A three-dimensional image data segmentation method and system

Abstract

The application provides a three-dimensional image data segmentation method and system, and belongs to the technical field of image processing, comprising: inputting three-dimensional image data into a first segmentation network to obtain a preliminary three-dimensional segmentation result; splicing the preliminary three-dimensional segmentation result and the three-dimensional image data in the channel dimension to obtain fusion data; inputting the fusion data into a second segmentation network to obtain a final three-dimensional segmentation result; wherein the second segmentation network is used for processing the fusion data to correct segmentation errors in the preliminary three-dimensional segmentation result. The application constructs a two-stage segmentation network, fuses the preliminary segmentation result and the original image data, and uses the second segmentation network to optimize segmentation errors such as false positives, poor continuity and inaccurate boundaries, thereby significantly improving the segmentation accuracy and robustness of the three-dimensional image data.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a three-dimensional image data segmentation method and system. Background Art

[0002] Three-dimensional image data segmentation technology is an important research direction in image processing and is widely used for target region extraction and analysis based on three-dimensional image data. In existing technologies, three-dimensional segmentation algorithms are usually developed based on traditional image processing methods or single deep learning models. Although traditional methods such as threshold segmentation, region growing, or graph-based segmentation methods have a certain theoretical basis, they often have difficulty in achieving satisfactory results when faced with complex target structures in three-dimensional images. These methods lack robustness to the boundary and regional information of specific targets and are easily affected by noise interference or other high-similarity regions, resulting in insufficient accuracy of segmentation results.

[0003] With the development of deep learning technology, 3D segmentation has gradually shifted to methods based on models such as convolutional neural networks (CNNs). These methods utilize multi-layer network structures to extract features from 3D images and can effectively capture detailed information in local areas. However, single network structures face certain challenges in complex 3D image data, such as insufficient understanding of global contextual information, resulting in defects in the continuity and integrity of segmentation results. In addition, due to the interference of non-target areas in the image data, existing technologies have difficulty effectively suppressing false positive errors, especially in scenes with complex shapes or similar signals, further reducing segmentation accuracy and boundary accuracy.

[0004] Therefore, how to improve the accuracy of three-dimensional image data segmentation has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The present invention provides a three-dimensional image data segmentation method, system, electronic device and storage medium to address the defects in the prior art and improve the accuracy of three-dimensional image data segmentation.

[0006] The present invention provides a three-dimensional image data segmentation method, comprising the following steps:

[0007] Inputting the three-dimensional image data into the first segmentation network to obtain a preliminary three-dimensional segmentation result;

[0008] splicing the preliminary 3D segmentation result with the 3D image data in the channel dimension to obtain fused data;

[0009] The fused data is input into a second segmentation network to obtain a final three-dimensional segmentation result; wherein the second segmentation network is used to process the fused data to correct segmentation errors in the preliminary three-dimensional segmentation result.

[0010] According to a three-dimensional image data segmentation method provided by the present invention, the second segmentation network is trained by the following method:

[0011] Determining a preset segmentation label for the three-dimensional image data sample;

[0012] performing a false positive simulation operation on the three-dimensional image data sample to obtain a first damage label of the three-dimensional image data sample;

[0013] performing a random morphological operation on the first damage label to obtain a second damage label for the three-dimensional image data sample;

[0014] performing a random masking operation on the second damage label to obtain an actual damage label of the three-dimensional image data sample;

[0015] The three-dimensional image data samples and the actual damage labels are spliced ​​in the channel dimension and input into the initial model for training. The preset segmentation labels are used as training targets to train the second segmentation network.

[0016] According to a 3D image data segmentation method provided by the present invention, performing a false positive simulation operation on the 3D image data sample to obtain a first damaged label of the 3D image data sample specifically includes:

[0017] determining a maximum voxel value of the three-dimensional image data sample;

[0018] calculating a screening threshold based on the maximum voxel value;

[0019] All voxel values ​​of the three-dimensional image data sample are binarized based on the screening threshold to obtain the first damage label; wherein the first damage label includes multiple voxel labels, when the voxel value is less than the screening threshold, the corresponding voxel label is set to 0, and when the voxel value is not less than the screening threshold, the corresponding voxel label is set to 1.

[0020] According to a three-dimensional image data segmentation method provided by the present invention, performing a random morphological operation on the first damaged label to obtain a second damaged label of the three-dimensional image data sample specifically includes:

[0021] Identify various morphological operations;

[0022] Use a 3D structure element of a preset size as the structure element for morphological operations;

[0023] A morphological operation is randomly selected, and the randomly selected morphological operation is performed on the first damaged label according to the structural element to obtain the second damaged label.

[0024] According to a 3D image data segmentation method provided by the present invention, performing a random masking operation on the second damaged label to obtain the actual damaged label of the 3D image data sample specifically includes:

[0025] Set the random masking ratio k∈[0,1];

[0026] generating a random mask having the same shape as the second damaged label according to the random masking ratio, wherein: the probability of setting the value of each position in the random mask to 1 is k, and the probability of setting it to 0 is 1-k;

[0027] The random mask is multiplied element-wise by the second damage label to obtain the actual damage label.

[0028] According to a 3D image data segmentation method provided by the present invention, the first segmentation network adopts an encoder-decoder structure; the 3D image data is input into the first segmentation network to obtain a preliminary 3D segmentation result, specifically comprising:

[0029] In the encoder, a dual-branch network is used to extract features from the three-dimensional image data, wherein the dual-branch network includes a convolutional neural network branch and a Transformer branch;

[0030] At different scales of the encoder, encoding and feature extraction are performed on the three-dimensional image data using a convolutional neural network branch and a Transformer branch, respectively, to obtain first feature maps at different scales and second feature maps at different scales;

[0031] Fusing the first feature map and the second feature map at the same scale to obtain multiple fused features;

[0032] In the decoder, all the fused features are decoded through convolution and upsampling operations, and a skip connection is established between the decoder and the corresponding scales of the encoder;

[0033] The preliminary three-dimensional segmentation result is obtained through the output of the decoder.

