CN109214397A - The dividing method of Lung neoplasm in a kind of lung CT image - Google Patents
The dividing method of Lung neoplasm in a kind of lung CT image Download PDFInfo
- Publication number
- CN109214397A CN109214397A CN201811073727.7A CN201811073727A CN109214397A CN 109214397 A CN109214397 A CN 109214397A CN 201811073727 A CN201811073727 A CN 201811073727A CN 109214397 A CN109214397 A CN 109214397A
- Authority
- CN
- China
- Prior art keywords
- lung
- image
- region
- lung neoplasm
- contour line
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06T2207/30064—Lung nodule
Abstract
The invention proposes a kind of dividing methods of Lung neoplasm in lung CT image, comprising the following steps: (1) obtains pulmonary parenchyma image comprising the pulmonary parenchyma contour line including Lung neoplasm from extracting in lung images；(2) the pulmonary parenchyma image in step (1) is split the region of doubtful Lung neoplasm using clustering algorithm；(3) interference of blood vessel or tracheae is excluded using the part of the nearly long strip type of shape in image circularity method removal region to the region of step (2) doubtful Lung neoplasm；(4) according to the size in the region of doubtful Lung neoplasm in step (3), Gaussian template is constructed, the higher region of related coefficient is selected to complete the extraction of Lung neoplasm.The present invention considers the features of shape of solitary pulmonary nodule and the feature of the image grayscale Distribution value formed by the variation of lung CT value, reduces the error of Lung neoplasm segmentation result.
Description
Technical field
The present invention relates to technical field of image segmentation, a kind of dividing method of Lung neoplasm in lung CT image is particularly related to.
Background technique
Lung cancer is lethal high-incidence reason in cancer, because the asymptomatic display of its early stage is so that disease is difficult to be detected, when
Cancer is transferred when disease is found, this makes lung cancer difficult to treat, and lung cancer illness early stage often has pulmonary nodule symptom, such as
Modern Lung neoplasm detection mostly uses Thoracic CT scan, and scanning result has the features such as image clearly, accurate, but the detection of a patient
As a result up to a hundred figures are had, are finding upper a large amount of cost radiologist's energy, while long-time repeated work is also possible to produce
Raw mistaken diagnosis is failed to pinpoint a disease in diagnosis.
Computer-aided diagnosis technology is the common trait that focal part is summarized by research, at medical image
Reason, the technology that auxiliary doctor diagnoses the state of an illness.By this technology can by doctor from it is a large amount of repeat, uninteresting interpreting blueprints
It is freed in journey, and further increases diagnosis speed and precision, also strive for more treatment times for a large amount of sufferers.
The area-of-interest of lung CT image is Lung neoplasm, and early stage Lung neoplasm shows as solitary pulmonary nodule, the application more
Selection for this kind of tubercle is studied.
Summary of the invention
The invention proposes a kind of dividing methods of Lung neoplasm in lung CT image, it is contemplated that the shape of solitary pulmonary nodule
The feature of feature and the image grayscale Distribution value formed by the variation of lung CT value reduces the error of Lung neoplasm segmentation result.
The technical scheme of the present invention is realized as follows: in a kind of lung CT image Lung neoplasm dividing method, including with
Lower step:
(1) pulmonary parenchyma image is obtained comprising the pulmonary parenchyma contour line including Lung neoplasm from extracting in lung images；
(2) the pulmonary parenchyma image in step (1) is split the region of doubtful Lung neoplasm using clustering algorithm；
(3) to the region of step (2) doubtful Lung neoplasm using the portion of the nearly long strip type of shape in image circularity method removal region
Point, exclude the interference of blood vessel or tracheae；
(4) according to the size in the region of doubtful Lung neoplasm in step (3), Gaussian template is constructed, selects related coefficient higher
Region complete Lung neoplasm extraction.
Further, in step (1), threshold value is calculated by iterative method or big law to lung images, carries out Threshold segmentation,
Obtain bianry image；The monolithic wheel profile in lung images is extracted by morphological method, then utilizes connected region mark point
Different part out, selects by profile, to obtain pulmonary parenchyma contour line.
Further, in contours extract, to obtain contour line complete and that interference is less, structural element in morphological method
The shape set is rectangular, and the 1/120 of size approximation lung images length.
Further, closed operation twice is carried out to monolithic wheel profile, filling tiny gap is connected to contour line, then right
Lung images carry out 8 connected region marks.
Further, the method for pulmonary parenchyma contour line is obtained, comprising the following steps:
1) judge on bianry image middle row pixel, first point of value from left to right or from right to left；
If 2) value is 1, the point that the second segment of this line from left to right or from right to left is 1 on profile diagram must be lung
Point on substantive contour line；
If 3) value is 0, the point that profile diagram first segment is 1 must be the point on pulmonary parenchyma contour line；
4) position of this two o'clock in the picture is found out, corresponding connected domain marked content finds out pulmonary parenchyma contour line.
Further, after obtaining pulmonary parenchyma contour line, first inside is filled, it is external extra to be removed using closing operation of mathematical morphology
Contour line, outside is modified, after modifying outside, external connected region is only chosen, then removes internal interference, obtain
Complete pulmonary parenchyma image.
Further, in step (2), classified using fuzzy C-means clustering method (FCM) to gray value, by gray scale
Value constantly updates cluster centre as standard, and cluster classification is set as 4, and fuzzy coefficient is set as 2, extracts the highest portion of gray value in figure
It is divided into the region of doubtful Lung neoplasm.
Further, in step (3), circularity image method the following steps are included:
1) regard the number of pixels in image as area, the pixel of connected region is found by find function in matlab
Number is denoted as m；
2) by the location information of the point in connected region, the column vector maximum value y of connected region is found outmaxMinimum value
ymin, row vector maximum value xmaxMinimum value xmin, select the maximum number of two groups of differences as diameter of a circle, reference area is denoted as M；
3) m/M and 1 is closer, and mark connected region is carried out closer to circle, each fraction highest to the gray value of extraction
It calculates, the obvious poor part of removal circularity.
Beneficial effects of the present invention: the present invention has studied Fuzzy C-Means Cluster Algorithm, is dropped using the concept of wherein fuzziness
The error rate of low segmentation result, because soft clustering algorithm can remove the marginal portion for being difficult to be classified in segmentation result；Lung
The general gray value of tubercle is higher, constantly updates cluster centre for gray value as standard, is divided by similar iterative calculation
As a result.The present invention divides lung CT image after having tested out relevant parameter, and obtaining to be the portion of tubercle
Point.Due to the high region of gray value it is also possible that blood vessel, and solitary pulmonary nodule is presented subcircular more, the nearly bar shaped of blood vessel, this
Invention is to be shown and stored this point by picture element matrix in a computer using image, using number of pixels as area into
Row calculates, and by connected domain and the ratio between the number of pixels of circle that the region surrounds completely can be judged each section and circular difference
Away from the part of the nearly long strip type of shape, excludes the interference of blood vessel in removal Probability Area.
In order to improve the accuracy of experimental result, the present invention is labeled using template matching method.Lung neoplasm part, by
Different piece penetrability is different during CT scan, in Lung neoplasm part, due to different piece penetrability during CT scan
The Gaussian Profile centered on centre is substantially presented in Lung neoplasm part, by the gray value of image to Lung neoplasm part in difference
The research of distribution judges its approximate Gaussian distribution, and according to Lung neoplasm size, selection construction Gaussian template is compared, and selects phase
The higher region of relationship number is completed Lung neoplasm and is extracted, and algorithm is simple and parameter is easily set.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the dividing method flow chart of Lung neoplasm in a kind of lung CT image of the present invention；
Fig. 2 is the bianry image of three kinds of methods in step (1)；
Fig. 3 is the acquisition image of monolithic wheel profile in step (1)；
Fig. 4 is the partial enlarged view of (b) and (c) after expansion after corroding in Fig. 3；
Fig. 5 is the acquisition image of pulmonary parenchyma contour line in step (1)；
Fig. 6 is the image after Fig. 5 (b) removal interference；
Fig. 7 is the pulmonary parenchyma image obtained in step (1)；
Fig. 8 is the image in step (2) after FCM cluster；
Fig. 9 is the extraction image in doubtful Lung neoplasm region in step (2)；
Figure 10 is the image that removal circularity crosses lower part in step (3)；
Figure 11 is 3 kinds of different size of Gaussian template images in step (4)；
Figure 12 is the image of Gaussian template matched indicia in step (4)；
Figure 13 is the image of dividing method standard Lung neoplasm of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under that premise of not paying creative labor
Embodiment shall fall within the protection scope of the present invention.
Data used in the present invention are LIDC-IDRI, this is by chest medical image files and corresponding diagnostic result group
At data set, the data set be by American science research institute National Cancer Institute initiate collect.
As shown in Figure 1, in a kind of lung CT image Lung neoplasm dividing method, comprising the following steps:
(1) it extracts from lung images comprising the pulmonary parenchyma contour line including Lung neoplasm
Step 1: image binaryzation
Since gray value gradient is larger on lung CT image boundary, can using global adaptive threshold, using iterative method,
Two-peak method and big law calculate separately threshold value T, carry out Threshold segmentation, obtain bianry image, bianry image refers to that gray scale only takes
The image of two probable values, two gray values taken are usually 0 and 1, and the point of value 1 corresponds to the point in target, value 0
Point correspond to the point in background.
As shown in Fig. 2, Two-peak method, iterative method and big law, which is respectively adopted, to lung CT original image (a) obtains threshold value and two-value
Change processing, (b) Two-peak method T=117, (c) iterative method T=89.6, (d) big law T=91.2, it follows that three kinds of methods obtain
The result gap arrived is little, but from the point of view of the threshold value that three kinds of methods obtain, iterative method is obviously more accurate than Two-peak method, anti-interference energy
Power is stronger, and big law result is not much different with Two-peak method, and on the segmentation result of lung CT image without influence, therefore the application is to lung
Portion's image calculates threshold value by iterative method or big law.
Step 2: extracting monolithic wheel profile
Morphology operations are one kind using bianry image as foundation, the figure realized by the set enumeration tree in mathematical morphology
As processing method, by setting the structural element of shape and size, morphology operations can be according to the shape and ruler in image
Very little geometrical characteristic similar with structural element remains required part, remaining feature of image is removed.Corroding with expansion is
The fundamental operation of morphology operations, opening operation are first to corrode the operation expanded afterwards, and closed operation is first to expand the operation of post-etching.It opens
Operation is used for the profile of smooth object target, removes tiny protrusion burr, truncation is got too close to but and disjunct part；It closes
Operation supplies the notch of profile for eliminating cavity and gully tiny in image.
Bright part operates in the bianry image obtained using expansion or corrosion to the first step, as shown in figure 3,
(a) bianry image (b) corrodes obtained profile, (c) expands obtained profile, removes original image after carrying out expansion process to image
It can be obtained profile in image from removal in original image again after carrying out corrosion treatment to image by picture.
Complete monolithic wheel profile is obtained by morphological method, it is mostly important to the selection of structural element, according to reality
Image size and content, by testing repeatedly, the shape set is rectangular, and the 1/120 of size approximation lung images length, from figure
3 as can be seen that the contour line obtained under this case is more complete and interference is less.
Step 3: extracting pulmonary parenchyma contour line
After obtaining monolithic wheel profile whole in image, it is only necessary to determine the contour line of pulmonary parenchyma part can be partitioned into
Pulmonary parenchyma.
I, connected domain marks
In bianry image, if the value of background area pixel is 0, the pixel value of target area is 1.According to pixel to a width
According to from left to right, route from the top down is scanned image, while marking pixel currently scanned, check it with
Whether several neighbor pixels being scanned between it have connectivity.
The selection of connected domain generally has 4 connections to be connected to 8, it is assumed that the pixel value X that Current Scan arrives is 0, is scanned for next
A pixel value checks adjacent pixel values scanned in its connected region, the two or four pictures if this pixel value is 1
Plain value will appear four kinds of possible outcomes:
1) adjacent pixel value is all 0.Illustrate the beginning of the stylish connected domain in the position, sets a new mark value.
2) adjacent pixel value any one be 1.Illustrate the pixel and adjacent pixel in the same connected domain, and they
It marks identical.
3) adjacent pixel value is all 1 and label is identical.Illustrate the pixel and adjacent pixel in the same connected domain, and
They mark identical.
4) adjacent pixel value is 1 and label is not identical.Current pixel marks wherein lesser mark value.
It is scanned by judgement and then since another side, until returning to the beginning pixel in region.Recall every time all
Aforementioned four step is executed to be judged.
Such step can guarantee that all connected domains can be labeled out, pass through different labels later
Different connected domains is obtained, contour line can be obtained by the connected domain of acquisition, but directly acquire connected domain meeting on contour line
Cause profile imperfect, some thins will disconnect, therefore carry out closed operation twice to monolithic wheel profile, fill tiny seam
Gap is connected to contour line.As shown in figure 4, (a) corrodes obtained partial contour, i.e. the partial enlarged view of (b) in Fig. 3 (b) expands
Obtained partial contour, the i.e. partial enlarged view of (c) in Fig. 3, the contour line for expanding generation generate in continuity not as good as corrosion
Contour line, while expanding method can bring more interference, therefore the overall profile that only selective etching obtains to segmentation result
The mark of line progress connected region.
To treated in second step, image carries out 8 connected component labelings, and different connected regions is designated as different digital
Afterwards, it is only necessary to learn the number where pulmonary parenchyma contour line it can be learnt that desired contour line.
II, the selection of pulmonary parenchyma contour line
The contour line of pulmonary parenchyma is obtained by exclusive PCR in numerous contour lines, to find being total to for all pulmonary parenchyma contour lines
Same part.For the image after connected region mark, the method for judging to obtain pulmonary parenchyma contour line, comprising the following steps:
1) judge on bianry image middle row pixel, first point of value from left to right or from right to left；
If 2) value is 1, the point that the second segment of this line from left to right or from right to left is 1 on profile diagram must be lung
Point on substantive contour line；
If 3) value is 0, the point that profile diagram first segment is 1 must be the point on pulmonary parenchyma contour line；
4) position of this two o'clock in the picture is found out, corresponding connected domain marked content finds out pulmonary parenchyma contour line.
Position of this two o'clock in the image after connected region mark is found out, the content of corresponding connected domain standard finds out lung reality
Matter contour line, as shown in figure 5, (a) monolithic wheel profile, the pulmonary parenchyma contour line (b) selected.
Step 4: pulmonary parenchyma is divided
The pulmonary parenchyma contour line selected in the third step, has some interference sometimes because and contour line from obtained it is close or
There is connected part to be retained, at this moment removes this part, while this being partially filled with.
The processing of 8 connected regions is made to pulmonary parenchyma profile diagram in the third step, on centerline than the point on pulmonary parenchyma profile
Again to the point in the close point in center necessarily pulmonary parenchyma region, two o'clock coordinate is similarly obtained, corresponding position finds out connected region
Reference numerals, corresponding pixel are the part of pulmonary parenchyma.
As shown in Fig. 5 (b), pulmonary parenchyma contour line can be seen that there are a tubercle and profile phase in inside there are two main interference
It is even external to have an interference.As shown in fig. 6, (a) removes external disturbance, internal interference (b) is removed, it, first will be interior as shown in Fig. 6 (a)
Portion's filling removes external extra contour line using morphology opening and closing operation, to achieve the purpose that be modified outside；Such as figure
After 6 (b) have modified outside, external connected region is only chosen, then internal interference can remove, to there is complete pulmonary parenchyma
Image；Corresponding region is intercepted out from original image, obtains the image in pulmonary parenchyma region, as shown in fig. 7, (a) is original image, (b)
For pulmonary parenchyma image.
(2) pulmonary parenchyma in step (1) is split the region of doubtful Lung neoplasm using clustering algorithm
The advantages of the application uses fuzzy C-means clustering method (FCM), this method essentially consists in the setting for avoiding legal system
And it is difficult to the multiple-limb segmentation problem solved by Threshold segmentation, when there is fuzzy or uncertain feature in image, can lead to
Cross the solution of FCM algorithm.
Need to provide initial degree of membership in FCM algorithm, subordinating degree function is under the jurisdiction of the journey of set A for description object x
Degree, is denoted as μ A (x), domain is point all in set A, and value range is [0,1], wherein when μ A (x)=1 item indicates x
∈ A, that is, x is under the jurisdiction of set A completely.This subordinating degree function has been defined on space X={ x } and has just defined fuzzy set A, or
Person is called fuzzy subset, and for limited object, fuzzy set can be indicated are as follows:
A={ (μ A (xi), xi)|xi∈X} (3-1)
The concept of fuzzy set makes cluster become soft, regards the cluster that production is clustered in clustering problem as fuzzy set
It closes, then its degree of membership is just inner in [0,1].
FCM determines that each data point belongs to the degree of some cluster by degree of membership, it is by n vector xi(i=1,
2 ..., n) it is divided into c ambiguity group, is classified at a distance from center according to sample, constantly update every group of cluster centre, make
The cost function of its non-similarity index obtains minimum value.Subject Matrix U value is in section [0,1], in addition normalization provides, one
The sum of all degrees of membership that a data are concentrated are equal to 1:
So, the cost function formula of FCM is exactly:
Here uijBetween [0,1]；ciIt is the cluster centre of ambiguity group I, dij=| | ci-xj| | it is in ith cluster
Euclidean distance between the heart and j-th of data point, it is a Weighted Index.
New objective function is constructed, the necessary condition for making (3-3) formula can achieve minimum value is sought:
To all input parameter derivations, formula (3-3) reaches the smallest necessary condition and is
With
Then formula (3-5) is cluster centre, and formula (3-6) is degree of membership, according to above-mentioned two necessary condition, it can be seen that mould
Pasting C means clustering algorithm is iterative process.
FCM algorithm flow
According to above-mentioned related content, for data set X, it is made of N number of L dimensional vector, to be classified as C class, first
Want customized subordinated-degree matrixSpecific step is as follows later:
1) setting needs the classification number C and fuzzy coefficient m being divided into；
2) subordinated-degree matrix is initialized, determines that matrix meets formula (3-2)；
3) by formula (3-5), the calculated result of formula (3-6) constantly updates subordinated-degree matrix；
4) subordinated-degree matrix of iteration is made comparisons with matrix norm, if | | U (t)-U (t-1) | | < ε, iteration stopping are no
It then returns 3).
Pass through these steps, it can be seen that first to set two parameters using FCM algorithm, one is clusters number C, another
A is fuzzy parameter m.In setting, it is desirable that C is far smaller than the total number for clustering sample, and C > 1；It is required that m wants moderate, because of m
Value determine the softness of algorithm, the excessive accuracy that will affect cluster result, the too small meaning for losing setting.
Output result one of algorithm is the vector of C cluster centre point, the other is the fuzzy partition matrix of C × N, it
Illustrate that each sample point to the degree of membership of each class, clusters sample point according to the maximum principle of degree of membership in matrix,
What cluster centre indicated is the average characteristics of classification where the center, can be used as such representative point.From the derivation of algorithm
It can be concluded that, the data for meeting normal distribution are relatively suitble to be clustered with FCM algorithm, in addition, algorithm is to isolated point in journey
Sensitive.
FCM arithmetic result
Note simulated program is set according to algorithm flow, sets classification number and fuzzy coefficient.Pulmonary parenchyma figure after input segmentation
Picture, according to feature of image, cluster is set to 4, and fuzzy coefficient is set as 2, as shown in figure 8,4 classes of setting are clustered according to gray value
Cluster result is respectively background, nonuseable part in pulmonary parenchyma, the interference sections such as tracheae, doubtful tubercle part from low to high.It will figure
The middle highest extracting section of gray value comes out, as a result as shown in Figure 9.
(3) to the region of step (2) doubtful Lung neoplasm using the portion of the nearly long strip type of shape in image circularity method removal region
Point, exclude the interference of blood vessel or tracheae
For the part that step (2) extracts, since wherein there are also the interference of pulmonary vascular or tracheae, by each company
The circularity judgement in logical region, can effectively remove the part of elongated.
If a connected region is rested entirely in a circle, the area that the area of this connected region accounts for this circle is got over
It is more, can approximation to regard this connected region as round.Image circularity method the following steps are included:
1) in order to which convenience of calculation regards the number of pixels in image as area, company is found by find function in matlab
The number of pixels in logical region, is denoted as m；
2) by the location information of the point in connected region, the column vector maximum value y of connected region is found outmaxMinimum value
ymin, row vector maximum value xmaxMinimum value xmin, select the maximum number of two groups of differences as diameter of a circle, reference area is denoted as M；
3) m/M and 1 is closer, the closer circle of mark connected region.
By the above-mentioned means, each fraction highest to the gray value of extraction calculates, removal circularity is obviously poor
Part, the results are shown in Figure 10, the image in the doubtful Lung neoplasm region (a), is (b) circularity treated image.
(4) according to the size in the region of doubtful Lung neoplasm in step (3), Gaussian template is constructed, selects related coefficient higher
Region complete Lung neoplasm extraction
The method of Gaussian template is constructed (see document: lung's suspected abnormality Region Segmentation Algorithm research based on ct images
[D] Wuwei, Northeastern University, 2008, the 45-47 pages): using traditional template sometimes cannot correct selection target, in order to solve
Small object image Gaussian template is normalized in this problem.
According to imaging theory, the imaging Gaussian function approximate representation of the Small object for being compared:
Wherein g0Indicate the gray scale of target picture central point, (x0, y0) be central point position, σxAnd σyIt is the direction x and y respectively
The standard deviation in direction；Due to pulmonary nodule very little, so also very little, the structure of transvers plate of relative configurations also compare tubercle radius out
It is small, if: Gk=Sk(g (x, y)), S in formulakThe uniform sampling function for being k for sampling number；The corresponding domain sampling of template are as follows:
Xk=[- xn-xn-1-xn-2…0…xn-2xn-1xn] (3-8)
If producing 3 × 3 Gaussian template, order can be turned to matrix, then mutually multiply, 3 × 3 matrixes can be obtained
(i.e. 3 × 3 Gaussian templates).
Since the size of Lung neoplasm is indefinite, appropriate adjusting size is needed in the selection of Gaussian template, for large-scale tubercle,
As shown in figure 11, image is typically chosen 30 × 30 template (a), the template (b) of medium-sized selection 10 × 10, small-sized selection 3
× 3 template (c).
Template matching is carried out in numerous doubtful tubercles are to favored area, selects the wherein highest region of similarity, as
Lung neoplasm region, as shown in figure 12, the Lung neoplasm image of (a) doctor mark, (b) image of the application matched indicia.
Using the above method, lung CT image is randomly selected from LIDC data set, carries out the segmentation of Lung neoplasm, as a result as schemed
Shown in 13, (a) marks original image, (b) pulmonary parenchyma extracted, and (c) Lung neoplasm chooses result.
As shown in figure 13, it can be seen that obtained preferable Lung neoplasm segmentation result.Pass through form in the segmentation of pulmonary parenchyma
Corrosion and expansive working, available relative smooth and the profile and edge that are connected to reduce Lung neoplasm and are divided
A possibility that pulmonary parenchyma.The features of shape of solitary pulmonary nodule is mainly considered in the cutting procedure of Lung neoplasm and by CT value
Change the feature of the image grayscale Distribution value formed, the results showed that, this method is feasible in the selection of Lung neoplasm.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (8)
1. the dividing method of Lung neoplasm in a kind of lung CT image, it is characterised in that: the following steps are included:
(1) pulmonary parenchyma image is obtained comprising the pulmonary parenchyma contour line including Lung neoplasm from extracting in lung images；
(2) the pulmonary parenchyma image in step (1) is split the region of doubtful Lung neoplasm using clustering algorithm；
(3) to the region of step (2) doubtful Lung neoplasm using the part of the nearly long strip type of shape in image circularity method removal region, row
Except the interference of blood vessel or tracheae；
(4) according to the size in the region of doubtful Lung neoplasm in step (3), Gaussian template is constructed, selects the higher area of related coefficient
The extraction of domain completion Lung neoplasm.
2. the dividing method of Lung neoplasm in a kind of lung CT image according to claim 1, it is characterised in that: step (1)
In, threshold value is calculated to lung images by iterative method, Threshold segmentation is carried out, obtains bianry image；It is extracted by morphological method
Then monolithic wheel profile in lung images separates different parts using connected region mark, selects by profile, thus
To pulmonary parenchyma contour line.
3. the dividing method of Lung neoplasm in a kind of lung CT image according to claim 2, it is characterised in that: contours extract
In, to obtain contour line complete and that interference is less, for the shape that structural element is set in morphological method to be rectangular, size is approximate
The 1/120 of lung images length.
4. the dividing method of Lung neoplasm in a kind of lung CT image according to claim 2 or 3, it is characterised in that: to whole
Body contour line carries out closed operation twice, and filling tiny gap is connected to contour line, then carries out 8 connections to overall profile line chart
Area marking.
5. the dividing method of Lung neoplasm in a kind of lung CT image according to claim 4, it is characterised in that: obtain lung reality
The method of matter contour line the following steps are included:
1) judge on bianry image middle row pixel, first point of value from left to right or from right to left；
If 2) value is 1, the point that the second segment of this line from left to right or from right to left is 1 on profile diagram must be pulmonary parenchyma
Point on contour line；
If 3) value is 0, the point that profile diagram first segment is 1 must be the point on pulmonary parenchyma contour line；
4) position of this two o'clock in the picture is found out, corresponding connected domain marked content finds out pulmonary parenchyma contour line.
6. the dividing method of Lung neoplasm in a kind of lung CT image according to claim 2, it is characterised in that: it is real to obtain lung
After matter contour line, first inside is filled, external extra contour line is removed using closing operation of mathematical morphology, outside is modified,
After modifying outside, external connected region is only chosen, internal interference is then removed, obtains complete pulmonary parenchyma image.
7. the dividing method of Lung neoplasm in a kind of lung CT image according to claim 2, it is characterised in that: step (2)
In, classified using fuzzy C-means clustering method to gray value, cluster classification is set as 4, and fuzzy coefficient is set as 2, and it is real to extract lung
The highest part of gray value is the region of doubtful Lung neoplasm in matter.
8. the dividing method of Lung neoplasm in a kind of lung CT image according to claim 2, it is characterised in that: step (3)
In, circularity image method the following steps are included:
1) regard the number of pixels in image as area, the pixel of connected region is found by find function in matlab
Number, is denoted as m；
2) by the location information of the point in connected region, the column vector maximum value y of connected region is found outmaxMinimum value ymin, row
Vector maximization xmaxMinimum value xmin, select the maximum number of two groups of differences as diameter of a circle, reference area is denoted as M；
3) m/M and 1 is closer, and mark connected region is calculated closer to circle, each fraction highest to the gray value of extraction,
Remove the obvious poor part of circularity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811073727.7A CN109214397A (en) | 2018-09-14 | 2018-09-14 | The dividing method of Lung neoplasm in a kind of lung CT image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811073727.7A CN109214397A (en) | 2018-09-14 | 2018-09-14 | The dividing method of Lung neoplasm in a kind of lung CT image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109214397A true CN109214397A (en) | 2019-01-15 |
Family
ID=64984031
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811073727.7A Pending CN109214397A (en) | 2018-09-14 | 2018-09-14 | The dividing method of Lung neoplasm in a kind of lung CT image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109214397A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816667A (en) * | 2019-01-21 | 2019-05-28 | 杭州英库医疗科技有限公司 | A kind of Lung neoplasm dividing method based on CT images |
| CN110898338A (en) * | 2019-12-16 | 2020-03-24 | 福建师范大学 | Intelligent photodynamic therapy illumination system based on LED array |
| CN110930381A (en) * | 2019-11-19 | 2020-03-27 | 南京大树智能科技股份有限公司 | Tobacco flake shape determination method for improving production quality of fine cigarettes |
| CN111105427A (en) * | 2019-12-31 | 2020-05-05 | 佛山科学技术学院 | Lung image segmentation method and system based on connected region analysis |
| CN111145185A (en) * | 2019-12-17 | 2020-05-12 | 天津市肿瘤医院 | Lung parenchyma segmentation method for extracting CT image based on clustering key frame |
| CN113112467A (en) * | 2021-04-06 | 2021-07-13 | 上海深至信息科技有限公司 | Plan drawing marking system |
| CN113129317A (en) * | 2021-04-23 | 2021-07-16 | 广东省人民医院 | Lung lobe automatic segmentation method based on watershed analysis technology |
| CN113657559A (en) * | 2021-10-18 | 2021-11-16 | 广州天鹏计算机科技有限公司 | Chest scanning image classification method based on machine learning |
| CN115272376A (en) * | 2022-09-27 | 2022-11-01 | 山东鑫科来信息技术有限公司 | Floating polishing head control method based on machine vision |
| CN116681701A (en) * | 2023-08-02 | 2023-09-01 | 青岛市妇女儿童医院（青岛市妇幼保健院、青岛市残疾儿童医疗康复中心、青岛市新生儿疾病筛查中心） | Children lung ultrasonic image processing method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020028008A1 (en) * | 2000-09-07 | 2002-03-07 | Li Fan | Automatic detection of lung nodules from high resolution CT images |
| CN102360495A (en) * | 2011-10-19 | 2012-02-22 | 西安电子科技大学 | Pulmonary nodule segmentation method based on average intensity projection and translation gaussian model |
| CN102429679A (en) * | 2011-09-09 | 2012-05-02 | 华南理工大学 | Computer-assisted emphysema analysis system based on chest CT (Computerized Tomography) image |
| CN103824295A (en) * | 2014-03-03 | 2014-05-28 | 天津医科大学 | Segmentation method of adhesion vascular pulmonary nodules in lung CT (computed tomography) image |
| CN105469419A (en) * | 2016-01-07 | 2016-04-06 | 于翠妮 | Pulmonary nodule detection method based on CT image |
-
2018
- 2018-09-14 CN CN201811073727.7A patent/CN109214397A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020028008A1 (en) * | 2000-09-07 | 2002-03-07 | Li Fan | Automatic detection of lung nodules from high resolution CT images |
| CN102429679A (en) * | 2011-09-09 | 2012-05-02 | 华南理工大学 | Computer-assisted emphysema analysis system based on chest CT (Computerized Tomography) image |
| CN102360495A (en) * | 2011-10-19 | 2012-02-22 | 西安电子科技大学 | Pulmonary nodule segmentation method based on average intensity projection and translation gaussian model |
| CN103824295A (en) * | 2014-03-03 | 2014-05-28 | 天津医科大学 | Segmentation method of adhesion vascular pulmonary nodules in lung CT (computed tomography) image |
| CN105469419A (en) * | 2016-01-07 | 2016-04-06 | 于翠妮 | Pulmonary nodule detection method based on CT image |
Non-Patent Citations (5)
| Title |
|---|
| 刘凤民等: "《岩心扫描及应用》", 30 September 2010 * |
| 杨恩烈: "《形位公差与技术测量》", 30 November 1981 * |
| 查宇飞等: "《视频目标跟踪方法》", 31 July 2015 * |
| 高园园 等: "肺结节检测算法研究", 《临床医学信息学》 * |
| 魏颖 等: "面向肺癌CAD的CT图像疑似病灶检测算法", 《仪器仪表学报》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816667A (en) * | 2019-01-21 | 2019-05-28 | 杭州英库医疗科技有限公司 | A kind of Lung neoplasm dividing method based on CT images |
| CN110930381A (en) * | 2019-11-19 | 2020-03-27 | 南京大树智能科技股份有限公司 | Tobacco flake shape determination method for improving production quality of fine cigarettes |
| CN110898338A (en) * | 2019-12-16 | 2020-03-24 | 福建师范大学 | Intelligent photodynamic therapy illumination system based on LED array |
| CN111145185A (en) * | 2019-12-17 | 2020-05-12 | 天津市肿瘤医院 | Lung parenchyma segmentation method for extracting CT image based on clustering key frame |
| CN111145185B (en) * | 2019-12-17 | 2023-12-22 | 天津市肿瘤医院 | Lung substance segmentation method for extracting CT image based on clustering key frame |
| CN111105427A (en) * | 2019-12-31 | 2020-05-05 | 佛山科学技术学院 | Lung image segmentation method and system based on connected region analysis |
| CN111105427B (en) * | 2019-12-31 | 2023-04-25 | 佛山科学技术学院 | Lung image segmentation method and system based on connected region analysis |
| CN113112467B (en) * | 2021-04-06 | 2023-04-07 | 上海深至信息科技有限公司 | Plan drawing labeling system |
| CN113112467A (en) * | 2021-04-06 | 2021-07-13 | 上海深至信息科技有限公司 | Plan drawing marking system |
| CN113129317A (en) * | 2021-04-23 | 2021-07-16 | 广东省人民医院 | Lung lobe automatic segmentation method based on watershed analysis technology |
| CN113657559A (en) * | 2021-10-18 | 2021-11-16 | 广州天鹏计算机科技有限公司 | Chest scanning image classification method based on machine learning |
| CN115272376B (en) * | 2022-09-27 | 2022-12-20 | 山东鑫科来信息技术有限公司 | Floating polishing head control method based on machine vision |
| CN115272376A (en) * | 2022-09-27 | 2022-11-01 | 山东鑫科来信息技术有限公司 | Floating polishing head control method based on machine vision |
| CN116681701A (en) * | 2023-08-02 | 2023-09-01 | 青岛市妇女儿童医院（青岛市妇幼保健院、青岛市残疾儿童医疗康复中心、青岛市新生儿疾病筛查中心） | Children lung ultrasonic image processing method |
| CN116681701B (en) * | 2023-08-02 | 2023-11-03 | 青岛市妇女儿童医院（青岛市妇幼保健院、青岛市残疾儿童医疗康复中心、青岛市新生儿疾病筛查中心） | Children lung ultrasonic image processing method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109214397A (en) | The dividing method of Lung neoplasm in a kind of lung CT image | |
| Wang et al. | Weakly supervised deep learning for whole slide lung cancer image analysis | |
| CN107886514B (en) | Mammary gland molybdenum target image lump semantic segmentation method based on depth residual error network | |
| CN107506761B (en) | Brain image segmentation method and system based on significance learning convolutional neural network | |
| Man et al. | Classification of breast cancer histopathological images using discriminative patches screened by generative adversarial networks | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| CN107369160A (en) | A kind of OCT image median nexus film new vessels partitioning algorithm | |
| CN112365464B (en) | GAN-based medical image lesion area weak supervision positioning method | |
| CN109300113B (en) | Pulmonary nodule auxiliary detection system and method based on improved convex hull method | |
| Wang et al. | Weakly supervised learning for whole slide lung cancer image classification | |
| CN111028206A (en) | Prostate cancer automatic detection and classification system based on deep learning | |
| CN110390665B (en) | Knee joint disease ultrasonic diagnosis method based on deep learning multichannel and graph embedding method | |
| CN112101451A (en) | Breast cancer histopathology type classification method based on generation of confrontation network screening image blocks | |
| CN111767952B (en) | Interpretable lung nodule benign and malignant classification method | |
| CN114972341B (en) | WSI image classification method, system and medium based on Bayesian assisted learning | |
| Balamurugan et al. | Brain tumor segmentation and classification using hybrid deep CNN with LuNetClassifier | |
| CN112215842B (en) | Malignant nodule edge detection image processing method based on benign thyroid template | |
| Chen et al. | Pulmonary nodule detection on chest radiographs using balanced convolutional neural network and classic candidate detection | |
| NL2029876B1 (en) | Deep residual network-based classification system for thyroid cancer computed tomography (ct) images | |
| Yonekura et al. | Improving the generalization of disease stage classification with deep CNN for glioma histopathological images | |
| Sreelakshmy et al. | An automated deep learning model for the cerebellum segmentation from fetal brain images | |
| Ananth et al. | Enhancing segmentation approaches from Gaussian mixture model and expected maximization to super pixel division algorithm | |
| Abdelhafeez et al. | Skin Cancer Detection using Neutrosophic c-means and Fuzzy c-means Clustering Algorithms | |
| US20220028068A1 (en) | Multi-scale tumor cell detection and classification | |
| Li et al. | Structure convolutional extreme learning machine and case-based shape template for HCC nucleus segmentation |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190115 |
|
| RJ01 | Rejection of invention patent application after publication |