Time-of-flight magnetic resonance blood vessel image cerebral vessel segmentation method, system, device and medium

Abstract

The invention discloses a time-of-flight magnetic resonance blood vessel image cerebral vessel segmentation method, system and device and a medium, and relates to the field of medical image processing, and the method comprises the steps: obtaining a to-be-processed time-of-flight magnetic resonance blood vessel image; reinforcing blood vessel features in the to-be-processed time-of-flight magnetic resonance blood vessel image to obtain a preprocessed image; according to the preprocessed image, performing cerebrovascular segmentation by adopting a few-sample segmentation model to obtain a blood vessel probability graph; the few-sample segmentation model is obtained by migrating knowledge of a pre-training video word segmentation device to train 3D U-Net; and performing post-processing on the blood vessel probability graph and the to-be-processed flight time magnetic resonance blood vessel image based on a human-computer interaction interface and a conditional random field to obtain a final segmented image. According to the method, a high-precision and high-robustness segmentation effect can be realized only by a small number of samples, and result optimization can be carried out through an efficient man-machine interaction mode.

Description

Technical Field

[0001] The present application relates to the field of medical image processing, and in particular to a method, system, device and medium for cerebral vascular segmentation in time-of-flight magnetic resonance vascular imaging based on human-computer interaction. Background Art

[0002] Time-of-Flight Magnetic Resonance Angiography (TOF-MRA), a contrast-free, non-invasive neurovascular imaging technique, leverages the magnetic resonance imaging of blood flow and static tissue to demonstrate significant clinical value in the diagnosis and treatment of cerebrovascular diseases. Accurately segmenting the three-dimensional structure of cerebral vessels in TOF-MRA images is a critical preprocessing step for precise quantitative analysis, surgical planning, and scientific research.

[0003] Despite significant progress in automated segmentation technology, TOF-MRA's inherent image quality remains a fundamental challenge to achieving accurate segmentation. TOF-MRA's imaging principle is based on the inflow effect, which, while providing excellent contrast between blood vessels and background tissue, inevitably introduces complex and diverse image quality issues, all of which directly undermine the performance and reliability of automated segmentation algorithms.

[0004] To address the above challenges, cerebral vascular segmentation technology has undergone a development process from traditional image processing methods to modern deep learning models. However, the standard supervised learning paradigm still faces some inherent limitations when applied to TOF-MRA segmentation: First, it heavily relies on large-scale, high-quality annotated data, but the annotation work is time-consuming, costly, and requires professional knowledge, resulting in a scarcity of public datasets; second, data scarcity limits the model's generalization ability, making it prone to overfitting and performance degradation when processing data from different sources; third, the model is sensitive to quality issues such as image noise and artifacts, and insufficient robustness affects segmentation quality; fourth, the ability to segment fine vascular systems is insufficient, and there are challenges in outlining tiny vessels, maintaining vascular network topology, and processing pathological vessels. Some loss functions ignore tiny vessels.

[0005] Furthermore, manual review and correction are essential for ensuring accurate TOF-MRA cerebral vascular segmentation results in clinical practice. While existing interactive segmentation tools offer manual and semi-automatic editing capabilities, these tools often face limitations such as high workload, difficulty learning, poor correction consistency, low efficiency in processing complex 3D structures, and insufficient intelligence.

[0006] In summary, the current automated segmentation methods for TOF-MRA cerebral vascular images face challenges in dealing with data scarcity, image quality variability, small vessel segmentation accuracy, and interactive correction efficiency. These problems limit their widespread application in clinical practice. Summary of the Invention

[0007] The purpose of this application is to provide a method, system, device and medium for cerebral vascular segmentation in time-of-flight magnetic resonance vascular imaging, which can reduce the dependence on large-scale labeled data, improve the robustness of segmentation, and efficiently obtain personalized segmentation results that meet clinical needs.

