CN110570523B - Bone medical semantic automatic extraction method based on Gaussian process gradient model - Google Patents
Bone medical semantic automatic extraction method based on Gaussian process gradient model Download PDFInfo
- Publication number
- CN110570523B CN110570523B CN201910841806.6A CN201910841806A CN110570523B CN 110570523 B CN110570523 B CN 110570523B CN 201910841806 A CN201910841806 A CN 201910841806A CN 110570523 B CN110570523 B CN 110570523B
- Authority
- CN
- China
- Prior art keywords
- model
- bone
- template
- target
- gaussian process
- 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
- 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
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Graphics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Architecture (AREA)
- Radiology & Medical Imaging (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- General Health & Medical Sciences (AREA)
- Epidemiology (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a method for automatically extracting skeletal medical semantics based on a Gaussian process gradient model, which adopts a group matching mode to construct a more average single-bone statistical shape template based on the Gaussian process gradient model and marks corresponding medical semantics; then, importing a medical CT image of the target bone and generating a three-dimensional grid model of the target bone; and fitting and mapping the template and the three-dimensional grid model of the target skeleton, so that the medical semantics of the target skeleton can be automatically extracted. The invention adopts the skeleton medical semantic extraction method based on the Gaussian process gradient model, the Gaussian process gradient model has stronger deformability, better statistical deformation can be generated under the condition of fewer statistical shape samples, the deformation can exceed the shape range of the statistical samples, the requirement of medical semantic extraction in medicine can be better met, and the automation degree and the efficiency are effectively improved.
Description
Technical Field
The invention belongs to the field of digital medical treatment, and particularly relates to a method for automatically extracting skeletal medical semantics based on a Gaussian process gradient model.
Background art:
the extraction of medical semantics can be used for orthopedic preoperative three-dimensional planning such as knee joint replacement, femoral head prosthesis design and the like. There are some typical works on the extraction of the skeletal medical semantics, which can be divided into two categories: the method is a template-free method and is mainly based on the significance of bone markers, the morphological characteristics of bones and the spatial position relationship among the bone markers; one is templated methods such as statistical shape analysis and multispectral analysis. The invention adopts a medical semantic extraction method based on a single bone template of a Gaussian process gradient model.
The existing extraction method has poor performance, needs a large amount of statistical shape data, so that the extraction of the bone medical semantics at present mainly depends on the experience of doctors, and has extremely low automation degree. The invention provides a method for automatically extracting skeletal medical semantics based on a Gaussian process gradient model, which meets the requirement of extracting medical semantics in medicine, greatly improves the degree of automation and improves the working efficiency.
Disclosure of Invention
The invention aims to provide a bone medical semantic automatic extraction method based on a Gaussian process gradient model, so as to overcome the defect that a large amount of statistical shape data is needed in the prior art.
A method for automatically extracting bone medical semantics based on a Gaussian process gradient model comprises the following steps:
constructing a grid model according to the skeleton image;
obtaining a single bone template through a grid model, and labeling corresponding medical semantics;
selecting a target grid model;
fitting the single bone template with a target grid model;
medical semantics are obtained on the target mesh model by mapping.
Further, the method for constructing the grid model comprises the following steps:
acquiring electronic scanning images of the same bone of different patients;
the electronically scanned image is constructed into a mesh model.
Further, the construction method of the single-bone template comprises the following steps:
s3.1, regarding the grid model as a set A, and selecting any element in the set A to construct a Gaussian model after pre-configuration;
s3.2, respectively labeling corresponding mark points on the grid models in the set A, performing deformation registration on all elements in the set A through the obtained models according to the corresponding mark points, and respectively generating corresponding models which are regarded as a set B;
s3.3, constructing a Principal Component Analysis (PCA) model based on the set B;
s3.4, generating a gradual change model from the PCA model by using a mathematical method;
s3.5 jumping to S3.2, repeatedly executing S3.2, S3.3 and S3.4, and jumping to S3.6 after at least 10 times of circulation;
and S3.6, outputting a Gaussian process gradient model, namely the required single bone template.
Further, the method for selecting the target mesh model comprises the following steps:
the mesh model generated from the piece of bone of any one patient is used as the target mesh model.
Further, the method for fitting the single-bone template and the target grid model comprises the following steps:
selecting a plurality of corresponding characteristic points;
selecting a corresponding marked point coordinate variance range;
and (5) giving a fitting error range, and fitting according to the characteristic points.
Further, the method for obtaining medical semantics on the target mesh model by mapping comprises the following steps:
calculating the distance time complexity from the centroid of each surface of the target model to the centroid of each surface of the template model, wherein the distance time complexity is as follows:
O(n)＝n 2 ；
by constructing the AVL tree, the time complexity is reduced as follows:
O(n)＝log(n)；
setting each face of the template model as a node, wherein the weight is the Z coordinate value of the centroid of the face; given an accuracy pre, with a certain face of the target model set as F1, the search algorithm using the AVL tree finds that:
|F1.Z-root.Z|<pre
the first node of (a);
traversing a subtree taking the node as a root node, and selecting a template surface closest to the Euclidean distance of the center of mass of F1, namely a mapping result of the F1 surface;
mapping other surfaces according to the mode;
and mapping the medical semantics on the template model to the target model according to the mapping result.
The invention has the advantages that:
1. the Gaussian process gradient model has stronger deformability, and can generate better statistical deformation under the condition of fewer statistical shape samples;
2. the range of template deformation may be outside of the sample, so malformed or diseased bone may be better treated;
3. the requirement of extracting medical semantics in medicine is met, the automation degree is greatly improved, and the working efficiency is improved.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
FIG. 2 is a schematic diagram of the construction process of the template in the present invention.
FIG. 3 is a schematic diagram of the algorithm flow of the present invention.
Fig. 4 is a schematic diagram of a statistical tibial shape template according to the present invention.
FIG. 5 is a schematic diagram of the template model of the present invention before fitting to the target model (normal tibia).
FIG. 6 is a schematic diagram of the template model of the present invention after fitting to the target model (normal tibia).
Fig. 7 is a schematic diagram of medical semantic effects labeled on a tibial statistical shape template according to the present invention.
Fig. 8 is a schematic diagram of the mapping effect of the medical semantics on the target bone (normal bone) in the invention.
Fig. 9 is a schematic diagram illustrating the mapping effect of the medical semantics on the target bone (deformed bone) in the present invention.
Reference numerals: 1-a target model; 2-template model; 3-tibial tubercle; 4-Gerdy nodules; 5-oblique line; 6-medial condyle; 7-lateral condyle; 8-medial condyle; 9-lateral intercondylar tuberosity; 10-medial intercondylar tuberosity; 11-medial articular surface; 12-lateral articular surface.
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further described with the specific embodiments.
As shown in fig. 1 to 9, an automatic extraction method of medical semantics based on a gaussian process gradient model skeleton, in combination with the flowchart shown in fig. 1, includes the following steps:
step 2: pre-registering a series of tibiae obtained in the step 1 by utilizing an ICP algorithm;
and step 3: constructing a single bone statistical shape template (as shown in figure 4) with stronger deformability based on a Gaussian process gradient model by using the series of tibial grid models generated in the step 2, and labeling corresponding medical semantics (as shown in figure 7); the method for constructing the single bone statistical shape template with stronger deformability based on the Gaussian process gradient model comprises the following steps:
s3.1, taking a series of three-dimensional grid models generated by the Mimics as a set A, and constructing any element in the set A into a Gaussian model after pre-configuration;
s3.2, the fitting effect of the unmarked mark points is far inferior to that of the marked points, so that the corresponding mark points are marked, all elements in the A are subjected to deformation registration by the obtained model through the corresponding mark points, and the corresponding models are respectively generated and are regarded as a set B;
s3.3, constructing a Principal Component Analysis (PCA) model based on the set B;
s3.4, generating a Gaussian process gradient model with enhanced deformability by utilizing a Gaussian kernel function based on the PCA model;
s3.5 jumping to S3.2, repeatedly executing S3.2, S3.3 and S3.4 until cycling to the specified times, and jumping to S3.6;
and S3.6, outputting a Gaussian process gradient model, namely the required template.
And 4, step 4: generating a three-dimensional grid model from a CT image of a tibia of a certain patient, and regarding the three-dimensional grid model as a target grid model;
and 5, fitting the template to a target grid model through iterative fitting (as shown in figure 6). In the process of fitting the target skeleton model by the template, in order to improve the fitting accuracy, before shape fitting, a characteristic point matching mode is utilized, and a maximum value is given to the coordinate variance of the corresponding marking point so as to control the fitting error and obtain a better fitting effect.
Step 6: and mapping by taking the shortest centroid Euclidean distance of the two surfaces as a mapping principle. Corresponding medical semantics can be generated on the target model (as shown in fig. 8);
and 7: constructing the template model into an AVL tree to optimize the mapping algorithm by combining the flow chart shown in FIG. 3, thereby reducing the time complexity; in the step 6, the complexity of the distance from the centroid of each surface of the target model to the centroid of each surface of the template model in time is calculated to be
O(n)＝n 2
The calculation amount is large, the running time is long, and in order to meet the actual requirement, a certain optimization strategy must be adopted for the algorithm so as to reduce some unnecessary calculation. The invention adopts the method of constructing the AVL tree to reduce the complexity of the searching time
O(n)＝log(n)
Setting each face of the template model as a node, wherein the weight is the Z coordinate value of the centroid of the face; given an accuracy pre, taking a certain face of the target model (set as F1) as an example, the search algorithm using the AVL tree finds a satisfaction
|F1.Z-root.Z|<pre
The first node of (2) is traversed through a subtree taking the node as a root node, and a template surface closest to the Euclidean distance of the centroid of F1 is selected, namely the mapping result of the F1 surface, and the other surfaces are similar.
FIG. 4 illustrates a tibial statistical shape template in accordance with an embodiment of the present invention; FIG. 5 is a pictorial representation of the template model before fitting to the target model (normal tibia), from which it can be seen that the template bone is significantly longer than the target bone, with a significant difference therebetween; fig. 6 is an illustration of the template model and the target model (normal tibia) after fitting, and it can be seen that the two can be better matched together after fitting.
Fig. 7 is a diagram of medical semantic effects marked on a statistical tibial shape template according to an embodiment of the present invention, where fig. a is a front view and fig. b is a top view.
Fig. 8 is a diagram illustrating the mapping effect of medical semantics on a target bone (normal bone), in which a is a front view and b is a top view, according to an embodiment of the present invention.
Fig. 9 is a diagram illustrating the mapping effect of medical semantics on a target bone (malformed bone), in which fig. a is a front view and fig. b is a top view, according to the embodiment of the present invention. It can be seen that the mapping of malformed or diseased bone can be done well using templates based on a gaussian process gradient model.
The PCA model is a linear parametric model suitable for mathematical image analysis algorithms, but it has a certain disadvantage that the PDM method can only represent shapes within a given training sample range, and it requires a lot of training data if it is to express all possible target shapes.
A PCA model can be regarded as superposing a transformation on an average model, and the Gaussian process gradient model is that the transformation is directly constructed into a Gaussian process, and then K-L Expansion (Karhunen-Loeve Expansion) is utilized to obtain
Where u ∈ GP (μ, k), μ is the mathematical expectation of the Gaussian process, and k is the covariance function. (lambda i ,φ i ) Is a eigenvalue/eigenfunction pair of the following integral operator
ρ (x) represents a metric. Due to the random coefficient alpha i Is uncorrelated, the variance of u can be given by the sum of the variances of the individual components. Therefore, the eigenvalue λ i Corresponding to the variance from the ith component. This indicates if λ i Decay fast enough, we can express this process using the following low rank approximation, i.e.
The covariance function (kernel function) is not limited by training samples, so that the Gaussian process gradient model can be made by using any effective positive definite covariance function, and a very effective prior model can be made under the condition of few or no samples.
The key to representing the process with low rank approximation is to find
We can approximate this with Nystrom method by which we can randomly sample points X ═ X 1 ,...,x n We can get
Ku i ＝λ i mat u i (3)
Wherein K il ＝k(x i ,x l ) In the form of a kernel matrix, the kernel matrix,
u i denotes the ith feature vector, λ i mat Indicates a corresponding characteristic value, λ i mat Is approximately equal to lambda i Intrinsic function phi i Is approximately equal to
Wherein k is x (x)＝(k(x 1 ,x),...,k(x n ,x))。
The form of the kernel function is diverse. A simpler gaussian process is one that performs a smooth deformation with the expectation of zero. The expectation of the gaussian process is often set to zero in model matching, which means we assume that the reference surface shape is very close to the average shape. A particularly simple smoothing kernel is defined as follows
k g (x,y)＝exp(-||x-y|| 2 /σ 2 ) (5)
Wherein sigma 2 The range of the deformation is defined. Thus, the larger the value of σ, the smoother the resulting deformation will be. To match using this scalar kernel, we can defineOne matrix value kernel is as follows
k(x,y)＝s·I 3×3 k g (x,y) (6)
Wherein the identity matrix I 3×3 The x, y and z components representing the model vector field are independent, the parameter s determines the variance of the deformation vector, and the structure can be generalized by defining the matrix value kernel as follows
k(x,y)＝Ak g (x,y)A T ,A∈R 3×3 (7)
Based on the above, in the invention, in order to realize the automatic extraction of the bone medical semantics, firstly, a group matching mode is adopted to construct a more average single bone statistical shape template based on a Gaussian process gradient model, and corresponding medical semantics are labeled; then, importing a medical CT image of the target bone and generating a three-dimensional grid model of the target bone; and fitting and mapping the template and the three-dimensional grid model of the target skeleton, so that the medical semantics of the target skeleton can be automatically extracted. The invention adopts the skeletal medicine semantic extraction method based on the Gaussian process gradient model, the Gaussian process gradient model has stronger deformability, better statistical deformation can be generated under the condition of fewer statistical shape samples, the deformation can exceed the shape range of the statistical samples, the requirement of extracting medical semantics in medicine can be better met, and the automation degree and the efficiency are effectively improved.
It will be appreciated by those skilled in the art that the invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The embodiments disclosed above are therefore to be considered in all respects as illustrative and not restrictive. All changes which come within the scope of or equivalence to the invention are intended to be embraced therein.
Claims (5)
1. A method for automatically extracting bone medical semantics based on a Gaussian process gradient model is characterized by comprising the following steps:
constructing a grid model according to the skeleton image;
obtaining a single bone template through a grid model, and labeling corresponding medical semantics;
selecting a target grid model;
fitting the single bone template with a target grid model;
obtaining medical semantics on the target grid model by mapping;
the method for obtaining the medical semantics on the target mesh model through mapping comprises the following steps:
calculating the distance time complexity from the centroid of each surface of the target model to the centroid of each surface of the template model, wherein the distance time complexity is as follows:
O(n)＝n 2 ；
by constructing the AVL tree, the time complexity is reduced to:
O(n)＝log(n)；
setting each face of the template model as a node, wherein the weight is the Z coordinate value of the centroid of the face; given an accuracy pre, with a certain face of the target model set as F1, the search algorithm using the AVL tree finds that:
|F1.Z-root.Z|＜pre
the first node of (a);
traversing a subtree taking the node as a root node, and selecting a template surface closest to the Euclidean distance of the center of mass of F1, namely a mapping result of the F1 surface;
mapping other surfaces according to the mode;
and mapping the medical semantics on the template model to the target model according to the mapping result.
2. The method for automatically extracting the bone medical semantics based on the Gaussian process gradient model as claimed in claim 1, wherein: the construction method of the grid model comprises the following steps:
acquiring electronic scanning images of the same bone of different patients;
the electronically scanned image is constructed into a mesh model.
3. The method for automatically extracting the bone medical semantics based on the Gaussian process gradient model as claimed in claim 1, wherein: the construction method of the single-bone template comprises the following steps:
s3.1, regarding the grid model as a set A, and selecting any element in the set A to construct a Gaussian model after pre-configuration;
s3.2, respectively labeling corresponding mark points on the grid models in the set A, performing deformation registration on all elements in the set A through the obtained models according to the corresponding mark points, and respectively generating corresponding models which are regarded as a set B;
s3.3, constructing a Principal Component Analysis (PCA) model based on the set B;
s3.4, generating a gradual change model from the PCA model by using a mathematical method;
s3.5 jumping to S3.2, repeatedly executing S3.2, S3.3 and S3.4, and jumping to S3.6 after at least 10 times of circulation;
and S3.6, outputting a Gaussian process gradient model, namely the required single bone template.
4. The method for automatically extracting the bone medical semantics based on the Gaussian process gradient model as claimed in claim 1, wherein: the method for selecting the target grid model comprises the following steps:
a mesh model generated from the bone of any one of the patients is used as a target mesh model.
5. The method for automatically extracting the bone medical semantics based on the Gaussian process gradient model as claimed in claim 1, wherein: the method for fitting the single-bone template and the target grid model comprises the following steps:
selecting a plurality of corresponding characteristic points;
selecting a corresponding marked point coordinate variance range;
and (5) giving a fitting error range, and fitting according to the characteristic points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910841806.6A CN110570523B (en) | 2019-09-06 | 2019-09-06 | Bone medical semantic automatic extraction method based on Gaussian process gradient model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910841806.6A CN110570523B (en) | 2019-09-06 | 2019-09-06 | Bone medical semantic automatic extraction method based on Gaussian process gradient model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110570523A CN110570523A (en) | 2019-12-13 |
| CN110570523B true CN110570523B (en) | 2022-08-16 |
Family
ID=68778165
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910841806.6A Active CN110570523B (en) | 2019-09-06 | 2019-09-06 | Bone medical semantic automatic extraction method based on Gaussian process gradient model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110570523B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104881875A (en) * | 2015-06-09 | 2015-09-02 | 河海大学常州校区 | Priori-knowledge-based reconstruction method of closed skeleton contour in medical fault image |
| CN108597017A (en) * | 2018-04-19 | 2018-09-28 | 河海大学常州校区 | A kind of textured bone template construction method based on measurement parameter |
-
2019
- 2019-09-06 CN CN201910841806.6A patent/CN110570523B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104881875A (en) * | 2015-06-09 | 2015-09-02 | 河海大学常州校区 | Priori-knowledge-based reconstruction method of closed skeleton contour in medical fault image |
| CN108597017A (en) * | 2018-04-19 | 2018-09-28 | 河海大学常州校区 | A kind of textured bone template construction method based on measurement parameter |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110570523A (en) | 2019-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Subburaj et al. | Automated identification of anatomical landmarks on 3D bone models reconstructed from CT scan images | |
| Davies et al. | A minimum description length approach to statistical shape modelling | |
| US20160210779A9 (en) | Single Action Three-Dimensional Model Printing Methods | |
| US20060111631A1 (en) | System, apparatus and method for forensic facial approximation | |
| CN102365655A (en) | Image matching device and device for positioning patient | |
| US20080260254A1 (en) | Automatic 3-D Object Detection | |
| CN112037200A (en) | Method for automatically identifying anatomical features and reconstructing model in medical image | |
| Adams et al. | Uncertain-deepssm: From images to probabilistic shape models | |
| Vezzetti et al. | Geometry-based 3D face morphology analysis: soft-tissue landmark formalization | |
| Dai et al. | Locating anatomical landmarks on 2D lateral cephalograms through adversarial encoder-decoder networks | |
| CN108597017A (en) | A kind of textured bone template construction method based on measurement parameter | |
| Aneja et al. | Automated detection of 3d landmarks for the elimination of non-biological variation in geometric morphometric analyses | |
| Grosland et al. | Automated hexahedral meshing of anatomic structures using deformable registration | |
| Marzola et al. | A reliable procedure for the construction of a statistical shape model of the cranial vault | |
| Lim et al. | Cube Kohonen Self-Organizing Map (CKSOM) model with new equations in organizing unstructured data | |
| CN110570523B (en) | Bone medical semantic automatic extraction method based on Gaussian process gradient model | |
| Wilson et al. | Voyaging into the third dimension: A perspective on virtual methods and their application to studies of juvenile sex estimation and the ontogeny of sexual dimorphism | |
| KR101444816B1 (en) | Image Processing Apparatus and Method for changing facial impression | |
| US8005652B2 (en) | Method and apparatus for surface partitioning using geodesic distance | |
| Wu et al. | A comparative analysis of global shape analysis methods for the assessment of the human right ventricle | |
| White et al. | A three-dimensional approach to visualize pairwise morphological variation and its application to fragmentary palaeontological specimens | |
| Kainmueller | Deformable meshes for medical image segmentation: accurate automatic segmentation of anatomical structures | |
| Krawczyk et al. | YOLO and morphing-based method for 3D individualised bone model creation | |
| CN114004940A (en) | Non-rigid generation method, device and equipment of face defect reference data | |
| CN106228180A (en) | The identification model parameter training method of a kind of lung image focus and recognition methods |
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 |