CN108171703A - A kind of method that tracheae tree is automatically extracted from chest CT image - Google Patents

Abstract

The invention belongs to the technical field of image processing based on medical image more particularly to a kind of methods that tracheae tree is automatically extracted from chest CT image.Main tracheae and main bronchus are obtained from chest CT image；According to 3D Region growing segmentations mode and main tracheae and main bronchus information have been obtained, has established adaptive threshold 3D Region growing segmentations model and adaptive threshold leak model；Using adaptive threshold 3D Region growing segmentations model and adaptive threshold leak model, the second class tracheae branch of chest CT image is extracted；According to the average information of the second class tracheae branch of extraction, the parameter of adaptive threshold 3D region growings model and adaptive threshold leakage model is adjusted, then extracts the third class tracheae branch of the chest CT image；Based on tracheae tree topology extraction tip tracheae branch has been obtained, the tracheae tree of the chest CT image is obtained.Method provided by the invention improves the tracheae segmentation precision that tracheae tree is extracted from CT images, while reduces extraction time.

Description

A kind of method that tracheae tree is automatically extracted from chest CT image

Technical field

The invention belongs to the technical field of image processing based on medical image more particularly to one kind from chest CT image from The method of dynamic extraction tracheae tree.

Background technology

Accurate lung qi pipe tree construction is obtained from CT images to be of great significance in medical field and computer application circle.Face Bed doctor can be to the common respiratory disease such as chronic obstructive pulmonary disease, bronchiectasis by tracheae parameter and rating information Pathological analysis and follow-up study is unfolded；Patient can also look into without intrusive virtual bronchoscopy with this；In addition, gas The one-to-one correspondence of the substates lung structure such as the Guan Shuyu lobes of the lung, lung section anatomically is also the segmentation of dependency structure image and analysis Important foundation.Therefore, the lung qi pipe segmentation based on CT images is always the hot spot of researcher's concern.

In chest CT image, lung qi pipe totally in tree, is split into left and right main bronchus, then by main tracheae Leaf, section, sub- segmental bronchus are split into, then is split into rudimentary bronchus and tip bronchus, is stretched into pulmonary parenchyma.Tracheae branch is certainly Substantially in club shaped structure after main bronchus division, tracheae chamber is in low gray value region, by the tracheal wall of higher gray value surround with It is isolated with pulmonary parenchyma.

Due to topological features special possessed by lung qi Guan Shu and gray scale textural characteristics, with tracheae tree tissue by Grade division, tube chamber attenuates, tube wall is thinning, and on CT images, the gray value of tracheal wall continuously decreases, conventionally used for pulmonary parenchyma, liver Some dividing methods of portion, brain are not ideally suited for the segmentation of lung qi pipe.Tracheorrhaphy based on conventional threshold values growth It cuts and easily crosses tracheal wall and diffuse into pulmonary parenchyma, form large-scale over-segmentation, i.e., " leak " phenomenon, it is also difficult to effectively identification Subtle or lesion branch, misses a large amount of detailed information, this is difficult to obtain accurate lung qi Guan Shu.

In current published tracheae tree processing method, patent CN201210423958.2 is increased using multiple dimensioned gray reconstruction Strong lumen area；On this basis, patent CN201510009239.X utilize multiple dimensioned tubular structure feature extraction tracheae tree, two Kind method is both needed to the consumption plenty of time；Patent CN201110405950.9 extracts the following bronchus of section grade, institute using fixed threshold Design leakage processing model is also only for Duan Ji branches, it is difficult to deeper bronchus and subtle tracheae is obtained, more without needle Processing method is specifically leaked to these branch designs；Patent CN201510224781.7 is rebuild subtle and last using energy function Tip tracheae, but there is still a need for fixed thresholds to judge tracheae degree of membership and leakage.The above method need stringent artificial parameter setting or It needs to carry out CT images complicated pretreatment enhancing operation or needs to extract the auxiliary tracheae segmentation such as tubulose structure feature, greatly Processing time is improved greatly, also constrains tracheae segmentation precision, it is difficult to the segmentation task of CT images under a variety of image-forming conditions is completed, Seriously constrain reliability and applicability in clinical practice.

Invention content

(1) technical problems to be solved

In order to improve the tracheae segmentation precision that tracheae tree is extracted from CT images, while the extraction time of tracheae tree is reduced, The present invention provides a kind of method that tracheae tree is automatically extracted from chest CT image.

(2) technical solution

To achieve the above object, the main technical schemes that the present invention uses include：

A kind of method that tracheae tree is automatically extracted from chest CT image, including：

101st, for chest CT image, the main tracheae of tracheae tree and main branch gas are obtained based on 3D Region growing segmentation modes Pipe；

102nd, based on described in the 3D Region growing segmentations mode, the average information of the main tracheae obtained and acquisition The average information of main bronchus is established to extract the adaptive threshold 3D region growings model of the second class tracheae branch and adaptive Answer sub-threshold leakage model；

103rd, according to the adaptive threshold 3D region growings model and the adaptive threshold leak model, described in extraction Second class tracheae branch of chest CT image；

104th, according to the average information of the second class tracheae branch of extraction, the adaptive threshold 3D regions life is adjusted The parameter of long model and adaptive threshold leakage model；

105th, it according to the adaptive threshold 3D region growings model of adjustment and the adaptive threshold leak model, carries Take the third class tracheae branch of the chest CT image；

106th, tracheae tree is formed according to the main tracheae, main bronchus, the second class tracheae branch and third class tracheae branch Topological structure, extract the tip tracheae branch of the tracheae tree, obtain the tracheae tree of the chest CT image.

Further, described 101 include：

1011st, the chest CT image is pre-processed；

1012nd, image binaryzation processing is carried out to the pretreated chest CT image, obtains the chest after binaryzation The foreground area of portion's CT images and background area；

1013rd, centered on the center of chest CT image described after binaryzation, the corresponding area of the foreground area is selected A maximum prospect connected domain；

1014th, judge whether the prospect connected domain of selection meets the first preset condition, if satisfied, then by the prospect Connected domain divides initial connected domain as main tracheae, and the prospect connected domain central point is as initial seed point s₀；

1015th, the initial seed point s is handled based on the 3D Region growing segmentations mode₀, the main gas of acquisition tracheae tree Pipe.

Further, described 1015 include：

S01, by the initial seed point s₀Segmentation queue is added in, based on the 3D Region growing segmentations mode, to chest bottom Direction carries out 3D Region growing segmentations；

S02, in the ith iteration of the 3D Region growing segmentations mode, take out it is described segmentation queue in the first kind Sub- pointThe pixel average gray value of the prospect connected domain is not more than default first gray threshold T_traAnd unlabelled institute It states26 neighborhood territory pixel point w₁,w₂,...,w_mIt marks as main tracheae；

S03 and using the 26 neighborhood territory pixel point as new seed pointAdd in the segmentation queue；After It is described to divide removing for queue in continuous processing ith iterationExcept remaining seed pointUntil the ith changes Whole seed points in generation are all handled through the 3D Region growing segmentations mode；

S04, iterations add 1 and repeat the iterative process of ith, and the main tracheae label described in the CT images is most lower When side is split into two connected domains, the main tracheae of the tracheae tree is obtained, wherein, the main tracheae is upper close to the one end in oral cavity Side, one end far from the oral cavity are downside, and the main tracheae lower side is the distalmost end of the main tracheae far from the oral cavity.

Further, it described 101 further includes；

The main bronchus includes left principal bronchus and right principal bronchus, is obtained based on the 3D Region growing segmentations mode The step of left principal bronchus, includes：

M01, using the central point of the connected domain on the left of the main tracheae anatomy as initial kind of left principal bronchus Sub- point S_left；

M02, the pixel average gray value μ according to the main tracheae region of acquisition_iniAnd standard deviation sigma_iniDetermine institute State the second gray threshold T of left principal bronchus segmentation_bro, T_bro=μ_ini+2σ_ini；

M03, by initial seed point S_leftSegmentation queue is added in, based on the 3D Region growing segmentations mode, and with reference to institute State whole seed points in the processing segmentation queue of the second gray threshold, the left principal bronchus label division described in the CT images During for two connected domains, the left principal bronchus segmentation terminates, and obtains the left principal bronchus；

And the step of based on the 3D Region growing segmentations mode obtaining the right principal bronchus, includes：

N01, using the central point of the connected domain on the right side of the main tracheae anatomy as initial kind of right principal bronchus Sub- point S_right；

N02, the pixel average gray value μ according to the right main tracheae region of acquisition_iniAnd standard deviation sigma_iniIt determines Second gray threshold T of main bronchus segmentation_bro, T_bro=μ_ini+2σ_ini；

N03, by initial seed point S_rightSegmentation queue is added in, based on the 3D Region growing segmentations mode, and with reference to institute State whole seed points in the processing segmentation queue of the second gray threshold, the right principal bronchus label division described in the CT images During for two connected domains, the right principal bronchus segmentation terminates, and obtains the right principal bronchus.

Further, described 102 include：Establish the adaptive threshold 3D region growing models：

It specifically includes：

1021：Obtain initial segmentation seed point set；

1022：Obtain gray threshold；

1023：Tracheae degree of membership rule is determined according to the gray threshold, including：

(1) first is subordinate to rule：The gray value I of seed point w_wNo more than third gray threshold T；

(2) second are subordinate to rule：The gray value I of the seed point w_wNo more than the 4th gray threshold, the 4th gray scale Threshold value is the gray value T for the seed point v for generating the seed point w_v；

(3) third is subordinate to rule：The pixel that tracheae is marked as in the 6 neighborhood territory pixel points of the seed point w is no less than 4 It is a, i.e.,：Wherein L represents dividing mark,Represent the 6 of the seed point w The set of neighborhood territory pixel point；

(4) the 4th are subordinate to rule：The 6 neighborhood territory pixel point gray values of the seed point w and seed point w and average value it is little In the third gray threshold T, and in the 6 neighborhood territory pixel points of the seed point w, gray value is not more than the third gray threshold The pixel of T is no less than 4, i.e.,：

1024：Obtain growing strategy：In ith iteration, segmentation first seed point of queue is taken outBy described inSymbol It is regular and unlabelled described to close the tracheae degree of membership26 neighborhood territory pixel point w₁,w₂,...,w_mLabeled as tracheae, and will The 26 neighborhood territory pixel point w₁,w₂,...,w_mAs new seed pointAdd in the segmentation queue；It continues with Remaining seed point in the segmentation queueUntilIt is all finished by growth, into i+1 time repeatedly Generation；This step is repeated until meeting termination rules.

Further, the initial segmentation seed point set includes：The left principal bronchus is divided in iteration ends Whole seed points and the right principal bronchus in queue divide whole seed points in queue in iteration ends.

Further, it described 102 further includes：Establish the adaptive threshold leak model；

It specifically includes：

1025：Leakage detection method：Record during 3D Region growing segmentations in described 1024 what iteration every time was divided New seed points C₁,C₂,...,C_NIf not leaked during iterative segmentation, with the adaptive leak threshold of this iterative segmentation Calculate the required adaptive leak threshold of leak judgement method；

Segmentation leakage phenomenon includes：Drastically leakage, progressive leakage and end leakage；

The adaptive leak threshold includes：

(1) for the first leak threshold E of drastically leakage phenomenon setting₁：First leak threshold is set as In M-R to the M times iterative process, adjacent secondary iteration generates the maximum value of new seed points difference, i.e.,：

(2) for the second leak threshold E of the progressive leakage phenomenon setting₂：Second leak threshold is set as In M-R to the M times iterative process, the maximum value of new seed points is generated, i.e.,：

(3) for the third leak threshold E of end leakage setting₃With the 4th leak threshold E₄：Iv-th iteration is grown E₃And E₄It is respectively set as：With

The Iter₁For the first iteration threshold, the Iter₁=R/2, wherein R be adaptive scale, R=(x_max-x_min+ y_max-y_min)/2, the x_min, y_min, x_max, y_max, the last layer of cleavage layer is marked in length and width two-dimensional directional for the main tracheae Maximum and min coordinates；

The Iter₂For secondary iteration threshold value, the Iter₂=R/4；

1026：Leak judgement rule is determined, if the 3D Region growing segmentations of current region have completed n-th and changed Generation, and last leakage phenomenon is happened at the M times iteration, then the leakage rule is：

Further, which is characterized in that described 102 further include：

1027：Determine leakage processing method, including：

10271：When iv-th iteration segmentation leaks, N-Iter is calculated₂With the maximum value N of M+1_max；If the N Secondary iteration is exactly the most last iteration of the main bronchus growth segmentation, then N_max=N；

10272：Remove N described in segmentation queue_maxThe tracheae label of whole seed points caused by after secondary iteration；

10273：To the N_maxThe seed point set of secondary iterationUsing the adaptive threshold 3D region growings model carries out simulation growth segmentation, and the maximum iteration of simulation growth cutting procedure is set as Iter₁；

10274：Single trachea-seed point v in described 10273_iFive leak threshold E of simulation growth regulation₅By v_iSurrounding is marked The neighborhood territory pixel points for being denoted as tracheae determine, count the seed point v_i6 neighborhood territory pixel points and 18 neighborhood territory pixels point in gas Pipe reference numerals, and weight W and obtain E₅：

10275：Record the seed point v_iThe new seed points c caused by each iteration in simulation growth segmentation₁, c₂,...,c_R/2If adjacent time the new seed points difference is not less than E₅, then fromThe middle removal v_iIf adjacent time described New seed points difference is less than E₅Then retain the v_i；

10276：Described 10272 are repeated, after growth segmentation is all simulated, after screeningAs it is described from Adapt to the initial segmentation seed point set of threshold value 3D region growing model next iterations.

Further, the termination rules are：

When adaptive threshold 3D region growings model growth cutting procedure is examined by the adaptive threshold leak model It measures when there is no new seed point that can be marked as tracheae in growth cutting procedure when leaking or in described 1024, institute Adaptive threshold 3D region growing models are stated to terminate.

Further, described 103 include：

1031：The adaptive gray threshold of left bronchus tree is worth to according to the average gray of the left principal bronchus T_left, and by T_leftThe tracheae degree of membership rule is determined as the third gray threshold, and according to the third gray threshold；

1032：Extract seed point set S caused by the most last iteration of the left principal bronchus growth segmentation_leftIf The seed point set that the fixed adaptive threshold 3D region growings model carries out simulation growth segmentation is S_pre=S_left；

1033：The S is screened using the leakage processing method of the adaptive threshold leak model_pre, obtain initial segmentation Seed point set S '_pre, wherein, the weight W is set as R；

1034：The S ' is divided using adaptive threshold 3D region growings model growth_pre；

1035：Growth using the leak judgement method detection described 1034 of the adaptive threshold leak model was divided If the growth cutting procedure leaks, the seed of the simulation growth segmentation is obtained using the leakage processing method for journey Point set S_pre, perform 1033；

If the growth cutting procedure does not leak, 1034 are continued to execute, until the adaptive threshold 3D regions are given birth to Model terminates, and completes to extract the second Lei Zuo tracheaes branch from the CT images；

The mode of the second Lei Zuo tracheaes branch is obtained based on 1031-1035 described above, by the T_leftIt replaces with T_right, the S_leftReplace with S_right, complete to extract the second Lei You tracheaes branch from the CT images, with described in realization The extraction of second class tracheae branch.

Further, described 105 include：

1051：Extraction has grown the whole edge pixel points of the second class tracheae branch divided and finished as seed point Set S_local, while by the S_localIt is set as the seed point set S of the simulation growth segmentation_pre；

1052：By the 1051 seed point set S_preThe gray value of each seed point itself be set as the seed point The local gray level threshold value of segmentation, and determine tracheae degree of membership rule according to the local gray level threshold value；

1053：The S is screened using the leakage processing method_pre, obtain initial segmentation seed point set S '_pre；

1054：Using S ' described in the adaptive threshold 3D region growing model treatments of adjustment_pre；

1055：Described 1054 growth is detected using the leak judgement method of the adaptive threshold leak model of adjustment Cutting procedure if the growth cutting procedure leaks, is obtained using the leakage processing method of the leak model of adjustment The seed point set S of the simulation growth segmentation_pre, perform described 1053；If the growth cutting procedure does not leak, after It is continuous to perform described 1054, until the adaptive threshold 3D region growings model of the adjustment terminates, complete to third class tracheorrhaphy The extraction of branch.

Further, described 106 include：

1061：Tracheae tree is formed according to the main tracheae, main bronchus, the second class tracheae branch and third class tracheae branch Topological structure, obtain the initial seed point of the whole tip tracheae dendritic growths segmentations of the tracheae tree and the initial seed point Initial growth direction, including：

10611：Calculate the source distance field and distance from boundary field of the tracheae tree；

10612：Extract all pixel S with local maximum of the tracheae tree source distance field_end={ s₁, s₂,...,s_nAs the tip tracheae dendritic growth segmentation initial seed point；

10613：To any initial seed point v₀∈S_end, find the seed point v₀26 neighborhood territory pixel points in have part The pixel u of maximum boundary distance field then sets the seed point v₀Initial growth direction vectorFor：

1062：By the seed point v₀Add in segmentation queue；

1063：Determine the direction of growth of the tip tracheae branch：To by direction of growth vectorIt obtains The seed point v obtained_i26 neighborhood territory pixel point w,It is the seed Point v_iThe direction vector of the seed point w is grown into, it is describedWill simultaneously with it is describedWithKeep close, the growth side It is to term restriction：

1064：Determine tip tracheae branch degree of membership rule：

If the seed point v_iThe 26 neighborhood territory pixel points for meeting 1063 condition beDue to the end The diameter of tip tracheae branch is very small, and the pixel of the tip tracheorrhaphy bronchial lumen is more most than the tip tracheae branch Several 26 neighborhood territory pixel point gray values are low, and appointing in 1 26 neighborhood territory pixel point for meeting condition of selection is only needed in comparing every time Neighborhood territory pixel point, specific rules are set as：

1065：In ith iteration, first seed point v of queue is taken out_i, search the v_i26 neighborhood territory pixel points in institute There is unlabelled pixelSelection meets the neighborhood territory pixel point of the tip tracheae branch degree of membership ruleBy described inAs new seed point v_i+1Add in segmentation queue；

1066：Repeat described 1065；If described 1065 have not had the generation of new seed point or described 1062 initial seed point The R times iteration is completed, stops the growth to the initial seed point；

1067：To the S_endIn all initial seed points carry out growth segmentation using described 1065, complete to the end The extraction of tip tracheae branch.

(3) advantageous effect

The beneficial effects of the invention are as follows：

A kind of method that tracheae tree is automatically extracted from chest CT image of the present invention, including：For chest CT image, it is based on 3D Region growing segmentation modes obtain the main tracheae and main bronchus of tracheae tree；Based on 3D Region growing segmentations mode, obtain The average information of the average information of main tracheae and the main bronchus obtained is established to extract the adaptive of the second class tracheae branch Threshold value 3D region growings model and adaptive threshold leak model；According to adaptive threshold 3D region growings model and adaptive thresholding It is worth leak model, extracts the second class tracheae branch of chest CT image；According to the average information of the second class tracheae branch of extraction, Adjust the parameter of adaptive threshold 3D region growings model and adaptive threshold leakage model；According to the adaptive threshold 3D of adjustment Region growing model and adaptive threshold leak model extract the third class tracheae branch of chest CT image；According to main tracheae, master The topological structure of bronchus, the second class tracheae branch and third class tracheae branch composition tracheae tree extracts the tip gas of tracheae tree Pipe branch obtains the tracheae tree of the chest CT image.Method provided by the invention improves the extraction tracheae tree from CT images Tracheae segmentation precision, while reduce the extraction time of tracheae tree.

Description of the drawings

Fig. 1 is to the 2D schematic diagrames of tracheae branch taxonomy to be extracted in the embodiment of the present invention；

Fig. 2 is the flow chart that the embodiment of the present invention automatically extracts tracheae tree method from chest CT image；

Fig. 3 is the main trachea-seed point schematic diagram of selection of the embodiment of the present invention；

Fig. 4 is the schematic diagram of the different leakage phenomenons of the embodiment of the present invention；

Fig. 5 is the process schematic that tracheae tree is automatically extracted in the slave chest CT image of the embodiment of the present invention；

Fig. 6 is the extraction result schematic diagram for CT image window tracheae trees in Fig. 1；

Fig. 7 is that method using the present invention extracts result schematic diagram to the 3D tracheaes tree of the chest CT image of different condition.

Specific embodiment

In order to preferably explain the present invention, in order to understand, below in conjunction with the accompanying drawings, by specific embodiment, to this hair It is bright to be described in detail.

(scanning device, breathing phases, reconstruction parameter, noise grade), different patients and different tissues are imaged in different CT Under the conditions of lesion, gray scale, form, the position difference of tracheae are very big.As shown in Figure 1, by tracheae tree tissue include branch according to Feature in CT images is divided into four classes, including：

(1) as shown in Figure 1A, main tracheae and main bronchus, i.e. first kind tracheae branch：The tracheae chamber of low gray value is high The tube wall of gray value is closely surrounded and is separated with pulmonary parenchyma；

(2) as shown in Figure 1B, the second class tracheae branch includes leaf, section and the sub- other tracheae branch of section grade：Main bronchus Pulmonary parenchyma is stretched by division, tube chamber and tube wall grey-scale contrast are relatively high；But the factors such as imaging and lesion make localized branches Variation takes place in place's intensity profile, and tracheae is lower with peripheral region contrast；

(3) as shown in Figure 1 C, third class tracheae branch, i.e., rudimentary tracheae branch：Second class tracheae forms grade by division Not lower tracheae branch, the intensity profile of these branches and first and second class branch gap bigger, tube chamber gray scale is more between each branch It is uneven；

(4) as shown in figure iD, tip tracheae branch, i.e. the 4th class tracheae branch：Third class tracheae is expanded to by division The tracheae tree tip that peripheral lung is formed；Diameter is no more than 2 pixels on CT images；Due to lesion or local volumetric effect shadow It rings, the raising of tube chamber gray scale is apparent with second and third class tube chamber gray difference.

Under normal conditions, for tracheae tree after dividing step by step, tracheal wall is thinning dimmed on CT images, it is difficult to distinguish its institute The tracheae chamber of encirclement and external pulmonary parenchyma.Tracheae tree gassy in air-breathing place CT images and turgor, tube chamber gray scale are very low, Tube wall gray scale is higher, and rudimentary tracheae is also more easy to identify；And expiration phase CT images are due to air exclusion in tracheae tree, tube chamber receipts Contracting, rudimentary tracheae are almost closed, and tracheae tree intensity profile and air-breathing place difference are larger.Therefore, according to man-machine interactively or system One preset threshold value can greatly improve complexity and the processing time of tracheae extraction；Improperly artificial setting may both cause big The mistake segmentation (leakage) of range, it is also possible to leak in order to avoid accidentally dividing and divide a large amount of branches.

In view of the above-mentioned problems, the present invention determines sequential extraction process according to the tracheae branch classification concluded, according to image Self information is obtained according to extracted tracheae branching characteristic and growth cutting procedure comprising adaptive in different extraction steps The model of parameter and rule obtains complete tracheae tree.

Embodiment 1

The technical issues of the present embodiment solves is to provide a kind of method that tracheae tree is automatically extracted in CT images from chest, profit The required man-machine interactively of the prior art and parameter setting are avoided with adaptive threshold, to adapt to a variety of image-forming conditions, lesion feelings The tracheae tree segmentation task of condition.As shown in Figure 1, the present invention provides a kind of sides that tracheae is automatically extracted from chest CT image Method, technical solution are as follows：

101st, for chest CT image, the first kind tracheae branch of tracheae tree is obtained based on 3D Region growing segmentation modes, I.e. main tracheae and main bronchus, wherein, main bronchus includes left principal bronchus and right principal bronchus；

102nd, based in 3D Region growing segmentations mode, the average information of main tracheae and the main bronchus of acquisition that obtain Between information, establish to extract the adaptive threshold 3D region growings model of the second class tracheae branch and adaptive threshold tunnelling ray Type；

103rd, according to adaptive threshold 3D region growings model and adaptive threshold leak model, chest CT image is extracted Second class tracheae branch, the second class tracheae branch include leaf, section and the sub- other tracheae branch of section grade；

104th, according to the average information of the second class tracheae branch of extraction, adjustment adaptive threshold 3D region growings model and Adaptive threshold reveals the parameter of model；

105th, according to the adaptive threshold 3D region growings model and adaptive threshold leak model of adjustment, chest CT is extracted The third class tracheae branch of image；

106th, opening up for tracheae tree is formed according to main tracheae, main bronchus, the second class tracheae branch and third class tracheae branch Structure is flutterred, extracts the tip tracheae branch of tracheae tree, obtains the tracheae tree of the chest CT image.

Embodiment 2

Main tracheae and main bronchus are surrounded and detached with pulmonary parenchyma by more complete highlighted tracheal walls, are extracted from CT images Main tracheae and main bronchus do not have to consider leakage situation, and difficulty is relatively low, therefore use common threshold value 3D Region growing segmentation sides Method can effectively obtain main tracheae and main bronchus.

In the present embodiment, first kind tracheae branch is obtained from chest CT image, including：

First, main tracheae is obtained from chest CT image：

1011st, chest CT image is read in, for the influence that CT image noises is avoided to divide subsequent growth, to chest CT image Carry out the pre-treatment job that three dimension scale is the Gaussian smoothing of σ=0.5mm；

1012nd, one layer of CT image of pretreatment potruncus CT images is obtained from chest top to chest bottom direction, and to the tomographic image Image binaryzation processing is carried out, obtains foreground area and the background area of binaryzation potruncus CT images, wherein, image binaryzation Including any binarization method, as shown in figure 3, the low dark gray values region of image is set to foreground area, highlighted gray value is put For background area；

1013rd, since the initial layers position of chest scan can reach oral cavity, after obtaining CT images and binaryzation by top, meeting Generate the very irregular shape prospect connected domain (oral cavity and nasal cavity are shown in Fig. 3 A) of large area or far from before the two dimension synusia center Scape connected domain (bottleneck throat or apex pulmonis are shown in Fig. 3 B)；These are not the present embodiment area of interest, therefore, are selected in the present embodiment It selects centered on the center of binaryzation potruncus CT images, setting image length and width each 1/4 is the window of the length of side, in this window, Select area the maximum in multiple prospect connected domains (see Fig. 3 C)；

1014th, judge whether the prospect connected domain of selection meets the first preset condition, if satisfied, then making prospect connected domain Divide initial connected domain for main tracheae, prospect connected domain central point is as initial seed point s₀, wherein, the first preset condition is is somebody's turn to do The perimeter l and area a of prospect connected domain meet condition：l²/ 4 π a >=0.6 (see Fig. 3 D)；

1015th, initial seed point s is handled based on 3D Region growing segmentation modes₀, the main tracheae of tracheae tree is obtained, it is specific to walk Suddenly it is (attached to be not shown in figure)：

S01, by initial seed point s₀Segmentation queue is added in, based on 3D Region growing segmentation modes, is carried out to chest bottom direction 3D Region growing segmentations calculate the pixel average gray value μ of initial connected domain_iniAnd standard deviation sigma_ini, 3D Region growing segmentations adopt It is with the first gray threshold：T_tra=μ_ini+2σ_ini；

S02, in the ith iteration of 3D Region growing segmentation modes, take out segmentation queue in first seed pointIt will The pixel average gray value of prospect connected domain is not more than default first gray threshold T_traAnd unlabelled seed point26 neighborhoods Pixel w₁,w₂,...,w_mIt marks as main tracheae；

S03 and using 26 neighborhood territory pixel points as new seed pointAdd in segmentation queue；Continue with this Divide removing for queue in ith iterationExcept remaining seed pointUntil whole seeds of the ith iteration Point is all handled through 3D Region growing segmentation modes；

S04, iterations add 1 and repeat the iterative process of ith, when tracheae label lower side main in CT images is split into During two connected domains, main tracheae growth segmentation finishes, and obtains the main tracheae of tracheae tree.Wherein, main tracheae is close to the one end in oral cavity For upside, one end far from oral cavity is downside, and main tracheae lower side is distalmost end of the main tracheae far from oral cavity.

2nd, main bronchus is obtained from chest CT image：

Main bronchus includes left principal bronchus and right principal bronchus, and left main branch gas is obtained based on 3D Region growing segmentation modes The step of pipe, includes (attached to be not shown in figure)：

M01, using the central point of the connected domain on the left of main tracheae anatomy as the initial seed point S of left principal bronchus_left；

M02, the pixel average gray value μ according to the main tracheae region of acquisition_iniAnd standard deviation sigma_iniDetermine left main branch Second gray threshold T of tracheae segmentation_bro, T_bro=μ_ini+2σ_ini；

M03, by initial seed point S_leftSegmentation queue is added in, based on 3D Region growing segmentation modes, and combines the second ash Whole seed points in threshold process segmentation queue are spent, when left principal bronchus label is split into two connected domains in CT images, Left principal bronchus segmentation terminates, and obtains left principal bronchus；

The step of obtaining right principal bronchus based on 3D Region growing segmentation modes includes (attached to be not shown in figure)：

N01, using the central point of the connected domain on the right side of main tracheae anatomy as the initial seed point S of right principal bronchus_right；

N02, the pixel average gray value μ according to the right main tracheae region of acquisition_iniAnd standard deviation sigma_iniDetermine main branch Second gray threshold T of tracheae segmentation_bro, T_bro=μ_ini+2σ_ini；

N03, by initial seed point S_rightSegmentation queue is added in, based on 3D Region growing segmentation modes, and combines the second ash Whole seed points in threshold process segmentation queue are spent, when right principal bronchus label is split into two connected domains in CT images, Right principal bronchus segmentation terminates, and obtains right principal bronchus.

Embodiment 3

In the present embodiment, according to the main tracheae and the tracheae intensity profile of main bronchus, space scale, segmentation obtained The average informations such as procedural information establish adaptive threshold 3D region growings model and adaptive threshold leak model.

Adaptive threshold 3D region growing models are established to be as follows：

1021：Obtain initial segmentation seed point set；Initial segmentation seed point set includes：Left principal bronchus is at iteration end Whole seed points and right principal bronchus when only in segmentation queue divide whole seed points in queue in iteration ends.

1022：Obtain gray threshold；The present embodiment sets certain gray threshold to determine tracheae to the segmentation of tracheae tree The upper limit of segmentation, while gray value more than the upper limit but is also divided into tracheae by the pixel that trachea area closely surrounds, so as to Extract the raised tracheae branch of tube chamber gray scale in Fig. 1 C, Fig. 1 D.

1023：Tracheae degree of membership rule is determined according to gray threshold, including：

(1) first is subordinate to rule：The gray value I of seed point w_wNo more than third gray threshold T；

(2) second are subordinate to rule：The gray value I of seed point w_wNo more than the 4th gray threshold, the 4th gray threshold is production The gray value T of the seed point v of raw seed point w_v；

(3) third is subordinate to rule：The pixel that tracheae is marked as in the 6 neighborhood territory pixel points of seed point w is no less than 4, I.e.：Wherein L represents dividing mark,Represent 6 neighborhoods of seed point w The set of pixel；

(4) the 4th are subordinate to rule：The 6 neighborhood territory pixel point gray values of seed point w and seed point w and average value no more than the Three gray threshold T, and in the 6 neighborhood territory pixel points of seed point w, pixel of the gray value no more than third gray threshold T is no less than 4 It is a, i.e.,：

1024：Obtain growing strategy：In ith iteration, segmentation first seed point of queue is taken outIt willMeet gas Pipe degree of membership is regular and unlabelled26 neighborhood territory pixel point w₁,w₂,...,w_mLabeled as tracheae, and by 26 neighborhood territory pixel points w₁,w₂,...,w_mAs new seed pointAdd in segmentation queue；Continue with remaining seed in segmentation queue PointUntilIt is all finished by growth, into i+1 time iteration；This step is repeated until meeting Termination rules；

Termination rules are：When adaptive threshold 3D region growings model grows cutting procedure by adaptive threshold leak model It detects when there is no new seed point that can be marked as tracheae in growth cutting procedure when leaking or in 1024, it is adaptive Threshold value 3D region growing models is answered to terminate.

Because the factors such as imaging and lesion influence, difference etc. in the tracheae gray feature, same CT images between different CT images Grade, the tracheae gray feature of different location are different, and the adaptive threshold 3D region growing models of the present embodiment can basis The gray feature of first kind tracheae calculates the threshold value of the second class tracheae segmentation, and third class is voluntarily determined according to each tracheae branch location The threshold value of branch's local segmentation.

In tracheae tree cutting procedure, by tracheae chamber cross thinning dimmed tracheal wall enter pulmonary parenchyma mistake segmentation be referred to as let out Leakage.Control leakage is to extract the significant challenge that tracheae tree faces, and segmentation leakage can cause extraction result form, textural anomaly, shadow Ring the clinical analysis of tracheae tree and other application expansion.Different CT images are because of factors such as imaging, lesions, in any branch, any Position can may all leak.Artificial given threshold condition excessively can strictly omit a large amount of tracheae branches；Excessively it is loose then without Method effectively avoids leakage, and artificial threshold value can not be widely used in the CT images of different condition.The present embodiment is using adaptive Sub-threshold leakage model is answered, the adaptive threshold and rule needed according to cutting procedure information acquisition model is to avoid leakage, it is ensured that It can continue to obtain more tracheae branches, and expand the scope of application of method on this basis.

In the present embodiment, the step of establishing adaptive threshold leak model is specific as follows：

1025：If the 3D Region growing segmentations of current region have completed iv-th iteration, and last leakage phenomenon occurs In the M times iteration, leakage detection method is：The novel species that iteration is divided every time during 3D Region growing segmentations in record 1024 Son points C₁,C₂,...,C_NIf not leaked during iterative segmentation, calculated with the adaptive leak threshold of this iterative segmentation The required adaptive leak threshold of leak judgement method；

The present embodiment will divide leakage phenomenon conclusion and be divided into three classes：

(1) it drastically leaks：As shown in Figure 4 A, the unexpected hair of new seed points caused by certain iteration in cutting procedure is grown It is raw to increase severely, indicate that the secondary growth has passed over tracheal wall and diffuses into pulmonary parenchyma；

(2) progressive leakage：As shown in Figure 4 B, seed points are gradually increased and are run up to very big caused by each iteration Value, different from the first kind, adjacent generation seed points difference keeps lower value, but a large amount of lung parenchyma tissues are accidentally divided into gas at this time Pipe；

(3) end leaks：When region growing is close to tracheae end, new seed points are gradually decrease to 0 under normal circumstances, But as shown in Figure 4 C, if seed points become larger again since small value；Region growing terminates after iteration several times, but at this time In part, tracheae terminal region leaks.

Adaptive leak threshold includes：

(1) for the first leak threshold E of drastically leakage phenomenon setting₁：First leak threshold is set as M-R to M In secondary iterative process, adjacent secondary iteration generates the maximum value of new seed points difference, i.e.,：

(2) for the second leak threshold E of progressive leakage phenomenon setting₂：Second leak threshold is set as M-R to M In secondary iterative process, the maximum value of new seed points is generated, i.e.,：

(3) for the third leak threshold E of end leakage setting₃With the 4th leak threshold E₄：The E of iv-th iteration growth₃ ForE₄It is set as

Iter₁For the first iteration threshold, Iter₁=R/2, wherein R be adaptive scale, R=(x_max-x_min+y_max-y_min)/ 2, x_min, y_min, x_max, y_max, the last layer of cleavage layer is marked in the maximum and min coordinates of length and width two-dimensional directional for main tracheae； Here the last layer of main tracheae label cleavage layer is meant：When main tracheae divides, left and right main bronchus is generated, for master Main tracheae is exactly last layer for bronchus；

Iter₂For secondary iteration threshold value, Iter₂=R/4；

1026：Determine leak judgement rule, if the 3D Region growing segmentations of current region have completed iv-th iteration, and Last leakage phenomenon is happened at the M times iteration, then leaks rule and be：

First leakage rule：Current iteration generates new seed points difference more than the first adaptive leakage threshold with preceding an iteration Value, and current iteration and the difference of last effective iterations are more than the first adaptive iteration threshold value；

Second leakage rule：Current iteration generates new seed points more than the second adaptive leak threshold；

Third leakage rule：It is more than the adaptive leak threshold of third or current iteration that current iteration, which generates new seed points, The whole new seeds divided in preceding first adaptive iteration threshold number are counted more than the 4th adaptive leak threshold；

The formula for leaking rule is as follows：

1027：Determine leakage processing method, including：

10271：When iv-th iteration segmentation leaks, N-Iter is calculated₂With the maximum value N of M+1_max；If n-th changes In generation, exactly main bronchus grew the most last iteration divided, then N_max=N；

10272：Remove N in segmentation queue_maxThe tracheae label of whole seed points caused by after secondary iteration；

10273：To N_maxThe seed point set of secondary iterationIt is given birth to using adaptive threshold 3D regions Long model carries out simulation growth segmentation, and the maximum iteration of simulation growth cutting procedure is set as Iter₁；

10274：Single trachea-seed point v in 10273_iFive leak threshold E of simulation growth regulation₅By v_iSurrounding is marked as The neighborhood territory pixel points of tracheae determine, count seed point v_i6 neighborhood territory pixel points and 18 neighborhood territory pixels point in tracheae reference numerals, And it weights W and obtains E₅：

Seed point v_iThe labeled neighborhood point of surrounding is more, it was demonstrated that seed point v_iThe space scale of the tracheae branch is bigger；

10275：Record seed point v_iThe new seed points c caused by each iteration in simulation growth segmentation₁,c₂,..., c_R/2If adjacent secondary new seed points difference is not less than E₅, then fromMiddle removal v_iIf adjacent secondary new seed points difference is less than E₅Then Retain v_i；

10276：10272 are repeated, after growth segmentation is all simulated, after screeningAs adaptive threshold The initial segmentation seed point set of 3D region growing model next iterations.

Embodiment 4

Bronchial tree middle period, section, the sub- other bronchus intensity profile of section grade are more uniform, using based on global threshold from Threshold value 3D region growings model extraction the second class tracheae branch is adapted to, leakage phenomenon may occur for localized region using adaptive Sub-threshold leakage model is handled.The present embodiment separately extracts left and right bronchial tree, including：

1031：According to the average gray value μ of left (right side) main bronchus_left, σ_left(μ_right, σ_right) obtain left (right side) branch gas The adaptive gray threshold T of Guan Shu_left=μ_left+2σ_left(T_right=μ_right+2σ_right), and by T_left(T_right) as third Gray threshold, and determine tracheae degree of membership rule according to third gray threshold；

1032：Seed point set S caused by the most last iteration of left (right side) main bronchus growth segmentation of extraction_left (S_right), it sets adaptive threshold 3D region growings model and carries out the seed point set of simulation growth segmentation as S_pre=S_left(S_pre =S_right)；

1033：S is screened using the leakage processing method of adaptive threshold leak model_pre, obtain initial segmentation seed point set Close S '_pre, wherein, the second class tracheae branch W is needed to take higher value in the present embodiment, so weight W is set as R,；

1034：Segmentation S ' is grown using adaptive threshold 3D region growings model_pre；

1035：Using the growth cutting procedure of the leak judgement method detection 1034 of adaptive threshold leak model, if raw Long cutting procedure leaks, and utilizes the seed point set S for leaking processing method acquisition simulation growth segmentation_pre, perform 1033；

If growth cutting procedure does not leak, 1034 are continued to execute, until adaptive threshold 3D regions life model terminates, Left (right side) the tracheae branch of the second class is extracted in completion from CT images；To realize the extraction of the second class tracheae branch.

Embodiment 5

Third class tracheae branch has been fully inserted into pulmonary parenchyma, although with using the adaptive threshold based on global threshold 3D region growings model cannot obtain these branches, but these branches are connected with tracheae tree, each branch internal, adjacent pixel The gray scale of point is still close to each other, therefore, is grown since extracted tracheae tree table face, according to the similarity of adjacent pixel gray scale It can be found that potential trachea area.The intensity profile of third class tracheae branch is more complicated, is also easier to leak, be still to herein Using adaptive threshold leak model.

Extraction third class tracheae branch is as follows：

1051：Extraction has grown the whole edge pixel points of the second class tracheae branch divided and finished as seed point set S_local, while by S_localIt is set as the seed point set S of simulation growth segmentation_pre；

1052：By the seed point set S in 1051_preThe gray value of each seed point itself be set as the seed point point The local gray level threshold value cut, and local gray level threshold value determines tracheae degree of membership rule according to this；

1053：S is screened using the leakage processing method of adaptive threshold leak model_pre, obtain initial segmentation seed point set Close S '_pre, wherein, smaller value should be taken for the weight W of third class tracheae branch in the present embodiment, so weight W is set asl_x、l_y、l_zIt is chest CT image single pixel in three directions On size；

1054：Using the adaptive threshold 3D region growing model treatments S ' of adjustment_pre；

1055：Growth using the leak judgement method detection 1054 of the adaptive threshold leak model of adjustment was divided Journey if growth cutting procedure leaks, is simulated using the leakage processing method of the adaptive threshold leak model of adjustment Grow the seed point set S of segmentation_pre, perform 1053；If growth cutting procedure does not leak, 1054 are continued to execute, until adjusting Whole adaptive threshold 3D region growings model terminates, and completes the extraction to third class tracheae branch.

Embodiment 6

Tip tracheae branch can not effectively be extracted using adaptive threshold 3D region growing models, using tracheae Tree topology determines the direction in space of each tip tracheae branch, under the constraint of this direction in space, is compared according to pixel grey scale Degree gradually extracts tip tracheae branch.

Tracheae topological structure refers to comprising the system including tracheae center path, branch's grade, branch's endpoint, branch 3D directions Row morphology, anatomy and geometry structure.Although the intensity profile of tip tracheae branch and other class tracheae intensity profile are poor It is different larger, but its tube chamber is still more slightly lower than tube wall gray value；Tip tracheae branch approximation is in rodlike, and same branch's starting terminates, Or the sub- grade bronchus after division is close with the parent bronchus direction before division；Such tracheal diameter is no more than 2 pixels.Cause This, continues search in CT image relays along tracheae branch direction has been extracted, searches for the trachea area for meeting above-mentioned gray feature, can be with Obtain including the tracheae tree of more final words.

Do not need to obtain complete topological structure in practical application, one embodiment of the invention only need tracheae branch endpoint and Center path point passes through source distance field DFS (english abbreviation of Distance Field of Source) and distance from boundary field DFB (english abbreviation of Distance Field of Boundary) can quick obtaining information needed.

In the present embodiment, tracheae is formed according to main tracheae, main bronchus, the second class tracheae branch and third class tracheae branch The topological structure of tree extracts the tip tracheae branch of tracheae tree, is as follows：

1061：Opening up for tracheae tree is formed according to main tracheae, main bronchus, the second class tracheae branch and third class tracheae branch Structure is flutterred, obtains the initial life of the initial seed point and the initial seed point of the whole tip tracheae dendritic growth segmentations of tracheae tree Length direction, including：

10611：Calculate the DFS and DFB of tracheae tree；

10612：Extract all pixel S with local maximum of tracheae tree DFS_end={ s₁,s₂,...,s_nConduct The initial seed point of tip tracheae dendritic growth segmentation,

10613：To any initial seed point v₀∈S_end, find seed point v₀26 neighborhood territory pixel points in have local maxima The pixel u of DFB, then set seed point v₀Initial growth direction vectorFor：

1062：By seed point v₀Add in growth segmentation queue；

1063：Determine the direction of growth of tip tracheae branch：

Direction condition is divided in the growth of tip tracheae branch：

A. to the required direction of growth, if its have on three-dimensional either direction with corresponding dimension target direction in the straight angle or The required direction of growth is vertical with target direction, then the required direction of growth does not meet the growth segmentation direction item of tip tracheae branch Part；

B. target direction includes the seed point initial segmentation direction of tip tracheae branch and the preceding an iteration direction of growth；

Specific explanations are：To by direction of growth vectorThe seed point v of acquisition_i26 neighborhood territory pixel points W,It is seed point v_iThe direction vector of seed point w is grown into, Will simultaneously withWithKeep close, direction of growth term restriction is：

1064：Determine tip tracheae branch degree of membership rule：

Tip tracheae branch degree of membership rule is (attached to be not shown in figure)：

A. it presses tip segmentation direction condition and obtains the required growth unlabelled 26 neighborhood point set of seed point；

B. to each point in step set A, condition need to further be met：Gray value is not more than the neighborhood of a point in its 26 neighborhood Points not more than 3；

C. using through the pixel with minimum gradation value that step B is constrained as new tip trachea-seed point；

Tip tracheae branch degree of membership rule is specially：If seed point v_iThe 26 neighborhood territory pixel points for meeting 1063 conditions beSince the diameter of tip tracheae branch is very small, the pixel of tip tracheorrhaphy bronchial lumen is than tip tracheae Most of 26 neighborhood territory pixel point gray values of branch are low, and 1 26 neighborhood territory pixel point for meeting condition of selection are only needed in comparing every time In any neighborhood territory pixel point, specific rules formula is set as：

1065：In ith iteration, first seed point v of queue is taken out_i, search v_i26 neighborhood territory pixel points in it is all not The pixel of labelSelection meets the neighborhood territory pixel point of tip tracheae branch degree of membership ruleIt willMake For new seed point v_i+1Add in segmentation queue；

1066：Repeat 1065；If 1065 surpass without the generation of new seed point or 1062 initial seed point iteration growth number Cross third iteration threshold (third iteration threshold Iter₃=R), i.e., 1062 initial seed point has completed the R times iteration, stopping pair The growth of initial seed point；

1067：To S_endIn all initial seed points carry out growth segmentation using 1065, complete to tip tracheae branch Extraction.

In embodiments of the present invention, after whole tracheae branch extractions, it can obtain required tracheae tree.Especially need to refer to Go out, part third class tracheae branch influences because the factors such as lesion or local volumetric effect occur, and also shows tip on CT images Therefore tracheae bifurcation state, can perform 105 and 106 again after performing 106, further improve segmentation precision.

As shown in figure 5, the process schematic of tracheae tree is automatically extracted from chest CT image for one embodiment of the invention, such as It is main tracheae and main bronchus shown in Fig. 5 A；As shown in Figure 5 B, newly-increased tissue part is through adaptive threshold 3D region growing moulds The bronchial tree for including the second class tracheae branches of level branches such as leaf, section, sub- section of type segmentation；As shown in Figure 5 C, it compares The tissue part that Fig. 5 B are increased newly is the third class tracheae branch that the segmentation of adaptive threshold 3D region growings model obtains；Such as Fig. 5 D institutes Show, the tissue part that Fig. 5 C are increased newly that compares is the tip tracheae branch extracted based on tracheae tree topology.

Fig. 6 lists the extraction result to CT image windows tracheae tree in Fig. 1.1 region of Fig. 6 A is the first kind in Figure 1A The segmentation result of tracheae branch；2 regions of Fig. 6 B and Fig. 6 C are the segmentation results of the second class tracheae branch in Figure 1B and Fig. 1 C, by Fig. 6 B through the present embodiment adaptive threshold leak model as it can be seen that handle, the adaptive threshold 3D region growings based on global threshold The tracheae branch of model extraction is not bled into pulmonary parenchyma at the tube wall place of obscuring；3 regions of Fig. 6 C are third class tracheae branches Segmentation result, can be effective using the adaptive threshold 3D region growings model based on local threshold and adaptive threshold leak model Obtain such branch；4 regions of Fig. 6 D are the segmentation results of tip tracheae branch, and the present embodiment has divided tracheae branch from one section End start, complete the extraction to narrow tip branch.

Fig. 7 is that method using the present invention extracts result schematic diagram to the 3D tracheaes tree of the chest CT image of different condition. The relevant information of each group CT images is as follows：

It is to be appreciated that it is described above to what specific embodiments of the present invention carried out simply to illustrate that the skill of the present invention Art route and feature, its object is to allow those skilled in the art that can understand present disclosure and implement according to this, but The present invention is not limited to above-mentioned particular implementations.Every various change made within the scope of the claims is repaiied Decorations should all be covered within the scope of the present invention.

Claims (12)

1. A kind of 1. method that tracheae tree is automatically extracted from chest CT image, which is characterized in that including：

   101st, for chest CT image, the main tracheae and main bronchus of tracheae tree are obtained based on 3D Region growing segmentation modes；

   102nd, based on the 3D Region growing segmentations mode, the average information of the main tracheae obtained and the main branch of acquisition The average information of tracheae establishes adaptive threshold 3D region growings model and adaptive thresholding for extracting the second class tracheae branch It is worth leak model；

   103rd, according to the adaptive threshold 3D region growings model and the adaptive threshold leak model, the chest is extracted Second class tracheae branch of CT images；

   104th, according to the average information of the second class tracheae branch of extraction, the adaptive threshold 3D region growing moulds are adjusted The parameter of type and adaptive threshold leakage model；

   105th, according to the adaptive threshold 3D region growings model of adjustment and the adaptive threshold leak model, institute is extracted State the third class tracheae branch of chest CT image；

   106th, opening up for tracheae tree is formed according to the main tracheae, main bronchus, the second class tracheae branch and third class tracheae branch Structure is flutterred, extracts the tip tracheae branch of the tracheae tree, obtains the tracheae tree of the chest CT image.
2. 2. according to the method described in claim 1, it is characterized in that, described 101 include：

   1011st, the chest CT image is pre-processed；

   1012nd, image binaryzation processing is carried out to the pretreated chest CT image, obtains the chest CT after binaryzation The foreground area of image and background area；

   1013rd, centered on the center of chest CT image described after binaryzation, select the corresponding area of the foreground area maximum A prospect connected domain；

   1014th, judge whether the prospect connected domain of selection meets the first preset condition, if satisfied, then connecting the prospect Initial connected domain is divided in domain as main tracheae, and the prospect connected domain central point is as initial seed point s₀；

   1015th, the initial seed point s is handled based on the 3D Region growing segmentations mode₀, the main tracheae of acquisition tracheae tree.
3. 3. according to the method described in claim 2, it is characterized in that, described 1015 include：

   S01, by the initial seed point s₀Add in segmentation queue, based on the 3D Region growing segmentations mode, to chest bottom direction into Row 3D Region growing segmentations；

   S02, in the ith iteration of the 3D Region growing segmentations mode, take out it is described segmentation queue in first seed pointThe pixel average gray value of the prospect connected domain is not more than default first gray threshold T_traIt is and unlabelled described's 26 neighborhood territory pixel point w₁,w₂,...,w_mIt marks as main tracheae；

   S03 and using the 26 neighborhood territory pixel point as new seed pointAdd in the segmentation queue；It continues with It is described to divide removing for queue in the ith iterationExcept remaining seed pointIt is complete until the ith iteration Portion's seed point is all handled through the 3D Region growing segmentations mode；

   S04, iterations add 1 and repeat the iterative process of ith, the main tracheae label lower side point described in the CT images When splitting for two connected domains, the main tracheae of the tracheae tree is obtained, wherein, the main tracheae is upside close to the one end in oral cavity, One end far from the oral cavity is downside, and the main tracheae lower side is the distalmost end of the main tracheae far from the oral cavity.
4. 4. according to the method described in claim 3, it is characterized in that, described 101 further include；

   The main bronchus includes left principal bronchus and right principal bronchus, is obtained based on the 3D Region growing segmentations mode described The step of left principal bronchus, includes：

   M01, using the central point of the connected domain on the left of the main tracheae anatomy as the initial seed point of left principal bronchus S_left；

   M02, the pixel average gray value μ according to the main tracheae region of acquisition_iniAnd standard deviation sigma_iniDetermine the left side Second gray threshold T of main bronchus segmentation_bro, T_bro=μ_ini+2σ_ini；

   M03, by initial seed point S_leftSegmentation queue is added in, based on the 3D Region growing segmentations mode, and with reference to described second Whole seed points in gray threshold processing segmentation queue, the left principal bronchus label described in the CT images are split into two During connected domain, the left principal bronchus segmentation terminates, and obtains the left principal bronchus；

   And the step of based on the 3D Region growing segmentations mode obtaining the right principal bronchus, includes：

   N01, using the central point of the connected domain on the right side of the main tracheae anatomy as the initial seed point of right principal bronchus S_right；

   N02, the pixel average gray value μ according to the right main tracheae region of acquisition_iniAnd standard deviation sigma_iniDetermine main branch Second gray threshold T of tracheae segmentation_bro, T_bro=μ_ini+2σ_ini；

   N03, by initial seed point S_rightSegmentation queue is added in, based on the 3D Region growing segmentations mode, and with reference to described the Whole seed points in the processing segmentation queue of two gray thresholds, the right principal bronchus label described in the CT images are split into two During a connected domain, the right principal bronchus segmentation terminates, and obtains the right principal bronchus.
5. 5. according to the method described in claim 4, it is characterized in that, described 102 include：Establish the adaptive threshold 3D regions Growth model specifically includes：

   1021：Obtain initial segmentation seed point set；

   1022：Obtain gray threshold；

   1023：Tracheae degree of membership rule is determined according to the gray threshold, including：

   (1) first is subordinate to rule：The gray value I of seed point w_wNo more than third gray threshold T；

   (2) second are subordinate to rule：The gray value I of the seed point w_wNo more than the 4th gray threshold, the 4th gray threshold is Generate the gray value T of the seed point v of the seed point w_v；

   (3) third is subordinate to rule：The pixel that tracheae is marked as in the 6 neighborhood territory pixel points of the seed point w is no less than 4, I.e.：Wherein L represents dividing mark,Represent the 6 of the seed point w The set of neighborhood territory pixel point；

   (4) the 4th are subordinate to rule：The 6 neighborhood territory pixel point gray values of the seed point w and seed point w and average value be not more than institute Third gray threshold T is stated, and in the 6 neighborhood territory pixel points of the seed point w, gray value is no more than the third gray threshold T's The pixel is no less than 4, i.e.,：

   1024：Obtain growing strategy：In ith iteration, segmentation first seed point of queue is taken outBy described inMeet institute It is regular and unlabelled described to state tracheae degree of membership26 neighborhood territory pixel point w₁,w₂,...,w_mLabeled as tracheae, and by described in 26 neighborhood territory pixel point w₁,w₂,...,w_mAs new seed pointAdd in the segmentation queue；It continues with described Divide remaining seed point in queueUntilIt is all finished by growth, into i+1 time iteration；Weight This step is performed again until meeting termination rules.
6. 6. according to the method described in claim 5, it is characterized in that, the initial segmentation seed point set includes：The left master Bronchus divides whole seed points in queue in iteration ends and the right principal bronchus divides queue in iteration ends In whole seed points.
7. 7. according to the method described in claim 5, it is characterized in that, described 102 further include：Establish the adaptive threshold leakage Model specifically includes：

   1025：Leakage detection method：Record the novel species that iteration is divided every time during 3D Region growing segmentations in described 1024 Son points C₁,C₂,...,C_NIf not leaked during iterative segmentation, calculated with the adaptive leak threshold of this iterative segmentation The required adaptive leak threshold of leak judgement method；

   Segmentation leakage phenomenon includes：Drastically leakage, progressive leakage and end leakage；

   The adaptive leak threshold includes：

   (1) for the first leak threshold E of drastically leakage phenomenon setting₁：First leak threshold is set as M-R and arrives In the M times iterative process, adjacent secondary iteration generates the maximum value of new seed points difference, i.e.,：

   (2) for the second leak threshold E of the progressive leakage phenomenon setting₂：Second leak threshold is set as M-R and arrives In the M times iterative process, the maximum value of new seed points is generated, i.e.,：

   (3) for the third leak threshold E of end leakage setting₃With the 4th leak threshold E₄：The E of iv-th iteration growth₃ And E₄It is respectively set as：With

   The Iter₁For the first iteration threshold, the Iter₁=R/2, wherein R be adaptive scale, R=(x_max-x_min+y_max- y_min)/2, the x_min, y_min, x_max, y_max, for the main tracheae mark the last layer of cleavage layer in length and width two-dimensional directional most Greatly and min coordinates；

   The Iter₂For secondary iteration threshold value, the Iter₂=R/4；

   1026：Determine leak judgement rule, if the 3D Region growing segmentations of current region have completed iv-th iteration, and Last leakage phenomenon is happened at the M times iteration, then the leakage rule is：
8. 8. the method according to the description of claim 7 is characterized in that described 102 further include：

   1027：Determine leakage processing method, including：

   10271：When iv-th iteration segmentation leaks, N-Iter is calculated₂With the maximum value N of M+1_max；If the n-th changes In generation, exactly the main bronchus grew the most last iteration divided, then N_max=N；

   10272：Remove N described in segmentation queue_maxThe tracheae label of whole seed points caused by after secondary iteration；

   10273：To the N_maxThe seed point set of secondary iterationUsing the adaptive threshold 3D areas Domain growth model carries out simulation growth segmentation, and the maximum iteration of simulation growth cutting procedure is set as Iter₁；

   10274：Single trachea-seed point v in described 10273_iFive leak threshold E of simulation growth regulation₅By v_iSurrounding is marked as The neighborhood territory pixel points of tracheae determine, count the seed point v_i6 neighborhood territory pixel points and 18 neighborhood territory pixels point in tracheae mark Numeration, and weight W and obtain E₅：

   10275：Record the seed point v_iThe new seed points c caused by each iteration in simulation growth segmentation₁,c₂,..., c_R/2If adjacent time the new seed points difference is not less than E₅, then fromThe middle removal v_iIf the adjacent new seed point Number difference is less than E₅Then retain the v_i；

   10276：Described 10272 are repeated, after growth segmentation is all simulated, after screeningAs described adaptive The initial segmentation seed point set of threshold value 3D region growing model next iterations.
9. 9. according to claim 5-8 any one of them methods, which is characterized in that the termination rules are：

   When adaptive threshold 3D region growings model growth cutting procedure is detected by the adaptive threshold leak model When there is no new seed point that can be marked as tracheae when leaking or in described 1024 in growth cutting procedure, it is described oneself Threshold value 3D region growing models are adapted to terminate.
10. 10. according to the method described in claim 9, it is characterized in that, described 103 include：

    1031：The adaptive gray threshold T of left bronchus tree is worth to according to the average gray of the left principal bronchus_left, and will T_leftThe tracheae degree of membership rule is determined as the third gray threshold, and according to the third gray threshold；

    1032：Extract seed point set S caused by the most last iteration of the left principal bronchus growth segmentation_left, described in setting The seed point set that adaptive threshold 3D region growings model carries out simulation growth segmentation is S_pre=S_left；

    1033：The S is screened using the leakage processing method of the adaptive threshold leak model_pre, obtain initial segmentation seed Point set S_p′_re, wherein, the weight W is set as R；

    1034：The S is divided using adaptive threshold 3D region growings model growth_p′_re；

    1035：Described 1034 growth cutting procedure is detected using the leak judgement method of the adaptive threshold leak model, If the growth cutting procedure leaks, the seed point set of the simulation growth segmentation is obtained using the leakage processing method Close S_pre, perform 1033；

    If the growth cutting procedure does not leak, 1034 are continued to execute, until model is given birth in the adaptive threshold 3D regions It terminates, completes to extract the second Lei Zuo tracheaes branch from the CT images；

    The mode of the second Lei Zuo tracheaes branch is obtained based on 1031-1035 described above, by the T_leftReplace with T_right, The S_leftReplace with S_right, complete to extract the second Lei You tracheaes branch from the CT images, to realize described second The extraction of class tracheae branch.
11. 11. according to the method described in claim 10, it is characterized in that, described 105 include：

    1051：Extraction has grown the whole edge pixel points of the second class tracheae branch divided and finished as seed point set S_local, while by the S_localIt is set as the seed point set S of the simulation growth segmentation_pre；

    1052：By the 1051 seed point set S_preThe gray value of each seed point itself be set as seed point segmentation Local gray level threshold value, and according to the local gray level threshold value determine tracheae degree of membership rule；

    1053：The S is screened using the leakage processing method_pre, obtain initial segmentation seed point set S_p′_re；

    1054：Using S described in the adaptive threshold 3D region growing model treatments of adjustment_p′_re；

    1055：Described 1054 growth segmentation is detected using the leak judgement method of the adaptive threshold leak model of adjustment Process, if the growth cutting procedure leaks, using described in the leakage processing method acquisition of the leak model of adjustment The seed point set S of simulation growth segmentation_pre, perform described 1053；If the growth cutting procedure does not leak, continue to hold Row described 1054 until the adaptive threshold 3D region growings model of the adjustment terminates, is completed to third class tracheae branch Extraction.
12. 12. according to the method for claim 11, which is characterized in that described 106 include：

    1061：Opening up for tracheae tree is formed according to the main tracheae, main bronchus, the second class tracheae branch and third class tracheae branch Flutter structure, obtain the whole tip tracheae dendritic growths segmentations of the tracheae tree initial seed point and the initial seed point just The beginning direction of growth, including：

    10611：Calculate the source distance field and distance from boundary field of the tracheae tree；

    10612：Extract all pixel S with local maximum of the tracheae tree source distance field_end={ s₁,s₂,..., s_nAs the tip tracheae dendritic growth segmentation initial seed point；

    10613：To any initial seed point v₀∈S_end, find the seed point v₀26 neighborhood territory pixel points in have local maxima The pixel u of distance from boundary field then sets the seed point v₀Initial growth direction vectorFor：

    1062：By the seed point v₀Add in segmentation queue；

    1063：Determine the direction of growth of the tip tracheae branch：To by direction of growth vectorThe kind of acquisition Sub- point v_i26 neighborhood territory pixel point w,It is the seed point v_iIt is raw The direction vector of the seed point w is grown to, it is describedWill simultaneously with it is describedWithKeep close, the direction of growth condition It is defined to：

    1064：Determine tip tracheae branch degree of membership rule：

    If the seed point v_iThe 26 neighborhood territory pixel points for meeting 1063 condition beDue to the tip gas The diameter of pipe branch is very small, and the pixel of the tip tracheorrhaphy bronchial lumen is more most of by 26 than the tip tracheae branch Neighborhood territory pixel point gray value is low, and any neighbour in 1 26 neighborhood territory pixel point for meeting condition of selection is only needed in comparing every time Domain pixel, specific rules are set as：

    1065：In ith iteration, first seed point v of queue is taken out_i, search the v_i26 neighborhood territory pixel points in it is all not The pixel of labelSelection meets the neighborhood territory pixel point of the tip tracheae branch degree of membership ruleBy institute It statesAs new seed point v_i+1Add in segmentation queue；

    1066：Repeat described 1065；If described 1065 have completed without the generation of new seed point or described 1062 initial seed point The R times iteration stops the growth to the initial seed point；

    1067：To the S_endIn all initial seed points carry out growth segmentation using described 1065, complete to the tip gas The extraction of pipe branch.