CN110930389B - Region segmentation method, device and storage medium for three-dimensional medical model data - Google Patents
Region segmentation method, device and storage medium for three-dimensional medical model data Download PDFInfo
- Publication number
- CN110930389B CN110930389B CN201911156860.3A CN201911156860A CN110930389B CN 110930389 B CN110930389 B CN 110930389B CN 201911156860 A CN201911156860 A CN 201911156860A CN 110930389 B CN110930389 B CN 110930389B
- Authority
- CN
- China
- Prior art keywords
- model data
- graph
- dimensional medical
- algorithm
- sampling
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Abstract
The invention provides a region segmentation method, a device and a storage medium of three-dimensional medical model data, wherein the method comprises the following steps: acquiring a medical image of a patient, and carrying out three-dimensional reconstruction on the medical image through an isosurface algorithm to obtain three-dimensional medical grid model data; converting the three-dimensional medical grid model data into a corresponding dual graph form, and setting the side weight of the graph according to the distance between triangular patches on the three-dimensional medical grid model data; sampling the approximate farthest points on the dual graph to obtain a plurality of seed points; a clustering algorithm based on pruning is operated on the dual graph, and the area controlled by each seed point is determined; and merging adjacent classes one by one according to the minimum merging cost principle, and determining a segmentation result according to the residual class number selected by the user. The method is quicker and more efficient, has accurate segmentation effect, is simple and convenient to operate, has simple equipment and low cost, and has higher practical value.
Description
Technical Field
The invention relates to the technical field of digital medical treatment, in particular to a region segmentation method of three-dimensional medical model data, a region segmentation device of the three-dimensional medical model data and a non-transitory computer readable storage medium.
Background
During a surgical planning procedure, a physician typically needs to carefully and omnidirectionally view certain tissue and organs of a patient in order to design an accurate surgical plan. For images obtained by high-precision electronic computer tomography or nuclear magnetic resonance technology, the three-dimensional structure of the tissue and organ of interest can be reconstructed according to the isosurface, or a volume rendering algorithm can be used for presenting the region which the user wants to observe. However, none of these approaches allow the physician to view all-round the substructures in a tissue organ, and thus cannot design an accurate surgical plan.
In the field of three-dimensional medical model data region segmentation, modes such as clustering, random walk, basic shape fitting, graph cutting and the like are mainly available at present. Firstly, a user is required to input a cluster number based on a clustering mode, then, a plurality of seed patches are selected, patches contained in each cluster are calculated iteratively, and cluster center patches are recalculated until a result is converged; firstly, calculating which cluster each patch belongs to according to a random walk mode, and obtaining an excessive result; the second part is subjected to hierarchical combination and small clustering so as to obtain a final segmentation result; the hierarchical clustering algorithm based on basic shape matching considers all the patches as a single cluster at first, and then performs hierarchical clustering by judging the matching degree of the combined patches and some basic shapes until the number of clusters reaches a target; the graph cut-based region segmentation algorithm selects an optimal cut from the random minimum cuts to determine a segmentation boundary, thereby dividing the mesh data into several parts. However, these approaches all have the common problem of being sensitive to noise and being less efficient, so the above-described techniques are difficult to apply directly in clinic.
Disclosure of Invention
The invention provides a region segmentation method, a device and a storage medium for three-dimensional medical model data, which are faster and more efficient, have accurate segmentation effect, are simple and convenient to operate, have simple equipment, have low cost and have higher practical value.
The technical scheme adopted by the invention is as follows:
a region segmentation method of three-dimensional medical model data, comprising the steps of: acquiring a medical image of a patient, and carrying out three-dimensional reconstruction on the medical image through an isosurface algorithm to obtain three-dimensional medical grid model data; converting the three-dimensional medical grid model data into a corresponding dual graph form, and setting the side weight of the graph according to the distance between triangular patches on the three-dimensional medical grid model data; sampling the approximate farthest points on the dual graph to obtain a plurality of seed points; operating a pruning-based clustering algorithm on the dual graph, and determining a region controlled by each seed point; and merging adjacent classes one by one according to the minimum merging cost principle, and determining a segmentation result according to the residual class number selected by the user.
The medical image is an electronic computed tomography image or a nuclear magnetic resonance image.
Performing three-dimensional reconstruction on the medical image through an isosurface algorithm, wherein the obtaining of the three-dimensional medical grid model data specifically comprises the following steps: and reconstructing the electronic computer tomography image volume data or the nuclear magnetic resonance image volume data into tissue and organs of interest to doctors by using a Marching cubes algorithm, and presenting the tissue and organs in a three-dimensional grid data form.
The side weights of the distance setting map between the triangular patches on the three-dimensional medical grid model data specifically comprise: and taking each triangular surface patch on the three-dimensional medical grid model data as a vertex on the graph, connecting a corresponding vertex between two adjacent triangular surface patches with an edge, and representing the edge weight of the graph by weighted summation of the physical distance and the angular distance.
Performing near furthest point sampling on the dual graph to obtain a plurality of seed points specifically comprises: dividing the three-dimensional medical grid model data into a plurality of sub-blocks, converting each sub-block into a small undirected graph through connectivity, obtaining the distance between the sub-blocks by running Dijkstra algorithm on the small undirected graph, obtaining sampling blocks by using furthest point sampling algorithm on the small undirected graph consisting of the sub-blocks, and randomly sampling points in the sampling blocks to obtain an approximate furthest seed point set by sampling.
Running a pruning-based clustering algorithm on the dual graph, wherein the determining of the area controlled by each seed point specifically comprises the following steps: and pruning the Dijkstra algorithm on the dual graph by taking the seed point set as a center and utilizing a geodesic distance pruning method based on the undirected graph to obtain an over-segmentation result.
Merging adjacent classes one by one according to the minimum merging cost principle, and determining a segmentation result according to the residual class number selected by a user specifically comprises: and defining merging cost for every two adjacent clusters according to the over-dividing result, storing all the merging cost into a minimum stack, merging only two parts with the minimum cost each time, updating the minimum stack, and the like to obtain a final region dividing result.
An area segmentation apparatus of three-dimensional medical model data, comprising: the modeling module is used for acquiring a medical image of a patient, and carrying out three-dimensional reconstruction on the medical image through an isosurface algorithm to obtain three-dimensional medical grid model data; the conversion module is used for converting the three-dimensional medical grid model data into a corresponding dual graph form and setting the side weight of the graph according to the distance between triangular patches on the three-dimensional medical grid model data; the sampling module is used for performing approximate furthest point sampling on the dual graph to obtain a plurality of seed points; the first determining module is used for running a pruning-based clustering algorithm on the dual graph and determining the area controlled by each seed point; and the second determining module is used for merging the adjacent classes one by one according to the minimum merging cost principle and determining a segmentation result according to the residual class number selected by the user.
The medical image is an electronic computed tomography image or a nuclear magnetic resonance image.
A non-transitory computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the region segmentation method of three-dimensional medical model data described above.
The invention has the beneficial effects that:
according to the invention, the three-dimensional medical grid model data are converted into the dual graph form, and the sub-structures of the three-dimensional medical grid model data are obtained by using the approximate furthest point sampling and the pruning-based clustering algorithm, so that the method is more rapid and efficient compared with other segmentation methods, the segmentation effect is accurate, the flow of the method is full-automatic, the number of the wanted segmentation areas is only required to be adjusted after the algorithm is operated, and less user interaction is required, so that the operation is simple and convenient, the used equipment is simple, the cost is low, and the method has higher practical value.
Drawings
FIG. 1 is a flow chart of a method for region segmentation of three-dimensional medical model data according to an embodiment of the present invention;
fig. 2 is a block diagram of a region segmentation apparatus for three-dimensional medical model data according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, the region segmentation method of three-dimensional medical model data according to the embodiment of the invention includes the following steps:
s1, acquiring a medical image of a patient, and performing three-dimensional reconstruction on the medical image through an isosurface algorithm to obtain three-dimensional medical grid model data.
In one embodiment of the invention, the medical image is a high resolution computerized tomography image or a nuclear magnetic resonance image.
After the electronic computer tomography image or the nuclear magnetic resonance image of the patient is obtained, the electronic computer tomography image volume data or the nuclear magnetic resonance image volume data can be used for extracting an isosurface by using a Maring cube algorithm, and the isosurface is reconstructed into tissues and organs of interest to doctors in a three-dimensional mode and is presented in a three-dimensional grid data mode.
S2, converting the three-dimensional medical grid model data into a corresponding dual graph form, and setting the side weight of the graph according to the distance between the triangular patches on the three-dimensional medical grid model data.
In the embodiment of the invention, the three-dimensional medical grid model data region segmentation problem can be abstracted into a graph partitioning problem, so that the three-dimensional medical grid model data M can be firstly converted into the form G (M) = { V of the dual graph M ,E M }. And taking each triangular surface patch on the three-dimensional medical grid model data as a vertex on the graph, connecting a side with the corresponding vertex between two adjacent triangular surface patches, and representing the side weight of the graph by weighted summation of the physical distance and the angular distance. Namely, the edge weight is divided into two terms of a physical distance and an angular distance, and can be defined according to the following expression:
wherein avg (·) represents the mean function, δ controls the weights of both terms. The physical distance PhyDist between two adjacent triangular patches represents the sum of the perpendicular distances from their centers to the common edge, while the angular distance AngDist is determined by the dihedral angle between them, expressed as follows:
AngDist(V i ，V j )＝η·(1-α ij )
where η varies according to the convexity of the dihedral angle, the value of η is much smaller when the two panels are convex than when the two panels are concave. This may make the algorithm more sensitive to the recessed portions of the model, thereby making the algorithm more accurate.
And S3, performing approximate furthest point sampling on the dual graph to obtain a plurality of seed points.
Since the region segmentation method of the embodiment of the invention depends on the distribution of sampling points on the graph, a set of points which are far away from each other on the three-dimensional medical grid model data are obtained as seed points in as short a time as possible. Specifically, the three-dimensional medical grid model data can be firstly split into a plurality of sub-blocks according to an octree, each sub-block is then converted into a small undirected graph through connectivity, and the distance between every two sub-blocks is obtained by running Dijkstra algorithm on the small undirected graph. After the connectivity and the geodesic distance between the sub-blocks are obtained, sampling blocks can be obtained by using the furthest point sampling algorithm on a small undirected graph consisting of the sub-blocks, and then sampling points are randomly sampled in the sampling blocks, so that an approximate furthest seed point set can be obtained in an extremely short time.
And S4, running a pruning-based clustering algorithm on the dual graph, and determining the area controlled by each seed point.
And pruning the Dijkstra algorithm by using the geodesic distance pruning method based on the undirected graph on the dual graph with the seed point set as the center to obtain an excessive result.
Specifically, according to the seed point set calculated in step S3, for each vertex on the graph G (M), the seed point closest to the vertex is calculated, so as to complete the clustering process. The original clustering requires that Dijkstra algorithm (time complexity O (nlogn)) be run once on the entire graph for each seed point, which is not available for data with a common level of millions or even tens of millions of top points.
In fact, for a single seed point, it is not necessary to calculate a single source shortest for the entire graph. A vertex that is too far from the seed point is practically meaningless for the seed point because it does not belong to the class represented by this seed point. According to the conditions, the embodiment of the invention adopts a zoning clustering algorithm based on the geodesic distance, which is a Dijkstra single-source shortest algorithm variant based on a pruning strategy. Formally, in the undirected weighted graph G M ＝{V M ,E M There are k non-repeating vertices (k < n) on }, and define the seed point set as S. Definition of the slave seed points s i To vertex v k Distance d of (2) ik . Given seed point s i If there are other seed points s j Satisfy d ik >d jk Then call v k For s i Is unnecessary. Then consider the seed point s i Is an unnecessary vertex v k For meeting condition d ik′ ＝d ik +w kk′ >d jk +w kk′ ≥d jk′ Neighborhood vertex v of (v) k′ ∈N(v k ) The following formula holds:
d ik′ ＝d ik +w kk′ ＞d jk +w kk′ ≥d jk′
that is, v k′ Is s i Is not necessary.
Based on the theorem, a considerable amount of peaks which do not need to be calculated at all can be considered less, so that the overall efficiency of the algorithm can be improved. Specifically, when the Dijkstra algorithm extracts the nearest vertex from the minimum heap (fibonacci heap) each time, judging whether the distance between the current seed point and the vertex is a preset value (infinity), and if so, ending the cycle; judging whether the precursor of the vertex is necessary relative to the current seed point, and skipping the vertex if not; otherwise, judging whether the vertex is necessary or not, and if not, skipping the vertex; otherwise the same as the original Dijkstra algorithm. It can be seen that the geodesic nearest class assignment algorithm calculates the shortest distance of the seed point relative to the whole graph when the loop is executed for the first time, and the calculated area gradually decreases each time as the seed point increases.
S5, merging adjacent classes one by one according to the minimum merging cost principle, and determining a segmentation result according to the number of the residual classes selected by the user.
Specifically, for the over-segmentation result obtained in step S4, merging costs are defined for every two adjacent clusters, all merging costs are stored in a minimum heap, only two parts with the minimum merging costs are merged each time, then the minimum heap is updated, and the final region segmentation result is obtained by analogy.
The definition of the merging cost is as follows: for two adjacent portions S i ,S j Definition of
Is S i And S is equal to j Boundary of adjacent parts->
Is S i And S is equal to j The merged boundary. Obviously, these boundaries are all spliced from edges of one or more panels. Each of the above-mentioned boundaries is connected with two panels belonging to different parts, then +.>
I.e. the length of the boundary. For->
The definition may be similarly performed. Then S i ,S j The merge cost of (2) may be defined as:
according to the region segmentation method of the three-dimensional medical model data, the three-dimensional medical grid model data are converted into the dual graph form, the substructures of the three-dimensional medical grid model data are obtained through the approximate furthest point sampling and the pruning-based clustering algorithm, compared with other segmentation methods, the method is more rapid and efficient, the segmentation effect is accurate, the flow of the method is full-automatic, the number of the wanted segmentation regions is only required to be adjusted after the algorithm is operated, and the user interaction is less, so that the operation is simple and convenient, the used equipment is simple, the cost is low, and the practical value is higher.
In order to realize the region segmentation method of the three-dimensional medical model data in the embodiment, the invention also provides a region segmentation device of the three-dimensional medical model data.
As shown in fig. 2, the region segmentation apparatus for three-dimensional medical model data according to an embodiment of the present invention includes a modeling module 10, a conversion module 20, a sampling module 30, a first determination module 40, and a second determination module 50. The modeling module 10 is used for acquiring a medical image of a patient, and performing three-dimensional reconstruction on the medical image through an isosurface algorithm to obtain three-dimensional medical grid model data; the conversion module 20 is used for converting the three-dimensional medical grid model data into a corresponding dual graph form, and setting the side weight of the graph according to the distance between the triangular patches on the three-dimensional medical grid model data; the sampling module 30 is configured to perform near furthest point sampling on the dual graph to obtain a plurality of seed points; the first determining module 40 is configured to run a pruning-based clustering algorithm on the dual graph, and determine a region controlled by each seed point; the second determining module 50 is configured to merge the adjacent classes one by one according to a minimum merging cost principle, and determine a segmentation result according to the number of remaining classes selected by the user.
In one embodiment of the invention, the medical image is a high resolution computerized tomography image or a nuclear magnetic resonance image.
After acquiring an electronic computed tomography image or a nuclear magnetic resonance image of the patient, the modeling module 10 may extract the iso-surface from the electronic computed tomography image volume data or the nuclear magnetic resonance image volume data by using a Marching cube algorithm, reconstruct the iso-surface into a tissue organ of interest to the doctor in three dimensions, and present the tissue organ in the form of three-dimensional grid data.
In an embodiment of the present invention, the three-dimensional medical mesh model data region segmentation problem can be abstracted into a graph partitioning problem, so the conversion module 20 can first convert the three-dimensional medical mesh model data M into the form G (M) = { W of its dual graph M ,E M And then, each triangular surface patch on the three-dimensional medical grid model data is regarded as a vertex on the graph, the corresponding vertex between two adjacent triangular surface patches is connected with an edge, and the edge weight of the graph is represented by weighted summation of the physical distance and the angular distance. Namely, the edge weight is divided into two terms of a physical distance and an angular distance, and can be defined according to the following expression:
wherein avg (·) represents the mean function, δ controls the weights of both terms. The physical distance PhyDist between two adjacent triangular patches represents the sum of the perpendicular distances from their centers to the common edge, while the angular distance AngDist is determined by the dihedral angle between them, expressed as follows:
AngDist(V i ，V j )＝η·(1-α ij )
where η varies according to the convexity of the dihedral angle, the value of η is much smaller when the two panels are convex than when the two panels are concave. This may make the algorithm more sensitive to the recessed portions of the model, thereby making the algorithm more accurate.
Since the region segmentation method of the embodiment of the invention depends on the distribution of sampling points on the graph, a set of points which are far away from each other on the three-dimensional medical grid model data are obtained as seed points in as short a time as possible. Specifically, the sampling module 30 may first divide the three-dimensional medical mesh model data into a plurality of sub-blocks according to an octree, then convert each sub-block into a small undirected graph through connectivity, and obtain the distance between every two sub-blocks by running Dijkstra algorithm on the small undirected graph. After obtaining connectivity and geodesic distance between sub-blocks, the sampling module 30 may obtain sampling blocks by using the furthest point sampling algorithm on a small undirected graph composed of sub-blocks, and then randomly sampling points in the sampling blocks, so as to obtain an approximately furthest seed point set in an extremely short time.
The first determining module 40 may perform pruning operation on the Dijkstra algorithm on the dual graph with the set of seed points as a center by using a geodesic distance pruning method based on the undirected graph, to obtain the over-segmentation result.
Specifically, the first determining module 40 may calculate, for each vertex on the graph G (M), a seed point closest to the vertex according to the seed point set obtained by the sampling module 30, so as to complete the clustering process. The original clustering requires that Dijkstra's algorithm (O (n log n) in time complexity) be run once on the entire graph for each seed point, which is not available for data with top points on the order of millions or even tens of millions.
In fact, for a single seed point, it is not necessary to calculate a single source shortest for the entire graph. A vertex that is too far from the seed point is practically meaningless for the seed point because it does not belong to the class represented by this seed point. According to the conditions, the embodiment of the invention adopts a zoning clustering algorithm based on the geodesic distance, which is a Dijkstra single-source shortest algorithm variant based on a pruning strategy. Formally, in the undirected weighted graph G M ＝{V M ，E M There are k non-repeating vertices (k < n) on }, and define the seed point set as S. Definition of the slave seed points s i To vertex v k Distance d of (2) ik . Given seed point s i If there are other seed points s j Satisfy d ik ＞d jk Then call v k For s i Is unnecessary. Then consider the seed point s i Is an unnecessary vertex v k For meeting condition d ik′ ＝d ik +w kk′ ＞d jk +w kk′ D jk′ Neighborhood vertex v of (v) k′ ∈N(v k ) The following formula holds:
d ik′ ＝d ik +w kk′ ＞d jk +w kk′ d jk′
That is, v k′ Is s i Is not necessary.
