Blood vessel image segmentation method and device, equipment, storage medium and program product

Abstract

The invention provides a blood vessel image segmentation method and device, equipment, a storage medium and a program product, and relates to the technical field of image processing. The method comprises the following steps: acquiring an original blood vessel image and an initial blood vessel segmentation image corresponding to the original blood vessel image; based on the blood vessel segmentation image, extracting a blood vessel skeleton structure, and determining endpoints on the blood vessel skeleton structure; superposing the original blood vessel image and the blood vessel segmentation image to obtain a dual-channel image; for each end point, determining a local segmentation window corresponding to the end point, performing iterative blood vessel segmentation on a local image corresponding to the local segmentation window in the dual-channel image, and updating a blood vessel segmentation image according to an iterative blood vessel segmentation result; and traversing all end points to obtain a final blood vessel segmentation image. According to the method, the segmentation precision can be improved, and meanwhile, computing resources are saved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a blood vessel image segmentation method, device, equipment, storage medium and program product. Background Art

[0002] In the field of medical image analysis, accurate vascular segmentation plays an indispensable role in the diagnosis, treatment plan formulation and surgical planning of various diseases. For example, in the diagnosis of cardiovascular diseases, clearly presenting the morphology, distribution and stenosis of blood vessels can assist doctors in accurately judging the condition and formulating appropriate treatment strategies; in tumor treatment, clarifying the distribution of blood vessels around the tumor helps determine the scope of surgical resection or evaluate the feasibility of interventional treatment.

[0003] Common vascular image segmentation methods mainly rely on deep learning models to perform global optimization on the original vascular images, that is, using the basic model to perform preliminary segmentation of medical images to obtain the binary or semantic segmentation results of the blood vessels. Then, the initial segmentation results are input into the overall optimization model for multiple iterative optimizations to gradually improve the quality of vascular segmentation.

[0004] However, this overall iterative optimization method that performs iterative optimization over the entire image cannot accurately locate unsegmented areas and is easily affected by background interference, thereby ignoring the local features of small blood vessels, making it difficult to identify vascular segments in low-contrast or small blood vessel areas, causing some vascular branches to never be correctly segmented, thereby reducing segmentation accuracy. In addition, the overall iterative optimization method of the entire image will cause repeated iterative calculations in areas that have been correctly segmented, resulting in a waste of computing resources. Summary of the invention

[0005] The embodiments of the present invention provide a blood vessel image segmentation method, apparatus, device and storage medium to solve the problems of low segmentation accuracy and waste of computing resources in the blood vessel image segmentation process.

[0006] In a first aspect, an embodiment of the present invention provides a blood vessel image segmentation method, comprising: Acquire an original blood vessel image and its corresponding initial blood vessel segmentation image; Based on the blood vessel segmentation image, a blood vessel skeleton structure is extracted, and endpoints on the blood vessel skeleton structure are determined; the blood vessel skeleton structure includes a plurality of blood vessel centerlines, and the endpoints are endpoints of the blood vessel centerlines; superimposing the original blood vessel image and the blood vessel segmentation image to obtain a dual-channel image; For each endpoint, determine a local segmentation window corresponding to the endpoint, perform iterative blood vessel segmentation on a local image corresponding to the local segmentation window in the dual-channel image, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result; Traverse all endpoints to obtain the final blood vessel segmentation image.

[0007] In a possible implementation, performing iterative blood vessel segmentation on the local image corresponding to the local segmentation window in the dual-channel image, and updating the blood vessel segmentation image according to the iterative blood vessel segmentation result includes: Initialize the number of iterations; Determine a local image corresponding to the local segmentation window in the dual-channel image, and input the local image into a blood vessel segmentation model to obtain a local blood vessel segmentation image; determining whether an iteration termination condition is satisfied according to the local blood vessel segmentation image and the number of iterations; If the iteration termination condition is not met, determining a target endpoint in the local blood vessel segmentation image; the target endpoint is an endpoint farthest from the blood vessel skeleton structure in the local image among all endpoints of the local blood vessel segmentation image; According to the target endpoint, updating the local segmentation window, updating the blood vessel segmentation image and the dual-channel image based on the local blood vessel segmentation image, and updating the current number of iterations; Jump to the step of determining the local image corresponding to the local segmentation window in the dual-channel image until the iteration termination condition is met, determine the current local blood vessel segmentation image as the iterative blood vessel segmentation result, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result.

[0008] In a possible implementation, the iteration termination condition includes: During N consecutive iterations, the newly added blood vessel segmentation area in the local blood vessel segmentation image is smaller than a preset value; N is an integer greater than or equal to 1; or The current number of iterations has reached the maximum limit.

[0009] In a possible implementation manner, determining the target endpoint in the local blood vessel segmentation image includes: Extracting a blood vessel skeleton structure in the local blood vessel segmentation image, and determining each endpoint in the blood vessel skeleton structure; An original blood vessel skeleton structure in a local image corresponding to the local blood vessel segmentation image is determined, and an endpoint farthest from the original blood vessel skeleton structure among all endpoints of the blood vessel skeleton structure in the local blood vessel segmentation image is determined as a target endpoint.

[0010] In a possible implementation, determining, for each endpoint, a local segmentation window corresponding to the endpoint includes: For each endpoint, extracting the actual blood vessel diameter at the location of the endpoint in the blood vessel segmentation image; Respectively obtaining a preset standard window size and a preset standard blood vessel diameter, and determining a proportional relationship between the standard window size and the standard blood vessel diameter; Determining a local window size corresponding to the endpoint based on the proportional relationship and the actual blood vessel diameter; The endpoint is used as the center of the local segmentation window, and the local segmentation window is determined according to the local window size.

[0011] In a possible implementation, extracting a blood vessel skeleton structure based on the blood vessel segmentation image and determining endpoints on the blood vessel skeleton structure includes: For each foreground pixel in the blood vessel segmentation image, detecting whether there is a background pixel in the neighborhood of the foreground pixel; the foreground pixel is a pixel used to characterize the blood vessel, and the background pixel is the remaining pixel except the foreground pixel; If there is a background pixel in the neighborhood of the foreground pixel, the foreground pixel is determined as a candidate pixel; Detect whether the topological structure in the blood vessel segmentation image changes before and after each candidate pixel is deleted; Determine the candidate pixel points whose topological structures in the blood vessel segmentation image before and after deletion are unchanged as target candidate pixel points, and delete them to obtain a new blood vessel segmentation image; Jump to the step of detecting whether there is a background pixel in the neighborhood of each foreground pixel in the blood vessel segmentation image until the target candidate pixel does not exist in the current blood vessel segmentation image, thereby obtaining the blood vessel skeleton structure; For each pixel point on the blood vessel skeleton structure, if the number of foreground pixel points in the neighborhood of the pixel point is 1, the pixel point is determined to be an endpoint.

[0012] In a second aspect, an embodiment of the present invention provides a blood vessel image segmentation device, comprising: An acquisition module, used for acquiring an original blood vessel image and its corresponding initial blood vessel segmentation image; An extraction module, used to extract a blood vessel skeleton structure based on the blood vessel segmentation image, and determine endpoints on the blood vessel skeleton structure; the blood vessel skeleton structure includes a plurality of blood vessel centerlines, and the endpoints are endpoints of the blood vessel centerlines; Segmentation module for: superimposing the original blood vessel image and the blood vessel segmentation image to obtain a dual-channel image; For each endpoint, determine a local segmentation window corresponding to the endpoint, perform iterative blood vessel segmentation on a local image corresponding to the local segmentation window in the dual-channel image, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result; Traverse all endpoints to obtain the final blood vessel segmentation image.

[0013] In a third aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method in the first aspect or any possible implementation of the first aspect is implemented.

[0014] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method in the first aspect or any possible implementation of the first aspect.

[0015] In a fifth aspect, an embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the method in the first aspect or any possible implementation manner of the first aspect.

[0016] In the embodiment of the present invention, the endpoints can be used as the critical points between the segmented area and the unsegmented area in the blood vessel segmentation image. By extracting the blood vessel skeleton structure from the blood vessel segmentation image and then determining the endpoints in the blood vessel skeleton structure, the unsegmented area in the blood vessel segmentation image can be accurately located to reduce the background interference in the blood vessel segmentation image. On this basis, by determining the local image corresponding to the local segmentation window corresponding to each endpoint in the dual-channel image, iterative blood vessel segmentation is performed, and the unsegmented area corresponding to each endpoint can be locally iteratively segmented respectively, thereby improving the segmentation accuracy of small blood vessels, thereby improving the segmentation accuracy, and saving computing resources. In addition, in the embodiment of the present invention, when the dual-channel image of the original blood vessel image and the blood vessel segmentation image is superimposed for local blood vessel segmentation, the segmented area in the blood vessel segmentation image can guide the current local blood vessel segmentation work, thereby improving the robustness of blood vessel segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of an implementation of a blood vessel image segmentation method provided by an embodiment of the present invention; Figure 2 is a schematic diagram of a blood vessel segmentation image provided by an embodiment of the present invention; Figure 3 is a schematic diagram of a blood vessel skeleton structure provided by an embodiment of the present invention; Figure 4 is a flowchart for implementing iterative blood vessel segmentation on a local image provided by an embodiment of the present invention; Figure 5 is a schematic diagram of determining a target endpoint provided by an embodiment of the present invention; Figure 6 is a schematic structural diagram of a blood vessel image segmentation device provided by an embodiment of the present invention; Figure 7 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0019] In the existing vascular segmentation technology, the deep learning model is mainly relied on for global optimization, that is, the basic model is used to perform preliminary segmentation of the vascular image to obtain the initial vascular segmentation result, and then the vascular segmentation result is input into the overall optimization model for multiple iterative optimizations to gradually improve the initial vascular segmentation result and obtain the final vascular segmentation result. However, this method cannot accurately locate the unsegmented areas in the initial vascular segmentation results, and can only perform overall iterative optimization within the entire image. It is easily affected by the image background and ignores the local features of small blood vessels, which leads to reduced vascular segmentation accuracy. In addition, during the overall iterative optimization process of the entire image, even if some areas have been correctly segmented, the overall optimization model will still repeat the calculation of the area, wasting computing resources.

[0020] In order to improve segmentation accuracy and save computing resources, the embodiment of the present invention can accurately locate the unsegmented area in the blood vessel segmentation image by extracting the endpoints in the blood vessel segmentation image. On this basis, by iteratively segmenting the local image corresponding to each endpoint, each segmented area can be accurately segmented respectively, thereby improving the segmentation accuracy of small blood vessels and improving segmentation accuracy. In addition, there is no need to repeatedly segment the segmented area, thereby saving computing resources.

[0021] See also Figure 1 , which shows a flow chart of the implementation of the blood vessel image segmentation method provided by an embodiment of the present invention, and is described in detail as follows: Step 101, obtaining an original blood vessel image and its corresponding initial blood vessel segmentation image; In the embodiment of the present invention, the original blood vessel image can be obtained by using angiography medical imaging technology. The original blood vessel image is usually a single-channel grayscale image. By inputting the original blood vessel image into an image segmentation model such as U-Net or DeepLabV3+, an initial blood vessel segmentation image can be obtained. For example, see Figure 2 The initial blood vessel segmentation image is usually a binary image, in which the grayscale value of the pixel points in the blood vessel area is usually 1, and the grayscale value of the pixel points in the remaining area is usually 0.

[0022] In order to ensure the accuracy of subsequent image processing, the embodiment of the present invention can pre-process the original blood vessel image and the blood vessel segmentation image respectively.

[0023] For the original vascular image, the original vascular image can be normalized to eliminate the pixel intensity distribution changes caused by different imaging devices, imaging parameters or individual physiological differences. The embodiment of the present invention adopts the Min-Max Normalization method to perform normalization, and its specific calculation formula is as follows:

[0024] in, Represents the normalized pixel value, whose value range is constrained between [0,1]. I Represents the original pixel value, and Represent the minimum pixel value and maximum pixel value in the original vascular image respectively. This normalization process can improve the consistency of the data, enhance the convergence of the model, and avoid the problem of unstable gradient update caused by too large or too small pixel value range.

[0025] In order to further enhance the edge information of blood vessels and improve the contrast between blood vessels and background, the embodiment of the present invention further adopts the contrast-limited adaptive histogram equalization (CLAHE) method on the basis of normalization to optimize the grayscale distribution of the image in a local adaptive manner. Among them, the CLAHE method divides the image into multiple small blocks (usually 8×8 or 16×16 grids), and performs histogram equalization in each small block separately, while applying contrast restrictions to prevent excessive contrast enhancement in local areas and cause noise amplification. Its mathematical expression is as follows:

[0026] in, Indicates that the pixel value is The pixel point, the pixel value after equalization, i Indicates from 0 to The gray level index between Indicates grayscale i The frequency of occurrence in the current small block, Indicates the total number of pixels in the current tile.

[0027] Through CLAHE processing, the separability of vascular structures in the input data is enhanced, which helps to improve the accuracy of subsequent vascular segmentation.

[0028] For the initial vascular segmentation image, since noise or local low contrast may cause discontinuity of vascular structure or missing of small areas, the embodiment of the present invention adopts morphological closing operation to post-process the vascular segmentation image. The closing operation consists of dilation operation and erosion operation, and its mathematical definition is as follows:

[0029] in, A represents a binary image, i.e., a blood vessel segmentation image, B Represents a structural element, represents the expansion operation, Represents an erosion operation.

[0030] This process is able to fill small holes in the blood vessel segments and enhance the connectivity of the vascular structure.

[0031] In addition, in order to reduce the boundary noise in the blood vessel segmentation image and improve the smoothness of the blood vessel contour, the embodiment of the present invention also uses Gaussian filtering to smooth the blood vessel segmentation image. The kernel function of Gaussian filtering is defined as follows:

[0032] in, The coordinates are Gaussian filtering result of the pixel at represents the standard deviation of the Gaussian kernel.

[0033] Gaussian filtering can effectively reduce high-frequency noise by performing weighted averaging of pixel values ​​while maintaining the overall shape of the blood vessel boundary, thereby improving segmentation accuracy.

[0034] Step 102: extract the blood vessel skeleton structure based on the blood vessel segmentation image, and determine the endpoints on the blood vessel skeleton structure. The blood vessel skeleton structure includes multiple blood vessel centerlines, and the endpoints are the endpoints of the blood vessel centerlines.

[0035] The purpose of extracting the vascular skeleton structure is to convert the vascular region in the vascular segmentation image into a thinned structure containing only the vascular centerline, so as to facilitate subsequent vascular morphology analysis, branch tracking and diameter measurement. The embodiment of the present invention uses a morphological thinning algorithm to skeletonize the vascular segmentation image to remove redundant pixels and retain only the vascular centerline structure.

[0036] In general, the process of extracting the vascular skeleton structure is carried out in an iterative manner. Each iteration scans the pixels in the image and deletes the boundary pixels that meet the requirements according to the preset removal conditions until the skeleton is stable.

[0037] Specifically, the main steps of the morphological thinning algorithm include: for each foreground pixel in the blood vessel segmentation image, detect whether there is a background pixel in the neighborhood of the foreground pixel; if there is a background pixel in the neighborhood of the foreground pixel, determine the foreground pixel as a candidate pixel. Then, detect whether the topological structure of the blood vessel segmentation image changes before and after each candidate pixel is deleted; determine the candidate pixels whose topological structure does not change before and after deletion as target candidate pixels, and delete them to obtain a new blood vessel segmentation image; Jump to the step of detecting whether there is a background pixel in the neighborhood of each foreground pixel in the blood vessel segmentation image, until there is no target candidate pixel in the current blood vessel segmentation image, and the blood vessel skeleton structure is obtained.

[0038] Here, the foreground pixel points refer to the pixel points used to characterize the blood vessels, and the background pixel points refer to the remaining pixel points except the foreground pixel points.

[0039] For each foreground pixel, if there is at least one background pixel within the eight-neighborhood range of the foreground pixel, the foreground pixel is marked as a candidate pixel. By performing connectivity analysis on the candidate pixels, it is detected that the deletion of the candidate pixels will not cause the vascular structure to break, that is, the topological structure of the original image remains unchanged after deletion. If the connectivity constraint is met (that is, the topological structure remains unchanged after deletion), the pixel is deleted and the next round of iteration is entered. When there are no more deletable pixels in the image, the iteration is stopped and the final skeletonization result is output.

[0040] Mathematically, the above process of extracting the vascular skeleton structure can be expressed as:

[0041] in, Indicates n +The vascular skeleton structure extracted after 1 iteration. Indicates n The vascular skeleton structure extracted after iterations, Indicates n The set of pixel points that satisfy the connectivity constraints in the vascular skeleton structure after iterations.

[0042] For example, Figure 2 After extracting the vascular skeleton structure from the vascular segmentation image, we can get Figure 3The above method can effectively remove redundant pixels in the vascular area while ensuring that the topological structure of the blood vessel remains unchanged.

[0043] Based on the extraction of the blood vessel skeleton structure, the embodiment of the present invention further performs endpoint detection on the blood vessel skeleton structure.

[0044] according to Figure 3 It can be seen that the vascular skeleton structure contains multiple vascular centerlines. Among them, the pixel points at the endpoints of each vascular centerline are the endpoints in the embodiment of the present invention. By performing endpoint detection, the broken area of ​​the blood vessel can be identified, thereby accurately locating the unsegmented area in the vascular segmentation image. In the embodiment of the present invention, the pixel points that are connected to only one pixel point among all the pixel points on the vascular centerline are determined as endpoints.

[0045] Here, endpoint detection can be performed based on the eight-neighborhood pixel counting method. Specifically, for each pixel point on the vascular skeleton structure, if the number of foreground pixels in the neighborhood of the pixel point is 1, the pixel point is determined to be an endpoint. The foreground pixel point is the pixel point used to characterize the blood vessel.

[0046] That is to say, for each pixel point on each blood vessel skeleton structure, the number of foreground pixels in its eight-neighborhood is calculated. If the number is equal to 1, the pixel point is determined to be an endpoint. The mathematical expression is as follows:

[0047] in, Represents pixel p The number of foreground pixels in the eight-neighborhood of Represents pixel p The first The pixel value of the pixel point, here, for the foreground pixel point, , for background pixels, .

[0048] Step 103, superimposing the original blood vessel image and the blood vessel segmentation image to obtain a dual-channel image; The original vascular image and the vascular segmentation image can be superimposed and fused by channel superposition to obtain a dual-channel image. According to the above, the original vascular image is a single-channel grayscale image, and the vascular segmentation image is a binary image, that is, it also belongs to a single-channel grayscale image.

[0049] The embodiment of the present invention aligns the original blood vessel image and the blood vessel segmentation image at the same spatial scale and splices them in the channel dimension to form a dual-channel image. That is, the original blood vessel image and the blood vessel segmentation image are adjusted to the same size, and the pixels at corresponding positions are channel-superimposed.

[0050] It can be understood that the dual-channel image has the same image size as the original blood vessel image and the blood vessel segmentation image, and each pixel point in the dual-channel image corresponds to the pixel value of two channels.

[0051] During the channel superposition process, the geometric consistency between the original vascular grayscale distribution and the segmentation result is maintained, so as to enhance the attention of the subsequent vascular segmentation model to the image edge information.

[0052] Step 104 , for each endpoint, determine the local segmentation window corresponding to the endpoint, perform iterative blood vessel segmentation on the local image corresponding to the local segmentation window in the dual-channel image, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result.

[0053] The local segmentation window corresponding to the endpoint is actually the local area where the endpoint is located, that is, the fracture area of ​​the current blood vessel segment. The local image corresponding to the local segmentation window in the dual-channel image is the unsegmented local area in the dual-channel image.

[0054] The embodiment of the present invention performs iterative vascular segmentation on the unsegmented local area, and can further improve and expand the vascular segmentation result in the local area on the basis of the initial vascular segmentation image. After the iterative segmentation of the local area is completed, the overall vascular segmentation image is updated according to the iterative vascular segmentation result of the local area.

[0055] To ensure that blood vessels of different scales maintain the same proportional relationship during iterative blood vessel segmentation, the embodiment of the present invention adopts a method based on adaptive window adjustment to dynamically adjust the size of the local segmentation window to adapt to blood vessel structures of different diameters.

[0056] In general, the local segmentation window is centered on the endpoint, and the window size is determined by the vessel diameter in the overall vessel segmentation image. The size of the local segmentation window is adjusted by determining the vessel diameter at the location of the endpoint.

[0057] Specifically, when determining the local segmentation window corresponding to each endpoint, for each endpoint, in the vascular segmentation image, the actual blood vessel diameter at the endpoint position is extracted; then, the preset standard window size and the preset standard blood vessel diameter are respectively obtained, and the proportional relationship between the standard window size and the standard blood vessel diameter is determined; then, based on the proportional relationship and the actual blood vessel diameter, the local window size corresponding to the endpoint is determined; finally, the endpoint is used as the center of the local segmentation window, and the local segmentation window is determined according to the local window size.

[0058] In the blood vessel segmentation image, for each endpoint of the blood vessel skeleton structure, the width of continuous foreground pixels (i.e., the segmented blood vessel area) centered on the endpoint is determined along a section perpendicular to the blood vessel direction to obtain the actual blood vessel diameter at the endpoint.

[0059] According to the actual blood vessel diameter, the local window size of the local segmentation window can be adjusted to ensure that the proportion of the blood vessel in the standard window size is consistent. The specific adjustment formula is as follows:

[0060] in, Indicates the local window size, Indicates the default standard window size. Indicates the preset standard blood vessel diameter. Indicates the actual blood vessel diameter.

[0061] Here, the standard window size and the standard blood vessel diameter can be determined according to actual conditions, and the embodiment of the present invention does not specifically limit this. For example, the standard window can be a square, 128×128 pixels, the standard window size is 128 pixels, and the standard blood vessel diameter can be 10 pixels.

[0062] The adjusted local window size can ensure that blood vessels of different scales maintain the same morphological features in the standard window, thereby improving the adaptability of the subsequent blood vessel segmentation model at different blood vessel scales.

[0063] The embodiment of the present invention determines the local segmentation window based on the size of the local window with the endpoint as the center, and determines the local image corresponding to the local segmentation window in the dual-channel image based on cropping.

[0064] After determining the local image, the local image can be further standardized. Specifically, for the endpoints close to the edge of the image, the image is expanded by mirror filling or zero filling to ensure the integrity of the extraction window. The local image is scaled to a standard window size of 128×128 pixels using bicubic interpolation. The pixel values ​​of the scaled image are normalized so that the pixel values ​​are distributed between [0,1] to improve the stability of blood vessel segmentation.

[0065] The traditional method adopts global optimization, which easily ignores the local features of small blood vessels, resulting in blurred blood vessel edges and missing branches. The embodiment of the present invention adopts the method of adaptively adjusting the local segmentation window to perform window extraction, which can focus on local segmentation and improve the segmentation accuracy of small blood vessels.

[0066] Step 105, traverse all endpoints to obtain the final blood vessel segmentation image.

[0067] After completing the local iterative segmentation of an endpoint, the overall vascular segmentation image is updated according to the iterative vascular segmentation result of the endpoint, and then the dual-channel image is updated. On the basis of the update completion, the iterative vascular segmentation of the next endpoint is performed until all endpoints are traversed to obtain the final vascular segmentation image.

[0068] Compared with the prior art, the embodiment of the present invention can accurately locate the unsegmented area in the vascular segmentation image by extracting the vascular skeleton structure from the vascular segmentation image and then determining the endpoints in the vascular skeleton structure, so as to reduce the background interference in the vascular segmentation image. Among them, the endpoints can be used as the critical points between the segmented area and the unsegmented area in the vascular segmentation image, and are used to locate the unsegmented area in the vascular segmentation image. On this basis, by determining the local image corresponding to the local segmentation window corresponding to each endpoint in the dual-channel image, iterative vascular segmentation is performed, and the unsegmented area corresponding to each endpoint can be locally iteratively segmented respectively, thereby improving the segmentation accuracy of small blood vessels, thereby improving the segmentation accuracy, and saving computing resources. In addition, when the embodiment of the present invention performs local vascular segmentation by superimposing the dual-channel image of the original vascular image and the vascular segmentation image, the segmented area in the vascular segmentation image can guide the current local vascular segmentation work, thereby improving the robustness of vascular segmentation.

[0069] In order to supplement and improve the incompletely segmented vascular regions and improve the integrity and accuracy of the segmentation results, and at the same time, to ensure the connectivity and integrity of the vascular structure, the embodiment of the present invention determines the local segmentation window according to the endpoints, and adopts a gradual expansion strategy based on the local segmentation window to supplement the incompletely segmented vascular regions.

[0070] Here, endpoint detection is used to determine the broken area of ​​the current blood vessel segmentation, local images are extracted with the endpoints as the center, and the blood vessel segmentation model is used to finely segment the local images. Based on the fine segmentation, the blood vessel segmentation results in the local images are updated, and new endpoints are extracted from the blood vessel segmentation results in the new local images, and the iteration continues until the termination condition is met.

[0071] In the embodiment of the present invention, the core idea of ​​iterative expansion is: in each round of iteration, the skeleton structure of the currently segmented blood vessel is used to guide the local segmentation, and the endpoints are continuously updated until the blood vessel structure is complete.

[0072] Optional, see Figure 4 ,The specific execution steps of iterative blood vessel segmentation are as follows: Step 401, initializing the number of iterations.

[0073] During the initialization phase, the number of iterations can be set to zero.

[0074] Step 402: determine the local image corresponding to the local segmentation window in the dual-channel image, and input the local image into the blood vessel segmentation model to obtain a local blood vessel segmentation image.

[0075] The blood vessel segmentation model is used to perform blood vessel segmentation on an input local image and output a blood vessel segmentation result corresponding to the local image, that is, a local blood vessel segmentation image.

[0076] The blood vessel segmentation model here can be any deep learning segmentation model, for example, U-Net, AttentionU-Net, DeepLabV3+, TransUNet, and a segmentation network based on a Transformer structure. The blood vessel segmentation model in the embodiment of the present invention is based on the U-Net structure, and the specific structure is as follows: The input of the vascular segmentation model consists of two channels. The first channel is used to receive the local image corresponding to the original vascular image. The original vascular image can provide the grayscale or color structure information of the blood vessels, so that the model can learn the vascular morphology and boundary features in the original image. The second channel is used to receive the local image corresponding to the vascular segmentation image. The vascular segmentation image contains the information of the segmented vascular area, which can guide the model to focus on the unsegmented area, suppress background noise, and improve the accuracy of segmentation.

[0077] The vascular segmentation model uses a dual-channel input strategy and comprehensively utilizes the original vascular image and the initial vascular segmentation image to enhance the model's ability to recognize small vascular structures and reduce background interference.

[0078] The model structure adopts the Encoder-Decoder framework and combines the Attention Mechanism to enhance the ability to focus on the edges of blood vessels. Its main components include: Encoder: It consists of multiple convolutional layers and pooling layers, extracting multi-scale features layer by layer. Batch normalization is performed after each convolutional layer to stabilize the gradient, and the ReLU (Rectified Linear Unit) activation function is used to enhance the nonlinear expression ability. Appropriate stride and padding strategies are used to ensure that the feature map size is gradually reduced while retaining key vascular feature information.

[0079] Skip Connection: A skip connection is introduced between the encoder and decoder to directly transfer low-level features from the encoding stage to the decoding stage to preserve fine-grained edge information and prevent blurring of blood vessel boundaries. A weighted fusion strategy is used to adaptively fuse features of different scales to optimize segmentation results.

[0080] Decoder: Use deconvolution or bilinear interpolation for upsampling to gradually restore the original resolution of the image. During the decoding process, an attention mechanism (such as SE-Net, CBAM, Transformer-based Attention) is introduced to enhance the ability to focus on microvascular structures and improve segmentation accuracy.

[0081] Output Layer: 1×1 convolution is used as the final output layer to map the high-dimensional features extracted in the previous layer to a single-channel segmentation probability map. The role of this 1×1 convolution is to compress multi-channel features into a single channel through linear transformation while retaining spatial information to ensure that the output pixel-level prediction has accurate segmentation boundaries. The sigmoid activation function is used to normalize the output of the 1×1 convolution and map the pixel values ​​to the [0,1] interval, so that the output result can represent the probability that each pixel belongs to the vascular area. The specific calculation formula is as follows:

[0082] in, Represents a pixel ( x , y ), Represents the original output value of the 1×1 convolution.

[0083] The output layer also uses a fixed threshold T=0.5 to binarize the probability map to generate the final segmentation mask (Binary Segmentation Mask). The specific calculation method is as follows:

[0084] in, Indicates the final segmentation result. When the value is 1, it means that the pixel belongs to the blood vessel area, and when the value is 0, it means that the pixel belongs to the background area.

[0085] It can be understood that the local blood vessel segmentation image output by the model is equivalent to further expanding the blood vessel area on the basis of the local image input by the model to further improve the blood vessel skeleton structure.

[0086] Step 403, judging whether an iteration termination condition is satisfied according to the local blood vessel segmentation image and the number of iterations; On the one hand, an iteration termination condition can be created according to the blood vessel expansion result in the local blood vessel segmentation and highlighting to determine whether to terminate the iteration. On the other hand, an iteration termination condition can be created according to the current number of iterations to determine whether to terminate the iteration.

[0087] Optionally, the iteration termination condition may include: during N consecutive iterations, the newly added blood vessel segmentation area in the local blood vessel segmentation image is less than a preset value; or the current number of iterations reaches a maximum limit. Here, N is an integer greater than or equal to 1.

[0088] Here, the newly added blood vessel segmentation area refers to the area difference between the blood vessel area in the local blood vessel segmentation image output by the model and the blood vessel area in the local image input to the model.

[0089] The preset value, N value and maximum limit here can be determined according to actual conditions. The preset value can be 1% of the blood vessel area in the local blood vessel segmentation image. The value of N can be 3. The maximum limit can be 10.

[0090] That is, in three consecutive iterations, the newly added blood vessel area accounts for less than 1%. At this time, it is determined that the blood vessel segmentation has converged and no further expansion is performed. Alternatively, when the current number of iterations reaches 10, the iteration is terminated to prevent the model from repeatedly calculating in invalid areas and improve calculation efficiency.

[0091] Step 404: If the iteration termination condition is not met, then determine the target endpoint in the local blood vessel segmentation image.

[0092] Here, the target endpoint is the endpoint farthest from the blood vessel skeleton structure in the local image among all the endpoints of the local blood vessel segmentation image.

[0093] Optionally, the specific steps of determining the target endpoint in the local blood vessel segmentation image are as follows: First, the vascular skeleton structure in the local vascular segmentation image is extracted, and each endpoint in the vascular skeleton structure is determined; then, the original vascular skeleton structure in the local image corresponding to the local vascular segmentation image is determined, and among all the endpoints of the vascular skeleton structure in the local vascular segmentation image, the endpoint farthest from the original vascular skeleton structure is determined as the target endpoint.

[0094] For ease of description, the vascular skeleton structure in the local vascular segmentation image output by the vascular segmentation model is defined as the first vascular skeleton structure. The vascular skeleton structure in the local image input to the vascular segmentation model is defined as the second vascular skeleton structure. The target endpoint refers to: among all endpoints of the first vascular skeleton structure, the endpoint farthest from the second vascular skeleton structure.

[0095] For example, see Figure 5 , Figure 5 The left side shows the blood vessel skeleton structure in the local image of the input blood vessel segmentation model, that is, the second blood vessel skeleton structure. According to the second blood vessel skeleton structure, the corresponding second blood vessel segment region can be determined. Figure 5 The right side shows the vascular skeleton structure in the local vascular segmentation image output by the vascular segmentation model, i.e., the first vascular skeleton structure. According to the first vascular skeleton structure, the corresponding first vascular segment region can be determined. Among all the endpoints of the first vascular skeleton structure, the endpoint farthest from the second vascular skeleton structure / second vascular segment region can be determined, i.e., the target endpoint.

[0096] Here, by determining the endpoint with the farthest distance as the target endpoint, it can be ensured that the blood vessel can be gradually expanded in the correct direction.

[0097] Step 405 , updating the local segmentation window according to the target endpoint, updating the blood vessel segmentation image and the dual-channel image based on the local blood vessel segmentation image, and updating the current number of iterations.

[0098] On the one hand, based on the determination of the target endpoint, the local segmentation window is re-determined with the target endpoint as the center; on the other hand, based on the local vascular segmentation image, the overall vascular segmentation image is updated, and the dual-channel image is updated accordingly. In addition, the current number of iterations can also be updated to prepare for the next iterative segmentation.

[0099] When merging the local vascular segmentation image into the overall vascular segmentation image to update the vascular segmentation image, the local vascular segmentation results are aligned according to the spatial coordinates and accumulated into the overall vascular segmentation image. On this basis, the morphological closing operation can also be used to remove small segmentation gaps to maintain the smoothness of the vascular boundary.

[0100] Step 406, jump to the step of determining the local image corresponding to the local segmentation window in the dual-channel image, until the iteration termination condition is met, determine the current local blood vessel segmentation image as the iterative blood vessel segmentation result, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result.

[0101] Based on the updated local segmentation window and dual-channel image, the local image is re-determined, and the subsequent vascular segmentation steps are continued until the iteration termination condition is met. The local vascular segmentation image currently output by the vascular segmentation model is used as the iterative vascular segmentation result, and the overall vascular segmentation image is updated according to the local iterative vascular segmentation result to complete the local iterative segmentation of an endpoint.

[0102] The above local iterative segmentation process ensures the gradual advancement of blood vessel segmentation, while avoiding repeated calculation of invalid areas, thereby improving the efficiency and accuracy of segmentation.

[0103] Due to the overall optimization, traditional methods are difficult to accurately focus on the unsegmented areas of blood vessels, which may lead to problems such as broken blood vessels and missing side branches in the segmentation results. The embodiment of the present invention uses the method of vascular skeleton structure extraction and endpoint iterative tracking segmentation to gradually expand the vascular segmentation area, which can effectively complete the segmentation results of low-contrast and small blood vessels and ensure the integrity and connectivity of the vascular network.

[0104] In view of the problem that traditional methods are prone to discontinuous blood vessels and serious mis-segmentation in low-contrast and noisy medical images (such as CTA, MRA, etc.), the embodiment of the present invention combines endpoint local segmentation, input dual-channel images, and endpoint tracking iterative segmentation to enable the model to more stably identify low-contrast blood vessels and improve the robustness of segmentation.

[0105] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0106] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0107] Figure 6 The structure diagram of the blood vessel image segmentation device provided by the embodiment of the present invention is shown. For the convenience of description, only the part related to the embodiment of the present invention is shown, which is described in detail as follows: like Figure 6 As shown, the blood vessel image segmentation device 6 includes: an acquisition module 61 , an extraction module 62 and a segmentation module 63 .

[0108] An acquisition module 61 is used to acquire an original blood vessel image and its corresponding initial blood vessel segmentation image; An extraction module 62 is used to extract the blood vessel skeleton structure based on the blood vessel segmentation image, and determine the endpoints on the blood vessel skeleton structure; the blood vessel skeleton structure includes multiple blood vessel centerlines, and the endpoints are the endpoints of the blood vessel centerlines; The segmentation module 63 is used for: Overlay the original vascular image and the vascular segmentation image to obtain a dual-channel image; For each endpoint, determine the corresponding local segmentation window, and perform iterative vascular segmentation on the local image corresponding to the local segmentation window in the dual-channel image, and update the vascular segmentation image according to the iterative vascular segmentation result; Traverse all endpoints to obtain the final vascular segmentation image.

[0109] In a possible implementation manner, the segmentation module 63 is specifically configured to: Initialize the number of iterations; Determine the local image corresponding to the local segmentation window in the dual-channel image, and input the local image into the vascular segmentation model to obtain a local vascular segmentation image; Judge whether the iterative termination condition is satisfied according to the local vascular segmentation image and the number of iterations; If the iterative termination condition is not satisfied, determine the target endpoint in the local vascular segmentation image; the target endpoint is the endpoint farthest from the vascular skeleton structure in the local image among all endpoints of the local vascular segmentation image; Update the local segmentation window according to the target endpoint, update the vascular segmentation image and the dual-channel image based on the local vascular segmentation image, and update the current number of iterations; Jump to execute the step of determining the local image corresponding to the local segmentation window in the dual-channel image, and until the iterative termination condition is satisfied, determine the current local vascular segmentation image as the iterative vascular segmentation result, and update the vascular segmentation image according to the iterative vascular segmentation result.

[0110] In a possible implementation manner, the iterative termination condition includes: In consecutive N iterations, the newly added vascular segmentation area in the local vascular segmentation image is less than a preset value; N is an integer greater than or equal to 1; Or The current number of iterations reaches the maximum limit.

[0111] In a possible implementation manner, the segmentation module 63 is specifically configured to: Extract the vascular skeleton structure in the local vascular segmentation image and determine each endpoint in the vascular skeleton structure; Determine the original vascular skeleton structure in the local image corresponding to the local vascular segmentation image, and determine the endpoint farthest from the original vascular skeleton structure among all endpoints of the vascular skeleton structure in the local vascular segmentation image as the target endpoint.

[0112] In a possible implementation manner, the segmentation module 63 is specifically configured to: For each endpoint, extract the actual blood vessel diameter at the location of the endpoint in the blood vessel segmentation image; Respectively obtaining a preset standard window size and a preset standard blood vessel diameter, and determining a proportional relationship between the standard window size and the standard blood vessel diameter; Based on the proportional relationship and the actual blood vessel diameter, a local window size corresponding to the endpoint is determined; The endpoint is used as the center of the local segmentation window, and the local segmentation window is determined according to the local window size.

[0113] In a possible implementation, the extraction module 62 is specifically configured to: For each foreground pixel in the blood vessel segmentation image, detect whether there is a background pixel in the neighborhood of the foreground pixel; the foreground pixel is the pixel used to characterize the blood vessel, and the background pixel is the remaining pixel except the foreground pixel; If there is a background pixel in the neighborhood of the foreground pixel, the foreground pixel is determined as a candidate pixel; Detect whether the topological structure in the blood vessel segmentation image changes before and after each candidate pixel is deleted; Determine the candidate pixel points whose topological structures in the blood vessel segmentation image before and after deletion are unchanged as target candidate pixel points, and delete them to obtain a new blood vessel segmentation image; Jump to the step of detecting whether there is a background pixel in the neighborhood of each foreground pixel in the blood vessel segmentation image, until there is no target candidate pixel in the current blood vessel segmentation image, and the blood vessel skeleton structure is obtained; For each pixel point on the blood vessel skeleton structure, if the number of foreground pixels in the neighborhood of the pixel point is 1, the pixel point is determined to be an endpoint.

[0114] The present device embodiment can be used to execute the above method embodiment. Its technical principle and technical effect are the same as those of the above method embodiment, and will not be described in detail here.

[0115] Figure 7 Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 7 As shown, the electronic device 7 of this embodiment includes: a processor 70 and a memory 71. The memory 71 stores a computer program 72. When the processor 70 executes the computer program 72, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 70 executes the computer program 72, the functions of each module / unit in the above-mentioned device embodiments are implemented.

[0116] Exemplarily, the computer program 72 may be divided into one or more modules / units, which are stored in the memory 71 and executed by the processor 70 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program 72 in the electronic device 7.

[0117] The electronic device 7 may include, but is not limited to, a processor 70 and a memory 71. Those skilled in the art will appreciate that Figure 7 It is only an example of the electronic device 7 and does not constitute a limitation of the electronic device 7. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device 7 may also include input and output devices, network access devices, buses, etc.

[0118] The processor 70 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0119] The memory 71 may be an internal storage unit of the electronic device 7, such as a hard disk or memory of the electronic device 7. The memory 71 may also be an external storage device of the electronic device 7, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 7. Further, the memory 71 may also include both an internal storage unit and an external storage device of the electronic device 7. The memory 71 is used to store the computer program 72 and other programs and data required by the electronic device 7. The memory 71 may also be used to temporarily store data that has been output or is to be output.

[0120] For the convenience and simplicity of description, only the division of the above functional modules / units is used as an example for illustration. In actual applications, the above functions can be assigned to different functional modules / units as needed. The above modules / units can be implemented in the form of hardware, software, or a combination of hardware and software.

[0121] The embodiment of the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the methods in the above method embodiments are implemented.

[0122] The embodiment of the present invention further provides a computer program product, including a computer program. When the computer program is executed by a processor, the methods in the above method embodiments are implemented.

[0123] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. Computer readable media may include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0124] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. If there is no special explanation and logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other. The technical features in different embodiments can be combined to form a new embodiment according to their internal logical relationship.

[0125] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A blood vessel image segmentation method, characterized in that: include: Acquire an original blood vessel image and its corresponding initial blood vessel segmentation image; Based on the blood vessel segmentation image, a blood vessel skeleton structure is extracted, and endpoints on the blood vessel skeleton structure are determined; the blood vessel skeleton structure includes a plurality of blood vessel centerlines, and the endpoints are endpoints of the blood vessel centerlines; superimposing the original blood vessel image and the blood vessel segmentation image to obtain a dual-channel image; For each endpoint, determine a local segmentation window corresponding to the endpoint, perform iterative blood vessel segmentation on a local image corresponding to the local segmentation window in the dual-channel image, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result; Traverse all endpoints to obtain the final blood vessel segmentation image.

2. The blood vessel image segmentation method according to claim 1, characterized in that: The iterative blood vessel segmentation is performed on the local image corresponding to the local segmentation window in the dual-channel image, and the blood vessel segmentation image is updated according to the iterative blood vessel segmentation result, including: Initialize the number of iterations; Determine a local image corresponding to the local segmentation window in the dual-channel image, and input the local image into a blood vessel segmentation model to obtain a local blood vessel segmentation image; determining whether an iteration termination condition is satisfied according to the local blood vessel segmentation image and the number of iterations; If the iteration termination condition is not met, determining a target endpoint in the local blood vessel segmentation image; the target endpoint is an endpoint farthest from the blood vessel skeleton structure in the local image among all endpoints of the local blood vessel segmentation image; According to the target endpoint, updating the local segmentation window, updating the blood vessel segmentation image and the dual-channel image based on the local blood vessel segmentation image, and updating the current number of iterations; Jump to the step of determining the local image corresponding to the local segmentation window in the dual-channel image until the iteration termination condition is met, determine the current local blood vessel segmentation image as the iterative blood vessel segmentation result, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result.

3. The blood vessel image segmentation method according to claim 2, characterized in that: The iteration termination condition includes: During N consecutive iterations, the newly added blood vessel segmentation area in the local blood vessel segmentation image is smaller than a preset value; N is an integer greater than or equal to 1; or The current number of iterations has reached the maximum limit.

4. The blood vessel image segmentation method according to claim 2 or 3, characterized in that: The step of determining a target endpoint in the local blood vessel segmentation image comprises: Extracting a blood vessel skeleton structure in the local blood vessel segmentation image, and determining each endpoint in the blood vessel skeleton structure; An original blood vessel skeleton structure in a local image corresponding to the local blood vessel segmentation image is determined, and an endpoint farthest from the original blood vessel skeleton structure among all endpoints of the blood vessel skeleton structure in the local blood vessel segmentation image is determined as a target endpoint.

5. The blood vessel image segmentation method according to any one of claims 1 to 3, characterized in that: The step of determining, for each endpoint, a local segmentation window corresponding to the endpoint includes: For each endpoint, extracting the actual blood vessel diameter at the location of the endpoint in the blood vessel segmentation image; Respectively obtaining a preset standard window size and a preset standard blood vessel diameter, and determining a proportional relationship between the standard window size and the standard blood vessel diameter; Determining a local window size corresponding to the endpoint based on the proportional relationship and the actual blood vessel diameter; The endpoint is used as the center of the local segmentation window, and the local segmentation window is determined according to the local window size.

6. The blood vessel image segmentation method according to any one of claims 1 to 3, characterized in that: The extracting of the blood vessel skeleton structure based on the blood vessel segmentation image and determining the endpoints on the blood vessel skeleton structure includes: For each foreground pixel in the blood vessel segmentation image, detecting whether there is a background pixel in the neighborhood of the foreground pixel; the foreground pixel is a pixel used to characterize the blood vessel, and the background pixel is the remaining pixel except the foreground pixel; If there is a background pixel in the neighborhood of the foreground pixel, the foreground pixel is determined as a candidate pixel; Detect whether the topological structure in the blood vessel segmentation image changes before and after each candidate pixel is deleted; Determine the candidate pixel points whose topological structures in the blood vessel segmentation image before and after deletion are unchanged as target candidate pixel points, and delete them to obtain a new blood vessel segmentation image; Jump to the step of detecting whether there is a background pixel in the neighborhood of each foreground pixel in the blood vessel segmentation image until the target candidate pixel does not exist in the current blood vessel segmentation image, thereby obtaining the blood vessel skeleton structure; For each pixel point on the blood vessel skeleton structure, if the number of foreground pixel points in the neighborhood of the pixel point is 1, the pixel point is determined to be an endpoint.

7. A blood vessel image segmentation device, characterized in that: include: An acquisition module, used for acquiring an original blood vessel image and its corresponding initial blood vessel segmentation image; An extraction module, used to extract a blood vessel skeleton structure based on the blood vessel segmentation image, and determine endpoints on the blood vessel skeleton structure; the blood vessel skeleton structure includes a plurality of blood vessel centerlines, and the endpoints are endpoints of the blood vessel centerlines; Segmentation module for: superimposing the original blood vessel image and the blood vessel segmentation image to obtain a dual-channel image; For each endpoint, determine a local segmentation window corresponding to the endpoint, perform iterative blood vessel segmentation on a local image corresponding to the local segmentation window in the dual-channel image, and update the blood vessel segmentation image according to the iterative blood vessel segmentation result; Traverse all endpoints to obtain the final blood vessel segmentation image.

8. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The method comprises a computer program, which implements the method according to any one of claims 1 to 6 when being executed by a processor.

Images