Blood vessel image processing and analyzing method and system combining magnetic resonance blood vessel and blood vessel wall imaging

Abstract

The invention provides a blood vessel image processing analysis method and system combining magnetic resonance blood vessel and blood vessel wall imaging, and the system comprises a data obtaining module which obtains a blood vessel enhancement image and a blood vessel wall image; the image registration module is used for registering the blood vessel wall image to the blood vessel enhanced image to obtain a registered blood vessel wall image; the segmentation module is used for obtaining a blood vessel lumen segmentation result and a blood vessel wall segmentation result; the fusion module is used for carrying out weighted fusion on the blood vessel enhanced image and the registered blood vessel wall image to obtain a fused image; the analysis module is used for calculating an intersection region of a blood vessel lumen segmentation result and a blood vessel wall segmentation result; and the three-dimensional visualization and curved surface reconstruction module is used for performing three-dimensional skeleton extraction and curved surface reconstruction on the three-dimensional connected domain of the suspected occluded blood vessel. According to the method, the information of the blood vessel lumen and the blood vessel wall can be obtained at the same time, the defect of a traditional imaging means in evaluating the condition of the blood vessel wall at the far end of the occluded blood vessel is overcome, and the time needed by preoperative evaluation is remarkably shortened through automatic processing and analysis.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and medical image processing technology, and in particular to a vascular image processing and analysis method and system combined with magnetic resonance vascular and vascular wall imaging. Background Art

[0002] With the development of medical imaging technology, artificial intelligence is being used more and more widely in medical image processing. Especially in acute endovascular treatment of stroke, time is of the essence and preoperative evaluation needs to be as concise and efficient as possible. In elective interventional recanalization treatment of non-acute large artery occlusion, detailed preoperative evaluation is particularly important because it can help doctors fully understand the details of the patient's pathological mechanism, thereby improving the safety and efficiency of the operation. Multimodal imaging evaluation plays a key role in this process. It can enable doctors to obtain important information such as cerebral infarction and blood perfusion in the patient's occluded artery blood supply area, the location and range of the occluded lesion and its cause, and whether there is thrombus in the occluded segment of the blood vessel.

[0003] Deficiencies of existing technology: 1. Limited information: In the case of occlusion of the head and neck arteries, conventional preoperative vascular assessment provides limited information, especially in the case of intracranial artery occlusion, when doctors often find it difficult to obtain detailed vascular information of the occluded segment; 2. Single function: Current post-processing software mainly focuses on one aspect of lumen imaging or vessel wall imaging and cannot provide a comprehensive evaluation tool; 3. Manual operation: Existing image processing methods often rely on manual segmentation and analysis, which is not only time-consuming and labor-intensive, but also prone to human errors; 4. Lack of integration: There is a lack of a method or software that can integrate the vascular lumen and vascular wall images for analysis, making the preoperative evaluation not comprehensive and accurate enough.

[0004] Therefore, the prior art has deficiencies and needs further improvement. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a vascular image processing and analysis method and system combining magnetic resonance vascular and vascular wall imaging.

[0006] To achieve the above object, the specific scheme of the present invention is as follows: The present invention provides a vascular image processing and analysis system combined with magnetic resonance vascular and vascular wall imaging, the system comprising: A data acquisition module, used for acquiring a blood vessel enhancement image A1 and a blood vessel wall image B1; An image registration module realizes registration between the vascular wall image and the vascular enhanced image, registers the vascular wall image B1 to the vascular enhanced image A1, and obtains the registered vascular wall image A2; The segmentation module uses a deep learning algorithm to segment the vascular enhanced image A1 and the registered vascular wall image A2 to obtain the vascular lumen segmentation result S1 and the vascular wall segmentation result S2 respectively; A fusion module performs weighted fusion on the vascular enhanced image A1 and the registered vascular wall image A2 to obtain a fused image A3; The analysis module calculates the intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2, and sets the label of the intersection pixel as the blood vessel lumen label; Calculate the union S4 of the intersection area S3 and the blood vessel wall segmentation result S2; Perform 3D connected domain analysis on S4 to identify occluded blood vessel areas, and classify and process them according to the distribution of different connected domain labels, remove false positive segmentation results, retain normal blood vessels, mark stenotic blood vessels, and retain suspected occluded blood vessels; 3D visualization and surface reconstruction module, used to extract 3D skeleton and reconstruct the 3D connected domain of suspected occluded blood vessels; The storage module is used to store raw image data, intermediate processing results and final analysis results.

[0007] Furthermore, the image registration module adopts rigid body registration to improve registration accuracy.

[0008] Furthermore, the segmentation module uses a pre-trained CNN model, and the model adopts data enhancement technology during training to improve segmentation accuracy.

[0009] Furthermore, the fusion module also includes a denoising algorithm for further removing small-area noise points.

[0010] Furthermore, the three-dimensional visualization and surface reconstruction module includes Laplace operator processing of smooth contour points and obtaining more accurate coordinates of the medial axis point by ellipse fitting.

[0011] The present invention also provides a blood vessel image processing and analysis method combined with magnetic resonance blood vessel and blood vessel wall imaging. Based on the above system, the method comprises the following steps: S101, acquiring a blood vessel enhanced image A1 and a blood vessel wall image B1; S102, registering the vascular wall image B1 to the vascular enhanced image A1 by rigid body registration to obtain a registered vascular wall image A2; S103, based on a deep learning method, by channel stitching, the blood vessel enhanced image A1 and the registered blood vessel wall image A2 data are used as network inputs to obtain a blood vessel lumen segmentation result S1 and a blood vessel wall segmentation result S2; S104, calculating the intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2, and setting the label of the intersection pixel as the blood vessel lumen label; S105, calculating the union S4 of the intersection area S3 and the blood vessel wall segmentation result S2, analyzing the three-dimensional connected domain of S4, and identifying the occluded blood vessel area; S106, for the three-dimensional connected domain of the suspected occluded blood vessel, use a three-dimensional skeleton extraction method to calculate the initial central axis X1 of the connected domain, and correct the coordinates X2 of the central axis point; S107, generating a weighted fusion image A3 based on the blood vessel enhanced image A1 and the registered blood vessel wall image A2; S108, based on the corrected coordinates of the central axis point X2, extract the local area image from the weighted fusion image A3 and perform surface reconstruction.

[0012] Furthermore, in step S102, the rigid body registration is implemented using an iterative closest point algorithm.

[0013] Furthermore, in step S103, the deep learning method adopts a convolutional neural network, and uses data enhancement technology during the network training process.

[0014] Furthermore, in step S104, the calculation of the intersection area S3 also includes removing small-area noise points based on morphological operations.

[0015] Furthermore, in step S106, the three-dimensional skeleton extraction method includes traversing the connected domain coordinate points along the central axis, and performing ellipse fitting on the mask contour points within the specified range to obtain the corrected central axis point coordinates X2.

[0016] The technical solution of the present invention has the following beneficial effects: 1. Comprehensive pathological information acquisition: Provided is a method and system for combined magnetic resonance vascular and vascular wall imaging, which can simultaneously obtain detailed information of the vascular lumen and vascular wall, making up for the shortcomings of traditional imaging methods in evaluating the condition of the distal vascular wall of occluded blood vessels.

[0017] 2. Automated segmentation and analysis: Through deep learning methods, automatic segmentation of vascular lumen and vascular wall is achieved, which improves the accuracy and efficiency of segmentation and reduces the need for manual intervention.

[0018] 3. 3D visualization and surface reconstruction: It realizes three-dimensional visualization and curved surface reconstruction of the vascular lumen and wall, allowing doctors to intuitively observe the degree of lesions and vascular morphology of the large blood vessels at the occlusion site and their distal and branch vessels, thereby improving the accuracy of preoperative evaluation.

[0019] 4. Improve diagnostic speed and accuracy: Through automated processing and analysis, the time required for preoperative evaluation is significantly shortened, while the accuracy and reliability of the evaluation are improved, which is conducive to the formulation of more reasonable treatment plans.

[0020] 5. Detailed connected domain analysis: By performing three-dimensional connected domain analysis on the segmentation results, different types of vascular lesions (such as false positive segmentation, normal blood vessels, stenotic blood vessels, and suspected occluded blood vessels) can be distinguished, providing doctors with more detailed pathological information.

[0021] 6. Central axis correction and partial reconstruction: The coordinates of the central axis point are corrected through the three-dimensional skeleton extraction method, and the surface reconstruction of the local area image is performed based on the corrected coordinates, which provides a more sophisticated means of analyzing the occluded area and helps to formulate a more accurate surgical plan.

[0022] 7. Integrated solution: It provides an integrated solution that covers the entire process from data acquisition, image registration, segmentation, analysis to final 3D visualization and surface reconstruction, providing a comprehensive tool for clinical applications.

[0023] 8. Improve surgical safety and efficiency: Through detailed preoperative evaluation, the uncertainty caused by insufficient information during the operation is reduced, thereby improving the safety and efficiency of the operation and reducing surgical risks.

[0024] 9. Improved Patient Management: More precise preoperative assessments can help optimize patient management and ensure that each patient receives the treatment best suited to their specific circumstances, thereby improving overall healthcare quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a schematic diagram of blood vessel lumen and blood vessel wall segmentation based on deep learning of the present invention; Figure 2 is a three-dimensional visualization of the segmentation result of the blood vessel lumen and the blood vessel wall of the present invention; Figure 3 is a surface reconstruction image of the suspected occlusion area of ​​the present invention; Figure 4 is a system block diagram of the present invention; Figure 5 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0026] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments; it can be understood that the specific embodiments described herein are only used to explain the present invention rather than to limit the present invention; it should also be noted that, for ease of description, only the parts related to the present invention rather than all are shown in the accompanying drawings.

[0027] Combination Figure 1-Figure 5 As shown, the present invention provides a vascular image processing and analysis system combined with magnetic resonance vascular and vascular wall imaging, the system comprising: A data acquisition module, used for acquiring a blood vessel enhancement image A1 and a blood vessel wall image B1; An image registration module realizes registration between the vascular wall image and the vascular enhanced image, registers the vascular wall image B1 to the vascular enhanced image A1, and obtains the registered vascular wall image A2; The segmentation module uses a deep learning algorithm to segment the vascular enhanced image A1 and the registered vascular wall image A2 to obtain the vascular lumen segmentation result S1 and the vascular wall segmentation result S2 respectively; A fusion module performs weighted fusion on the vascular enhanced image A1 and the registered vascular wall image A2 to obtain a fused image A3; The analysis module calculates the intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2, and sets the label of the intersection pixel as the blood vessel lumen label; Calculate the union S4 of the intersection area S3 and the blood vessel wall segmentation result S2; Perform 3D connected domain analysis on S4 to identify occluded blood vessel areas, and classify and process them according to the distribution of different connected domain labels, remove false positive segmentation results, retain normal blood vessels, mark stenotic blood vessels, and retain suspected occluded blood vessels; 3D visualization and surface reconstruction module, used to extract 3D skeleton and reconstruct the 3D connected domain of suspected occluded blood vessels; The storage module is used to store raw image data, intermediate processing results and final analysis results.

[0028] The image registration module adopts rigid body registration to improve the registration accuracy.

[0029] The segmentation module uses a pre-trained CNN model, and the model adopts data enhancement technology during training to improve segmentation accuracy.

[0030] The fusion module also includes a denoising algorithm for further removing small-area noise points.

[0031] The three-dimensional visualization and surface reconstruction module includes Laplace operator processing of smooth contour points and obtaining more accurate coordinates of the medial axis point through ellipse fitting.

[0032] The present invention also provides a blood vessel image processing and analysis method combined with magnetic resonance blood vessel and blood vessel wall imaging. Based on the above system, the method comprises the following steps: S101, acquiring a blood vessel enhanced image A1 and a blood vessel wall image B1; S102, registering the vascular wall image B1 to the vascular enhanced image A1 by rigid body registration to obtain a registered vascular wall image A2; S103, based on a deep learning method, by channel stitching, the blood vessel enhanced image A1 and the registered blood vessel wall image A2 data are used as network inputs to obtain a blood vessel lumen segmentation result S1 and a blood vessel wall segmentation result S2; S104, calculating the intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2, and setting the label of the intersection pixel as the blood vessel lumen label; S105, calculating the union S4 of the intersection area S3 and the blood vessel wall segmentation result S2, performing a three-dimensional connected domain analysis on S4, and identifying the occluded blood vessel area; S106, for the three-dimensional connected domain of the suspected occluded blood vessel, use a three-dimensional skeleton extraction method to calculate the initial central axis X1 of the connected domain, and correct the coordinates X2 of the central axis point; S107, generating a weighted fusion image A3 based on the blood vessel enhanced image A1 and the registered blood vessel wall image A2; S108, based on the corrected coordinates of the central axis point X2, extract the local area image from the weighted fusion image A3 and perform surface reconstruction.

[0033] In step S102, rigid body registration is implemented using an iterative closest point algorithm.

[0034] In step S103, the deep learning method adopts a convolutional neural network, and uses data enhancement technology during the network training process.

[0035] In step S104, the calculation of the intersection area S3 also includes removing small-area noise points based on morphological operations.

[0036] In step S106, the three-dimensional skeleton extraction method includes traversing the connected domain coordinate points along the central axis, and performing ellipse fitting on the mask contour points within the specified range to obtain the corrected central axis point coordinates X2.

[0037] The system of the present invention mainly includes the following modules: Data acquisition module: Used to obtain vascular enhanced images (MRA) and vascular wall images (vascular wall MRI).

[0038] Image registration module: The registration between the vascular wall image and the vascular enhanced image is achieved, and the vascular wall image B1 is registered with the vascular enhanced image A1 to obtain the registered vascular wall image A2.

[0039] Segmentation module: Using a deep learning algorithm, the vascular enhanced image A1 and the registered vascular wall image A2 are segmented to obtain the vascular lumen segmentation result S1 and the vascular wall segmentation result S2, respectively.

[0040] Fusion Module: The vascular enhanced image A1 and the registered vascular wall image A2 are weighted fused to obtain a fused image A3.

[0041] Analysis Module: The intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2 is calculated, and the label of the intersection pixel is set as the blood vessel lumen label.

[0042] Perform three-dimensional connected domain analysis to identify occluded blood vessel areas and classify and process them according to the distribution of different connected domain labels (such as removing false positive segmentation results, retaining normal blood vessels, marking narrowed blood vessels, and retaining suspected occluded blood vessels).

[0043] 3D visualization and surface reconstruction module: The 3D skeleton of the 3D connected domain of the suspected occluded blood vessel is extracted, the central axis is calculated and the coordinates of the central axis point are corrected.

[0044] Extract local area images and perform surface reconstruction to achieve detailed display of occluded areas.

[0045] User Interface Module (optional): Provides a user interactive interface that allows doctors to view and operate segmentation results and reconstructed images, facilitating preoperative evaluation.

[0046] Storage Module: Stores raw image data, intermediate processing results, and final analysis results.

[0047] These modules work together to achieve accurate reconstruction and analysis of the lumen and walls of the head and neck blood vessels, thereby assisting doctors in analyzing the degree of lesions and vascular morphology of the large blood vessels at the occlusion site and their distal and branch vessels, so as to complete preoperative assessment decisions more quickly and accurately.

[0048] The method of the present invention comprises the following steps: Data acquisition: Obtain vascular enhanced image (MRA) A1 and vascular wall image B1.

[0049] Image registration: The vascular wall image B1 is registered to the vascular enhanced image A1 by rigid body registration to obtain the registered vascular wall image A2.

[0050] Vascular lumen and vascular wall segmentation based on deep learning: Using deep learning methods, through channel splicing, the vascular enhanced image A1 and the registered vascular wall image A2 data are used as network input to obtain the vascular lumen segmentation result S1 and the vascular wall segmentation result S2.

[0051] Segmentation result processing: The intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2 is calculated, and the label of the intersection pixel is set as the blood vessel lumen label.

[0052] The union S4 of the intersection area S3 and the blood vessel wall segmentation result S2 is calculated, and a three-dimensional connected domain analysis is performed on S4 to identify the occluded blood vessel area.

[0053] Connected domain analysis and classification: Perform three-dimensional connected domain analysis on S4, count the distribution of segmentation labels in each connected domain, and classify them: Remove false positive segmentation results (connected domains with only vessel wall segmentation labels).

[0054] Normal blood vessels (connected domains with only the segmentation labels of the vascular lumen) are retained.

[0055] Mark stenotic vessels (both vessel lumen and vessel wall segmentation labels exist, but the vessel wall segmentation volume is less than 20% of the connected domain volume).

[0056] Suspected occluded vessels are retained (both vessel lumen and vessel wall segmentation labels exist, and the vessel wall segmentation volume accounts for more than 20% of the connected domain volume).

[0057] 3D visualization and surface reconstruction: For the three-dimensional connected domain of the suspected occluded blood vessel, the three-dimensional skeleton extraction method is used to calculate the initial central axis X1 of the connected domain, and the coordinates of the central axis point X2 are corrected.

[0058] Based on the corrected coordinates of the central axis point X2, the corresponding local area image is extracted and the surface reconstruction is performed.

[0059] These steps together constitute a vascular image processing and analysis method for combined magnetic resonance vascular and vascular wall imaging, which aims to assist doctors in analyzing the degree of lesions and vascular morphology of large blood vessels at the occluded site and their distal blood vessels and branch vessels, so as to complete preoperative evaluation decisions more quickly and accurately.

[0060] How it works 1. Data Acquisition Obtaining magnetic resonance imaging (MRA): First, obtain the patient's magnetic resonance imaging (MRA), which is a commonly used vascular imaging technology that can clearly show the morphology and structure of blood vessels.

[0061] Obtain images of the blood vessel wall (vascular wall MRI): Images of the blood vessel wall (vascular wall MRI) are obtained at the same time. This image can provide detailed information about the blood vessel wall, including changes in the intima, media, and adventitia.

[0062] 2. Image Registration Rigid body registration: The vascular wall image (B1) is registered to the vascular enhanced image (A1) by rigid body registration to obtain the registered vascular wall image (A2). This process is to ensure the spatial correspondence between the two images for subsequent processing.

[0063] 3. Segmentation based on deep learning Deep learning network input: The vascular enhanced image (A1) and the registered vascular wall image (A2) are input into the pre-trained deep learning network through channel splicing.

[0064] Segmentation result output: The network outputs the vascular lumen segmentation result (S1) and the vascular wall segmentation result (S2). The segmentation results distinguish different tissue structures by labels, where the label of the vascular lumen is 1 and the label of the vascular wall is 2.

[0065] 4. Segmentation result processing Intersection calculation: Calculate the intersection area (S3) of the vascular lumen segmentation result (S1) and the vascular wall segmentation result (S2), and set the label of the intersection pixel as the vascular lumen label (1).

[0066] Union calculation and three-dimensional connected domain analysis: Calculate the union (S4) of the intersection area (S3) and the vascular wall segmentation result (S2), perform three-dimensional connected domain analysis on S4, and count the distribution of segmentation labels in each connected domain.

[0067] 5. Connected domain analysis and classification Classification processing: False positive segmentation results: When only the vessel wall segmentation label exists in the connected domain, it is considered as an erroneous segmentation result and removed.

[0068] Normal blood vessels: When only the vascular lumen segmentation label exists in the connected domain, it is regarded as a normal blood vessel.

[0069] Stenotic vessels: When the connected domain has both vascular lumen and vascular wall segmentation labels, but the vascular wall segmentation volume is less than 20% of the connected domain volume, it is considered as a stenotic vessel and retained.

[0070] Suspected occluded vessel: When the connected domain has both vascular lumen and vascular wall segmentation labels, and the vascular wall segmentation volume accounts for more than 20% of the connected domain volume, it is considered as a suspected occluded vessel.

[0071] 6. 3D visualization and surface reconstruction Three-dimensional skeleton extraction and central axis correction: For the three-dimensional connected domain of suspected occluded blood vessels, the three-dimensional skeleton extraction method is used to calculate the initial central axis (X1) of the connected domain, and by traversing the coordinate points of the central axis of the connected domain, an ellipse is fitted to the mask contour points within the specified range to obtain the corrected coordinates of the central axis point (X2).

[0072] Fusion image generation: The vascular enhancement image (A1) and the registered vascular wall image (A2) are weighted fused to obtain a fused image (A3).

[0073] Local area image extraction and surface reconstruction: Based on the fused image (A3) and the corrected mid-axis point coordinates (X2), the corresponding local area image is extracted to achieve surface reconstruction of the suspected occluded area.

[0074] Through the above steps, the method and system of the present invention can achieve accurate reconstruction and analysis of the lumen and wall of the head and neck blood vessels, and assist doctors in completing preoperative evaluation decisions more quickly and accurately.

[0075] Embodiment 1: Acute endovascular treatment of stroke is a battle of encounters, time is brain, and preoperative evaluation should be as simple as possible. Elective interventional recanalization of non-acute large artery occlusion is a positional battle, and the operation is complex and difficult. Detailed preoperative evaluation can provide key information for safe and efficient surgery, especially advanced multimodal imaging evaluation, which enables the surgeon to fully understand the details of the patient's pathological mechanism.

[0076] Compared with head and neck artery stenosis, cases of head and neck artery occlusion are much more complicated. When the lumen of the head and neck arteries is in an occluded state, especially when the intracranial artery is occluded, the information obtained from conventional preoperative vascular assessment is limited. Interventional doctors usually face a "dark room" without windows. During the operation, they rely on operation feedback and angiography to continuously and deeply understand the changes in the vascular conditions of the occluded segment of the large artery and adjust the treatment strategy. Uncertainty is inevitable. Therefore, in order to safely and efficiently implement interventional recanalization treatment for non-acute occlusion of the head and neck arteries, it is necessary to understand its pathological mechanism as much as possible before surgery, including cerebral infarction and blood perfusion in the blood supply area of ​​the occluded artery, the location, range and cause of the large artery occlusion lesions, whether the intracranial and extracranial branches of the occluded artery are combined with stenosis-occlusion lesions, and whether there is thrombosis in the lumen of the occluded artery. The cerebral infarction and blood perfusion in the blood supply area of ​​the occluded artery can be decided based on the method of volume mismatch between the cerebral infarction area and the low perfusion area by cerebral perfusion examination. Due to the complete occlusion of the middle cerebral artery, traditional vascular assessment methods such as CTA or MRA cannot accurately determine the condition of the distal vascular wall and its pathological characteristics, while MRI of the vessel wall can provide information of potential value. Magnetic resonance vascular wall imaging can display the changes in the endothelium, subendothelium structure and vessel wall through images, helping doctors observe the condition of the blood vessels.

[0077] Currently, there are only post-processing and analysis software for lumen imaging or vascular wall imaging. The former is mainly used for three-dimensional reconstruction and visualization of blood vessels, and identification of lesions such as vascular stenosis and aneurysms, while the latter is mainly focused on the evaluation of the pathological properties and stability of plaques. There is a lack of a method and software that can simultaneously analyze and process the lumen and vascular wall images of blood vessels, realize three-dimensional and curved reconstruction and visualization of the lumen and vascular wall of blood vessels, and assist doctors in analyzing the degree of lesions and vascular morphology of large blood vessels and their distal and branch vessels at the occluded site, so as to complete preoperative evaluation decisions more quickly and accurately.

[0078] It provides a fully automatic, fast and accurate reconstruction of the lumen and wall volume of the head and neck blood vessels, and combines the morphology of the blood vessel wall to perform surface reconstruction of the suspected occluded blood vessel area, assisting doctors in analyzing the degree of lesions and vascular morphology of the large blood vessels and their distal and branch vessels at the occluded site, so as to complete preoperative assessment and decision-making more quickly and accurately: (1) Taking the vascular enhancement (MRA) as the main image A1, the vascular wall image B1 is registered to the vascular enhancement image through rigid body registration to obtain image A2.

[0079] (2) If Figure 1As shown in the figure, based on the deep learning method, the three-dimensional vascular enhancement and vascular wall image sequence (A1, A2) data is used as the network input through channel splicing, and the region of interest segmentation result of the corresponding data is used as the network output target (the vascular lumen segmentation result label is 1, and the vascular wall segmentation result label is 2). The segmentation network is repeatedly trained until the network segmentation result is consistent with the region of interest.

[0080] (3) Calculate the intersection area S3 = S1∩S2 of the vascular lumen segmentation result S1 and the vascular wall segmentation result S2, and set the segmentation label of the intersection voxel to 1, that is, the intersection voxel is regarded as a lumen voxel. Calculate the union S4 = S3∪S2 of the intersection area and the vascular wall segmentation result, and output a three-dimensional visualization diagram of the intersection, as shown in Figure 2 Then, a three-dimensional connected domain analysis is performed on S4, and the distribution of segmentation labels in each connected domain is counted and divided into the following four cases: ① False positive segmentation results: When only the vascular wall segmentation label exists in the connected domain, the connected domain is regarded as an erroneous segmentation result and removed.

[0081] ② When only the vascular lumen segmentation label exists in the connected domain, the connected domain is regarded as a normal blood vessel. ③ When the connected domain has both vascular lumen and vascular wall segmentation labels, but the vascular wall segmentation volume is less than 20% of the volume of the connected domain, the connected domain is regarded as a stenotic vessel.

[0082] ④ When the connected domain has both vascular lumen and vascular wall segmentation labels, and the vascular wall segmentation volume accounts for more than 20% of the connected domain volume, the connected domain is considered as a suspected occluded vessel retention.

[0083] (4) For the three-dimensional connected domain suspected of having occluded blood vessels, the three-dimensional skeleton extraction method is used to calculate the initial central axis X1 of the connected domain. The coordinate points of the central axis of the connected domain are traversed along the z-axis direction from the bottom to the brain. With the central axis coordinate point area as the center, the connected domain mask submask_D within the range of 80x80 pixels is extracted, the contour points of the mask are extracted, and an ellipse is fitted to fit the center point coordinates of the ellipse to obtain the corrected central axis point coordinates X2. The enhanced blood vessel image A1 and the aligned blood vessel wall image A2 are weighted fused to obtain the fused image A3 = 0.7*A1 + 0.3*A2. Based on the fused image A3 and the corrected central axis point coordinates X2, the corresponding local area image is extracted to achieve surface reconstruction of the corresponding suspected occluded area, such as Figure 3 shown.

[0084] Example 2: Segmentation and analysis of vascular lumen and vascular wall Step 1: Get the data Obtaining vascular enhanced imaging (MRA) A1: A 3T MRI scanner was used to acquire vascular enhancement images A1 of the patient's head, ensuring that the image covered the entire head and neck vascular area.

[0085] File format: DICOM format, resolution: 0.5mm × 0.5mm × 2mm.

[0086] Obtaining blood vessel wall images (vascular wall MRI) B1: A high-resolution vessel wall MRI scanner is used to acquire the vessel wall image B1 to ensure that the image clearly displays the vessel wall structure.

[0087] File format: NIFTI format, resolution: 0.3mm × 0.3mm × 0.3mm.

[0088] Step 2: Image Registration Rigid body registration: The vascular wall image B1 is registered to the vascular enhanced image A1 using a rigid body registration algorithm based on mutual information (MI) to obtain the registered vascular wall image A2.

[0089] Ensure that the registration error is less than 1mm.

[0090] Step 3: Deep Learning-Based Segmentation Deep learning network input: Using the trained Convolutional Neural Network (CNN), the vascular enhanced image A1 and the registered vascular wall image A2 are used as network input through channel splicing.

[0091] Input size: 256×256×16 (XYZ dimensions).

[0092] Segmentation result output: The network outputs the vascular lumen segmentation result S1 (labeled 1) and the vascular wall segmentation result S2 (labeled 2).

[0093] The segmentation accuracy reaches over 95%.

[0094] Step 4: Segmentation result processing Intersection calculation: The intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2 is calculated, and the label of the intersection pixel is set as the blood vessel lumen label (1).

[0095] The volume of the intersection area S3 accounts for about 5% of the total segmentation area.