[0008] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a method for cerebral vascular segmentation in time-of-flight magnetic resonance angiography, comprising: acquiring a time-of-flight magnetic resonance angiographic image to be processed; Enhance the vascular features in the to-be-processed time-of-flight magnetic resonance angiography image to obtain a pre-processed image; Based on the preprocessed image, a few-shot segmentation model is used to segment cerebral blood vessels to obtain a blood vessel probability map; the few-shot segmentation model is obtained by training a 3D U-Net by transferring knowledge from a pre-trained video segmenter; The blood vessel probability map and the to-be-processed time-of-flight magnetic resonance vascular image are post-processed based on a human-computer interaction interface and a conditional random field to obtain a final segmented image.

[0009] In a second aspect, the present application provides a time-of-flight magnetic resonance angiography cerebral vascular segmentation system, comprising: An image acquisition module, used for acquiring a time-of-flight magnetic resonance vascular image to be processed; a preprocessing module for enhancing vascular features in a to-be-processed time-of-flight magnetic resonance angiography image to obtain a preprocessed image; A preliminary segmentation module is configured to segment cerebral blood vessels using a few-shot segmentation model based on the preprocessed image to obtain a blood vessel probability map; the few-shot segmentation model is obtained by training a 3D U-Net by transferring knowledge from a pre-trained video segmenter; The post-processing module is used to post-process the vascular probability map and the time-of-flight magnetic resonance vascular image to be processed based on the human-computer interaction interface and the conditional random field to obtain a final segmented image.

[0010] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for cerebral vascular segmentation in time-of-flight magnetic resonance vascular imaging.

[0011] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for cerebral vascular segmentation in time-of-flight magnetic resonance angiography.

[0012] According to the specific embodiments provided in this application, this application has the following technical effects: The present application provides a method, system, device and medium for cerebral vascular segmentation in time-of-flight magnetic resonance angiography. By introducing a preprocessing process to strengthen vascular features and utilizing the knowledge transfer of pre-trained video segmenters, the few-sample segmentation model only requires a very small number of completely annotated time-of-flight magnetic resonance angiography images to complete training and achieve robust segmentation performance, significantly reducing the dependence on large-scale annotated data. The interactive post-processing process based on the human-computer interaction interface and conditional random fields provides an intuitive visualization interface and convenient modification tools. Users can quickly locate and correct segmentation errors, and can efficiently obtain personalized segmentation results that meet clinical needs. In summary, the present application only requires a small number of samples to achieve high-precision and high-robustness segmentation effects, and can optimize the results through efficient human-computer interaction. It has important research value and broad clinical application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 This is a diagram of the application environment of a method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography in one embodiment of the present application.

[0015] Figure 2 This is a schematic diagram of the overall process of a method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography provided by one embodiment of the present application.

[0016] Figure 3 A detailed flowchart of a method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography provided in one embodiment of the present application is provided.

[0017] Figure 4 FIG. 1 is a flow chart of image preprocessing in one embodiment of the present application.

[0018] Figure 5 Schematic diagram of the network structure of the residual encoder U-Net in one embodiment of the present application.

[0019] Figure 6 Schematic diagram of the fusion process of the features extracted by the pre-trained video segmenter and the feature map extracted by the residual encoder U-Net in one embodiment of the present application.

[0020] Figure 7 This is a schematic diagram of the functional modules of a cerebral vessel segmentation system for time-of-flight magnetic resonance angiography provided by one embodiment of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0022] This application aims to address the problems of poor robustness and low accuracy of existing TOF-MRA cerebrovascular segmentation methods under small sample conditions, as well as the complex operation and low efficiency of existing interactive segmentation tools. Specifically, this application is dedicated to providing a cerebrovascular image segmentation process that requires only a very small number of fully annotated TOF-MRA images for model training to achieve robust segmentation results, and can optimize and verify the results through efficient human-computer interaction.

[0023] The core goal of Few-Shot Learning (FSL) is to enable the model to learn and generalize effectively using a very small number of labeled samples. Common implementation strategies include meta-learning, model fine-tuning, data augmentation, and metric learning.

[0024] Transfer Learning (TL) transfers and applies the knowledge learned by a model pre-trained on a large-scale source domain dataset (such as ImageNet in the field of natural images or Kinetics in the field of videos) to the target medical imaging task.

[0025] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0026] The cerebral blood vessel segmentation method of time-of-flight magnetic resonance angiography provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the time-of-flight magnetic resonance angiography image to be processed to the server 104. After receiving the time-of-flight magnetic resonance angiography image to be processed, the server 104 enhances the vascular features in the time-of-flight magnetic resonance angiography image to be processed to obtain a preprocessed image; based on the preprocessed image, a few-sample segmentation model is used to perform cerebral vascular segmentation to obtain a vascular probability map; based on the human-computer interaction interface and conditional random field, the vascular probability map and the time-of-flight magnetic resonance angiography image to be processed are post-processed to obtain a final segmented image. The server 104 can feed back the obtained final segmented image to the terminal 102. In addition, in some embodiments, the method for cerebral vascular segmentation in time-of-flight magnetic resonance angiography can also be implemented separately by the server 104 or the terminal 102.

[0027] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or a cloud server.

[0028] In an exemplary embodiment, Figure 2 and Figure 3 As shown, a method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, and can also be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, including the following steps 201 to 204.

[0029] Step 201: Acquire a time-of-flight magnetic resonance angiography image to be processed.

[0030] Step 202 : Enhance the blood vessel features in the to-be-processed time-of-flight magnetic resonance angiography image to obtain a pre-processed image.

[0031] In a specific application example, the pre-processed images include time-of-flight MRI angiography images after skull removal and contrast-enhanced time-of-flight MRI angiography images. Figure 4 As shown, the present application designs a set of pre-processing processes including multimodal image registration, skull removal, and histogram normalization to significantly enhance the vascular features in the image and lay the foundation for subsequent accurate segmentation. Specifically, step 202 includes the following steps 21 to 23.

[0032] Step 21: Register the to-be-processed time-of-flight MRI angiography image with the T1-weighted image to obtain a registered time-of-flight MRI angiography image and a registered T1-weighted image. The T1-weighted image and the to-be-processed time-of-flight MRI angiography image are images of the same subject.

[0033] Specifically, based on the Advanced Normalization Tools (ANTs), rigid and non-rigid registration of TOF-MRA images and T1-weighted images is completed, and the spatial correspondence between images of different modalities is accurately established to ensure that subsequent processing is performed in a unified spatial coordinate system.

[0034] The goal of image registration is to find a spatial transformation , making the floating image (such as TOF-MRA images) and the reference image after transformation (e.g. T1-weighted images). This is usually achieved by optimizing a cost function, which can be expressed as: ;in, is the cost function value, is a similarity metric function (e.g. mutual information, normalized cross-correlation), Represents applying a spatial transformation to a floating image , is a regularization term used to constrain the smoothness or rationality of the transformation, is the regularization coefficient.

[0035] For rigid registration, spatial transformation including rotation and translation; for non-rigid registration, spatial transformation Usually a more complex deformation field.

[0036] Step 22: remove the skull from the registered T1-weighted image to generate a binary brain tissue mask, and apply the binary brain tissue mask to the time-of-flight MRI angiography image to be processed to obtain the time-of-flight MRI angiography image after the skull is removed.

[0037] Specifically, the Brain Extraction Tool 2 (BET2) was used to remove skull bones from the registered T1-weighted images (automatically extracting brain tissue regions) to generate a binary brain tissue mask. This binary brain tissue mask was then applied to the TOF-MRA images in the same space to remove interference from non-brain tissue such as the skull and scalp, retaining only vascular and brain parenchyma information, resulting in a TOF-MRA image containing only brain parenchyma and blood vessels.

[0038] Step 23 , performing histogram normalization processing on the time-of-flight MRI angiography image after the skull is removed to obtain a contrast-enhanced time-of-flight MRI angiography image.

[0039] Specifically, the team used a histogram normalization method implemented in the medical image processing library Torchio to globally match and normalize the grayscale distribution of skull-removed TOF-MRA images, significantly improving the contrast of vascular details. By mapping the image grayscale intensity to a predefined standard range or reference histogram, the grayscale distribution between different images was standardized, enhancing the contrast between blood vessels and background tissue.

[0040] The goal of histogram normalization is to transform the input image Histogram of TOF-MRA image after skull removal Transformed into a target histogram , thus obtaining the standardized image (Contrast-enhanced time-of-flight magnetic resonance angiography). This can be achieved using the Cumulative Distribution Function (CDF). Let and The input image and target histogram are in grayscale The cumulative distribution function value at . For each pixel gray value in the input image , its standardized gray value It can be found by: ;in, is the inverse of the target cumulative distribution function. In addition, the implementation in Torchio may involve more complex techniques such as histogram matching based on landmark points.

[0041] Step 203 : Based on the preprocessed image, a few-shot segmentation model is used to segment cerebral blood vessels to obtain a blood vessel probability map.

[0042] The few-shot segmentation model is trained by transferring knowledge from a pre-trained video word segmenter to a 3D U-Net. The few-shot segmentation model is based on the U-Net architecture and incorporates features extracted by the pre-trained video word segmenter at different scales.

[0043] In a specific application example, the pre-trained video segmenter is pre-trained using natural scene video streams and features corresponding to different resolutions. Efficient few-shot learning is achieved by embedding the pre-trained knowledge in natural scene videos. The 3D U-Net is a residual encoder U-Net.

[0044] The training process for the few-shot segmentation model involves obtaining and preprocessing a small dataset of time-of-flight (TOF)-MRA images with complete pixel-level annotations of cerebral vascular vessels. To expand the training set, the TOF-MRA dataset is augmented using data augmentation techniques including random rotation, flipping, elastic deformation, and intensity transformation. All input images (whether single-modal or multimodal) are normalized using the Z-score method before entering the few-shot segmentation model to generate the few-shot training set. Parameters of the residual encoder U-Net are optimized based on the few-shot training set and a pretrained video segmenter.

[0045] During the training of the few-shot segmentation model, the parameters of the pre-trained video word segmenter are frozen, and its extracted features are injected into specific layers of the encoder or decoder of the residual encoder U-Net. Only the parameters of the residual encoder U-Net and, if necessary, the parameters of the feature adaptation layer are trained. This strategy helps to efficiently transfer the general visual features learned by the pre-trained video word segmenter to the medical image segmentation task, especially when training samples are scarce. This ensures the efficient use of pre-trained features, guides the network to train quickly, reduces the interference of different network parameter initializations on the pre-trained features, and improves the robustness of the model.

[0046] In the choice of loss function, for example, a combination of Dice loss and cross entropy loss can be used as the total loss function. It can be a weighted sum of the two: ;in, is the Dice loss, is the cross entropy loss, is the weight coefficient of Dice loss, is the weight coefficient of the cross entropy loss.

[0047] Dice loss is used to measure the predicted segmentation results and the true label The overlap between , for the two-classification problem, is defined as: .

[0048] Or the more general form: ;in, The few-shot segmentation model divides pixels The probability of being predicted as foreground (vessel), It's a pixel The true label (0 or 1, 1 represents blood vessels, 0 represents non-blood vessels), is the total number of pixels, is a small smoothing constant used to prevent the denominator from reaching zero.

[0049] For a binary classification problem (e.g., each pixel is classified as a vessel or non-vessel), the cross-entropy loss takes the form of a binary cross-entropy loss: .

[0050] The optimizer can use the Adaptive Moment Estimation (Adam) optimizer or the Adaptive Moment Estimation Weight (AdamW) optimizer with weight decay. The Adam optimizer combines the advantages of the Adaptive Gradient (AdaGrad) and the Root Mean Square Propagation (RMSProp) algorithms.

[0051] The parameter update rule is: At the iteration, for the parameter Follow the steps (1) to (6) below.

[0052] (1) Calculate the gradient: .

[0053] (2) Update the first-order moment estimate (momentum): .

[0054] (3) Update the second-order moment estimate (speed): .

[0055] (4) Calculate the bias-corrected first-order moment estimate: .

[0056] (5) Calculate the bias-corrected second-order moment estimate: .

[0057] (6) Update parameters: .

[0058] in, It is The loss function value at the iteration, It is The parameters at the iteration, It is The parameters at the iteration, It is The gradient at iteration , It is The first-order moment estimate at the iteration, It is The first-order moment estimate at the iteration, It is The second-order moment estimate at the iteration, It is The second-order moment estimate at the iteration, and is the decay rate (usually close to 1, such as 0.9 and 0.999), It is The first-order moment estimate after bias correction at the iteration, It is The bias-corrected second-order moment estimate at the iteration, is the learning rate.

[0059] Regarding training parameters, for example, if the input 3D data block size is [112, 256, 256] voxels, the batch size can be set to 3. Parameters such as learning rate and training rounds are adjusted according to the specific hardware resources.

[0060] In this application, step 203 includes the following steps 31 and 32.

[0061] Step 31 : Using a pre-trained video segmenter to perform feature encoding on the pre-processed image to obtain multiple features of different resolutions.

[0062] The pre-processed images (e.g., skull-removed time-of-flight MRI and contrast-enhanced time-of-flight MRI, which can be used as multi-channel input or processed separately and then fused) are regarded as a video stream and a pre-trained video segmenter (Vidtok) is used, e.g. vidtok_kl_causal_ 488 _ 16 chn_v 1_1 and vidtok_kl_causal_ 41616_16 chn_v 1_1, the input pre-processed image is encoded into the latent space at different compression ratios (for example, 4×8×8 and 4×16×16), forming features of different resolutions. Specifically, the pre-trained video segmenter includes at least two encoders with different compression ratios, which respectively encode the input 3D TOF-MRA image into the latent space and fuse them with the feature maps of different depths of the residual encoder U-Net.

[0063] For example, vidtok_kl_causal_ 488 _ 16 chn_v 1_1The preprocessed image is compressed and encoded into the latent space at a rate of 4×8×8, and its output features are concatenated and fused with the feature maps of the penultimate layer of the residual encoder U-Net decoder (or the deeper layer of the encoder). vidtok_kl_causal_ 41616_16 chn_v1_1The preprocessed image is compressed and encoded into the latent space at a rate of 4×16×16, and its output features are concatenated and fused with the feature maps of the penultimate layer of the residual encoder U-Net decoder (or the deeper layer of the encoder, that is, before the first level of upsampling after the bottleneck layer).

[0064] In step 32, a pre-trained 3D U-Net is used to segment cerebral blood vessels on the pre-processed image. At the same time, multiple features of different resolutions are fused into the feature maps of different downsampling stages of the 3D U-Net (for example, the second to last layer and the first to last layer of the network) to obtain a blood vessel probability map.

[0065] like Figure 5 As shown, the encoder architecture of the residual encoder U-Net consists of seven processing stages, with input, for example, a 3D TOF-MRA image patch of size [112, 256, 256]. The number of feature maps (channels) in each stage is 32, 64, 128, 256, 256, 256, and 256 (bottleneck layer), respectively. The number of convolutional blocks varies with depth: the first stage has one, the second stage has three, the third stage has four, and the fourth through seventh stages have six each. All convolution operations are based on 3×3×3 three-dimensional convolution kernels ( torch.nn.modules.conv.Conv3d ) with bias enabled ( conv_bias: true ). Regarding the downsampling strategy, the transition from the first to the fifth stage uses convolutions with a stride of [2,2,2] to achieve isotropic downsampling; the transition from the fifth to the sixth stage and from the sixth to the seventh stage uses convolutions with a stride of [1,2,2] to achieve anisotropic downsampling, which means that the resolution is maintained in the Z axis, while the X and Y axes are downsampled by a factor of 2. The stride of the initial convolution (i.e., the first convolution block in the first stage) is [1,1,1]. Residual connections are introduced between the feature extraction modules of the encoder path (for example, within or between each convolution block) to facilitate the training of deep networks and improve gradient flow.

[0066] The decoder architecture of the residual encoder U-Net consists of six upsampling stages, symmetrically structured with the encoder. Each decoder stage performs upsampling via transposed convolution or interpolation combined with convolution, restoring the resolution of the feature map. This is then concatenated with the feature map from the corresponding encoder layer via skip connections, and the concatenated features are then processed by a convolutional block.

[0067] The general layer components in the residual encoder U-Net follow the following design: each convolution operation is connected to a three-dimensional instance normalization (InstanceNorm3d, eps:1e-05, affine:true), followed by a LeakyReLU activation function (torch.nn.LeakyReLU, inplace:true), and no Dropout layer is configured.

[0068] like Figure 6 As shown, to It is the feature map obtained at each processing stage of the residual encoder U-Net encoder, to It is the feature map obtained at each processing stage of the residual encoder U-Net decoder, S is the feature map finally obtained by the encoder.

[0069] Step 204 : post-processing the blood vessel probability map and the to-be-processed time-of-flight magnetic resonance vascular image based on the human-computer interaction interface and the conditional random field to obtain a final segmented image.

[0070] In one specific application example, the human-computer interaction interface uses a 3D Slicer plug-in to display the current segmentation results and receive user correction instructions. Using the vessel probability map as initial input, the interface, integrated into a 3D visualization platform (such as 3DSlicer), allows users to visually inspect and interactively correct the segmentation results. Conditional random fields are then used to optimize and generate a globally optimal segmentation mask based on user corrections. Step 204 includes steps 41 through 45.

[0071] Step 41: Generate a 3D reconstruction model of cerebral blood vessels based on the blood vessel probability map.

[0072] Step 42: Generate a two-dimensional view based on the time-of-flight MRI angiography image to be processed, specifically including two-dimensional views of the time-of-flight MRI angiography image to be processed on three standard sections: sagittal, coronal, and transverse planes.

[0073] Specifically, the TOF-MRA image to be processed and the vascular probability map are loaded into the 3D Slicer environment, and the vascular probability map is used as the initial input of the unary potential of the conditional random field.

[0074] Step 43 : For any round of interaction, based on the current 3D reconstructed model of cerebral blood vessels and the current 2D view, a human-computer interaction interface is used to display the current segmentation result and receive the user's correction instructions.

[0075] Specifically, the current segmentation result is displayed (for example, as a colored outline or semi-transparent mask) by overlaying the current 3D reconstruction model of the cerebral vessels with the current 2D view. Users can use operations such as zooming, panning, and rotating to examine the segmentation result from different angles and levels, quickly locating possible over-segmentation (false positive) or under-segmentation (false negative) areas.

[0076] Step 44 : Modify the current cerebral blood vessel three-dimensional reconstructed model and / or the current two-dimensional view based on the modification instruction to obtain a modified blood vessel probability map.

[0077] Users can use the interactive tools provided by 3D Slicer (such as brushes, erasers, and region selections, which can be reused or customized in this application) to make corrections to the 2D view or the 3D reconstruction model of the cerebral vascular system: For under-segmented areas (where blood vessels are not detected), users can mark the area as "vessels," and the plug-in will increase the probability value of the corresponding voxel (for example, setting it to a higher probability value close to 1); for over-segmented areas (where non-vessels are mistakenly detected as blood vessels), users can mark the area as "background" or "non-vessels," and the plug-in will reduce the probability value of the corresponding voxel (for example, setting it to a lower probability value close to 0). The interactive points or regions provided by the user will serve as hard constraints or strong priors in the conditional random field, directly modifying the unary potential.

[0078] In step 45, the vascular probability map is used as the initial input for the unary potential of the conditional random field (CRF). The user's correction instructions are used as hard constraints or strong priors for the CRF. Based on the modified vascular probability map, the CRF is used to calculate a global optimal segmentation mask. This global optimal segmentation mask is then fed back into the 3D reconstruction model and 2D image of the cerebral vasculature for the next round of interaction. After multiple rounds of interaction, the final segmented image is obtained.

[0079] After the user completes a round of modifications, the CRF optimization process is triggered. By minimizing an energy function consisting of unary and binary potentials, the CRF takes into account the vessel probability map, the user-provided correction instructions, and the spatial continuity between pixels to calculate a globally optimal segmentation mask. This globally optimal segmentation mask is then updated in real time and fed back to each view of the 3D Slicer for user evaluation.

[0080] Specifically, the goal of the conditional random field is to find a label configuration , so that the Gibbs energy function Minimize: ;in, It's a pixel Tags, is a unary potential, representing the pixel Get label The cost is usually derived from the prediction probability of the few-shot segmentation model, and the correction of user interaction can directly affect the unary potential. is a binary potential, representing adjacent pixels and pixels ( , It's a pixel Neighborhood) take labels respectively and The cost is usually used to encourage the smoothness of the segmentation results, such as using the Potts model or a model based on image features (such as intensity, gradient): ;in, Is an indicator function, which is 1 when the condition is met and 0 otherwise. is the weight, which can depend on the pixel and pixels The similarities between them.

[0081] Based on the feedback from the conditional random field, users can perform multiple rounds of interactive corrections and optimizations until a satisfactory final segmentation mask is obtained. The final result can be saved in standard medical imaging formats such as Neuroimaging Informatics Technology Initiative (NIfTI) and Nearly Raw Raster Data (NRRD).

[0082] Through the above steps, this application realizes a complete TOF-MRA cerebral vascular segmentation process from image preprocessing, automatic segmentation based on few-sample learning to efficient human-computer interaction post-processing.

[0083] Compared with the prior art, this application has the following beneficial effects.

[0084] (1) Efficient few-shot learning capability: By introducing a preprocessing process to enhance vascular features and utilizing knowledge transfer from pre-trained video segmenters, the few-shot segmentation model only requires a very small number of fully annotated TOF-MRA images to complete training and achieve robust segmentation performance, significantly reducing the dependence on large-scale annotated data.

[0085] (2) Improved segmentation accuracy and robustness: The preprocessing process and deep learning network architecture, especially the effective integration of residual connections and pre-trained features, help the few-shot segmentation model capture more subtle vascular structures, improving the segmentation accuracy and robustness to different data qualities.

[0086] (3) Convenient and efficient human-computer interaction: The interactive post-processing process implemented based on the 3D Slicer plug-in provides an intuitive visual interface and convenient modification tools. Users can quickly locate and correct segmentation errors. Combined with conditional random field optimization, it can efficiently obtain personalized segmentation results that meet clinical needs.

[0087] (4) Ease of use and integration: The entire process (preprocessing, segmentation, and post-processing) is integrated, especially the post-processing part as a 3D Slicer plug-in, providing an out-of-the-box solution that is convenient for clinicians and researchers to use.

[0088] Based on the same inventive concept, embodiments of the present application also provide a time-of-flight MRI cerebral vessel segmentation system for implementing the aforementioned time-of-flight MRI cerebral vessel segmentation method. The solution provided by this system is similar to the solution described in the aforementioned method. Therefore, the specific limitations in one or more embodiments of the time-of-flight MRI cerebral vessel segmentation system provided below can be found in the limitations of the time-of-flight MRI cerebral vessel segmentation method described above and will not be further elaborated here.

[0089] In an exemplary embodiment, Figure 7 As shown, a time-of-flight magnetic resonance angiography cerebral blood vessel segmentation system is provided, which includes: an image acquisition module 701 , a pre-processing module 702 , a preliminary segmentation module 703 and a post-processing module 704 .

[0090] The image acquisition module 701 is used to acquire the time-of-flight magnetic resonance angiography image to be processed.

[0091] The pre-processing module 702 is used to enhance the blood vessel features in the to-be-processed time-of-flight magnetic resonance angiography image to obtain a pre-processed image.

[0092] The preliminary segmentation module 703 is used to segment cerebral blood vessels based on the pre-processed image using a few-shot segmentation model to obtain a blood vessel probability map. The few-shot segmentation model is obtained by training a 3D U-Net by transferring knowledge from a pre-trained video segmenter.

[0093] The post-processing module 704 is used to post-process the blood vessel probability map and the to-be-processed time-of-flight magnetic resonance vascular image based on the human-computer interaction interface and the conditional random field to obtain a final segmented image.

[0094] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0095] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0096] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0098] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0099] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0100] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0101] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography, characterized in that: The method comprises: acquiring a time-of-flight magnetic resonance angiographic image to be processed; Enhance the vascular features in the to-be-processed time-of-flight magnetic resonance angiography image to obtain a pre-processed image; Based on the preprocessed image, a few-shot segmentation model is used to segment cerebral blood vessels to obtain a blood vessel probability map; the few-shot segmentation model is obtained by training a 3D U-Net by transferring knowledge from a pre-trained video segmenter; The blood vessel probability map and the to-be-processed time-of-flight magnetic resonance vascular image are post-processed based on a human-computer interaction interface and a conditional random field to obtain a final segmented image.

2. The method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography according to claim 1, characterized in that: The pre-processed images include time-of-flight magnetic resonance angiography images after skull removal and contrast-enhanced time-of-flight magnetic resonance angiography images; The vascular features in the to-be-processed time-of-flight magnetic resonance angiography image are enhanced to obtain a pre-processed image, specifically including: registering the to-be-processed time-of-flight magnetic resonance angiography image with the T1-weighted image to obtain a registered time-of-flight magnetic resonance angiography image and a registered T1-weighted image; wherein the T1-weighted image and the to-be-processed time-of-flight magnetic resonance angiography image are images of the same subject; performing skull removal on the registered T1-weighted image to generate a binary brain tissue mask, and applying the binary brain tissue mask to the time-of-flight magnetic resonance angiography image to be processed to obtain a skull-removed time-of-flight magnetic resonance angiography image; The time-of-flight MRI angiogram images after skull removal were histogram-normalized to obtain contrast-enhanced time-of-flight MRI angiogram images.

3. The method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography according to claim 1, characterized in that: Based on the preprocessed image, a few-shot segmentation model is used to segment cerebral blood vessels to obtain a blood vessel probability map, specifically including: Using a pre-trained video segmenter to perform feature encoding on the pre-processed image to obtain multiple features with different resolutions; A pre-trained 3D U-Net is used to segment cerebral blood vessels on the pre-processed image, and multiple features of different resolutions are fused into feature maps of different downsampling stages of the 3D U-Net to obtain a blood vessel probability map.

4. The method for cerebral vessel segmentation in time-of-flight magnetic resonance angiography according to claim 1, characterized in that: The pre-trained video word segmenter is trained in advance using natural scene video streams and features corresponding to different resolutions.

5. The method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography according to claim 1, characterized in that: The 3D U-Net is a residual encoder U-Net.

6. The method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography according to claim 1, characterized in that: Post-processing the vascular probability map and the to-be-processed time-of-flight magnetic resonance vascular image based on the human-computer interaction interface and the conditional random field to obtain a final segmented image specifically includes: generating a 3D reconstruction model of cerebral blood vessels based on the blood vessel probability map; generating a two-dimensional view based on the time-of-flight magnetic resonance angiography image to be processed; For any round of interaction, based on the current 3D reconstruction model of cerebral blood vessels and the current 2D view, the human-computer interaction interface is used to display the current segmentation results and receive the user's correction instructions; Modifying the current cerebral vascular three-dimensional reconstruction model and / or the current two-dimensional view based on the modification instruction to obtain a modified vascular probability map; The vascular probability map is used as the initial input of the unary potential of the conditional random field, and the user's correction instruction is used as the hard constraint or strong prior of the conditional random field. Based on the modified vascular probability map, the conditional random field is used to calculate the global optimal segmentation mask, and the global optimal segmentation mask is fed back to the 3D reconstruction model and 2D view of the cerebral vascular system for the next round of interaction. After multiple rounds of interaction, the final segmented image is obtained.

7. The method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography according to claim 6, characterized in that: The human-computer interaction interface uses a 3D Slicer plug-in to display the current segmentation results and receive user correction instructions.

8. A system for cerebral vessel segmentation in time-of-flight magnetic resonance angiography, applied to the method for cerebral vessel segmentation in time-of-flight magnetic resonance angiography according to any one of claims 1 to 7, characterized in that: The system comprises: An image acquisition module, used for acquiring a time-of-flight magnetic resonance vascular image to be processed; a preprocessing module for enhancing vascular features in a to-be-processed time-of-flight magnetic resonance angiography image to obtain a preprocessed image; A preliminary segmentation module is configured to segment cerebral blood vessels using a few-shot segmentation model based on the preprocessed image to obtain a blood vessel probability map; the few-shot segmentation model is obtained by training a 3D U-Net by transferring knowledge from a pre-trained video segmenter; The post-processing module is used to post-process the vascular probability map and the time-of-flight magnetic resonance vascular image to be processed based on the human-computer interaction interface and the conditional random field to obtain a final segmented image.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for cerebral blood vessel segmentation in time-of-flight magnetic resonance angiography according to any one of claims 1 to 7 is implemented.

Images