Blood vessel image segmentation method and related equipment

Abstract

The invention discloses a blood vessel image segmentation method and related equipment, and the method comprises the steps: carrying out the multi-direction feature extraction processing of obtained to-be-segmented blood vessel image data through the scheme, and obtaining a plurality of different levels of direction perception feature data, and taking the data as multi-scale direction perception feature data; performing feature decoding on the obtained multi-scale direction sensing feature data to obtain a plurality of different levels of spatial features as spatial feature data; performing feature fusion on the obtained multi-scale direction sensing feature data and spatial feature data to obtain different levels of fusion feature data as fusion data; and performing up-sampling reconstruction processing on the obtained fusion data to obtain reconstruction data, and performing convolution activation processing on the obtained reconstruction data to obtain a blood vessel segmentation result. According to the embodiment of the invention, the blood vessel segmentation precision and continuity can be improved. The method can be widely applied to the technical field of image data processing.

Description

Technical Field

[0001] This application relates to the field of image data processing technology, and in particular to a method and related equipment for segmenting blood vessel images. Background Technology

[0002] MRA (Magnetic Resonance Angiography) is a non-invasive vascular imaging technique that provides important support for the diagnosis and treatment of vascular diseases. However, due to the complex three-dimensional tree-like structure of intracranial blood vessels and their significant anisotropy, it is difficult to identify and segment blood vessels from MRA images. At the same time, hemodynamic effects can also easily lead to the loss of image signals, thereby reducing the accuracy and continuity of blood vessel segmentation. Summary of the Invention

[0003] The main objective of this application is to propose a blood vessel image segmentation method and related equipment, which can improve the accuracy and continuity of blood vessel segmentation.

[0004] To achieve the above objectives, one aspect of this application proposes a blood vessel image segmentation method, the method comprising: The process involves acquiring blood vessel image data to be segmented, extracting multi-directional features from the blood vessel image data to determine multi-scale directional sensing feature data, wherein the multi-scale directional sensing feature data includes several different levels of directional sensing feature data. The multi-scale orientation sensing feature data is decoded to determine spatial feature data; wherein, the spatial feature data includes several spatial features at different levels; Feature fusion is performed based on the multi-scale directional sensing feature data and the spatial feature data to determine fused data; wherein, the fused data includes several fused feature data at different levels; The fused data is upsampled and reconstructed to determine the reconstructed data, and then convolutional activation processing is performed on the reconstructed data to determine the blood vessel segmentation result.

[0005] In some embodiments, the step of extracting multi-directional features from the blood vessel image data to be segmented to determine multi-scale directional sensing feature data specifically includes: The three-dimensional orientation features of the blood vessel image data to be segmented are extracted to determine the orientation feature data, and the orientation feature data is pooled to obtain the first feature data. The first feature data is subjected to three-dimensional residual convolution processing to determine semantic feature data, and the semantic feature data is combined with the first feature data to determine the multi-scale orientation perception feature data.

[0006] In some embodiments, the step of extracting three-dimensional orientation features from the blood vessel image data to be segmented and determining the orientation feature data specifically includes: The blood vessel image data to be segmented is used as input data, and multi-branch convolution processing is performed on the input data to determine several branch output data; wherein, the branch output data includes directional feature data in the X direction, directional feature data in the Y direction, directional feature data in the Z direction, and global spatial context information; The branch output data are fused according to the channel dimension to determine the branch fused data, and the branch fused data is subjected to nonlinear activation and channel normalization processing to determine the first output data. The directional feature data is determined by performing a residual connection based on the first output data and the input data.

[0007] In some embodiments, performing three-dimensional residual convolution processing on the first feature data to determine semantic feature data specifically includes: Feature extraction is performed on the first feature data to determine local spatial feature data, and semantic enhancement is performed on the local spatial feature data to determine the main path data; The first feature data is subjected to feature mapping to determine the mapped data, and the mapped data is channel aligned to determine the projection path data; The semantic feature data is determined by performing feature concatenation on the trunk path data and the projected path data.

[0008] In some embodiments, the upsampling and decoding of the multi-scale orientation sensing feature data to determine spatial feature data specifically includes: Using the multi-scale orientation perception feature data as the second input data, three-dimensional orientation feature extraction is performed on the second input data to determine the second feature data; Alternatively, perform three-dimensional residual convolution processing on the second input data to determine the third feature data; The second feature data or the third feature data is upsampled to determine the spatial feature data.

[0009] In some embodiments, the step of fusing features based on the multi-scale orientation sensing feature data and the spatial feature data to determine the fused data specifically includes: Based on the multi-scale directional sensing feature data and the spatial feature data, feature stitching is performed to determine the stitched feature data, and channel weighting processing is performed on the stitched feature data according to the normalized weights to determine the weighted feature map; Perform multi-branch convolution on the weighted feature map to determine several branch convolution data; and concatenate the several branch convolution data to determine the concatenated data; Perform 3D convolution on the spliced ​​data to determine convolution feature data, and perform activation processing on the convolution feature data to determine feature output data; The sub-feature map is determined by multiplying the feature output data and the weighted feature map; the fused data is determined by performing residual connection based on the preset attention weights, the sub-feature map, and the weighted feature map.

[0010] To achieve the above objectives, another aspect of this application proposes a blood vessel image segmentation system, the system comprising: The feature extraction module is used to acquire blood vessel image data to be segmented, perform multi-directional feature extraction on the blood vessel image data to be segmented, and determine multi-scale directional sensing feature data; wherein, the multi-scale directional sensing feature data includes several different levels of directional sensing feature data; The feature decoding module is used to perform feature decoding on the multi-scale orientation sensing feature data to determine spatial feature data; wherein, the spatial feature data includes several spatial features at different levels; The feature fusion module is used to perform feature fusion based on the multi-scale directional sensing feature data and the spatial feature data to determine the fused data; wherein, the fused data includes several fused feature data at different levels; The segmentation output module is used to upsample and reconstruct the fused data, determine the reconstructed data, and perform convolutional activation processing on the reconstructed data to determine the blood vessel segmentation result.

[0011] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0012] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.

[0013] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0014] The embodiments of this application include at least the following beneficial effects: This application provides a method, system, electronic device, storage medium, and program product for segmenting blood vessel images. This method performs multi-directional feature extraction processing on acquired blood vessel image data to be segmented, obtaining multiple levels of directional sensing feature data as multi-scale directional sensing feature data; it then performs feature decoding on the obtained multi-scale directional sensing feature data to obtain multiple levels of spatial features as spatial feature data; it further performs feature fusion on the obtained multi-scale directional sensing feature data and spatial feature data to obtain fused feature data at different levels as fused data; it then performs upsampling reconstruction processing on the obtained fused data to obtain reconstructed data, and finally performs convolution activation processing on the reconstructed data to obtain the blood vessel segmentation result. By extracting multi-scale directional features and spatial features and performing feature fusion for comprehensive analysis, the accuracy and continuity of blood vessel segmentation are improved. Attached Figure Description

[0015] Figure 1 This is a flowchart of a blood vessel image segmentation method provided in an embodiment of this application; Figure 2 yes Figure 1 The flowchart of step S101 in the text; Figure 3 yes Figure 2 The flowchart of step S201 in the text; Figure 4 yes Figure 2 The flowchart of step S202 in the document; Figure 5 yes Figure 1 The flowchart of step S102 in the document; Figure 6 yes Figure 1 The flowchart of step S103 in the process; Figure 7 This application provides a flowchart of the overall network structure in a specific embodiment; Figure 8 This application provides a schematic diagram of the 3D-MDC module structure in the overall network structure of a specific embodiment. Figure 9 This application provides a schematic diagram of the ResConv3D module structure in the overall network structure of a specific embodiment. Figure 10 This is a schematic diagram of the vascular collaborative attention fusion module in the overall network structure of a specific embodiment of this application; Figure 11 This is a schematic diagram of the structure of a blood vessel image segmentation system provided in an embodiment of this application; Figure 12This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0017] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0018] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] This application provides a method for segmenting blood vessel images, relating to the field of information technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing a blood vessel image segmentation method, but is not limited to the above forms.

[0021] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0022] Figure 1 This is an optional flowchart of a blood vessel image segmentation method provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0023] Step S101: Obtain the blood vessel image data to be segmented, perform multi-directional feature extraction on the blood vessel image data to be segmented, and determine multi-scale directional perception feature data; wherein, the multi-scale directional perception feature data includes several different levels of directional perception feature data. Step S102: Decode the multi-scale orientation sensing feature data to determine the spatial feature data; wherein, the spatial feature data includes several spatial features at different levels. Step S103: Perform feature fusion based on multi-scale directional sensing feature data and spatial feature data to determine fused data; wherein, the fused data includes fused feature data at several different levels; Step S104: Upsample and reconstruct the fused data to determine the reconstructed data, and perform convolution activation processing on the reconstructed data to determine the blood vessel segmentation result.

[0024] Steps S101 to S104 of this embodiment involve inputting the acquired magnetic resonance angiography image data into a pre-constructed and trained image processing model. This model uses the mU-Net model as a baseline and, based on an adaptive configuration mechanism, introduces a three-dimensional multi-directional core module, a three-dimensional residual convolution module, and a blood vessel collaborative attention fusion module to construct a convolutional neural network. Sample data is collected to train the constructed convolutional neural network, resulting in the image processing model. The image processing model then segments the input blood vessel imaging image data. It extracts multi-directional features from the input blood vessel image data to obtain directional perception features at different levels, serving as multi-scale directional perception feature data. Simultaneously, the image processing model decodes the obtained multi-scale directional perception feature data and processes it to obtain spatial feature data at different levels. The multi-scale directional perception feature data at different levels and the corresponding spatial feature data are then fused to improve the image processing model's perception and learning of the complex directionality and spatial distribution of blood vessels, thereby improving the accuracy and continuity of blood vessel segmentation. Finally, the image processing model upsamples the fused feature data to gradually restore the image resolution, resulting in the blood vessel segmentation result.

[0025] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S201 to S202: Step S201: Extract three-dimensional orientation features from the blood vessel image data to be segmented, determine the orientation feature data, and perform pooling processing on the orientation feature data to obtain the first feature data; Step S202: Perform three-dimensional residual convolution processing on the first feature data to determine semantic feature data, and perform feature combination between the semantic feature data and the first feature data to determine multi-scale directional perception feature data.

[0026] In step S201 of some embodiments, the image processing model uses a three-dimensional multi-directional core module and a three-dimensional residual convolution module to process the input vascular imaging image data; the three-dimensional multi-directional core module displays and models the vascular orientation in the vascular imaging image data, extracts the directional features of the blood vessels in the orthogonal directions, and performs pooling processing on the extracted directional features to avoid gradient explosion problems in subsequent processing.

[0027] In step S202 of some embodiments, the input data information is processed by a three-dimensional residual convolution module to further extract feature data and enrich local feature details, thereby improving the accuracy of subsequent blood vessel segmentation. At the same time, the semantic information in the input data is preserved through three-dimensional residual convolution processing to improve the continuity and consistency of subsequent blood vessel segmentation. Then, the extracted feature data and the preserved semantic information are combined as the output of the three-dimensional residual convolution module to obtain multi-scale directional perception feature data.

[0028] Please see Figure 3 In some embodiments, step S201 may include, but is not limited to, steps S301 to S303: Step S301: The blood vessel image data to be segmented is used as input data, and multi-branch convolution processing is performed on the input data to determine several branch output data; wherein, the branch output data includes directional feature data in the X direction, directional feature data in the Y direction, directional feature data in the Z direction, and global spatial context information. Step S302: Fuse the output data of several branches according to the channel dimension to determine the branch fused data, and perform nonlinear activation and channel normalization processing on the branch fused data to determine the first output data; Step S303: Perform residual connection based on the first output data and input data to determine the directional feature data.

[0029] In step S301 of some embodiments, the three-dimensional multi-directional core module is provided with two stacked three-dimensional multi-directional core units, which are connected in series through residual connections. After receiving the input vascular imaging image data, the first unit performs multi-branch parallel convolution processing on the vascular imaging image data to simultaneously extract the directional features of blood vessels in different directions in the image. In this embodiment, the input vascular imaging image is subjected to four-branch parallel convolution processing, wherein three parallel branches extract the directional features of blood vessels in the three orthogonal directions of X, Y and Z in the vascular imaging image, respectively, and capture the course pattern of blood vessels in different anatomical planes to improve the accuracy and continuity of subsequent blood vessel segmentation. The other branch uses an isotropic convolutional layer to extract the spatial context information of the vascular imaging image at the global level to improve the accuracy and continuity of subsequent blood vessel segmentation.

[0030] In step S302 of some embodiments, the feature data or context information extracted from the four branches are fused along the channel dimension to obtain branch fused data, which is then output as branch output data to the next layer of the three-dimensional multi-directional core unit for processing. Through feature fusion along the channel dimension, rich spatial details and high-level semantic information are preserved. The fused branch output data is subjected to nonlinear function activation and channel normalization to avoid overfitting problems in subsequent feature extraction and to effectively characterize complex anisotropic blood vessel orientations, thereby improving the accuracy and continuity of segmentation of small and tortuous blood vessels.

[0031] In step S303 of some embodiments, the branch output data after nonlinear function activation and channel normalization is residually connected with the input data of the input three-dimensional multi-directional core unit and used as the input data of the next three-dimensional multi-directional core unit. The previous steps are repeated to obtain directional feature data.

[0032] Please see Figure 4 In some embodiments, step S202 may include, but is not limited to, steps S401 to S403: Step S401: Extract features from the first feature data to determine local spatial feature data, and perform semantic enhancement on the local spatial feature data to determine the backbone path data; Step S402: Perform feature mapping on the first feature data to determine the mapped data, and perform channel alignment on the mapped data to determine the projection path data; Step S403: Perform feature concatenation on the backbone path data and the projected path data to determine the semantic feature data.

[0033] In step S401 of some embodiments, the image processing model inputs the directional feature data output by the three-dimensional multi-directional core module into the three-dimensional residual convolution module for processing, so as to further improve the expressive ability of directional features in blood vessels and enhance the corresponding semantic information, thereby improving the continuity and consistency of subsequent blood vessel segmentation. In this embodiment, the three-dimensional residual convolution module sets up dual paths to process the input feature data; the input feature data is convolved through the main path to extract local spatial feature data and perform semantic information enhancement processing, and the output is the main path data.

[0034] In step S402 of some embodiments, the three-dimensional residual convolution module also performs feature mapping processing on the input feature data through the projection path to enrich the semantic information in the feature data, and performs channel alignment on the feature data after feature mapping processing as the output data of the projection path.

[0035] In step S403 of some embodiments, the three-dimensional residual convolution module performs feature concatenation on the backbone path data and the projection path data to obtain semantic feature data; while preserving the original information, the feature diversity and robustness are enhanced through feature concatenation.

[0036] Please see Figure 5 In some embodiments, step S102 may include, but is not limited to, steps S501 to S503: Step S501: Use the multi-scale orientation perception feature data as the second input data, extract three-dimensional orientation features from the second input data, and determine the second feature data; Step S502, or, perform three-dimensional residual convolution processing on the second input data to determine the third feature data; Step S503: Upsample the second feature data or the third feature data to determine the spatial feature data.

[0037] In step S501 of some embodiments, the image processing model decodes the obtained multi-scale directional sensing feature data, gradually improving the resolution to obtain spatial features. In this embodiment, the features extracted by the three-dimensional multi-directional core module and the three-dimensional residual convolution module are used as encoded features. The encoded features are input into the three-dimensional multi-directional core module for processing to extract the spatial feature data. Then, the spatial feature data is processed by the upsampling module to improve the resolution, so as to improve the accuracy of blood vessel segmentation.

[0038] In step S502 of some embodiments, the image processing model can also input the obtained encoded features into the three-dimensional residual convolution module for processing, extract the spatial feature data therein, and then process it through the upsampling module to improve the resolution, so as to improve the accuracy of blood vessel segmentation.

[0039] In step S503 of some embodiments, the image processing model upsamples the feature data output by the three-dimensional multi-directional core module or the three-dimensional residual convolution module to improve the feature resolution; in this embodiment, transposed convolution is used to process the feature data.

[0040] Please see Figure 6 In some embodiments, step S103 may include, but is not limited to, steps S601 to S604: Step S601: Based on the multi-scale directional sensing feature data and spatial feature data, feature stitching is performed to determine the stitched feature data, and channel weighting processing is performed on the stitched feature data according to the normalized weights to determine the weighted feature map. Step S602: Perform multi-branch convolution on the weighted feature map to determine several branch convolution data; and concatenate the several branch convolution data to determine the concatenated data; Step S603: Perform 3D convolution on the spliced ​​data, determine the convolution feature data, perform activation processing on the convolution feature data, and determine the feature output data; Step S604: Multiply the feature output data and the weighted feature map to determine the sub-feature map; perform residual connection based on the preset attention weights, the sub-feature map and the weighted feature map to determine the fused data.

[0041] In step S601 of some embodiments, the feature fusion module adaptively fuses the obtained encoded features and decoded features at the corresponding levels to improve the segmentation consistency and topological continuity of the model in complex blood vessel bifurcation and weak signal regions. In this embodiment, the feature fusion module performs feature stitching on multi-scale directional sensing feature data and spatial feature data, and then performs channel weighting on the stitched feature data to obtain a weighted feature map. In step S602 of some embodiments, the feature fusion module performs branch convolution processing on the obtained weighted feature map in the spatial dimension to extract contextual relationships, and splices the extracted contextual relationships to obtain spliced ​​data, which is used to highlight the target region and suppress background interference in order to improve the accuracy of blood vessel segmentation.

[0042] In step S603 of some embodiments, the stitched data is subjected to three-dimensional convolution processing to extract spatial feature data, and the extracted spatial feature data is processed by an activation function to obtain feature output data.

[0043] In step S604 of some embodiments, the feature fusion module performs residual fusion processing on the obtained feature output data and the obtained weighted feature map through the adaptive residual fusion module; the context information in the original weighted feature map is added to the feature output data to improve the vascular feature representation in the feature map, thereby improving the accuracy of vascular segmentation; then, the feature data output by residual fusion is subjected to weighted residual fusion to balance the ratio of original features to enhanced features, thereby achieving synergistic optimization of channel attention and spatial attention, thereby improving the accuracy and continuity of vascular segmentation.

[0044] The following is a detailed description and explanation of the solutions in the embodiments of the present invention, using specific application examples: Please see Figure 7 , Figure 7 This document describes the overall network flow of a blood vessel image segmentation method provided in this application, applied in a specific embodiment. The flow includes three stages: feature extraction, feature fusion, and segmentation reconstruction. In this flow, the input MRA data undergoes feature extraction through a hierarchical structure composed of a ResConv3D module and a 3D-MDC module. This hierarchical structure can be configured as follows: Figure 7As shown; the outputs of each ResConv3D module and 3D-MDC module are connected to a max pooling layer, and the output feature data is pooled before being input into the next module; the feature data of different levels directly output by each ResConv3D module or 3D-MDC module are also output to the corresponding level feature fusion module for feature fusion; the structure of the 3D-MDC module is as follows. Figure 8 As shown, the 3D-MDC module contains four parallel convolutional branches. Three of them employ anisotropic branches to extract directional features of blood vessels from the input MRA image data along the three orthogonal directions of X, Y, and Z, respectively, to capture the course patterns in different anatomical planes. The other branch employs an isotropic convolutional layer to simultaneously perform convolution processing on the input MRA image to extract global spatial context information. The outputs of the four branches are fused along the channel dimension and then processed through nonlinear activation and channel normalization to obtain directional features. Simultaneously, during the feature extraction stage, the ResConv3D module is used to process the input MRA data or the feature data output from the 3D-MDC module. The structure of the ResConv3D module is as follows: Figure 9 As shown, the ResConv3D module adopts a dual-path structure design, including a backbone path and a projection path, to maintain the network's lightweight nature while improving feature representation capabilities. The backbone path performs convolution processing on the input MRA image or feature data, extracting local spatial features and enhancing semantic information. The projection path simultaneously processes the input MRA image or feature data, achieving feature mapping and channel alignment. The outputs of the two paths are fused and combined with the input feature data to form the module output, preserving original information while enhancing feature diversity and robustness. The multi-scale orientation perception features extracted by the ResConv3D and 3D-MDC modules are decoded through the ResConv3D and 3D-MDC modules and upsampling processing to obtain spatial feature data. Figure 7 In the illustrated process, the ConvTranspose3d module upsamples the output features of the ResConv3D and 3D-MDC modules to obtain spatial features, which are then input into the feature fusion module or other levels of the ResConv3D or 3D-MDC modules for further processing. The feature fusion module receives multi-scale directional sensing feature data and spatial feature data for feature fusion. In a specific embodiment, a vascular collaborative attention fusion module is used for feature fusion, and the structure of this module is as follows: Figure 10As shown, it includes three functional subunits responsible for channel weighting, spatial attention, and residual fusion, respectively. The channel weighting subunit weights different channels according to the importance of the input feature data, thereby enhancing key semantic information and suppressing redundant features. It concatenates the input feature data to obtain concatenated feature data, then normalizes the set learnable weights, and performs channel weighting on the concatenated feature data based on the normalized weights to obtain a weighted feature map. The spatial attention subunit processes the obtained weighted feature map, extracts the spatial context of features, highlights the target region, suppresses background interference, and improves the recognition of small blood vessels. The localization accuracy is improved by performing dual-path convolution on the weighted feature map and concatenating the path outputs to obtain output feature data. This output feature data is then subjected to 3D convolution, with non-linearity added through an activation function to avoid overfitting, resulting in convolutional output data. A residual fusion subunit performs residual fusion on the convolutional output data to balance the ratio of original features to enhanced features, achieving synergistic optimization of channel attention and spatial attention. This module multiplies the weighted feature map with the convolutional output data to obtain corresponding feature data. Then, based on the set learnable weights, a residual connection is performed between this feature data and the weighted feature map, outputting a fused feature. The final fused feature is then processed by the 3D-MDC module, sequentially through 3D convolution and activation function processing, outputting a vessel segmentation map corresponding to the input MRA image.

[0045] In a specific embodiment, in order to obtain an image processing model and improve its generalization ability, the constructed image processing model is trained by integrating three publicly available datasets: MIDAS, Cerebral Artery Segmentation (CAS2023) Challenge, and Information eXtraction from Images (IXI) dataset. The training and validation sets of the datasets are combined into a unified training pool, while independent test sets of MIDAS and CAS2023 are retained for evaluation. The image sample data in the training pool are all rigidly matched and resampled to a uniform voxel spacing of 0.5×0.5×0.8mm^3. The size is cropped with the geometric center of each image sample data as the origin to obtain image patches of size 192×224×56 voxels. Zero padding is performed on sample data with insufficient image boundaries to complete the data. The following dataset is obtained: MIDAS dataset: 44 manually annotated MRA cases are included for the unified training pool, and 10 cases are retained as independent test sets. CAS2023 dataset: 63 high-quality annotated cases were used for training and validation, and 15 cases were reserved for the test set to ensure the reliability of the evaluation results. IXI dataset: The original IXI dataset did not provide vascular annotations, so expert annotations published by ŽigaBizjak et al. were used. After rigorous quality control, 90 high-quality MRA cases were selected and included in a unified training pool. To avoid evaluation bias introduced by third-party annotations, no independent test set was set for this dataset.

[0046] The model is trained using the constructed training pool, and the Dice loss and cross-entropy loss are calculated using the following formulas: , , in, For the first The true label of an image For the model in the first Predicted values ​​at individual elements This represents the total number of speed-ups in the image. It is a constant; after calculating the Dice loss and cross-entropy loss, the final loss is calculated using the following formula: , in, This is the weighting factor for the loss factor; The trained image processing model is evaluated using metrics to determine its performance, including Dice similarity coefficient, Jaccard index, precision, recall, and skeleton accuracy ratio. The Dice similarity coefficient, used to assess the spatial overlap between the predicted results and the ground truth labels, is calculated using the following formula: , in, To predict the segmentation results, The labels are real labels, and DSC is the Dice similarity coefficient. The Jaccard index is used to assess the similarity between the predicted result and the reference segment, and is calculated using the following formula: , Where Precision is the accuracy rate; Recall rate represents the proportion of all real blood vessel voxels that are correctly identified, and is used to characterize the completeness of the detection. It is calculated using the following formula: , Where Recall is the recall rate; The skeleton accuracy ratio assesses the degree of preservation of vascular topology by comparing the predicted skeleton with a reference skeleton, and is determined by the following formula: , SAR stands for skeleton accuracy ratio.

[0047] The image processing model provided in this application is compared with models such as nnU-Net, Swin UNETR, and ER-Net. Ablation experiments are set up to remove different modules from the image processing model. The MIDAS and CAS2023 datasets are processed, and their respective evaluation metrics are calculated. The following information table is obtained: Table 1

[0048] Table 2

[0049] Tables 1 and 2 show the evaluation metrics of each model on the MIDAS and MCAS2023 datasets, respectively. According to Tables 1 and 2, the model in this application outperforms other models on multiple metrics and has the best segmentation effect. Table 3

[0050] Table 3 (continued)

[0051] Table 4

[0052] Tables 3 and 4 present the performance metrics of the image processing model after removing the backbone network and functional modules through ablation experiments. According to Tables 3 and 4, introducing the 3D-MDC module alone significantly improves the model's segmentation accuracy, effectively extracting multi-directional vascular features and enhancing the continuity of vascular branches. Introducing the V-YAF module alone also improves segmentation performance, demonstrating its adaptive enhancement capability for features in both spatial and channel dimensions. Simultaneous introduction of both the 3D-MDC and V-YAF modules improves overall structural consistency and segmentation stability; the 3D-MDC and V-YAF modules exhibit a synergistic gain effect.

[0053] The embodiments of this application include at least the following beneficial effects: This application provides a method, system, electronic device, storage medium, and program product for segmenting blood vessel images. This method performs multi-directional feature extraction processing on acquired blood vessel image data to be segmented, obtaining multiple levels of directional sensing feature data as multi-scale directional sensing feature data; it then performs feature decoding on the obtained multi-scale directional sensing feature data to obtain multiple levels of spatial features as spatial feature data; it further performs feature fusion on the obtained multi-scale directional sensing feature data and spatial feature data to obtain fused feature data at different levels as fused data; it then performs upsampling reconstruction processing on the obtained fused data to obtain reconstructed data, and finally performs convolution activation processing on the reconstructed data to obtain the blood vessel segmentation result. By extracting multi-scale directional features and spatial features and performing feature fusion for comprehensive analysis, the accuracy and continuity of blood vessel segmentation are improved.

[0054] Please see Figure 11 This application also provides a blood vessel image segmentation system that can implement the above-described method. The system includes: The feature extraction module is used to acquire blood vessel image data to be segmented, perform multi-directional feature extraction on the blood vessel image data to be segmented, and determine multi-scale directional sensing feature data; wherein, the multi-scale directional sensing feature data includes several different levels of directional sensing feature data; The feature decoding module is used to perform feature decoding on the multi-scale orientation sensing feature data to determine spatial feature data; wherein, the spatial feature data includes several spatial features at different levels; The feature fusion module is used to perform feature fusion based on the multi-scale directional sensing feature data and the spatial feature data to determine the fused data; wherein, the fused data includes several fused feature data at different levels; The segmentation output module is used to upsample and reconstruct the fused data, determine the reconstructed data, and perform convolutional activation processing on the reconstructed data to determine the blood vessel segmentation result.

[0055] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0056] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0057] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0058] Please see Figure 12 , Figure 12 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 1201 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, GPU (Graphics Processing Unit), application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 1202 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1202 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1202 and is called and executed by the processor 1201 using the methods described in the embodiments of this application. The input / output interface 1203 is used to implement information input and output; The communication interface 1204 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 1205 transmits information between various components of the device (e.g., processor 1201, memory 1202, input / output interface 1203, and communication interface 1204); The processor 1201, memory 1202, input / output interface 1203 and communication interface 1204 are connected to each other within the device via bus 1205.

[0059] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0060] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0061] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0062] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0063] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0064] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0065] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0066] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0067] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0068] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0069] It should be understood that in this application, "at least one (item)" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0070] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0071] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0072] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for segmenting blood vessel images, characterized in that, The method includes: The process involves acquiring blood vessel image data to be segmented, extracting multi-directional features from the blood vessel image data to determine multi-scale directional sensing feature data, wherein the multi-scale directional sensing feature data includes several different levels of directional sensing feature data. The multi-scale orientation sensing feature data is decoded to determine spatial feature data; wherein, the spatial feature data includes several spatial features at different levels; Feature fusion is performed based on the multi-scale directional sensing feature data and the spatial feature data to determine fused data; wherein, the fused data includes several fused feature data at different levels; The fused data is upsampled and reconstructed to determine the reconstructed data, and then convolutional activation processing is performed on the reconstructed data to determine the blood vessel segmentation result.

2. The method according to claim 1, characterized in that, The step of extracting multi-directional features from the blood vessel image data to be segmented, and determining multi-scale directional sensing feature data, specifically includes: The three-dimensional orientation features of the blood vessel image data to be segmented are extracted to determine the orientation feature data, and the orientation feature data is pooled to obtain the first feature data. The first feature data is subjected to three-dimensional residual convolution processing to determine semantic feature data, and the semantic feature data is combined with the first feature data to determine the multi-scale orientation perception feature data.

3. The method according to claim 2, characterized in that, The step of extracting three-dimensional orientation features from the blood vessel image data to be segmented, and determining the orientation feature data, specifically includes: The blood vessel image data to be segmented is used as input data, and multi-branch convolution processing is performed on the input data to determine several branch output data; wherein, the branch output data includes directional feature data in the X direction, directional feature data in the Y direction, directional feature data in the Z direction, and global spatial context information; The branch output data are fused according to the channel dimension to determine the branch fused data, and the branch fused data is subjected to nonlinear activation and channel normalization processing to determine the first output data. The directional feature data is determined by performing a residual connection based on the first output data and the input data.

4. The method according to claim 2, characterized in that, The step of performing three-dimensional residual convolution processing on the first feature data to determine semantic feature data specifically includes: Feature extraction is performed on the first feature data to determine local spatial feature data, and semantic enhancement is performed on the local spatial feature data to determine the main path data; The first feature data is subjected to feature mapping to determine the mapped data, and the mapped data is channel aligned to determine the projection path data; The semantic feature data is determined by performing feature concatenation on the trunk path data and the projected path data.

5. The method according to claim 1, characterized in that, The step of upsampling and decoding the multi-scale orientation sensing feature data to determine the spatial feature data specifically includes: Using the multi-scale orientation perception feature data as the second input data, three-dimensional orientation feature extraction is performed on the second input data to determine the second feature data; Alternatively, perform three-dimensional residual convolution processing on the second input data to determine the third feature data; The second feature data or the third feature data is upsampled to determine the spatial feature data.

6. The method according to claim 1, characterized in that, The step of fusing features based on the multi-scale directional sensing feature data and the spatial feature data to determine the fused data specifically includes: Based on the multi-scale directional sensing feature data and the spatial feature data, feature stitching is performed to determine the stitched feature data, and channel weighting processing is performed on the stitched feature data according to the normalized weights to determine the weighted feature map; Perform multi-branch convolution on the weighted feature map to determine several branch convolution data; and concatenate the several branch convolution data to determine the concatenated data; Perform 3D convolution on the spliced ​​data to determine convolution feature data, and perform activation processing on the convolution feature data to determine feature output data; The sub-feature map is determined by multiplying the feature output data and the weighted feature map; the fused data is determined by performing residual connection based on the preset attention weights, the sub-feature map, and the weighted feature map.

7. A blood vessel image segmentation system, characterized in that, The system includes: The feature extraction module is used to acquire blood vessel image data to be segmented, perform multi-directional feature extraction on the blood vessel image data to be segmented, and determine multi-scale directional sensing feature data; wherein, the multi-scale directional sensing feature data includes several different levels of directional sensing feature data; The feature decoding module is used to perform feature decoding on the multi-scale orientation sensing feature data to determine spatial feature data; wherein, the spatial feature data includes several spatial features at different levels; The feature fusion module is used to perform feature fusion based on the multi-scale directional sensing feature data and the spatial feature data to determine the fused data; wherein, the fused data includes several fused feature data at different levels; The segmentation output module is used to upsample and reconstruct the fused data, determine the reconstructed data, and perform convolutional activation processing on the reconstructed data to determine the blood vessel segmentation result.

8. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images