Vessel segmentation method and system based on slice sequence modeling and structure perception memory

Abstract

The application discloses a blood vessel segmentation method and system based on slice sequence modeling and structure perception memory, and relates to the technical field of medical image processing. The method comprises the following steps: converting three-dimensional medical image data into a two-dimensional slice image sequence for layer-by-layer processing; extracting current slice query features to generate a blood vessel segmentation mask, determining a blood vessel region according to the mask and adaptively expanding the blood vessel region, and extracting key-value features to write into a memory unit; calculating attention weights by using subsequent slice query features and historical key features, weighting and aggregating historical value features to obtain context features, and fusing to generate a blood vessel segmentation mask of a current slice image; processing all slice images in sequence and integrating to obtain a three-dimensional segmentation result. The application improves the continuity and accuracy of blood vessel segmentation while maintaining the calculation efficiency of a two-dimensional method, reduces blood vessel rupture and missed detection, and can still realize high-precision segmentation under resource-limited conditions, and can be widely applied to blood vessel analysis and auxiliary diagnosis of medical images such as CTA and MRA.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, specifically relating to a method and system for blood vessel segmentation based on slice sequence modeling and structure-aware memory. Background Technology

[0002] Computed tomography angiography (CTA) and magnetic resonance angiography (MRA) are widely used medical imaging technologies in clinical practice, playing a crucial role in the diagnosis, treatment planning, and preoperative assessment of vascular diseases. Accurate segmentation of vascular structures from these three-dimensional medical images is fundamental for quantitative vascular analysis, lesion detection, and three-dimensional reconstruction, directly impacting the accuracy of clinical diagnosis and treatment.

[0003] For blood vessel segmentation, existing technologies mainly fall into three categories. The first category is segmentation methods based on two-dimensional images. This method cuts the three-dimensional volume data into multiple two-dimensional slices along a certain axis, uses a two-dimensional convolutional neural network (CNN) to independently extract features and segment each slice, and finally stacks the two-dimensional results into a three-dimensional structure. This type of method is computationally efficient and easy to implement, but its fundamental drawback lies in the lack of modeling for the structural continuity between slices. Since blood vessels have a slender, tortuous tubular topology in space, processing each slice independently can easily lead to breaks or missed detections in small branches, weak connections, or complex morphological regions, thus disrupting the global topological consistency of blood vessels.

[0004] The second category is segmentation methods based on 3D volumetric data. These methods directly take 3D image blocks or complete images as input and utilize 3D convolutional neural networks (3D CNNs) to simultaneously model the contextual information in three spatial dimensions. While theoretically they can better capture the three-dimensional extension relationships of blood vessels, their computational complexity and memory consumption are extremely high. In high-resolution medical imaging scenarios, limited by hardware resources, image cropping or block-based strategies are often necessary, which introduces new problems: the discontinuity of information between different local blocks leads to artifacts or inconsistencies in the overall segmentation results, thus weakening the ability to model long-range, continuous vascular structures.

[0005] The third category is the sequence modeling method based on cross-slice information propagation, which has emerged in recent years. To strike a balance between 2D computational efficiency and 3D spatial modeling capabilities, this type of method treats 3D data as a sequence of slices arranged axially. It constructs memory units or context modules to pass historical slice information during layer-by-layer processing. However, most existing sequence modeling methods are designed for general target segmentation tasks and lack specific optimization for vascular structural characteristics. They suffer from two core drawbacks: First, during memory storage, all pixel features of the entire slice are typically densely written, while vascular regions occupy only a small number of voxels in the image. This introduces a large amount of background noise and redundant information, reducing the density of effective information. Second, during historical information retrieval and utilization, there is a lack of a dynamic filtering mechanism based on structural relevance. All stored historical information, regardless of its relevance to the current slice, is used equally or inefficiently. This not only increases the computational burden but may also cause crucial structural information to be submerged in redundant noise. Summary of the Invention

[0006] The purpose of this invention is to overcome the above-mentioned problems and provide a blood vessel segmentation method and system based on slice sequence modeling and structure-aware memory, so as to solve the technical problems of insufficient cross-slice structure modeling capability, high computational complexity of three-dimensional methods, redundant memory storage, and low efficiency of historical information utilization in the prior art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a blood vessel segmentation method based on slice sequence modeling and structure-aware memory, comprising the following steps: The three-dimensional medical image data is converted into a two-dimensional slice image sequence along a preset direction, and the current slice image is selected in sequence. The query features of the current slice image are extracted, and a blood vessel segmentation mask of the current slice image is generated based on the query features. The blood vessel region is determined based on the blood vessel segmentation mask of the current slice image. The blood vessel region is adaptively spatially expanded to obtain an expanded region. Key features and value features are extracted from the current slice image features within the expanded region, and the key features and value features are written into the memory unit as structure-aware memory information. The next slice image to be processed is selected from the slice image sequence, its query features are extracted, and the similarity between the query features and the historical key features stored in the memory unit is calculated to obtain the attention weight. The historical value features stored in the memory unit are weighted and aggregated using the attention weight to obtain the context features. The contextual features are fused with the query features of the next slice image to be processed, and a blood vessel segmentation mask for the slice image is generated by the decoding module. All slice images are processed sequentially, and the blood vessel segmentation masks of each slice image are integrated to obtain the three-dimensional blood vessel segmentation result.

[0008] A further improvement of the present invention is that the adaptive spatial expansion of the vascular region to obtain the expanded region specifically includes: A pixel-level radius range map is generated from the query features of the current slice image using function mapping; Based on the extended radius in the radius range map, a distance mapping corresponding to the local offset is defined, and a soft expansion function is constructed; Using the soft dilation function and the blood vessel segmentation mask of the current slice image, an extended mask is obtained through morphological dilation, and the area covered by the extended mask is the extended region.

[0009] A further improvement of this invention is that the soft expansion function is expressed as:

[0010] in, Indicates spatial location, Indicates local offset. Indicates the location Offset at location The coefficient of thermal expansion, This represents the Sigmoid function. Indicates the adjustment parameter. Indicates position The radius of expansion at that location, Indicates offset Corresponding distance mapping; The extended mask is calculated using the following formula:

[0011] in, Indicates position The extended mask value at that location, This indicates the location of the blood vessel segmentation mask in the current slice image. The value at that location, Indicates all local offsets Perform a series of multiplications.

[0012] A further improvement of this invention is that the attention weight is calculated as follows:

[0013] in, Indicates attention weights, This represents the historical key features stored in the memory unit. This indicates the query features of the next slice to be processed. Represents the square of the Euclidean distance. This represents the normalized exponential function.

[0014] A further improvement of this invention is that, before using the attention weights to weight and aggregate historical value features to obtain contextual features, a dynamic memory filtering step is included: Based on the attention weight, the historical key features and historical value features stored in the memory unit are filtered, and only features with attention weights higher than a preset threshold or ranked higher are retained to participate in subsequent weighted aggregation calculations.

[0015] A further improvement of this invention is that the method for extracting query features, key features, and value features is as follows: For each slice image, a depth feature map is first extracted through an encoder network, and then mapped to query features, key features and value features respectively through independent linear projection layers.

[0016] A further improvement of this invention is that the process of generating the blood vessel segmentation mask for the slice image through the decoding module specifically involves: The decoding module adopts a convolutional neural network structure, which upsamples the fused features step by step, outputs a blood vessel probability map with the same size as the input slice image, and then obtains the final blood vessel segmentation mask through threshold binarization.

[0017] Secondly, the present invention provides a blood vessel segmentation system based on slice sequence modeling and structure-aware memory, comprising the following modules: The serialization module is used to convert three-dimensional medical image data into a two-dimensional slice image sequence along a preset direction, select the current slice image in sequence, extract the query features of the current slice image, and generate a blood vessel segmentation mask of the current slice image based on the query features. The memory storage module is used to determine the blood vessel region based on the blood vessel segmentation mask of the current slice image, and to adaptively expand the blood vessel region to obtain an expanded region; extract key features and value features from the current slice image features within the expanded region, and write the key features and value features as structure-aware memory information into the memory unit; The context retrieval module is used to select the next slice image to be processed from the slice image sequence, extract its query features, and calculate the similarity between the query features and the historical key features stored in the memory unit to obtain the attention weight; the attention weight is used to perform weighted aggregation of the historical value features stored in the memory unit to obtain the context features; The segmentation and fusion module is used to fuse the context features with the query features of the next slice image to be processed, and generate a blood vessel segmentation mask for the slice image through the decoding module. The result integration module is used to process all slice images in sequence and integrate the blood vessel segmentation masks of each slice image to obtain the three-dimensional blood vessel segmentation result.

[0018] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described blood vessel segmentation method based on slice sequence modeling and structure-aware memory.

[0019] Fourthly, the present invention provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described blood vessel segmentation method based on slice sequence modeling and structure-aware memory.

[0020] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a blood vessel segmentation method based on slice sequence modeling and structure-aware memory. By converting three-dimensional medical image data into a two-dimensional slice image sequence and processing it sequentially, this method maintains the low computational resource consumption and high processing efficiency of two-dimensional methods. It effectively establishes structural continuity between slices through a cross-slice information propagation mechanism, thus overcoming the problems of small blood vessel breaks and missed detections caused by the lack of spatial correlation in traditional two-dimensional segmentation methods. This significantly improves the completeness and topological consistency of blood vessel segmentation results. Furthermore, this method accurately locates blood vessel regions based on the blood vessel segmentation mask of the current slice image and performs adaptive spatial expansion. Only key key features and value features are extracted from the expanded region and written into the memory unit as structure-aware memory information. This avoids the large amount of background redundancy caused by dense storage of the entire slice image, increases the density of effective information in the memory unit, and makes the structural information transmitted across slices more accurate and stable. Further, for subsequent slice images, this method calculates attention weights by comparing query features with historical key features in the memory unit, and then weights and aggregates historical value features to obtain contextual features. This achieves dynamic retrieval and selective utilization of historical information based on structural relevance, effectively suppressing interference from irrelevant information and reducing computational redundancy in long sequence processing. Finally, the vascular segmentation mask is generated by fusing contextual features with the query features of the current slice image, and the segmentation masks of all slice images are integrated in sequence to obtain a complete three-dimensional vascular structure. This allows the segmentation performance to be better than some mainstream three-dimensional methods under the condition of relying only on two-dimensional computing resources, and has good engineering application prospects. It can be widely used in vascular analysis and auxiliary diagnosis scenarios of medical images such as CTA and MRA. Attached Figure Description

[0021] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Furthermore, the shapes and proportions of the components in the drawings are merely illustrative to aid in understanding the invention and do not specifically limit the shapes and proportions of the components of the invention.

[0022] Figure 1 The flowchart shows a blood vessel segmentation method based on slice sequence modeling and structure-aware memory. Figure 2 A data flow diagram for a blood vessel segmentation method based on slice sequence modeling and structure-aware memory; Figure 3 A flowchart of a blood vessel segmentation system based on slice sequence modeling and structure-aware memory; Figure 4 This is a schematic diagram of an electronic device. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0025] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0026] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0028] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0029] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1 like Figure 1 and Figure 2 As shown, this invention provides a blood vessel segmentation method based on slice sequence modeling and structure-aware memory, comprising the following steps: S1, convert the three-dimensional medical image data into a two-dimensional slice image sequence along a preset direction, the preset direction including at least one of axial, coronal and sagittal directions, and select the current slice image in sequence order, extract the query features of the current slice image, and generate a blood vessel segmentation mask of the current slice image based on the query features. Specifically, the input three-dimensional medical image data is first processed into slice sequences to obtain an ordered slice sequence. The data is then input into the model layer by layer in sequence for processing. For any slice image... The corresponding query features are obtained through feature extraction and projection mapping. Blood vessel segmentation mask based on the current slice image. Determine the vascular region It retains only the features located within the region as candidate memory information, thereby achieving sparse feature selection based on the target structure.

[0030] S2, determine the blood vessel region based on the blood vessel segmentation mask of the current slice image, perform adaptive spatial expansion on the blood vessel region to obtain the expanded region, extract key features and value features from the current slice image features in the expanded region, and write the key features and value features as structure-aware memory information into the memory unit, while avoiding writing non-blood vessel background region features of the current slice image into the memory unit. To further supplement the vascular neighborhood context information, an adaptive spatial expansion mechanism is introduced. For the current slice image features... Pixel-level radius range map is generated through function mapping: ,in, Indicates the radius of expansion at different spatial locations.

[0031] Based on this radius, a local offset is defined. Corresponding distance mapping And construct the soft expansion function: ,in, This represents the Sigmoid function. This indicates the adjustment parameter.

[0032] Furthermore, the expanded mask is obtained in the following manner: .

[0033] According to the extended area Extract the corresponding key-value pair features from the original features: It is then written into the memory unit to achieve structure-aware sparse memory storage.

[0034] S3, Select the next slice image to be processed from the slice image sequence, extract its query features, and calculate the similarity between the query features and the historical key features stored in the memory unit to obtain the attention weight. Use the attention weight to perform weighted aggregation on the historical value features stored in the memory unit to obtain the context features. For the slice image to be processed First, extract its query features. and sparse key-value pairs with historical slice images in memory cells. Interact with the system. Calculate attention weights based on the similarity between query features and historical key features: ,in, Represents the square of the Euclidean distance. This represents the normalized exponential function. Historical features are weighted and aggregated using this weight to obtain contextual features: .

[0035] Before using the attention weights to perform weighted aggregation of historical value features to obtain context features, a dynamic memory filtering step is included: based on the magnitude of the attention weights, the historical key features and historical value features stored in the memory unit are filtered, retaining only features with attention weights higher than a preset threshold or those ranked highly to participate in subsequent weighted aggregation calculations. As the number of slice images increases, historical information in the memory unit gradually accumulates. Considering that different historical slice images have different importance to the current prediction, this step reduces redundant calculations and improves information utilization efficiency through dynamic filtering.

[0036] S4, the context features are fused with the query features of the next slice image to be processed, and a blood vessel segmentation mask for the slice image is generated by the decoding module; context features Features of the current slice image The fusion process is performed, and a blood vessel segmentation mask is generated through the decoding module. The decoding module uses a convolutional neural network structure to upsample the fused features step by step, outputting a blood vessel probability map with the same size as the input slice image, and then performing threshold binarization to obtain the final blood vessel segmentation mask.

[0037] S5 processes all slice images sequentially and integrates the blood vessel segmentation masks of each slice image to obtain the three-dimensional blood vessel segmentation result.

[0038] Through the progressive processing described above, the vascular segmentation masks of the sequentially generated slice images are stacked axially to reconstruct a complete three-dimensional vascular segmentation result. This invention achieves an effective balance between computational efficiency and structural representation capability by organically combining key technologies such as slice sequence modeling, structure-aware sparse memory, adaptive spatial expansion, and dynamic information retrieval.

[0039] This invention transforms the 3D blood vessel segmentation task into a progressively layered sequence modeling problem by constructing an information propagation mechanism based on slice sequences. This achieves continuous modeling of cross-slice structural information while maintaining computational efficiency, overcoming the limitations of traditional 2D methods lacking spatial correlation and 3D methods having high computational complexity. Addressing the elongated, sparse, and topologically continuous characteristics of blood vessel structures in medical images, this invention proposes a structure-aware sparse memory storage mechanism. By utilizing a blood vessel segmentation mask to locate the target region and combining it with an adaptive spatial expansion strategy, only blood vessel-related regions and their neighborhood features are filtered and stored. This significantly reduces redundant background information while increasing the density of effective structural information in the memory unit, thereby enhancing the stability and robustness of cross-slice information transmission.

[0040] Furthermore, this invention constructs a dynamic memory retrieval mechanism based on feature correlation. During cross-slice information interaction, historical information is selectively utilized based on the matching relationship between the query features of the current slice image and historical features. The feature representations most relevant to the current vascular structure are retained first, thereby effectively suppressing redundant information interference and reducing computational complexity, and achieving efficient information propagation under long sequence conditions.

[0041] Furthermore, this invention introduces a spatial adaptive modeling strategy into the overall framework. By generating differentiated extended ranges for different spatial locations, the model can adaptively adjust the context-sensing region according to the local morphological features of the vascular structure, thereby improving the modeling ability for small vascular branches and complex topological structures.

[0042] In summary, this invention achieves an effective balance between computational efficiency and structural representation capability by organically combining key technologies such as slice sequence modeling, structure-aware sparse memory, adaptive spatial expansion, and dynamic information retrieval. It also possesses good generalization ability and practical application value.

[0043] Example 2 like Figure 3 As shown, the present invention also provides a blood vessel segmentation system based on slice sequence modeling and structure-aware memory, comprising the following modules: The serialization module is used to convert three-dimensional medical image data into a two-dimensional slice image sequence along a preset direction, select the current slice image in sequence, extract the query features of the current slice image, and generate a blood vessel segmentation mask of the current slice image based on the query features. The memory storage module is used to determine the blood vessel region based on the blood vessel segmentation mask of the current slice image, and to adaptively expand the blood vessel region to obtain an expanded region; extract key features and value features from the current slice image features within the expanded region, and write the key features and value features as structure-aware memory information into the memory unit; The context retrieval module is used to select the next slice image to be processed from the slice image sequence, extract its query features, and calculate the similarity between the query features and the historical key features stored in the memory unit to obtain the attention weight; the attention weight is used to perform weighted aggregation of the historical value features stored in the memory unit to obtain the context features; The segmentation and fusion module is used to fuse the context features with the query features of the next slice image to be processed, and generate a blood vessel segmentation mask for the slice image through the decoding module. The result integration module is used to process all slice images in sequence and integrate the blood vessel segmentation masks of each slice image to obtain the three-dimensional blood vessel segmentation result.

[0044] Example 3 Please see Figure 4 As shown, the present invention also provides an electronic device 100 for a blood vessel segmentation method based on slice sequence modeling and structure-aware memory; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0045] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the blood vessel segmentation method based on slice sequence modeling and structure-aware memory described in Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0046] The at least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor. The processor 102 is the control center of the electronic device 100, connecting various parts of the electronic device 100 via various interfaces and lines.

[0047] The memory 101 in the electronic device 100 stores multiple instructions to implement a blood vessel segmentation method based on slice sequence modeling and structure-aware memory, and the processor 102 can execute the multiple instructions to achieve the following: The three-dimensional medical image data is converted into a two-dimensional slice image sequence along a preset direction, and the current slice image is selected in sequence. The query features of the current slice image are extracted, and a blood vessel segmentation mask of the current slice image is generated based on the query features. The blood vessel region is determined based on the blood vessel segmentation mask of the current slice image. The blood vessel region is adaptively spatially expanded to obtain an expanded region. Key features and value features are extracted from the current slice image features within the expanded region, and the key features and value features are written into the memory unit as structure-aware memory information. The next slice image to be processed is selected from the slice image sequence, its query features are extracted, and the similarity between the query features and the historical key features stored in the memory unit is calculated to obtain the attention weight. The historical value features stored in the memory unit are weighted and aggregated using the attention weight to obtain the context features. The contextual features are fused with the query features of the next slice image to be processed, and a blood vessel segmentation mask for the slice image is generated by the decoding module. All slice images are processed sequentially, and the blood vessel segmentation masks of each slice image are integrated to obtain the three-dimensional blood vessel segmentation result.

[0048] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0049] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0052] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0054] Many embodiments and applications beyond the examples provided will be apparent to those skilled in the art upon reading the foregoing description. Therefore, the scope of this teaching should not be determined by reference to the foregoing description, but rather by reference to the foregoing claims and the full scope of their equivalents. For purposes of completeness, all articles and references, including patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the foregoing claims is not intended as a waiver of that subject matter, nor should it be construed as an indication that the applicant has not considered that subject matter as part of the disclosed inventive subject matter.

[0055] The above content provides a further detailed description of the present invention. It should not be construed that the specific embodiments of the present invention are limited to this. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention as defined by the submitted claims.

Claims

1. A blood vessel segmentation method based on slice sequence modeling and structure-aware memory, characterized in that, Includes the following steps: The three-dimensional medical image data is converted into a two-dimensional slice image sequence along a preset direction, and the current slice image is selected in sequence. The query features of the current slice image are extracted, and a blood vessel segmentation mask of the current slice image is generated based on the query features. The blood vessel region is determined based on the blood vessel segmentation mask of the current slice image. The blood vessel region is adaptively spatially expanded to obtain an expanded region. Key features and value features are extracted from the current slice image features within the expanded region, and the key features and value features are written into the memory unit as structure-aware memory information. The next slice image to be processed is selected from the slice image sequence, its query features are extracted, and the similarity between the query features and the historical key features stored in the memory unit is calculated to obtain the attention weight. The historical value features stored in the memory unit are weighted and aggregated using the attention weight to obtain the context features. The contextual features are fused with the query features of the next slice image to be processed, and a blood vessel segmentation mask for the slice image is generated by the decoding module. All slice images are processed sequentially, and the blood vessel segmentation masks of each slice image are integrated to obtain the three-dimensional blood vessel segmentation result.

2. The blood vessel segmentation method based on slice sequence modeling and structure-aware memory according to claim 1, characterized in that, The adaptive spatial expansion of the vascular region to obtain the expanded region specifically includes: A pixel-level radius range map is generated from the query features of the current slice image using function mapping; Based on the extended radius in the radius range map, a distance mapping corresponding to the local offset is defined, and a soft expansion function is constructed; Using the soft dilation function and the blood vessel segmentation mask of the current slice image, an extended mask is obtained through morphological dilation, and the area covered by the extended mask is the extended region.

3. The blood vessel segmentation method based on slice sequence modeling and structure-aware memory according to claim 2, characterized in that, The soft expansion function is expressed as: in, Indicates spatial location, Indicates local offset. Indicates the location Offset at location The coefficient of thermal expansion, This represents the Sigmoid function. Indicates the adjustment parameter. Indicates position The radius of expansion at that location, Indicates offset Corresponding distance mapping; The extended mask is calculated using the following formula: in, Indicates position The extended mask value at that location, This indicates the location of the blood vessel segmentation mask in the current slice image. The value at that location, Indicates all local offsets Perform a series of multiplications.

4. The blood vessel segmentation method based on slice sequence modeling and structure-aware memory according to claim 1, characterized in that, The attention weights are calculated as follows: in, Indicates attention weights. This represents the historical key features stored in the memory unit. This indicates the query features of the next slice to be processed. Represents the square of the Euclidean distance. This represents the normalized exponential function.

5. The blood vessel segmentation method based on slice sequence modeling and structure-aware memory according to claim 1, characterized in that, Before using the attention weights to weight and aggregate historical features to obtain contextual features, a dynamic memory filtering step is also included: Based on the attention weight, the historical key features and historical value features stored in the memory unit are filtered, and only features with attention weights higher than a preset threshold or ranked higher are retained to participate in subsequent weighted aggregation calculations.

6. The blood vessel segmentation method based on slice sequence modeling and structure-aware memory according to claim 1, characterized in that, The method for extracting query features, key features, and value features is as follows: For each slice image, a depth feature map is first extracted through an encoder network, and then mapped to query features, key features and value features respectively through independent linear projection layers.

7. The blood vessel segmentation method based on slice sequence modeling and structure-aware memory according to claim 1, characterized in that, The process of generating a blood vessel segmentation mask from the sliced ​​image via the decoding module is as follows: The decoding module adopts a convolutional neural network structure, which upsamples the fused features step by step, outputs a blood vessel probability map with the same size as the input slice image, and then obtains the final blood vessel segmentation mask through threshold binarization.

8. A blood vessel segmentation system based on slice sequence modeling and structure-aware memory, characterized in that, Includes the following modules: The serialization module is used to convert three-dimensional medical image data into a two-dimensional slice image sequence along a preset direction, select the current slice image in sequence, extract the query features of the current slice image, and generate a blood vessel segmentation mask of the current slice image based on the query features. The memory storage module is used to determine the blood vessel region based on the blood vessel segmentation mask of the current slice image, and to adaptively expand the blood vessel region to obtain an expanded region; extract key features and value features from the current slice image features within the expanded region, and write the key features and value features as structure-aware memory information into the memory unit; The context retrieval module is used to select the next slice image to be processed from the slice image sequence, extract its query features, and calculate the similarity between the query features and the historical key features stored in the memory unit to obtain the attention weight; the attention weight is used to perform weighted aggregation of the historical value features stored in the memory unit to obtain the context features; The segmentation and fusion module is used to fuse the context features with the query features of the next slice image to be processed, and generate a blood vessel segmentation mask for the slice image through the decoding module. The result integration module is used to process all slice images in sequence and integrate the blood vessel segmentation masks of each slice image to obtain the three-dimensional blood vessel segmentation result.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the blood vessel segmentation method based on slice sequence modeling and structure-aware memory as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the blood vessel segmentation method based on slice sequence modeling and structure-aware memory as described in any one of claims 1 to 7.

Images