A two-stage segmentation method and related device for vascular structure in medical CT images

Abstract

This application discloses a two-stage segmentation method for vascular structures in medical CT images and related devices. The method includes: acquiring a medical CT image and its coarse segmentation image; performing vascular morphology interpolation processing on the coarse segmentation image and performing HU value similarity determination to generate first annotation data and HU value similarity mapping data; performing vascular connectivity determination on the HU value similarity mapping data, extracting conditional labels, and generating second annotation data; and using a boundary erosion-based vascular extraction method to extract the target area in the repaired and compensated data, ultimately obtaining the target segmentation result. By combining HU value determination, vascular connectivity analysis, and boundary erosion technology, the present invention improves the accuracy and connectivity of vascular segmentation and is suitable for the fine extraction of complex vascular structures.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a two-stage segmentation method for blood vessel structures in medical CT images and a related device. Background Art

[0002] As a core component of the human circulatory system, human blood vessels are responsible for transporting blood, oxygen, nutrients, and metabolic waste, crucial for maintaining human life. According to WHO statistics, cardiovascular diseases such as arteriosclerosis, hypertension, and venous thrombosis cause approximately 18 million deaths each year. Therefore, early and immediate detection and intervention of vascular diseases are crucial for maintaining human health. Digital twin reconstruction of the main abdominal blood vessels is a key technology for the detection and treatment of vascular diseases, providing important material support for preoperative planning for patients with vascular diseases.

[0003] Currently, CTA (Computed Tomography Angiography) is an important non-invasive medical imaging technology for observing the major abdominal blood vessels. It uses CT scans combined with contrast enhancement to clearly visualize the vascular structure of the major abdominal vessels and is used to diagnose vascular diseases. Post-processing techniques then generate a three-dimensional model of the major abdominal vessels to assist physicians in analysis. While the commonly used MIP (Maximum Intensity Projection) technique can generate a two-dimensional projection image by taking the maximum CT value of each pixel along the line of sight (usually for contrast-enhanced vessels), this method loses depth information and requires multi-angle rotation to observe the vascular structure. Furthermore, VR (Volume Rendering) technology assigns varying degrees of transparency / color to all voxel data, generating three-dimensional images. Manually adjusting the transparency threshold allows physicians to more fully visualize the major abdominal vascular structures and their distribution within the body. However, this manual adjustment inevitably leaves some organ parenchyma (particularly bone and cardiac regions) in the image, as the threshold is close to that of the major abdominal vessels. This also creates obstacles for the subsequent reconstruction and structural alignment of the entire human vascular system.

[0004] With the continuous development and iteration of artificial intelligence technology, its application in various fields has achieved significant breakthroughs. In recent years, deep learning-based medical image processing technology has been a research hotspot in the field of intelligent healthcare. Image segmentation, a key component of medical image processing, has been automated and intelligently implemented, providing effective technical support for the automated reconstruction of digital twins of major abdominal blood vessels. Numerous studies have shown that, with sufficient training data, deep learning-based medical image segmentation methods (such as UNet, 3D U-Net, and nnUNet) can achieve comparable results in labeling target regions in medical images as manual annotation. However, this requires the construction of a sufficiently rich and high-quality manually annotated dataset to support the training of high-precision segmentation models. However, the complex internal structure of the human body and the complex tree-like structure of major abdominal blood vessels, characterized by a single trunk and multiple branches, complicate the creation of vascular annotation datasets and the high-precision segmentation of deep learning models. This, in turn, leads to a scarcity of high-quality manually annotated vascular segmentation datasets, resulting in problems such as the frequent disconnection of secondary vascular branches from the main trunk and incomplete labeling of thin branches in model segmentation results.

[0005] To address the scarcity of 3D vascular segmentation datasets, some researchers have used transfer learning methods to train models. This involves training the model on a large, publicly available dataset of vascular annotations from various parts of the human body, enabling it to acquire knowledge of slender tubular structures. Dynamic fine-tuning strategies are then used to dynamically freeze or fine-tune the filters in the model. Other researchers have used teacher-assisted learning models, which convert noise in low-quality vascular segmentation data into useful guidance for vascular segmentation through pixel-by-pixel soft correction of the teacher model, thereby improving the segmentation accuracy of the student model in high-quality, low-sample vascular segmentation datasets.