[0034] According to a three-dimensional image data segmentation method provided by the present invention, the fused data is input into a second segmentation network to obtain a final three-dimensional segmentation result, specifically comprising:

[0035] Inputting the fused data into a second segmentation network, correcting segmentation errors in the preliminary three-dimensional segmentation result in the second segmentation network, including eliminating false positive areas in the preliminary three-dimensional segmentation result, repairing broken parts in the preliminary three-dimensional segmentation result caused by discontinuity, and refining the boundaries of the preliminary three-dimensional segmentation result;

[0036] Output the corrected final 3D segmentation result.

[0037] The present invention also provides a three-dimensional image data segmentation system, comprising the following modules:

[0038] a processing module, configured to input the 3D image data into a first segmentation network to obtain a preliminary 3D segmentation result;

[0039] a splicing module, configured to splice the preliminary 3D segmentation result with the 3D image data in a channel dimension to obtain fused data;

[0040] The processing module is further used to input the fused data into a second segmentation network to obtain a final three-dimensional segmentation result; wherein the second segmentation network is used to process the fused data to correct segmentation errors in the preliminary three-dimensional segmentation result.

[0041] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the above-described three-dimensional image data segmentation methods is implemented.

[0042] The present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements any of the above-mentioned three-dimensional image data segmentation methods.

[0043] The present invention further provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the above-mentioned three-dimensional image data segmentation methods.

[0044] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0045] By inputting 3D image data into the first segmentation network, the network's dual-branch structure extracts local features and global contextual information, respectively, generating a preliminary 3D segmentation result. This preliminary segmentation result provides basic contours and semantic information about the target region, laying the foundation for subsequent optimization. However, due to the complexity of 3D image data, such as background noise and subtle features of the target structure, relying solely on preliminary segmentation can lead to high false positives, discontinuous target regions, and blurred boundaries, necessitating further optimization. By concatenating the preliminary 3D segmentation result with the original 3D image data in the channel dimension to generate fused data, the semantic information of the segmentation result is organically combined with the detailed features of the image data, enriching the feature representation of the network input. The fused data not only provides preliminary localization information for the target region, but also preserves texture details and contrast information in the image, providing more comprehensive contextual support for subsequent processing. This design effectively compensates for the lack of accuracy in the preliminary segmentation due to the single feature set. The fused data is then input into the second segmentation network, where its encoder-decoder structure performs deep feature extraction and optimization on the fused data, thereby correcting segmentation errors in the preliminary 3D segmentation result. The second segmentation network, by specifically addressing false-positive regions, discontinuous target structures, and fuzzy boundaries, gradually corrects errors in the initial segmentation results, significantly improving the overall segmentation quality. False-positive regions are effectively eliminated, broken target structures are repaired, and boundary clarity and fit are significantly enhanced. This design significantly improves the accuracy of the final 3D segmentation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 This is one of the flow charts of the three-dimensional image data segmentation method provided by the present invention.

[0048] Figure 2 This is the second flow chart of the three-dimensional image data segmentation method provided by the present invention.

[0049] Figure 3 This is the third flow chart of the three-dimensional image data segmentation method provided by the present invention.

[0050] Figure 4 This is the fourth flow chart of the three-dimensional image data segmentation method provided by the present invention.

[0051] Figure 5 This is the fifth flow chart of the three-dimensional image data segmentation method provided by the present invention.

[0052] Figure 6 It is a structural diagram of the three-dimensional image data segmentation system provided by the present invention.

[0053] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0055] It should be noted that, in the description of the present invention, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. The orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0056] The terms "first," "second," and so forth, used herein are used to distinguish similar objects, not to describe a specific order or precedence. It should be understood that such terms are interchangeable where appropriate, allowing embodiments of the present invention to be implemented in an order other than that illustrated or described herein. Furthermore, the terms "first," "second," and so forth generally distinguish objects of a single type, and do not limit the number of objects. For example, the first object may be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates an "or" relationship between the connected objects.

[0057] The following combination Figure 1-Figure 7The present invention describes a three-dimensional image data segmentation method, system, electronic device, and storage medium.

[0058] Figure 1 This is one of the flow charts of the three-dimensional image data segmentation method provided by the present invention. Figure 1 As shown, including but not limited to the following steps:

[0059] Step 11: Input the 3D image data into the first segmentation network to obtain a preliminary 3D segmentation result.

[0060] Specifically, in this solution, 3D image data covers a wide range of application scenarios and data types, including medical imaging data, industrial inspection imaging data, autonomous driving perception imaging data, and other types of data. Due to the complexity of imaging targets, imaging conditions, and the shortcomings of existing segmentation methods, 3D image data segmentation often faces common problems such as a high number of false positive segmentation errors, poor continuity, and inaccurate boundaries. The specific manifestations are as follows:

[0061] In medical imaging, 3D image data is used to segment the 3D structure of cells or tissues. However, due to the high similarity in signal intensity between cells or tissues, non-target areas can be easily misidentified as target areas, resulting in false positives. Furthermore, cell segmentation suffers from poor continuity, especially in areas with complex morphology or weak signal strength. Furthermore, fuzzy boundaries make it difficult for segmentation results to reconcile with the actual morphology.

[0062] Industrial CT or nondestructive testing (NDT) is widely used in industrial inspection image data to detect internal defects (such as pores, cracks, or inclusions). Due to the complexity of material structures and noise interference, the segmentation process can easily misidentify non-defective areas as defects, resulting in a large number of false positives. Furthermore, existing methods struggle to ensure segmentation continuity for elongated objects like cracks, resulting in fragmented or incomplete objects. Furthermore, the boundaries of defective areas often lack accurate segmentation results due to weak signal gradients.

[0063] Object segmentation based on LiDAR point clouds or RGB-D images is a critical task in autonomous driving perception imaging data. Due to point cloud sparsity and environmental interference (such as rain, fog, or dynamic objects), non-target areas (such as ground noise or reflected signals) are easily misclassified as target objects, resulting in false positives. Furthermore, segmentation continuity is poor when processing complex 3D structures (such as pedestrians, vehicles, or trees), and some objects are prone to fragmentation. Furthermore, the limited resolution of point clouds or depth maps makes it difficult to clearly segment the boundaries of target objects.

[0064] Therefore, based on the problems of the above-mentioned three-dimensional image data, such as a large number of false positive segmentation errors, poor continuity, and inaccurate boundaries, when processing three-dimensional image data, the task of the first segmentation network in this solution is to perform preliminary segmentation of the input data to extract the main features of the target area, providing a basis for subsequent optimized segmentation. However, three-dimensional image data usually has a complex spatial structure and diverse intensity distribution. It is difficult to effectively capture local details and global relationships in the image by relying solely on traditional single methods, which may lead to problems such as blurred segmentation boundaries and structural discontinuities. To address these problems, the first segmentation network is designed as a dual-branch structure that integrates convolutional neural networks (CNNs) and Transformers. Combining the advantages of both, it performs comprehensive feature extraction and processing on three-dimensional image data.

[0065] In a possible implementation, step 11 specifically includes the following steps:

[0066] In the encoder, a dual-branch network is used to extract features from 3D image data. The dual-branch network includes a convolutional neural network branch and a Transformer branch.

[0067] At different scales of the encoder, the convolutional neural network branch and the Transformer branch are used to encode and extract features of the 3D image data, respectively, to obtain the first feature map at different scales and the second feature map at different scales;

[0068] Fusing the first feature map and the second feature map at the same scale to obtain multiple fused features;

[0069] In the decoder, all fused features are decoded through convolution and upsampling operations, and skip connections are established between the corresponding scales of the decoder and the encoder;

[0070] The preliminary three-dimensional segmentation result is obtained through the output of the decoder.

[0071] In the segmentation and processing method of three-dimensional image data provided in the above embodiment, the core task of the first segmentation network is to encode and decode the input three-dimensional image data, extract the key features of the target area and generate a preliminary three-dimensional segmentation result. In order to make full use of the complementarity of local and global information in three-dimensional image data, the encoder adopts a dual-branch network structure, which includes a convolutional neural network branch and a Transformer branch. The convolutional neural network branch can efficiently extract detailed features such as edges and textures in the image through the local receptive field, while the Transformer branch captures long-range spatial dependencies through the self-attention mechanism and models the global context information of the image. Through this design, the dual-branch network can realize the joint modeling of local details and global features, making up for the deficiency that traditional methods are difficult to take into account both types of information.

[0072] At different scales within the encoder, 3D image data is first fed into the convolutional neural network branch and the Transformer branch for encoding, generating the first and second feature maps, respectively. This multi-scale encoding approach progressively extracts image features from low-level to high-level layers, capturing salient features of target regions at varying resolutions. This multi-scale feature extraction approach also helps improve the network's ability to represent complex target shapes, enabling it to capture boundary details while maintaining global structural integrity.

[0073] To further enhance the ability to collaboratively express local and global information, the first and second feature maps at the same scale are integrated through a feature fusion module. This fusion can be achieved by element-by-element summation or channel-wise concatenation, combining the two features into a unified fused feature. The fused feature incorporates both the boundary details provided by the convolutional branch and the contextual relationships provided by the Transformer branch, enabling the network to utilize richer semantic information for reconstruction during the subsequent decoding phase.

[0074] After entering the decoder, the fused features are gradually restored to their original resolution through convolution and upsampling. Skip connections are introduced during the decoding phase to directly transfer high-resolution features of the corresponding scale from the encoder to the decoder. The skip connection design effectively mitigates the resolution loss caused by multi-layer downsampling, enhancing the network's ability to recover image detail. This design also improves the segmentation network's accuracy in boundary regions, reducing blurred or discontinuous boundaries.

[0075] The decoder ultimately outputs a feature map reconstructed through layer-by-layer decoding, generating a preliminary 3D segmentation result. This segmentation result, through multi-scale feature extraction and fusion in the encoding phase, and resolution restoration and detail enhancement in the decoding phase, demonstrates high accuracy in terms of boundary clarity, overall continuity, and detail integrity in the target region. This design not only improves the quality of the preliminary segmentation result but also provides reliable input for subsequent optimization of false positives and imprecise boundaries in the secondary segmentation network.

[0076] Step 12: Combine the preliminary 3D segmentation results with the 3D image data in the channel dimension to obtain fused data.

[0077] In the task of 3D image data segmentation, although the preliminary 3D segmentation results provide basic outline information of the target area, they still suffer from a high number of false positive segmentation errors, blurred boundaries, and poor continuity. These problems are often caused by the complex image background and subtle features of the target structure, and are difficult to resolve with preliminary segmentation alone. Therefore, to improve the segmentation effect, the preliminary 3D segmentation results are fused with the original 3D image data. This can simultaneously utilize the semantic information in the segmentation results and the detailed features in the original data, further enhancing the network's ability to express the target area.

[0078] Specifically, the initial 3D segmentation results are aligned with the 3D image data and then spliced ​​together along the channel dimension to generate fused data with multiple channels. During the fusion process, the initial segmentation results serve as additional semantic information to guide the network's focus on the location and basic outline of the target area, while the original 3D image data provides high-resolution details and texture information of the target area. Through this splicing operation, the fused data can provide comprehensive input to the second segmentation network, allowing the network to rely on both semantic information and image details for further optimization.

[0079] Step 13: Input the fused data into the second segmentation network to obtain the final three-dimensional segmentation result; wherein the second segmentation network is used to process the fused data to correct segmentation errors in the preliminary three-dimensional segmentation result.

[0080] During the segmentation process of 3D image data, although the initial 3D segmentation results contain the general outline of the target area, they still cannot directly meet the requirements for accurate segmentation due to issues such as a high number of false positive segmentation errors, structural discontinuities, and imprecise boundaries. To further optimize the segmentation results, inputting the fused data into a secondary segmentation network is a key step. This secondary segmentation network is designed to correct segmentation errors in the initial 3D segmentation results. Through deep feature extraction and detail optimization, it eliminates incorrectly segmented areas, enhances the continuity and boundary accuracy of the segmentation results, and thus produces a more accurate final 3D segmentation result.

[0081] The following embodiment is a specific training method for the second segmentation network.

[0082] In one possible implementation, refer to Figure 2 , Figure 2 This is the second flow chart of the three-dimensional image data segmentation method provided by the present invention. Figure 2 As shown, the method for training the second segmentation network specifically includes steps 21-25:

[0083] Step 21: Determine the preset segmentation labels of the 3D image data samples.

[0084] In training the second segmentation network, it is essential to determine the preset segmentation label of the three-dimensional image data sample. The preset segmentation label is the true segmentation result of the three-dimensional image data sample, which is used to guide the network to learn how to identify the target region and correct segmentation errors in training.

[0085] Specifically, the preset segmentation label usually comes from manual annotation or high-precision reference method. In order to ensure the reliability of the label, each voxel of the three-dimensional image data will be assigned a class label to indicate whether the voxel belongs to the target region. For complex three-dimensional image data samples, the preset segmentation label needs to fully reflect the boundary information, spatial structure and distinguishing characteristics of the target region from the background. These labels are not only the basis for network training, but also the benchmark for subsequent damage operation and network optimization.

[0086] By establishing a corresponding relationship between the preset segmentation label and the three-dimensional image data sample, the voxel class information of each data sample is clear and searchable. In the training process, the output result of the network is compared with the preset segmentation label, the error is calculated and the model parameters are adjusted through back propagation.

[0087] Step 22: Perform false positive simulation operation on the three-dimensional image data sample to obtain the first damage label of the three-dimensional image data sample.

[0088] In training the second segmentation network, in order to enhance its ability to correct false positive errors, it is necessary to generate a first damage label with false positive errors by performing false positive simulation operation on the three-dimensional image data sample. The purpose of this process is to simulate the common false positive segmentation errors in the real segmentation process, so that the network can be exposed to these problems in the training stage and master the ability to correct false positive errors through learning.

[0089] In one possible implementation, with reference to Figure 3 , Figure 3 is a flowchart of the three-dimensional image data segmentation method provided by the present application, as shown in Figure 3 Step 22 specifically includes steps 221-223:

[0090] Step 221: Determine the maximum voxel value of the three-dimensional image data sample.

[0091] Determining the maximum voxel value of a 3D image data sample is a critical first step in performing a false-positive simulation. The primary purpose of this process is to provide a basis for subsequent calculations of the screening threshold, so that the generated first damaged label can accurately simulate the distribution characteristics of false-positive segmentation errors. 3D image data typically contains voxel values ​​of varying intensities. The maximum voxel value represents the highest brightness area or signal intensity in the data sample, reflecting the extreme values ​​of the target area or highlighted background. By obtaining this value, the threshold calculation can be dynamically adjusted, making the false-positive simulation operation more closely resemble the error characteristics in the actual segmentation task.

[0092] Specifically, for a given 3D image data sample , it is necessary to traverse the positions (i, j, k) of all voxels in the data sample and find the maximum value of the image data This calculation method can efficiently determine the highest intensity value in the image data without being affected by background noise or local features. At the same time, the determination of the maximum voxel value m is global, ensuring that the subsequent threshold calculation is based on the characteristics of the entire data sample, rather than being limited to a local area.

[0093] Step 222: Calculate a screening threshold based on the maximum voxel value.

[0094] The purpose of the false-positive simulation operation, based on the maximum voxel value of the 3D image data samples, is to dynamically adjust the threshold so that the false-positive simulation can adapt to the brightness distribution characteristics of different types of 3D image data, thereby generating a first corrupted label that is closer to the actual segmentation error. The setting of the screening threshold not only affects the distribution of false-positive regions but also determines the relevance and effectiveness of the simulation results for training the network.

[0095] Specifically, the screening threshold is calculated based on the determined maximum voxel value m. By setting a fixed ratio r∈[0,1], the screening threshold t=r×m is calculated. The ratio r is an adjustable parameter that controls the range and density of generated false-positive regions. When r is small, the screening threshold t is low, more voxels are marked as target regions, and the false-positive distribution in the simulation results is denser. When r is large, the screening threshold t is high, only voxels in high-intensity regions are marked as targets, and the false-positive distribution in the simulation results is relatively sparse.

[0096] Step 223: Binarize all voxel values ​​of the three-dimensional image data sample based on the screening threshold to obtain a first damage label; wherein the first damage label includes multiple voxel labels, when the voxel value is less than the screening threshold, the corresponding voxel label is set to 0, and when the voxel value is not less than the screening threshold, the corresponding voxel label is set to 1.

[0097] In the false-positive simulation operation, the core purpose of binarizing the 3D image data samples based on the screening threshold is to simulate the distribution characteristics of false-positive segmentation errors by setting a classification standard for voxel intensity to distinguish high-intensity areas from low-intensity areas. The binarization design can effectively separate the target area from the background area in the 3D image data, while introducing mislabeled areas to provide training data for the network to prevent false-positive errors.

[0098] Specifically, the calculated screening threshold is used as the criterion for voxel classification, and each voxel value V(i,j,k) in the 3D image data sample is compared. <t时,将该体素标记为背景(标签值为0);当体素值V(i,j,k)≥t时,将该体素标记为目标(标签值为1)。这一操作生成了包含全体体素分类结果的第一损坏标签 , which has the same shape as the original image data, but the label value is only 0 or 1.

[0099] This binarization logic accurately simulates common false positive errors in segmentation tasks. By adjusting the screening threshold t, the number and distribution of false positive regions can be flexibly controlled. For example, when the threshold is low, high-intensity background regions are more likely to be mislabeled as targets, thereby expanding the distribution of false positives. Conversely, when the threshold is high, only some high-intensity regions are mislabeled, making the distribution of false positives more sparse. This dynamic adaptation ensures that simulation results cover a wide range of false positive scenarios, enhancing the diversity of network training.

[0100] The first damaged labels generated through binarization successfully introduce false positives into the damaged labels, providing realistic simulated segmentation error samples for the second segmentation network. This design effectively improves the network's ability to identify and correct false positive regions during training, laying a solid foundation for high accuracy and low error rates in the final segmentation results. Furthermore, the damaged label generation method is simple and efficient, making it applicable to various types of 3D image data.

[0101] Step 23: Perform a random morphological operation on the first damaged label to obtain a second damaged label of the three-dimensional image data sample.

[0102] After the false positive simulation operation generates the first corrupted label, in order to further enhance the second segmentation network's ability to learn complex segmentation errors, it is necessary to perform a random morphological operation on the first corrupted label to generate a second corrupted label. The purpose of this operation is to simulate common regional morphological irregularities in segmentation tasks by randomly introducing morphological changes, such as morphological changes in the target area, boundary errors, or local structural damage, thereby increasing the diversity of the training data and improving the network's robustness to complex morphological segmentation errors.

[0103] In one possible implementation, refer to Figure 4 , Figure 4 This is the fourth flow chart of the three-dimensional image data segmentation method provided by the present invention. Figure 4 As shown, step 23 specifically includes steps 231-233:

[0104] Step 231: Determine multiple morphological operations.

[0105] When performing random morphological operations on the first corrupted label, identifying multiple morphological operations is a key step in achieving morphological diversity simulation. Common errors in segmentation tasks include enlargement, reduction, fracture, or morphological irregularities in the target region, which are difficult to fully simulate with a single morphological operation. Therefore, by pre-defining multiple morphological operations such as dilation, erosion, opening, and closing, we lay the foundation for generating a richer second corrupted label, enabling the network to learn more segmentation error types and their correction strategies during training.

[0106] Specifically, the various morphological operations include but are not limited to the following: no morphological operation; dilation is used to expand the boundary of the target region, simulating the expansion of false positive regions; erosion is used to shrink the boundary of the target region, simulating errors such as shrinkage or incomplete boundaries; opening is performed by first eroding and then dilating, simulating the removal of small false positive regions or the elimination of background noise; closing is performed by first dilating and then eroding, simulating the smoothing or locally broken connection of the boundary of the target region. By selecting these classic morphological operations, we ensure that the simulated segmentation errors cover a variety of common scenarios.

[0107] Step 232: Use a three-dimensional structure element of a preset size as a structure element for morphological operations.

[0108] Specifically, in random morphological operations, the selection of appropriate structuring elements is the key to the morphological processing's ability to accurately simulate segmentation errors. The shape and size of the structuring element directly determine the scope and effect of the morphological operation on the first damaged label. In order to ensure that the morphological operation can reflect the local characteristics of the segmentation error while avoiding excessive modification of the target area, a three-dimensional cross structuring element is selected as the three-dimensional structuring element. The three-dimensional cross structuring element consists of a central voxel and its neighboring voxels in six main directions. This design can perform detailed processing on the local features of the target area, such as expanding boundaries, shrinking areas, or smoothing edges, without causing excessive impact on non-local areas in three dimensions. The preset size of the structuring element is preferably 3×3×3, which not only ensures the high efficiency of the morphological operation, but also can flexibly adapt to features of different scales in three-dimensional image data.

[0109] The reason for selecting this structural element is that it can accurately simulate the morphological problems commonly seen in segmentation, such as insufficient boundary expansion or connection of false positive regions. By limiting the scope of morphological operations, it is ensured that the generated second damage label not only contains representative segmentation errors, but also maintains a high degree of consistency with the characteristics of the true data. This design avoids the distortion of segmentation errors caused by oversized or undersized structural elements, while improving the authenticity and diversity of the damage label.

[0110] Step 233: Randomly select a morphological operation and perform a random selection of morphological operations on the first damage label according to the structural element to obtain a second damage label.

[0111] Specifically, when randomly selecting, a morphological operation is uniformly sampled from no operation, dilation, erosion, opening operation and closing operation. The dilation operation expands the target region boundary to simulate the segmentation error of target region expansion or false positive region enlargement; the erosion operation shrinks the target region boundary to simulate the segmentation error of target region shrinkage or boundary loss; the opening operation processes by first erosion and then dilation to simulate the removal of small false positive regions; the closing operation processes by first dilation and then erosion to simulate the connection of boundary smoothing or local fracture. Random selection ensures that the generation of different segmentation error types has uncertainty, making the training data more diverse.

[0112] Once the randomly selected operation is determined, a preset 3x3x3 three-dimensional cross-shaped structural element is used to perform morphological transformation on the first damage label to obtain a label damaged by random morphological operation The structural element limits the operation range, so that the change is mainly concentrated in the local area, thereby avoiding excessive impact on the overall morphology. The second damage label after transformation can more realistically simulate the morphological errors in the target region, including insufficient boundary expansion, connection of false positive regions, or residual small noise, etc.

[0113] By randomly selecting morphological operations, the generation of the second damage label is no longer limited to fixed error forms, but can dynamically present different segmentation error characteristics. This design significantly enhances the diversity of training data, improves the ability of the second segmentation network to deal with different segmentation errors during training, and enables it to better identify and correct various types of segmentation errors in actual tasks, providing an important guarantee for the high precision and robustness of the final segmentation result.

[0114] Step 24: Perform a random masking operation on the second damage label to obtain the actual damage label of the three-dimensional image data sample.

[0115] After generating the second corrupted label, in order to further enrich the types and manifestations of segmentation errors, a random masking operation needs to be performed on the second corrupted label to simulate the common target region information loss or partial occlusion in the segmentation task. The core purpose of this operation is to introduce uncertainty in the second corrupted label through a randomized masking strategy, generate actual corrupted labels with higher complexity and diversity, and thus enhance the adaptability of the second segmentation network to segmentation errors.

[0116] In a possible implementation, referring to Figure 5 , Figure 5 is a flowchart of the three-dimensional image data segmentation method provided by the present application, as shown in Figure 5 , step 24 specifically includes the following steps:

[0117] Step 241: Set a random masking ratio k e [0, 1].

[0118] In the random masking operation, setting the random masking ratio k e [0, 1] is a key step for generating a random mask. The setting of the masking ratio directly affects the proportion of voxels that are retained and masked in the actual corrupted label, thereby controlling the range and degree of target region information loss. Flexible adjustment of this ratio can simulate the partial absence or occlusion of the target region under different scenarios, making the generated actual corrupted label more diverse and representative, thereby providing more targeted training samples for the second segmentation network.

[0119] Specifically, the masking ratio k represents the probability that each voxel value in the generated random mask is 1 (retained), and 1-k represents the probability that the voxel value is 0 (masked). When the masking ratio k is small, most voxels will be masked, and the degree of loss of target region information in the actual corrupted label is high; when the masking ratio k is large, the information retention degree of the target region is high, and only a small amount of voxels are masked. This flexible ratio control allows the random masking operation to adapt to segmentation scenarios of different complexity.

[0120] Step 242: According to the random masking ratio, generate a random mask with the same shape as the second corrupted label, wherein: for each position in the random mask, the probability of setting the value to 1 is k, and the probability of setting the value to 0 is 1-k.

[0121] In the random masking operation, generating a random mask for the second corrupted label according to the set random masking ratio k is a key step. The generated random mask determines which voxels in the second corrupted label will be retained and which voxels will be masked, thereby simulating the loss of target region information. The design of the random mask needs to fully consider the diversity of segmentation errors, and through the introduction of randomness, the complexity of the corrupted label is enhanced, providing more challenging samples for the training of the second segmentation network.

[0122] Specifically, the random mask has the same shape D x H x W as the second corrupted label. The mask value at each position is determined by random sampling, where the probability of the value being 1 is k (indicating that the corresponding voxel is retained), and the probability of the value being 0 is 1-k (indicating that the corresponding voxel is masked). This randomized design can introduce uncertainty at different positions of the label, making the target region exhibit multiple information loss distribution characteristics in different samples.

[0123] The process of generating the random mask ensures the diversity of the masking effect. For example, at a lower masking proportion k, most voxels are masked, and the generated corrupted label can simulate a scenario of severe target region loss; while at a higher masking proportion k, only a small number of voxels are masked, simulating a scenario of slight missing or noise occlusion of the target region. This flexibility enables the generated corrupted label to cover different degrees of segmentation errors, providing a more comprehensive error sample for the learning process of the network.

[0124] Step 243: Element-wise multiplication of the random mask and the second corrupted label to obtain the actual corrupted label.

[0125] In the random masking operation, the purpose of this operation of element-wise multiplication of the generated random mask and the second corrupted label is to control the retention and masking of voxels of the target region through the mask, generate an actual corrupted label with random information loss characteristics, and thus simulate scenarios of partial occlusion or loss of the target region in the segmentation task, providing diverse input samples for the training of the second segmentation network.

[0126] Specifically, the second corrupted label L2 and the random mask M have the same shape, both D x H x W. The voxels of the random mask M with a value of 1 indicate that the label at the corresponding position is retained, and the voxels with a value of 0 indicate that the label at the corresponding position is masked. Through element-wise multiplication, the actual corrupted label L actual is generated, and its calculation formula is:

[0127] ;

[0128] where (i, j, k) is the position of the voxel. When , the label value remains unchanged; when , the label value is forced to be 0, indicating that the voxel is masked.

[0129] This process can randomly introduce information loss in the second corrupted label, and the generated actual corrupted label contains random distribution of target region loss characteristics. By adjusting the masking proportion k in the mask generation, the degree of target region information loss in the actual corrupted label can be flexibly controlled, from slight loss to large-scale occlusion.

[0130] The element-wise multiplication of the random mask with the second corrupted label imbues the actual corrupted label with greater randomness and complexity. This design allows the second segmentation network to be exposed to a variety of segmentation error scenarios during training, including partial loss of target regions and loss of boundary information. This significantly improves the network's ability to cope with complex segmentation errors in real-world tasks, providing a crucial guarantee for the high accuracy and robustness of the final segmentation results.

[0131] Step 25: The 3D image data samples and the actual damage labels are spliced ​​in the channel dimension and input into the initial model for training. The preset segmentation labels are used as training targets to train a second segmentation network.

[0132] Specifically, the 3D image data sample V and the actual damage label L actual First, the image data is concatenated along the channel dimension to generate fused data F, where each channel contains both the intensity information of the original image data and the category information of the damaged labels. This fused data preserves the detailed features of the image data while also introducing simulated segmentation error features through the actual damaged labels, providing rich input information for network learning.

[0133] The fused data F is input to the initial model for training. During the training process, the network extracts the features of the input data through the encoder, gradually identifies the characteristic patterns of the target area and the distribution characteristics of segmentation errors, and generates optimized segmentation results in the decoding stage. The network output result is consistent with the preset segmentation label L true The loss function is generated by comparing the two and calculating the difference between them. The loss function is often in the form of cross entropy loss or Dice coefficient loss, which is used to quantify the deviation between the network prediction result and the true label.

[0134] The loss value is used to update the network parameters through a back-propagation mechanism, allowing the network to gradually learn how to extract effective features from the fused data and correct segmentation errors. Specifically, by learning the error characteristics of the actual damaged labels, the network can specifically eliminate false positive areas, repair discontinuities in the target structure, and optimize the accuracy of the target boundary.

[0135] Through this training process, the second segmentation network gradually acquires the ability to identify and correct various types of segmentation errors. The fused data design ensures rich input information for the network. The generation of actual damage labels simulates typical error scenarios in segmentation tasks, while the pre-set segmentation labels provide a clear learning objective. Ultimately, the trained second segmentation network significantly improves segmentation accuracy and robustness, providing a reliable solution for high-quality segmentation of 3D imaging data.

[0136] In a possible implementation, step 13 specifically includes the following steps:

[0137] Inputting the fused data into the second segmentation network, correcting the segmentation errors in the preliminary three-dimensional segmentation result in the second segmentation network, including eliminating false positive areas in the preliminary three-dimensional segmentation result, repairing the broken parts caused by discontinuity in the preliminary three-dimensional segmentation result, and refining the boundaries of the preliminary three-dimensional segmentation result;

[0138] Output the corrected final 3D segmentation result.

[0139] Specifically, the fused data is generated by concatenating the initial 3D segmentation results and the original 3D image data in the channel dimension. This data contains preliminary semantic information about the target region while preserving the detailed features of the original image. The second segmentation network uses an encoder-decoder structure to extract deep features from the fused data through multi-layer convolution and feature aggregation. During the encoding phase, the network identifies possible false positive regions in the fused data, breakpoints in the target structure, and the location and characteristics of regions with blurred boundaries.

[0140] During the decoding phase, the network utilizes skip connections to transfer high-resolution features from the encoding phase directly to the decoding phase, while simultaneously restoring the spatial resolution of the fused data through gradual upsampling. During this process, the network uses contextual information from the feature maps to eliminate false positives, repair broken target structures, and refine boundary regions, ensuring a more complete target region with clearer boundaries. By combining the encoding and decoding processes, the second segmentation network is able to comprehensively consider both global and local information, accurately correcting errors in the initial segmentation results.

[0141] After processing by the second segmentation network, the resulting optimized output is the final 3D segmentation result. This optimized segmentation result eliminates the main errors in the initial segmentation result, including redundant false-positive regions, discontinuities in the target structure, and blurred boundaries. The generation of the final segmentation result completes the segmentation optimization process, achieving significantly better accuracy and robustness than the initial segmentation result.

[0142] By outputting a corrected final 3D segmentation result, the second segmentation network not only addresses the shortcomings of the initial segmentation stage but also achieves high-precision processing of complex segmentation tasks by learning the characteristics of the fused data. This result significantly improves the practicality of segmentation tasks and provides a reliable foundation for subsequent analysis and applications based on the segmentation results.

[0143] Reference Figure 6 , Figure 6 Schematic diagram of the structure of the 3D image data segmentation system provided by the present invention, the system includes:

[0144] a processing module, configured to input the 3D image data into a first segmentation network to obtain a preliminary 3D segmentation result;

[0145] a splicing module configured to splice the preliminary three-dimensional segmentation result and the three-dimensional image data in a channel dimension to obtain fusion data;

[0146] The processing module is further configured to input the fusion data into a second segmentation network to obtain a final three-dimensional segmentation result; the second segmentation network is configured to process the fusion data to correct segmentation errors in the preliminary three-dimensional segmentation result.

[0147] In a possible implementation, the system further comprises a model training module configured to:

[0148] determine a preset segmentation label of a three-dimensional image data sample;

[0149] perform a false positive simulation operation on the three-dimensional image data sample to obtain a first damage label of the three-dimensional image data sample;

[0150] perform a random morphological operation on the first damage label to obtain a second damage label of the three-dimensional image data sample;

[0151] perform a random masking operation on the second damage label to obtain an actual damage label of the three-dimensional image data sample;

[0152] splice the three-dimensional image data sample and the actual damage label in the channel dimension, input the three-dimensional image data sample and the actual damage label into an initial model for training, and use the preset segmentation label as a training target to train the second segmentation network.

[0153] In a possible implementation, the model training module is further configured to:

[0154] determine a maximum voxel value of the three-dimensional image data sample;

[0155] calculate a screening threshold based on the maximum voxel value;

[0156] perform a binaryzation processing on all voxel values of the three-dimensional image data sample based on the screening threshold to obtain the first damage label; the first damage label comprises a plurality of voxel labels, when a voxel value is less than the screening threshold, a corresponding voxel label is set to 0, and when the voxel value is not less than the screening threshold, the corresponding voxel label is set to 1.

[0157] In a possible implementation, the model training module is further configured to:

[0158] determine a plurality of morphological operations;

[0159] use a three-dimensional structure element with a preset size as a structure element of the morphological operation;

[0160] A morphological operation is randomly selected, and the randomly selected morphological operation is performed on the first damaged label according to the structural element to obtain a second damaged label.

[0161] In one possible implementation, the model training module is further configured to:

[0162] Set the random masking ratio k∈[0,1];

[0163] Generate a random mask with the same shape as the second damaged label according to the random mask ratio, where: the probability of setting the value of each position in the random mask to 1 is k, and the probability of setting it to 0 is 1-k;

[0164] The random mask is element-wise multiplied with the second damaged label to obtain the actual damaged label.

[0165] In a possible implementation, the processing module is further configured to:

[0166] In the encoder, a dual-branch network is used to extract features from 3D image data. The dual-branch network includes a convolutional neural network branch and a Transformer branch.

[0167] At different scales of the encoder, the convolutional neural network branch and the Transformer branch are used to encode and extract features of the 3D image data, respectively, to obtain the first feature map at different scales and the second feature map at different scales;

[0168] Fusing the first feature map and the second feature map at the same scale to obtain multiple fused features;

[0169] In the decoder, all fused features are decoded through convolution and upsampling operations, and skip connections are established between the corresponding scales of the decoder and the encoder;

[0170] The preliminary three-dimensional segmentation result is obtained through the output of the decoder.

[0171] In a possible implementation, the processing module is further configured to:

[0172] Inputting the fused data into the second segmentation network, correcting the segmentation errors in the preliminary three-dimensional segmentation result in the second segmentation network, including eliminating false positive areas in the preliminary three-dimensional segmentation result, repairing the broken parts caused by discontinuity in the preliminary three-dimensional segmentation result, and refining the boundaries of the preliminary three-dimensional segmentation result;

[0173] Output the corrected final 3D segmentation result.

[0174] It should be noted that the three-dimensional image data segmentation system provided by the present application can execute the three-dimensional image data segmentation method of any of the above embodiments during specific operation, and the present embodiment will not be described here.

[0175] Figure 7 is a structural schematic diagram of an electronic device provided by the present application, as Figure 7 shown, the electronic device can include a processor 710, a communications interface 720, a memory 730, and a communications bus 740, wherein the processor 710, the communications interface 720, and the memory 730 complete mutual communication through the communications bus 740. The processor 710 can call the logic instructions in the memory 730 to execute the three-dimensional image data segmentation method, which includes: inputting three-dimensional image data to a first segmentation network to obtain a preliminary three-dimensional segmentation result; splicing the preliminary three-dimensional segmentation result and the three-dimensional image data in the channel dimension to obtain fusion data; inputting the fusion data to a second segmentation network to obtain a final three-dimensional segmentation result; wherein the second segmentation network is used to process the fusion data to correct segmentation errors in the preliminary three-dimensional segmentation result.

[0176] In addition, the logic instructions in the memory 730 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0177] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the three-dimensional image data segmentation method provided by the above-mentioned embodiments, the method including: inputting the three-dimensional image data into a first segmentation network to obtain a preliminary three-dimensional segmentation result; splicing the preliminary three-dimensional segmentation result with the three-dimensional image data in the channel dimension to obtain fused data; inputting the fused data into a second segmentation network to obtain a final three-dimensional segmentation result; wherein the second segmentation network is used to process the fused data to correct segmentation errors in the preliminary three-dimensional segmentation result.

[0178] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by the processor 710, it is implemented to perform the three-dimensional image data segmentation method provided in the above-mentioned embodiments, the method comprising: inputting the three-dimensional image data into a first segmentation network to obtain a preliminary three-dimensional segmentation result; splicing the preliminary three-dimensional segmentation result with the three-dimensional image data in the channel dimension to obtain fused data; inputting the fused data into a second segmentation network to obtain a final three-dimensional segmentation result; wherein the second segmentation network is used to process the fused data to correct segmentation errors in the preliminary three-dimensional segmentation result.

[0179] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0180] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods of each embodiment or certain portions of the embodiments.

[0181] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A three-dimensional image data segmentation method, characterized in that: include: Inputting the three-dimensional image data into the first segmentation network to obtain a preliminary three-dimensional segmentation result; splicing the preliminary 3D segmentation result with the 3D image data in the channel dimension to obtain fused data; The fused data is input into a second segmentation network to obtain a final 3D segmentation result; wherein the second segmentation network is used to process the fused data to correct segmentation errors in the preliminary 3D segmentation result; the second segmentation network is trained by the following method: Determining a preset segmentation label for the three-dimensional image data sample; performing a false positive simulation operation on the three-dimensional image data sample to obtain a first damage label of the three-dimensional image data sample; performing a random morphological operation on the first damage label to obtain a second damage label for the three-dimensional image data sample; performing a random masking operation on the second damage label to obtain an actual damage label of the three-dimensional image data sample; The three-dimensional image data samples and the actual damage labels are spliced ​​in the channel dimension and input into the initial model for training. The preset segmentation labels are used as training targets to train the second segmentation network.

2. The three-dimensional image data segmentation method according to claim 1, characterized in that: The performing a false positive simulation operation on the three-dimensional image data sample to obtain a first damaged label of the three-dimensional image data sample specifically includes: determining a maximum voxel value of the three-dimensional image data sample; calculating a screening threshold based on the maximum voxel value; All voxel values ​​of the three-dimensional image data sample are binarized based on the screening threshold to obtain the first damage label; wherein the first damage label includes multiple voxel labels, when the voxel value is less than the screening threshold, the corresponding voxel label is set to 0, and when the voxel value is not less than the screening threshold, the corresponding voxel label is set to 1.

3. The three-dimensional image data segmentation method according to claim 1, characterized in that: The performing a random morphological operation on the first damaged label to obtain a second damaged label of the three-dimensional image data sample specifically includes: Identify various morphological operations; Use a 3D structure element of a preset size as the structure element for morphological operations; A morphological operation is randomly selected, and the randomly selected morphological operation is performed on the first damaged label according to the structural element to obtain the second damaged label.

4. The three-dimensional image data segmentation method according to claim 1, wherein: The performing a random masking operation on the second damaged label to obtain the actual damaged label of the three-dimensional image data sample specifically includes: Set the random masking ratio k∈[0,1]; generating a random mask having the same shape as the second damaged label according to the random masking ratio, wherein: the probability of setting the value of each position in the random mask to 1 is k, and the probability of setting it to 0 is 1-k; The random mask is multiplied element-wise by the second damage label to obtain the actual damage label.

5. The three-dimensional image data segmentation method according to claim 1, characterized in that: The first segmentation network adopts an encoder-decoder structure; the step of inputting the three-dimensional image data into the first segmentation network to obtain a preliminary three-dimensional segmentation result specifically includes: In the encoder, a dual-branch network is used to extract features from the three-dimensional image data, wherein the dual-branch network includes a convolutional neural network branch and a Transformer branch; At different scales of the encoder, encoding and feature extraction are performed on the three-dimensional image data using a convolutional neural network branch and a Transformer branch, respectively, to obtain first feature maps at different scales and second feature maps at different scales; Fusing the first feature map and the second feature map at the same scale to obtain multiple fused features; In the decoder, all the fused features are decoded through convolution and upsampling operations, and a skip connection is established between the decoder and the corresponding scales of the encoder; The preliminary three-dimensional segmentation result is obtained through the output of the decoder.

6. The three-dimensional image data segmentation method according to claim 1, characterized in that: Inputting the fused data into the second segmentation network to obtain the final three-dimensional segmentation result specifically includes: Inputting the fused data into a second segmentation network, correcting segmentation errors in the preliminary three-dimensional segmentation result in the second segmentation network, including eliminating false positive areas in the preliminary three-dimensional segmentation result, repairing broken parts in the preliminary three-dimensional segmentation result caused by discontinuity, and refining the boundaries of the preliminary three-dimensional segmentation result; Output the corrected final 3D segmentation result.

7. A three-dimensional image data segmentation system, characterized in that: include: a processing module, configured to input the 3D image data into a first segmentation network to obtain a preliminary 3D segmentation result; a splicing module, configured to splice the preliminary 3D segmentation result with the 3D image data in a channel dimension to obtain fused data; The processing module is further configured to input the fused data into a second segmentation network to obtain a final three-dimensional segmentation result; wherein the second segmentation network is configured to process the fused data to correct segmentation errors in the preliminary three-dimensional segmentation result; A model training module is used to determine the preset segmentation label of the three-dimensional image data sample; perform a false positive simulation operation on the three-dimensional image data sample to obtain a first damaged label of the three-dimensional image data sample; perform a random morphological operation on the first damaged label to obtain a second damaged label of the three-dimensional image data sample; perform a random masking operation on the second damaged label to obtain an actual damaged label of the three-dimensional image data sample; splice the three-dimensional image data sample and the actual damaged label in the channel dimension, input the concatenation into the initial model for training, and use the preset segmentation label as the training target to train and obtain the second segmentation network.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the three-dimensional image data segmentation method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the three-dimensional image data segmentation method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the three-dimensional image data segmentation method according to any one of claims 1 to 6 is implemented.

Images