provided, the method being applied to a Unet network and comprising:
[0006] S1, obtaining at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set;
[0007] S2, inputting the degraded micro-CT images in the data set into the Unet network, subjecting the degraded micro-CT images in the data set to feature extraction by two layers of feature extraction modules, outputting a first feature map and inputting the feature map into a maximum pooling layer, wherein the maximum pooling layer performs pooling processing on the feature map and continues to input the feature map into the next feature extraction module for feature extraction;
[0008] S3, repeating step S2 until the output second feature map is processed by four pooling processes and the second feature map is input to the image restoration module, the image restoration module expands the second feature map to a preset size and inputs the expanded third feature map to the attention module;
[0009] S4, inputting the first feature map, the second feature map after the feature extraction module and the third feature map after the image restoration module into the attention module for fusion to form a fourth feature map, and inputting the fourth feature map into the image restoration module, the image restoration module splicing the feature map before inputting into the attention module and the fourth feature map, and inputting the spliced image into the next layer image restoration module;
[0010] S5, performing convolution processing on the spliced image output by the last layer image restoration module and activating it using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel, and generating a network binary segmentation image according to the probability that each pixel belongs to a blood vessel;
[0011] S6. Reconstruct the 3D structure by using the binarized segmentation images of each network to generate a 3D microvascular structure.
[0012] Further, obtaining at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set includes:
[0013] The data enhancement processing includes: performing horizontal flipping, vertical flipping, and horizontal flipping plus vertical flipping on the degraded micro-CT image and the binary segmented image respectively.
[0014] Furthermore, the spliced image output by the last layer of image restoration module is convolved and activated using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel. According to the probability that each pixel belongs to a blood vessel, a network prediction binary segmentation image is generated, including:
[0015] Obtaining the probability that the current pixel belongs to a blood vessel, and determining whether the probability that the current pixel belongs to a blood vessel is less than a preset threshold;
[0016] If the probability that the current pixel belongs to a blood vessel is less than a preset threshold, the current pixel is the background, the current pixel value is modified to 0, and the next pixel is determined;
[0017] If the probability that the current pixel belongs to a blood vessel is greater than or equal to the preset threshold, the current pixel is a blood vessel, the current pixel value is modified to 255, and the next pixel is determined;
[0018] After traversing all pixels, the network predicts a binary segmentation image.
[0019] Furthermore, the preset threshold is 0.5.
[0020] Preferably, a 3D microvascular structure segmentation method based on degraded micro-CT images further includes:
[0021] The network is used to predict the binary segmentation image and the real binary segmentation image to calculate the loss function and obtain the loss function value;
[0022] The loss function value is combined with the back propagation algorithm to update various parameters in the Unet network and optimize the network prediction results;
[0023] Among them, the loss functions are Focalloss and Diceloss.
[0024] Furthermore, the expression of the Focalloss is:
[0025]
[0026] Among them, x is any pixel, p is the predicted probability value of x taking 1, α is 0.25, and γ is 2.
[0027] Furthermore, the expression of Diceloss is:
[0028]
[0029] Among them, X is the real segmentation image, To predict the segmented image;
[0030] The expression of the loss function is:
[0031] LoSS0.5×Focal loss+0.5×Dice loss
[0032] Among them, Loss represents the loss function.
[0033] According to a second aspect of an embodiment of the present invention, a 3D microvascular structure segmentation system based on degraded micro-CT images is provided, the system comprising:
[0034] A preprocessing unit, used for acquiring at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set;
[0035] A feature extraction unit is used to input the degraded micro-CT images in the data set into the Unet network, so that the degraded micro-CT images in the data set are subjected to feature extraction by two layers of feature extraction modules, output a first feature map and input the feature map into a maximum pooling layer, and the maximum pooling layer performs pooling processing on the feature map and continues to input the feature map into a next feature extraction module for feature extraction;
[0036] An image enlargement unit, configured to repeat step S2 until the output second feature map is subjected to four pooling processes and the second feature map is input to an image restoration module, the image restoration module enlarges the second feature map to a preset size and inputs the enlarged third feature map to an attention module;
[0037] A splicing unit, used for inputting the first feature map and the second feature map after the feature extraction module and the third feature map after the image restoration module into the attention module for fusion to form a fourth feature map and inputting the fourth feature map into the image restoration module, the image restoration module splices the feature map before inputting into the attention module and the fourth feature map and inputs the spliced image into the image restoration module of the next layer;
[0038] The prediction unit is used to perform convolution processing on the spliced image output by the last layer of image restoration module and activate it using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel, and generate a binary segmentation image according to the probability that each pixel belongs to a blood vessel;
[0039] The 3D microvascular structure reconstruction unit is used to perform 3D structure reconstruction using each of the network binary segmentation images to generate a 3D microvascular structure.
[0040] The embodiments of the present invention have the following advantages:
[0041] The embodiment of the present invention utilizes convolutional layers and maximum pooling layers to collect complex image features, and adopts a supervised training strategy to enable a neural network to learn the mapping relationship between degraded images and known real data, and ultimately achieves that the segmentation result on the test set is not affected by artifacts, thereby improving the accuracy of segmentation; Focal loss and Dice loss are selected as weighted loss functions, Focal loss assigns different weights to the target and background in the process of calculating the loss, and Dice loss evaluates the overlap between the prediction result and the label, thereby also achieving a higher segmentation accuracy in images with an unbalanced target and background, and at the same time introducing an attention module to enable the network to focus more on the target area, so that the segmentation edges can be continuous and accurate even in images with complex edges. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.
[0043] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.
[0044] Figure 1 A schematic diagram of the logical structure of a 3D microvascular structure segmentation system based on degraded micro-CT images provided by an embodiment of the present invention;
[0045] Figure 2 A schematic flow chart of a 3D microvascular structure segmentation method based on degraded micro-CT images provided by an embodiment of the present invention;
[0046] Figure 3 A schematic diagram of data flow in a 3D microvascular structure segmentation method based on degraded micro-CT images provided by an embodiment of the present invention;
[0047] Figure 4 A schematic diagram of a process of 3D structure reconstruction in a 3D microvascular structure segmentation method based on degraded micro-CT images provided by an embodiment of the present invention;
[0048] Figure 5 A schematic diagram of a feature extraction module in a 3D microvascular structure segmentation method based on degraded micro-CT images provided by an embodiment of the present invention;
[0049] Figure 6 A schematic diagram of an attention module in a 3D microvascular structure segmentation method based on degraded micro-CT images provided by an embodiment of the present invention;
[0050] Figure 7 A schematic diagram of an image restoration module in a 3D microvascular structure segmentation method based on degraded micro-CT images provided by another embodiment of the present invention;
[0051] Figure 8 A schematic diagram of an output prediction module in a 3D microvascular structure segmentation method based on degraded micro-CT images provided by another embodiment of the present invention;
[0052] Fig. 9 A schematic diagram of experimental results of a 3D microvascular structure segmentation method based on degraded micro-CT images provided in another embodiment of the present invention. DETAILED DESCRIPTION
[0053] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0054] Traditional CT image vascular segmentation methods include threshold segmentation, edge detection, watershed algorithm, etc. Among them, the threshold segmentation method is simple to calculate and has the advantage of high efficiency, but it only considers the characteristics of the pixel gray value itself, so it is easily disturbed by artifacts; the edge detection method can quickly and accurately locate the segmentation edge, but it cannot guarantee the continuity of the edge, and often has poor results in high-detail areas with a large number of broken edges; the segmentation edge continuity of the watershed algorithm is good, but it is easy to over-segment the background as the foreground target. When the image data contains vascular structures of various scales and the CT image is degraded by noise, artifacts, etc., the existing traditional methods often have poor generalization ability and are difficult to accurately extract the characteristics and details of the microvascular structure, so it is difficult to obtain good segmentation results.
[0055] Traditional methods often determine one or more thresholds through grayscale differences, and then set image pixels to 0 or 1 according to the thresholds, ultimately obtaining a binary segmented image. When such methods are applied to CT images, they are easily affected by degradation conditions such as noise and artifacts, resulting in poor segmentation accuracy and discontinuous segmentation edges.
[0056] In order to solve the above-mentioned technical problems of poor generalization ability, difficulty in accurately extracting the characteristics and details of microvascular structures, and poor segmentation results caused by determining the threshold value for segmenting images through grayscale differences.
[0057] refer to Figure 1 The embodiment of the present invention discloses a 3D microvascular structure segmentation system based on degraded micro-CT images, the system comprising: a preprocessing unit 1; a feature extraction unit 2; an image enlargement unit 3; a splicing unit 4; a prediction unit 5; and a 3D microvascular structure reconstruction unit 6.
[0058] Corresponding to the above-disclosed 3D microvascular structure segmentation system based on degraded micro-CT images, the embodiment of the present invention further discloses a 3D microvascular structure segmentation method based on degraded micro-CT images. The following describes in detail the 3D microvascular structure segmentation method based on degraded micro-CT images disclosed in the embodiment of the present invention in combination with the above-described 3D microvascular structure segmentation system based on degraded micro-CT images.
[0059] refer to Figure 2 and Figure 3The present invention discloses a 3D microvascular structure segmentation method based on degraded micro-CT images, the method is applied to a Unet network, and comprises:
[0060] S1, obtaining at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set;
[0061] refer to Figure 5 S2, inputting the degraded micro-CT images in the data set into the Unet network, subjecting the degraded micro-CT images in the data set to feature extraction by two layers of feature extraction modules, outputting a first feature map and inputting the feature map into a maximum pooling layer, wherein the maximum pooling layer performs pooling processing on the feature map and continues to input the feature map into the next feature extraction module for feature extraction;
[0062] S3, repeating step S2 until the output second feature map is processed by four pooling processes and the second feature map is input to the image restoration module, the image restoration module expands the second feature map to a preset size and inputs the expanded third feature map to the attention module;
[0063] refer to Figure 6 and Figure 7 S4, input the first feature map, the second feature map after the feature extraction module and the third feature map after the image restoration module into the attention module for fusion to form a fourth feature map, and input the fourth feature map into the image restoration module, the image restoration module splices the feature map before inputting the attention module and the fourth feature map, and inputs the spliced image into the next layer image restoration module;
[0064] refer to Figure 8 , S5, convolution processing is performed on the spliced image output by the last layer of image restoration module and activated using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel, and a network binary segmentation image is generated according to the probability that each pixel belongs to a blood vessel;
[0065] refer to Figure 4 , S6, using each of the network binarization segmentation images to perform 3D structure reconstruction to generate a 3D microvascular structure.
[0066] Furthermore, at least one set of paired degraded micro-CT images and true binary segmented images are obtained and data enhancement processing is performed on the degraded micro-CT images and the binary segmented images to expand the data set, including: the data enhancement processing includes: horizontally flipping, vertically flipping, and horizontally flipping plus vertically flipping the degraded micro-CT images and the binary segmented images respectively.
[0067] Furthermore, the spliced image output by the last layer of image restoration module is convolved and activated by a Sigmoid function to obtain the probability that each pixel belongs to a blood vessel, and a network-predicted binary segmentation image is generated according to the probability that each pixel belongs to a blood vessel, including: obtaining the probability that the current pixel belongs to a blood vessel, and judging whether the probability that the current pixel belongs to a blood vessel is less than a preset threshold; if the probability that the current pixel belongs to a blood vessel is less than the preset threshold, the current pixel is the background, the current pixel value is modified to 0, and the next pixel is judged; if the probability that the current pixel belongs to a blood vessel is greater than or equal to the preset threshold, the current pixel is a blood vessel, the current pixel value is modified to 255, and the next pixel is judged; and after traversing all pixels, a network-predicted binary segmentation image is generated.
[0068] Furthermore, the preset threshold is 0.5.
[0069] Preferably, a 3D microvascular structure segmentation method based on degraded micro-CT images also includes: using the network to predict the binary segmentation image and the real binary segmentation image to calculate the loss function to obtain the loss function value; combining the loss function value with the back propagation algorithm to update each parameter in the Unet network and optimize the network prediction result; wherein the loss function is Focalloss and Diceloss.
[0070] Furthermore, the expression of the Focalloss is:
[0071]
[0072] Among them, x is any pixel, p is the predicted probability value of x taking 1, α is 0.25, and γ is 2.
[0073] Furthermore, the expression of Diceloss is:
[0074]
[0075] Among them, X is the real segmentation image, To predict the segmented image;
[0076] The expression of the loss function is:
[0077] LoSS0.5×Focal loss+0.5×Dice loss
[0078] Among them, Loss represents the loss function.
[0079] refer to Fig. 9, (a) is a CT image under degraded conditions, (b) is a true binary segmentation image, and (c) is a network predicted binary segmentation image according to an embodiment of the present invention. It can be seen that the segmentation edges in (c) are continuous and relatively clear, which proves the effectiveness of the embodiment of the present invention.
[0080] Fig. 9 Figure (d) shows, from left to right, a degraded CT image, a real 3D microvascular structure, and a 3D microvascular structure reconstructed according to the experimental results of an embodiment of the present invention. It can be seen that the reconstructed structure of the degraded CT image is affected by the degradation conditions, while the reconstructed structure of the experimental results of the embodiment of the present invention eliminates the influence of the degradation conditions and has a high similarity with the real 3D microvascular structure.
[0081] The embodiments of the present invention have the following advantages:
[0082] 1) The degraded-binarized paired CT images are used to train the Unet network to obtain better test results. The Unet network can output clear vascular segmentation images according to the input degraded CT images, so that the performance of the network does not depend on the high-quality input images;
[0083] 2) For micro-CT images with a small number of images, data augmentation is performed before the paired images are input into the network for learning, which increases the diversity of the data and improves the generalization ability of the network model;
[0084] 3) For the case of sample imbalance, such as CT images with a small proportion of blood vessel area, the traditional loss function is easily affected by the background other than the blood vessels, resulting in most of the loss function being occupied by background pixels, making the model biased towards the background, and ultimately leading to poor segmentation effect prediction. The weighted loss function used in the embodiment of the present invention combines Focalloss and Dice loss, which can adjust the weights of blood vessels and background in CT images when calculating the loss function value, and can effectively prevent the network model from tilting towards the background;
[0085] 4) The introduction of the attention module can make the network pay more attention to the areas in the CT images that are more valuable for network training and improve the segmentation effect of the network.
[0086] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.
Claims
1. A 3D microvascular structure segmentation method based on degraded micro-CT images, characterized in that: The method is applied to a Unet network, which comprises: S1, obtaining at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set; S2, inputting the degraded micro-CT images in the data set into the Unet network, subjecting the degraded micro-CT images in the data set to feature extraction by two layers of feature extraction modules, outputting a first feature map and inputting the feature map into a maximum pooling layer, wherein the maximum pooling layer performs pooling processing on the feature map and continues to input the feature map into the next feature extraction module for feature extraction; S3, repeating step S2 until the output second feature map is processed by four pooling processes and the second feature map is input to the image restoration module, the image restoration module expands the second feature map to a preset size and inputs the expanded third feature map to the attention module; S4, inputting the first feature map, the second feature map after the feature extraction module and the third feature map after the image restoration module into the attention module for fusion to form a fourth feature map, and inputting the fourth feature map into the image restoration module, the image restoration module splicing the feature map before inputting into the attention module and the fourth feature map, and inputting the spliced image into the next layer image restoration module; S5, performing convolution processing on the spliced image output by the last layer image restoration module and activating it using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel, and generating a network binary segmentation image according to the probability that each pixel belongs to a blood vessel; S6, performing 3D structure reconstruction using each of the network binarized segmentation images to generate a 3D microvascular structure; Acquiring at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set, including: The data enhancement processing includes: horizontally flipping, vertically flipping, and horizontally flipping plus vertically flipping the degraded micro-CT image and the binary segmented image respectively; The spliced image output by the last layer of image restoration module is convolved and activated using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel. Based on the probability that each pixel belongs to a blood vessel, a network prediction binary segmentation image is generated, including: Obtaining the probability that the current pixel belongs to a blood vessel, and determining whether the probability that the current pixel belongs to a blood vessel is less than a preset threshold; If the probability that the current pixel belongs to a blood vessel is less than a preset threshold, the current pixel is the background, the current pixel value is modified to 0, and the next pixel is determined; If the probability that the current pixel belongs to a blood vessel is greater than or equal to the preset threshold, the current pixel is a blood vessel, the current pixel value is modified to 255, and the next pixel is determined; After traversing all pixels, the network predicts a binary segmentation image.
2. A 3D microvascular structure segmentation method based on degraded micro-CT images as claimed in claim 1, characterized in that: The preset threshold is 0.
5.
3. The 3D microvascular structure segmentation method based on degraded micro-CT images according to claim 2, characterized in that: The method further comprises: The network is used to predict the binary segmentation image and the real binary segmentation image to calculate the loss function and obtain the loss function value; The loss function value is combined with the back propagation algorithm to update various parameters in the Unet network and optimize the network prediction results; Among them, the loss functions are Focalloss and Diceloss.
4. The 3D microvascular structure segmentation method based on degraded micro-CT images according to claim 3, characterized in that: The expression of the Focalloss is: Among them, x is any pixel, p is the predicted probability value of x taking 1, α is 0.25, and γ is 2.
5. The 3D microvascular structure segmentation method based on degraded micro-CT images according to claim 4, characterized in that: The expression of Diceloss is: Among them, X is the real segmentation image, To predict the segmented image; The expression of the loss function is: Loss=0.5×Focalloss+0.5×Diceloss Among them, Loss represents the loss function.
6. A 3D microvascular structure segmentation system based on degraded micro-CT images, characterized in that: The system comprises: A preprocessing unit, used for acquiring at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set; A feature extraction unit is used to input the degraded micro-CT images in the data set into the Unet network, so that the degraded micro-CT images in the data set are subjected to feature extraction by two layers of feature extraction modules, output a first feature map and input the feature map into a maximum pooling layer, and the maximum pooling layer performs pooling processing on the feature map and continues to input the feature map into a next feature extraction module for feature extraction; An image enlargement unit, configured to repeat step S2 until the output second feature map is subjected to four pooling processes and the second feature map is input to an image restoration module, the image restoration module enlarges the second feature map to a preset size and inputs the enlarged third feature map to an attention module; A splicing unit, used for inputting the first feature map and the second feature map after the feature extraction module and the third feature map after the image restoration module into the attention module for fusion to form a fourth feature map and inputting the fourth feature map into the image restoration module, the image restoration module splices the feature map before inputting into the attention module and the fourth feature map and inputs the spliced image into the image restoration module of the next layer; The prediction unit is used to perform convolution processing on the spliced image output by the last layer of image restoration module and activate it using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel, and generate a binary segmentation image according to the probability that each pixel belongs to a blood vessel; A 3D microvascular structure reconstruction unit is used to perform 3D structure reconstruction using each network binary segmentation image to generate a 3D microvascular structure; Acquiring at least one set of paired degraded micro-CT images and true binary segmented images and performing data enhancement processing on the degraded micro-CT images and the binary segmented images to expand the data set, including: The data enhancement processing includes: horizontally flipping, vertically flipping, and horizontally flipping plus vertically flipping the degraded micro-CT image and the binary segmented image respectively; The spliced image output by the last layer of image restoration module is convolved and activated using the Sigmoid function to obtain the probability that each pixel belongs to a blood vessel. Based on the probability that each pixel belongs to a blood vessel, a network prediction binary segmentation image is generated, including: Obtaining the probability that the current pixel belongs to a blood vessel, and determining whether the probability that the current pixel belongs to a blood vessel is less than a preset threshold; If the probability that the current pixel belongs to a blood vessel is less than a preset threshold, the current pixel is the background, the current pixel value is modified to 0, and the next pixel is determined; If the probability that the current pixel belongs to a blood vessel is greater than or equal to the preset threshold, the current pixel is a blood vessel, the current pixel value is modified to 255, and the next pixel is determined; After traversing all pixels, the network predicts a binary segmentation image.
Citation Information
Patent Citations
Image processing method and system for CT (Computed Tomography) image of intestinal part
CN115439471A
Positional convolutional attention network-based vascular plaque CT image segmentation method
WO2022247168A1
