Method for constructing CT-MRI personalized three-dimensional heart model based on multi-modal imaging

Abstract

The invention provides a method for constructing a CT-MRI (Computed Tomography-Magnetic Resonance Imaging) personalized three-dimensional heart model based on multi-modal imaging, which comprises the following steps of: firstly, carrying out myocardial and infarction region segmentation on a CMR-LGE image, and carrying out myocardial and intramyocardial adipose tissue segmentation on a CE-CT image to generate a corresponding Label image; then, a high-precision three-dimensional volume mesh model of the corresponding ventricle is constructed based on the segmentation results of the two types of images; and then, mapping the corresponding Label image with the tissue type to a corresponding finite element model according to the coordinates of the grid, and obtaining a corresponding personalized ventricular model with the tissue type. An interpolation method based on a general ventricular coordinate system is adopted, data of an MR ventricular model with infarct tissue attributes is transferred to a CT ventricular model, and therefore a mixed CT-MRI ventricular model is constructed. And finally, storing the constructed CT-MRI ventricular model data with the integrated tissue attributes as a standard geometric and topological file format.

Description

Technical Field

[0001] This invention relates to the fields of digital healthcare and image processing technology, specifically to a method for constructing a personalized 3D heart model based on multimodal imaging for CT-MRI. Background Technology

[0002] Personalized heart models can be used for non-invasive assessment of patients' risk of ventricular tachycardia through electrophysiological simulation and to help determine the location of lesions, enabling precise clinical diagnosis, treatment planning, and efficacy evaluation.

[0003] Clinically, various medical imaging techniques are used to acquire cardiac information. CMR-LGE sequences can provide detailed information on myocardial tissue characteristics, such as the region, extent, and transmural nature of myocardial infarction, but their spatial resolution is relatively limited, making it difficult to accurately depict the fine anatomical structures of the heart. CT, on the other hand, has advantages in fine cardiac anatomy and adipose tissue regions due to its high spatial resolution. During cardiac repair after myocardial infarction, the formation of fatty infiltration and infarcted tissue significantly affects the electrophysiological characteristics of the heart. Studies have shown that intramyocardial adipose tissue and myocardial fibrosis coexist in post-myocardial infarction patients, indicating that intramyocardial adipose infiltration is more important than fibrotic scarring in causing arrhythmias. However, traditional imaging techniques struggle to accurately capture the distribution of both tissues simultaneously. Therefore, establishing accurate infarct and fatty infiltration regions in a patient's personalized cardiac model is crucial for the accuracy of subsequent electrophysiological simulations. Summary of the Invention

[0004] This invention proposes a method for constructing personalized 3D cardiac models based on multimodal imaging for CT-MRI. The method first segments the myocardium and infarct region in CMR-LGE images and segments the myocardium and intramyocardial adipose tissue in CE-CT images, generating corresponding label images. Next, a high-precision 3D volumetric mesh model of the corresponding ventricle is constructed based on the segmentation results of the CMR-LGE and CE-CT images. Subsequently, the corresponding label images with tissue types are mapped to the corresponding finite element models according to the mesh coordinates, resulting in personalized ventricular models with tissue types. Further, to integrate multimodal information and generate a unified CT-MRI ventricular model, an interpolation method based on a universal ventricular coordinate system is used to transfer the data of the MRI ventricular model with infarct tissue attributes to the CT ventricular model, thereby constructing a hybrid CT-MRI ventricular model. Finally, the constructed CT-MRI ventricular model data with integrated tissue attributes is saved in a standard geometric and topological file format for subsequent electrophysiological simulations.

[0005] The technical solution of the present invention is as follows:

[0006] A method for constructing a personalized 3D cardiac model based on CT-MRI using multimodal imaging, comprising the following steps:

[0007] S1, Data Preprocessing;

[0008] S11, Image Preprocessing

[0009] Deep learning-based methods were used to segment the acquired cardiac CMR-LGE and CE-CT images to obtain images of the patient's ventricular myocardium; the CE-CT images were resampled from short-axis to long-axis cardiac images.

[0010] S12, Ventricular cavity interpolation reconstruction

[0011] Based on the segmentation results of the CMR-LGE images, constraint points located at the myocardial boundaries were selected in the acquired myocardial region. Variational implicit function interpolation was used to interpolate these discrete points, generating fitted surfaces for the left ventricular endocardium, right ventricular endocardium, and aorta, respectively. The left and right ventricular endocardium surfaces were combined to construct a biventricular cavity model, which was saved as a high-resolution MRI-Label image. The aorta and left ventricular endocardium surfaces were combined to construct an aorta-left ventricular endocardium model, which was saved as an MRI-Aorta image of the same resolution. The CE-CT images themselves have high resolution, but for the convenience of subsequent mapping, they were interpolated and saved as high-resolution CT-Label images. The aorta was saved as a CT-Aorta image of the same resolution.

[0012] S13, Reconstructing infarcted tissue

[0013] The myocardial infarction region was extracted and reconstructed from CMR-LGE images. First, a Gaussian mixture model (GMM) based method or threshold segmentation was used to analyze the myocardial region in the CMR-LGE images to obtain preliminary segmentation results of the myocardial infarction region. Then, the segmented infarct region was interpolated and reconstructed using a probabilistic logarithmic interpolation algorithm (LogOdds) to generate high-resolution three-dimensional infarct tissue image data. Finally, the interpolated infarct tissue image data was merged with the aforementioned MRI-Label image to generate an MRI-Scar image.

[0014] S14, Reconstructing adipose tissue

[0015] Segmentation and extraction of intramyocardial adipose tissue were performed on CE-CT images. First, a threshold segmentation method was used to analyze the myocardial region, defining low-attenuation areas between -180 HU and -5 HU as adipose tissue. This range was further subdivided into two intervals: -180 HU to -50 HU defined as dense adipose tissue, and -50 HU to -5 HU defined as a mixed adipose-myocardial region. Subsequently, to eliminate noise artifacts in the CE-CT images that might be misidentified as fat, volumetric filtering was performed on the segmentation results, removing areas with a volume less than or equal to 1 mm³. After processing, denoised dense adipose tissue images and mixed adipose-myocardial tissue images were obtained, and these two images were merged with the CT-Label image to generate a CT-Fat image.

[0016] S15. Establish a finite element mesh model of the heart.

[0017] Based on the aforementioned MRI-Label and CT-Label images, a high-precision cardiac cavity was constructed. The construction process included surface mesh generation and smoothing optimization, followed by volume mesh generation, thereby generating MRI and CT ventricular models. Then, based on the aforementioned MRI-Aorta and CT-Aorta images, the corresponding aorta was constructed. Rigid registration was then performed on the MRI and CT ventricular models using the aorta, and excess portions of the basal portion of the CT ventricular model were removed. Finally, the generated model data was saved as MRI.pts and CT.pts files recording the point coordinates of the ventricular model, MRI.elem and CT.elem files recording the vertex indices of the tetrahedral mesh, and MRI.vtu and CT.vtu files recording the volume mesh.

[0018] S16. Extract the heart surface mesh.

[0019] From the volumetric mesh of MRI and CT ventricular models, the corresponding four boundary surfaces—Base, Epicardial surface, Left Ventricular Endocardial surface (LV), and Right Ventricular Endocardial surface (RV)—are extracted. The extracted data are then saved as MRI-Base.ply, MRI-Epi.ply, MRI-LV.ply, MRI-RV.ply, CT-Base.ply, CT-Epi.ply, CT-LV.ply, and CT-RV.ply files, which record the coordinates of the grid points on the cardiac surfaces.

[0020] S2, mapping of infarcted and fatty regions;

[0021] S21, Infarct Region Mapping

[0022] The infarcted region in the high-resolution MRI-SCAR image is mapped onto an MRI ventricular model. The centroid of each tetrahedral grid in the model is calculated, and the centroid coordinates are divided by the image resolution and rounded down to obtain the corresponding pixel in the MRI-SCAR image for each grid. The MRI-SCAR value of a pixel is 1 for normal myocardial tissue, 2 for hemi-infarcted myocardial tissue, and 3 for complete infarcted tissue. If the pixel is not a myocardial region, non-zero pixels in the first to third order neighborhoods are searched sequentially, and their MRI-SCAR values ​​are used as the type of the grid.

[0023] S22, Fat Region Mapping

[0024] The fatty regions in the high-resolution CT-Fat image are mapped onto the CT ventricular model. The centroid of each tetrahedral grid in the model is calculated, and the centroid coordinates are divided by the image resolution and rounded down to obtain the corresponding pixel in the CT-Fat image for each grid. A pixel with a CT-Fat value of 1 represents normal myocardial tissue, 2 represents mixed fat-myocardial tissue, and 3 represents dense adipose tissue. If the pixel is not a myocardial region, non-zero pixels in the first to third order neighborhoods are searched sequentially, and their CT-Fat values ​​are used as the type of the grid.

[0025] S3. Establish universal ventricular coordinates

[0026] Biventricular coordinates were established for both MRI and CT ventricular models. The MRI.vtu and CT.vtu files, recording the volumetric mesh, and the corresponding cardiac surface mesh files (including four surface types: Base, Epi, LV, and RV, in .ply format) were used. Next, transventricular coordinates were established by calculating and binarizing the transventricular Laplace solution, followed by an initial mesh reconstruction to extract the surface and curves of the interventricular septum. Then, the transmural trajectory distance was calculated and normalized based on the inner and outer boundaries and the interventricular septum to obtain the transmural coordinates. Simultaneously, the anatomical axis and apex of the heart were determined. Further, the ridge surface was extracted by calculating the Laplace solution between the epicardium and the interventricular septum and performing a second mesh reconstruction. The rotational trajectory distance was calculated using the gradient of the transmural coordinates and the apical-base Laplace solution, and normalized and adjusted to obtain the rotational coordinates. Finally, the normalized distance along the contour lines of the transmural and rotational coordinates was calculated, and the apical-base coordinates were obtained using Laplace extrapolation. Generate MRI-UVC.vtu and CT-UVC.vtu files respectively, as well as store the matrix that associates the established coordinates with the original grid data structure.

[0027] S4. Establish a CT-MRI ventricular model;

[0028] S41. Convert cell data to point data

[0029] The data of infarct cells stored in the MRI.vtu file is converted into data defined on the vertices of the dataset. For each vertex in the dataset, the vertex data value corresponding to that vertex is determined by averaging the cell data values ​​of the cells containing that vertex.

[0030] S42, Model Fusion

[0031] A hybrid CT-MRI ventricular model is constructed by mapping the infarct distribution in an MRI ventricular model to the geometry of a CT ventricular model. Using a linear interpolation method based on ventricular coordinates, for each node in the CT ventricular model mesh, a set of candidate MRI ventricular model mesh elements corresponding to the CT ventricular model node coordinates is determined in the ventricular coordinate space. The centroid coordinates corresponding to the CT ventricular model node coordinates are then calculated within these candidate elements. Subsequently, for each node data of the CT ventricular model, based on the calculated centroid coordinates, the MRI ventricular model mesh element that minimizes the maximum absolute deviation of the centroid coordinates is selected as the optimal element from the candidate element set. Using the node index of the selected optimal element and its corresponding calculated centroid coordinates, a mapping matrix is ​​constructed that linearly interpolates data from the MRI ventricular model mesh nodes to the CT mesh nodes. The point data values ​​of the infarct units on the CT ventricular model mesh are obtained through the mapping matrix, generating the CT-MRI.vtu file.

[0032] S43. Point data is converted into cell data.

[0033] The point data of infarct units stored in the CT-MRI.vtu file are converted into data defined at the units in the dataset. For each unit in the dataset, a set of points constituting the unit is determined, and the point data values ​​in the set are subject to majority voting, where in the event of a tie, the smaller value is selected to compute the unit data value corresponding to the unit.

[0034] S44, Assigning Labels

[0035] In the hybrid CT-MRI ventricular model, the labels are as follows: 1 represents normal myocardial tissue, 2 represents a mixture of dense fatty infiltration and scar tissue, 3 represents fibrofatty infiltration myocardial tissue, 4 represents completely infarcted tissue, 5 represents dense fatty tissue, 6 represents fat-free hemi-infarcted myocardial tissue, and 7 represents non-fibrotic fatty-myocardial mixed tissue. Finally, the CT-MRI.pts file containing the point coordinates of this ventricular model and the CT-MRI.elem file containing the tetrahedral mesh vertex index are saved.

[0036] The beneficial effects of this invention are:

[0037] This invention integrates a personalized 3D heart model construction method based on CT-MRI, which effectively integrates information from CMR-LGE and CE-CT modal images. Through an efficient and accurate attribute mapping process, it reliably and accurately assigns voxel-level or pixel-level tissue classification results from the images to the units or vertices of the 3D mesh model, while ensuring the accuracy of the mapping and the integrity of the topological structure, thus providing support for subsequent electrophysiological simulations. Attached Figure Description

[0038] Figure 1 This is a flowchart of a method for constructing a personalized 3D cardiac model based on multimodal imaging using CT-MRI.

[0039] Figure 2 This is a schematic diagram of data preprocessing; (a) is the MRI-Label image obtained by CMR-LGE image segmentation and interpolation; (b) is the CT-Label image obtained by CE-CT image segmentation; (c) is the MRI-Label image after infarct region segmentation, interpolation, and merging; (d) is the CT-Label image after fat region segmentation, denoising, and merging; (e) is the MRI three-dimensional volumetric mesh model of the patient's ventricle; (f) is the CT three-dimensional volumetric mesh model of the patient's ventricle; and (g) is the CT three-dimensional volumetric mesh model of the patient's ventricle after removing the excess basal portion.

[0040] Figure 3 The diagram shows the model fusion; (a) shows the infarct distribution in the MRI ventricular model; (b) shows the fat distribution in the CT ventricular model; and (c) shows the fat and infarct distribution in the hybrid CT-MRI ventricular model. Detailed Implementation

[0041] The following uses data from a single patient as an example to further illustrate the invention in conjunction with specific implementation steps, such as... Figure 1 As shown, the method for constructing a personalized 3D cardiac model based on multimodal imaging for CT-MRI includes the following steps:

[0042] Step 1: Data preprocessing.

[0043] S11, Image Preprocessing

[0044] Deep learning-based methods were used to segment CMR-LGE and CE-CT images of patients acquired clinically to obtain ventricular myocardial images; CE-CT images were resampled from short-axis to long-axis cardiac images.

[0045] S12, Ventricular cavity interpolation reconstruction

[0046] Model reconstruction was performed on the segmentation results of the CMR-LGE images. Constraint points located at the myocardial boundaries were selected within the obtained myocardial segmentation regions. Variational implicit function interpolation was used to interpolate these discrete points, generating fitted surfaces for the left ventricular endocardium, left ventricular endocardium, right ventricular endocardium, right ventricular endocardium, and aorta, respectively. The left ventricular endocardium and right ventricular endocardium surfaces were combined to construct a biventricular cavity model, which was saved as a high-resolution (0.35×0.35×0.35 mm³) MRI-Label image, as shown below. Figure 2 As shown in (a), the aorta-left ventricular endocardium model was constructed by combining the curved surfaces of the aorta and the left ventricular endocardium, and saved as an MRI-Aorta image of the same resolution. Considering that the CE-CT image itself has a high resolution, it was interpolated to the same voxel size (0.35×0.35×0.35 mm³) for easier subsequent spatial mapping processing, and saved as CT-Label images, as shown in (a). Figure 2 As shown in (b), its aorta was saved as a CT-Aorta image of the same resolution;

[0047] S13, Reconstructing infarcted tissue

[0048] The CMR-LGE images are segmented using a Gaussian mixture model (GMM) based method or a thresholding method. Then, a logarithmic probabilistic algorithm (LogOdds) is used to interpolate the segmented infarct regions to obtain high-resolution three-dimensional image data of the infarct tissue. This interpolated result is then merged with the corresponding biventricular high-resolution MRI-Label image to obtain an MRI-SCAR image, such as... Figure 2 As shown in (c);

[0049] S14, Reconstructing adipose tissue

[0050] A thresholding method was used to segment the adipose tissue region in the myocardium in CE-CT images. Low-attenuation regions between -180 HU and -5 HU were defined as adipose tissue. This range was further subdivided into -180 HU to -50 HU (defined as dense adipose region) and -50 HU to -5 HU (defined as adipose-myocardial mixed region). To remove noise artifacts in CE-CT images that might be misidentified as adipose tissue, the segmentation results were processed to remove regions with a volume less than or equal to 1 mm³. This yielded denoised dense adipose and adipose-myocardial mixed images. These images were then merged with the corresponding CT-Label images to obtain the CT-Fat image, as shown below. Figure 2 As shown in (d);

[0051] S15. Establish a finite element mesh model of the heart.

[0052] Based on the aforementioned MRI-Label and CT-Label images, corresponding high-precision cardiac cavities were constructed. The construction process included surface mesh generation and smoothing optimization, followed by volume mesh generation, thereby generating MRI and CT ventricular models, as shown below. Figure 2 As shown in Figure 2(e) and Figure 2(f). Then, based on the aforementioned MRI-Aorta and CT-Aorta images, the corresponding aorta is constructed. Then, the MRI and CT ventricular models are rigidly registered using the aorta, and the excess portion of the CT ventricular model's base is removed, as shown in Figure 2(f). Figure 2 As shown in (g). Finally, the generated model data is saved as MRI.pts and CT.pts files recording the coordinates of the ventricular model points, MRI.elem and CT.elem files recording the vertex indices of the tetrahedral mesh, and MRI.vtu and CT.vtu files recording the volumetric mesh;

[0053] S16. Extract the heart surface mesh.

[0054] From the volumetric meshes of MRI and CT ventricular models, four main boundary surfaces were extracted: the base surface (Base), the epicardial surface (Epi), the left ventricular endocardial surface (LV), and the right ventricular endocardial surface (RV). Corresponding surface mesh coordinate files were generated for each extracted surface and saved, including MRI-Base.ply, MRI-Epi.ply, MRI-LV.ply, MRI-RV.ply, and CT-Base.ply, CT-Epi.ply, CT-LV.ply, and CT-RV.ply.

[0055] Step 2: Mapping the infarcted area to the fatty area;

[0056] S21, Infarct Region Mapping

[0057] The infarcted region in the high-resolution MRI-SCAR image is mapped onto an MRI ventricular model. The centroid of each tetrahedral grid in the model is calculated, and the centroid coordinates are divided by the image resolution and rounded down to obtain the corresponding pixel in the MRI-SCAR image for each grid. The MRI-SCAR value of a pixel is 1 for normal myocardial tissue, 2 for hemi-infarcted myocardial tissue, and 3 for complete infarcted tissue. If the pixel is not a myocardial region, non-zero pixels in the first to third order neighborhoods are searched sequentially, and their MRI-SCAR values ​​are used as the type of the grid.

[0058] S22, Fat Region Mapping

[0059] The fatty regions in the high-resolution CT-Fat image are mapped onto the CT ventricular model. The centroid of each tetrahedral grid in the model is calculated, and the centroid coordinates are divided by the image resolution and rounded down to obtain the corresponding pixel in the CT-Fat image for each grid. A pixel with a CT-Fat value of 1 represents normal myocardial tissue, 2 represents mixed fat-myocardial tissue, and 3 represents dense adipose tissue. If the pixel is not a myocardial region, non-zero pixels in the first to third order neighborhoods are searched sequentially, and their CT-Fat values ​​are used as the type of the grid.

[0060] Step 3: Establish universal ventricular coordinates

[0061] Biventricular coordinates were established for both MRI and CT ventricular models. The MRI.vtu and CT.vtu files, recording the volumetric mesh, and the corresponding cardiac surface mesh files (including four surface types: Base, Epi, LV, and RV, in .ply format) were used. Next, transventricular coordinates were established by calculating and binarizing the transventricular Laplace solution, followed by an initial mesh reconstruction to extract the surface and curves of the interventricular septum. Then, the transmural trajectory distance was calculated and normalized based on the inner and outer boundaries and the interventricular septum to obtain the transmural coordinates. Simultaneously, the anatomical axis and apex of the heart were determined. Further, the ridge surface was extracted by calculating the Laplace solution between the epicardium and the interventricular septum and performing a second mesh reconstruction. The rotational trajectory distance was calculated using the gradient of the transmural coordinates and the apical-base Laplace solution, and normalized and adjusted to obtain the rotational coordinates. Finally, the normalized distance along the contour lines of the transmural and rotational coordinates was calculated, and the apical-base coordinates were obtained using Laplace extrapolation. Generate MRI-UVC.vtu and CT-UVC.vtu files respectively, and store the matrix that associates the established coordinates with the original grid data structure;

[0062] Step 4: Establish a CT-MRI ventricular model

[0063] S41. Convert cell data to point data

[0064] The data of infarct cells stored in the MRI.vtu file is converted into data defined on the vertices of the dataset. For each vertex in the dataset, the vertex data value corresponding to that vertex is determined by averaging the cell data values ​​of the cells containing that vertex;

[0065] S42, Model Fusion

[0066] A hybrid CT-MRI ventricular model is constructed by mapping the infarct distribution in an MRI ventricular model to the geometry of a CT ventricular model. Using a linear interpolation method based on ventricular coordinates, for each node in the CT ventricular model mesh, a set of candidate MRI ventricular model mesh elements corresponding to the CT ventricular model node coordinates is determined in the ventricular coordinate space. The centroid coordinates corresponding to the CT ventricular model node coordinates are then calculated within these candidate elements. Subsequently, for each node data of the CT ventricular model, based on the calculated centroid coordinates, the MRI ventricular model mesh element that minimizes the maximum absolute deviation of the centroid coordinates is selected as the optimal element from the candidate element set. Using the node index of the selected optimal element and its corresponding calculated centroid coordinates, a mapping matrix is ​​constructed to linearly interpolate data from the MRI ventricular model mesh nodes to the CT mesh nodes. The point data values ​​of the infarct units on the CT ventricular model mesh are obtained through the mapping matrix, generating the CT-MRI.vtu file.

[0067] S43. Point data is converted into cell data.

[0068] The point data of infarct units stored in the CT-MRI.vtu file are converted into data defined at the units in the dataset. For each unit in the dataset, the set of points constituting the unit is determined, and the point data values ​​in the set are subject to majority voting, wherein in the case of a tie, the smaller value is selected to calculate the unit data value corresponding to the unit.

[0069] S44, Assigning Labels

[0070] In the hybrid CT-MRI ventricular model, the labels are as follows: 1 represents normal myocardial tissue, 2 represents a mixture of dense fatty infiltration and scar tissue, 3 represents fibrofatty infiltration myocardial tissue, 4 represents completely infarcted tissue, 5 represents dense fatty tissue, 6 represents fat-free hemi-infarcted myocardial tissue, and 7 represents non-fibrotic fatty-myocardial mixed tissue. Finally, save the CT-MRI.pts file containing the point coordinates of the ventricular model and the CT-MRI.elem file containing the tetrahedral mesh vertex index, as follows: Figure 3 As shown, the hybrid CT-MRI ventricular model achieves precise fusion of multiple tissue types in terms of structure. Different labeled regions are clearly distributed in space with continuous interfaces, which can effectively distinguish different pathological tissues such as normal myocardium, fibrosis, and fatty infiltration. This provides a high-precision anatomical basis and histological evidence for subsequent electrophysiological simulation and myocardial pathological mechanism analysis.

Claims

1. A method for constructing a personalized 3D cardiac model based on CT-MRI using multimodal imaging, characterized in that, The steps are as follows: S1, Data Preprocessing; S11, Image Preprocessing Deep learning-based methods were used to segment the acquired cardiac CMR-LGE and CE-CT images to obtain images of the patient's ventricular myocardium; the CE-CT images were resampled from short-axis to long-axis cardiac images. S12, Ventricular cavity interpolation reconstruction Based on the segmentation results of the CMR-LGE images, constraint points located at the myocardial boundaries were selected within the acquired myocardial region. Variational implicit function interpolation was used to interpolate these discrete points, generating fitted surfaces for the left ventricular endocardium, right ventricular endocardium, and aorta, respectively. The left and right ventricular endocardium surfaces were combined to construct a biventricular cavity model, which was saved as an MRI-Label image. The aorta and left ventricular endocardium surfaces were combined to construct an aorta-left ventricular endocardium model, which was saved as an MRI-Aorta image of the same resolution. The CE-CT images were interpolated and saved as CT-Label images, and the aorta was saved as a CT-Aorta image of the same resolution. S13, Reconstructing infarcted tissue The myocardial infarction region was extracted and reconstructed from CMR-LGE images. First, the myocardial region in the CMR-LGE images was analyzed using a Gaussian mixture model (GMM) or threshold segmentation method to obtain preliminary segmentation results of the myocardial infarction region. Then, the segmented infarction region was interpolated and reconstructed using the logarithmic interpolation algorithm LogOdds to generate three-dimensional infarct tissue image data. Finally, this interpolation result was merged with a high-resolution MRI-Label image of both ventricles to obtain an MRI-Scar image. S14, Reconstructing adipose tissue The myocardial adipose tissue was segmented and extracted from CE-CT images. First, the myocardial region was analyzed using a threshold segmentation method, and the region with attenuation between -180 HU and -5 HU was defined as adipose tissue. Further, this range was subdivided into two intervals: -180 HU to -50 HU was defined as dense adipose region, and -50 HU to -5 HU was defined as adipose-myocardial mixed region. Subsequently, regions with a volume less than or equal to 1 mm³ were removed. After processing, a denoised dense fat image and a fat-myocardial hybrid image were obtained, and the two were merged with the CT-Label image to generate a CT-Fat image; S15. Establish a finite element mesh model of the heart. Based on the aforementioned MRI-Label and CT-Label images, the corresponding cardiac cavities were constructed. The construction process included surface mesh generation and smoothing optimization, followed by volume mesh generation, thereby generating MRI and CT ventricular models. Then, based on the aforementioned MRI-Aorta and CT-Aorta images, the corresponding aorta was constructed. The MRI and CT ventricular models were then rigidly registered using the aorta, and excess portions of the CT ventricular model's base were removed. Finally, the generated model data were saved as MRI.pts and CT.pts files recording the coordinates of the ventricular model points, MRI.elem and CT.elem files recording the tetrahedral mesh vertex indices, and MRI.vtu and CT.vtu files recording the volume mesh. S16. Extract the heart surface mesh. From the volumetric mesh of MRI and CT ventricular models, the corresponding four boundary surfaces—Base, Epicardial surface, Left Ventricular Endocardial surface (LV), and Right Ventricular Endocardial surface (RV)—are extracted. The extracted data are saved as MRI-Base.ply, MRI-Epi.ply, MRI-LV.ply, MRI-RV.ply, CT-Base.ply, CT-Epi.ply, CT-LV.ply, and CT-RV.ply files, which record the coordinates of the mesh points on the cardiac surfaces. S2, mapping of infarcted and fatty regions; S21, Infarct Region Mapping The infarcted region in the MRI-Scar image is mapped to the MRI ventricular model. The centroid of each tetrahedral grid in the model is calculated, and the centroid coordinates are divided by the image resolution and rounded down to obtain the corresponding pixel in the MRI-Scar image for each grid. The MRI-Scar value of a pixel is 1 for normal myocardial tissue, 2 for hemi-infarcted myocardial tissue, and 3 for complete infarcted tissue. If the pixel is a non-myocardial region, non-zero pixels in the first to third order neighborhoods are searched sequentially, and their MRI-Scar values ​​are used as the type of the grid. S22, Fat Region Mapping The fatty regions in the CT-Fat image are mapped to the CT ventricular model. The centroid of each tetrahedral grid in the model is calculated, and the centroid coordinates are divided by the image resolution and rounded down to obtain the corresponding pixel of each grid in the CT-Fat image. A pixel with a CT-Fat value of 1 represents normal myocardial tissue, 2 represents mixed fat-myocardial tissue, and 3 represents dense adipose tissue. If the pixel is not a myocardial region, non-zero pixels in the first to third order neighborhoods are searched sequentially, and their CT-Fat values ​​are used as the type of the grid. S3. Establish universal ventricular coordinates Establish biventricular coordinates for both the MRI and CT ventricular models. S4. Establish a CT-MRI ventricular model; S41. Convert cell data to point data S42, Model Fusion A hybrid CT-MRI ventricular model is constructed by mapping the infarct distribution in the MRI ventricular model to the geometry of the CT ventricular model. By using a linear interpolation method based on ventricular coordinates, for each node in the CT ventricular model mesh, a set of candidate cell volumes of the MRI ventricular model mesh corresponding to the node coordinates of the CT ventricular model is determined in the ventricular coordinate space, and the centroid coordinates corresponding to the node coordinates of the CT ventricular model are calculated within these candidate cell volumes. Subsequently, for each node data of the CT ventricular model, based on the calculated centroid coordinates, the MRI ventricular model grid cell that minimizes the maximum absolute deviation of the centroid coordinates is selected as the optimal cell from the candidate cell set; using the node index of the selected optimal cell and its corresponding calculated centroid coordinates, a mapping matrix is ​​constructed to linearly interpolate the data from the MRI ventricular model grid nodes to the CT grid nodes; the point data values ​​of the infarct cells on the CT ventricular model grid are obtained through the mapping matrix, and the CT-MRI.vtu file is generated; S43. Point data is converted into cell data. The point data of the infarcted cells stored in the CT-MRI.vtu file are converted into data defined at the cells in the dataset; for each cell in the dataset, the set of points constituting the cell is determined, and the point data values ​​in the set of points are subject to majority voting, wherein in the case of a tie, the smaller value is selected to calculate the cell data value corresponding to the cell; S44, Assigning Labels In the hybrid CT-MRI ventricular model, the labels are as follows: 1 represents normal myocardial tissue, 2 represents a mixture of dense fatty infiltration and scar tissue, 3 represents fibrofatty infiltration myocardial tissue, 4 represents completely infarcted tissue, 5 represents dense fatty tissue, 6 represents fat-free semi-infarcted myocardial tissue, and 7 represents non-fibrotic fatty-myocardial mixed tissue. Finally, the CT-MRI.pts file containing the point coordinates of the ventricular model and the CT-MRI.elem file containing the tetrahedral mesh vertex index are saved.

2. The method for constructing a personalized CT-MRI three-dimensional heart model based on multimodal imaging according to claim 1, characterized in that, S3 is as follows: The above-mentioned MRI.vtu and CT.vtu files record the volume grid and the corresponding heart surface grid file; then, by calculating the transventricular Laplace solution and binarizing it, the transventricular coordinates are established, and then the first grid reconstruction is performed to extract the surface and curves of the interventricular septum; Then, based on the inner and outer boundaries and the interventricular septum, partial differential equations are solved to calculate the transmural trajectory distance and normalize it to obtain the transmural coordinates; simultaneously, the anatomical axis and apex of the heart are determined; further, the spinal surface is extracted by calculating the Laplace solution between the epicardium and the interventricular septum and performing a second mesh reconstruction; the rotational trajectory distance is calculated using the trajectory field defined by the gradient of the transmural coordinates and the apical-base Laplace solution, and the rotational coordinates are obtained after normalization and adjustment; finally, by extracting the contour lines of the transmural and rotational coordinates, the normalized distance along the contour lines is calculated, and the apical-base coordinates are obtained using Laplace extrapolation. Generate MRI-UVC.vtu and CT-UVC.vtu files respectively, and store the matrix that associates the established coordinates with the original grid data structure.

3. The method for constructing a personalized CT-MRI three-dimensional heart model based on multimodal imaging according to claim 1, characterized in that, S41 is as follows: The data of the infarcted cells stored in the MRI.vtu file is converted into data defined on the vertices of the dataset; for each vertex in the dataset, the vertex data value corresponding to the vertex is determined by averaging the cell data values ​​of the cells containing that vertex.

Images