A method, device, equipment and medium for measuring hippocampal substructure morphology

Abstract

The embodiment of the present invention provides a method, device, equipment and medium for measuring the morphology of hippocampal substructures, which relates to the field of image processing technology. The method includes: segmenting a magnetic resonance standard template image, constructing a boundary optimization surface, and determining a skeleton representation; sampling the skeleton representation, determining the subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of the template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation; determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template; determining the morphological characteristics of the target hippocampus based on the positions of the subregion skeleton points in the target skeleton representation. The embodiment of the present invention can perform refined measurements of hippocampal substructures and achieve stable correspondence of local morphologies across individuals.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method, device, equipment and medium for measuring the morphology of hippocampal substructures. Background Art

[0002] The hippocampus, also known as the hippocampus, hippocampal area, and brain hippocampus, is located between the thalamus and the medial temporal lobe of the brain. It is part of the limbic system and is mainly responsible for the storage, conversion, and orientation of short-term memory. Different subregions of the hippocampus are heterogeneous in anatomical structure, functional properties, and pathological changes. Quantitative measurement of local thickness, morphology, and other characteristics of different subregions of the hippocampus can help with early identification of the disease and mechanism research.

[0003] Because the hippocampal structure is highly curled and its morphology varies greatly across individuals, for example, the CA1 region in the hippocampal head may have 1 to 4 finger-like protrusions depending on the individual, and characteristics such as thickness are difficult to compare between different individuals. The measurement methods in related technologies can only perform morphological measurements of the hippocampus as a whole, and cannot achieve accurate and stable thickness and morphological feature measurements of internal subregions (such as CA1, CA3, DG, etc.) at the individual level, and lack stable correspondence between local morphologies across individuals. Summary of the Invention

[0004] The problem solved by the present invention is how to perform refined measurements of the hippocampus and achieve stable correspondence of local morphology across individuals.

[0005] To solve the above problems, in some embodiments, a method, device, equipment and medium for measuring the morphology of hippocampal substructures are provided.

[0006] In a first aspect, in some embodiments, a method for measuring hippocampal substructure morphology is provided, comprising:

[0007] Segment the magnetic resonance standard template image, construct the boundary optimization surface, and determine the skeleton representation;

[0008] Sampling the skeleton representation to determine subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of the template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation;

[0009] determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template;

[0010] The morphological features of the target hippocampus are determined based on the positions of the subregion skeleton points in the target skeleton representation.

[0011] Optionally, segmenting the magnetic resonance standard template image, constructing the boundary optimization surface, and determining the skeleton representation include:

[0012] Segmenting the magnetic resonance standard template image to generate segmentation label images of each hippocampal subregion within the template hippocampus;

[0013] generating the boundary optimization surface according to the segmented label image;

[0014] The boundary optimization surface is converted into a Voronoi diagram and fitted to generate the skeleton representation.

[0015] Optionally, the skeleton representation includes a skeleton surface and a radial axis vector, the skeleton points are located on the skeleton surface, and the converting of the boundary optimization surface into a Voronoi diagram and fitting using a non-uniform rational B-spline surface fitting technique to generate the skeleton representation includes:

[0016] Converting the boundary optimization surface into the Voronoi diagram, retaining the Voronoi vertices in the Voronoi diagram that are within the boundary optimization surface, and constructing a point cloud;

[0017] Fitting the point cloud to generate the skeleton surface;

[0018] Based on the skeleton face, the spoke axis vector is determined according to a restriction criterion and a penalty function.

[0019] Optionally, fitting the point cloud to generate the skeleton surface includes:

[0020] dividing the hippocampal subregions in the template hippocampus into regular hippocampal subregions and complex hippocampal subregions;

[0021] Fitting the point cloud of the regular hippocampal subregion to generate a skeleton surface of the regular hippocampal subregion;

[0022] The complex hippocampal subregion is cut into at least two complex sub-subregions, the point cloud of each complex sub-subregion is fitted to generate a sub-subregion skeleton surface of the complex sub-subregion, and each sub-subregion skeleton surface of the complex sub-subregion is merged to generate a skeleton surface of the complex hippocampal subregion.

[0023] Optionally, sampling the skeleton representation to determine subregion thickness measurement points of each hippocampal subregion includes:

[0024] According to the layered distribution law of the hippocampus morphological model, the radial axis of the hippocampus morphological model passes through the skeleton surface of the skeleton representation, and the intersection point of the radial axis of the hippocampus morphological model and the skeleton surface is determined;

[0025] The sub-region thickness measurement point is determined according to the intersection point.

[0026] Optionally, the morphological features include at least one of hippocampal surface parameters, hippocampal thickness parameters, hippocampal length parameters, hippocampal long axis parameters and subregion thickness.

[0027] Optionally, determining the morphological features of the target hippocampus according to the positions of the subregion skeleton points in the target skeleton representation includes:

[0028] When the sub-region skeleton point is located inside the target boundary optimization surface, optimizing the target radial axis vector in the target skeleton representation, and determining the sub-region thickness according to the optimized target radial axis vector;

[0029] When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is greater than or equal to two, the sub-region skeleton point and its corresponding radial axis end point are updated, and the sub-region thickness is determined based on the updated sub-region skeleton point and its corresponding radial axis end point, wherein the skeleton radial axis line is a line connecting the sub-region skeleton point and its corresponding radial axis end point;

[0030] When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is less than two, the sub-region thickness is set to zero.

[0031] In a second aspect, in some embodiments, a device for measuring hippocampal substructure morphology is provided, comprising:

[0032] Segmentation module, used to segment the magnetic resonance standard template image, construct the boundary optimization surface, and determine the skeleton representation;

[0033] a template module for sampling the skeleton representation, determining subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of the template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation;

[0034] a mapping module for determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template;

[0035] A morphological module is used to determine the morphological characteristics of the target hippocampus according to the positions of the sub-region skeleton points in the target skeleton representation.

[0036] In a third aspect, in some embodiments, an electronic device is provided, including a memory and a processor;

[0037] The memory is used to store computer programs;

[0038] The processor is configured to implement the hippocampal substructure morphology measurement method as described in the first aspect when executing the computer program.

[0039] In a fourth aspect, in some embodiments, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the hippocampal substructure morphology measurement method as described in the first aspect is implemented.

[0040] In some embodiments, a skeleton representation is determined by segmenting the magnetic resonance image, providing a stable internal reference frame for thickness measurement; the skeleton representation is then sampled to determine subregion thickness measurement points and a thickness measurement template is constructed, so that the subregion thickness measurement points across individuals have a stable correspondence, achieving unified measurement standards and solving the problem of difficulty in corresponding measurement points caused by individual morphological differences; based on the target boundary optimization surface and thickness measurement template, a target skeleton representation is generated, which can maintain a one-to-one correspondence between the template and the target hippocampus under large deformation, ensuring the anatomical consistency of subregion thickness measurement points in different individuals; finally, the morphological characteristics are determined according to the position of the subregion skeleton points in the target skeleton representation, and the accuracy of the morphological measurement is further improved by targeted optimization of the skeleton points in different situations inside and outside the boundary, thereby achieving overall accurate and stable measurement of hippocampal subregions and reliable correspondence of local morphology across individuals, providing more sensitive structural indicators for early diagnosis and mechanism research of neurodegenerative diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of a process for measuring the morphology of hippocampal substructures provided by an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of segmentation label images of 10 sub-regions provided in an embodiment of the present invention;

[0043] Figure 3 Schematic diagram of boundaries with different resolutions provided by an embodiment of the present invention;

[0044] Figure 4 A schematic diagram of a Voronoi diagram provided by an embodiment of the present invention;

[0045] Figure 5 A schematic diagram of a fitting surface of a regular sub-region provided in an embodiment of the present invention;

[0046] Figure 6 A schematic diagram of a fitting surface of a complex sub-region provided in an embodiment of the present invention;

[0047] Figure 7 A schematic diagram of a skeleton representation provided by an embodiment of the present invention;

[0048] Figure 8A schematic diagram of the structure of a device for measuring the morphology of hippocampal substructures provided by an embodiment of the present invention;

[0049] Figure 9 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for exemplary purposes only and are not intended to limit the scope of protection of the present invention. It should be understood that the various steps described in the method embodiments of the present invention can be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0051] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to"; the term "based on" means "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0052] It should be noted that the modifications of "one" and "multiple" mentioned in the present invention are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0053] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0054] like Figure 1 As shown, an embodiment of the present invention provides a method for measuring the morphology of hippocampal substructures, comprising the following steps:

[0055] Segment the magnetic resonance standard template image, construct the boundary optimized surface, and determine the skeleton representation.

[0056] Specifically, the magnetic resonance standard template image is a standard template generated from magnetic resonance T1 images of multiple people, also called an MNI152 image, that is, a standard template generated from T1 MRI images of 152 people, and is stored in the form of a nii file.

[0057] In one embodiment, the steps of segmenting the magnetic resonance standard template image, constructing the boundary optimized surface, and determining the skeleton representation include:

[0058] Segmenting the magnetic resonance standard template image to generate segmentation label images of each hippocampal subregion within the template hippocampus;

[0059] generating the boundary optimization surface according to the segmented label image;

[0060] The boundary optimization surface is converted into a Voronoi diagram and fitted to generate the skeleton representation.

[0061] Specifically, the magnetic resonance standard template image can be segmented by segmentation software, such as Freesurfer software, based on the FS60 segmentation protocol, and a total of 12 hippocampal subregion segmentation label images can be generated, that is, the template hippocampal substructure is segmented into 12 hippocampal subregions, and the hippocampal subregions are CA1 (Cornu Ammonis 1), CA3 (Cornu Ammonis3), CA4 (Cornu Ammonis 4), GC-ML-DG (Granule Cell layer-Molecular Layer-Dentate Gyrus), molecular_layer (molecular layer), HATA (Hippocampal-Amygdala Transition Area), subiculum (subiculum), parasubiculum (parasubiculum), presubiculum (presubiculum), tail (hippocampal tail), fimbria (hippocampal fimbria) and hippocampal Since the structures of the fimbria and hippocampal fissure subregions are very thin, the segmented voxels are discrete blocks and cannot form a connected surface. Therefore, in this embodiment, only the remaining 10 subregions are selected for subsequent steps. The segmentation label images of the remaining 10 subregions are as follows: Figure 2As shown. In addition, since the hippocampal subregion accounts for a very small proportion in the human brain, the integrity and accuracy of the generated boundary surface strongly depend on the resolution of the label image. For magnetic resonance imaging, the resolution of the original image is mostly around 1mm. The segmented label image segmented by Freesurfer is in the same space as the original image and has a resolution of 1mm. For example, Figure 3 The initial boundary surface of the reconstructed CA3 subregion is provided when the label image resolution is 1mm, 0.5mm, and 0.1mm, respectively. Figure 3As shown, when the resolution is 1mm, due to the small number of voxels, the number of points in the coordinate space is too small, and the subsequent steps cannot determine enough valid connection relationships, causing the reconstructed boundary surface to be fragmented and unable to form a closed spatial object. As the resolution of the label file gradually increases, the number of points in the coordinate space increases, and at 0.5mm, a smooth closed surface is formed, and at 0.1mm, the hippocampal boundary with a stepped voxel morphology is basically reproduced. Based on the feasibility of subsequent thickness feature measurement, this embodiment selects the resolution of the segmented label image to be 0.5mm, which can improve the accuracy of the hippocampal subregion morphology and the closedness of the boundary surface. Therefore, the low-resolution segmented label images are uniformly resampled to 0.5mm for the construction of the subregion thickness measurement template and the cross-individual step. Freesurfer is an open-source software widely used for brain image processing and analysis. It is primarily used for automated brain structure segmentation, cortical reconstruction, and morphometric analysis of magnetic resonance imaging (MRI) data (particularly T1-weighted images). It is commonly used in neuroscience and clinical research for tasks such as identifying brain region boundaries, measuring cortical thickness, and delineating subregions. In this embodiment, Freesurfer's core function is to provide accurate segmentation results for hippocampal subregions. Its built-in FS60 segmentation scheme processes the MNI152 standard template (or T1 MRI images of the target individual) to generate labeled images for the 12 hippocampal subregions (in this paper, 10 of these subregions, such as CA1, CA3, GC-ML-DG, etc., are selected to form connected surfaces). These labeled images are stored in voxel format at the same resolution as the original images (after resampling to 0.5 mm for subsequent analysis). 3D modeling software, such as Rhino, can be used. Rhino is a 3D modeling software based on non-uniform rational B-splines (Nurbs), also known as Rhino. The segmented label image is optimized using Rhino to eliminate errors in reconstructing the boundary surface from the segmented label image, thereby obtaining the optimized boundary surface for each hippocampal subregion. The skeleton representation embodies geometric symmetry. For a closed, thin, sheet-like object, such as the hippocampus, the skeleton surface should be located in the middle of the closed surface. In geometric modeling, it is generally desirable to establish it in a position that exhibits morphological symmetry. First, the optimized boundary surface is converted into a Voronoi diagram. A Voronoi diagram, also known as a Thiessen polygon, is a spatial partitioning method based on a discrete point set. The principle is to construct a spatial boundary using the perpendicular bisectors of the lines connecting adjacent control points, dividing the boundary into multiple non-overlapping convex polygonal regions. Each region corresponds to a unique generating point, and the Euclidean distance from any spatial point in the region to the generating point is less than the distance to all other generating points. The Voronoi diagram is as follows: Figure 4As shown, the non-uniform rational B-spline surface fitting technology is then used to fit and generate a skeleton representation.

[0062] The skeleton representation is sampled to determine subregion thickness measurement points of each hippocampal subregion, and a thickness measurement template of the template hippocampus is constructed based on the subregion thickness measurement points and the skeleton points in the skeleton representation.

[0063] Specifically, the skeleton representation can be sampled using ARMM technology. ARMM stands for the Axis-referenced Morphometric Model of the hippocampus. Based on medial axis geometry theory, the ARMM creates an intrinsic, unified coordinate system for the hippocampus. This coordinate system, developed around the inscribed skeletal plane of the entire hippocampus, incorporates the histological definition of the hippocampal long axis and uses geometric methods such as conformal mapping to precisely partition the hippocampus into layers. Using the ARMM coordinate system and its radial axes, it is possible to measure and compare multiple morphological indices across individuals, including hippocampal thickness, length, width, and longitudinal curvature. The morphological correspondence between individuals reconstructed using this model is stronger than that of advanced hippocampal morphological models, such as continuous internal fitting models and discrete skeleton fitting models. Therefore, the skeleton representation is sampled according to the ARMM technology, and the correspondence between the skeleton representation of the subregion and the ARMM skeleton is established, so as to determine the subregion thickness measurement points of each hippocampal subregion, and extract the skeleton points corresponding to the subregion thickness measurement points. A set of measurement sites and corresponding skeleton points of each hippocampal subregion constitute the thickness measurement template of the hippocampal subregion. The thickness measurement templates of all hippocampal subregions are combined with the boundary optimization surface of the magnetic resonance standard template image to obtain the thickness measurement template of the template hippocampus.

[0064] A target boundary optimization surface corresponding to the target magnetic resonance image of the target hippocampus is determined, and a target skeleton representation is generated based on the target boundary optimization surface and the thickness measurement template.

[0065] Specifically, the target hippocampus refers to the hippocampus whose thickness is to be measured. First, the target MRI image of the target hippocampus is preprocessed. The image is segmented using Freesurfer software through the aforementioned segmentation and optimization steps, and resampled by 0.5 mm to obtain a segmented label image of the target hippocampus. This image is stored as a nii file, and then a target boundary optimized surface is generated using Rhino software. Based on the target boundary optimized surface and the thickness measurement template, a large-scale deformation diffeomorphic metric mapping technique (LDDMM) is then used to guide the deformation from the template to the target hippocampus, i.e., the transformation from the thickness measurement template to the target hippocampus. The large-scale deformation diffeomorphic metric mapping technique can adjust the velocity and momentum field of the control points during the deformation process to obtain a deformation result that optimally matches the target shape. It also achieves a one-to-one correspondence between corresponding vertices during the deformation process, generating a target skeleton representation.

[0066] The morphological features of the target hippocampus are determined based on the positions of the subregion skeleton points in the target skeleton representation.

[0067] In one embodiment, the morphological features include at least one of hippocampal surface parameters, hippocampal thickness parameters, hippocampal length parameters, hippocampal long axis parameters, and subregion thickness.

[0068] Specifically, the morphological features include at least one of hippocampus surface parameters, hippocampus thickness parameters, hippocampus length parameters, hippocampus long axis parameters and sub-region thickness, among which the hippocampus surface parameters, hippocampus thickness parameters, hippocampus length parameters and hippocampus long axis parameters can be determined based on the target skeleton representation or target boundary optimization surface obtained in this embodiment, using methods such as surface morphological measurement methods based on the longitudinal structure of the human hippocampus, and the sub-region thickness can be determined based on the position of the sub-region skeleton point in the target skeleton representation.

[0069] Specifically, according to the positions of the subregion skeleton points in the target skeleton representation, the skeleton points and the radial axis are optimized and updated, and the thickness of the subregion of the target hippocampus is determined using a thickness formula. The thickness formula includes:

[0070] ;

[0071] Among them, T n is the thickness of the nth sub-region, is the coordinate of the upper skeleton point updated in the radial axis optimization of the nth original skeleton point, is the coordinate of the lower skeleton point updated in the lower radial axis optimization of the nth original skeleton point, for The coordinates of the corresponding upper spoke vertex, is the coordinate of the corresponding lower spoke vertex.

[0072] It should be understood that the segmented label image, the boundary optimized surface, the Voronoi diagram, and the skeleton representation are all corresponding processing of the magnetic resonance image to obtain different representations of the hippocampal morphology.

[0073] In this embodiment, the magnetic resonance image is segmented by Freesurfer software to generate segmentation label images of each hippocampal subregion, which can accurately locate the boundaries of subregions such as CA1, GC-ML-DG, provide a reliable initial segmentation basis for subsequent fine analysis at the subregion level, and avoid neglect of subregion details by overall segmentation; based on the segmentation label image, a three-dimensional modeling software is used to generate a boundary optimization surface, which can effectively remove topological defects such as free grids and self-intersecting grids, obtain a smooth and closed surface, eliminate the radial axis optimization error caused by incomplete surface, and lay a geometric foundation for accurate skeleton construction and thickness measurement; the boundary optimization surface is converted into a Voronoi diagram and a skeleton representation is generated by fitting a non-uniform rational B-spline surface. The Voronoi diagram can accurately extract the point cloud of the central axis position inside the subregion, and the fitted skeleton surface conforms to the geometric characteristics of the central axis. Combined with the radial axis that meets the boundary convergence, orthogonality and non-self-intersection criteria, a stable internal reference frame is provided for thickness measurement; according to the ARMM technique The technique samples the skeleton representation to determine the subregion thickness measurement points and constructs a thickness measurement template. With the help of the anatomical coordinate system of ARMM based on longitudinal laminar tissue, the subregion thickness measurement points across individuals have a stable correspondence, which realizes the unification of measurement standards and solves the problem of difficulty in corresponding measurement points caused by individual morphological differences. Based on the target boundary optimization surface and thickness measurement template, the target skeleton representation is generated using large-scale deformation differential homeomorphism metric mapping technology, which can maintain a one-to-one correspondence between the template and the target hippocampus under large deformation, ensuring the anatomical consistency of the subregion thickness measurement points of different individuals. Finally, the subregion thickness is determined according to the position of the subregion skeleton points in the target skeleton representation. The accuracy of thickness measurement is further improved by targeted optimization of the skeleton points in different situations inside and outside the boundary, thereby realizing the overall accurate and stable measurement of hippocampal subregions and reliable correspondence of local morphology across individuals, providing a more sensitive structural indicator for the early diagnosis and mechanism research of neurodegenerative diseases.

[0074] Optionally, generating the boundary optimization surface according to the segmented label image includes:

[0075] Extracting a binary image of the segmented label image, and performing voxel reconstruction on the binary image to generate a boundary surface;

[0076] Rhino software is used to perform free cutting, shrink wrapping, and rule rearrangement on the boundary surface to generate the boundary optimized surface.

[0077] Specifically, the binary image of the corresponding subregion is first extracted from the segmented label image. This can be achieved using the mri_extract_label command integrated into the Freesurfer software. A custom Python script is then used to convert the binary image into a meshed VTK file. VTK (Visualization Toolkit) is an open-source software system widely used in scientific visualization. A VTK file is a text file used in the VTK system to store the coordinates of points in a three-dimensional graphical space and the connections between them. By defining the outermost non-zero voxel in the binary hippocampal image as a point in three-dimensional coordinate space and generating connections between points based on the relative positional relationships between voxels, the hippocampal subregion can be reconstructed as a triangular mesh in the three-dimensional coordinate space using text data. This triangular mesh represents the outermost part of the hippocampal subregion and is therefore called the boundary surface. For complex hippocampal subregions, such as GC-ML-DG and molecular_layer, the boundary surfaces generated using the above method are prone to two types of errors. One is the presence of free closed meshes independent of the hippocampal subregion boundary surface. These free closed meshes significantly affect the construction of the subregion thickness measurement template. The other is that complex hippocampal subregions may cause unnecessary self-intersections when generating the boundary surface due to the close distance between voxels, which manifests as unreasonable void areas. To eliminate these two errors, the Python language, VTK toolkit, and Rhino built-in functions can be used to cut free closed meshes from the hippocampal subregion boundary surface. After cutting, the entire mesh is shrink-wrapped and regularly rearranged to obtain a smooth, closed boundary optimized surface.

[0078] When performing free cutting, the input of the Python language script is the boundary surface. By calling the SplitDisjointMesh command, the input mesh can be cut into several disconnected parts. When the mesh vector after cutting contains more than 2 elements, call mesh.Vertices.Count to return the number of vertices of each mesh in the mesh vector. Select the element with the largest number of vertices as the mesh object for subsequent processing. The reason for selecting the element with the largest number of vertices is that the free mesh originates from the segmentation noise, and the free mesh composed of it is usually much smaller than the main mesh of the hippocampal subregion. When the mesh vector after cutting contains only 1 element, it proves that there is no free mesh on the input surface and the free cutting is completed.

[0079] When shrink-wrapping a main mesh, the ShrinkWrap command can be used to transform a mesh with internally intersecting, self-intersecting structures into a smooth, closed mesh. The ShrinkWrap parameter settings determine the effectiveness of the shrink-wrap. Key parameters include: FillHolesInInputObjects. When True, the algorithm internally fills holes caused by self-intersections; Offset, representing the offset. A positive value indicates that the output mesh is larger than the input, meaning that the output wraps around the input; a negative value indicates the opposite. Typically, Offset is set to 0 to improve mesh size consistency before and after optimization; PolygonOptimization, representing the polygon optimization ratio, controls the number of coplanar meshes. A larger value reduces the number of coplanar meshes, but the mesh and shape of non-coplanar areas remain largely unchanged. A typical setting is 10; TargetEdgeLength, controlling the edge length of the output mesh. Longer edges result in fewer polygons, a coarser mesh shrinkage, and less computational complexity. Due to the morphological inconsistencies among hippocampal subregions, the setting of this parameter varies. For subregions with simpler morphology, such as CA1, CA3, and SUB, this parameter is set high enough to improve shrinkage accuracy while increasing computational efficiency. For subregions with more complex morphology, such as GC-ML-DG and molecular_layer, this parameter is set appropriately low to ensure the accuracy of the output mesh.

[0080] When performing regular rearrangement of the mesh, the QuadRemesh command can be called to retopologize the shrink-wrapped mesh while maintaining its original shape, generating a reasonable square unit mesh structure. The core parameter is TargetQuadCount, which controls the estimated number of mesh faces after rearrangement, that is, the fineness of the mesh. For sub-regions with simpler morphology, such as CA1, CA3, SUB, etc., this parameter is relatively small, and as few points and faces as possible are sufficient to completely reconstruct the true boundary surface of the sub-region. For sub-regions with more complex morphology, such as GC-ML-DG, molecular_layer, this parameter is appropriately increased to ensure the accuracy of the detailed depiction of the mesh edge. After the QuadRemesh rule rearrangement, it is exported to the STL file format, and the square mesh automatically degenerates into a triangular mesh, and the optimization process is completed.

[0081] In this embodiment, the boundary surface is free-cut, shrink-wrapped, and rule-rearranged using Rhino software to generate a boundary optimized surface method, which can be called the RBOS (Rhino-based Optimized Surface) method. In addition, the empirical values ​​of the parameters set for each hippocampal subregion, such as TargetEdgeLength and TargetQuadCount, are 0.4 and 2500, respectively. Depending on the complexity and size of the subregion, they can be adjusted up and down by 0.2 and 1500, respectively, based on visual evaluation.

[0082] Optionally, the skeleton representation includes a skeleton surface and a radial axis vector, the skeleton points are located on the skeleton surface, and converting the boundary optimization surface into a Voronoi diagram and performing fitting to generate the skeleton representation includes:

[0083] Converting the boundary optimization surface into the Voronoi diagram, retaining the Voronoi vertices in the Voronoi diagram that are within the boundary optimization surface, and constructing a point cloud;

[0084] Fitting the point cloud to generate the skeleton surface;

[0085] Based on the skeleton face, the spoke axis vector is determined according to a restriction criterion and a penalty function.

[0086] Specifically, the skeleton representation consists of skeleton faces and radial axis vectors, with the skeleton points located on the skeleton faces. This means that the skeleton faces are composed of skeleton points. A spatial partitioning method can be used to convert the optimized boundary surface into a Voronoi diagram. A filtering algorithm can then be used to retain the Voronoi vertices in the Voronoi diagram that lie within the optimized boundary surface. Non-Uniform Rational B-Splines (Nurbs) surface fitting techniques are then used to construct a continuous surface, or skeleton face, that optimally approximates the Voronoi points within the subregion. Voronoi vertices are generated using a Delaunay triangulation-based method. First, a Delaunay triangulation is constructed from a given set of points, namely the points in the optimized boundary surface of the hippocampal subregion. This triangulation ensures that the circumcircle of any triangle does not contain any other points, thereby improving the uniqueness and stability of the triangulation. Furthermore, the centers of the circumcircles of the Delaunay triangles are calculated, and the connections between these centers are used to generate the edges of the Voronoi diagram, thus forming a spatial partition centered on each point. The Voronoi vertices located within the boundary surfaces of the hippocampal subregions were selected as reference point clouds for subsequent single surface fitting. Subsequently, the patch command in Rhino was used to implement Nurbs single surface fitting. Finally, the skeleton surface of each subregion was obtained. The radial axis was then installed based on the skeleton surface. The length and direction of the radial axis were optimized according to the constraint criteria and penalty function, and the radial axis vector was determined. The constraint criteria are also called the medial axis geometric constraints. The radial axis vector consists of the radial axis vertex and the radial axis starting point. The radial axis vertex also serves as the subregion thickness measurement point.

[0087] Optionally, fitting the point cloud to generate the skeleton surface includes:

[0088] dividing the hippocampal subregions in the template hippocampus into regular hippocampal subregions and complex hippocampal subregions;

[0089] Fitting the point cloud of the regular hippocampal subregion to generate a skeleton surface of the regular hippocampal subregion;

[0090] The complex hippocampal subregion is cut into at least two complex sub-subregions, the point cloud of each complex sub-subregion is fitted to generate a sub-subregion skeleton surface of the complex sub-subregion, and each sub-subregion skeleton surface of the complex sub-subregion is merged to generate a skeleton surface of the complex hippocampal subregion.

[0091] Specifically, based on the structure of the hippocampal subregion and manual experience, the hippocampal subregions are divided into regular hippocampal subregions and complex hippocampal subregions. For regular hippocampal subregions with regular shapes, such as CA3 and subiculum, their Voronoi point clouds and Nurbs surfaces are as follows: Figure 5 As shown, Figure 5The white transparent outline in the middle and outer sides is the boundary surface of the subregion, and the filtered Voronoi vertices are approximately distributed at the inner median axis position of the closed boundary surface. The single surface after Nurbs fitting is smooth and inscribed in the subregion boundary surface. However, for complex hippocampal subregions with complex shapes, such as the CA1 subregion, GC-ML-DG and molecular_layer subregion, it is difficult to fit a single inscribed skeleton surface because the CA1 subregion has a large degree of twisting from its head to the body, and the layered structure of the GC-ML-DG and molecular_layer subregions. Therefore, these three subregions are cut into two parts and divided into anterior (anterior), posterior (posterior), superior (superior), and inferior (inferior) according to the orientation of the hippocampus in the human brain, and are called sub-subregions. By manually cutting the sub-subregions, generating Voronoi vertices and Nurbs fitting, the skeleton surfaces of each sub-subregion are generated. The two sub-subregion skeleton surfaces of the complex hippocampal subregion are merged as the combined skeleton surface of the three subregions, as shown in Figure 6 As shown, Figure 6 This section illustrates the difference between single and combined skeleton surfaces for CA1 and molecular_layer subregions. Combined skeleton surfaces address the problem of CA1 subregions, where distortion of the head and body creates local subregions that are unable to fit within the skeleton surface, and the problem of single skeleton surfaces protruding from the boundary surface due to the layered structure of the GC-ML-DG and molecular_layer subregions. For any single subregion, the subregion skeleton surface is strictly inscribed within the subregion boundary surface, ultimately yielding a skeleton surface for each subregion.

[0092] Optionally, the restriction criterion includes a boundary convergence criterion, a radial axis orthogonality criterion, and a radial axis non-self-intersection criterion, and the penalty function includes a first penalty function corresponding to the boundary convergence criterion, a second penalty function corresponding to the radial axis orthogonality criterion, and a third penalty function corresponding to the radial axis non-self-intersection criterion, wherein the first penalty function, the second penalty function, and the third penalty function are respectively:

[0093] ;

[0094] ;

[0095] ;

[0096] in, i is a point on the skeleton surface, S i for i The spoke axis on the point, The size of the surface is optimized for the deviation of the endpoints of the spokes from the boundary, Spoke Si The Euclidean distance from the end point to the boundary of the optimized surface, is the magnitude of the deviation of the radial axis from the radial axis orthogonality criterion, is the angle between the direction of the radial axis Si and the surface normal vector of its nearest boundary point, is the magnitude of the radial axis deviation from the radial axis non-self-intersection criterion, is the ReLU function value corresponding to the radial axis Si, r i for i The length of the radial axis at point for i The first principal radial curvature at point , is a matrix The maximum eigenvalue of is the radial shape operator, for:

[0097] ;

[0098] ;

[0099] in, S is a continuous vector-valued function consisting of all spokes on the skeleton surface, r i represent S The length of , U is the unit vector of S, is the velocity direction of the curve passing through point i on the tangent plane corresponding to point i on the skeleton surface, and I is the unit matrix.

[0100] Specifically, the boundary convergence criterion states that hippocampal subregion thickness measurements rely on the geometric representation of their topological skeletons, requiring that the endpoints of the radial axes strictly converge to the target boundary surface. This criterion ensures spatial consistency between thickness measurements and subregion segmentation results. The radial axis orthogonality criterion states that, to establish a uniform radial axis direction field, the radial axes must be forced to remain orthogonal to the local boundary tangent plane at their endpoints, ensuring uniformity of measurement benchmarks across different subregions. The radial axis non-self-intersection criterion states that, when measuring radial axis-based thickness, local self-intersections should be avoided to improve reciprocity between boundary surface vertices and skeleton vertices. That is, any point on the subregion boundary surface corresponds to only one radial axis vertex, resulting in a unique thickness measurement. In practice, both the radial axis orthogonality and the radial axis non-self-intersection criteria are often difficult to satisfy simultaneously. This is because the radial axis density is too high on highly curved boundary surfaces. If the radial axes are strictly forced to remain orthogonal to the local boundary tangent plane at their endpoints, self-intersections are likely to occur at their ends. However, if the radial axes intersect with each other, it is easy to cause multiple radial axes to exist at the same boundary surface location, resulting in multiple thickness measurements, making the thickness at the same observation location non-unique. To solve this problem, based on the satisfaction of three constraints, three penalty functions are proposed to meet the radial axis orthogonality criterion and the radial axis non-self-intersection criterion as much as possible. By combining the second penalty function and the third penalty function, the total penalty function for optimizing the direction of each radial axis can be obtained:

[0101] ;

[0102] E is the total penalty function, α is the penalty weight corresponding to the deviation of the radial axis from the radial axis orthogonality criterion, β is the penalty weight corresponding to the deviation of the radial axis from the non-self-intersection criterion.

[0103] For example, when determining the radial axis vector according to the restriction criteria and the penalty function, the first penalty function is first executed on all radial axes so that the error between the radial axis endpoint and the corresponding vertex on the boundary surface is sufficiently small. The total penalty function is then used to adjust the direction of each radial axis. The adjustment of the direction means a change in the position of the radial axis end point, which may cause the radial axis end point to deviate from the boundary surface after the direction is corrected, thereby no longer meeting the boundary convergence criteria. Therefore. The first penalty function is executed again on the radial axis after the total penalty function is executed to fine-tune the radial axis end point, and the initialized radial axis of each sub-region, that is, the radial axis vector, can be obtained. Among them, each radial axis starting point on the skeleton surface corresponds to two radial axis end points, and the vectors from the two radial axis starting points to the radial axis end points are called the upper radial axis and the lower radial axis, respectively. Figure 7The initialization of the radial axes for the CA4 subregion with 100 sampling points is shown. The endpoints of the radial axes (i.e., where the arrows terminate) all conform to the gray boundary surface, with no self-intersections. The radial axes are not perpendicular to the skeleton plane but maintain a certain angle. Where the radial axes approach the subregion boundary surface, they are ensured to be as orthogonal as possible to the tangent vector of the corresponding vertex on the boundary surface.

[0104] Optionally, sampling the skeleton representation to determine subregion thickness measurement points of each hippocampal subregion includes:

[0105] According to the layered distribution law of the hippocampus morphological model, the radial axis of the hippocampus morphological model passes through the skeleton surface of the skeleton representation, and the intersection point of the radial axis of the hippocampus morphological model and the skeleton surface is determined;

[0106] The sub-region thickness measurement point is determined according to the intersection point.

[0107] Specifically, the hippocampal morphological model includes the hippocampal axis reference morphological model (ARMM), the continuous internal measurement fitting model, and the discrete skeleton fitting model, etc. In this embodiment, the hippocampal axis reference morphological model is preferred. The skeleton surface of the ARMM contains a total of 549 measurement points, each of which corresponds to an upper radial axis and a lower radial axis, so there are 1098 ARMM radial axes. The 1098 ARMM radial axes sample the skeleton surface of each subregion respectively, and by locating the coordinates of the position where the ARMM radial axis passes through the subregion skeleton surface, and finding the subregion skeleton surface mesh vertex closest to the intersection as the subregion thickness measurement point, generally the radial axis vertex, the subregion measurement site is located. Since each subregion measurement site has only one ARMM radial axis corresponding to it, the subregion can establish a connection with the overall laminar structure of the hippocampus, retaining the better inter-individual morphological correspondence of ARMM, and greatly enhancing the comparability of morphological characteristics of hippocampal subregions across individuals. For example, the subregion thickness measurement points of each hippocampal subregion can be 130 for the CA1 subregion, 83 for the CA3 subregion, 92 for the CA4 subregion, 110 for the GC-ML-DG subregion, 95 for the HATA subregion, 201 for the molecular_layer subregion, 17 for the parasubiculum subregion, 102 for the presubiculum subregion, 60 for the subiculum subregion, and 87 for the tail subregion. For the subregion thickness measurement points of any subregion, , the sub-region thickness at the measurement point can be expressed as:

[0108] ;

[0109] Among them, Ti is the sub-area thickness measurement point Sub-area thickness at, sub-area thickness measurement point In this formula, it is represented by the coordinates corresponding to the point. Sub-area thickness measurement point The coordinates of the corresponding upper spoke vertex, Sub-area thickness measurement point The coordinates of the corresponding vertex of the inferior radial axis are shown in Figure 3. By integrating the subregional thickness measurement points of the 10 subregions with the optimized surface of the overall hippocampal boundary, a thickness measurement template was ultimately constructed. This template provides a total of 1954 bilateral hippocampal thickness features for subsequent cross-individual comparison of hippocampal subregional thickness.

[0110] Optionally, generating a target skeleton representation according to the target boundary optimization surface and the thickness measurement template includes:

[0111] Using a rigid registration strategy, the target boundary optimization surface is linearly transformed to the spatial position of the thickness measurement template;

[0112] Using a large-scale deformation differential homeomorphism metric mapping technique, the template boundary surface of the thickness measurement template is deformed to the target boundary optimization surface to generate a deformation field;

[0113] According to the deformation field, the sub-region thickness measurement points of the thickness measurement template are deformed to the spatial position of the target hippocampus to generate the target skeleton representation.

[0114] Specifically, to improve the deformation accuracy of the large-scale deformation diffeomorphism metric mapping technique, a rigid registration strategy is used to linearly transform the optimized boundary surfaces of the individual hippocampus and the target to the template location, that is, the spatial location of the thickness measurement template. This achieves spatial alignment between the individual hippocampus and the template hippocampus, thereby enabling the standardized measurement points of the template to be "mapped" to the corresponding spatial locations of the target individual, achieving spatial correspondence across individual morphologies. During this process, the subregion boundary surfaces should be transformed along with the overall boundary surface. If each subregion boundary surface is transformed separately to the subregion boundary surface of the thickness measurement template, spatial correspondence between the subregions and the overall hippocampus after rigid registration cannot be achieved. In other words, significant gaps remain between the boundary surfaces of the hippocampal subregions, preventing the accurate formation of the overall hippocampus. Furthermore, the boundary surface of the thickness measurement template (that is, the optimized boundary surface of the standard magnetic resonance image template) is deformed to the optimized boundary surface of the target hippocampus using the large-scale deformation diffeomorphism metric mapping technique. Based on this deformation field, each subregion thickness measurement point within the template is deformed to the spatial location of the target hippocampus. Among them, the large-scale deformation differential homeomorphism metric mapping technology is implemented through the integrated software deformetrica. By adjusting the Gaussian kernel width of the model, different degrees of deformation can be achieved. When applied to the hippocampus, this parameter is set to 3~20 according to experience, and it is iterated from low to high. By calculating the average distance between the generated surface and the target surface after each deformation, until it is less than the preset threshold, the deformation effect closest to the target boundary is obtained. In addition, dense mode is an important parameter in the deformetrica deformation process. This mode only supports accurate deformation under single surface input. If multiple hippocampus sub-region boundary surfaces are input at the same time, deformation is difficult to achieve regardless of the number of Gaussian kernel iterations. This also corresponds to why the thickness measurement template only contains the skeleton points and radial axis end points of the sub-region, but not the boundary surface of the template sub-region.

[0115] Optionally, determining the morphological features of the target hippocampus according to the positions of the subregion skeleton points in the target skeleton representation includes:

[0116] When the sub-region skeleton point is located inside the target boundary optimization surface, optimizing the target radial axis vector in the target skeleton representation, and determining the sub-region thickness according to the optimized target radial axis vector;

[0117] When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is greater than or equal to two, the sub-region skeleton point and its corresponding radial axis end point are updated, and the sub-region thickness is determined based on the updated sub-region skeleton point and its corresponding radial axis end point, wherein the skeleton radial axis line is a line connecting the sub-region skeleton point and its corresponding radial axis end point;

[0118] When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is less than two, the sub-region thickness is set to zero.

[0119] Specifically, the morphological features include the thickness of the sub-region. After deformation using the large-scale deformation differential homeomorphism metric mapping technology, there are three situations: In the first case, the deformed sub-region skeleton point is located inside the sub-region boundary surface. At this time, the skeleton point is a valid measurement point, but the deformation may cause the end point of the radial axis to be exposed on the sub-region boundary surface, the radial axis direction and the tangent plane of the nearest point on the boundary surface to be non-perpendicular, or the radial axes to intersect with each other. For these cases, the first penalty function, the second penalty function, and the third penalty function are used for optimization. The optimized radial axis and skeleton point are calculated using the above-mentioned thickness formula or the thickness formula of the first case; In the second case, the sub-region skeleton point is located outside the target boundary optimization surface, and the intersection of the skeleton radial axis line and the target boundary optimization surface is greater than or equal to two. At this time, the original skeleton point and the radial axis end point are invalidated, and the direction from the original skeleton point to the original radial axis end point is taken. The first intersection point is used as the updated skeleton point, and the second intersection point is used as the updated radial axis end point, that is, the length of the radial axis passing through the boundary surface is taken as the thickness of the measurement point, or the thickness formula of the second case is used to calculate the thickness. When the number of intersection points is greater than two, it means that the boundary surface near the skeleton point is layered, and only the boundary surface layer closest to the first intersection point and the second intersection point is measured. The length of the radial axis passing through the boundary surface can be taken as the thickness of the measurement point, or the thickness formula of the second case is used to calculate the thickness; in the third case, the sub-region skeleton point is located outside the target boundary optimization surface, and the intersection points of the skeleton radial axis line and the target boundary optimization surface are less than two. At this time, the radial axis is tangent to or separated from the boundary surface, and the skeleton point is a waste measurement point. The coordinates of the end points of the radial axes on both sides are updated to the coordinates of the skeleton point, the radial axis length is reset to zero, and the thickness is set to zero here. For example, the thickness formula of the first case includes:

[0120] ;

[0121] in, is the thickness calculated in the first case, the sub-area thickness measurement point In this formula, it is represented by the coordinates corresponding to the point. Sub-area thickness measurement point The coordinates of the corresponding upper spoke vertex, Sub-area thickness measurement point The coordinates of the corresponding lower spoke vertex.

[0122] The second case thickness formula includes:

[0123] ;

[0124] in, is the thickness calculated in the second case, and They are and The upper and lower spokes are located, Optimize the surface for the boundary, The spoke axis and The coordinates of the first intersection point.

[0125] like Figure 8 As shown, an embodiment of the present invention provides a hippocampal substructure morphology measurement device, comprising:

[0126] Segmentation module, used to segment the magnetic resonance standard template image, construct the boundary optimization surface, and determine the skeleton representation;

[0127] a template module for sampling the skeleton representation, determining subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of the template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation;

[0128] a mapping module for determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template;

[0129] A morphological module is used to determine the morphological characteristics of the target hippocampus according to the positions of the sub-region skeleton points in the target skeleton representation.

[0130] like Figure 9 As shown, an electronic device 900 provided by an embodiment of the present invention includes a memory 910 and a processor 920; the memory 910 is used to store a computer program; the processor 920 is used to implement the above-mentioned hippocampal substructure morphology measurement method when executing the computer program.

[0131] In other words, an electronic device 900 includes a memory 910 and a processor 920 coupled to the memory 910; the memory 910 is configured to store a computer program; and the processor 920 is configured to perform the following operations when executing the computer program:

[0132] Segment the magnetic resonance standard template image, construct the boundary optimization surface, and determine the skeleton representation;

[0133] Sampling the skeleton representation to determine subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of the template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation;

[0134] determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template;

[0135] The morphological features of the target hippocampus are determined based on the positions of the subregion skeleton points in the target skeleton representation.

[0136] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned hippocampal substructure morphology measurement method is implemented.

[0137] In other words, a non-volatile computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to perform the following operations:

[0138] Segment the magnetic resonance standard template image, construct the boundary optimization surface, and determine the skeleton representation;

[0139] Sampling the skeleton representation to determine subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of the template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation;

[0140] determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template;

[0141] The morphological features of the target hippocampus are determined based on the positions of the subregion skeleton points in the target skeleton representation.

[0142] An electronic device 900 that can serve as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device 900 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 900 can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0143] Electronic device 900 includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). The RAM can also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. An input / output (I / O) interface is also connected to the bus.

[0144] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM). In this application, the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network elements. Some or all of these units can be selected based on actual needs to achieve the objectives of the embodiments of the present invention. Furthermore, the functional units in the various embodiments of the present invention can be integrated into a single processing unit, each unit can exist physically separately, or two or more units can be integrated into a single unit. These integrated units can be implemented in either hardware or software functional units.

[0145] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A method for measuring the morphology of hippocampal substructures, characterized in that: include: Segment the magnetic resonance standard template image, construct the boundary optimization surface, and determine the skeleton representation; Sampling the skeleton representation to determine subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of the template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation; determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template; determining morphological features of the target hippocampus based on positions of subregion skeleton points in the target skeleton representation; The skeleton representation includes a skeleton surface and a radial axis vector, the skeleton point is located on the skeleton surface, and the radial axis vector is determined based on the skeleton surface according to a restriction criterion and a penalty function; Determining the morphological features of the target hippocampus according to the positions of the subregion skeleton points in the target skeleton representation includes: When the sub-region skeleton point is located inside the target boundary optimization surface, optimizing the target radial axis vector in the target skeleton representation, and determining the sub-region thickness according to the optimized target radial axis vector; When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is greater than or equal to two, the sub-region skeleton point and its corresponding radial axis end point are updated, and the sub-region thickness is determined based on the updated sub-region skeleton point and its corresponding radial axis end point, wherein the skeleton radial axis line is a line connecting the sub-region skeleton point and its corresponding radial axis end point; When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is less than two, the sub-region thickness is set to zero.

2. The hippocampal substructure morphology measurement method according to claim 1, characterized in that: The method of segmenting the magnetic resonance standard template image, constructing the boundary optimization surface, and determining the skeleton representation includes: Segmenting the magnetic resonance standard template image to generate segmentation label images of each hippocampal subregion within the template hippocampus; generating the boundary optimization surface according to the segmented label image; The boundary optimization surface is converted into a Voronoi diagram and fitted to generate the skeleton representation.

3. The method for measuring the morphology of hippocampal substructures according to claim 2, wherein: The step of converting the boundary optimization surface into a Voronoi diagram and performing fitting to generate a skeleton representation includes: Converting the boundary optimization surface into the Voronoi diagram, retaining the Voronoi vertices in the Voronoi diagram that are within the boundary optimization surface, and constructing a point cloud; Fitting the point cloud to generate the skeleton surface; Based on the skeleton face, the spoke axis vector is determined according to a restriction criterion and a penalty function.

4. The method for measuring the morphology of hippocampal substructures according to claim 3, wherein: The step of fitting the point cloud to generate the skeleton surface includes: dividing the hippocampal subregions in the template hippocampus into regular hippocampal subregions and complex hippocampal subregions; Fitting the point cloud of the regular hippocampal subregion to generate a skeleton surface of the regular hippocampal subregion; The complex hippocampal subregion is cut into at least two complex sub-subregions, the point cloud of each complex sub-subregion is fitted to generate a sub-subregion skeleton surface of the complex sub-subregion, and each sub-subregion skeleton surface of the complex sub-subregion is merged to generate a skeleton surface of the complex hippocampal subregion.

5. The method for measuring the morphology of hippocampal substructures according to claim 1, wherein: Sampling the skeleton representation to determine subregion thickness measurement points of each hippocampal subregion includes: According to the layered distribution law of the hippocampus morphological model, the radial axis of the hippocampus morphological model passes through the skeleton surface of the skeleton representation, and the intersection point of the radial axis of the hippocampus morphological model and the skeleton surface is determined; The sub-region thickness measurement point is determined according to the intersection point.

6. The method for measuring the morphology of hippocampal substructures according to claim 1, wherein: The morphological characteristics include at least one of hippocampal surface parameters, hippocampal thickness parameters, hippocampal length parameters, hippocampal long axis parameters and subregion thickness.

7. A device for measuring the morphology of hippocampal substructures, characterized in that: include: Segmentation module, used to segment the magnetic resonance standard template image, construct the boundary optimization surface, and determine the skeleton representation; a template module for sampling the skeleton representation, determining subregion thickness measurement points of each hippocampal subregion, and constructing a thickness measurement template of a template hippocampus based on the subregion thickness measurement points and the skeleton points in the skeleton representation; a mapping module for determining a target boundary optimization surface corresponding to a target magnetic resonance image of a target hippocampus, and generating a target skeleton representation based on the target boundary optimization surface and the thickness measurement template; a morphological module for determining morphological features of the target hippocampus based on positions of subregion skeleton points in the target skeleton representation; The skeleton representation includes a skeleton surface and a radial axis vector, the skeleton point is located on the skeleton surface, and the radial axis vector is determined based on the skeleton surface according to a restriction criterion and a penalty function; Determining the morphological features of the target hippocampus according to the positions of the subregion skeleton points in the target skeleton representation includes: When the sub-region skeleton point is located inside the target boundary optimization surface, optimizing the target radial axis vector in the target skeleton representation, and determining the sub-region thickness according to the optimized target radial axis vector; When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is greater than or equal to two, the sub-region skeleton point and its corresponding radial axis end point are updated, and the sub-region thickness is determined based on the updated sub-region skeleton point and its corresponding radial axis end point, wherein the skeleton radial axis line is a line connecting the sub-region skeleton point and its corresponding radial axis end point; When the sub-region skeleton point is located outside the target boundary optimization surface, and the number of intersections between the skeleton radial axis line and the target boundary optimization surface is less than two, the sub-region thickness is set to zero.

8. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the hippocampal substructure morphology measurement method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method for measuring the morphology of hippocampal substructures according to any one of claims 1 to 6 is implemented.

Images