CN107230204B - A kind of method and device for extracting the lobe of the lung from chest CT image - Google Patents
A kind of method and device for extracting the lobe of the lung from chest CT image Download PDFInfo
- Publication number
- CN107230204B CN107230204B CN201710374181.8A CN201710374181A CN107230204B CN 107230204 B CN107230204 B CN 107230204B CN 201710374181 A CN201710374181 A CN 201710374181A CN 107230204 B CN107230204 B CN 107230204B
- Authority
- CN
- China
- Prior art keywords
- pixel
- lung
- point
- lobe
- subelement
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20036—Morphological 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
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20101—Interactive definition of point of interest, landmark or seed
-
- 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
- G06T2207/30061—Lung
Abstract
The present invention discloses a kind of method and device that the lobe of the lung is extracted from chest CT image, is related to field of computer technology.This method includes the pulmonary parenchyma extraction process based on 3D region growing, the left and right adhesion of lung tracheae based on provincial characteristics rejects process, process, the extracted process in Pulmonary Vascular Center Road based on topological thinning and the lobe of the lung partitioning algorithm based on support vector cassification are rejected in Pulmonary Vascular root, to obtain lobe of the lung tissue.The present invention can accurately extract the lobe of the lung from chest CT image, accurately complete the qualitative assessment to the severity extent of each lobe of the lung, more accurate and effective to the diagnosing and treating of pulmonary disease.
Description
Technical field
The invention belongs to field of computer technology, are related to a kind of method and device that the lobe of the lung is extracted from chest CT image.
Background technique
CT (Computed Tomography) is using together with the X-ray beam of the Accurate collimation detector high with sensitivity
Profile scanning one by one is done at a certain position around human body, indicates the suction of organ and tissue to x-ray with different gray scales
Receipts degree, for example, on chest CT image, region the expression tracheae, pulmonary parenchyma of low-density, highdensity region expression blood vessel,
The features such as thoracic cavity, bone etc. have sweep time fast, image clearly, can be used for the inspection of a variety of diseases, be doctor's inspections and examinations
Disease provides convenience reliable foundation.
Since CT equipment can get clearly chest CT image, by means of CT image at the diagnosis chronic obstructive pulmonary disease state of an illness
One main means, however current medical level can only accomplish the quantitative Diagnosis to single lung, for the single lobe of the lung state of an illness
Qualitative assessment can not be accomplished.The treatment of chronic obstructive pulmonary disease usually requires lobe of the lung volume reduction surgery, however operation consent can not but carry out the lobe of the lung subtracts
The qualitative assessment of appearance, this certainly will will affect the therapeutic effect of chronic obstructive pulmonary disease.If pulmo and left and right can be accurately partitioned into
The various pieces of lung, the i.e. lobe of the lung accurately complete that chronic obstructive pulmonary disease will be examined the qualitative assessment of the severity extent of each lobe of the lung
Disconnected and treatment is of great importance.
Summary of the invention
(1) technical problems to be solved
In order to solve the above problem of the prior art, the present invention provides a kind of side that the lobe of the lung is extracted from chest CT image
Method, this method can accurately extract the lobe of the lung from chest CT image, accurately complete to the severity extent of each lobe of the lung
It is quantitatively evaluated, it is more accurate and effective to the diagnosing and treating of pulmonary disease.
The present invention also provides a kind of from chest CT image extracts the device of the lobe of the lung, which can be accurately from chest CT
Extract the lobe of the lung in image, accurately complete the qualitative assessment to the severity extent of each lobe of the lung, diagnosis to pulmonary disease and
Treatment is more accurate and effective.
(2) technical solution
In order to achieve the above object, the main technical schemes that the present invention uses include:
A method of extracting the lobe of the lung from chest CT image, comprising the following steps:
S1, the n-layer chest CT image for receiving input, obtain intermediate image layer, and the specified pixel point for choosing lung areas is
Seed point carries out the growth of the region 3D according to setting segmentation threshold and initial seed point, obtains without pulmonary vascular pulmonary parenchyma area
Domain, wherein n is natural number;
S2, pass through the left and right adhesion of lung elimination method for corroding approximate adhesion boundary in the resulting pulmonary parenchyma region step S1,
Eliminating conglutination obtains two independent pulmonary parenchymas;
S3, it is filled into the resulting pulmonary parenchyma of step S2 using the operation handlebar blood vessel for first expanding post-etching, passes through maximum kind
Between variance method automatically obtain optimum segmentation threshold value T0;
S4, with the resulting optimum segmentation threshold value T of step S30As the threshold value of segmentation blood vessel, the point work for being greater than threshold value is calculated
For vessel seed point, the growth of the region 3D is carried out, obtains Pulmonary Vascular;
S5, the blood vessel that the different lobes of the lung are disconnected from vascular root make the blood vessel in each lobe of the lung become an independent company
Logical domain;
S6, according to the resulting Pulmonary Vascular of step S5, Pulmonary Vascular center path is extracted using elimination approach;
S7, two category support vector machines models of training, obtain two category support vector machines classifiers, are looked for by the classifier
Interface between different lobe of the lung blood vessels using this interface as the interface of the lobe of the lung and calculates acquisition pulmonary parenchyma pixel
Discriminant classification function differentiates each pixel in pulmonary parenchyma, obtains lobe of the lung tissue.
Preferably, the step S1 the following steps are included:
S11, n (n is natural number) the layer chest CT image for receiving input obtain intermediate image layer, on intermediate image layer
The specified pixel point of lung areas is chosen as initial seed point;
S12, segmentation threshold is arranged according to the characteristics of pulmonary parenchyma;
26 neighborhoods of the labeled initial seed point of S13, search, choose one of pixel;
Whether pixel selected by S14, judgement has been labeled, if then return step S13, no to then follow the steps
S15;
Whether the gray value of pixel selected by S15, judgement is less than segmentation threshold, if so, the pixel is marked
Remember and be added mark queue, otherwise the stop flag pixel, executes step S16;
S16, judge whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish, if so then execute step S17,
Otherwise return step S13;
Whether S17, judge mark queue are sky, if being not sky, then a mark point are taken out from the queue as initially
Labeled seed point, return step S13, the pixel collection being otherwise labeled is exactly to be partitioned into without pulmonary vascular lung
Parenchyma section.
Preferably, the step S2 the following steps are included:
S21, to the pulmonary parenchyma region decision of acquisition, whether it is the approximately equal connected domain of two sizes, if then tying
Beam, it is no to then follow the steps S22;
S22, the connected domain number for counting intermediate image layer reject the lesser connected domain region of pixel number, retain pixel
Count biggish connected domain region, and judge the number of connected domain and the size of each connected domain, if there are two connected domains simultaneously
And size is approximately equal, terminates, it is no to then follow the steps S23;
S23, successively statistics connected domain number, if only one connected domain, pulmonary parenchyma one is scheduled on this layer of adhesion, executes
Otherwise step S24 records the number of each connected domain pixel, maximum two connected domains of pixel number are found, if face
The maximum two connected regions pixel number number of product is suitable, then this layer of CT image or so pulmonary parenchyma adhesion, executes step
S26, otherwise adhesion and biggish connected domain are pulmonary parenchyma, execute step S24;
S24, adhesion region is replaced with line segment approximation, is converted into the position of determining straight line, highest point of the lower boundary in the direction y
The as lower extreme point of near linear, the point nearest from lower extreme point is then the upper extreme point of near linear in all the points of coboundary, is sought
Straight line is determined after finding this two o'clock, executes step S25;
S25, the near linear boundary on every layer of pulmonary parenchyma region adhesion image and 26 around straight border is eroded
The mark value of neighborhood point obtains the pulmo region of two adhesions, executes step S26;
Whether S26, to judge all layers scanned, no to then follow the steps S23 if then terminating.
Preferably, the step S3 the following steps are included:
S31, by choose suitable structural element executed on all chest CT image layers first expand post-etching close fortune
Operation is calculated, Pulmonary Vascular is filled into segmented good pulmonary parenchyma;
S32, one layer close to centre is selected in CT image, automatically obtain optimum segmentation using maximum variance between clusters
Threshold value T0。
Preferably, the step S4 the following steps are included:
S41, obtaining step S3 the pulmonary parenchyma with blood vessel in grey scale pixel value be greater than segmentation threshold T0All pixels point,
And it is labeled as blood vessel.
S42, a pixel is chosen in the resulting blood vessel of step S41 as initial seed point;
26 neighborhoods of the labeled initial seed point of S43, search, choose one of pixel;
Whether pixel selected by S44, judgement has been labeled, if then return step S43, no to then follow the steps
S45;
Whether the gray value of pixel selected by S45, judgement is greater than segmentation threshold, if so, the pixel is marked
Remember and be added mark queue, otherwise the stop flag pixel, executes step S46;
S46, judge whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish, if so then execute step S47,
Otherwise return step S43;
Whether S47, judge mark queue are sky, if being not sky, then a mark point are taken out from the queue as initially
Labeled seed point, return step S43, the pixel collection being otherwise labeled is exactly the Pulmonary Vascular being partitioned into.
Preferably, the step S5 the following steps are included:
S51, lung edge and the Pulmonary Vascular pixel close to tracheae are found, thens follow the steps S52 if finding, otherwise continues to sweep
Retouch lung tissue;
S52, Pulmonary Vascular is disconnected using the blood vessel at the edge of corrosion pulmonary parenchyma, and counts of the connected domain of vascularization
Number thens follow the steps S53 if the number of connected domain is five, no to then follow the steps S51;
S53, the color different to each connected component labeling, all mark finish beam.
Preferably, the step S6 the following steps are included:
S61, scanning Pulmonary Vascular pixel, judge whether it is boundary point, if 6 neighborhood of pixel or 26 neighborhoods have back
Scenic spot pixel is then boundary point, executes step S62, otherwise marks scanned, skips and continue to execute step S61;
Otherwise S62, the Euler's characteristic for judging boundary point are skipped and are continued to execute if Euler's characteristic is constant to then follow the steps S63
Step S61;
S63, judge whether boundary point is simple point, if the point on former three-dimensional space topological structure without influence if be simple
Point, while by the point deletion, execute step S64;It is no to then follow the steps S61;
S64, judging blood vessel pixel, whether there are also other simple points, if nothing, centerline extraction terminates, no to then follow the steps
S61。
Preferably, the step S7 the following steps are included:
S71, according to the interface between the lobe of the lung closest to the track of radial basis function, select Radial basis kernel function as to height
The function of dimension space mapping, it is determined that after kernel function, execute step S72;
S72, for left lung, it is assumed that the pixel of left lung be N1, blood vessel pixel number be n1, the institute of the upper leaf of left lung
There is blood vessel pixel label to be positive, while blood vessel pixels all on inferior lobe label is negative, obtains a discriminant function, execute
Step S73;
S73, it is gone to judge the symbol of each pixel in left pulmonary parenchyma with discriminant function, if symbol is positive, the point
Belong to leaf, otherwise the point belongs to inferior lobe, and the number of statistical pixel, and if N1, then left lung segmentation is completed, step S74 is executed,
Otherwise step S73 is continued to execute;
S74, for right lung, it is assumed that the pixel of right lung be N2, blood vessel pixel number be n2, for use two classification support
The upper leaf of right lung and middle period are combined into an entirety and regarded as " upper leaf ", execution step S75 by vector machine classification;
S75, a discriminant classification function is obtained using the method as the segmentation of left lung, inferior lobe is extracted first, step
All blood vessel pixels label of " upper leaf " that rapid S74 is regarded as is positive, while blood vessel pixels all on inferior lobe are labeled as
It is negative, a discriminant function is obtained, step S76 is executed;
S76, it is gone to judge the symbol of each substantial pixel of right lung with discriminant function, if symbol is positive, the point
Belong to " upper leaf " that step S74 is regarded as, otherwise the point belongs to inferior lobe, and the number of statistical pixel, and then inferior lobe is divided if N2
It completes, executes step S77, otherwise continue to execute step S76;
After inferior lobe is completed in S77, segmentation, the blood vessel Training Support Vector Machines model of leaf and middle period is then used again, obtains one
A differentiation surface function removes the partial segmentation of inferior lobe to right lung, extracts leaf and middle period, executes step S78;
S78, for the upper leaf of right lung and middle period, it is assumed that the pixel of right lung is N3, and blood vessel pixel number is n3, right
All blood vessel pixels label of the upper leaf of lung is positive, while the middle period of right lung upper all blood vessel pixel labels are negative, and obtains
To a discriminant function, step S79 is executed;
S79, it is gone to judge the symbol of each substantial pixel of right lung with discriminant function, if symbol is positive, the point
Belong to leaf, otherwise the point belongs to middle period, and the number of statistical pixel, and if N3, then superior lobe of right lung and middle period segmentation are completed, until
This pulmo segmentation lobe of the lung terminates.
A kind of device extracting the lobe of the lung from chest CT image, comprising:
CT image input units, n-layer chest CT image for receiving input, wherein n is natural number;
Without pulmonary vascular pulmonary parenchyma acquiring unit, the specified pixel point for choosing lung areas is seed point, according to setting
Determine segmentation threshold and initial seed point carries out the growth of the region 3D, obtains without pulmonary vascular pulmonary parenchyma region;
Two independent pulmonary parenchyma acquiring units, after obtaining left and right pulmonary parenchyma, judge left and right pulmonary parenchyma whether adhesion, if
Adhesion then obtains two independent pulmonary parenchymas by rejecting the adhesion boundary of tracheae and near linear;
Pulmonary Vascular segmentation threshold acquiring unit is filled into pulmonary parenchyma using the operation handlebar blood vessel for first expanding post-etching,
Optimum segmentation threshold value T is automatically obtained by maximum variance between clusters0;
Pulmonary Vascular extraction unit, with resulting optimum segmentation threshold value T0As the threshold value of segmentation blood vessel, calculates and be greater than threshold
The point of value carries out the growth of the region 3D as vessel seed point, obtains Pulmonary Vascular;
Pulmonary Vascular is divided into independent communication domain unit, and the blood vessel of the different lobes of the lung is disconnected from vascular root, makes each lung
The blood vessel of leaf becomes an independent connected domain;
Pulmonary Vascular centerline extraction unit is divided into the resulting lung blood of independent communication domain unit according to the Pulmonary Vascular
Pipe, Pulmonary Vascular center path is extracted using elimination approach;
Lobe of the lung cutting unit, by two sorter model of blood vessel Training Support Vector Machines of the adjacent lobe of the lung, left lung has two
A lobe of the lung, training one, right lung is there are three the lobe of the lung, and twice, three two classifiers of support vector machines are obtained in pulmo for training, leads to
It crosses support vector machines and finds interface between different lobe of the lung blood vessels, be partitioned into the lobe of the lung.
(3) beneficial effect
The beneficial effects of the present invention are:
The present invention provides a kind of from chest CT image extracts the method and device of the lobe of the lung, wherein method includes being based on
The pulmonary parenchyma extraction process of 3D region growing, the left and right adhesion of lung tracheae based on provincial characteristics rejects process, Pulmonary Vascular root is picked
The extracted process in Pulmonary Vascular Center Road except process, based on topological thinning and the lobe of the lung based on support vector cassification, which are divided, to be calculated
Method obtains lobe of the lung tissue by journey processed above.The present invention can accurately extract the lobe of the lung from chest CT image, quasi-
True completion is more accurate to the diagnosing and treating of pulmonary disease and effectively to the qualitative assessment of the severity extent of each lobe of the lung.
Detailed description of the invention
Fig. 1 is the committed step flow chart that the lobe of the lung is extracted in the slave chest CT image of preferred embodiment.
Fig. 2 is the method flow diagram that the lobe of the lung is extracted in the slave chest CT image of preferred embodiment.
Fig. 3 is the grey level histogram of preferred embodiment chest CT image.
Fig. 4 is pulmonary parenchyma area results schematic diagram of the preferred embodiment without blood vessel.
Fig. 5 is pulmonary parenchyma area results schematic diagram of the preferred embodiment with tracheae.
Fig. 6 is the pulmonary parenchyma area results schematic diagram that preferred embodiment rejects tracheae.
Fig. 7 is the pulmonary parenchyma area results schematic diagram that preferred embodiment boundary is adhered.
Fig. 8 is the pulmonary parenchyma area results schematic diagram of preferred embodiment blood vessel filling.
Fig. 9 is that preferred embodiment Pulmonary Vascular extracts result schematic diagram.
Figure 10 is preferred embodiment Pulmonary Vascular centerline extraction result schematic diagram.
Figure 11 is that preferred embodiment Pulmonary Vascular disconnects root extraction result schematic diagram.
Figure 12 is that the preferred embodiment lobe of the lung extracts result schematic diagram.
Specific embodiment
In order to preferably explain the present invention, in order to understand, with reference to the accompanying drawing, by specific embodiment, to this hair
It is bright to be described in detail.
Preferred embodiment
As depicted in figs. 1 and 2, the present embodiment proposes a kind of method that the lobe of the lung is preferably extracted from chest CT image,
This method for the lobe of the lung is extracted from chest CT image the following steps are included:
S1, n (n is natural number) the layer chest CT image for receiving input obtain intermediate image layer, choose the finger of lung areas
Determining pixel is seed point, carries out the growth of the region 3D according to setting segmentation threshold and initial seed point, obtains without Pulmonary Vascular
Pulmonary parenchyma region.
Specifically, step S1 the following steps are included:
S11, n (n is natural number) the layer chest CT image for receiving input obtain intermediate image layer, on intermediate image layer
The specified pixel point of lung areas is chosen as initial seed point;
S12, segmentation threshold is arranged according to the characteristics of pulmonary parenchyma;
26 neighborhoods of the labeled initial seed point of S13, search, choose one of pixel;
Whether pixel selected by S14, judgement has been labeled, if then return step S13, no to then follow the steps
S15;
Whether the gray value of pixel selected by S15, judgement is less than segmentation threshold, if so, the pixel is marked
Remember and be added mark queue, otherwise the stop flag pixel, executes step S16;
S16, judge whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish, if so then execute step S17,
Otherwise return step S13;
Whether S17, judge mark queue are sky, if being not sky, then a mark point are taken out from the queue as initially
Labeled seed point, return step S13, the pixel collection being otherwise labeled is exactly to be partitioned into without pulmonary vascular lung
Parenchyma section.
S2, pass through the left and right adhesion of lung elimination method for corroding approximate adhesion boundary in the resulting pulmonary parenchyma region step S1,
Eliminating conglutination obtains two independent pulmonary parenchymas.
Some researchers are first corroded with pulmonary parenchyma of the morphologic method to adhesion, until pulmo adhesion, then
It marks respectively, reflation is returned original size.Although this method has successfully separated left lung and right lung essence, lose
The details on boundary, makes the boundary of some lungs deviate original position, increases the error of segmentation.Also some calculate complicated side
Although method can accurately separate left lung and right lung essence, but pay a large amount of calculating time cost, so that answering
With being difficult to meet the requirement of time.
By finding to a large amount of adhesion regional observation, the adhesion of left lung and right lung essence is simply present in certain layers of CT figure
As upper, the case where adhesion is not present in most of layers of CT image, even adhesion, also only one piece of region adhesion of very little.Therefore
The specific layer CT image of adhesion can be found first, then layer-by-layer eliminating conglutination.
The method for finding adhering layer is connected domain number successively to be counted, if only one connected domain, pulmonary parenchyma are certain
In this layer of adhesion.If there is multiple connected domains, then the number of each connected domain pixel is recorded, it is maximum to find pixel number
Two connected domains, pixel number use N respectively1And N2It indicates, if
Threshold is usually arranged as 10 in above formula, and formula specifically means that if maximum two connected regions of area
Domain pixel number number is suitable, then this layer of CT image or so pulmonary parenchyma adhesion, because left lung is real on the CT image of same layer
The area of matter region and right lung parenchyma section is comparable;, whereas if two maximum two connected regions pixel numbers
Ratio great disparity, then this layer of CT image or so pulmonary parenchyma adhesion, at this time a pulmonary parenchyma region letter of guarantee left side for biggish connection domain representation
Pulmonary parenchyma region and right lung parenchyma section, and it is lesser connection domain representation tissue regions some be noise interference, some are
Tiny bronchus, and these bronchuses will not make left and right adhesion of lung.
It is then to find specific adhesion boundary position after determining specific adhering layer CT image.Because adhesion is only very
Small region, and the boundary of adhesion is a very short curve, therefore can be replaced with line segment approximation.The benefit done so
It is, it can reduce calculation amount, to reduce the runing time of algorithm, and have little influence on the precision of pulmonary parenchyma segmentation.
Specifically, step S2 the following steps are included:
S21, to the pulmonary parenchyma region decision of acquisition, whether it is the approximately equal connected domain of two sizes, if then tying
Beam, it is no to then follow the steps S22;
S22, the connected domain number for counting intermediate image layer reject the lesser connected domain region of pixel number, retain pixel
Count biggish connected domain region, and judge the number of connected domain and the size of each connected domain, if there are two connected domains simultaneously
And size is approximately equal, terminates, it is no to then follow the steps S23;
S23, successively statistics connected domain number, if only one connected domain, pulmonary parenchyma one is scheduled on this layer of adhesion, executes
Otherwise step S24 records the number of each connected domain pixel, maximum two connected domains of pixel number are found, if face
The maximum two connected regions pixel number number of product is suitable, then this layer of CT image or so pulmonary parenchyma adhesion, executes step
S26, otherwise adhesion and biggish connected domain are pulmonary parenchyma, execute step S24;
S24, adhesion region is replaced with line segment approximation, is converted into the position of determining straight line, highest point of the lower boundary in the direction y
The as lower extreme point of near linear, the point nearest from lower extreme point is then the upper extreme point of near linear in all the points of coboundary, is sought
Straight line is determined after finding this two o'clock, executes step S25;
S25, the near linear boundary on every layer of pulmonary parenchyma region adhesion image and 26 around straight border is eroded
The mark value of neighborhood point obtains the pulmo region of two adhesions, executes step S26;
Whether S26, to judge all layers scanned, no to then follow the steps S23 if then terminating.
S3, it is filled into the resulting pulmonary parenchyma of step S2 using the operation handlebar blood vessel for first expanding post-etching, passes through maximum kind
Between variance method automatically obtain optimum segmentation threshold value T0。
Morphologic first expansion post-etching operation is carried out on the image and is known as closed operation, it is therefore an objective to which packing ratio structural element is small
Hole and smoothed image edge.
Expansion is " lengthening " or the operation of " thicker " in bianry image.This special mode and thicker degree are by one
A set control for being known as structural element, structural element are indicated with the matrix of 0 and 1.For corrosion, as in expansion,
Contraction mode and degree are controlled by a structural element.
Maximum variance between clusters are that original image is divided into foreground and background two parts using threshold value.Remember that T is prospect and background
Segmentation threshold, prospect points accounts for image scaled be w0, average gray u0;It is w that background points, which account for image scaled,1, average gray
For u1, then the overall average gray scale of image are as follows:
U=w0u0+w1u1 (1.1)
The variance of foreground and background image:
σ2=w0(u0-u)2+w1(u1-u)2 (1.2)
Formula (1.1) bring into formula (1.2) obtain inter-class variance calculation formula it is as follows:
σ2=w0w1(u0-u1)2 (1.3)
Another σ is found out using formula (1.3)2Maximum threshold value Tmax, threshold value as segmented image foreground and background.
Specifically, step S3 the following steps are included:
S31, by choose suitable structural element executed on all chest CT image layers first expand post-etching close fortune
Operation is calculated, Pulmonary Vascular is filled into segmented good pulmonary parenchyma;
S32, one layer close to centre is selected in CT image, automatically obtain optimum segmentation using maximum variance between clusters
Threshold value T0。
S4, with the resulting optimum segmentation threshold value T of step S30As the threshold value of segmentation blood vessel, the point work for being greater than threshold value is calculated
For vessel seed point, the growth of the region 3D is carried out, obtains Pulmonary Vascular.Wherein, Pulmonary Vascular extraction effect is as shown in Figure 9.
Specifically, step S4 the following steps are included:
S41, obtaining step S3 the pulmonary parenchyma with blood vessel in grey scale pixel value be greater than segmentation threshold T0All pixels point,
And it is labeled as blood vessel.
S42, a pixel is chosen in the resulting blood vessel of step S41 as initial seed point;
26 neighborhoods of the labeled initial seed point of S43, search, choose one of pixel;
Whether pixel selected by S44, judgement has been labeled, if then return step S43, no to then follow the steps
S45;
Whether the gray value of pixel selected by S45, judgement is greater than segmentation threshold, if so, the pixel is marked
Remember and be added mark queue, otherwise the stop flag pixel, executes step S46;
S46, judge whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish, if so then execute step S47,
Otherwise return step S43;
Whether S47, judge mark queue are sky, if being not sky, then a mark point are taken out from the queue as initially
Labeled seed point, return step S43, the pixel collection being otherwise labeled is exactly the Pulmonary Vascular being partitioned into.
S5, the blood vessel that the different lobes of the lung are disconnected from vascular root make the blood vessel in each lobe of the lung become an independent company
Logical domain.
Specifically, step S5 the following steps are included:
S51, lung edge and the Pulmonary Vascular pixel close to tracheae are found, thens follow the steps S52 if finding, otherwise continues to sweep
Retouch lung tissue;
S52, Pulmonary Vascular is disconnected using the blood vessel at the edge of corrosion pulmonary parenchyma, and counts of the connected domain of vascularization
Number thens follow the steps S53 if the number of connected domain is five, no to then follow the steps S51;
S53, the color different to each connected component labeling, all mark finish beam.
S6, according to the resulting Pulmonary Vascular of step S5, Pulmonary Vascular center path is extracted using elimination approach.
Specifically, step S6 the following steps are included:
S61, scanning Pulmonary Vascular pixel, judge whether it is boundary point, if 6 neighborhood of pixel or 26 neighborhoods have back
Scenic spot pixel is then boundary point, executes step S62, otherwise marks scanned, skips and continue to execute step S61;
Otherwise S62, the Euler's characteristic for judging boundary point are skipped and are continued to execute if Euler's characteristic is constant to then follow the steps S63
Step S61;
S63, judge whether boundary point is simple point, if the point on former three-dimensional space topological structure without influence if be simple
Point, while by the point deletion, execute step S64;It is no to then follow the steps S61;
S64, judging blood vessel pixel, whether there are also other simple points, if nothing, centerline extraction terminates, no to then follow the steps
S61。
S7, two category support vector machines models of training, obtain two category support vector machines classifiers, are looked for by the classifier
Interface between different lobe of the lung blood vessels using this interface as the interface of the lobe of the lung and calculates acquisition pulmonary parenchyma pixel
Discriminant classification function differentiates each pixel in pulmonary parenchyma, obtains lobe of the lung tissue.Wherein, the lobe of the lung extracts result such as
Shown in Figure 12.
Support vector machines (SVM, Support Vector Machines) is proposed first by Vapnik et al., be from
The optimal classification surface of linear separability gradually develops.
In three dimensions, between the Pulmonary Vascular of two adjacent lobes of the lung and be not present a plane, make on a lobe of the lung
Each pulmonary vascular pixel makes each pulmonary vascular pixel on another lobe of the lung flat in the side of plane
The other side in face, therefore this is a linearly inseparable problem.
For solving the problems, such as linearly inseparable with svm classifier, one appropriate kernel function of selection is first had to, these three-dimensionals are made
Data in space are mapped in the feature space of a more higher-dimension, so that these sample datas become linear separability.It is logical
It crosses and a large amount of lung tissue CT data observation is found, the interface between the lobe of the lung is selected closest to the track of radial basis function
Radial basis kernel function is as the function mapped to higher dimensional space.
The form of Radial basis kernel function are as follows:
The discriminant function that support vector machines at this time constructs are as follows:
Wherein, s is the number of supporting vector, and supporting vector can determine the center of radial basis function.Radial base core
The dimension of the corresponding feature space of function is infinity, and limited data sample must be linear separability in this feature space,
Therefore Radial basis kernel function is most commonly used.
Algorithm of support vector machine is finally attributed to the problem of seeking quadratic programming (QP), can be rewritten as following rectangular
Formula.Given sample (xi,yi), i=1,2,3...n, kernel function K (xi,xj) and adjustment parameter C, minimizing.
Constraint condition are as follows:
αTY=0,0≤α≤C (1.7)
Kernel matrix H's is defined as:
H=[hij]=yiyjK(xi,xj) (i, j=1,2,3...n) (1.8)
Wherein, α=(α1,α2,...αn)T, α=(y1,y2,...yn)T
If training data is on a grand scale, because the scale of matrix H is square of training data, matrix H is possible to big
To can not normally be handled with computer.
In view of the above problems, it has been proposed that method below for large-scale data sample training.
1.Chunking algorithm
A kind of method that Vapnik et al. first proposed solution SVM training memory space problem, referred to as Chunking are calculated
Method.In formula (1.6), if removing row and column corresponding with zero Lagrange multiplier, it is worth constant.Therefore it can will solve
The QP PROBLEM DECOMPOSITION of support vector machines is at a series of lesser QP problems.It is true for solving the final goal of these lesser QP problems
Fixed all non-zero Lagrange multipliers, and remove zero all Lagrange multipliers.
2. decomposition algorithm
Osuan et al. first proposed decomposition algorithm, decomposition algorithm be the QP PROBLEM DECOMPOSITION of solution support vector machines at
A series of lesser QP problems, but its working set size remains unchanged, the demand to memory becomes line at quadratic relationship from s
Sexual intercourse.It can handle data sample point up to 110,000, supporting vector is more than 100,000 problems.
3.SMO algorithm
Serial minimum optimization algorithm (Sequential Minimal Optimization, SMO) is mentioned first by Platt
Out, SMO algorithm also belongs to a kind of decomposition algorithm, and working space only includes two data samples, only right in every single-step iteration
Two Lagrange multipliers optimize.Although QP problem increases in SMO, total calculating speed is substantially increased,
And this algorithm, completely without big matrix, thus the requirement not additional to memory space is handled, very big SVM is trained to be asked
Topic can also be run on a personal computer.Due to above advantage, SMO algorithm becomes widest a kind of in practical applications
Method.
In the present embodiment, for left lung, it is assumed that the blood vessel pixel number of left lung is n, because only that two lobes of the lung, because
All blood vessel pixels of this upper leaf are labeled as (xi,yi),i≤n,xi∈R3,yi=+1, while blood vessel pictures all on inferior lobe
Vegetarian refreshments is labeled as (xi,yi),i≤n,xi∈R3,yi=-1.Obtain a discriminant function f1(x).With discriminant function f1(x) it goes to sentence
Break in left lung tissue often with the symbol of a pixel, if symbol is positive, which belongs to leaf, and otherwise the point belongs to inferior lobe.
Upper lobe of left lung and inferior lobe in this way can be divided out.
And for right lung, because there are three the lobe of the lung, respectively upper leaf, middle period and inferior lobe.In order to use two classification
The upper leaf of right lung and middle period are combined into an entirety first by support vector cassification, thus can be used and left lung segmentation one
The method of sample obtains a discriminant classification function f2(x), inferior lobe is extracted first, then again with leaf and the training of the blood vessel in middle period
Supporting vector machine model obtains a differentiation surface function f3(x), the part subseries again that inferior lobe is removed to right lung, extracts
Leaf and middle period.
Specifically, step S7 the following steps are included:
S71, according to the interface between the lobe of the lung closest to the track of radial basis function, select Radial basis kernel function as to height
The function of dimension space mapping, it is determined that after kernel function, execute step S72;
S72, for left lung, it is assumed that the pixel of left lung be N1, blood vessel pixel number be n1, the institute of the upper leaf of left lung
There is blood vessel pixel label to be positive, while blood vessel pixels all on inferior lobe label is negative, obtains a discriminant function, execute
Step S73;
S73, it is gone to judge the symbol of each pixel in left pulmonary parenchyma with discriminant function, if symbol is positive, the point
Belong to leaf, otherwise the point belongs to inferior lobe, and the number of statistical pixel, and if N1, then left lung segmentation is completed, step S74 is executed,
Otherwise step S73 is continued to execute;
S74, for right lung, it is assumed that the pixel of right lung be N2, blood vessel pixel number be n2, for use two classification support
The upper leaf of right lung and middle period are combined into an entirety and regarded as " upper leaf ", execution step S75 by vector machine classification;
S75, a discriminant classification function is obtained using the method as the segmentation of left lung, inferior lobe is extracted first, step
All blood vessel pixels label of " upper leaf " that rapid S74 is regarded as is positive, while blood vessel pixels all on inferior lobe are labeled as
It is negative, a discriminant function is obtained, step S76 is executed;
S76, it is gone to judge the symbol of each substantial pixel of right lung with discriminant function, if symbol is positive, the point
Belong to " upper leaf " that step S74 is regarded as, otherwise the point belongs to inferior lobe, and the number of statistical pixel, and then inferior lobe is divided if N2
It completes;Execute step S77;
After inferior lobe is completed in S77, segmentation, the blood vessel Training Support Vector Machines model of leaf and middle period is then used again, obtains one
A differentiation surface function removes the partial segmentation of inferior lobe to right lung, extracts leaf and middle period, executes step S78;
S78, for the upper leaf of right lung and middle period, it is assumed that the pixel of right lung is N3, and blood vessel pixel number is n3, right
All blood vessel pixels label of the upper leaf of lung is positive, while the middle period of right lung upper all blood vessel pixel labels are negative, and obtains
To a discriminant function, step S79 is executed;
S79, it is gone to judge the symbol of each substantial pixel of right lung with discriminant function, if symbol is positive, the point
Belong to leaf, otherwise the point belongs to middle period, and the number of statistical pixel, and if N3, then superior lobe of right lung and middle period segmentation are completed, until
This pulmo segmentation lobe of the lung terminates.
Meanwhile the present embodiment additionally provides a kind of device that the lobe of the lung is preferably extracted from chest CT image, including CT figure
It is obtained as input unit, without pulmonary vascular pulmonary parenchyma acquiring unit, two independent pulmonary parenchyma acquiring units, Pulmonary Vascular segmentation thresholds
Unit, Pulmonary Vascular extraction unit, Pulmonary Vascular is taken to be divided into independent communication domain unit, Pulmonary Vascular centerline extraction unit and the lobe of the lung
Cutting unit.
Wherein, CT image input units, n-layer chest CT image for receiving input, wherein n is natural number.
Without pulmonary vascular pulmonary parenchyma acquiring unit, the specified pixel point for choosing lung areas is seed point, according to setting
Determine segmentation threshold and initial seed point carries out the growth of the region 3D, obtains without pulmonary vascular pulmonary parenchyma region.
Specifically, including following each unit without pulmonary vascular pulmonary parenchyma acquiring unit:
Initial seed point selection unit chooses the specified pixel point of lung areas as initial kind on intermediate tomographic image
Sub- point.Segmentation threshold obtains subelement, and segmentation threshold is arranged according to the characteristics of pulmonary parenchyma.Pixel chooses subelement, searches
26 neighborhoods of the labeled initial seed point of rope, choose one of pixel.Pixel marker for judgment subelement, judgement
Whether selected pixel has been labeled, if so, notice pixel chooses subelement, otherwise notifies pixel value judgement
Unit.Whether the gray value of pixel value judgment sub-unit, the selected pixel of judgement is less than segmentation threshold, if so, handle
The pixel marks and mark queue is added, otherwise the stop flag pixel, notifies 26 neighborhood territory pixel point search of seed point
Subelement.26 neighborhood search subelement of seed point, judges whether 26 neighborhood territory pixel points of seed point are all searched for and judged
Finish, if so, notification indicia queue judgment sub-unit, otherwise notifies the pixel to choose subelement.Mark queue judgement
Whether unit, judge mark queue are sky, if being not sky, then a mark point are taken out from the queue as initially labeled
Seed point, notice pixel choose subelement, and the pixel collection being otherwise labeled is exactly to be partitioned into without Pulmonary Vascular
Pulmonary parenchyma.
Two independent pulmonary parenchyma acquiring units, after obtaining left and right pulmonary parenchyma, judge left and right pulmonary parenchyma whether adhesion, if
Adhesion then obtains two independent pulmonary parenchymas by rejecting the adhesion boundary of tracheae and near linear.
Specifically, two independent pulmonary parenchyma acquiring units include following each unit:
Pulmonary parenchyma connected domain number judgment sub-unit judges whether it is two size approximation phases to the pulmonary parenchyma of acquisition
Deng connected domain, if then terminating, otherwise notice reject tracheae subelement.Tracheae subelement is rejected, intermediate image layer is counted
Connected domain number, reject the lesser connected domain region of pixel number, retain the biggish connected domain region of pixel number, and judge
The size of the number of connected domain and each connected domain, if there are two connected domain and size it is approximately equal if terminate, otherwise lead to
Know a layer connected domain number judgment sub-unit.Layer connected domain number judgment sub-unit, successively counts connected domain number, if only
One connected domain, then pulmonary parenchyma one is scheduled on this layer of adhesion, and notice straight line determines subelement, otherwise records each connected domain pixel
The number of point, finds maximum two connected domains of pixel number, if the maximum two connected regions pixel number number of area
Quite, then this layer of CT image or so pulmonary parenchyma adhesion, notice pixel layer scan judgment sub-unit, otherwise adhesion and biggish company
Logical domain is pulmonary parenchyma, and notice straight line determines subelement.Straight line determines subelement, replaces adhesion region with line segment approximation, conversion
For determine straight line position, lower boundary in the lower extreme point that the highest point in the direction y is near linear, all the points of coboundary from
The nearest point of lower extreme point is then the upper extreme point of near linear, determines straight line, notice corrosion near linear side after searching out this two o'clock
Boundary's subelement.Corrode near linear boundary subelement, erodes the near linear side on every layer of pulmonary parenchyma region adhesion image
The mark value of 26 neighborhood points around boundary and straight border, obtains the pulmo region of two adhesions, and notice pixel layer is swept
Retouch judgment sub-unit.Pixel layer scans judgment sub-unit, whether scanned judges all layers, if then terminating, otherwise leads to
Know a layer connected domain number judgment sub-unit.
Pulmonary Vascular segmentation threshold acquiring unit is filled into pulmonary parenchyma using the operation handlebar blood vessel for first expanding post-etching,
Optimum segmentation threshold value T is automatically obtained by maximum variance between clusters0。
Specifically, Pulmonary Vascular segmentation threshold acquiring unit includes following each unit:
Closed operation operating unit is first expanded by choosing suitable structural element and executing on all chest CT image layers
The closed operation of post-etching operates, and Pulmonary Vascular is filled into segmented good pulmonary parenchyma.Optimum segmentation threshold value acquiring unit,
One layer close to centre is selected in CT image, automatically obtains optimum segmentation threshold value T using maximum variance between clusters0。
Pulmonary Vascular extraction unit, with resulting optimum segmentation threshold value T0As the threshold value of segmentation blood vessel, calculates and be greater than threshold
The point of value carries out the growth of the region 3D as vessel seed point, obtains Pulmonary Vascular.
Specifically, Pulmonary Vascular extraction unit includes following each unit:
Middle layer Pulmonary Vascular extracts subelement, obtains in middle layer grey scale pixel value in the pulmonary parenchyma with blood vessel and is greater than point
Cut threshold value T0All pixels point, and be labeled as blood vessel.Initial seed point obtains subelement, mentions in the middle layer Pulmonary Vascular
It takes and chooses a pixel in the resulting blood vessel of subelement as initial seed point.Pixel obtains subelement, and search is marked
26 neighborhoods of the initial seed point of note, choose one of pixel.Pixel marker for judgment subelement, judgement selected by
Pixel whether be labeled, if notice pixel obtain subelement, otherwise notify pixel judgment sub-unit.Picture
Element value judgment sub-unit, whether the gray value of the selected pixel of judgement is greater than segmentation threshold, if so, the pixel
It marks and mark queue is added, notify 26 neighborhood search subelement of seed point, otherwise the stop flag pixel.Seed point 26
Neighborhood search subelement, judges whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish, if then notifying to mark
Remember queue judgment sub-unit, otherwise pixel is notified to obtain subelement.Mark queue judgment sub-unit, judge mark queue are
It is no a mark point then to be taken out from the queue as initially labeled seed point, notice pixel obtains if being not sky for sky
Subelement, the pixel collection being otherwise labeled are exactly the Pulmonary Vascular being partitioned into.
Pulmonary Vascular is divided into independent communication domain unit, and the blood vessel of the different lobes of the lung is disconnected from vascular root, makes each lung
The blood vessel of leaf becomes an independent connected domain.
Specifically, it includes following each unit that Pulmonary Vascular, which is divided into independent communication domain unit:
Preliminary sweep pulmonary parenchyma unit is found lung edge and is notified close to the Pulmonary Vascular pixel of tracheae if finding
Pulmonary Vascular root subelement is disconnected, otherwise notifies preliminary sweep pulmonary parenchyma unit.Pulmonary Vascular root subelement is disconnected, corruption is used
The blood vessel at the edge of pulmonary parenchyma is lost to disconnect Pulmonary Vascular, and counts the number of the connected domain of vascularization, if the number of connected domain
It is five, then Pulmonary Vascular is notified to mark subelement, otherwise notifies preliminary sweep pulmonary parenchyma unit.Pulmonary Vascular marks subelement,
Mark different colors to the blood vessel of each connected domain, all mark finishes beam.
Pulmonary Vascular centerline extraction unit is divided into the resulting lung blood of independent communication domain unit according to the Pulmonary Vascular
Pipe, Pulmonary Vascular center path is extracted using elimination approach.
Specifically, Pulmonary Vascular centerline extraction unit includes following each unit:
Judge boundary point subelement, scan Pulmonary Vascular pixel, judge whether it is boundary point, if the pixel 6 is adjacent
Domain or 26 neighborhoods are then boundary point there are background area pixel, notify the first judgment sub-unit, otherwise mark and scanned, skip after
Continuous notice judges boundary point subelement.First judgment sub-unit judges Euler's characteristic of boundary point, if Euler's characteristic is constant
It notifies the second judgment sub-unit, otherwise skips and continue to notify to judge boundary point subelement.Second judgment sub-unit, judges boundary
Whether point is simple point, if the point on former three-dimensional space topological structure without influencing if be simple point, while by the point deletion, notice
Third judgment sub-unit, otherwise notice judges boundary point subelement.Also whether third judgment sub-unit judge blood vessel pixel
Other simple points, if nothing, centerline extraction terminates, and otherwise notice judges boundary point subelement.
Lobe of the lung cutting unit, by two sorter model of blood vessel Training Support Vector Machines of the adjacent lobe of the lung, left lung has two
A lobe of the lung, training one, right lung is there are three the lobe of the lung, and twice, three two classifiers of support vector machines are obtained in pulmo for training, leads to
It crosses support vector machines and finds interface between different lobe of the lung blood vessels, be partitioned into the lobe of the lung.
Specifically, lobe of the lung cutting unit includes following each unit:
Support vector machines Selection of kernel function unit, according to the interface between the lobe of the lung closest to the track of radial basis function,
Select Radial basis kernel function as the function mapped to higher dimensional space, it is determined that after kernel function, left lung discriminant function to be notified to seek
Subelement.Left lung discriminant function seeks subelement, for left lung, it is assumed that the pixel of left lung is N1, blood vessel pixel number
For n1, all blood vessel pixels label of the upper leaf of left lung is positive, while blood vessel pixels all on inferior lobe label is negative,
A discriminant function is obtained, notifies left lung upper and lower lobes segmentation subelement.Left lung upper and lower lobes divide subelement, are gone with discriminant function
Judge the symbol of each pixel in left pulmonary parenchyma, if symbol is positive, which belongs to leaf, under otherwise the point belongs to
Leaf, and the number of statistical pixel, if N1, then left lung segmentation is completed, and the notice superior lobe of right lung middle period merges subelement.Superior lobe of right lung
Middle period merges subelement, for right lung, it is assumed that the pixel of right lung is N2, and blood vessel pixel number is n2, to use two points
The upper leaf of right lung and middle period are combined into an entirety and regarded as " upper leaf ", notice right lung discriminant function by class support vector machines classification
Seek subelement.Right lung discriminant function seeks subelement, obtains a discriminant classification using the method as the segmentation of left lung
Function extracts inferior lobe first, all blood vessel pixels label of " the upper leaf " regarded as is positive, while all on inferior lobe
Blood vessel pixel label is negative, and obtains a discriminant function, and notice inferior lobe of right lung divides subelement.Inferior lobe of right lung segmentation is single
Member removes the symbol for judging each substantial pixel of right lung with discriminant function, if symbol is positive, which, which belongs to, is seen
" the upper leaf " done, otherwise the point belongs to inferior lobe, and the number of statistical pixel, and if N2, then inferior lobe segmentation is completed, and notifies on right lung
Leaf middle period discriminant function seeks subelement.Superior lobe of right lung middle period discriminant function seeks subelement, after inferior lobe is completed in segmentation, then
The blood vessel Training Support Vector Machines model for using leaf and middle period again, obtains a differentiation surface function, and the portion of inferior lobe is removed to right lung
Segmentation extracts leaf and middle period, and the notice superior lobe of right lung middle period divides subelement.The superior lobe of right lung middle period divides subelement,
For the upper leaf of right lung and middle period, it is assumed that the pixel of right lung is N3, and blood vessel pixel number is n3, the institute of the upper leaf of right lung
There is blood vessel pixel label to be positive, while the middle period of right lung upper all blood vessel pixel labels are negative, obtains a differentiation letter
Number, and with discriminant function remove judge the symbol of each pixel of right lung substantially, if symbol is positive, which belongs to
Leaf, otherwise the point belongs to middle period, and the number of statistical pixel, and if N3, then superior lobe of right lung and middle period segmentation are completed.
In the present embodiment, since the Density Distribution of pulmonary parenchyma is relatively uniform, the pulmonary parenchyma gray value on CT image
Distribution is also relatively uniform, can replace entire pulmonary parenchyma with the threshold value approximation of middle layer CT image when obtaining initial segmentation threshold value
Initial segmentation threshold value.The advantage of doing so is that being greatly reduced in the case where influencing very little to initial segmentation threshold accuracy
The calculation amount of the initial segmentation threshold value in entire pulmonary parenchyma region is calculated, to meet the requirement of practical application.Because of pulmonary parenchyma area
The gray value of the sum of the grayscale values background area (pulmonary parenchyma surrounding tissue) in domain has apparent difference, can show in image graph
One apparent trough.Therefore, the method that can use statistic histogram (see Fig. 3) finds the trough between two wave crests,
It is exactly the region excessive to surrounding tissue gray value from pulmonary parenchyma gray value, so that it is determined that initial segmentation threshold value T out0。
Fig. 4 is the pulmonary parenchyma area results schematic diagram without blood vessel, and as can be seen from Figure 4, there are holes to ask in pulmonary parenchyma region
Topic can be used morphology operations filling blood vessel method and carry out lung areas holes filling.Wherein, it carries out on the image morphologic
First expansion post-etching operation is known as closed operation, it is therefore an objective to the edge of the small hole of packing ratio structural element and smoothed image.Cause
This closing operation of mathematical morphology can fill Pulmonary Vascular.
Because tracheae makes pulmo be sticked together, the gas for making left and right adhesion of lung must be rejected first to separation pulmo
Pipe, it is bright there is no one between pulmonary parenchyma gray value and tracheae gray value although the gray value of tracheae is smaller than pulmonary parenchyma
Aobvious boundary.However tracheae has oneself unique distribution characteristics, the top of tracheae only has the centre for being distributed in two lungs,
Two are then divided into, respectively enters left lung and right lung, from the top of tracheae to the portion being divided into before two tracheaes enter pulmo
Point be not with pulmonary parenchyma adhesion, as shown in figure 5, intermediate independent fritter connected region is exactly tracheae.It is mentioned in the pulmonary parenchyma of Fig. 5
On the basis of taking result, closing operation of mathematical morphology is executed on all chest CT image bearing layers by choosing suitable structural element, finally
Obtaining includes pulmonary vascular lung tissue region, as a result as shown in Figure 8.
By counting the number of each connected domain pixel, the lesser connected domain region of pixel number is rejected, pixel is retained
It counts biggish two connected domain regions.After rejecting tracheae, for some data, pulmo has been two independent connection
Region, as shown in Figure 6.
Fig. 7 gives with numerical differentiation (DDA) near linear that pulmonary parenchyma adhesion region is drawn on single layer CT image
Boundary.The near linear boundary on every layer of pulmonary parenchyma region adhesion image and 26 neighbours around straight border are fallen in final etching
The mark value of domain point, so that it may obtain the pulmo region of two adhesions, mark point is finally directly counted in three-dimensional data
Connected domain, two disjunct connected domains can be obtained, and mark respectively.
The root of blood vessel and the root of tracheae are tightly adjacent, and all in the edge of lung, the diameter of pulmonary vascular root
Also bigger than the diameter of other parts blood vessel, therefore the blood vessel that can corrode the edge of lung tissue disconnects pulmonary vascular root, it is real
The existing different pulmonary vascular separation of the lobe of the lung, obtain five mutual disjunct connected domains, and marked respectively with different colors, such as Figure 11
It is shown.
Since blood vessel has too many pixel, if directly training will affect trained rate, therefore first have to reduce blood
The pixel number of pipe, but the distribution of blood vessel cannot be influenced.The extraction of center line just can satisfy the two conditions, because
Center path can remove the marginal portion of column blood vessel and will not influence the tendency of blood vessel, can using eliminate thinning method come
Extract the center path of blood vessel.For smooth Pulmonary Vascular region, loophole that may be present inside Pulmonary Vascular is filled, first to lung blood
Pipe does a closing operation of mathematical morphology, i.e., first expands post-etching, then extracts center path, final result with cancellation thinning method again
As shown in Figure 10.
The present invention is based on the pulmonary parenchyma extraction process of 3D region growing, the left and right adhesion of lung tracheae based on provincial characteristics reject
Process, Pulmonary Vascular root reject process, the extracted process in Pulmonary Vascular Center Road based on topological thinning and are based on support vector machines
The treatment processes such as the lobe of the lung partitioning algorithm of classification accurately extract the lobe of the lung from chest CT image, accurately complete to each lung
The qualitative assessment of the severity extent of leaf, more accurate to the diagnosing and treating of pulmonary disease and effectively, whether each single lung
Leaf, or entirely the average segmentation accuracy rate of five lobes of the lung of lung tissue meets expected segmentation and requires all 85% or more.
It is to be appreciated that describing the skill simply to illustrate that of the invention to what specific embodiments of the present invention carried out above
Art route and feature, its object is to allow those skilled in the art to can understand the content of the present invention and implement it accordingly, but
The present invention is not limited to above-mentioned particular implementations.All various changes made within the scope of the claims are repaired
Decorations, should be covered by the scope of protection of the present invention.
Claims (8)
1. a kind of method for extracting the lobe of the lung from chest CT image, which comprises the following steps:
S1, the n-layer chest CT image for receiving input obtain intermediate image layer, and the specified pixel point for choosing lung areas is seed
Point carries out the growth of the region 3D according to setting segmentation threshold and initial seed point, obtains without pulmonary vascular pulmonary parenchyma region,
In, n is natural number;
S2, pass through the left and right adhesion of lung elimination method on corrosion approximate adhesion boundary, rejecting in the resulting pulmonary parenchyma region step S1
Adhesion obtains two independent pulmonary parenchymas;
S3, it is filled into the resulting pulmonary parenchyma of step S2 using the operation handlebar blood vessel for first expanding post-etching, passes through side between maximum kind
Poor method automatically obtains optimum segmentation threshold value T0;
S4, with the resulting optimum segmentation threshold value T of step S30As the threshold value of segmentation blood vessel, the point for being greater than threshold value is calculated as blood vessel
Seed point carries out the growth of the region 3D, obtains Pulmonary Vascular;
S5, the blood vessel that the different lobes of the lung are disconnected from vascular root make the blood vessel in each lobe of the lung become an independent connected domain;
S6, according to the resulting Pulmonary Vascular of step S5, Pulmonary Vascular center path is extracted using elimination approach;
S7, two category support vector machines models of training, obtain two category support vector machines classifiers, are found not by the classifier
With the interface between lobe of the lung blood vessel, using this interface as the interface of the lobe of the lung and calculate obtain pulmonary parenchyma pixel classification
Discriminant function differentiates each pixel in pulmonary parenchyma, obtains lobe of the lung tissue;
The step S3 the following steps are included:
S31, the closed operation for executing first expansion post-etching on all chest CT image layers by choosing suitable structural element are grasped
Make, Pulmonary Vascular is filled into segmented good pulmonary parenchyma;
S32, one layer close to centre is selected in CT image, automatically obtain optimum segmentation threshold value using maximum variance between clusters
T0;
The step S6 the following steps are included:
S61, scanning Pulmonary Vascular pixel, judge whether it is boundary point, if there are background areas for 6 neighborhood of pixel or 26 neighborhoods
Pixel is then boundary point, executes step S62, otherwise marks scanned, skips and continue to execute step S61;
Otherwise S62, the Euler's characteristic for judging boundary point are skipped if Euler's characteristic is constant to then follow the steps S63 and continue to execute step
S61;
S63, judge whether boundary point is simple point, if the point on former three-dimensional space topological structure without influence if be simple point, together
When by the point deletion, execute step S64;It is no to then follow the steps S61;
S64, judging blood vessel pixel, whether there are also other simple points, if nothing, centerline extraction terminates, no to then follow the steps S61.
2. the method according to claim 1 for extracting the lobe of the lung from chest CT image, it is characterised in that: the step S1 packet
Include following steps:
S11, n (n is natural number) the layer chest CT image for receiving input obtain intermediate image layer, choose on intermediate image layer
The specified pixel point of lung areas is as initial seed point;
S12, segmentation threshold is arranged according to the characteristics of pulmonary parenchyma;
26 neighborhoods of the labeled initial seed point of S13, search, choose one of pixel;
Whether pixel selected by S14, judgement has been labeled, if then return step S13, no to then follow the steps S15;
Whether the gray value of pixel selected by S15, judgement is less than segmentation threshold, if so, the pixel is marked simultaneously
Mark queue is added, otherwise the stop flag pixel, executes step S16;
S16, judge whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish, if so then execute step S17, otherwise
Return step S13;
Whether S17, judge mark queue are empty, are not such as sky, then take out a mark point from the queue and be used as and initially marked
Remember seed point, return step S13, the pixel collection being otherwise labeled is exactly to be partitioned into without pulmonary vascular pulmonary parenchyma
Region.
3. the method according to claim 1 for extracting the lobe of the lung from chest CT image, it is characterised in that: the step S2 packet
Include following steps:
S21, to the pulmonary parenchyma region decision of acquisition, whether it is the approximately equal connected domain of two sizes, no if then terminating
Then follow the steps S22;
S22, the connected domain number for counting intermediate image layer reject the lesser connected domain region of pixel number, retain pixel number
Biggish connected domain region, and judge the number of connected domain and the size of each connected domain, if there are two connected domain and greatly
It is small approximately equal, terminate, it is no to then follow the steps S23;
S23, successively statistics connected domain number, if only one connected domain, pulmonary parenchyma one is scheduled on this layer of adhesion, executes step
Otherwise S24 records the number of each connected domain pixel, maximum two connected domains of pixel number are found, if area is most
Two big connected region pixel number numbers are suitable, then this layer of CT image or so pulmonary parenchyma adhesion, execute step S26, no
Then adhesion and biggish connected domain are pulmonary parenchyma, execute step S24;
S24, adhesion region is replaced with line segment approximation, is converted into the position of determining straight line, lower boundary is in the highest point in the direction y
The lower extreme point of near linear, the point nearest from lower extreme point is then the upper extreme point of near linear in all the points of coboundary, is searched out
Straight line is determined after this two o'clock, executes step S25;
S25, the near linear boundary on every layer of pulmonary parenchyma region adhesion image and 26 neighborhoods around straight border are eroded
The mark value of point obtains the pulmo region of two adhesions, executes step S26;
Whether S26, to judge all layers scanned, no to then follow the steps S23 if then terminating.
4. the method according to claim 1 for extracting the lobe of the lung from chest CT image, it is characterised in that: the step S4 packet
Include following steps:
S41, obtaining step S3 the pulmonary parenchyma with blood vessel in grey scale pixel value be greater than segmentation threshold T0All pixels point, and mark
It is denoted as blood vessel;
S42, a pixel is chosen in the resulting blood vessel of step S41 as initial seed point;
26 neighborhoods of the labeled initial seed point of S43, search, choose one of pixel;
Whether pixel selected by S44, judgement has been labeled, if then return step S43, no to then follow the steps S45;
Whether the gray value of pixel selected by S45, judgement is greater than segmentation threshold, if so, the pixel is marked simultaneously
Mark queue is added, otherwise the stop flag pixel, executes step S46;
S46, judge whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish, if so then execute step S47, otherwise
Return step S43;
Whether S47, judge mark queue are empty, are not such as sky, then take out a mark point from the queue and be used as and initially marked
Remember seed point, return step S43, the pixel collection being otherwise labeled is exactly the Pulmonary Vascular being partitioned into.
5. the method according to claim 1 for extracting the lobe of the lung from chest CT image, it is characterised in that: the step S5 packet
Include following steps:
S51, lung edge and the Pulmonary Vascular pixel close to tracheae are found, thens follow the steps S52 if finding, otherwise continues to scan on lung
Portion's tissue;
S52, Pulmonary Vascular is disconnected using the blood vessel at the edge of corrosion pulmonary parenchyma, and counts the number of the connected domain of vascularization,
If the number of connected domain is five, S53 is thened follow the steps, it is no to then follow the steps S51;
S53, the color different to each connected component labeling, all mark finish beam.
6. the method according to claim 1 for extracting the lobe of the lung from chest CT image, it is characterised in that: the step S7 packet
Include following steps:
S71, according to the interface between the lobe of the lung closest to the track of radial basis function, select Radial basis kernel function as to higher-dimension sky
Between the function that maps, it is determined that after kernel function, execute step S72;
S72, for left lung, it is assumed that the pixel of left lung be N1, blood vessel pixel number be n1, all blood of the upper leaf of left lung
Pipe pixel label is positive, while blood vessel pixels all on inferior lobe label is negative, and obtains a discriminant function, executes step
S73;
S73, it is gone to judge the symbol of each pixel in left pulmonary parenchyma with discriminant function, if symbol is positive, which belongs to
Upper leaf, otherwise the point belongs to inferior lobe, and the number of statistical pixel, and if N1, then left lung segmentation is completed, and executes step S74, otherwise
Continue to execute step S73;
S74, for right lung, it is assumed that the pixel of right lung be N2, blood vessel pixel number be n2, to use two class Support Vectors
The upper leaf of right lung and middle period are combined into an entirety and regarded as " upper leaf ", execution step S75 by machine classification;
S75, a discriminant classification function is obtained using the method as the segmentation of left lung, inferior lobe is extracted first, step S74
All blood vessel pixels label of " the upper leaf " regarded as is positive, while blood vessel pixels all on inferior lobe label is negative, and obtains
To a discriminant function, step S76 is executed;
S76, it is gone to judge the symbol of each substantial pixel of right lung with discriminant function, if symbol is positive, which belongs to
" the upper leaf " that step S74 is regarded as, otherwise the point belongs to inferior lobe, and the number of statistical pixel, and if N2, then inferior lobe segmentation is completed,
Step S77 is executed, step S76 is otherwise continued to execute;
After inferior lobe is completed in S77, segmentation, the blood vessel Training Support Vector Machines model of leaf and middle period is then used again, one is obtained and sentences
Other surface function removes the partial segmentation of inferior lobe to right lung, extracts leaf and middle period, executes step S78;
S78, for the upper leaf of right lung and middle period, it is assumed that the pixel of right lung be N3, blood vessel pixel number be n3, right lung
All blood vessel pixels label of upper leaf is positive, while the middle period of right lung upper all blood vessel pixel labels are negative, and obtains one
A discriminant function executes step S79;
S79, it is gone to judge the symbol of each substantial pixel of right lung with discriminant function, if symbol is positive, which belongs to
Upper leaf, otherwise the point belongs to middle period, and the number of statistical pixel, and if N3, then superior lobe of right lung and middle period segmentation are completed, so far left
The right lung segmentation lobe of the lung terminates.
7. a kind of device for extracting the lobe of the lung from chest CT image, it is characterised in that: include:
CT image input units, n-layer chest CT image for receiving input, wherein n is natural number;
Without pulmonary vascular pulmonary parenchyma acquiring unit, the specified pixel point for choosing lung areas is seed point, according to setting point
It cuts threshold value and initial seed point carries out the growth of the region 3D, obtain without pulmonary vascular pulmonary parenchyma region;
Two independent pulmonary parenchyma acquiring units, after obtaining left and right pulmonary parenchyma, judge left and right pulmonary parenchyma whether adhesion, if viscous
Even, then two independent pulmonary parenchymas are obtained by rejecting the adhesion boundary of tracheae and near linear;
Pulmonary Vascular segmentation threshold acquiring unit is filled into pulmonary parenchyma using the operation handlebar blood vessel for first expanding post-etching, is passed through
Maximum variance between clusters automatically obtain optimum segmentation threshold value T0, comprising: closed operation operating unit, by choosing suitable structure
Element executes the closed operation operation for first expanding post-etching on all chest CT image layers, and it is segmented good that Pulmonary Vascular is filled into
Pulmonary parenchyma in;Optimum segmentation threshold value acquiring unit selects one layer close to centre, using between maximum kind in CT image
Variance method automatically obtains optimum segmentation threshold value T0;
Pulmonary Vascular extraction unit, with resulting optimum segmentation threshold value T0As the threshold value of segmentation blood vessel, the point for being greater than threshold value is calculated
As vessel seed point, the growth of the region 3D is carried out, obtains Pulmonary Vascular;
Pulmonary Vascular is divided into independent communication domain unit, and the blood vessel of the different lobes of the lung is disconnected from vascular root, makes each lobe of the lung
Blood vessel becomes an independent connected domain;
Pulmonary Vascular centerline extraction unit is divided into the resulting Pulmonary Vascular of independent communication domain unit according to the Pulmonary Vascular,
Pulmonary Vascular center path is extracted using elimination approach, comprising: judge boundary point subelement, scan Pulmonary Vascular pixel, judgement
Whether it is boundary point, is boundary point if 6 neighborhood of pixel or 26 neighborhoods are there are background area pixel, notifies that the first judgement is sub
Otherwise unit is marked and has been scanned, skip and continue to notify to judge boundary point subelement;First judgment sub-unit, judges boundary
Euler's characteristic of point, notifies the second judgment sub-unit if Euler's characteristic is constant, otherwise skips and continues to notify to judge boundary idea
Unit;Second judgment sub-unit judges whether boundary point is simple point, if the point is on former three-dimensional space topological structure without influence
Then it is simple point, while by the point deletion, notifies third judgment sub-unit, otherwise notice judges boundary point subelement;Third is sentenced
Disconnected subelement, judging blood vessel pixel, whether there are also other simple points, if nothing, centerline extraction terminates, otherwise notice judgement
Boundary point subelement;
Lobe of the lung cutting unit, by two sorter model of blood vessel Training Support Vector Machines of the adjacent lobe of the lung, there are two lungs for left lung
Leaf, training one, right lung is there are three the lobe of the lung, and twice, three two classifiers of support vector machines are obtained in pulmo, pass through branch for training
It holds vector machine and finds interface between different lobe of the lung blood vessels, be partitioned into the lobe of the lung.
8. the device according to claim 7 for extracting the lobe of the lung from chest CT image, it is characterised in that:
It is described to include: without pulmonary vascular pulmonary parenchyma acquiring unit
Initial seed point selection unit chooses the specified pixel point of lung areas as initial seed on intermediate tomographic image
Point;
Segmentation threshold obtains subelement, and segmentation threshold is arranged according to the characteristics of pulmonary parenchyma;
Pixel chooses subelement, and 26 neighborhoods of the labeled initial seed point of search choose one of pixel;
Whether pixel marker for judgment subelement, the selected pixel of judgement have been labeled, if so, notice pixel
Subelement is chosen, otherwise notifies pixel value judgment sub-unit;
Whether the gray value of pixel value judgment sub-unit, the selected pixel of judgement is less than segmentation threshold, if so, this
Pixel marks and mark queue is added, otherwise the stop flag pixel, 26 neighborhood territory pixel point search of notice seed point
Unit;
26 neighborhood search subelement of seed point, judges whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish,
If so, notification indicia queue judgment sub-unit, otherwise notifies the pixel to choose subelement;
Whether mark queue judgment sub-unit, judge mark queue are sky, if being not sky, then a mark are taken out from the queue
Note point chooses subelement as initially labeled seed point, notice pixel, and the pixel collection being otherwise labeled is exactly
Be partitioned into without pulmonary vascular pulmonary parenchyma;
Described two independent pulmonary parenchyma acquiring units include:
Pulmonary parenchyma connected domain number judgment sub-unit judges whether it is that two sizes are approximately equal to the pulmonary parenchyma of acquisition
Connected domain, if then terminating, otherwise notice rejects tracheae subelement;
Tracheae subelement is rejected, the connected domain number of intermediate image layer is counted, rejects the lesser connected domain region of pixel number,
Retain the biggish connected domain region of pixel number, and judges the number of connected domain and the size of each connected domain, if there are two
Connected domain and size is approximately equal, terminates, and otherwise notifies layer connected domain number judgment sub-unit;
Layer connected domain number judgment sub-unit, successively counts connected domain number, if only one connected domain, pulmonary parenchyma one
It is scheduled on this layer of adhesion, notice straight line determines subelement, otherwise records the number of each connected domain pixel, find pixel number
Maximum two connected domains, if the maximum two connected regions pixel number number of area is suitable, this layer of CT image or so
Pulmonary parenchyma adhesion, notice pixel layer scan judgment sub-unit, and otherwise adhesion and biggish connected domain are pulmonary parenchyma, notify straight line
Determine subelement;
Straight line determines subelement, replaces adhesion region with line segment approximation, is converted into the position of determining straight line, lower boundary is in the side y
To highest point be near linear lower extreme point, the point nearest from lower extreme point is then near linear in all the points of coboundary
Upper extreme point determines straight line, notice corrosion near linear boundary subelement after searching out this two o'clock;
Corrode near linear boundary subelement, erodes the near linear boundary on every layer of pulmonary parenchyma region adhesion image, with
And the mark value of 26 neighborhood points around straight border, the pulmo region of two adhesions is obtained, notice pixel layer scanning is sentenced
Disconnected subelement;
Pixel layer scans judgment sub-unit, whether scanned judges all layers, if then terminating, otherwise notifies layer connected domain
Number judgment sub-unit;
The Pulmonary Vascular extraction unit includes:
Middle layer Pulmonary Vascular extracts subelement, and grey scale pixel value is greater than segmentation threshold in the pulmonary parenchyma with blood vessel in acquisition middle layer
Value T0All pixels point, and be labeled as blood vessel;
Initial seed point obtains subelement, extracts in the middle layer Pulmonary Vascular and chooses a picture in the resulting blood vessel of subelement
Vegetarian refreshments is as initial seed point;
Pixel obtains subelement, and 26 neighborhoods of the labeled initial seed point of search choose one of pixel;
Whether pixel marker for judgment subelement, the selected pixel of judgement have been labeled, if notice pixel
Subelement is obtained, otherwise notifies pixel judgment sub-unit;
Whether the gray value of pixel value judgment sub-unit, the selected pixel of judgement is greater than segmentation threshold, if so, this
Pixel marks and mark queue is added, and notifies 26 neighborhood search subelement of seed point, otherwise the stop flag pixel;
26 neighborhood search subelement of seed point, judges whether 26 neighborhood territory pixel points of seed point are all searched for and judge to finish,
If then notification indicia queue judgment sub-unit, otherwise pixel is notified to obtain subelement;
Whether mark queue judgment sub-unit, judge mark queue are sky, if being not sky, then a mark are taken out from the queue
Note point obtains subelement as initially labeled seed point, notice pixel, and the pixel collection being otherwise labeled is exactly
The Pulmonary Vascular being partitioned into;
The Pulmonary Vascular is divided into independent communication domain unit
Preliminary sweep pulmonary parenchyma unit finds lung edge and notifies to disconnect if finding close to the Pulmonary Vascular pixel of tracheae
Otherwise Pulmonary Vascular root subelement notifies preliminary sweep pulmonary parenchyma unit;
Pulmonary Vascular root subelement is disconnected, uses the blood vessel at the edge of corrosion pulmonary parenchyma to disconnect Pulmonary Vascular, and count blood vessel
The number of the connected domain of formation notifies Pulmonary Vascular to mark subelement if the number of connected domain is five, and otherwise notice is initially swept
Retouch pulmonary parenchyma unit;
Pulmonary Vascular marks subelement, marks different colors to the blood vessel of each connected domain, and all mark finishes beam;
The lobe of the lung cutting unit includes:
Support vector machines Selection of kernel function unit is selected according to the interface between the lobe of the lung closest to the track of radial basis function
Radial basis kernel function is as the function mapped to higher dimensional space, it is determined that after kernel function, it is sub single to notify that left lung discriminant function is sought
Member;
Left lung discriminant function seeks subelement, for left lung, it is assumed that the pixel of left lung is N1, and blood vessel pixel number is
N1 is positive all blood vessel pixels label of the upper leaf of left lung, while blood vessel pixels all on inferior lobe label is negative, and obtains
To a discriminant function, left lung upper and lower lobes segmentation subelement is notified;
Left lung upper and lower lobes divide subelement, and the symbol for judging each pixel in left pulmonary parenchyma is removed with discriminant function, if
Symbol is positive, then the point belongs to leaf, and otherwise the point belongs to inferior lobe, and the number of statistical pixel, and if N1, then left lung is segmented
At the notice superior lobe of right lung middle period merges subelement;
The superior lobe of right lung middle period merges subelement, for right lung, it is assumed that the pixel of right lung is N2, and blood vessel pixel number is
The upper leaf of right lung and middle period are combined into an entirety and regarded as " upper leaf ", notice by n2 to use two category support vector machines to classify
Right lung discriminant function seeks subelement;
Right lung discriminant function seeks subelement, obtains a discriminant classification function using the method as the segmentation of left lung, first
Inferior lobe is first extracted, all blood vessel pixels label of " the upper leaf " regarded as is positive, while blood vessel pictures all on inferior lobe
Vegetarian refreshments label is negative, and obtains a discriminant function, and notice inferior lobe of right lung divides subelement;
Inferior lobe of right lung divides subelement, and the symbol for judging each substantial pixel of right lung is removed with discriminant function, if symbol
Number be positive, then the point belongs to " the upper leaf " regarded as, and otherwise the point belongs to inferior lobe, and the number of statistical pixel, if N2 then under
Leaf segmentation is completed, and notice superior lobe of right lung middle period discriminant function seeks subelement;
Superior lobe of right lung middle period discriminant function seeks subelement, after inferior lobe is completed in segmentation, then uses the blood vessel of leaf and middle period again
Training Support Vector Machines model obtains a differentiation surface function, and the partial segmentation of inferior lobe is removed to right lung, extracts Ye Hezhong
Leaf, notice superior lobe of right lung middle period divide subelement;
The superior lobe of right lung middle period divides subelement, for the upper leaf of right lung and middle period, it is assumed that the pixel of right lung is N3, blood vessel picture
Vegetarian refreshments number is n3, and all blood vessel pixels label of the upper leaf of right lung is positive, while the middle period of right lung upper all blood vessels
Pixel label is negative, and obtains a discriminant function, and is gone to judge each substantial pixel of right lung with discriminant function
Symbol, if symbol is positive, which belongs to leaf, and otherwise the point belongs to middle period, and the number of statistical pixel, if N3, then right
Leaf and middle period segmentation are completed on lung.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710374181.8A CN107230204B (en) | 2017-05-24 | 2017-05-24 | A kind of method and device for extracting the lobe of the lung from chest CT image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710374181.8A CN107230204B (en) | 2017-05-24 | 2017-05-24 | A kind of method and device for extracting the lobe of the lung from chest CT image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107230204A CN107230204A (en) | 2017-10-03 |
CN107230204B true CN107230204B (en) | 2019-11-22 |
Family
ID=59934015
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710374181.8A Active CN107230204B (en) | 2017-05-24 | 2017-05-24 | A kind of method and device for extracting the lobe of the lung from chest CT image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107230204B (en) |
Families Citing this family (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107909581B (en) * | 2017-11-03 | 2019-01-29 | 杭州依图医疗技术有限公司 | Lobe of the lung section dividing method, device, system, storage medium and the equipment of CT images |
CN108171703B (en) * | 2018-01-18 | 2020-09-15 | 东北大学 | Method for automatically extracting trachea tree from chest CT image |
CN108961273B (en) * | 2018-07-03 | 2021-10-29 | 东北大学 | Method and system for segmenting pulmonary artery and pulmonary vein from CT image |
CN109191436A (en) * | 2018-08-15 | 2019-01-11 | 复旦大学 | The low-dose CT Lung neoplasm detection algorithm of view-based access control model conspicuousness spectrum residual error method |
CN109410166B (en) * | 2018-08-30 | 2022-09-09 | 中国科学院苏州生物医学工程技术研究所 | Full-automatic segmentation method for lung parenchyma CT image |
CN110956604B (en) * | 2018-09-27 | 2022-12-16 | 上海蔚来汽车有限公司 | Method and system for counting effective bonding area of bonding glue |
CN109360219A (en) * | 2018-10-23 | 2019-02-19 | 东北大学 | A kind of augmented reality auxiliary operation method and system |
CN110060262A (en) * | 2019-04-18 | 2019-07-26 | 北京市商汤科技开发有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
CN110246144A (en) * | 2019-06-17 | 2019-09-17 | 吉林大学第一医院 | A kind of the blood vessel Enhancement Method and system of lung CT image |
CN110246145B (en) * | 2019-06-21 | 2023-02-21 | 福州大学 | Segmentation method of abdominal CT image |
CN110473207B (en) * | 2019-07-30 | 2022-05-10 | 赛诺威盛科技(北京)股份有限公司 | Method for interactively segmenting lung lobes |
CN112150477B (en) * | 2019-11-15 | 2021-09-28 | 复旦大学 | Full-automatic segmentation method and device for cerebral image artery |
CN110942453A (en) * | 2019-11-21 | 2020-03-31 | 山东众阳健康科技集团有限公司 | CT image lung lobe identification method based on neural network |
CN111179298B (en) * | 2019-12-12 | 2023-05-02 | 深圳市旭东数字医学影像技术有限公司 | Three-dimensional lung automatic segmentation and left and right lung separation method and system based on CT image |
CN111275722A (en) * | 2020-02-18 | 2020-06-12 | 广州柏视医疗科技有限公司 | Lung segment and liver segment segmentation method and system |
CN111563902B (en) * | 2020-04-23 | 2022-05-24 | 华南理工大学 | Lung lobe segmentation method and system based on three-dimensional convolutional neural network |
CN112884792B (en) * | 2021-02-02 | 2022-10-25 | 青岛海信医疗设备股份有限公司 | Lung image segmentation method and device, electronic equipment and storage medium |
CN112950554B (en) * | 2021-02-05 | 2021-12-21 | 慧影医疗科技(北京)有限公司 | Lung lobe segmentation optimization method and system based on lung segmentation |
CN113012114B (en) * | 2021-03-02 | 2021-12-03 | 推想医疗科技股份有限公司 | Blood vessel identification method and device, storage medium and electronic equipment |
CN113129317B (en) * | 2021-04-23 | 2022-04-08 | 广东省人民医院 | Lung lobe automatic segmentation method based on watershed analysis technology |
CN113160186B (en) * | 2021-04-27 | 2022-10-25 | 青岛海信医疗设备股份有限公司 | Lung lobe segmentation method and related device |
CN113223013B (en) * | 2021-05-08 | 2022-02-22 | 推想医疗科技股份有限公司 | Method, device, equipment and storage medium for pulmonary vessel segmentation positioning |
CN115049807B (en) * | 2022-06-28 | 2023-05-09 | 上海市胸科医院 | Method, device and server for establishing pulmonary vessel model |
CN116228767B (en) * | 2023-05-09 | 2023-06-27 | 北京易优联科技有限公司 | X-ray lung tumor image processing method based on computer vision |
CN116402816B (en) * | 2023-06-08 | 2023-08-15 | 中国人民解放军海军青岛特勤疗养中心 | Physical examination CT image data management method and system |
CN117830302A (en) * | 2024-03-04 | 2024-04-05 | 瀚依科技(杭州)有限公司 | Optimization method and device for lung segment segmentation, electronic equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1447772A1 (en) * | 2003-02-11 | 2004-08-18 | MeVis GmbH | A method of lung lobe segmentation and computer system |
JP2008142481A (en) * | 2006-12-13 | 2008-06-26 | Med Solution Kk | Apparatus and program for carrying out segmentation of lungs to units of pulmonary segment automatically |
CN102208105A (en) * | 2010-03-31 | 2011-10-05 | 富士胶片株式会社 | Medical image processing technology |
CN103310457A (en) * | 2013-06-18 | 2013-09-18 | 哈尔滨工程大学 | Lung parenchyma segmentation method based on convex hulls correcting by parabolic |
CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7711167B2 (en) * | 2005-12-07 | 2010-05-04 | Siemens Medical Solutions Usa, Inc. | Fissure detection methods for lung lobe segmentation |
-
2017
- 2017-05-24 CN CN201710374181.8A patent/CN107230204B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1447772A1 (en) * | 2003-02-11 | 2004-08-18 | MeVis GmbH | A method of lung lobe segmentation and computer system |
JP2008142481A (en) * | 2006-12-13 | 2008-06-26 | Med Solution Kk | Apparatus and program for carrying out segmentation of lungs to units of pulmonary segment automatically |
CN102208105A (en) * | 2010-03-31 | 2011-10-05 | 富士胶片株式会社 | Medical image processing technology |
CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
CN103310457A (en) * | 2013-06-18 | 2013-09-18 | 哈尔滨工程大学 | Lung parenchyma segmentation method based on convex hulls correcting by parabolic |
Non-Patent Citations (5)
Title |
---|
3D lung image retrieval usinglocalized features;Adrien Depeursinge 等;《Medical Imaging 2011》;20110309;第7963卷;79632E-1至79632E-14 * |
Lung Segmentation and Tumor Identification from CT Scan Images Using SVM;Ganesh Sable 等;《International Journal of Science and Research》;20141031;第3卷(第10期);64-69 * |
Segmentation of Lungs from CT Scan Images for Early Diagnosis of Lung Cancer;Nisar Ahmed Memon 等;《PROCEEDINGS OF WORLD ACADEMY OF SCIENCE, ENGINEERING AND TECHNOLOGY》;20060831;第14卷;228-233 * |
基于区域生长法的三维心血管模型的分割;谭启路 等;《北京生物医学工程》;20161031;第35卷(第5期);469-475 * |
基于漫水填充算法的肺实质分割方法;冯宗雪,董晓;《电脑知识与技术》;20160630;第12卷(第18期);205-207、210 * |
Also Published As
Publication number | Publication date |
---|---|
CN107230204A (en) | 2017-10-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107230204B (en) | A kind of method and device for extracting the lobe of the lung from chest CT image | |
CN109102502B (en) | Pulmonary nodule detection method based on three-dimensional convolutional neural network | |
CN110276745B (en) | Pathological image detection algorithm based on generation countermeasure network | |
CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
CN109727235B (en) | Organ automatic delineation algorithm based on deep learning | |
CN107341812B (en) | A kind of sequence Lung neoplasm image partition method based on super-pixel and Density Clustering | |
CN102663416B (en) | Segmentation method of viscera and internal blood vessels thereof in surgical planning system | |
JP2007537812A (en) | How to detect nodule boundaries | |
Maitra et al. | Automated digital mammogram segmentation for detection of abnormal masses using binary homogeneity enhancement algorithm | |
CN109300113A (en) | A kind of Lung neoplasm assisted detection system and method based on improvement Convex Hull Method | |
CN108537751A (en) | A kind of Thyroid ultrasound image automatic segmentation method based on radial base neural net | |
Maitra et al. | Detection of abnormal masses using divide and conquer algorithmin digital mammogram | |
Maitra et al. | Accurate breast contour detection algorithms in digital mammogram | |
Lee et al. | An adaptive resource-allocating network for automated detection, segmentation, and classification of breast cancer nuclei topic area: image processing and recognition | |
CN106295639A (en) | A kind of virtual reality terminal and the extracting method of target image and device | |
CN108765409A (en) | A kind of screening technique of the candidate nodule based on CT images | |
CN113077441A (en) | Coronary artery calcified plaque segmentation method and method for calculating coronary artery calcified score | |
CN113096184A (en) | Diatom positioning and identifying method under complex background | |
CN115471701A (en) | Lung adenocarcinoma histology subtype classification method based on deep learning and transfer learning | |
CN110459303B (en) | Medical image abnormity detection device based on depth migration | |
CN111383759A (en) | Automatic pneumonia diagnosis system | |
CN110517766A (en) | Identify the method and device of encephalatrophy | |
CN112348826A (en) | Interactive liver segmentation method based on geodesic distance and V-net | |
CN116433649A (en) | Micro-bubble counting method for patent foramen ovale and intelligent diagnosis device | |
CN105956587A (en) | Method for automatically extracting meniscus from knee-joint magnetic resonance image sequence based on shape constraint |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |