CN104021539B - System for detecting tumour automatically in ultrasound image - Google Patents
System for detecting tumour automatically in ultrasound image Download PDFInfo
- Publication number
- CN104021539B CN104021539B CN201310064300.1A CN201310064300A CN104021539B CN 104021539 B CN104021539 B CN 104021539B CN 201310064300 A CN201310064300 A CN 201310064300A CN 104021539 B CN104021539 B CN 104021539B
- Authority
- CN
- China
- Prior art keywords
- candidate region
- tumor
- tumour
- region
- ultrasound image
- 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.)
- Expired - Fee Related
Links
Landscapes
- Ultra Sonic Daignosis Equipment (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
A kind of system for detecting tumour automatically in ultrasound image is provided.The system comprises: candidate region detection device, for including at least one candidate region of tumour from ultrasound image detection based on deformable part model (DPM)；Tumor region positioning device determines tumor region at least one candidate region described in detect；Tumor's profiles separator, for detecting tumour by the profile for separating tumour based on determining tumor region.By the system, can in the case where not needing artificially to participate in, effectively detected from ultrasound image include tumour region, and relatively accurately isolate the profile of tumour on this basis, using as tumour make a definite diagnosis according to one of.
Description
Technical field
The present invention relates to Medical Image Processing more particularly to it is a kind of tumor section is detected from ultrasound image be
System.
Background technique
In modern medicine, ultrasonic imaging is the weight for diagnosing various tumours (for example, the thoracic cavities such as tumor of breast tumour)
Means are wanted, because of ultrasonic examination relatively convenient, human body is not invaded and cost is relatively low.However, the doctor of image department needs for every
Width ultrasound image is artificially labeled, to reflect the image characteristics of corresponding organ, as the image basis for judging tumour.So
And the result of ultrasonography of a large amount of patients is carried out to mark manually one by one and needs to take a substantial amount of time, human cost is higher.
Therefore, people focus on the scheme for detecting tumour automatically from ultrasound image.For example, Drukker, K. et al.
In " Computerized lesion detection on breastultrasound ", Med.Phys., 29 (7): 1438-46
(2002) it proposes that diseased region is almost more darker than background area to some extent in, and tumour is generated based on this stringent hypothesis
Region.
However, in the automatic detection scheme of existing tumour for example described above, due to the figure of ultrasonic testing results itself
As second-rate, moreover, the form of tumour is extremely complex, not only mostly irregular figure, but also it is often associated with calcification, therefore,
It is difficult to effectively detect the region where tumour in ultrasound image.
Further, since whether relevant range is that tumour is also woven with substantial connection with the specific group where it, it is therefore, existing
Also it is easy to appear erroneous judgements in lesion detection technology.
Summary of the invention
The purpose of the present invention is to provide a kind of systems that tumour can be effectively detected automatically from ultrasound image.
According to an aspect of the present invention, a kind of system for detecting tumour automatically in ultrasound image is provided, comprising: wait
Constituency domain detection device, for including at least one time of tumour from ultrasound image detection based on deformable part model (DPM)
Favored area；Tumor region positioning device determines tumor region at least one candidate region described in detect；Tumour wheel
Wide separator, for detecting tumour by the profile for separating tumour based on determining tumor region.
Candidate region detection device can include swollen from ultrasound image detection by changing the ratio of width to height of the root template in DMP
At least one candidate region of tumor.
Tumor region positioning device can be used based on support vector machines (SVM) two-value classifier from described in detecting to
A few candidate region determines tumor region.
The feature vector of the two-value classifier can be based on the contextual feature of candidate region.
Described eigenvector may include at least one of following item: the DPM of candidate region detects score value, candidate region
All parts are relative to the intensity in the offset of root, candidate region between prospect and background in position and size, candidate region
Part coexists between difference, candidate region and the DPM detection highest candidate region of score value.
It is multiple that Multiple Kernel Learning (MKL) method can be used to be defined as the kernel function of two-value classifier for tumor region positioning device
The linear combination of basic kernel function.
Tumor region positioning device can be by the Polynomial kernel function progress of the RBF kernel function and three dimensions of three kinds of bandwidth
Linear combination, to be integrally trained for each of feature vector characteristic component and feature vector, to obtain two
It is worth the kernel function of classifier.
Tumor's profiles separator can be used Level Set Method separation tumour profile, wherein the contour curve of tumour with
The boundary of tumor region is iterated as initial curve.
Tumor's profiles separator can be constructed by maximizing the distance between the prospect of tumor image and background
The energy function used in Level Set Method.
According to another aspect of the present invention, a kind of system for test object automatic in ultrasound image is provided, comprising:
Candidate region detection device, for detecting at least one including object from ultrasound image based on deformable part model (DPM)
Candidate region；Subject area positioning device determines subject area at least one candidate region described in detect；Object
Profile separator, for the profile by separating object based on determining subject area come test object.
Candidate region detection device can include pair from ultrasound image detection by changing the ratio of width to height of the root template in DMP
At least one candidate region of elephant.
Subject area positioning device can be used based on support vector machines (SVM) two-value classifier from described in detecting to
A few candidate region determines subject area.
The feature vector of the two-value classifier can be based on the contextual feature of candidate region.
It is multiple that Multiple Kernel Learning (MKL) method can be used to be defined as the kernel function of two-value classifier for subject area positioning device
The linear combination of basic kernel function.
According to another aspect of the present invention, a kind of method for test object automatic in ultrasound image is provided, comprising:
At least one candidate region including object is detected from ultrasound image based on deformable part model (DPM)；From the institute detected
It states at least one candidate region and determines subject area；It is detected by separating the profile of object based on determining subject area pair
As.
According to the above exemplary embodiments, can effectively be examined from ultrasound image in the case where not needing artificially to participate in
Measure the region including tumour, and relatively accurately isolate the profile of tumour on this basis, using as tumour make a definite diagnosis according to
One of according to.
Detailed description of the invention
By with reference to the accompanying drawing to exemplary embodiment carry out detailed description, above and other objects of the present invention and
Feature will become apparent, in which:
Fig. 1 shows the block diagram of lesion detection system according to an exemplary embodiment of the present invention；
Fig. 2 shows the process flows according to an exemplary embodiment of the present invention that tumour is detected by lesion detection system；
Fig. 3 shows the example in the tumor candidate region detected according to an exemplary embodiment of the present；
Fig. 4 shows the example of tumor region determining according to an exemplary embodiment of the present；
Fig. 5 shows the example of the tumor's profiles separated according to an exemplary embodiment of the present；
Fig. 6 shows the block diagram of object detection systems according to an exemplary embodiment of the present invention；And
Fig. 7 show it is according to an exemplary embodiment of the present invention by object detection systems come the process flow of test object.
Specific embodiment
The embodiment of the present invention will now be described in detail, examples of the embodiments are shown in the accompanying drawings, wherein identical mark
Number identical component is referred to always.It will illustrate the embodiment, by referring to accompanying drawing below to explain the present invention.
Fig. 1 shows the block diagram of lesion detection system according to an exemplary embodiment of the present invention.As shown in Figure 1, according to this hair
The lesion detection system of bright exemplary embodiment includes: candidate region detection device 10, for being based on deformable part model
(DPM) from least one candidate region of ultrasound image detection including tumour；Tumor region positioning device 20, for from detecting
At least one described candidate region determine tumor region；Tumor's profiles separator 30, for by based on determining tumour
The profile of region disconnecting tumour detects tumour.Here, as an example, ultrasound image can indicate that the chest for tumor of breast surpasses
Acoustic inspection as a result, it should however be noted that: the present invention is not limited to this, lesion detection system according to an exemplary embodiment of the present invention
It can be applied to the lesion detection to other organs.
In above-mentioned lesion detection system, candidate region detection device 10 utilizes deformable part model (DPM) method, energy
The candidate region for being likely to occur tumour is enough effectively detected out, on this basis, then it is further by tumor region positioning device 20
It determines tumor region, and tumor's profiles is extracted by tumor's profiles separator 30, ultrasound is schemed so as to effectively complete
The automatic detection of tumour as in.
Hereinafter, by the example for carrying out lesion detection according to an exemplary embodiment of the present is described in conjunction with Fig. 2.
Fig. 2 shows the process flows according to an exemplary embodiment of the present invention that tumour is detected by lesion detection system.
Referring to Fig. 2, in step S100, deformable part model (DPM) is based on from ultrasound by candidate region detection device 10
Image detection includes at least one candidate region of tumour.
Particularly, as it is known by the man skilled in the art, in DPM method, there are the more rough root filters of a precision
Wave device and the more fine component filter of multiple precision, are accordingly, there are a root and multiple components, wherein p0Refer to
Show and is intended to Inertial manifolds for the root of detected entire object (for example, whole region where tumour), and p1、p2、…、pn
It indicates to be intended to subtly cover by each different components of detected object, correspondingly, candidate region detection device 10 can lead to
Equation (1) is crossed to calculate the DPM of each candidate region detection score value:
From equation (1) as can be seen that the DPM of candidate region detects score value score (p0,p1,...,pn) can mainly be expressed
Appearance score for root filter/component filter in respective position(wherein, Fi(pi) be root and
The filter of all parts responds, Gi(pi) be root and all parts feature) with deformatter (for indicating all parts
Deviate cost in the position for deviateing its anchor station)(wherein, xiAnd yiIndicate the pixel position of all parts
Set) between difference on this basis can be by the way that by real value offset term b(, wherein, b can be determined by testing) be added to institute
Difference is stated to obtain final DPM and detect score value.
It is preferred that can be improved to above-mentioned DPM method, to be more in line with the growth characteristics of tumour itself.?
That is since tumour itself comes in every shape and aspect ratio change is larger, and root template number corresponding with root filter in DPM
Less (usually 2 or 3), therefore, it is very easy to can't detect the tumour of different the ratio of width to height.For above situation, according to this hair
The candidate region detection device 10 of bright exemplary embodiment can be by changing the ratio of width to height of the root template in DMP come from ultrasound image
Detection includes at least one candidate region of tumour.
Particularly, it is being directed to root p0Calculate F0(p0) and G0(p0) dot product when, can be according to equation (2) by the dot product
As a result the maximum value being determined as in specific range of deflection:
From equation (2) as can be seen that by by F0(p0) and G0(p0) dot product be determined as within the scope of predetermined migration (its
In, τ indicates range of deflection, and the specific value that different applications determine τ can be directed to according to experiment) maximum point product value so that
Root template in DMP is no longer limited by fixed value, and can carry out a degree of adjustment within a predetermined range, to facilitate needle
The tumour of various the ratio of width to height is effectively detected.
It is preferred that in the manner described above obtain DMP detection the higher each candidate region of score value it
Afterwards, the candidate region of acquisition can further be screened, to obtain more structurally sound testing result, for example, can remove with
DPM detects the higher particular candidate region overlapping of score value and detects the lower candidate region of score value less than 50% DPM, to realize
To the Effective selection of candidate region.
Fig. 3 shows the example in the tumor candidate region detected according to an exemplary embodiment of the present.As shown in figure 3, phase
For the region of practical tumour and its place, several tumor candidate regions can be detected, wherein each rectangular area corresponds to
The candidate region detected in step S100.
Then, it in step S200, is determined and is swollen from least one candidate region detected by tumor region positioning device 20
Tumor region.
Particularly, tumor region positioning device 20 can be used the two-value classifier based on support vector machines (SVM) from inspection
At least one candidate region measured determines tumor region.
Various modes appropriate can be used to obtain the feature vector for the two-value classifier, for example, can be from correspondence
Described eigenvector is extracted in root template/component model position that root filter/component filter exports and size.
In addition, in order to further increase the accuracy of tumor region positioning, the two-value based on support vector machines (SVM)
The feature vector of classifier can be based on the contextual feature of candidate region.For example, equation (3) below can indicate the two-value point
The feature vector f of class device:
f=(s,r,offseti,I,SMAX,rMAX) (3)
Features described above vector f is related to following item: the DPM detection score value s of candidate region, indicating the position of candidate region and big
Offset offset of all parts relative to root in small vector r, candidate regioni, in candidate region between prospect and background
Intensity difference I, candidate region and DPM detection the highest candidate region of score value between coexist part (SMAX,rMAX).Pass through basis
The features described above vector f of exemplary embodiment of the present can be classified using the contextual feature of candidate region, thus not
The position of candidate region is only accounted for size, it is also contemplated that the perienchyma where candidate region.Further, since in same width
Usually there is no three or more tumours in ultrasound image, accordingly, it is considered to which current candidate region and DPM detection score value are highest
Part coexists classify and help to further strengthen the accuracy of classification between candidate region.
However, it will be understood by those skilled in the art that: feature vector f is not limited to items listed above, for example, feature
Vector f can only include in items listed above at least one of or it is multinomial, without including the whole listed in equation (3)
?.In addition, the associated vector of any contextual feature for embodying candidate region can be applied to feature vector f.
Based on as above determining feature vector f, tumor region positioning device 20 can be used Multiple Kernel Learning (MKL) method by two
The kernel function of value classifier is defined as the linear combination of multiple basic kernel functions.Particularly, in Multiple Kernel Learning MKL method,
It can learn the parameter of the multiple basic kernel function and the weight of each basic kernel function by training data.For example, swollen
Tumor regional positioning device 20 can be by the RBF(radial direction base of three kinds of bandwidth) Polynomial kernel function of kernel function and three dimensions carries out
Linear combination, to be integrally trained for each of feature vector f characteristic component and feature vector f, to obtain
The kernel function of two-value classifier.Here, the specific bandwidth of RBF kernel function can be determined according to experiment.
After the feature vector and kernel function of two-value classifier has been determined, tumor region positioning device 20 can be used corresponding
Two-value classifier determine tumor region from least one candidate region detected, that is, occur tumour in candidate region
The candidate region of maximum probability.
Fig. 4 shows the example of tumor region determining according to an exemplary embodiment of the present.As shown in figure 4, being included in a left side
Irregular figure in the rectangular area of top indicates actual tumour, and the rectangular area of lower right is tumor region positioning dress
Set 20 tumor regions determined in step S200.
Then, in step S300, by tumor's profiles separator 30 by separating tumour based on determining tumor region
Profile detects tumour.
Particularly, by using Level Set Method, the contour curve of tumour can be expressed as higher dimensional function (referred to as
Level set function) zero level collection.Here, level set function may make to be developed or changed according to the evolution equation that it is met
Generation (wherein, the boundary of the tumor region to determine in step S200 is iterated as initial curve), due to level set function
Constantly develop, so corresponding zero level collection is also constantly changing, when level set movements tend to be steady, develop and stop,
Obtain the final profile curve of tumour.
An exemplary embodiment of the present invention can be obtained by minimizing energy function represented by equation (4)
Obtain the contour curve of tumour:
In equation (4), C indicates the contour curve of tumour, wherein the contour curve C of tumour is with the boundary of tumor region
It is iterated as initial curve, c1Indicate the average strong of the pixel of the interior zone of the contour curve C of tumour in the t times iteration
Degree, c2Indicate the mean intensity of the pixel of the perimeter of the contour curve C of tumour in the t times iteration, inside (C) instruction is swollen
The interior zone of the contour curve C of tumor, outside (C) indicate the perimeter of the contour curve C of tumour, u0(X, t) instruction the
Positioned at the intensity of the pixel of X position in t iteration, Length (C) indicates the length of the contour curve C of tumour, Area (inside
(C)) area for indicating the interior zone of the contour curve C of tumour, in addition, λ1、λ2, μ, v be greater than or equal to 0 preset parameter.
From equation (4) as can be seen that the energy function used in Level Set Method is represented by the sum of following items: the profile of tumour is bent
Image pixel intensities otherness in the interior zone of line CThe outside area of the contour curve C of tumour
Image pixel intensities otherness in domainThe length factor μ Length of the contour curve C of tumour
(C), the area factor vArea (inside (C)) of the interior zone of the contour curve C of tumour.
In above-mentioned equation (4), energy function F (c1,c2, C) mainly reflect the contour curve that is iterated internal and
External global image statistics and characteristic.As an example, can be by parameter lambda1、λ2It is respectively set to 1, and sets 0 for parameter v.
On this basis, in order to further increase the accuracy of separation tumour, the characteristics of for tumor image, for example, swollen
Often cause the intensity of image uneven because of calcification among tumor, tumor's profiles separator 30 is preferably by swell
The distance between prospect and background of tumor image maximize to construct the energy function used in Level Set Method.
As an example, tumor's profiles separator 30 can be by adding foundation on the basis of the energy function of equation (4)
The regular terms of refined distance (Bhattacharyya distance) comes so that between the prospect and background of tumor image in Ba Ta is proper
Distance maximize, the energy function accordingly obtained is optimised for as shown in equation below (5):
E(C)=βF(c1,c2, C) and+(1- β) B (C) (5)
In the equation (5),Indicate the probability density function p of contour curve Cin(X) with
Probability density function pout(X) refined distance in Ba Ta between is proper, wherein pin(X) indicate the pixel for being located at X position in profile song
The probability density function of (that is, the prospect for being located at tumor image), p inside line Cout(X) indicate the pixel for being located at X position in profile
The probability density function of (that is, the background for being located at tumor image) outside curve C.In addition, β ∈ [0,1] for control Ba Ta it is proper in
The degree of participation of refined item B (C).
On this basis, in order to meet the attribute of symbolic measurement φAnd it avoids again initial
Change, it is contemplated that the energy function of peer-to-peer (5) increases additional penalty Rp(φ), to obtain as shown in equation (6)
Energy function:
E(C)=βF(c1,c2,C)+(1-β)B(C)+αRp(φ) (6)
In equation (6), α > 0, also, penaltyWherein, Ω indicates whole picture figure
Picture.
It is preferred that above-mentioned processing can be carried out in narrow bandwidth range, to improve the speed of iterative processing.By above-mentioned
Mode, tumor's profiles separator 30 may separate out the profile of tumour, using as the tumour detected.
Fig. 5 shows the example of the tumor's profiles separated according to an exemplary embodiment of the present.As shown in figure 5, in step
For the tumor's profiles that S300 is isolated compared with actual tumour, degree of approximation is higher.
It should be noted that method shown in Fig. 2 can also be by computer programming come real other than detection system shown in FIG. 1
It is existing, it can also be executed via the proprietary processor that specific hardware is constituted.Those skilled in the art can be used it is any in the art
Known or available means execute method shown in Fig. 2.
By design special experiment (for example, for from 1941 patients (wherein, 913 benign tumour patients,
1028 malignant tumor patients) 2758 width ultrasound images), it will thus be seen that with existing tumor region positioning method and tumour
Separate mode is compared, and tumor region positioning method according to an exemplary embodiment of the present invention is obvious not only for the hit rate of tumour
Increase, and the situation for omitting or positioning mistake significantly reduces, and tumour separation side according to an exemplary embodiment of the present invention
Method can obtain accuracy rate more better than existing way.
It can be seen that the system of automatic detection tumour as described above, can in the case where not needing artificially to participate in,
Effectively detected from ultrasound image include tumour region, and relatively accurately isolate the wheel of tumour on this basis
Exterior feature, using one of the foundation made a definite diagnosis as tumour.
Automatic checkout system and method according to an exemplary embodiment of the present invention are described above in relation to lesion detection, so
And, it should be appreciated that the above exemplary embodiments apply also for other object detections in ultrasound image, particularly, ultrasound figure
As can be applied to numerous areas as a kind of nondestructiving detecting means, e.g., can be used for detecting mechanical, metallurgy, as empty space flight, railway,
Casting, forging, plate, composite material, tubing, bar, profile, weldment, the machine of the industries such as coal, non-ferrous metal, building add
Above-mentioned workpiece sensing in workpiece and use.Therefore, tumour automatic checkout system described above and method can be applied to needle
To other object detections in ultrasound image, without the object being detected is limited to tumour.
Object detection systems according to an exemplary embodiment of the present invention are described hereinafter with reference to Fig. 6 and Fig. 7 and by right
As detection system carrys out the process flow of test object.
Fig. 6 shows the block diagram of object detection systems according to an exemplary embodiment of the present invention.As shown in fig. 6, according to this hair
The object detection systems of bright exemplary embodiment include: candidate region detection device 101, for being based on deformable part model
(DPM) from least one candidate region of ultrasound image detection including object；Subject area positioning device 201 is used for from detection
To at least one described candidate region determine subject area；Object outline separator 301, for by based on determining pair
As the profile of region disconnecting object carrys out test object.
Fig. 7 show it is according to an exemplary embodiment of the present invention by object detection systems come the process flow of test object.
Referring to Fig. 7, in step S101, deformable part model (DPM) is based on from ultrasound by candidate region detection device 101
Image detection includes at least one candidate region of object.
Then, it in operation S201, is determined pair by subject area positioning device 201 from least one candidate region detected
As region.
Finally, in operation S301, by object outline separator 301 by separating object based on determining subject area
Profile carrys out test object.
It should be noted that detection system shown in fig. 6 in each device when executing operating process shown in Fig. 7, use and Fig. 1
With similar fashion shown in Fig. 2, difference, which is only that, to be changed to about the parameters of tumour and processing for any object
Parameter and processing.
Although being particularly shown and describing the present invention, those skilled in the art referring to its exemplary embodiment
It should be understood that in the case where not departing from the spirit and scope of the present invention defined by claim form can be carried out to it
With the various changes in details.
Claims (16)
1. a kind of system for detecting tumour automatically in ultrasound image, comprising:
Candidate region detection device includes at least the one of tumour for being detected based on deformable part model DPM from ultrasound image
A candidate region；
Tumor region positioning device determines tumor region at least one candidate region described in detect；
Tumor's profiles separator, for detecting tumour by the profile for separating tumour based on determining tumor region,
Wherein, candidate region detection device includes swollen from ultrasound image detection by changing the ratio of width to height of the root template in DPM
At least one candidate region of tumor.
2. the system as claimed in claim 1, wherein tumor region positioning device uses two based on support vector machines (SVM)
Value classifier at least one candidate region described in detect determines tumor region.
3. system as claimed in claim 2, wherein the feature vector of the two-value classifier is based on the context of candidate region
Feature.
4. system as claimed in claim 3, wherein described eigenvector includes at least one of following item: candidate region
DPM detection score value, candidate region position and size, candidate region in offset of all parts relative to root, candidate regions
Part coexists between intensity difference, candidate region in domain between prospect and background and the DPM detection highest candidate region of score value.
5. system as claimed in claim 4, wherein tumor region positioning device uses Multiple Kernel Learning (MKL) method by two-value
The kernel function of classifier is defined as the linear combination of multiple basic kernel functions.
6. system as claimed in claim 5, wherein tumor region positioning device is by the RBF kernel function and three of three kinds of bandwidth
The Polynomial kernel function of a dimension carries out linear combination, thus for each of feature vector characteristic component and feature to
Amount is whole to be trained, to obtain the kernel function of two-value classifier.
7. the system as claimed in claim 1, wherein tumor's profiles separator separates the wheel of tumour using Level Set Method
It is wide, wherein the boundary of the contour curve of tumour using tumor region is iterated as initial curve.
8. system as claimed in claim 7, wherein tumor's profiles separator is by making the prospect and background of tumor image
The distance between maximize to construct the energy function used in Level Set Method.
9. a kind of system for test object automatic in ultrasound image, comprising:
Candidate region detection device includes at least the one of object for being detected based on deformable part model DPM from ultrasound image
A candidate region；
Subject area positioning device determines subject area at least one candidate region described in detect；
Object outline separator, for by the profile that separates object based on determining subject area come test object,
Wherein, candidate region detection device includes pair from ultrasound image detection by changing the ratio of width to height of the root template in DPM
At least one candidate region of elephant.
10. system as claimed in claim 9, wherein subject area positioning device uses two based on support vector machines (SVM)
Value classifier at least one candidate region described in detect determines subject area.
11. system as claimed in claim 10, wherein the feature vector of the two-value classifier is based on the upper and lower of candidate region
Literary feature.
12. system as claimed in claim 11, wherein subject area positioning device uses Multiple Kernel Learning (MKL) method by two
The kernel function of value classifier is defined as the linear combination of multiple basic kernel functions.
13. a kind of method for test object automatic in ultrasound image, comprising:
At least one candidate region including object is detected from ultrasound image based on deformable part model DPM；
At least one candidate region described in detect determines subject area；
By separating the profile of object based on determining subject area come test object,
Wherein, at least one candidate including object is detected from ultrasound image by changing the ratio of width to height of the root template in DPM
Region.
14. method as claimed in claim 13, wherein at least one candidate region described in detect determines subject area
The step of include: using at least one candidate region described in detect of the two-value classifier based on support vector machines (SVM)
Determine subject area.
15. method as claimed in claim 14, wherein the feature vector of the two-value classifier is based on the upper and lower of candidate region
Literary feature.
16. method as claimed in claim 15, further includes: use Multiple Kernel Learning (MKL) method by the core letter of two-value classifier
Number is defined as the linear combination of multiple basic kernel functions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310064300.1A CN104021539B (en) | 2013-02-28 | 2013-02-28 | System for detecting tumour automatically in ultrasound image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310064300.1A CN104021539B (en) | 2013-02-28 | 2013-02-28 | System for detecting tumour automatically in ultrasound image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104021539A CN104021539A (en) | 2014-09-03 |
| CN104021539B true CN104021539B (en) | 2019-07-16 |
Family
ID=51438278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310064300.1A Expired - Fee Related CN104021539B (en) | 2013-02-28 | 2013-02-28 | System for detecting tumour automatically in ultrasound image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104021539B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10172517B2 (en) * | 2016-02-25 | 2019-01-08 | Samsung Electronics Co., Ltd | Image-analysis for assessing heart failure |
| WO2018176189A1 (en) | 2017-03-27 | 2018-10-04 | 上海联影医疗科技有限公司 | Image segmentation method and system |
| CN107369154B (en) * | 2017-07-19 | 2020-05-05 | 电子科技大学 | Image detection device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102068281A (en) * | 2011-01-20 | 2011-05-25 | 深圳大学 | Processing method for space-occupying lesion ultrasonic images |
| CN102300505A (en) * | 2009-06-30 | 2011-12-28 | 株式会社东芝 | Ultrasonic diagnostic device and control program for displaying image data |
| CN102622604A (en) * | 2012-02-14 | 2012-08-01 | 西安电子科技大学 | Multi-angle human face detecting method based on weighting of deformable components |
| CN102855483A (en) * | 2011-06-30 | 2013-01-02 | 北京三星通信技术研究有限公司 | Method and device for processing ultrasonic images and breast cancer diagnosis equipment |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5586203B2 (en) * | 2009-10-08 | 2014-09-10 | 株式会社東芝 | Ultrasonic diagnostic apparatus, ultrasonic image processing apparatus, and ultrasonic image processing program |
-
2013
- 2013-02-28 CN CN201310064300.1A patent/CN104021539B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102300505A (en) * | 2009-06-30 | 2011-12-28 | 株式会社东芝 | Ultrasonic diagnostic device and control program for displaying image data |
| CN102068281A (en) * | 2011-01-20 | 2011-05-25 | 深圳大学 | Processing method for space-occupying lesion ultrasonic images |
| CN102855483A (en) * | 2011-06-30 | 2013-01-02 | 北京三星通信技术研究有限公司 | Method and device for processing ultrasonic images and breast cancer diagnosis equipment |
| CN102622604A (en) * | 2012-02-14 | 2012-08-01 | 西安电子科技大学 | Multi-angle human face detecting method based on weighting of deformable components |
Non-Patent Citations (2)
| Title |
|---|
| 乳腺超声图像中的肿瘤区域定位与肿瘤分类技术研究;刘博;《中国博士学位论文全文数据库》;20110415;第2011卷;第1-145页 |
| 分割位置提示的可变形部件模型快速目标检测;杨扬等;《自动化学报》;20120430;第38卷(第4期);第540-548页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104021539A (en) | 2014-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Schlegl et al. | f-AnoGAN: Fast unsupervised anomaly detection with generative adversarial networks | |
| Peng et al. | A review of heart chamber segmentation for structural and functional analysis using cardiac magnetic resonance imaging | |
| JP6467041B2 (en) | Ultrasonic diagnostic apparatus and image processing method | |
| El-Baz et al. | Appearance analysis for the early assessment of detected lung nodules | |
| US9277902B2 (en) | Method and system for lesion detection in ultrasound images | |
| CN103425986B (en) | Mammary gland tumor image characteristic extracting method based on edge neighborhood weighting | |
| JP5498299B2 (en) | System and method for providing 2D CT images corresponding to 2D ultrasound images | |
| EP2846310A2 (en) | Method and apparatus for registering medical images | |
| US8588519B2 (en) | Method and system for training a landmark detector using multiple instance learning | |
| Hogeweg et al. | Clavicle segmentation in chest radiographs | |
| CN105701331A (en) | Computer-aided diagnosis apparatus and computer-aided diagnosis method | |
| JP2013525010A (en) | Malignant mass detection and classification in radiographic images | |
| CN107708550A (en) | For the surface modeling for the segmentation acoustic echo structure for detecting and measuring anatomic abnormalities | |
| JP2014009945A (en) | Image processing program, recording medium, image processing device, and image processing method | |
| US8340385B2 (en) | Method and system for left ventricle detection in 2D magnetic resonance images using ranking based multi-detector aggregation | |
| CN104021539B (en) | System for detecting tumour automatically in ultrasound image | |
| US9763636B2 (en) | Method and system for spine position detection | |
| Sammouda | Segmentation and analysis of CT chest images for early lung cancer detection | |
| Oliveira et al. | Simultaneous detection of prominent points on breast cancer conservative treatment images | |
| CN103455821A (en) | Device and method for analyzing images on basis of BI-RADS (breast imaging reporting and data system) | |
| EP3454301A1 (en) | Method for detecting and labelling coronary artery calcium | |
| US8761480B2 (en) | Method and system for vascular landmark detection | |
| Shankar et al. | Femur bone volumetric estimation for osteoporosis classification using optimization-based deep belief network in X-ray images | |
| CN112515705B (en) | Method and system for projection profile enabled computer-aided detection | |
| US8165375B2 (en) | Method and system for registering CT data sets |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190716 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |