CN103218800B - Method and apparatus for automatic rib central line pick-up - Google Patents
Method and apparatus for automatic rib central line pick-up Download PDFInfo
- Publication number
- CN103218800B CN103218800B CN201210368926.7A CN201210368926A CN103218800B CN 103218800 B CN103218800 B CN 103218800B CN 201210368926 A CN201210368926 A CN 201210368926A CN 103218800 B CN103218800 B CN 103218800B
- Authority
- CN
- China
- Prior art keywords
- rib
- centrage
- volume
- voxel
- template
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 210000000038 chest Anatomy 0.000 claims abstract description 20
- 210000000988 Bone and Bones Anatomy 0.000 claims description 15
- 238000006243 chemical reaction Methods 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 15
- 238000000034 method Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 6
- 230000004044 response Effects 0.000 claims description 6
- 238000003325 tomography Methods 0.000 abstract description 3
- 238000001514 detection method Methods 0.000 description 10
- 230000001131 transforming Effects 0.000 description 10
- 238000002372 labelling Methods 0.000 description 8
- 238000004590 computer program Methods 0.000 description 5
- 229920001707 polybutylene terephthalate Polymers 0.000 description 5
- 210000002356 Skeleton Anatomy 0.000 description 4
- 238000002591 computed tomography Methods 0.000 description 4
- 230000001808 coupling Effects 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 4
- 238000005859 coupling reaction Methods 0.000 description 4
- 230000003647 oxidation Effects 0.000 description 4
- 238000007254 oxidation reaction Methods 0.000 description 4
- 231100000915 pathological change Toxicity 0.000 description 4
- 230000036285 pathological change Effects 0.000 description 4
- 210000001185 Bone Marrow Anatomy 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 210000003109 Clavicle Anatomy 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000000712 assembly Effects 0.000 description 2
- 238000005295 random walk Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 208000010392 Bone Fractures Diseases 0.000 description 1
- 206010005949 Bone cancer Diseases 0.000 description 1
- 241001272996 Polyphylla fullo Species 0.000 description 1
- 210000001991 Scapula Anatomy 0.000 description 1
- 241000270295 Serpentes Species 0.000 description 1
- 102000004357 Transferases Human genes 0.000 description 1
- 108090000992 Transferases Proteins 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000000875 corresponding Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing Effects 0.000 description 1
- 210000000056 organs Anatomy 0.000 description 1
- 239000002965 rope Substances 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Abstract
The present invention relates to the method and system for automatic rib central line pick-up.Disclose a kind of method and system for extracting rib centrage in 3D volume (such as 3D computer tomography (CT) volume).Detector based on study is utilized to detect rib centrage voxel in described 3D volume.The rib centrage for whole Rib cage is extracted subsequently by the template of the rib centrage for whole Rib cage and described 3D volume being matched based on detected rib centrage voxel.Each the rib centrage extracted individually is refined followed by active contour model.
Description
This application claims the rights and interests of the U.S. Provisional Application No. 61/539,561 of JIUYUE in 2011 submission on the 27th, in it is open
Hold and be incorporated in this by reference.
Technical field
The present invention relates to the rib centrage in 3D medical image data (such as 3D computer tomography (CT) scanning)
Lines extract, and more particularly to utilize deformable template based on study coupling to carry out automatic rib centrage to carry
Take.
Background technology
In Thoracic CT scan, positioning rib Bone tumour and fracture are usually directed to read hundreds of axial CT section, with
The most visually follow the tracks of the change that frame section surface is long-pending.Artificial reading accordingly for CT scan is relatively time consuming, and in practice
In usually miss rib due to human error abnormal.Rib anatomical center line is automatically extracted and can be used to enhancing exhibition
The visualization of the Rib cage opened, this is so that routine skeleton reading task is more efficient and effective for radiologist.
That extracted and tagged rib centrage is also used as positioning organ, registration pathological changes and guide interval and changes
The reference of the correspondence between the serial thorax CT scan analyzed.Additionally, the information derived of rib geometry can be helped
Help Rib cage fracture fixing operation.
Although having the importance of previously described clinicing aspect, but detection automatically and the mark to rib in CT scan
Note remains a challenging task.Rib is modeled as elongated tubular construction by most of traditional methods, and uses
Hai Sai (Hessian) or structure tensor proper system analysis are to carry out back voxel detection.But, these algorithms are generally calculating
Aspect spends the highest, and possibly cannot all obtain consistent result for all patients.Fig. 1 shows the difference of different patient
The example of frame section surface.As shown in fig. 1, the frame section surface 104 shown in image (b) is than the frame section surface shown in image (a)
102 have the dark bone marrow become apparent from.In many cases, rib bone marrow may be more darker than rib border；Consequently, it is possible to cannot
Rib central point is as one man detected as back voxel.In order to construct rib centrage, generally use based on the method (ratio followed the tracks of
Such as Kalman filter) detected rib central point is tracked to next one section from a section.But, some pass
The method based on tracking of system needs artificial initial seed point, and such point-to-point tracking is for such as fracture etc
The On Local Fuzzy or the most extremely sensitive that caused of rib pathological changes (this is probably what radiologist was most interested in).Fig. 2
Show the example that may cause different rib pathological changes based on the inaccurate rib centrage in the method followed the trail of.Such as Fig. 2
Shown in, image (a) shows the rib of disappearance rib sections 202, and image (b) shows have rib transferase 12 04
Rib.Additionally, in traditional rib tracing algorithm, each root bone is detected separately and is followed the trail of；Therefore rib labelling needs
Want single heuristic method.
Summary of the invention
The invention provides a kind of method of automatic rib central line pick-up in 3D CT volume and labelling and set
Standby.The embodiment of the embodiment of the present invention utilizes rib center line detecting based on study to detect seed points, and subsequently by whole
The deformable template of individual Rib cage and seed points match to extract rib centrage.By simultaneously rather than individually
Extract the rib centrage of whole Rib cage, embodiments of the invention can apply to retrain in advance between adjacent rib with
Just improve the robustness during rib is followed the tracks of, and rib labelling is provided while rib is followed the tracks of.
In one embodiment of the invention, 3D volume detects rib centrage voxel.By based on detected
Rib centrage voxel the template of the rib centrage of many roots bone and described 3D volume are matched and extract in rib
Heart line.The template of rib centrage can be the template of the rib centrage for whole Rib cage.Active shape can be utilized
Model individually refines extracted rib centrage.
By referring to following detailed description and drawings, it will be appreciated by those of ordinary skill in the art that the present invention these and
Other advantages.
Accompanying drawing explanation
Fig. 1 shows the example of the different frame section surfaces in different patient；
Fig. 2 shows the example of different rib pathological changes；
Fig. 3 shows the method for extracting rib centrage from 3D volume according to an embodiment of the invention；
Fig. 4 shows exemplary rib centrage voxel testing result；
Fig. 5 shows the example of the rib centrage for different Rib cage shapes；
Fig. 6 shows the example results of articulated type rib segment matches method；
Fig. 7 shows the exemplary rib centrage refinement utilizing actively profile；
Fig. 8 shows exemplary rib central line pick-up result；
Fig. 9 shows the exemplary rib central line pick-up result of the challenging situation of the method for utilizing Fig. 3；
And
Figure 10 is the high-level block diagram of the computer that can implement the present invention.
Detailed description of the invention
The present invention is directed to for extracting rib in 3D medical image volume (such as 3D computer tomography (CT) volume)
The method and apparatus of bone centrage.It is described herein as embodiments of the invention to provide for described rib central line pick-up side
The visual analysis of method.Digital picture is usually made up of the numeral expression of one or more objects (or shape).Here may often be such that
The numeral expression of description object is carried out in terms of mark and manipulating objects.Such manipulation be computer system memorizer or
The virtual manipulation realized in other circuit/hardware.Accordingly, it can be appreciated that be, it is possible to use be stored in computer system
Data in described computer system, implement embodiments of the invention.
Embodiments of the invention utilize deformable template matching process to extract rib center in 3D medical image data
Line.The rib central point detection method based on study utilizing the efficient Like-Fenton Oxidation of calculating aspect can be used for 3D volume
In rib central point detection.In order to further speed up operation time detecting, it is possible to use pyramid study from coarse to fine
Structure.The probability respondence figure obtained from rib central point detection based on study can be used to mate whole Rib cage template,
To extract rib centrage.By extracting all rib centrages rather than individually following the trail of each rib center simultaneously
Line, embodiments of the invention are possible not only to apply to retrain in advance and thus significantly improve rib centrage between adjacent rib
The robustness followed the tracks of, but also follow the tracks of with rib and rib labelling is provided simultaneously.In order to apply the deformation of Rib cage and keep high
Computational efficiency, long rib is resolved into shorter rib sections by embodiments of the invention, and by according to being similar to limit
The mode of space learning (MSL) searches for optimum similarity transformation parameter to implement articulated type rib segment matches.With traditional point
Comparing to some rib tracing algorithm, embodiments of the invention are for On Local Fuzzy and discontinuous more robust.Once utilize described
After template matching extracts centrage for all rib sections, embodiments of the invention just can utilize active contour model
Refinement and the rib centrage of smooth rigidity coupling.
Fig. 3 shows the method for extracting rib centrage from 3D volume according to an embodiment of the invention.Fig. 3
Method 3D medical image volume (such as CT volume) is transformed into the rib centrage extracted of the rib representing patient
Set.As shown in Figure 3,3D volume is received in step 302 place.According to an advantageous embodiment, described 3D volume can be CT
Volume, but the invention is not restricted to this.Other medical imaging modalities can also be utilized to gather 3D volume, such as magnetic resonance (MR), x
Ray, ultrasonic etc..3D volume can be directly received from image acquisition device (such as CT scanner), or (example can be passed through
From the storage device of computer system or memorizer) load previously stored 3D volume to 3D volume as described in receiving.
In step 304 place, trained rib center line detecting device is utilized to detect rib centerline hull in 3D volume
Element.Dark bone marrow within rib, therefore can not be modeled as rib solid bright tubular structures simply.Correspondingly, logical
The peak response crossing wave filter based on sea match proper system the most reliably identifies rib centrage.Additionally, such as Fig. 1
Shown in, the rib crossing over different volumes shows various size, shape and contrast on border.According to the present invention one
Individual advantageous embodiment, uses the center line detecting method specific to object based on study of robust to detect rib centerline hull
Element, and use the probability graph obtained from detection based on study to follow the tracks of and labelling rib centrage.
Rib center line detecting device is trained by the training data utilizing band to annotate.In a kind of Advantageous implementations,
3D Like-Fenton Oxidation is used to train rib center line detecting device.3D Like-Fenton Oxidation considers the specific location in detection window
Neighbouring blocked areas, and calculate the difference between the image pixel intensities summation in each region.Summation area table can be utilized quick
Ground calculates described feature.In order to train rib center line detecting device, extract at the positive and negative training sample in training data
3D Like-Fenton Oxidation, and based on the features training rib center line detecting device extracted.In a possible embodiment,
Rib center line detecting device can be trained to the probability boosted tree grader with multiple AdaBoost grader node
(PBT).But, the invention is not restricted to PBT grader, and also other machines learning algorithm can be used to train rib center
Thread detector.
In an advantageous embodiment, replace simply use single PBT(or other kinds of grader), can train by
Thick to thin pyramid PBT grader and use it for the detection of rib centrage voxel, the most not only can pass through under image
Sample thus reduce the number of samples in classification commitment and substantially speed up detection, but also under make use of low resolution
The Space Elements of longer scope.In the case of given received 3D volume, trained to described 3D volume applications
Grader, and each voxel for 3D volume determines that this voxel is the probability of rib centrage voxel.Correspondingly, pass through
The rib center line detecting device generating probability response diagram of training, it shows that each voxel is possible of rib centrage
Property.Fig. 4 shows exemplary rib centrage voxel testing result.As shown in Figure 4, image 400 shows received
CT volume, and image 402 shows the probability respondence figure utilizing trained rib center line detecting device to generate.
Return to Fig. 3, in step 306 place, based on detected rib centrage voxel by rib centrage template and 3D
Volume matches.The probability graph generated by rib center line detecting device may be not always reliable, and this is due to from adjacent
The distraction of class skeleton (such as clavicle and scapula) or by the rib On Local Fuzzy that causes of damage or do not connect
Continue and cause.Correspondingly, the rib using a kind of robust follows the tracks of and labeling method extracts rib based on described probability graph
Centrage.Traditional point-to-point method for tracing (such as Kalman filter or region growing process) height abnormal for local rib
Sensitive and be therefore susceptible to error propagation.
Embodiments of the invention utilize matching process based on template to carry out rib central line pick-up.Rib centrage mould
Plate is by 12 pairs of rib centrages from normal Rib cage carrying out manual annotation and labelling and off-line structure, and rib
Each root bone of bone centrage template By the centrage set of voxels of uniform sampling
Represent, whereinIt it is ribLength.Described template is such as stored in memorizer or the storage device of computer system
In, and it is subsequently based on the probability graph and received 3D volume phase utilizing trained rib center line detecting device to generate
Coupling.In order to template is matched with 3D volume so that the response summation of the template through conversion on the probability graph generated
Maximized optimal mappingTIt is:
。
The template using whole Rib cage makes this method be with the difference of tradition rib detection method, all rib quilts
Follow the tracks of simultaneously or mate.Whole Rib cage template allows following the tracks of or during coupling, adjacent rib is being applied shape constraining, thus
Overcome the distraction from adjacent bones structure (such as clavicle or neighbouring rib).Additionally, utilize the mould of whole Rib cage
The template matching of plate is automatic labelling rib in 3D volume, this is because the rib of the rib in the rib centrage template stored
Bone label is known.
Fig. 5 shows the example of the rib centrage for different Rib cage shapes.In the image (a) of Fig. 5 and (b)
Shown in, between different patients, Rib cage there may be significantly deformation, and the spacing between neighbouring rib may
Change significantly.Due to the spacing change between substantially deformation and the neighbouring rib of the Rib cage of different patients, the most just
Property registration possibly cannot produce result accurately.Traditional non-rigid transformation or method for registering spend in terms of calculating the highest and
The most fragile for local minimum, thus cause inaccurate result.In an advantageous embodiment of the present invention, use and close
Joint formula matching process matches rib centrage template with 3D volume.Specifically, each root boneIt is broken down into several
Individual relatively short segment .Therefore, the degree of crook of each rib sections centrage is relatively low, and such that it is able to warp
By rigid transformation by the optimum similarity transformation parameter of searchMatch approximately with, whereinRepresent translation, orientation and zooming parameter respectively, as expressed by equation (2):
(2)
Wherein,It it is similarity conversionTSet.In a kind of Advantageous implementations, replace exhaustive searchNine dimension parameter spaces, only in accordance be similar to limit space learning (MSL) strategy search
Space, rope low-dimensional limit.Specifically, estimate to be divided into three steps by the conversion for each rib sections: location estimation, position-
Orientation is estimated, and position-orientation-scaling is estimated.First space, limit, exhaustive search position, and retain many optimum positions
Candidate.It follows that for space, each position candidate's exhaustive search orientation limit (i.e. by raw for each position candidate
Become many position-directional hypothesis), and retain multiple optimum position-orientation candidate.It follows that for each position-orientation
Space, candidate search scaling limit is (i.e. false by generating many position-orientation-scalings for each position-orientation candidate
If), thus obtain one or more optimal candidate of the conversion completely (position-orientation-scaling) for described rib sections.Right
Training data that band annotates can be utilized to train list in each stage (position, position-orientation and position-orientation-scaling)
Only trained grader (such as PBT grader).
In equation (2), each rib sections is by roving commission, and consequently, it is possible to due to the phase between neighbouring rib
The rib of mistake it is matched like property.In order to avoid this problem, embodiments of the invention utilize Markov random field (MRF)
Model is applied with paired smoothness constraint to the transformation parameter of adjacent rib sections:
Wherein,It is regularization parameter,Represent adjacent rib pairSet, and L is two be defined below
The similarity function of individual transformation parameter:
(4)
Wherein, L measures and passes through different parameters respectivelyWithTwo adjacent rib sections of conversionWithFlat
All Euclidean distances.
So that equation (3) maximizes and finds the optimal mapping set for all rib sections, for each
Individual rib sectionsRoving commission transformation parameterAnd keep transformation parameterMultiple top candidate.Specifically, right
In each rib sections can by utilize this sections is trained grader set (position, position-orientation and position-
Orientation-scaling) searching probability response diagram detects transformation parameter.Dynamic programming can be utilized subsequently to determine efficiently in order to maximum
Change all rib sections conversion of equation (3) Optimum combination.Fig. 6 shows that described articulated type rib saves
The example results of section matching process.As shown in Figure 6, image 600,602,604 and 606 shows the template of rib centrage
Articulated type rib segment matches, each of which root bone is broken down into four sections。
Return to Fig. 3, in step 308 place, utilize active contour model to refine single rib centrage.Due to each rib
Bone segment is to transform rigidly, so articulated type rib centrage smooths piecemeal, and due to each rib sections
Micro-strain and the finite resolving power of discrete transformation parameter search volume and stand less deviation.Correspondingly, the present invention
Embodiment uses active contour model or contours model (snakes) to refine template matching results further.Described active
Skeleton pattern is alone applied in each the rib centrage extracted, in order to thin by minimizing based on probability respondence figure
Change each rib centrage and energy function be:
(5)
Wherein,αWithβControl elastic force and rigidity power respectively, andRepresent rib centerline points.Applying such master
Possible problem during dynamic skeleton pattern is associated with initialization.In general, initial profile must be close to real border,
Or described profile will be likely to be converging on the result of mistake.But this is not a problem in an embodiment of the present invention, because articulated type
Template matching provides the good initialization for rib centrage.Minimizing of the energy function of equation (5) causes solving phase
The Eulerian equation answered, and well-known iterative numerical methods can be used to solve Eulerian equation.Fig. 7 shows that utilization is main
The exemplary rib centrage refinement that driving wheel is wide.As shown in Figure 7, image 700 shows and utilizes articulated type template matching to extract
The rib centrage gone out, and image 710 show by application active contour model obtain through refinement rib centrage.
It can be seen that the rib centrage 702 disconnected in image 700, and show corresponding through refinement in image 710
Rib centrage 712.
Return to Fig. 3, export rib central line pick-up result.Can be by such as showing on the display of computer system
The rib centrage extracted exports extracted rib centrage.Can also be by such as the rib extracted
Centrage be stored in computer system memorizer or storage device in and export extracted rib centrage.
Fig. 8 shows exemplary rib central line pick-up result.As shown in Figure 8, image 800 shows utilization tradition
The rib centrage that extracts in a CT volume of method for tracing based on random walk (walker) segmentation, and image
802 show the rib centrage that the method utilizing Fig. 3 extracts in a CT volume.Image 804 shows utilization tradition
The rib centrage that extracts in the 2nd CT volume of method for tracing based on random walk segmentation, and image 806 illustrates
Utilize the rib centrage that the method for Fig. 3 extracts in the 2nd CT volume.It can be seen in fig. 8 that according to the present invention's
The method of Fig. 3 of one embodiment is substantially than traditional method more robust based on tracking.
Fig. 9 shows the exemplary rib central line pick-up result of the challenging situation of the method for utilizing Fig. 3.
As shown in Figure 9, image (a) shows the rib center that the rib set of the rib sections 904 for having disappearance extracts
Line 902.Image (b) shows the rib centrage 912 extracted for the rib with rib transfer 914.Image (c) illustrates
For having the rib centrage 922 that the rib of the large pitch 924 of abnormality extracts between adjacent rib.Image (d)
Show for including the rib centrage 932 that the rib set of the rib 934 being connected extracts.
The previously described method for rib central line pick-up can utilize well-known computer processor, storage
Device unit, memory device, computer software and other assemblies are implemented on computers.Figure 10 illustrates such computer
High-level block diagram.Computer 1002 comprises processor 1004, and it is by performing the overall operation of definition computer 1002
Computer program instructions controls such operation.Computer program instructions can be stored in memory device 1012 or other meters
In calculation machine computer-readable recording medium (such as disk, CD ROM etc.), and it is loaded when hope performs described computer program instructions
In memorizer 1010.Therefore, the step of the method for Fig. 3 can be by being stored in memorizer 1010 and/or storage device 1012
Computer program instructions definition, and by perform computer program instructions processor 1004 control.Image acquisition device
1020(such as CT scanner) may be coupled to computer 1002 to computer 1002 input picture.Likely image is adopted
Collection device 1020 and computer 1002 are embodied as a device.It is also possible to make image acquisition device 1020 lead to computer 1002
Cross network and carry out radio communication.Computer 1002 also includes that one or more network interface 1006 is will pass through network and other devices
Part communicates.Computer 1002 also includes other input/output devices 1008(such as display, keyboard, mouse, speaker, button
Etc.), thus allow and computer 1002 to carry out user mutual.It would be recognized by those skilled in the art that the computer of reality
Implementation can also comprise other assemblies, and Figure 10 is the some of them assembly of this type of computer for explanatory purposes
High-level represent.
Detailed description above is appreciated that it is the most all illustrative and exemplary and nonrestrictive, and
And the scope of the present invention disclosed herein is not determined by described detailed description in detail, but the widest by allowed according to Patent Law
Claims that degree is explained determine.It should be appreciated that embodiment shown and described herein is merely illustrative the present invention
Principle, and in the case of without departing substantially from scope and spirit of the present invention, those skilled in the art can implement various amendment.
In the case of without departing substantially from scope and spirit of the present invention, those skilled in the art can implement other features various combination.
Claims (16)
1. for the method extracting rib centrage in 3D volume, including:
Rib centrage voxel is detected in described 3D volume；And
By based on detected rib centrage voxel by the template of the rib centrage of many roots bone and described 3D volume
Match and extract rib centrage.
2. the process of claim 1 wherein, the template of the rib centrage of many roots bone includes the rib for whole Rib cage
The template of centrage.
3. the method for claim 1, also includes:
Active contour model is utilized individually to refine each the rib centrage extracted.
4. the process of claim 1 wherein, the step detecting rib centrage voxel in described 3D volume includes:
Trained rib centrage voxel detector is utilized to detect rib centrage voxel in described 3D volume.
5. the method for claim 4, wherein, described trained rib center line detecting device is training number based on band annotation
Probability boosted tree grader according to training.
6. the method for claim 4, wherein, described trained rib center line detecting device includes pyramid from coarse to fine
The trained grader of shape.
7. the process of claim 1 wherein, in described 3D volume detect rib centrage voxel step include by for
Each in the middle of multiple voxels in described 3D volume determines and is the probability of rib centrage voxel about described voxel and gives birth to
Become probability respondence figure, and pass through the template of the rib centrage of many roots bone based on detected rib centrage voxel
Match with described 3D volume and extract the step of rib centrage and include:
The template of rib centrage is matched with described probability respondence figure.
8. the method for claim 7, wherein, it is multiple that the template of described rib centrage includes for each rib centrage
Rib sections, and the step that the template of rib centrage and probability respondence figure match is included:
Each in the middle of multiple rib sections of each rib centrage is determined that multiple conversion candidate is so that this rib
Bone segment matches with described probability respondence figure；And
Multiple rib sections for each rib centrage determine the combination of described conversion candidate, in order to based on described probability
Response diagram maximizes the rib centrage total probability by the paired smoothness constraint of the conversion for adjacent rib sections.
9. the method for claim 8, wherein, described for each in the middle of multiple rib sections of each rib centrage
Determine that multiple conversion candidate is so that the step that this rib sections and described probability respondence figure are matched includes for each rib
Bone segment enforcement herein below:
The first trained grader is utilized to detect position candidate in described probability respondence figure；
Utilize the second trained grader based on detected position candidate detect in described probability respondence figure position-
Orientation candidate；And
The 3rd trained grader is utilized to detect in described probability respondence figure based on detected position-orientation candidate
Conversion candidate.
10. the method for claim 8, wherein, described multiple rib sections for each rib centrage determine described change
Change the combination of candidate to maximize rib centrage based on described probability respondence figure and being converted by for adjacent rib sections
The step of total probability of paired smoothness constraint include:
Dynamic programming is utilized to determine the described combination converting candidate for multiple rib sections of each rib centrage, in order to
The rib centrage paired smoothness constraint by the conversion for adjacent rib sections is maximized based on described probability respondence figure
Total probability.
11. 1 kinds of equipment being used for extracting rib centrage in 3D volume, including:
For detecting the device of rib centrage voxel in described 3D volume；And
For passing through based on detected rib centrage voxel the template of the rib centrage of many roots bone and described 3D
Volume matches to extract the device of rib centrage.
The equipment of 12. claim 11, wherein, the template of the rib centrage of described many roots bone includes for whole Rib cage
The template of rib centrage.
The equipment of 13. claim 11, also includes:
For utilizing active contour model individually to refine the device of each the rib centrage extracted.
The equipment of 14. claim 11, wherein, the described device for detecting rib centrage voxel in 3D volume includes:
For utilizing trained rib centrage voxel detector to detect rib centrage voxel in described 3D volume
Device.
The equipment of 15. claim 11, wherein, the described device for detecting rib centrage voxel in 3D volume includes using
It is the general of rib centrage voxel in generating about described voxel for each in the middle of the multiple voxels in described 3D volume
The device of the probability respondence of rate, and described for by based on detected rib centrage voxel the rib of many roots bone
The template of bone centrage matches with described 3D volume so that the device extracting rib centrage includes:
For the device that the template of described rib centrage is matched with described probability respondence figure.
The equipment of 16. claim 15, wherein, the template of described rib centrage includes for each rib centrage many
Individual rib sections, and the described device for the template of rib centrage and described probability respondence figure are matched includes:
For determining that multiple conversion candidate is so that handle for each in the middle of multiple rib sections of each rib centrage
The device that this rib sections matches with described probability respondence figure；And
For determining that for multiple rib sections of each rib centrage the combination of conversion candidate is so that based on described probability
Response diagram maximizes rib centrage by filling for the total probability of the paired smoothness constraint of the conversion of adjacent rib sections
Put.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US201161539561P | 2011-09-27 | 2011-09-27 | |
| US61/539561 | 2011-09-27 | ||
| US13/602,730 US8989471B2 (en) | 2011-09-27 | 2012-09-04 | Method and system for automatic rib centerline extraction using learning based deformable template matching |
| US13/602730 | 2012-09-04 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103218800A CN103218800A (en) | 2013-07-24 |
| CN103218800B true CN103218800B (en) | 2016-11-30 |
Family
ID=
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006050102A2 (en) * | 2004-10-28 | 2006-05-11 | Siemens Medical Solutions Usa, Inc. | 2d visualization for rib analysis |
| CN101782968A (en) * | 2010-02-03 | 2010-07-21 | 北京航空航天大学 | Human skeleton extracting and orientation judging method based on geodetic survey model |
| CN101923713A (en) * | 2010-08-04 | 2010-12-22 | 中国科学院自动化研究所 | Method for extracting central line of coronary artery vessel |
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006050102A2 (en) * | 2004-10-28 | 2006-05-11 | Siemens Medical Solutions Usa, Inc. | 2d visualization for rib analysis |
| CN101782968A (en) * | 2010-02-03 | 2010-07-21 | 北京航空航天大学 | Human skeleton extracting and orientation judging method based on geodetic survey model |
| CN101923713A (en) * | 2010-08-04 | 2010-12-22 | 中国科学院自动化研究所 | Method for extracting central line of coronary artery vessel |
Non-Patent Citations (2)
| Title |
|---|
| Initialization, Noise, Singularities, and Scale in Height Ridge Traversal for Tubular Object Centerline Extraction;Stephen R. Aylward 等;《IEEE TRANSACTIONS ON MEDICAL IMAGING》;20020228;第21卷(第2期);61-75 * |
| 螺旋CT及三维成像在诊断胸骨肋软骨骨折中的价值;张治邦 等;《医学影像学杂志》;20101231;第20卷(第7期);1031-1033 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8437521B2 (en) | Systems and methods for automatic vertebra edge detection, segmentation and identification in 3D imaging | |
| US7876938B2 (en) | System and method for whole body landmark detection, segmentation and change quantification in digital images | |
| US8958614B2 (en) | Image-based detection using hierarchical learning | |
| CN111862249A (en) | System and method for generating canonical imaging data for medical image processing using deep learning | |
| US8989471B2 (en) | Method and system for automatic rib centerline extraction using learning based deformable template matching | |
| Wang et al. | A review of deformation models in medical image registration | |
| Habijan et al. | Overview of the whole heart and heart chamber segmentation methods | |
| Xiang et al. | CorteXpert: A model-based method for automatic renal cortex segmentation | |
| Yu et al. | Deformable models with sparsity constraints for cardiac motion analysis | |
| Bach Cuadra et al. | Atlas-based segmentation | |
| Sundar et al. | Estimating myocardial motion by 4D image warping | |
| Göçeri et al. | Fully automated liver segmentation from SPIR image series | |
| Zheng et al. | Marginal space learning for medical image analysis | |
| Zeng et al. | Imagecas: A large-scale dataset and benchmark for coronary artery segmentation based on computed tomography angiography images | |
| Sreeja et al. | Image fusion through deep convolutional neural network | |
| Lu et al. | Automatic delineation of left and right ventricles in cardiac MRI sequences using a joint ventricular model | |
| Chen et al. | Functional region annotation of liver CT image based on vascular tree | |
| Donner12 et al. | Sparse MRF appearance models for fast anatomical structure localisation | |
| Kamencay et al. | Improved feature point algorithm for 3D point cloud registration | |
| Alam et al. | Intrinsic registration techniques for medical images: A state-of-the-art review | |
| Zhuang et al. | Automatic segmentation of left ventricle in echocardiography based on YOLOv3 model to achieve constraint and positioning | |
| Tan et al. | Automatic localization of the left ventricular blood pool centroid in short axis cardiac cine MR images | |
| Aishwariya et al. | Computer-aided fracture detection of x-ray images | |
| CN103218800B (en) | Method and apparatus for automatic rib central line pick-up | |
| Duan et al. | Surface function actives |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220208 Address after: new jersey Patentee after: SIEMENS AG Patentee after: SIEMENS Medical Co., Ltd. Address before: new jersey Patentee before: SIEMENS AG Patentee before: siemens |