[0006] When it comes to current vascular segmentation tasks, the mainstream approach is to guide the network model architecture or training loss function in a specific sense, so that it can better focus on and capture the multi-branched, tree-like tubular structure unique to blood vessels during training. Among them, snake convolution is a method architecture specifically designed to learn this type of structure. The proposed dynamic snake convolution can adaptively capture slender and tortuous local structures. At the same time, a continuity-constrained loss function is also proposed to better constrain the topological continuity of segmentation during training. In order to better model the connectivity constraints for vascular segmentation, some researchers have used graph neural networks to represent the connectivity prior of blood vessels and embedded this network into a graph attention network. This network architecture has also been applied as a plug-in to the training process of 3D U-Net, thereby improving the 3D U-Net's attention to vascular structures.

[0007] In response to the challenges encountered in the task of blood vessel segmentation, the above methods have completed improvements in the model architecture or training methods based on their own thinking, including the migration of prior features of tubular structures and the prior embedding of aspects such as connectivity prior constraints. First of all, how to utilize the multi-branch-single-trunk structural characteristics of the main abdominal blood vessels to design and build an efficient network model to achieve high-precision segmentation of blood vessels is the first problem to be solved by the present invention; secondly, in response to the scarcity of high-quality annotated data of multiple samples, how to utilize the characteristics of CTA images to design a novel segmentation result repair and optimization method, and further obtain higher-quality segmentation effects on the basis of achieving high-precision segmentation of small-sample data sets is another outstanding contribution of the present invention. Therefore, there is an urgent need for a new method that can achieve high-precision blood vessel segmentation and repair optimization in a small-sample scenario based on the structural characteristics of the main abdominal blood vessels, so as to overcome the above-mentioned deficiencies in the existing technology. Summary of the Invention

[0008] The present invention relates to a two-stage segmentation method for vascular structures in medical CT images and related devices, which include but are not limited to segmentation equipment, electronic devices, computer-readable storage media and computer program products.

[0009] A first aspect provides a two-stage segmentation method for vascular structures in medical CT images, the method comprising:

[0010] Acquire a medical CT image of a target object; automatically segment the medical CT image of the target object based on a dual-path topology structure decoding and segmentation method and extract the maximum connected domain to obtain a medical CT coarse segmentation target object image;

[0011] performing vascular morphology interpolation processing on the medical CT coarsely segmented target object image to generate first annotation data; performing mask extraction on the medical CT image based on position parameters of the first annotation data to generate region-limited data; and generating HU value similarity mapping data based on HU value distribution of the region-limited data;

[0012] Converting the HU value similarity mapping data into binary data; performing vascular network unit labeling on the binary data to generate connected domain label data; performing mask extraction on the connected domain label data based on the position parameters of the first labeled data, and performing statistical label value information to obtain a conditional label; extracting all voxels whose label values ​​are equal to the conditional label from the connected domain label data to generate second labeled data; performing mask extraction on the HU value similarity mapping data based on the position parameters of the second labeled data to obtain repair compensation data;

[0013] The first annotated data is subjected to vascular morphology interpolation processing to obtain vascular expansion data; the vascular expansion data is subjected to a differential operation with the first annotated data to obtain boundary ablation data; the repair compensation data is subjected to erosion processing based on the boundary ablation data to obtain corrected area data; and the corrected area data is subjected to mask extraction based on position parameters of the second annotated data to obtain a target segmentation result.

[0014] In combination with any embodiment of the present application, the dual-path topology decoding and segmentation method includes:

[0015] A dual-pathway decoder structure is constructed. The first path is used to recover the category information of the vascular region, and the second path is used to maintain the skeleton connectivity of the vascular centerline. Feature information is interacted and fused between the first and second paths to complete the structural information of the connection between secondary vessels and main vessels.

[0016] In combination with any embodiment of the present application, the vascular morphology interpolation processing includes:

[0017] A target interval is divided into a plurality of subspaces, a third-order polynomial is constructed on each of the subspaces, and coefficients of the third-order polynomial are determined based on a least squares method or an interpolation condition, so that function values, first-order derivatives, and second-order derivatives of the third-order polynomial are continuous between the subspaces.

[0018] In combination with any embodiment of the present application, the generating of HU value similarity mapping data based on the HU value distribution of the region-limited data includes:

[0019] Counting the HU value distribution of the regionally restricted data, and calculating the key quantile x of the HU value;

[0020] Set x-50 as the lower threshold value Xdown, and x+50 as the upper threshold value Xup;

[0021] Voxels with HU values ​​lower than Xdown are mapped to 0, voxels with HU values ​​higher than Xup are mapped to 255, and voxels between Xdown and Xup are linearly mapped to the interval (0, 255) to obtain HU value similarity mapping data.

[0022] In combination with any embodiment of the present application, the vascular network unit marker includes:

[0023] Dividing voxels in the three-dimensional grid of the target area according to adjacency relationships, wherein the adjacency relationships include face adjacency, edge adjacency, and vertex adjacency;

[0024] The face-adjacent relationship refers to the 6-neighborhood relationship between voxels formed by face connections, the edge-adjacent relationship refers to the 12 directions of face-adjacent relationships further including edge connections, and the vertex-adjacent relationship refers to the 8 directions of vertex connections further including edge connections.

[0025] Connectivity division is performed on the voxels based on the adjacency relationship, and voxels belonging to the same connected domain are classified into one category and assigned the same label value to distinguish multiple different connected domains.

[0026] In a second aspect, a blood vessel structure segmentation device is provided, the blood vessel structure segmentation device comprising:

[0027] An acquisition unit is configured to acquire a medical CT image of a target object; automatically segment the medical CT image of the target object based on a dual-path topology structure decoding and segmentation method and extract the maximum connected domain to obtain a medical CT coarse segmentation target object image;

[0028] A determination unit is configured to perform vascular morphology interpolation processing on the medical CT coarsely segmented target object image to generate first annotation data; perform mask extraction on the medical CT image based on the position parameters of the first annotation data to generate region-limited data; generate HU value similarity mapping data based on the HU value distribution of the region-limited data; and convert the HU value similarity mapping data into binary data; perform vascular network unit labeling on the binary data to generate connected domain label data; perform mask extraction on the connected domain label data based on the position parameters of the first annotation data, and perform statistical label value information to obtain a conditional label; extract all voxels whose label values ​​are equal to the conditional label from the connected domain label data to generate second annotation data; perform mask extraction on the HU value similarity mapping data based on the position parameters of the second annotation data to obtain repair compensation data;

[0029] The extraction unit is configured to perform vascular morphology interpolation processing on the first annotated data to obtain vascular expansion data; perform a differential operation on the vascular expansion data and the first annotated data to obtain boundary ablation data; perform an erosion processing on the repair compensation data based on the boundary ablation data to obtain corrected area data; and perform mask extraction on the corrected area data based on position parameters of the second annotated data to obtain a target segmentation result.

[0030] In a third aspect, an electronic device is provided, comprising: a processor, a communication module, a sensor, a user interface, and a storage unit, wherein the storage unit is configured to store computer program code, wherein the program code comprises computer instructions. When the processor executes these instructions, the electronic device performs the method described in the second aspect and any embodiment thereof.

[0031] In a fourth aspect, another electronic device is provided, comprising: a processor, a wireless communication module, a touch screen, a speaker, and a storage unit, wherein the storage unit is configured to store computer program code, wherein the program code comprises computer instructions. When the processor executes these instructions, the electronic device performs the method described in the second aspect and any embodiment thereof.

[0032] In a fifth aspect, a computer-readable storage medium is provided, wherein a computer program is stored, wherein the program includes program instructions. When these instructions are executed by a processor, the processor will perform the method described in the second aspect and any embodiment thereof.

[0033] In a sixth aspect, a computer program product is provided, wherein the computer program product comprises a computer program or instructions. When the computer program or instructions are run on a computer, the computer will execute the method described in the second aspect and any embodiment thereof.

[0034] It should be understood that the above general description and the following detailed description are only used as examples and explanations and do not limit the present application in any way.

[0035] Compared with the prior art, the present invention first obtains a medical CT image of the target object and a medical CT coarse segmentation image. The coarse segmentation image is obtained by segmenting and extracting the maximum connected domain using a dual-path topology decoding method. The medical CT image is then subjected to HU value similarity constraint processing: the coarse segmentation image is interpolated to generate first annotation data, which is then subjected to mask extraction to obtain region-limited data. HU value similarity mapping data is generated based on its HU value distribution. The mapping data is then subjected to vascular connectivity constraint processing: it is binarized, connected domains are labeled, labels associated with the first annotated region are extracted to generate second annotation data, and mask extraction is performed again to obtain repaired and compensated data. Finally, a boundary erosion-based extraction method is used: boundary ablation data is obtained through vascular interpolation and differential operations, and the repaired and compensated data is eroded to obtain the corrected region, and a final, more complete segmentation result is further extracted. By introducing the dual constraints of HU value distribution and vascular connectivity, combined with a boundary erosion extraction strategy, the present invention achieves a more accurate and coherent automatic segmentation effect in the target region, significantly outperforming the segmentation results of traditional threshold-based or morphological methods, and is suitable for vascular structure extraction tasks in complex organs. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background technology, the drawings required for use in the embodiments of the present application or the background technology will be described below.

[0037] The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and, together with the specification, are used to illustrate the technical solutions of the present application.

[0038] Figure 1 A schematic flow chart of a two-stage segmentation method for vascular structures in medical CT images provided in an embodiment of the present application;

[0039] Figure 2 A schematic diagram of a network structure of a dual-path topology decoding and segmentation method provided in an embodiment of the present application;

[0040] Figure 3 A schematic diagram of a dual-path topology decoder provided in an embodiment of the present application;

[0041] Figure 4 A logic block diagram of a two-stage method of coarse segmentation followed by repair and optimization provided in an embodiment of the present application;

[0042] Figure 5 A schematic diagram of the structure of a blood vessel structure segmentation device provided in an embodiment of the present application;

[0043] Figure 6 A schematic diagram of the hardware architecture of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to allow professionals in this technical field to more fully understand the technical solution of the present application, the technical solution of the present application will be explained in detail and clearly with the help of the accompanying drawings. It should be noted that the described embodiments are only some examples of the present application and do not represent all. Based on these embodiments, those skilled in the art can directly deduce all other possible implementation plans without engaging in creative thinking, and these are also included in the scope of protection of the present application.

[0045] In the specification, claims, and related drawings of this application, the terms "first," "second," and the like are used solely to distinguish between different elements and do not imply any particular order. Furthermore, the use of "including," "having," and their variations denotes non-exclusive inclusion. This means that if a process, method, system, product, or apparatus includes a series of steps or components, the process, method, system, product, or apparatus is not limited to the enumerated steps or components and may include other steps or components not listed, or other steps or units inherent to the process, method, system, product, or apparatus.

[0046] The “embodiment” mentioned in this document refers to any instance in which a particular feature, structure or characteristic is combined, and these instances may belong to at least one embodiment of the present application. The “embodiment” mentioned in this document does not necessarily refer to the same specific case, nor does it mean that they are independent or exclusive alternatives. It should be understood by those skilled in the art that the embodiments described herein can be used in conjunction with other embodiments. It should be understood that in this application, “at least one” includes one or more instances, “a plurality” means two or more instances, and “at least two” means two or more instances.

[0047] It should be understood that the method embodiment of the present application can also be implemented by a processor executing computer program code. The embodiment of the present application is described below in conjunction with the drawings in the embodiment of the present application.

[0048] See also Figure 1 , Figure 1 A schematic flow chart of a two-stage segmentation method for vascular structures in medical CT images and related devices provided in an embodiment of the present application.

[0049] Obtain a medical CT image of the target object, use a dual-path topology structure decoding and segmentation method to automatically perform coarse segmentation on the CT image, extract the maximum connected domain, and obtain a medical CT coarse segmentation image.

[0050] In the embodiment of the present application, the medical CT image is an image obtained by performing a CT scan on the target object. The medical CT image includes the target object, wherein the image area of ​​the target object includes a vascular structure segmentation target object image.

[0051] In an embodiment of the present application, a medical CT coarse segmentation target object image is obtained by automatically segmenting the target object's medical CT image and extracting the largest connected domain using a dual-path topology decoding and segmentation method, validated by the DICOM-RT standard. The dual-path topology decoding and segmentation method includes: first constructing a dual-path decoder structure, with the first path used to restore the classification information of the vascular region and the second path used to maintain the skeleton connectivity of the vascular centerline; and then interacting and fusing feature information between the first and second paths to complete the structural information of the connection between the secondary and main vessels.

[0052] For details, please refer to Figure 2 , Figure 2 A schematic diagram of a network structure of a dual-path topology decoding and segmentation method provided in an embodiment of the present application; wherein the structural encoder first uses a convolution with a convolution kernel size of 7×7 ( Figure 2 The Large Kernel in the 3D image data is mapped to the encoder layer; then the four-layer 3DUX-Net module layer ( Figure 2The 3DUX-Net Block in the encoder completes the downsampling operation on the input, that is, extracts the characteristic information of the target blood vessel, and the output results of each layer of the encoder are processed by the residual unit ( Figure 2 ResidualUnit in) and is copied and propagated ( Figure 2 The decoder then receives the output from each layer of the encoder and performs upsampling operations ( Figure 2 The decoder is designed as a dual-pathway structure, in which the other path is the vascular topology analysis path, which is an upsampling path built by stacking deconvolution layers. It receives the encoder features transmitted from each layer of the encoder and merges them with the output of each layer of its own ( Figure 2 Concentrate in ); This path is lost during training by the centerline distance loss ( Figure 2 The other is the vascular category feature parsing pathway, which is controlled by the cross-fusion attention ( Figure 2 CFA Block in) and self-attention ( Figure 2 The SA Block in the channel is stacked into a unit block, which outputs the features of the previous layer of the channel ( Figure 2 in , i=0,1,2,3) and another pathway feature ( Figure 2 in , i=0, 1, 2, 3) are cross-fused attention modulation, and the useful feature information learned by both is used to restore the feature information of the target blood vessel and construct the final output ( Figure 2 Output in ).

[0053] For details, please refer to Figure 3 , Figure 3 A schematic diagram of a dual-path topology decoder provided in an embodiment of the present application; the figure describes in more detail the cross-fusion attention structure of the encoder output layer ( Figure 3 The CFA Block in the two pathways completes the calculation of its own attention map in turn and assigns attention weights to each other. Finally, the two cross-fusion features are added and fused as the final output. The vascular topology analysis pathway is trained by the centerline distance loss function ( Figure 3 The centerline distance loss in the function is used to complete the regulation. The real label used in the calculation process of this function represents the minimum distance between the target blood vessel area and the blood vessel centerline. The conversion formula is:

[0054] ,

[0055] in It represents the vascular area. represents the centerline of the blood vessel. The function maps the vessel area to a distribution map of the minimum distance between it and the center line.

[0056] The calculation formula of the centerline distance loss function is:

[0057] ,

[0058] in is the number of predicted categories, is the predicted output of the model, is the true label of the blood vessel region.

[0059] Figure 3 The vascular category feature parsing pathway in is regulated during the training process using the Dice cross entropy loss function (DiceCELoss), and its output is also used as the final prediction output of the network model and .

[0060] 102. Perform vascular morphology interpolation on the coarsely segmented image to generate first annotated data; perform mask extraction based on the first annotated data to generate region-limited data; and generate HU value similarity mapping data based on regional HU value distribution.

[0061] For details, please refer to Figure 4 , Figure 4 This is a logic diagram of a two-stage method for coarse segmentation followed by restoration optimization provided in an embodiment of the present application. First, the medical CT coarse segmentation target image is interpolated to generate first annotated data. A mask region is extracted based on its spatial position parameters to obtain limited data containing only the candidate vascular region. By statistically analyzing the voxel distribution of HU values ​​within the region, the key quantile x is calculated and a dynamic threshold range (X) is set. down =x-50,X up =x+50), set the HU value lower than X down The voxels above X are set to zero. up The voxels are saturated to 255, and the intermediate values ​​are linearly mapped to the interval (0, 255), and the HU value similarity map data is finally output. This step preliminarily screens out areas with vascular tissue density characteristics based on the physiological characteristics of the HU value.

[0062] In the embodiments of this application, vascular morphology interpolation processing involves first dividing the target interval into several subspaces, constructing a third-order polynomial function in each subspace, and then solving the polynomial coefficients using the least squares method or interpolation conditions to minimize local fitting errors while ensuring the continuity of the function value, first-order derivative, and second-order derivative at the junction of each subspace. This achieves a smooth transition and expansion of the overall vascular morphology curve, effectively improving the continuity and realism of the model. This method is suitable for processing vascular image data with irregular edges or missing areas, with good geometric fidelity and smoothness, ultimately achieving data expansion and smoothing of the target area.

[0063] In one possible implementation, a density interpolation method based on bilateral filtering can be used to achieve data expansion and smoothing in the target area.

[0064] In another possible implementation method, 3D B-spline interpolation combined with control point optimization can be used to achieve data expansion and smoothing in the target area.

[0065] In another possible implementation, by constructing a distance field and using a level-set evolution algorithm to control the contour expansion and contraction process, the target area data can be expanded and smoothed.

[0066] 103. Binarize the HU value similarity mapping data and mark the connected domains of the vascular network; filter the connected domains based on the first labeled data to generate second labeled data; further mask extract the HU value mapping data to obtain repair compensation data.

[0067] In the embodiment of the present application, vascular connectivity determination constraints are performed on the HU value similarity mapping data to further eliminate discontinuous, noisy or artifact areas, thereby improving the coherence and anatomical rationality of the segmented areas.

[0068] First, the HU similarity map data is binarized to obtain preliminary binary data of the vascular region. This binarization process determines whether each voxel belongs to a suspected vascular region by setting a threshold, thereby forming a binary image with a contrast between foreground (vascular) and background (non-vascular).

[0069] Next, a 3D connected domain analysis is performed based on the binary data. Using a connected component labeling algorithm, each vascular structure in the binary volume data is identified and encoded, generating connected domain label data. Each connected region is assigned a unique label value to indicate its spatial structural independence.

[0070] Next, a spatial mask is constructed based on the position parameters of the first annotation data. The connected domain label data is masked and extracted, retaining only the label data within the region defined by the first annotation data. The label values ​​of all voxels in this region are further counted, and these label values ​​are defined as "conditional labels" for subsequent connectivity screening.

[0071] Then, within the global connected domain label data, all voxels with the same conditional label value are extracted. These voxels belong to vascular regions connected to the initial labeled region, generating the second labeled data. This process effectively eliminates pseudo-vascular structures that are not directly connected to the target region in space.

[0072] Finally, according to the position parameters of the second annotated data, the original HU value similarity mapping data is masked again, retaining only the spatial area corresponding to the second annotated data, and finally obtaining the repair compensation data with coherent structure and fewer artifacts.

[0073] In an embodiment of the present application, the vascular network connected domain labeling is to divide the voxels in the three-dimensional grid of the target area into multiple independent connected domains according to their face adjacency, edge adjacency and vertex adjacency adjacency relationships, and mark each independent connected domain with a different label value for subsequent vascular network structure analysis or connectivity enhancement processing; the face adjacency is a 6-neighborhood; the edge adjacency adds 12 directions of shared edges on the basis of the 6-neighborhood; the vertex adjacency further includes 8 angular directions of shared vertices on the basis of edge adjacency.

[0074] 104. Use the boundary difference method to generate boundary ablation data, perform corrosion processing on the repair compensation data, and obtain the corrected area data; finally, extract the target area based on the second annotation data position to obtain a more complete target segmentation result.

[0075] In the embodiment of the present application, in order to further improve the integrity and accuracy of vascular structure segmentation, after obtaining the repair compensation data, a vascular extraction method based on boundary erosion is used to perform refined extraction of the target area.

[0076] First, vascular morphology interpolation processing is performed on the first annotated data to fill in missing structures in certain directions or slices. Specifically, a third-order polynomial interpolation function is constructed and locally fitted to the target region, maintaining the continuity of the function value and its first- and second-order derivatives, thereby achieving smooth expansion of the vascular boundaries. The image data obtained through this processing is called vascular expansion data.

[0077] Next, the vessel expansion data obtained above is subtracted from the original first-labeled data to extract the edge portion introduced during the expansion process, which is called the boundary ablation data. This difference region reflects the newly generated area at the vessel boundary and represents the potential range of vessel expansion.

[0078] Next, a morphological erosion process is performed on the inpainted and compensated data, using the boundary ablation data as a reference template for the erosion operation. This process removes potential artifacts and non-target areas, further highlighting the true vascular structure associated with the edges of the first annotated data. The resulting image data is the corrected region data.

[0079] Finally, based on the position parameters provided by the second annotation data, a mask extraction operation is performed on the corrected area data, retaining only the valid voxels within the area defined by the second annotation data, ultimately obtaining a more complete and better connected target segmentation result.

[0080] Those skilled in the art will understand that in the above-mentioned method of the specific implementation method, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0081] The above describes in detail the method of the embodiment of the present application, and the following provides an apparatus of the embodiment of the present application.

[0082] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of a blood vessel structure segmentation device provided in an embodiment of the present application. The blood vessel structure segmentation device 1 includes: an acquisition unit 11, a judgment unit 12, and an extraction unit 13. Specifically:

[0083] An acquisition unit 11 is configured to acquire a medical CT image of a target object; automatically segment the medical CT image of the target object based on a dual-path topology structure decoding and segmentation method and extract the maximum connected domain to obtain a medical CT coarse segmentation target object image;

[0084] Determination unit 12: configured to perform vascular morphology interpolation processing on the medical CT coarsely segmented target object image to generate first annotation data; perform mask extraction on the medical CT image based on the position parameters of the first annotation data to generate region-limited data; generate HU value similarity mapping data based on the HU value distribution of the region-limited data; and convert the HU value similarity mapping data into binary data; perform vascular network unit labeling on the binary data to generate connected domain label data; perform mask extraction on the connected domain label data based on the position parameters of the first annotation data, and calculate label value information to obtain conditional labels; extract all voxels whose label values ​​are equal to the conditional labels from the connected domain label data to generate second annotation data; perform mask extraction on the HU value similarity mapping data based on the position parameters of the second annotation data to obtain repair compensation data;

[0085] Extraction unit 13 is configured to perform vascular morphology interpolation processing on the first annotated data to obtain vascular expansion data; perform a differential operation on the vascular expansion data and the first annotated data to obtain boundary ablation data; perform erosion processing on the repair compensation data based on the boundary ablation data to obtain corrected region data; and perform mask extraction on the corrected region data based on position parameters of the second annotated data to obtain a target segmentation result.

[0086] In some embodiments, the functions or modules included in the device provided in the embodiments of the present application can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0087] Figure 6 A schematic diagram of the hardware architecture of an electronic device provided in an embodiment of the present application is shown. The electronic device 2 is mainly composed of a processor 21 and a memory 22. In addition, the device may also include an input device 23 and an output device 24. The processor 21, the memory 22, the input device 23 and the output device 24 are interconnected through connecting components. These connecting components can be various interfaces, data lines or communication buses, etc., and the embodiments of the present application do not make specific provisions for this. It should be clear that in multiple embodiments of the present application, the so-called connection refers to the mutual connection achieved by a specific method, which can be a direct connection or an indirect connection through other devices, for example, through various interfaces, data lines, communication buses, etc.

[0088] Processor 21 may be one or more graphics processing units (GPUs). If processor 21 is a GPU, the GPU may be single-core or multi-core. Optionally, processor 21 may comprise a processor group consisting of multiple GPUs, interconnected via one or more buses. Furthermore, the processor may be other types of processors, which are not specifically limited in this embodiment of the present application.

[0089] Memory 22 is designed to store computer program instructions and various program codes required to execute the present invention. Optionally, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or compact disc read-only memory (CD-ROM), which are used to store relevant instructions and data.

[0090] The input device 23 is used to input data and / or signals, and the output device 24 is used to output data and / or signals. The input device 23 and the output device 24 can be independent devices or an integrated device.

[0091] It should be appreciated that in the embodiment of the present application, the memory 22 can store not only relevant instructions but also relevant data. The embodiment of the present application does not specify the specific data content stored in the memory.

[0092] You should understand that Figure 6 Only a simplified design of an electronic device is shown. In actual use, the electronic device may also include other necessary components, such as different numbers of input / output devices, processors, memories, etc. All electronic devices that can implement the embodiments of this application are within the scope of protection of this application.

[0093] Those skilled in the art will recognize that, according to the components and algorithm steps of each example described in the embodiments disclosed herein, these functions can be implemented by electronic hardware or by combining computer software and electronic hardware. Whether these functions are performed by hardware or software will be determined based on the specific application requirements and design limitations of the technical solution. Technicians can adopt different implementation methods according to the requirements of each specific application, but such implementation methods should not be considered to exceed the scope of protection of this application.

[0094] Professionals should understand that, for the sake of ease of description and simplification, the specific operating procedures of the above-mentioned systems, devices, and components can refer to the corresponding steps in the previous method embodiments and will not be repeated here. At the same time, professionals should also understand that each embodiment in this application has its own focus. For the sake of ease of description and simplification, the same or similar content may not be repeated in different embodiments. Therefore, if a part is not mentioned or not explained in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0095] In the several embodiments provided in this application, it should be recognized that the disclosed systems, devices and methods can also be implemented in other ways. For example, the device embodiments described are only exemplary, wherein the division of the units is only a division of logical functions, and there may be different division methods in actual implementation. For example, multiple units or components may be merged or integrated into another system, or certain features may be omitted, or certain steps may not be performed. In addition, the connections between each other shown or discussed, whether direct or indirect, whether coupling or communication connection, may be implemented in electrical, mechanical or other forms through interfaces, devices or units.

[0096] Units described as independent components may or may not actually be physically separate; parts presented as units may or may not be physical entities, i.e., they may be centralized in one location or distributed across multiple network nodes. Depending on actual needs, some or all of these units may be selected to achieve the objectives of this embodiment.

[0097] Furthermore, in the various embodiments of the present application, the various functional units may be integrated into a single processing unit, physically exist independently, or two or more units may be combined into a single unit. In the aforementioned embodiments, the relevant functions may be implemented in whole or in part through software, hardware, firmware, or any combination thereof. If software implementation is chosen, it may be implemented in whole or in part in the form of a computer program product. This computer program product comprises one or more computer instructions. When these instructions are loaded and executed on a computer, they will generate, in whole or in part, the processes or functions described in the embodiments of this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. These computer instructions may be stored in a computer-readable storage medium or transmitted via such a medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, DSL) or wireless (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium may be any computer-accessible, usable medium, or a data storage facility such as a server or data center that integrates one or more usable media. These available media may include magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), semiconductor media (e.g., SSDs), etc. Those skilled in the art will appreciate that all or part of the process steps for implementing the above-described method embodiments can be accomplished through hardware associated with computer program instructions. These programs can be stored on computer-readable storage media. When executed, these programs will contain the processes for each of the above-described method embodiments. These storage media include, but are not limited to, various media capable of storing program code, such as read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A two-stage segmentation method for vascular structures in medical CT images, characterized in that: The following steps are involved: Acquire a medical CT image of a target object; automatically segment the medical CT image of the target object based on a dual-path topology structure decoding and segmentation method and extract the maximum connected domain to obtain a medical CT coarse segmentation target object image; performing blood vessel morphology interpolation processing on the medical CT coarsely segmented target object image to generate first annotated data; performing mask extraction on the medical CT image based on the position parameters of the first annotation data to generate region-limited data; generating HU value similarity mapping data based on the HU value distribution of the region-limited data; Converting the HU value similarity mapping data into binary data; labeling the binary data as vascular network units to generate connected domain label data; performing mask extraction on the connected domain label data based on the position parameters of the first labeled data, and performing statistical label value information to obtain a conditional label; Extracting all voxels whose label values ​​are equal to the conditional label from the connected domain label data to generate second labeled data; performing mask extraction on the HU value similarity mapping data based on the position parameter of the second annotation data to obtain repair compensation data; performing vascular morphology interpolation processing on the first labeled data to obtain vascular expansion data; performing a differential operation on the vascular expansion data and the first labeled data to obtain boundary ablation data; and performing an erosion processing on the repair compensation data based on the boundary ablation data to obtain corrected region data; Mask extraction is performed on the corrected area data based on the position parameters of the second labeled data to obtain a target segmentation result.

2. The method according to claim 1, characterized in that The dual-path topology decoding and segmentation method comprises: A dual-pathway decoder structure is constructed. The first path is used to recover the category information of the vascular region, and the second path is used to maintain the skeleton connectivity of the vascular centerline. Feature information is interacted and fused between the first and second paths to complete the structural information of the connection between secondary vessels and main vessels.

3. The method according to claim 1, characterized in that The blood vessel morphology interpolation process includes: The target interval is divided into several subspaces, a third-order polynomial is constructed on each of the subspaces, and the coefficients of the third-order polynomial are determined based on the least squares method or the interpolation condition, so that the function value, first-order derivative, and second-order derivative of the third-order polynomial are continuous between the subspaces.

4. The method according to claim 1, wherein The generating HU value similarity mapping data based on the HU value distribution of the area-limited data includes: Counting the HU value distribution of the regionally restricted data, and calculating the key quantile x of the HU value; Set x-50 as the lower threshold X down , set x+50 as the upper threshold X up ; Set the HU value below X down The voxel map is 0, and the voxels above X up The voxel map is 255, between X down With X up The voxels between are linearly mapped to the interval (0, 255) to obtain the HU value similarity mapping data.

5. The method according to claim 1, wherein The vascular network unit marker includes: Dividing voxels in the three-dimensional grid of the target area according to adjacency relationships, wherein the adjacency relationships include face adjacency, edge adjacency, and vertex adjacency; The face-adjacent relationship refers to the 6-neighborhood relationship between voxels formed by face connections, the edge-adjacent relationship refers to the 12 directions of face-adjacent relationships further including edge connections, and the vertex-adjacent relationship refers to the 8 directions of vertex connections further including edge connections. Connectivity division is performed on the voxels based on the adjacency relationship, and voxels belonging to the same connected domain are classified into one category and assigned the same label value to distinguish multiple different connected domains.

6. A two-stage segmentation device for vascular structures in medical CT images, characterized in that: The device comprises: An acquisition unit is configured to acquire a medical CT image of a target object; automatically segment the medical CT image of the target object based on a dual-path topology structure decoding and segmentation method and extract the maximum connected domain to obtain a medical CT coarse segmentation target object image; A determination unit is configured to perform vascular morphology interpolation processing on the medical CT coarsely segmented target object image to generate first annotation data; perform mask extraction on the medical CT image based on the position parameters of the first annotation data to generate region-limited data; generate HU value similarity mapping data based on the HU value distribution of the region-limited data; and convert the HU value similarity mapping data into binary data; perform vascular network unit labeling on the binary data to generate connected domain label data; perform mask extraction on the connected domain label data based on the position parameters of the first annotation data, and perform statistical label value information to obtain a conditional label; extract all voxels whose label values ​​are equal to the conditional label from the connected domain label data to generate second annotation data; perform mask extraction on the HU value similarity mapping data based on the position parameters of the second annotation data to obtain repair compensation data; The extraction unit is configured to perform vascular morphology interpolation processing on the first annotated data to obtain vascular expansion data; perform a differential operation on the vascular expansion data and the first annotated data to obtain boundary ablation data; perform an erosion processing on the repair compensation data based on the boundary ablation data to obtain corrected area data; and perform mask extraction on the corrected area data based on position parameters of the second annotated data to obtain a target segmentation result.

7. An electronic device, characterized in that: include: A processor and a memory, the memory is used to store computer program code, the computer program code includes computer instructions, and when the processor executes the computer instructions, the electronic device executes the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor, the processor is caused to execute the method according to any one of claims 1 to 5.

9. A computer program product, characterized in that The computer program product includes a computer program or instructions; when the computer program or instructions are run on a computer, the computer is caused to execute the method according to any one of claims 1 to 5.

Images