[0096] Union calculation and three-dimensional connected domain analysis: The union S4 of the intersection area S3 and the blood vessel wall segmentation result S2 is calculated, and the occluded blood vessel area is identified based on the three-dimensional connected domain analysis.

[0097] Statistics on the distribution of segmentation labels in each connected domain and classify them: Remove false positive segmentation results (connected domains with only vessel wall segmentation labels).

[0098] Normal blood vessels (connected domains with only the segmentation labels of the vascular lumen) are retained.

[0099] Mark stenotic vessels (both vessel lumen and vessel wall segmentation labels exist, but the vessel wall segmentation volume is less than 20% of the connected domain volume).

[0100] Suspected occluded vessels are retained (both vessel lumen and vessel wall segmentation labels exist, and the vessel wall segmentation volume accounts for more than 20% of the connected domain volume).

[0101] Step 5: 3D visualization and surface reconstruction 3D skeleton extraction and central axis correction: For the three-dimensional connected domain of the suspected occluded blood vessel, the initial central axis X1 of the connected domain is calculated using the three-dimensional skeleton extraction method, and the coordinates of the central axis point X2 are corrected by ellipse fitting.

[0102] The center axis correction error is controlled within 0.5mm.

[0103] Fusion image generation: The vascular enhanced image A1 and the registered vascular wall image A2 are weighted fused with a weight ratio of 0.7:A1 +0.3:A2 to obtain a fused image A3.

[0104] Local area image extraction and surface reconstruction: Based on the fused image A3 and the corrected coordinates of the central axis point X2, the local area image within the range of 80x80 pixels is extracted and the surface reconstruction is performed.

[0105] The surface reconstruction accuracy reaches sub-millimeter level.

[0106] Example 2: System Integration and User Interface Step 1: Get the data Obtaining vascular enhanced imaging (MRA) A1: A 1.5T MRI scanner is used to acquire a vascular enhanced image A1 of the patient's head.

[0107] File format: DICOM format, resolution: 1mm × 1mm × 3mm.

[0108] Obtaining blood vessel wall images (vascular wall MRI) B1: A high-resolution vessel wall MRI scanner is used to acquire the vessel wall image B1.

[0109] File format: NIFTI format, resolution: 0.5mm × 0.5mm × 0.5mm.

[0110] Step 2: Image Registration Rigid body registration: The vascular wall image B1 is registered to the vascular enhanced image A1 using a rigid body registration algorithm based on mutual information (MI) to obtain the registered vascular wall image A2.

[0111] Ensure that the registration error is less than 2mm.

[0112] Step 3: Deep Learning-Based Segmentation Deep learning network input: Using the trained deep residual network (ResNet), the vascular enhanced image A1 and the registered vascular wall image A2 are used as network input through channel splicing.

[0113] Input size: 512×512×16 (XYZ dimensions).

[0114] Segmentation result output: The network outputs the vascular lumen segmentation result S1 (labeled 1) and the vascular wall segmentation result S2 (labeled 2).

[0115] The segmentation accuracy reaches over 90%.

[0116] Step 4: Segmentation result processing Intersection calculation: The intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2 is calculated, and the label of the intersection pixel is set as the blood vessel lumen label (1).

[0117] The volume of the intersection area S3 accounts for about 10% of the total segmentation area.

[0118] Union calculation and three-dimensional connected domain analysis: The union S4 of the intersection area S3 and the blood vessel wall segmentation result S2 is calculated, and the occluded blood vessel area is identified based on the three-dimensional connected domain analysis.

[0119] Statistics on the distribution of segmentation labels in each connected domain and classify them: Remove false positive segmentation results (connected domains with only vessel wall segmentation labels).

[0120] Normal blood vessels (connected domains with only the segmentation labels of the vascular lumen) are retained.

[0121] Mark stenotic vessels (both vessel lumen and vessel wall segmentation labels exist, but the vessel wall segmentation volume is less than 20% of the connected domain volume).

[0122] Suspected occluded vessels are retained (both vessel lumen and vessel wall segmentation labels exist, and the vessel wall segmentation volume accounts for more than 20% of the connected domain volume).

[0123] Step 5: 3D visualization and surface reconstruction 3D skeleton extraction and central axis correction: For the three-dimensional connected domain of the suspected occluded blood vessel, the initial central axis X1 of the connected domain is calculated using the three-dimensional skeleton extraction method, and the coordinates of the central axis point X2 are corrected by ellipse fitting.

[0124] The center axis correction error is controlled within 1mm.

[0125] Fusion image generation: The vascular enhanced image A1 and the registered vascular wall image A2 are weighted fused with a weight ratio of 0.6:A1 +0.4:A2 to obtain a fused image A3.

[0126] Local area image extraction and surface reconstruction: Based on the fused image A3 and the corrected coordinates of the central axis point X2, the local area image within the range of 80x80 pixels is extracted and the surface reconstruction is performed.

[0127] The surface reconstruction accuracy reaches sub-millimeter level.

[0128] user interface: A user interface module is integrated into the system to display the processing results and allow doctors to interact, view and analyze the segmentation results and reconstructed images.

[0129] The user interface supports zooming, rotating, and panning functions, making it easy for doctors to observe vascular structures from different angles.

[0130] The above embodiments demonstrate the specific application of the present invention under different equipment configurations, ensuring the versatility and flexibility of the method. Through these detailed steps, doctors can obtain comprehensive and accurate vascular information, thereby better performing preoperative evaluation and decision-making.

[0131] The above description is only a preferred embodiment of the present invention, and does not limit the scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the protection scope of the present invention.

Claims

1. A vascular image processing and analysis system combining magnetic resonance vascular and vascular wall imaging, characterized in that: The system includes: A data acquisition module, used for acquiring a blood vessel enhancement image A1 and a blood vessel wall image B1; An image registration module realizes registration between the vascular wall image and the vascular enhanced image, registers the vascular wall image B1 to the vascular enhanced image A1, and obtains the registered vascular wall image A2; The segmentation module uses a deep learning algorithm to segment the vascular enhanced image A1 and the registered vascular wall image A2 to obtain the vascular lumen segmentation result S1 and the vascular wall segmentation result S2 respectively; A fusion module performs weighted fusion on the vascular enhanced image A1 and the registered vascular wall image A2 to obtain a fused image A3; The analysis module calculates the intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2, and sets the label of the intersection pixel as the blood vessel lumen label; Calculate the union S4 of the intersection area S3 and the blood vessel wall segmentation result S2; Perform 3D connected domain analysis on S4 to identify occluded blood vessel areas, and classify and process them according to the distribution of different connected domain labels, remove false positive segmentation results, retain normal blood vessels, mark stenotic blood vessels, and retain suspected occluded blood vessels; 3D visualization and surface reconstruction module, used to extract 3D skeleton and reconstruct the 3D connected domain of suspected occluded blood vessels; The storage module is used to store raw image data, intermediate processing results and final analysis results.

2. The system according to claim 1, characterized in that The image registration module adopts rigid body registration to improve the registration accuracy.

3. The system according to claim 1, characterized in that The segmentation module uses a pre-trained CNN model, and the model adopts data enhancement technology during training to improve segmentation accuracy.

4. The system according to claim 1, characterized in that The fusion module also includes a denoising algorithm for further removing small-area noise points.

5. The system according to claim 1, wherein: The three-dimensional visualization and surface reconstruction module includes Laplace operator processing of smooth contour points and obtaining more accurate coordinates of the medial axis point through ellipse fitting.

6. A method for processing and analyzing vascular images in combination with magnetic resonance vascular and vascular wall imaging, based on the system according to any one of claims 1 to 5, characterized in that: The method comprises the following steps: S101, acquiring a blood vessel enhanced image A1 and a blood vessel wall image B1; S102, registering the vascular wall image B1 to the vascular enhanced image A1 by rigid body registration to obtain a registered vascular wall image A2; S103, based on a deep learning method, by channel stitching, the blood vessel enhanced image A1 and the registered blood vessel wall image A2 data are used as network inputs to obtain a blood vessel lumen segmentation result S1 and a blood vessel wall segmentation result S2; S104, calculating the intersection area S3 of the blood vessel lumen segmentation result S1 and the blood vessel wall segmentation result S2, and setting the label of the intersection pixel as the blood vessel lumen label; S105, calculating the union S4 of the intersection area S3 and the blood vessel wall segmentation result S2, performing a three-dimensional connected domain analysis on S4, and identifying the occluded blood vessel area; S106, for the three-dimensional connected domain of the suspected occluded blood vessel, use a three-dimensional skeleton extraction method to calculate the initial central axis X1 of the connected domain, and correct the coordinates X2 of the central axis point; S107, generating a weighted fusion image A3 based on the blood vessel enhanced image A1 and the registered blood vessel wall image A2; S108, based on the corrected coordinates of the central axis point X2, extract the local area image from the weighted fusion image A3 and perform surface reconstruction.

7. The method according to claim 6, characterized in that In step S102, rigid body registration is implemented using an iterative closest point algorithm.

8. The method according to claim 6, characterized in that In step S103, the deep learning method adopts a convolutional neural network, and uses data enhancement technology during the network training process.

9. The method according to claim 6, characterized in that In step S104, the calculation of the intersection area S3 also includes removing small-area noise points based on morphological operations.

10. The method according to claim 6, characterized in that In step S106, the three-dimensional skeleton extraction method includes traversing the connected domain coordinate points along the central axis, and performing ellipse fitting on the mask contour points within the specified range to obtain the corrected central axis point coordinates X2.

Images