Based on the theorem, a considerable amount of peaks which do not need to be calculated at all can be considered less, so that the overall efficiency of the algorithm can be improved. Specifically, the first determining module 40 may determine whether the distance between the current seed point and the vertex is a preset value (infinity) each time the Dijkstra algorithm extracts the nearest vertex from the minimum heap (fibonacci heap), and if so, the loop ends; judging whether the precursor of the vertex is necessary relative to the current seed point, and skipping the vertex if not; otherwise, judging whether the vertex is necessary or not, and if not, skipping the vertex; otherwise the same as the original Dijkstra algorithm. It can be seen that the geodesic nearest class assignment algorithm calculates the shortest distance of the seed point relative to the whole graph when the loop is executed for the first time, and the calculated area gradually decreases each time as the seed point increases.
For the over-segmentation result obtained by the first determining module 40, the second determining module 50 may define merging costs for every two adjacent clusters, store all merging costs into a minimum heap, merge only two parts with the minimum cost each time, then update the minimum heap, and so on, to obtain a final region segmentation result.
The definition of the merging cost is as follows: for two adjacent portions S i ,S j Definition of
Is S i And S is equal to j Boundary of adjacent parts->
Is S i And S is equal to j The merged boundary. Obviously, these boundaries are all spliced by edges of one or more patchesAnd is made up of the steps of. Each of the above-mentioned boundaries is connected with two panels belonging to different parts, then +.>
I.e. the length of the boundary. For->
The definition may be similarly performed. Then S i ,S j The merge cost of (2) may be defined as:
according to the region segmentation device for the three-dimensional medical model data, the three-dimensional medical grid model data are converted into the dual graph form through the conversion module, the sampling module and the determining module calculate and obtain each substructure of the three-dimensional medical grid model data by using the approximate farthest point sampling and pruning-based clustering algorithm, compared with other segmentation devices, the region segmentation device is more rapid and efficient, the segmentation effect is accurate, the flow executed by each module of the device is fully automatic, only the number of the needed segmentation regions is required to be adjusted after the algorithm operation is completed, and the user interaction is less, so that the device is simple to operate, low in cost and high in practical value.
The present invention also proposes a non-transitory computer-readable storage medium corresponding to the above-described embodiments.
The non-transitory computer-readable storage medium of the embodiment of the present invention stores thereon a computer program that, when executed by a processor, can implement the region segmentation method of three-dimensional medical model data of any of the above embodiments.
According to the non-transitory computer readable storage medium, the stored computer program is executed by the processor, so that the method has the advantages of rapider and more efficient region segmentation of the three-dimensional medical model data, accurate segmentation effect, simple and convenient operation, simple equipment, low cost and higher practical value.
In the description of the present invention, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. The meaning of "a plurality of" is two or more, unless specifically defined otherwise.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (8)
1. A method for region segmentation of three-dimensional medical model data, comprising the steps of:
acquiring a medical image of a patient, and carrying out three-dimensional reconstruction on the medical image through an isosurface algorithm to obtain three-dimensional medical grid model data;
converting the three-dimensional medical grid model data into a corresponding dual graph form, and setting the side weight of the graph according to the distance between triangular patches on the three-dimensional medical grid model data;
performing approximate farthest point sampling on the dual graph to obtain a plurality of seed points: dividing the three-dimensional medical grid model data into a plurality of sub-blocks, converting each sub-block into a small undirected graph through connectivity, obtaining the distance between the sub-blocks by running Dijkstra algorithm on the small undirected graph, obtaining sampling blocks by using furthest point sampling algorithm on the small undirected graph consisting of the sub-blocks, and randomly sampling points in the sampling blocks to obtain an approximate furthest seed point set by sampling;
running a pruning-based clustering algorithm on the dual graph, and determining the area controlled by each seed point: carrying out pruning operation on the Dijkstra algorithm on the dual graph by taking the seed point set as a center and utilizing a geodesic distance pruning method based on an undirected graph to obtain an over-segmentation result;
and merging adjacent classes one by one according to the minimum merging cost principle, and determining a segmentation result according to the residual class number selected by the user.
