CN107424162A - A kind of image partition method and system - Google Patents
A kind of image partition method and system Download PDFInfo
- Publication number
- CN107424162A CN107424162A CN201710311908.8A CN201710311908A CN107424162A CN 107424162 A CN107424162 A CN 107424162A CN 201710311908 A CN201710311908 A CN 201710311908A CN 107424162 A CN107424162 A CN 107424162A
- Authority
- CN
- China
- Prior art keywords
- model
- image
- edge
- point
- module
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
-
- 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
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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
- G06T2207/10104—Positron emission tomography [PET]
-
- 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
- G06T2207/10108—Single photon emission computed tomography [SPECT]
-
- 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
- G06T2207/10116—X-ray image
-
- 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
- G06T2207/10132—Ultrasound image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20076—Probabilistic image processing
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The present invention provides a kind of image partition method, including:Obtain view data;Based on described image data, reconstruction image, wherein, described image includes one or more first edges;A model is obtained, wherein, the model includes the one or more second edges corresponding with one or more of first edges;Match the model and the image after the reconstruction;And according to one or more of first edges, adjust one or more second edges of the model.
Description
【Technical field】
The present invention relates to a kind of image processing method, more particularly to a kind of images match dividing method based on probability and it is
System.
【Background technology】
With the raising of human living standard and the extension of life expectancy, angiocardiopathy turns into the No.1 cause of the death of the mankind,
Therefore the early diagnosis of angiocardiopathy can effectively reduce case fatality rate.Understand the Radiologic imaging and its performance data of cardiac structure
It is the important prerequisite of correct Diagnosing Cardiac disease, the development of CT technologies, hence it is evident that improve time sense, reduce heartbeat
Artifact, good application potential is shown in terms of heart fine structure is shown.
Image Segmentation Technology is the key technology of graphical analysis link, and it plays increasing work in medical imaging
With.Image segmentation is to extract the indispensable means of the quantitative information of particular tissues in imaged image, while is also visualization
The pre-treatment step and premise of realization.Image after segmentation is just being widely used in various occasions, such as quantitative point of tissue volume
Analysis, diagnosis, the positioning of pathological tissues, the study of anatomical structure, treatment planning, the local volume effect calibration of functional imaging data and
Computer guidance is performed the operation.
, it is necessary to carry out fixation and recognition to the heart chamber on image after the reconstructed acquisition of CT images.The positioning of heart chamber
Identification needs to carry out heart rim detection.Variable model is the more general way in heart chamber segmentation field.Heart chamber mould
Type is averagely obtained based on view data corresponding to more set clinical heart chamber models.Obtained by the matching of model and image
The image matched somebody with somebody.
【The content of the invention】
A kind of image processing method, including:Obtain view data;Based on described image data, reconstruction image, wherein, institute
Stating image includes one or more first edges;A model is obtained, wherein, the model includes and one or more of the
The corresponding one or more second edges in one edge;Match the model and the image after the reconstruction;And according to described
One or more first edges, adjust one or more second edges of the model.
Further, adjusting one or more second edges of the model includes:Determine one in the second edge
Individual reference point;It is determined that the target point corresponding with the reference point;And according to the target point, adjust the second of the model
Edge.
Further, it is described to determine that the target point corresponding with the reference point includes:Determine the normal of the reference point;
Obtain step-length and hunting zone;According to the step-length and hunting zone, one or more candidate points are determined along normal;Obtain one
First grader;According to first grader, determine that one or more of candidate points correspond to the general of the first edge
Rate;And correspond to the probability of the first edge based on one or more of candidate points, determine the target point.
Further, the normal for determining the reference point includes:It is determined that one or more adjacent with the reference point
Individual polygonal mesh;Determine one or more normals corresponding to one or more of polygonal mesh;And according to described one
Individual or multiple normals, determine the normal of the reference point.
Further, the second edge of the adjustment model includes:Similarity transformation is carried out to the second edge;
The control point of the model is determined, one or more second edges according to the control point of the model and the model
Relation, the association factor at the control point of the model is generated, according to association factor, affine transformation is carried out to the second edge;
Or based on an energy function, the second edge is finely adjusted.
Further, described image data include brain image, skull image, chest image, cardiac image, mammary gland figure
Picture, abdomen images, renal image, liver image, pelvis image, perineum image, limbs image, vertebra image or osteogram
Picture.
Further, an image processing system, including:One memory, is configured as data storage and instruction;One
The processor to communicate is established with memory, wherein, when performing the instruction in memory, the processor is configured as:Obtain
View data;Based on described image data, reconstruction image, wherein, described image includes one or more first edges;Obtain one
Individual model, wherein, the model includes the one or more second edges corresponding with one or more of first edges;
With the image after the model and the reconstruction;And according to one or more of first edges, adjust the one of the model
Individual or multiple second edges.
Further, the processor is further configured to:Determine a reference point in the second edge;It is determined that
The target point corresponding with the reference point;And according to the target point, adjust the second edge of the model.
Further, the processor is further configured to:Determine the normal of the reference point;Obtain step-length and search
Scope;According to the step-length and hunting zone, one or more candidate points are determined along normal;Obtain first grader;Root
According to first grader, determine that one or more of candidate points correspond to the probability of the first edge;And based on institute
The probability that one or more candidate points correspond to the first edge is stated, determines the target point.
Further, the second edge of the adjustment model includes:Similarity transformation is carried out to the second edge;
The control point of the model is determined, one or more second edges according to the control point of the model and the model
Relation, the association factor at the control point of the model is generated, according to association factor, affine transformation is carried out to the second edge;
Or based on an energy function, the second edge is finely adjusted.
【Brief description of the drawings】
Fig. 1 is a kind of application scenarios signal of the example control and processing system according to some embodiments of the present application
Figure;
Fig. 2 is a kind of schematic diagram of example system configuration of the processing equipment according to some embodiments of the present application;
Fig. 3 is that one kind for being used to implement some particular systems in the application according to some embodiments of the present application is shown
Example mobile device schematic diagram;
Fig. 4 is the schematic diagram of the example process equipment according to some embodiments of the present application;
Fig. 5 is the exemplary process diagram of the implementation processing equipment according to some embodiments of the present application;
Fig. 6 is the schematic diagram of the example model structure module according to some embodiments of the present application;
Fig. 7 is the exemplary process diagram of the structure averaging model according to some embodiments of the present application;
Fig. 8 is the schematic diagram of the example training module according to some embodiments of the present application;
Fig. 9 is the exemplary process diagram of the training grader according to some embodiments of the present application;
Figure 10 is the structural representation of the example model matching module according to some embodiments of the present application;
Figure 11 is the matching averaging model and the exemplary stream for the image rebuild according to some embodiments of the present application
Cheng Tu;
Figure 12 is the structural representation of the example adjusting module according to some embodiments of the present application;
Figure 13 is the exemplary process diagram of the adjustment model according to some embodiments of the present application;
Figure 14 is the exemplary process diagram to be set the goal really according to some embodiments of the present application a little;
Figure 15 is the exemplary of one averaging model marginal point normal of determination according to some embodiments of the present application
Flow chart;
Figure 16 is the exemplary process diagram of the averaging model edge point transformation according to some embodiments of the present application;
Figure 17 is the sharp keen degree schematic diagram of an image of the application;
Figure 18 is the embodiment of the image border classification based training of the application;
Figure 19 is the embodiment of the model meshes classification of the application;
Figure 20 is the embodiment of the model meshes division of the application;
Figure 21 is the embodiment of the grid model based on association factor of the application;
Figure 22 is the image border exemplary plot based on the classification of sharp keen degree;
Figure 23 is the model example figure based on the classification of sharp keen degree;
Figure 24 is the image probability graph embodiment that the present embodiment obtains according to grader;
Figure 25 is the exemplary plot of the average meshes model and images match after Hough transformation;
Figure 26 is the image chamber segmentation result exemplary plot accurately matched after adjusting;
Figure 27 A are to be not based on the image segmentation figure that association factor is divided;And
Figure 27 B are the image segmentation figures divided based on association factor.
【Embodiment】
Many details are elaborated in the following description in order to fully understand the present invention.But the present invention can be with
Much it is different from other manner described here to implement, those skilled in the art can be in the situation without prejudice to intension of the present invention
Under do similar popularization, therefore the present invention is not limited to the specific embodiments disclosed below.
Secondly, the present invention is described in detail using schematic diagram, when the embodiment of the present invention is described in detail, for purposes of illustration only, institute
It is example to state schematic diagram, and it should not limit the scope of protection of the invention herein.
Fig. 1 is a kind of application scenarios signal of the example control and processing system according to some embodiments of the present application
Figure.As shown in figure 1, control and processing system 100 include 110, databases 120 of an imaging device and a processing equipment
130。
Imaging device 110 can generate image by scanning a target object.Described image can be various medical science figures
Picture.For example, head image, chest image, abdomen images, pelvis image, perineum image, limbs image, vertebra image, vertebra
Image etc..Wherein, head image can include brain image, skull image etc..Chest image can include whole chest image,
Cardiac image, galactophore image etc..Abdomen images can include whole abdomen images, renal image, liver image etc..Cardiac image
Comprehensive digitlization cardiod diagram, digitlization heart chromatographical X-ray figure, heart phase contrast figure, X ray image can be included but is not limited to
(CR) figure, multi-modality images etc..The medical image can be two dimensional image or 3-D view.The form of the medical image can
With including jpeg format, tiff format, GIF forms, FPX forms etc..The medical image can be stored in database 120,
It can also transmit to processing equipment 130 and carry out image procossing.The application will illustrate by taking cardiac image as an example, but this area
Technical staff is it is understood that the present processes can be used for other images.
Database 120 can store the information of image and/or image correlation.The described image information related to image can be with
There is provided, can also be obtained outside system 100 by imaging device 110 and processing equipment 130, for example, user inputs information, from network
Obtain information etc..The related information of described image can include handling in the algorithm of image, sample, model, parameter, processing procedure
Real time data etc..Database 120 can be hierarchical database, network database or relational database.Database 120
Can be local data base or remote data base.Other storage devices can be by Information Number in database 120 or system
The storage device to be worked in a manner of electricity, light or magnetic etc. is recycled to be stored after word.In certain embodiments, database 120
Or other storage devices can be the equipment using electric energy mode storage information in system, for example, it is random access memory (RAM), read-only
Memory (ROM) etc..Random access memory can include but is not limited to dekatron, selectron, delay line storage, William
This pipe, dynamic RAM (DRAM), SRAM (SRAM), IGCT random access memory (T-RAM), zero capacitance
One or more combinations in random access memory (Z-RAM) etc..Read-only storage includes but is not limited to magnetic bubble memory, A.O.
It is linear memory, thin-film memory, magnetic plated wire memeory, magnetic core internal memory, magnetic drum memory, CD drive, hard disk, tape, non-
It is volatile storage (NVRAM), phase-change memory element, reluctance type random storage formula internal memory, ferroelectric random stored memory, non-volatile
SRAM, programmable read only memory, Mask ROM, floating connection door random access memory, nanometer are random
One or more combinations in memory, racing track internal memory, variable resistance type internal memory, programmable metallization unit etc..At some
In embodiment, other storage devices can be the equipment using magnetic energy mode storage information in database 120 or system, such as firmly
Disk, floppy disk, tape, core memory, magnetic bubble memory, USB flash disk, internal memory etc..In certain embodiments, database 120 or system
Other interior storage devices can be the equipment using optical mode storage information, such as CD, DVD etc..In certain embodiments, number
Can be the equipment using magneto-optic mode storage information according to storehouse 120, such as magneto-optic disk etc..Other are deposited in database 120 or system
The access mode of storage equipment can be one or more combinations in random storage, serial access storage, read-only storage etc..Number
Can be impermanent memory memory or permanent memory memory according to other storage devices in storehouse 120 or system.It is above-mentioned
Storage device simply lists some examples, and the storage device that database 120 can use is not limited to this.
Database 120 can be a part for processing equipment 130 or a part for imaging device 110, can also
Exist independently of processing equipment 130 and imaging device 110.In certain embodiments, database 120 can by network 150 with
Control and the connection of other modules in processing system 100 or equipment.The connected mode can include wired connection, wireless connection
Or both combination.
Processing equipment 130 can obtain view data from imaging device 110, can also obtain picture number from database 120
According to.Processing equipment 130 can implement a variety of processing to the image of acquisition.The processing can include grey level histogram processing, return
One change processing, geometric transformation, spatial alternation, picture smooth treatment, image enhancement processing, image dividing processing, at image conversion
Reason, image recovery, compression of images, image characteristics extraction etc..Processing equipment 130 can be by the view data storage after processing to number
According to storehouse 120, can also be transferred in the equipment outside control and processing system 100.
In certain embodiments, processing equipment 130 can include one or more processors, memory etc..For example, processing
Equipment 130 can include central processing unit (CPU), application specific integrated circuit (ASIC), ASIP (ASIP), image
Processor (GPU), physical manipulations processor (PPU), digital signal processor (DSP), field programmable gate array (FPGA), can
In programmed logic device (PLD), controller, micro-control unit, processor, microprocessor, Advance RISC Machine processor etc.
One or more combinations.
In certain embodiments, control and processing system 100 can also include a terminal device 140.The terminal is set
It is standby to carry out information exchange with imaging device 110, database 120 and processing equipment 130.For example, the terminal device 140 can
With the view data after the acquisition processing from processing equipment 130.In certain embodiments, terminal device 140 can be set from imaging
Standby 110 obtain view data, and view data is transferred into processing equipment 130 and carries out image procossing.The terminal device 140 can
With including one or more input equipments, control panel etc..For example, the input equipment can include keyboard, touch-screen, mouse
Mark, voice-input device, scanning device, information identification equipment (such as human eye recognition system, fingerprint recognition system, brain monitoring system
Deng), remote controllers etc..
Control and processing system 100 can be connected with network 150.The network 150 can be wireless network, mobile network
Network, finite element network or other connections.Wherein, wireless network can includeWLAN, Wi-Fi, WiMax etc..
Mobile network can include 2G signals, 3G signals, 4G signals etc..Cable network can include LAN (LAN), wide area network
(WAN), proprietary network etc..
Control and the database 120 in processing system 100 and processing equipment 130 can perform operation by cloud computing platform
Instruction.Cloud computing platform can include the storage-type cloud platform based on data storage, the calculation type cloud based on data processing
The synthesis cloud computing platform that platform and calculating and data storage processing are taken into account.For example, caused by control and processing system 100
Some view data can be calculated or stored by cloud computing platform.
It should be noted that above to control and the description of processing system 100, only for convenience of description, can not be this Shen
It please be limited within cited scope of embodiments.
Fig. 2 is a kind of schematic diagram of example system configuration of the processing equipment according to some embodiments of the present application.
As shown in Fig. 2 processing equipment 130 can include a read-only storage of processor 220, one of data/address bus 210, one
(ROM) 230,260, hard disks of input/output end port of COM1 250, one of random access memory (RAM) 240, one
270 and a display 280 being connected with input/output end port 260.Connection in the processing equipment 130 between each hardware
Mode can be wired, wireless or both combination.Any one hardware can be local, long-range or both
With reference to.
Data/address bus 210 can be used for transmitting data information.In certain embodiments, in processing equipment 130 each hardware it
Between can pass through the data/address bus 210 carry out data transmission.For example, processor 220 can pass through the data/address bus 210
Transmit data in other hardware such as memory or input/output end port 260.It should be noted that the data can be
Real data or instruction code, status information or control information.In certain embodiments, data/address bus 210 can be with
It is mutual for industrial standard (I SA) bus, extension industrial standard (EISA) bus, video electronics standard (VESA) bus, external component
Connection standard (PCI) bus etc..
Processor 220 can be used for logical operation, data processing and instruction generation.In certain embodiments, processor 220
Data/commands can be obtained from internal storage, the internal storage can include read-only storage (ROM), deposit at random
Reservoir (RAM), cache memory (Cache) (not shown in FIG.) etc..In certain embodiments, processor 220 can be with
Including multiple sub-processors, the sub-processor can be used for the difference in functionality for realizing system.
Read-only storage 230 is used for the initial of each functional module in the Power-On Self-Test of processing equipment 130, processing equipment 130
Change, the driver of basic input/output of processing equipment 130 etc..In certain embodiments, read-only storage can include can
Program read-only memory (PROM), programmable and erasable read-only storage (EPROM) etc..Random access memory 240, which is used to deposit, to be grasped
Make system, various application programs, data etc..In certain embodiments, random access memory 240 can include SRAM
(SRAM), dynamic RAM (DRAM) etc..
COM1 250 is used for attended operation system and external network, realizes the communication exchanges between them.In some realities
Apply in example, COM1 250 can include FTP ports, http port or DNS ports etc..Input/output end port 260 is used for outer
Between portion's equipment or circuit and processor 210 carry out data, information exchange and control.In certain embodiments, input/output
Port 260 can include USB port, PCI port, I DE ports etc..
Hard disk 270 is used to store received information caused by processing equipment 130 or outside processing equipment 130 and number
According to.In certain embodiments, hard disk 270 can include mechanical hard disk (HDD), solid state hard disc (SSD) or hybrid hard disk (HHD)
Deng.Information, the data that display 280 is used to generate system 130 are presented to user.In certain embodiments, display 280 can
With including a physical display, such as display with loudspeaker, LCD display, light-emitting diode display, OLED display, electronic ink
Water display (E-Ink) etc..
Fig. 3 is that one kind for being used to implement some particular systems in the application according to some embodiments of the present application is shown
Example mobile device schematic diagram.As shown in figure 3, mobile device 350 can include a terminal device 150.In certain embodiments,
User can be received or be sent the information related to control and processing system 100 by mobile device 350.Mobile device 350 can
With including smart mobile phone, personal digital assistant (PDA), tablet personal computer, handheld device, intelligent glasses, intelligent watch, wearable
One or more in equipment, virtual reality device or display enhancing equipment etc..In certain embodiments, mobile device 350 can
With including 356, one or more central processing units (CPUs) 358, one or more image processors (GPUs) displays
354th, a memory 368 of communication platform 352, one of internal memory 362, one and one or more input-output apparatus 360.Enter
One step, mobile device 350 can also include system bus, controller etc..As shown in figure 3, CPU can be from storage
Module 368 should by mobile device operation system (for example, iOS, Android, Windows Phone etc.) 364 and one or more
Downloaded to 366 in internal memory 362.One or more of applications 366 can include a webpage or other for receiving and passing
Pass and control and the Mobile solution software (App) of information that processing system 100 is related.User can pass through input-output apparatus
360 obtain or provide information, and described information can further be transferred to the equipment in control and processing system 100 and/or system
Unit.
In embodiments herein, computer hardware platforms may be used as one or more elements (for example, control and place
Reason system 100 and its inside other parts) hardware platform, implement various modules, unit and their function.It is described hard
Part element, operating system and programming language are inherently traditional, and those skilled in the art are possible to adapt these technologies
And established and edge segmentation applied to cardiac image model.Computer with user interface can be used as PC (PC),
Other work stations or terminal device, properly programmed computer can also be used as server.Because those skilled in the art are to this
Structure, programming and the general operation of computer equipment used in application should be all very familiar with, therefore, no longer for other attached
Figure makees related specific explanations.
Fig. 4 is the schematic diagram of the example process equipment according to some embodiments of the present application.Processing equipment 130 can be with
Including 430, model construction modules 430 of a memory module of image reconstruction module 420, one of acquisition module 410, one,
One training module, 450, matching modules 460 and an adjusting module 470.In the processing equipment 130 between each module
Connected mode can be wired, wireless or both combination.Any one module can be it is local, long-range or
Both combinations.
Memory module 430 can be used for storage image data or information, and its function can be by hard disk in Fig. 2 270, read-only
One or more combinations in memory 230, random access memory 240 etc. are realized.Memory module 430 can store processing and set
The information of module or equipment in standby 130 outside other modules or processing equipment 130.The information that memory module 430 stores can be with
Scan data including imaging device 110, process part in the control command or parameter information, processing equipment 130 of user's input
The intermediate data of generation or partial data information etc..In certain embodiments, memory module 430 can send the information of storage
Image procossing is carried out to process part.In certain embodiments, the information that memory module 430 can be generated with storing process part,
Such as data are calculated in real time.Memory module 430 can include but is not limited to common all kinds of storage devices such as solid state hard disc, machinery
Hard disk, USB flash memory, SD storage cards, CD, random access memory (RAM) or read-only storage (ROM) etc..Memory module 430 can be with
It is the storage device or its exterior or external storage device of internal system, such as the storage on cloud storage service device
Device.
Acquisition module 410 can be used for obtaining the view data that imaging device 110 gathers, the image that database 120 stores
Data outside data, or control and processing system 100, its function can be realized by the processor 220 in Fig. 2.It is described
View data can include the view data of the collection of imaging device 110, the algorithm for handling image, sample, model, parameter, processing
During real time data etc..In certain embodiments, acquisition module 410 can send the view data got or information
Handled to image reconstruction module 420.In certain embodiments, acquisition module 410 can be by the processing image got
The information such as algorithm, parameter are sent to model construction module 440.In certain embodiments, acquisition module 410 will can be got
View data or information are sent to memory module 370 and stored.In certain embodiments, acquisition module 410 will can obtain
To sample, parameter, model, the information such as real time data be sent to training module 450, matching module 460 or adjusting module 470.
In certain embodiments, acquisition module 410 can receive the data acquisition instruction from processor 220, and complete corresponding
Data acquisition operations.In certain embodiments, acquisition module 410 can be carried out pre- after view data or information is obtained to it
Processing.
Image reconstruction module 420 can be used for building a medical image, and its function can be by the processor 220 in Fig. 2
To realize.In certain embodiments, image reconstruction module 420 can obtain image from acquisition module 410 or memory module 430
Data or information, and the medical image according to described image data or information architecture.The medical image can be a people
Body 3 D medical image.Described image data can include different time, diverse location, the scan data of different angle.According to
The scan data, image reconstruction module 420 can calculate the feature or state of human body corresponding position, such as human body corresponding position
Density of absorbability, human body corresponding position tissue to ray etc., so as to construct the human body three-dimensional medical image.Enter one
Step ground, the human body three-dimensional medical image can be shown by display 280, or be deposited by memory module 430
Storage.In certain embodiments, the human body three-dimensional medical image can also be sent to model as the pending image after rebuilding
Structure module 440 is further handled.
Model construction module 440 can be used for the three-dimensional averaging model for establishing target object.In certain embodiments, it is described
Target object can be heart, and the three-dimensional averaging model can be that the heart chamber based on more set reference model structures is three-dimensional flat
Equal grid model.In certain embodiments, model construction module 440 can pass through acquisition module 410, memory module 430 or use
The mode of family input obtains the reference model of at least one heart chamber and the information related to reference model.It is described with refer to mould
The related information of type can include locus of the size of image, pixel, pixel etc..In certain embodiments, model construction
Module 440 can be according to the reference model and the information related to reference model of at least one heart chamber of acquisition to reference to mould
Type carries out the pretreatment such as registration so that the directions of all reference models, ratio etc. are consistent.The pretreated image can enter
One step is manually or the mode of processor automatic marking marks cavity margin, and heart reference model is divided into several sub- hearts
Splanchnocoel room, and heart chamber average meshes model is built according to the edge point data of each chamber.Model construction module 440 can be with
The heart chamber average meshes model of structure is sent into memory module 430 to be stored, training module 450 can also be sent to
Or matching module 460 is further operated.In certain embodiments, model construction module 440 can also refer to according to more sets
Model data determines the relation between each chamber on averaging model.For example, model construction module 440 can build association factor
Matrix, the association factor matrix can represent influence of each chamber to some or multiple number of edges strong points.Pass through structure
Association factor matrix, chamber boundary separation situation can be improved.Model construction module 440 can be by the association factor matrix of structure
It is sent to memory module 430 to be stored, matching module 460 can also be sent to or adjusting module 470 is used for calculation process.
Training module 450 can be used for training grader.Training module 450 can will likely marginal point be divided into difference
In chamber classification.For example, a range of data point of reference model adjacent edges can be respectively divided to a left side for training module 450
In ventricle, atrium sinistrum, right ventricle, atrium dextrum, left cardiac muscle or six chamber classifications of sustainer.In another example training module 450 can be with
Based on the intensity of variation of cavity margin by a range of data point of reference model adjacent edges be respectively divided left ventricle edge,
Atrium sinistrum sharpened edge, the non-sharpened edge in atrium sinistrum, right ventricle sharpened edge, the non-sharpened edge of right ventricle, the sharp keen side in atrium dextrum
Edge, the non-sharpened edge in atrium dextrum, sustainer edge, left myocardium sharpened edge and left cardiac muscle 10 chamber classifications of non-sharpened edge
In.In certain embodiments, training module 450 can be inputted by memory module 430, model construction module 440 or user
Mode obtains the reference model of at least one heart chamber and the information related to the reference model.It is described related to reference model
Information can include edge point data etc. of each chamber in reference model.In certain embodiments, training module 450 can be with
Point near cavity margin is divided into by positive sample and negative sample according to the point near cavity margin and the distance of cavity margin.
In some embodiments, the positive sample can include the data point in the certain threshold range of cavity margin, the negative sample
The data point of other random sites farther out and in space apart from edge can be included.In certain embodiments, training module 450
The point of reference model or averaging model upper chamber adjacent edges can be trained according to positive and negative sample point, and obtain one or more points
Class device.In certain embodiments, training module 450 can utilize Probabilistic Boosting-Tree (PBT) training point
Class device.The PBT can include two-stage PBT algorithms or multistage PBT algorithms.The grader that training module 450 will can train
It is sent to memory module 430 to be stored, adjusting module 470 can also be sent to and be used for calculation process.
Matching module 460 can be used for the averaging model progress for establishing pending image and model construction module 440
Match somebody with somebody, build three-dimensional grid model corresponding with pending image.The pending image from image reconstruction module 420 or can be deposited
Storage module 430 obtains.In certain embodiments, matching module 460 can pass through the methods of Hough transformation and match averaging model
Onto pending image, obtain and the heart chamber three-dimensional grid model after pending image rough matching.Matching module 460 can
Parameter required for the Hough transformation is obtained by way of acquisition module 410, memory module 430 or user's input etc.
Information.Heart chamber three-dimensional grid model after matching can be sent to memory module 430 and be stored by matching module 460,
470 further optimization processing of adjusting module can also be sent to.
Adjusting module 470 can be used for Optimized model, make model closer to real heart (pending cardiac image number
According to).Adjusting module 470 can be from the heart chamber grid mould after matching module 460 or the acquisition rough matching of memory module 430
Type.In certain embodiments, adjusting module 470 can be according to the gained cardiac module upper chamber a range of number in edge after matching
Strong point belongs to the optimal heart chamber edge of determine the probability of cavity margin.Adjusting module 470 further can be adjusted accurately
Heart chamber three-dimensional grid model.The accurate adjustment can include similarity transformation, piecewise affine transformations and/or based on energy
Micro- change of function etc..In certain embodiments, adjusting module 470 can be by the heart chamber three-dimensional grid mould of accurate adjustment gained
Type carries out image format conversion, obtains heart chamber edge segmentation figure (as shown in figure 26).Adjusting module 470 will can be adjusted accurately
Heart chamber model or heart chamber segmentation figure after whole are sent to memory module 430 and stored, and can also be sent to display
Device 280 is shown.
It should be noted that the above-mentioned description for processing equipment 130, only for convenience of description, can not limit the application
System is within the scope of illustrated embodiment.It is appreciated that for those skilled in the art, it is former in the work for understanding the equipment
After reason, modules may be combined in the case of without departing substantially from this principle, or form subsystem and other
Module connects, and makees various modifications and variations in the form and details to implementing the said equipment.For example, model construction module 440 and/
Or training module 450 can remove, or merge with memory module 430.Such deformation, the protection model in the application
Within enclosing.
Fig. 5 is the exemplary process diagram of the implementation processing equipment according to some embodiments of the present application.In step 510
In, view data can be obtained.In certain embodiments, step 510 can be realized by acquisition module 410.Described image number
According to can from imaging device 110, database 120 or control and processing system 100 outside obtain.Described image data can include
CT, positron emission chromatography imaging technique (PET), single photon emission tomography (SPECT), MRI (mr imaging technique),
Ultrasonic (Ultrasound) and the raw image data of other medical imaging devices collection.In certain embodiments, described image
Data can be the local raw image data of heart or heart.In certain embodiments, step 510 can include to acquisition
Heart raw image data is pre-processed, and pretreated raw image data is sent into the figure in processing equipment 130
As rebuilding module 420 or memory module 430.The pretreatment can include distortion correction, denoising, smooth, enhancing of image etc..
In step 520, cardiac image can be rebuild according to cardiac image data.The step can be by processing equipment
Image reconstruction module 420 in 130 is completed based on Image Reconstruction Technology.The cardiac image data can pass through acquisition module 410
Or memory module 430 obtains.The cardiac image can include comprehensive digitlization cardiod diagram, digitlization heart chromatographical X-ray
Figure, heart phase contrast figure, X ray cardiac imaging (CR) figure, multi-modal cardiac image etc..The cardiac image can be two dimensional image
Or 3-D view.The form of the cardiac image can include jpeg format, tiff format, GIF forms, FPX forms etc..It is described
Image Reconstruction Technology can include solution of simultaneous equations, Fourier transformation reconstruction method, direct back projection reconstruction method, the anti-throwing of filtering
Shadow reconstruction method, Fourier backprojection reconstruction method or convolution inverse projection reconstruction method, iterative reconstruction etc..In certain embodiments, walk
Rapid 520 can pre-process to the cardiac image data of acquisition, and obtain multiple heart sectional views or perspective view.In some realities
Apply in example, the cardiac image data of acquisition or pretreated cardiac image data can include multiple heart sectional views.Image
Cardiac image or model can be rebuild according to a series of information that heart sectional views provide by rebuilding module 420.The heart
The information that sectional view provides can include the tissue density of Heart tissue, to information such as the absorbabilities of ray.Again structure
The cardiac image built can be shown by display 280, or be stored by memory module 430.Rebuild
Cardiac image can also carry out further image procossing by the model construction module 440 in processing equipment 130.
In step 530, a three-dimensional cardiac average meshes model can be built.The step can be by processing equipment 130
In model construction module 440 according to multiple reference models complete.Step 530 can by module 410, memory module 430 or
The mode of user's input obtains multiple reference models.In certain embodiments, step 530 can include entering multiple reference models
Row image registration.Described image registration can be included based on gray level image registration, based on transform domain image registration, feature based figure
As registration etc..Wherein, feature can include characteristic point, characteristic area, edge feature etc..In certain embodiments, it is the multiple
Reference model can be the heart chamber partition data or reference model that cavity margin is crossed by user annotation.The averaging model
Point Distribution Model (PDM), Active Shape Model (ASM), Active Contour can be included
The model that Model (also referred to as Snakes), Active Appearance Model (AAM) etc. are calculated.In some embodiments
In, step 530 can include each on the averaging model constructed according to the cavity margin data determination on multiple reference models
Relation between chamber, and establish association factor two-dimensional matrix.In certain embodiments, three-dimensional cardiac average meshes model or contain
The memory module 430 that the averaging model of relevant factor information can be transmitted directly in processing equipment 130 by processor 220 is entered
Row storage, or be sent to matching module 460 and further handle.
Cardiac image data can be matched with three-dimensional cardiac average meshes model in step 540.Further,
The matching can include the matching of first edge point and three-dimensional cardiac average meshes model in cardiac image data.In some realities
Apply in example, step 540 can be completed by matching module 460 by image matching method.Described image matching process can include
Matching process based on NNDR, the searching algorithm of adjacent features point, target detection based on Hough transformation etc..In some embodiments
In, the heart averaging model that model construction module 440 is established can be matched by image reconstruction module by generalised Hough transform
In first edge on the 420 obtained cardiac image datas of processing, and the cardiac module after being matched.In some embodiments
In, the probability that edge can be belonged to based on each point on cardiac image data to be matched implements weighting generalised Hough transform.It is described general
The grader that rate can train according to training module 450, each point on cardiac image data to be matched is inputted into classification
Device is calculated.In certain embodiments, a heart to be matched can be built according to the probability of each point on gained heart to be matched
Dirty marginal probability figure.The marginal probability figure can include shade of gray figure, color gradient figure (as shown in figure 24) etc..One
, can be to the cardiod diagram before each point on cardiac image data to be matched is calculated is as the probability at edge in a little embodiments
As being pre-processed.For example, the position for being unlikely to be heart edge completely can be excluded, so as to reduce the calculating of grader
Amount.For example, for CT images, the CT values of musculature are generally higher than -50, then CT values can be less than to -50 position
Marked by mask, grader is calculated the point at the position.In certain embodiments, the matching in processing equipment 130
Cardiac module after matching or three-dimensional cardiac grid model can be sent to memory module 430 and be stored by module 460, also may be used
To be sent to 470 further optimization processing of adjusting module.
In step 550, the heart chamber segmentation figure after accurate adjustment can be obtained.The step can be by processing equipment
Adjusting module 470 in 130 is completed.In certain embodiments, step 550 can be according to after matching on three-dimensional cardiac grid model
Cavity margin determine edge destination point., can be according to after matching on three-dimensional cardiac grid model for example, in certain embodiments
The a range of second edge point of cavity margin determine the probability edge destination point.In certain embodiments, the probability
The second classifier calculated based on the training of second edge point can be used.In certain embodiments, the probability can call base
In the first classifier calculated that multiple reference models or averaging model are trained.In certain embodiments, step 550 can be based on true
Fixed edge destination point deforms to three-dimensional cardiac grid model, so as to obtain the three-dimensional heart after cavity margin further adjusts
Dirty grid model.The deformation can include similarity transformation, affine transformation and other image Light deformation methods etc..For example,
In some embodiments, edge destination point that can be based on determination carries out similarity transformation, piecewise affine transformations and/or is based on successively
Micro- change of energy function.In certain embodiments, the adjusting module 470 in processing equipment 130 can be by the three-dimensional heart after adjustment
Dirty grid model is converted into heart chamber segmentation figure picture (as shown in figure 26) by mask (mask).The chamber segmentation figure as
Different chamber can be marked with different colors.In certain embodiments, the adjusting module 470 in processing equipment 130 can incite somebody to action
Heart chamber model or heart chamber segmentation figure after accurate adjustment are sent to memory module 430 and stored, and can also send
Shown to display 280.
It should be noted that the above-mentioned description that chamber cutting procedure is carried out for processing equipment 130, only for convenience of description,
The application can not be limited within the scope of illustrated embodiment.It is appreciated that for those skilled in the art,
After the operation principle for solving the equipment, any tune may be carried out to the order of each step in the case of without departing substantially from this principle
It is whole, or some steps of addition deletion.For example, the step of structure averaging model 530, can be removed.In another example adjusting module 470
One or both of above-mentioned several deformations, or micro- change using other forms can be only carried out to grid model.It is all such
The deformation of class, within the protection domain of the application.
Fig. 6 is the schematic diagram of the example model structure module according to some embodiments of the present application.Model construction mould
Block 440 can include a registration unit 620, one of acquiring unit 610, one and mark 630, model generation units of unit
640 and an association factor generation unit 650.Connected mode in the model construction module 440 between each module can be
Wired, wireless or both combination.Any one module can be local, long-range or both combination.
Acquiring unit 610 can be used for obtaining multiple reference models.Acquiring unit 610 can pass through database 120, control
And storage device outside processing system 100 or the mode of user's input obtain above- mentioned information.Its function can pass through the place in Fig. 2
Device 220 is managed to realize.In certain embodiments, the multiple reference model can include a patient in different time, difference
Position, the cardiac image data of different angle scanning.In certain embodiments, more set cardiac datas can include different diseases
The cardiac image data that people is scanned in diverse location, different angle.In some embodiments, acquiring unit 610 can be used for obtaining
Take the information such as modeling algorithm, parameter.The multiple reference models and/or other information of acquisition can be sent to by acquiring unit 610
Registration unit 620, mark unit 630, averaging model generation unit 640 or association factor generation unit 650.
Registration unit 620 can be used for adjusting by method for registering images multiple refers to mould acquired in acquiring unit 610
Type, and make the position of multiple reference models, ratio etc. consistent.Described image registration can be included based on space dimensionality registration, base
In feature registration, based on Transformation Properties registration, based on optimized algorithm registration, based on image modalities registration, based on main body registration etc..
In certain embodiments, multiple reference models can be registrated in an identical coordinate system by registration unit 620.Registration unit
Multiple reference models after registration can be sent to memory module 430 by 620 to be stored, and can also be sent to mark unit
630 and/or averaging model generation unit 640 further handle.
Mark unit 630 can be used for the multiple data points of cavity margin (the also referred to as point for marking multiple reference models
Collection).The cardiac image or model can be registration unit 620 carry out image registration after multiple reference models or
The averaging model that averaging model generation unit 640 is built.For example, cavity margin can carry out figure by user in registration unit 620
As being marked manually on multiple reference models after registration.In another example cavity margin can be by mark unit 630 according to significantly different
Cavity margin feature automatic marking.In certain embodiments, marking unit 630 can be by the whole heart in multiple reference models
Dirty image or model are divided into six parts according to chamber, respectively left ventricle, atrium sinistrum, right ventricle, atrium dextrum, cardiac muscle and
Sustainer.In certain embodiments, mark unit 630 (can also be claimed according to the intensity of variation at reference model upper chamber edge
For gradient), it is sharp keen class and non-sharp keen class by the whole cardiac image or model partition on multiple reference models.Specifically, mark
Note unit 630 can be communicated with the outside the marginal point of several chambers or be class with the less mark of external change degree,
To be connected with internal other chambers or be sharp keen class with more mark of external change, such as two arrow institutes in Figure 17
Show.For example, the whole cardiac image or model partition on multiple reference models can be 10 classifications by mark unit 630:It is left
Ventricle edge, atrium sinistrum sharpened edge, the non-sharpened edge in atrium sinistrum, right ventricle sharpened edge, the non-sharpened edge of right ventricle, the right heart
Room sharpened edge, the non-sharpened edge in atrium dextrum, sustainer edge, left myocardium sharpened edge and the left non-sharpened edge of cardiac muscle are (as schemed
Shown in 18).In certain embodiments, unit 630 is marked by multiple reference model registrations into a same coordinate system to lead to
The position of more multiple reference models and each point on the averaging model of the gained of averaging model generation unit 640 is crossed, marks multiple ginsengs
Examine the cavity margin on model.For example, mark unit 630 can will be closest with corresponding points on reference model on averaging model
Point belonging to classification of the classification as the point on reference model.Mark unit 630 can will be labeled with cavity margin point set
Multiple reference models are sent to memory module 430 and stored, and can also be sent to training module 450, averaging model generation list
Member 640 and/or association factor generation unit 650 are further handled or for calculating.
Averaging model generation unit 640 can be used for building three-dimensional cardiac average meshes model.In certain embodiments, put down
Equal model generation unit 640 can extract the cavity margin in multiple reference models or averaging model after marking, by every
Cavity margin model in individual reference model or averaging model carries out processing and obtains multiple grid of reference models, and passes through image mould
Average meshes model is calculated in type construction method.Described image model building method can include Point Distr
Ibution Model (PDM), Active Shape Model (ASM), Active Contour Model are (also referred to as
Snakes), Active Appearance Model (AAM) etc..In certain embodiments, averaging model generation unit 640 can be with
Whole heart averaging model after chamber is marked is divided into six submodels that are independent or be combineding with each other.For example, the left heart
(as shown in figure 20) such as room model, left atrial model, right ventricle model, atrium dextrum model, left myocardial model and sustainer models.
In certain embodiments, averaging model generation unit 640 can extract multiple cavity margins, and determine on multiple cavity margins
Control point is distributed, and network is formed by connecting control point.In certain embodiments, averaging model generation unit 640 can be based on
Grid model obtains the average meshes model of heart chamber, and corresponding characteristic value, characteristic vector etc. by ASM modeling methods
Model parameter.In certain embodiments, averaging model generation unit 640 can add association factor pair in averaging model calculating
The influence at control point.For example, in ASM calculating, averaging model generation unit 640 can utilize weighted average (i.e. Σ (Fi*
Wi)) the adjustment result at control point is calculated, wherein, Fi is the deformation parameter of some chamber, and Wi is shadow of the chamber to control point
Ring coefficient or weighted value.By the weighted average calculation based on association factor can on model control point adjustment by
To the influence of multiple chamber results, so as to reach the purpose for associating multiple chambers.Averaging model generation unit 640 will can obtain
Three-dimensional cardiac average meshes model be sent to memory module 430 stored or association factor generation unit 650 be used for calculate.
Averaging model generation unit 640 obtained three-dimensional cardiac average meshes model can also be sent to training module 450 and/or
Further handled with module 460.
Association factor generation unit 650 can be used for the relation for establishing control point on each chamber and average grid model.
In some embodiments, the relation can be the two-dimentional association factor matrix of chamber and control point as ranks, and the value of matrix can
To represent influence coefficient or weight of each chamber to each control point.In certain embodiments, the value of the matrix can be 0-1 it
Between any real number.
In certain embodiments, association factor generation unit 650 can belong to according to the chamber at control point on grid model,
And control point and the position relationship of other chambers, establish association factor matrix.In certain embodiments, association factor generation is single
Member 650 can calculate the coverage of association factor according to the distance of control point and other chambers or influence coefficient.For example, association
Factor generation unit 650 can control the meter of association factor influence coefficient by control point apart from the ultimate range of other chambers
Calculate.In certain embodiments, association factor generation unit 650 can adjust different chamber according to the tightness degree between each chamber
Between coverage and influence coefficient.As shown in figure 21, in grid control point model, light control point is represented only by place
The influence of chamber, and dark chamber junction then represents that control point is influenceed by the chamber of multiple connections, wherein color is deeper
Representative is influenceed bigger by other chambers.Obtained two-dimentional association factor matrix can be sent to by association factor generation unit 650
Memory module 430 is stored, and can also be sent to averaging model generation unit 640 and/or adjusting module 470 based on weighting
Calculate.
It should be noted that the above-mentioned description for model construction module 440, only for convenience of description, can not be this Shen
It please be limited within the scope of illustrated embodiment.It is appreciated that for those skilled in the art, understanding the work of the module
After making principle, unit in the module may be combined in the case of without departing substantially from this principle, or formed
Subsystem is connected with other units, makees various modifications and variations in the form and details to implementing above-mentioned module.For example, registration is single
Member 620 and/or mark unit 630 can remove, or merge with acquiring unit 610, memory module 430.It is in another example described more
Individual reference model or averaging model can include the cardiac data or model for having carried out edge mark via user.In another example institute
State the cardiac data that multiple reference models or averaging model can include having carried out rough or fine chamber segmentation.Such as
Such deformation, within the protection domain of the application.
Fig. 7 is the exemplary process diagram of the structure averaging model according to some embodiments of the present application.In step 710
In, multiple heart reference models can be obtained.The multiple heart reference model can by database 120, user input or
Storage device outside control and processing system 100 obtains.In certain embodiments, the multiple heart reference model can include
The cardiac image data that one patient is scanned in different time, diverse location, different angle.In certain embodiments, it is described more
Individual heart reference model can include the cardiac image data that different patients are scanned in diverse location, different angle.In some realities
Apply in example, the multiple heart reference model can include the cardiac data or model for having carried out edge mark via expert.
In some embodiments, the multiple heart reference model can include the heart for having carried out rough or fine chamber segmentation
Data.
In step 720, image registration can be carried out to multiple reference models of acquisition.The step can be by model construction
Registration unit 620 in module 440 is completed.In certain embodiments, can by translate, rotate, the mode such as scale will be any
Two reference models are transformed in the same coordinate system, and cause the point for corresponding to space same position in above-mentioned two reference model
Correspond, so as to realize information fusion.In certain embodiments, described image registration can include matching somebody with somebody based on space dimensionality
Standard, feature based are registering, match somebody with somebody based on Transformation Properties registration, based on optimized algorithm registration, based on image modalities registration, based on main body
Standard etc..Wherein, it is described that 2D/2D registrations, 2D/3D registrations or 3D/3D registrations can be included based on space dimensionality registration.It is described to be based on
Feature registration can include distinguished point based (such as the turning point of discontinuity point, figure, line crosspoint etc.) registration, based on face area
Domain (such as curve, curved surface etc.) registration, based on pixel value registration, based on surface registration etc..It is described to be matched somebody with somebody based on Transformation Properties
Will definitely with including based on rigid transformation registration, based on affine transformation registration, based on projective transformation registration and/or based on curve convert
Registration etc..It is described based on optimized algorithm registration can include based on gradient descent method registration, based on Newton method registration, be based on
Powell methods registration, based on genetic algorithm registration etc..It is described based on image modalities registration can include based on single mode registration and/
Or based on multi-mode registration.It is described based on main body registration can include based on the image registration from same patient, based on from
Different patient image registrations and/or the registration based on patient data and collection of illustrative plates.
In step 730, cavity margin is marked on multiple reference models that can be after registration.The step 730 can be by
Mark unit 630 in model construction module 440 is completed.In certain embodiments, can be by being joined by user in multiple hearts
Examine and mark cavity margin point on model manually, heart can be divided into six portions by the edge point set formed on each reference model
Divide, respectively left ventricle, atrium sinistrum, right ventricle, atrium dextrum, cardiac muscle and sustainer.In certain embodiments, can be according to chamber
Heart is divided into 10 classifications by edge relative to intensity of variation outwardly and inwardly:Left ventricle edge, the sharp keen side in atrium sinistrum
Edge, the non-sharpened edge in atrium sinistrum, right ventricle sharpened edge, the non-sharpened edge of right ventricle, atrium dextrum sharpened edge, atrium dextrum are non-sharp
Sharp edge edge, sustainer edge, left myocardium sharpened edge and the left non-sharpened edge of cardiac muscle (as shown in figure 18).The sharpened edge can
It is communicated with the outside or changed unobvious with referring to the edge of chamber.The non-sharp keen edge that can refer to chamber with it is internal or other
Chamber is connected or change is obvious.
In step 740, it may be determined that the control point on multiple reference models.The step can be by model construction module
Averaging model generation unit 640 in 440 is completed according to the multiple reference models marked by image registration and cavity margin.
In some embodiments, each chamber can be determined according to the image registration results and cavity margin markup information of multiple reference models
Axle.The axle can be 2 points of line direction being arbitrarily designated on chamber.For example, identified axle can be on chamber away from
The major axis formed from 2 points farthest of line.In certain embodiments, after multiple reference model marks can be extracted respectively
Cavity margin, the cross-sectional direction of identified axis is cut into slices to each chamber along each chamber, and according to cross section
Intensive point set is formed in slicing edge with curved surface features, forms the point model (as shown in figure 19) of averaging model.In some realities
Apply in example, the control point on each chamber can be determined according to point model.The control point can be the son of point set on point model
Collection.For example, the subset is bigger, grid model is bigger, and the amount of calculation during cardiac segmentation is bigger, and segmentation effect is better;Institute
The subset of selection is smaller, and grid model is smaller, and the amount of calculation during cardiac segmentation is smaller, and splitting speed is very fast.In some realities
Apply in example, the number at control point can change on chamber.For example, roughly segmentation the stage, control count out can with less, so as to
Fast positioning is to cavity margin;In the fine segmentation stage, control is counted out can be with more, so as to realize fine point of cavity margin
Cut.
In step 750, heart average meshes model can be built according to control point.In certain embodiments, step 750
Difference can be connected into by polygonal network according to the relation between control point.For example, in certain embodiments, it can pass through
The adjacent control point connected on contiguous slices forms triangular net.In certain embodiments, image distortion method can be passed through
Obtain average meshes model.Described image deformation method can include Point Distribution Model (PDM), Active
Shape Model (ASM), Active Contour Model (also referred to as Snakes), Active Appearance Model
(AAM) etc..For example, in certain embodiments, the triangular net that can be built based on control point is obtained by ASM computational methods
The average meshes model (as shown in figure 20) of multiple heart reference models.In certain embodiments, step 750 can be based on two dimension
Association factor matrix is weighted averaging model calculating to net of control points lattice model.For example, in ASM calculating, averaging model life
The adjustment result at control point can be calculated using weighted average (i.e. Σ (Fi*Wi)) into unit 640, wherein, Fi is some chamber
The deformation parameter of room, Wi are influence coefficient or weighted value of the chamber to control point.
It should be noted that the description of the above-mentioned process that averaging model is built for model construction module 440, is only described
It is convenient, the application can not be limited within the scope of illustrated embodiment.It is appreciated that come for those skilled in the art
Say, after the operation principle of the module is understood, the order of each step may be carried out in the case of without departing substantially from this principle
Some steps are deleted in any adjustment, or addition.For example, step 710 and step 720 can merge.In another example step 730 arrives
Step 750 can circulate repeatedly.Such deformation, within the protection domain of the application.
Fig. 8 is the schematic diagram of the example training module according to some embodiments of the present application.Training module 450 can be with
Including a taxon 810 and a grader generation unit 820.In the model construction module 440 between each module
Connected mode can be wired, wireless or both combination.Any one module can be local, long-range or two
The combination of person.
Taxon 810 can be used for the possibility cavity margin point on multiple reference models or averaging model being divided into not
In same chamber classification.The function can be realized by processor 220.In certain embodiments, taxon 810 can basis
The chamber classification that mark unit 630 divides is classified (such as Figure 22 institutes to possible marginal point on reference model or averaging model
Show).For example, possible marginal point near reference model or averaging model upper chamber can be divided into 10 by taxon 810
In chamber classification, it is respectively:Left ventricle edge, atrium sinistrum sharpened edge, the non-sharpened edge in atrium sinistrum, right ventricle sharpened edge,
The non-sharpened edge of right ventricle, atrium dextrum sharpened edge, the non-sharpened edge in atrium dextrum, sustainer edge, left myocardium sharpened edge and
The left non-sharpened edge of cardiac muscle.The classification can be realized by a variety of sorting techniques, including but not limited to Decision Tree Algorithm,
Bayes (Bayes) sorting algorithm, artificial neural network (ANN) sorting algorithm, k- neighbouring (kNN), SVMs (SVM),
Sorting algorithm based on correlation rule, integrated study sorting algorithm etc..In certain embodiments, taxon 810 can basis
Point near cavity margin is divided into positive sample and negative sample by the distance of point and cavity margin near cavity margin.For example,
The positive sample can be the data point in the certain threshold range of cavity margin, the negative sample can be apart from edge compared with
The data point of other random sites in remote and space.In certain embodiments, taxon 810 can be by multiple reference models
Or the classification results of probable edge point or data are sent to memory module 430 and stored on averaging model, can also be sent to
Grader generation unit 820 is further handled.
Grader generation unit 820 can be used for obtaining the grader trained.In certain embodiments, grader generates
Unit 820 can be according to the marginal point classification that taxon 810 divides to the marginal point on multiple reference models or averaging model
Carry out classifier training, and the grader (as shown in figure 23) trained.In certain embodiments, grader generation unit
820 can utilize PBT training graders.In certain embodiments, the grader trained can receive any one seat
After punctuate, probability corresponding to the coordinate points is exported.The probability refers to probability of the certain point as cavity margin.In some implementations
In example, the grader trained can be sent to memory module 430 and be stored by grader generation unit 820, can also be sent out
Give matching module 460 and/or adjusting module 470 is used to calculate.
It should be noted that the above-mentioned description for training module 450, only for convenience of description, can not limit the application
System is within the scope of illustrated embodiment.It is appreciated that for those skilled in the art, it is former in the work for understanding the module
After reason, unit in the module may be combined in the case of without departing substantially from this principle, or form subsystem
System is connected with other units, makees various modifications and variations in the form and details to implementing above-mentioned module.For example, taxon
810 can carry out chamber division to multiple reference models or averaging model, the chamber classification after division is divided relative to mark
Chamber classification it is finer.Such deformation, within the protection domain of the application.
Fig. 9 is the exemplary process diagram of the training grader according to some embodiments of the present application.In step 910,
Taxon 810 in training module 450 can obtain the sample point in multiple reference models or averaging model.In some implementations
In example, training module 450 can extract chamber based on the chamber segmentation result on the multiple reference models or averaging model after mark
Room edge (as shown in figure 22), and a range of point is as positive sample using near each cavity margin, apart from cavity margin
The point of other random sites is as negative sample farther out and in space.For example, the cavity margin certain limit can be
0.1cm, 0.5cm, 1cm, 2cm etc..
Taxon 810 in step 920, training module 450 can classify to the positive and negative sample point of acquisition.
In some embodiments, positive and negative sample point can be added in different chamber classifications by training module 450 according to sorting technique.
In some embodiments, positive sample can be a range of point in averaging model edge, and negative sample can be some averaging models
Point outside edge certain limit.In certain embodiments, the certain limit at averaging model edge could be arranged to zero, now positive sample
This is averaging model marginal point.In certain embodiments, positive negative sample can be based on sharp keen degree and sample point present position
Classified.In certain embodiments, sample point present position is the affiliated chamber of positive negative sample.For example, training module 450 can be with
According to the chamber classification of mark, positive and negative sample point is divided into 10 chamber classifications:Left ventricle edge, the sharp keen side in atrium sinistrum
Edge, the non-sharpened edge in atrium sinistrum, right ventricle sharpened edge, the non-sharpened edge of right ventricle, atrium dextrum sharpened edge, atrium dextrum are non-sharp
Sharp edge edge, sustainer edge, left myocardium sharpened edge and the left non-sharpened edge of cardiac muscle.The sorting technique can include decision tree
Sorting algorithm, Bayes (Bayes) sorting algorithm, artificial neural network (ANN) sorting algorithm, k- neighbouring (kNN), supporting vector
Machine (SVM), the sorting algorithm based on correlation rule, integrated study sorting algorithm etc..Wherein, Decision Tree Algorithm can include
ID3, C4.5, C5.0, CART, PUBLIC, SLIQ, SPRINT algorithm etc..Bayesian Classification Arithmetic can include naive Bayesian
Algorithm, TAN algorithms (tree augmented Bayes network)) etc..Artificial neural network sorting algorithm can include BP
Network, radial direction base RBF networks, Hopfield networks, stochastic neural net (such as Boltzmann machines), Competitive ANN (example
Such as Hamming networks, self-organized mapping network) etc..Sorting algorithm based on correlation rule can include CBA, ADT, CMAR
Deng.Integrated study sorting algorithm can be including Bagging, Boosting, AdpBoosting, PBT etc..
In step 930, training module 450 can obtain the grader by classification based training.In certain embodiments, train
Grader generation unit 820 in module 450 can be by the above-mentioned sample point classification of PBT Algorithm for Training, and obtains one or more
The individual grader (as shown in figure 23) trained.The PBT can include two-stage PBT algorithms or multistage PBT algorithms.In some realities
Apply in example, the grader can include one or more with a range of point of multiple reference models or averaging model edge
The grader (also referred to as " the first grader ") for training to obtain for positive sample.In certain embodiments, the grader can wrap
It using a range of point in pending image border is that the grader that positive sample trains to obtain (is also referred to as " the to include one or more
Two graders ").
It should be noted that the description of the above-mentioned process for the training grader of training module 450, only for convenience of description,
The application can not be limited within the scope of illustrated embodiment.It is appreciated that for those skilled in the art,
After the operation principle for solving the module, any tune may be carried out to the order of each step in the case of without departing substantially from this principle
It is whole, or some steps of addition deletion.For example, positive sample and negative sample can not be differentiated between in step 910 and step 920, and directly
Connect and all points near cavity margin are classified.In another example positive and negative sample point can be with apart from the ultimate range of cavity margin
It is 2cm.Such deformation, within the protection domain of the application.
Figure 10 is the structural representation of the example model matching module according to some embodiments of the present application.Such as Figure 10
Shown, matching module 460 can include an acquiring unit 1010, a picture point extraction unit 1020, a Hough transformation
Unit 1030 and a Model Matching unit 1040.Connected mode in the matching module 460 between each unit can be had
Line, wireless or both combination.Any one unit can be local, long-range or both combination.
Acquiring unit 1010 can obtain image.The image of the acquisition is pending image.In certain embodiments, institute
It can be the image rebuild based on view data to state image.The image of the reconstruction can be from other modules of processing equipment 130
Middle acquisition.Obtained for example, the image of the reconstruction can be acquiring unit 1010 from image reconstruction module 420.For another example institute
The image for stating reconstruction can be the image being stored in after the reconstruction image of image reconstruction module 420 in memory module 430.In some realities
Apply in example, described image can be the image being input to via external equipment in system.For example, external equipment passes through COM1
250 input an image into system.In certain embodiments, acquiring unit 1010 can obtain averaging model.The average mould
Type can be the three-dimensional cardiac average meshes model that averaging model generation unit 640 generates.In certain embodiments, acquiring unit
1010 can obtain the first grader that training module 450 trains.
In certain embodiments, acquiring unit 1010 can obtain required during the progress images match of model fitting module 460
The parameter wanted.For example, acquiring unit 1010 can obtain the parameter for generalised Hough transform.In certain embodiments, it is described
The parameter of generalised Hough transform can be based on three-dimensional average meshes model and its cavity margin control point obtains.For example, by true
Allocate the barycenter of equal edge of model, calculate all control points on averaging model edge relative to the offset of barycenter and relative to
The gradient direction of barycenter, can obtain corresponding to each gradient direction control point offset vector (hereinafter referred to as gradient to
Amount).In certain embodiments, averaging model can be placed in x-y-z coordinate system, and determines that each gradient vector is sat in x-y-z
Coordinate under mark system.In certain embodiments, the coordinate of each gradient vector can be converted to the coordinate under polar coordinate system.Specifically
Ground, can using gradient vector in the projection of x-y plane and the angle of x coordinate axle as first angle, θ, span is -180
Spend 180 degree.Can be using the angle of gradient vector and x-y plane as second angle φ, span is -90 degree to 90
Degree.In certain embodiments, sliding-model control can be carried out to two angle, θs and φ of above-mentioned expression gradient vector, obtained such as
Under the form (being also known as R-table).In certain embodiments, the offset on R-table can be zoomed in and out or
Different angles is rotated to detect the shape of different size or different angle.
Gradient related angle φ, θ | The offset of corresponding points |
0,90 | (x0,y0,z0),(x3,y3,z3),… |
0,80 | (x2,y2,z2),(x5,y5,z5),… |
… | … |
10,90 | (x4,y4,z4),(x6,y6,z6),… |
… | … |
Picture point extraction unit 1020 can obtain the marginal probability figure of pending image.Specifically, in certain embodiments,
Picture point extraction unit 1020 can be by by the grader that obtains of the coordinate put on pending image input acquiring unit 1010
In, probability of each point as cavity margin on pending image is calculated, and obtain according to the probability distribution of each point pending
The marginal probability figure of image.In certain embodiments, the marginal probability figure can include shade of gray figure, color gradient figure
(as shown in figure 24) etc..In certain embodiments, picture point extraction unit 1020 can be by pending image border probability graph
Probable value is more than the point of certain threshold value as first edge point.The threshold value can be any real number between 0-1, for example,
0.3rd, 0.5 etc..
Model Matching unit 1030 can match averaging model on pending image.Specifically, in some embodiments
In, the edge that Model Matching unit 1030 can match pending image by weighting generalised Hough transform by averaging model is general
On rate figure.The weighting generalised Hough transform, which can include being obtained according to first edge point on pending image and R-table, to be treated
All possible edge reference point on image is handled, the probability that all edge reference points are obtained by the method for weighted accumulation adds up
Value, and the barycenter using the maximum edge reference point of probability accumulated value as image.The conversion of model barycenter to image centroid is joined
Transformation parameter of the number as model.The edge reference point can be by pending image first edge point according in R-table
Parameter carry out coordinate transform after obtain.The weighted accumulation can (will refer to first edge positioned at identical edge reference point
Point falls on the behavior of same edge reference point after being shifted according to the parameter on R-table) corresponding to first edge point probability
Cumulative process.According to the image centroid of acquisition, can according to transformation parameter by the centroid transformation of model to image centroid weight
The position of conjunction.The transformation parameter can include the anglec of rotation and scaling etc..In certain embodiments, Model Matching unit
1030 can be rotated according to the transformation parameter of determination to the point on model, scaling is handled etc., so as to obtain and pending figure
As the model (as shown in figure 25) of matching.
It should be noted that the above-mentioned description for model fitting module 460, only for convenience of description, can not be this Shen
It please be limited within the scope of illustrated embodiment.It is appreciated that for those skilled in the art, understanding the work of the module
After making principle, unit in the module can be combined in the case of without departing substantially from this principle, or formed
Subsystem is connected with other units, makees various modifications and variations in the form and details to implementing above-mentioned module.For example, picture point
Extraction unit 1020 can remove, and the marginal probability figure of pending image can be obtained directly by training module 450.It is such
Deformation, within the protection domain of the application.
Figure 11 is the matching averaging model and the exemplary stream for the image rebuild according to some embodiments of the present application
Cheng Tu.In step 1110, averaging model, pending image and the second grader trained can be obtained.In some implementations
In example, the averaging model can be that averaging model generation unit 640 passes through iconic model structure side based on multiple reference models
The three-dimensional cardiac average meshes model that method obtains.Described image model building method can include Point Distribution
Model (PDM), Active Shape Model (ASM), Active Contour Model (also referred to as Snakes), Active
Appearance Model (AAM) etc..Step 1110 can be realized by acquiring unit 1010.In certain embodiments, obtain single
The pending image that member 1010 obtains can be the image that image reconstruction module 420 is rebuild.In certain embodiments, step 1110
The R-table based on averaging model can be obtained.
In step 1120, it may be determined that the parameter of generalised Hough transform.Specifically, in certain embodiments, step
1110 can the marginal probability figure based on pending image obtain the first edge point of pending image.The first edge point can
To be probability is more than certain threshold value on pending image border probability graph point, such as the probability can be 0.3.In some realities
Apply in example, the marginal probability figure can be by by the classification that obtains of the coordinate put on pending image input acquiring unit 1010
Probability of each point as cavity margin on pending image is calculated in device, and is obtained according to the probability distribution of each point.In some realities
Apply in example, angle, θ and φ corresponding to first edge point gradient direction on pending image can be calculated, and it is true according to R-table
Determine the offset of first edge point, be possible to side is used as using the difference of the coordinate value of first edge point and all corresponding offsets
The coordinate value of edge reference point.It is possible to further the probable value of voted according to edge reference point number and corresponding first edge point
All edge reference points are weighted cumulative.The weighted accumulation can be by positioned at first corresponding to same edge reference point
The probability of marginal point adds up.In certain embodiments, can be by the R- corresponding to the maximum edge reference point of probability accumulated value
Transformation parameter of the parameter as pending image in table.The transformation parameter can include the anglec of rotation and scaling
Deng.The method formula of the weighted accumulation can be expressed as:
Wherein, i is the index of first edge point, and j is the index for the probable edge reference point being voted-for on ballot image, and p is every
The probable value of individual first edge point, the two-valued functions of σ 0,1, i.e., when i-th of first edge o'clock is in j-th of probable edge reference point
When having ballot contribution, the value is 1, is otherwise 0.
In step 1130, model corresponding to pending image can be obtained.Specifically, can be weighted based on determined by
Generalised Hough transform parameter, line translation is clicked through to the first edge on pending image.For example, can be according to edge reference point pair
Angle and scaling in the R-table answered, the coordinate of first edge point on pending image is converted, and on averaging model
Corresponding information correspond on pending image, obtain pending image corresponding with average meshes model.
Figure 12 is the structural representation of the example adjusting module according to some embodiments of the present application.Such as Figure 12 institutes
Show, the adjusting module 470 can include an acquiring unit 1210, a target point determining unit 1220, and a model becomes
Change unit 1230.Connected mode in the adjusting module 470 between each unit can be wired, wireless or both knot
Close.Any one unit can be local, long-range or both combination.
Acquiring unit 1210 can obtain model and the second grader trained.Specifically, acquiring unit 1210 can be with
Obtain the coordinate data of second edge point on model.In certain embodiments, the second edge point of the model can be model
On control point.In certain embodiments, acquiring unit 1210 can obtain the second grader that training module 450 trains.
The grader can be that 10 chamber classifications based on chamber and the division of clear-cut margin degree are obtained by PBT classification algorithm trainings
To 10 graders, such as left ventricle edge, atrium sinistrum sharpened edge, the non-sharpened edge in atrium sinistrum, right ventricle sharpened edge, the right side
The non-sharpened edge of ventricle, atrium dextrum sharpened edge, the non-sharpened edge in atrium dextrum, sustainer edge, left myocardium sharpened edge and a left side
The non-sharpened edge of cardiac muscle.Because the grey scale change unobvious inside and outside some cavity margin, sharp keen degree is relatively low, because
This, does not classify to it according to sharp keen degree.In certain embodiments, acquiring unit 1210 can be obtained by model converter unit
Model after 1230 processing.
Target point determining unit 1220 can determine target point corresponding to second edge point on model.With one on model
Exemplified by second edge point, target point determining unit 1220 can determine multiple candidates around one model second edge point
Point.In certain embodiments, target point determining unit 1220 can be by around one model second edge point of determination
Multiple candidate points are input in the grader of the acquisition of acquiring unit 1210, determine one model second edge point and its surrounding
Multiple candidate points correspond to the probability of image border, and according to the target of the one model second edge point of the determine the probability
Point.In certain embodiments, target point determining unit 1220 can determine the corresponding target point of all second edge points on model.
Model converter unit 1230 can be adjusted to model.In certain embodiments, model converter unit 1230 can
With the position of the target point adjustment model edge point based on determined by target point determining unit 1220.The adjustment can include phase
Like property conversion, piecewise affine transformations and/or micro- change based on energy function etc..In certain embodiments, model converter unit
1230 can repeatedly adjust model, and each adjustment is required to redefine target point.Specifically, in some embodiments
In, whether model converter unit 1230 meets preparatory condition after may determine that model adjustment.For example, whether model adjustment number reaches
To certain threshold value.If model adjustment number reaches certain threshold value, the model accurately matched is exported;If model adjustment number is less than
The predetermined threshold value, then target point determining unit 1220 is sent a signal to, carry out the determination of target point again, then become by model
Change the conversion that unit 1230 carries out model edge point again.In certain embodiments, model converter unit 1230 can obtain essence
The really heart chamber model after adjustment.Heart chamber model after the accurate adjustment can be very close with true heart.
It should be noted that the above-mentioned description for adjusting module 470, only for convenience of description, can not limit the application
System is within the scope of illustrated embodiment.It is appreciated that for those skilled in the art, it is former in the work for understanding the module
After reason, unit in the module may be combined in the case of without departing substantially from this principle, or form subsystem
System is connected with other units, makees various modifications and variations in the form and details to implementing above-mentioned module.For example, model conversion is single
Member 1230 can preset cycle-index, the circulation time without determining accurate adjusting module 470 by threshold decision
Number.Such deformation, within the protection domain of the application.
Figure 13 is the exemplary process diagram of the adjustment model according to some embodiments of the present application.In step 1310
In, the point of the second edge on model and the grader trained can be obtained.In certain embodiments, acquiring unit 1210 and obtain
The grader for taking unit 1010 to obtain is not same type.The grader that the acquiring unit 1010 obtains can be training module
450 take a range of point in average grid model edge to train to obtain for positive sample.Point that the acquiring unit 1210 obtains
Class device can take a range of point in pending image border to train to obtain for positive sample.In certain embodiments, obtain
The grader that unit 1010 obtains can be the first grader, and the grader that acquiring unit 1210 obtains can be the second classification
Device.
In step 1320, the target point of second edge point on model can be determined based on the second grader.In some realities
To apply in example, an a range of candidate point of model second edge point can be input in the second grader by step 1320,
And obtain the probability that a range of candidate point of the model second edge point belongs to image border.In certain embodiments,
The target point of one model second edge point can be determined by target point determining unit 1220 based on identified probability.
In step 1330, can the target point based on determination by the second edge point transformation on model to target point.
In some embodiments, step 1330 can use a variety of mapping modes to click through line translation to model second edge.For example, it can lead to
Model converter unit 1230 is crossed to be modified model second edge point using similitude change and affine transformation.
In step 1340, it can be determined that whether adjustment result meets preparatory condition.In certain embodiments, preparatory condition
It can be whether adjustment number reaches certain threshold value.In certain embodiments, the threshold value is adjustable.When adjustment number
When reaching certain threshold value, into step 1350, and the model after accurate matching is exported;When adjustment number is less than certain threshold value,
Return to step 1320, new model edge point pair can be determined by target point determining unit 1220 based on new model edge point
The target point answered.
Figure 14 is the exemplary process diagram to be set the goal really according to some embodiments of the present application a little.Flow 1400 can
To be that target point determining unit 1220 is realized.The process of the respective objects point of any on averaging model edge is to determine in Figure 14,
It will be understood by those skilled in the art that this method can be used for obtaining multiple target points corresponding to multiple marginal points.
In certain embodiments, flow 1400 can be corresponding with step 1320.
In step 1410, it may be determined that the normal of an averaging model marginal point.In certain embodiments, the normal
Direction be that outside is internally pointed to by averaging model.Specific normal acquisition methods may refer to, for example, flow 1500 and its retouching
State.
In step 1420, step-length and search model along one averaging model marginal point normal direction can be obtained
Enclose.In certain embodiments, the step-length and hunting zone can be pre-set values.In certain embodiments, it is described
Step-length and hunting zone can be that user inputs.For example, user can be inputted everywhere by external equipment by COM1 250
Manage in equipment 130.In certain embodiments, the hunting zone is using one model edge point as starting point, along normal institute
The line segment at least one direction in straight line both direction (on the outside of to model or inner side).
In step 1430, based on step-length and hunting zone, it may be determined that one or more candidate points.For example, search model
Enclose for 10 centimetres, step-length is arranged to 1 centimetre, can respectively determine 10 points along straight line both direction where normal, totally 21 candidates
Point (including marginal point is in itself).In some embodiments, it is also possible to determine step-length and step number, and determine to wait according to step-length and step number
Reconnaissance.For example, step-length is arranged to 0.5 centimetre, step number is arranged to 3, can respectively determine 3 along straight line both direction where normal
Point, farthest candidate point is apart from marginal point 1.5cm, totally 7 candidate points.
In step 1440, it may be determined that it is a range of general that one or more of candidate points correspond to image border
Rate.In certain embodiments, the second grader is to take a range of point in image border to train to obtain for positive sample.Described one
Surely a preset value is may range to be set by machine or user.For example, the preset value can be 1 centimetre.
In step 1450, it is a range of general corresponding to image border that one or more of candidate points can be based on
Rate, it is target point to determine one in one or more of candidate points.In certain embodiments, target point can be based on following
Function obtains:
Fi=max (Pi-λ*di 2) (2)
Wherein, PiCorrespond to a range of probability in image border for candidate point;diFor candidate point and one average mould
The Euclidean distance of type marginal point;λ is weight, is constant to equilibrium distance and the relation of probable value.
In certain embodiments, multiple target points of multiple model edge points can be determined based on flow 1400, then
Line translation is entered to multiple model edge points and model according to the multiple target point.Specific conversion process may refer to, for example,
Figure 16 and its description.
Figure 15 is the exemplary process diagram of the determination marginal point normal according to some embodiments of the present application.At some
In embodiment, flow 1500 can be corresponding with step 1420.
In step 1510, multiple polygons can be determined according to multiple marginal points of averaging model.In some embodiments
In, the multiple polygon can be formed by connecting the multiple marginal point.The multiple polygon can be triangle, four
The shapes such as side shape, polygon.In certain embodiments, the process for determining multiple polygons according to multiple marginal points is also referred to as
Gridding is handled.Wherein, the multiple polygon can be referred to as grid, and the multiple marginal point can be referred to as node.
In some embodiments, averaging model surface may form multiple polygons corresponding with the averaging model marginal point,
In this case, step 1510 can be omitted.
In step 1520, it may be determined that adjacent multiple polygons with an averaging model marginal point.
In step 1530, it may be determined that multiple normals corresponding to the affiliated plane of the multiple polygon.In some realities
Apply in example, it is (on the outside of averaging model or interior that multiple normal directions corresponding to the affiliated plane of the multiple polygon are located at homonymy
Side).In certain embodiments, multiple normal line vectors corresponding to the affiliated plane of the multiple polygon are unit vector.
In step 1540, the normal of the marginal point can be determined based on the multiple normal.In certain embodiments,
Multiple normal line vectors corresponding to the multiple polygon can be added or are averaged.
Figure 16 is the exemplary process diagram of the conversion averaging model marginal point according to some embodiments of the present application.
In some embodiments, flow 1600 can be that model converter unit 1230 is realized.
In step 1610, similarity transformation can be performed to averaging model marginal point.For example, can be by averaging model side
The grid of edge point composition the target point direction determined according to cavity margin point, is integrally carried out as an entirety to averaging model
Conversion, it is main to include the operations such as translation, rotation, scaling.
In step 1620, piecewise affine transformations can be performed to averaging model marginal point.In certain embodiments, it is average
The grid of model edge point composition can be divided according to certain rule.For example, can be according to heart chamber to heart mould
Type is divided.As shown in figure 24, model meshes can be divided into left ventricle, atrium sinistrum, right ventricle, the right heart according to chamber
Room, sustainer and left myocardium six parts.In certain embodiments, piecewise affine transformations are referred to the various pieces of division
Grid carry out affine transformation respectively.The affine transformation can refer to moves conversion respectively to multiple nodes of various pieces
And shape conversion.In certain embodiments, averaging model marginal point may be influenceed by multiple chambers.Averaging model marginal point
The effect influenceed by different chamber can be showed in the form of association factor.When carrying out affine transformation, averaging model
Marginal point can be changed towards target point.During conversion, averaging model marginal point by multiple chambers due to being influenceed.Close
Join the weighted value that the factor can turn into conversion parameter (such as moving displacement, proportion of deformation etc.).According to corresponding to marginal point target point and
Association factor, model converter unit 1230 are changed the marginal point on the more segment mesh of averaging model using piecewise affine transformations respectively
To its corresponding position.
In step 1630, micro- change based on energy function can be performed to averaging model marginal point.In some embodiments
In, energy function can be expressed as:
Wherein, EextFor external energy, represent current point and detect the relation of target point;EintFor internal energy, represent
Current point and the relation of a marginal point of the averaging model;α is weight, and for balancing inside and outside energy, different chamber makes
With different weights;C represents each chamber.When current point not only close to target point but also close to a marginal point of the averaging model
When, then energy function is minimum, that is, tries to achieve optimum coordinates point.Gross energy E is smaller, as a result more accurate.
External energy function can be expressed as:
Wherein, i is each;wiFor the weight (i.e. the reliability of the point) shared by each point;Current point coordinates is vi, classify through PBT
The point that device detects isFor the gradient (vector) of point,For Grad size.Internal energy function can be with
It is expressed as:
Eint=∑i∑j∑kwI, k((vi-vj)-TAffine, k(mi-mj))2 (5)
Wherein, i is each point, and j is point i neighborhood (then vi-vjCorresponding to the side of each triangle in current point position);wI, kTo close
Join the factor (factors of each chamber k to current point i);mi, mjFor the point (being tried to achieve by PDM/ASM) on averaging model;mi-mjIt is corresponding
In the side of each triangle of mesh averaging models), TAffine, kThe transformation relation tried to achieve by each chamber k affine transformations PAT.Wherein,
Point coordinates viAll it is space three-dimensional.
Generalised Hough transform, model adjustment and model conversion are weighted, the model and image accurately matched can be obtained.
As shown in figure 25, each chamber of phantom heart after accurate matching clearly, is clearly split.
Basic conception is described above, it is clear that to those skilled in the art, foregoing invention discloses only
As an example, and the restriction to the application is not formed.Although do not clearly state herein, those skilled in the art may
Various modifications are carried out to the application, improves and corrects.Such modification, improvement and amendment are proposed in this application, so such
Modification, improve, correct the spirit and scope for still falling within the application example embodiment.
Meanwhile the application has used particular words to describe embodiments herein.Such as " one embodiment ", " one implements
Example ", and/or " some embodiments " mean a certain feature, structure or the feature related at least one embodiment of the application.Cause
This, it should be highlighted that and it is noted that " embodiment " or " implementation that are referred to twice or repeatedly in diverse location in this specification
Example " or " alternate embodiment " are not necessarily meant to refer to the same embodiment.In addition, in one or more embodiments of the application
Some features, structure or feature can carry out appropriate combination.
In addition, it will be understood by those skilled in the art that each side of the application can be by some with patentability
Species or situation are illustrated and described, including any new and useful process, the combination of machine, product or material, or right
Their any new and useful improvement.Correspondingly, the various aspects of the application can be performed completely by hardware, can be complete
Performed, can also be performed by combination of hardware by software (including firmware, resident software, microcode etc.).Hardware above is soft
Part is referred to alternatively as " data block ", " module ", " engine ", " unit ", " component " or " system ".In addition, each side of the application
The computer product being located in one or more computer-readable mediums may be shown as, the product includes computer-readable program
Coding.
Computer-readable signal media may include the propagation data signal containing computer program code in one, such as
A part in base band or as carrier wave.The transmitting signal may have many forms, including electromagnetic form, light form etc.
Deng or suitable combining form.Computer-readable signal media can be any meter in addition to computer-readable recording medium
Calculation machine computer-readable recording medium, the medium can by be connected to an instruction execution system, device or equipment with realize communication, propagate or
Transmit the program for using.Program coding in computer-readable signal media can be carried out by any suitable medium
Propagate, include the combination of radio, cable, fiber optic cables, RF or similar mediums or any of above medium.
In addition, except clearly stating in non-claimed, the order of herein described processing element and sequence, digital alphabet
Using or other titles use, be not intended to limit the order of the application flow and method.Although by each in above-mentioned disclosure
Kind of example discusses some it is now recognized that useful inventive embodiments, but it is to be understood that, such details only plays explanation
Purpose, appended claims are not limited in the embodiment disclosed, on the contrary, claim is intended to cover and all meets the application
The amendment of embodiment spirit and scope and equivalent combinations.For example, although system component described above can be set by hardware
It is standby to realize, but only can also be achieved by the solution of software, such as pacify on existing server or mobile device
The described system of dress.
Although the present invention is disclosed as above with preferred embodiment, it is not for limiting the present invention, any this area
Technical staff without departing from the spirit and scope of the present invention, may be by the methods and technical content of the disclosure above to this hair
Bright technical scheme makes possible variation and modification, therefore, every content without departing from technical solution of the present invention, according to the present invention
Any simple modifications, equivalents, and modifications made to above example of technical spirit, belong to technical solution of the present invention
Protection domain.
Claims (10)
1. a kind of image partition method, including:
Obtain view data;
Based on described image data, reconstruction image, wherein, described image includes one or more first edges;
A model is obtained, wherein, the model includes the one or more corresponding with one or more of first edges
Second edge;
Match the model and the image after the reconstruction;And
According to one or more of first edges, one or more second edges of the model are adjusted.
2. according to the method for claim 1, it is characterised in that one or more second edges of the adjustment model
Including:Determine a reference point in the second edge;It is determined that the target point corresponding with the reference point;And according to institute
Target point is stated, adjusts the second edge of the model.
3. according to the method for claim 2, it is characterised in that described to determine the target point bag corresponding with the reference point
Include:
Determine the normal of the reference point;
Obtain step-length and hunting zone;
According to the step-length and hunting zone, one or more candidate points are determined along normal;
Obtain first grader;
According to first grader, determine that one or more of candidate points correspond to the probability of the first edge;And
Correspond to the probability of the first edge based on one or more of candidate points, determine the target point.
4. according to the method for claim 3, it is characterised in that the normal for determining the reference point includes:It is determined that with
The adjacent one or more polygonal mesh of the reference point;Determine corresponding to one or more of polygonal mesh one or
Multiple normals;And according to one or more of normals, determine the normal of the reference point.
5. according to the method for claim 2, it is characterised in that the second edge of the adjustment model includes:
Similarity transformation is carried out to the second edge;
The control point of the model is determined, one or more second sides according to the control point of the model and the model
The relation of edge, the association factor at the control point of the model is generated, according to association factor, affine change is carried out to the second edge
Change;Or
Based on an energy function, the second edge is finely adjusted.
6. according to the method for claim 1, it is characterised in that described image data include brain image, skull image, chest
Portion's image, cardiac image, galactophore image, abdomen images, renal image, liver image, pelvis image, perineum image, limbs
Image, vertebra image or osteogram picture.
7. an image segmentation system, including:
One memory, is configured as data storage and instruction;
One is established the processor to communicate with memory, wherein, when performing the instruction in memory, the processor is configured
For:
Obtain view data;
Based on described image data, reconstruction image, wherein, described image includes one or more first edges;
A model is obtained, wherein, the model includes the one or more corresponding with one or more of first edges
Second edge;
Match the model and the image after the reconstruction;And
According to one or more of first edges, one or more second edges of the model are adjusted.
8. system according to claim 7, it is characterised in that the processor is further configured to:
Determine a reference point in the second edge;
It is determined that the target point corresponding with the reference point;And
According to the target point, the second edge of the model is adjusted.
9. system according to claim 8, it is characterised in that the processor is further configured to:
Determine the normal of the reference point;
Obtain step-length and hunting zone;
According to the step-length and hunting zone, one or more candidate points are determined along normal;
Obtain first grader;
According to first grader, determine that one or more of candidate points correspond to the probability of the first edge;And
Correspond to the probability of the first edge based on one or more of candidate points, determine the target point.
10. system according to claim 7, it is characterised in that the second edge of the adjustment model includes:To institute
State second edge and carry out similarity transformation;The control point of the model is determined, according to the control point of the model and the model
Described in one or more second edges relation, generate the association factor at the control point of the model, it is right according to association factor
The second edge carries out affine transformation;Or based on an energy function, the second edge is finely adjusted.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710311908.8A CN107424162B (en) | 2017-05-05 | 2017-05-05 | Image segmentation method and system |
US15/710,815 US10482604B2 (en) | 2017-05-05 | 2017-09-20 | Systems and methods for image processing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710311908.8A CN107424162B (en) | 2017-05-05 | 2017-05-05 | Image segmentation method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107424162A true CN107424162A (en) | 2017-12-01 |
CN107424162B CN107424162B (en) | 2019-12-20 |
Family
ID=60425385
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710311908.8A Active CN107424162B (en) | 2017-05-05 | 2017-05-05 | Image segmentation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107424162B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685816A (en) * | 2018-12-27 | 2019-04-26 | 上海联影医疗科技有限公司 | Image segmentation method, device, equipment and storage medium |
CN110675486A (en) * | 2019-08-28 | 2020-01-10 | 电子科技大学 | Frequency domain reconstruction method for non-rigid human body movement |
CN111242877A (en) * | 2019-12-31 | 2020-06-05 | 北京深睿博联科技有限责任公司 | Mammary X-ray image registration method and device |
WO2024002110A1 (en) * | 2022-06-27 | 2024-01-04 | Shanghai United Imaging Healthcare Co., Ltd. | Methods and systems for determining image control point |
CN117635942A (en) * | 2023-12-05 | 2024-03-01 | 齐鲁工业大学(山东省科学院) | Cardiac MRI image segmentation method based on edge feature enhancement |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
US20160063726A1 (en) * | 2014-08-28 | 2016-03-03 | Koninklijke Philips N.V. | Model-based segmentation of an anatomical structure |
CN105719278A (en) * | 2016-01-13 | 2016-06-29 | 西北大学 | Organ auxiliary positioning segmentation method based on statistical deformation model |
CN105976384A (en) * | 2016-05-16 | 2016-09-28 | 天津工业大学 | Human body thoracic and abdominal cavity CT image aorta segmentation method based on GVF Snake model |
-
2017
- 2017-05-05 CN CN201710311908.8A patent/CN107424162B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
US20160063726A1 (en) * | 2014-08-28 | 2016-03-03 | Koninklijke Philips N.V. | Model-based segmentation of an anatomical structure |
CN105719278A (en) * | 2016-01-13 | 2016-06-29 | 西北大学 | Organ auxiliary positioning segmentation method based on statistical deformation model |
CN105976384A (en) * | 2016-05-16 | 2016-09-28 | 天津工业大学 | Human body thoracic and abdominal cavity CT image aorta segmentation method based on GVF Snake model |
Non-Patent Citations (1)
Title |
---|
凌华强 等: "主动形体模型法在肝脏CT图像分割中的应用", 《浙江工业大学学报》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685816A (en) * | 2018-12-27 | 2019-04-26 | 上海联影医疗科技有限公司 | Image segmentation method, device, equipment and storage medium |
CN109685816B (en) * | 2018-12-27 | 2022-05-13 | 上海联影医疗科技股份有限公司 | Image segmentation method, device, equipment and storage medium |
CN110675486A (en) * | 2019-08-28 | 2020-01-10 | 电子科技大学 | Frequency domain reconstruction method for non-rigid human body movement |
CN110675486B (en) * | 2019-08-28 | 2023-03-07 | 电子科技大学 | Frequency domain reconstruction method for non-rigid human body movement |
CN111242877A (en) * | 2019-12-31 | 2020-06-05 | 北京深睿博联科技有限责任公司 | Mammary X-ray image registration method and device |
WO2024002110A1 (en) * | 2022-06-27 | 2024-01-04 | Shanghai United Imaging Healthcare Co., Ltd. | Methods and systems for determining image control point |
CN117635942A (en) * | 2023-12-05 | 2024-03-01 | 齐鲁工业大学(山东省科学院) | Cardiac MRI image segmentation method based on edge feature enhancement |
CN117635942B (en) * | 2023-12-05 | 2024-05-07 | 齐鲁工业大学(山东省科学院) | Cardiac MRI image segmentation method based on edge feature enhancement |
Also Published As
Publication number | Publication date |
---|---|
CN107424162B (en) | 2019-12-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107220965A (en) | A kind of image partition method and system | |
US10482604B2 (en) | Systems and methods for image processing | |
US10546014B2 (en) | Systems and methods for segmenting medical images based on anatomical landmark-based features | |
US20190355117A1 (en) | Techniques for Segmentation of Lymph Nodes, Lung Lesions and Other Solid or Part-Solid Objects | |
CN107424162A (en) | A kind of image partition method and system | |
CN109215032A (en) | The method and system of image segmentation | |
US8577130B2 (en) | Hierarchical deformable model for image segmentation | |
US9218542B2 (en) | Localization of anatomical structures using learning-based regression and efficient searching or deformation strategy | |
Xing et al. | Lesion segmentation in ultrasound using semi-pixel-wise cycle generative adversarial nets | |
Li et al. | Learning image context for segmentation of the prostate in CT-guided radiotherapy | |
US11935246B2 (en) | Systems and methods for image segmentation | |
Kieselmann et al. | Cross‐modality deep learning: contouring of MRI data from annotated CT data only | |
CN111709485B (en) | Medical image processing method, device and computer equipment | |
EP3493154A1 (en) | Segmentation system for segmenting an object in an image | |
US20220301224A1 (en) | Systems and methods for image segmentation | |
CN107220984A (en) | A kind of image partition method, system and grid model | |
US20220222873A1 (en) | Devices and process for synthesizing images from a source nature to a target nature | |
Wu et al. | Prostate segmentation based on variant scale patch and local independent projection | |
Feng et al. | Supervoxel based weakly-supervised multi-level 3D CNNs for lung nodule detection and segmentation | |
CN113570627A (en) | Training method of deep learning segmentation network and medical image segmentation method | |
CN111798424A (en) | Medical image-based nodule detection method and device and electronic equipment | |
CN107230211A (en) | A kind of image partition method and system | |
CN111724395A (en) | Heart image four-dimensional context segmentation method, device, storage medium and device | |
Gao et al. | Hybrid decision forests for prostate segmentation in multi-channel MR images | |
Yang et al. | Automatic segmentation of the clinical target volume and organs at risk for rectal cancer radiotherapy using structure-contextual representations based on 3D high-resolution network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CP01 | Change in the name or title of a patent holder |
Address after: 201807 Shanghai city Jiading District Industrial Zone Jiading Road No. 2258 Patentee after: Shanghai Lianying Medical Technology Co., Ltd Address before: 201807 Shanghai city Jiading District Industrial Zone Jiading Road No. 2258 Patentee before: SHANGHAI UNITED IMAGING HEALTHCARE Co.,Ltd. |
|
CP01 | Change in the name or title of a patent holder |