2. The method of claim 1, wherein the medical image is an electronic computed tomography image or a nuclear magnetic resonance image.
3. The method for region segmentation of three-dimensional medical model data according to claim 2, wherein the three-dimensional reconstruction of the medical image by an isosurface algorithm, the obtaining of three-dimensional medical mesh model data specifically comprises:
and reconstructing the electronic computer tomography image volume data or the nuclear magnetic resonance image volume data into tissue and organs of interest to doctors by using a Marching cubes algorithm, and presenting the tissue and organs in a three-dimensional grid data form.
4. The method for region segmentation of three-dimensional medical model data according to claim 3, wherein the side weights of the set-up graph according to the distance between the triangular patches on the three-dimensional medical mesh model data specifically comprises:
and taking each triangular surface patch on the three-dimensional medical grid model data as a vertex on the graph, connecting a corresponding vertex between two adjacent triangular surface patches with an edge, and representing the edge weight of the graph by weighted summation of the physical distance and the angular distance.
5. The method for region segmentation of three-dimensional medical model data according to claim 4, wherein merging adjacent classes one by one according to a minimum merging cost principle, and determining a segmentation result according to the number of remaining classes selected by a user specifically comprises:
and defining merging cost for every two adjacent clusters according to the over-dividing result, storing all the merging cost into a minimum stack, merging only two parts with the minimum cost each time, updating the minimum stack, and the like to obtain a final region dividing result.
6. A region segmentation apparatus for three-dimensional medical model data, comprising:
the modeling module is used for acquiring a medical image of a patient, and carrying out three-dimensional reconstruction on the medical image through an isosurface algorithm to obtain three-dimensional medical grid model data;
the conversion module is used for converting the three-dimensional medical grid model data into a corresponding dual graph form and setting the side weight of the graph according to the distance between triangular patches on the three-dimensional medical grid model data;
the sampling module is used for performing approximate furthest point sampling on the dual graph to obtain a plurality of seed points: dividing the three-dimensional medical grid model data into a plurality of sub-blocks, converting each sub-block into a small undirected graph through connectivity, obtaining the distance between the sub-blocks by running Dijkstra algorithm on the small undirected graph, obtaining sampling blocks by using furthest point sampling algorithm on the small undirected graph consisting of the sub-blocks, and randomly sampling points in the sampling blocks to obtain an approximate furthest seed point set by sampling;
the first determining module is used for running a pruning-based clustering algorithm on the dual graph and determining the area controlled by each seed point: carrying out pruning operation on the Dijkstra algorithm on the dual graph by taking the seed point set as a center and utilizing a geodesic distance pruning method based on an undirected graph to obtain an over-segmentation result;
and the second determining module is used for merging the adjacent classes one by one according to the minimum merging cost principle and determining a segmentation result according to the residual class number selected by the user.
7. The region segmentation apparatus in accordance with claim 6, wherein the medical image is an electronic computed tomography image or a nuclear magnetic resonance image.
8. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that the program, when executed by a processor, implements a region segmentation method of three-dimensional medical model data according to any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911156860.3A CN110930389B (en) | 2019-11-22 | 2019-11-22 | Region segmentation method, device and storage medium for three-dimensional medical model data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911156860.3A CN110930389B (en) | 2019-11-22 | 2019-11-22 | Region segmentation method, device and storage medium for three-dimensional medical model data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110930389A CN110930389A (en) | 2020-03-27 |
| CN110930389B true CN110930389B (en) | 2023-06-30 |
Family
ID=69850750
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911156860.3A Active CN110930389B (en) | 2019-11-22 | 2019-11-22 | Region segmentation method, device and storage medium for three-dimensional medical model data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110930389B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106373168A (en) * | 2016-11-24 | 2017-02-01 | 北京三体高创科技有限公司 | Medical image based segmentation and 3D reconstruction method and 3D printing system |
| CN109064471A (en) * | 2018-07-18 | 2018-12-21 | 中北大学 | A kind of three-dimensional point cloud model dividing method based on skeleton |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9978150B2 (en) * | 2015-08-05 | 2018-05-22 | Algotec Systems Ltd. | Method and system for spatial segmentation of anatomical structures |
| US20170091948A1 (en) * | 2015-09-30 | 2017-03-30 | Konica Minolta Laboratory U.S.A., Inc. | Method and system for automated analysis of cell images |
| CN105354846A (en) * | 2015-11-05 | 2016-02-24 | 沈阳东软医疗系统有限公司 | Method and apparatus for segmenting three-dimensional medical image |
| CN108805886B (en) * | 2018-05-30 | 2021-09-03 | 中北大学 | Persistent clustering segmentation method for multi-fusion physical signatures |
-
2019
- 2019-11-22 CN CN201911156860.3A patent/CN110930389B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106373168A (en) * | 2016-11-24 | 2017-02-01 | 北京三体高创科技有限公司 | Medical image based segmentation and 3D reconstruction method and 3D printing system |
| CN109064471A (en) * | 2018-07-18 | 2018-12-21 | 中北大学 | A kind of three-dimensional point cloud model dividing method based on skeleton |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110930389A (en) | 2020-03-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yoo | Three-dimensional surface reconstruction of human bone using a B-spline based interpolation approach | |
| US8571278B2 (en) | System and methods for multi-object multi-surface segmentation | |
| US7995810B2 (en) | System and methods for image segmentation in n-dimensional space | |
| CN106887000B (en) | Gridding processing method and system for medical image | |
| Pahr et al. | From high-resolution CT data to finite element models: development of an integrated modular framework | |
| WO2007058993A2 (en) | Surface-based characteristic path generation | |
| CN105740533B (en) | CT gray scale-material properties assignment the finite element modeling method orthopedic for osteotomy | |
| Zachow et al. | 3D reconstruction of individual anatomy from medical image data: Segmentation and geometry processing | |
| CN114155376A (en) | Target feature point extraction method and device, computer equipment and storage medium | |
| WO2021098617A1 (en) | Method for generating tumor in digital blood vessel model, system, and electronic device | |
| CN112102491B (en) | Skin damage surface three-dimensional reconstruction method based on surface structured light | |
| CN110930389B (en) | Region segmentation method, device and storage medium for three-dimensional medical model data | |
| CN103903255A (en) | Ultrasound image segmentation method and system | |
| CN110689080B (en) | Planar atlas construction method of blood vessel structure image | |
| US10593112B1 (en) | Chamber reconstruction from a partial volume | |
| CN111105476A (en) | Three-dimensional reconstruction method for CT image based on Marching Cubes | |
| Zhao et al. | A frame of 3D printing data generation method extracted from CT data | |
| US20140032180A1 (en) | Method and apparatus for computing deformation of an object | |
| CN107085866A (en) | The analog simulation method and system of a kind of straight bone CT images of human body | |
| CN112201329A (en) | Template-guided and data-driven three-dimensional broken bone segmentation and splicing method | |
| Callahan et al. | A meshing pipeline for biomedical computing | |
| Robb et al. | Efficient shape-based algorithms for modeling patient specific anatomy from 3D medical images: applications in virtual endoscopy and surgery | |
| CN116681705B (en) | Surface morphology measurement method and processing equipment based on longitudinal structure of human brain hippocampus | |
| CN116129060B (en) | Heart three-dimensional anatomical model construction method and heart three-dimensional mapping system | |
| CN110073408A (en) | Method and apparatus for dividing the two dimensional image of anatomical structure |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |