CN104504720A - New prostate ultrasonoscopy segmentation technique - Google Patents
New prostate ultrasonoscopy segmentation technique Download PDFInfo
- Publication number
- CN104504720A CN104504720A CN201510006415.4A CN201510006415A CN104504720A CN 104504720 A CN104504720 A CN 104504720A CN 201510006415 A CN201510006415 A CN 201510006415A CN 104504720 A CN104504720 A CN 104504720A
- Authority
- CN
- China
- Prior art keywords
- prostate
- image
- class
- segmentation
- pixel
- 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.)
- Granted
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
- A61B8/5223—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for extracting a diagnostic or physiological parameter from medical diagnostic data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- 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
- G06T2207/30081—Prostate
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Pathology (AREA)
- Radiology & Medical Imaging (AREA)
- General Physics & Mathematics (AREA)
- Heart & Thoracic Surgery (AREA)
- Physiology (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Image Processing (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a new prostate ultrasonoscopy segmentation technique for accurately segmenting a prostate image in a transrectal ultrasound (TRUS) image. A traditional view is that the spots in the TRUS image have intense side effect to target segmentation. In contrast, the segmentation problem is simplified by using the intrinsic attribute of the spots. The technique comprises the steps: firstly, the sizes and the directions of the spots are not entirely distributed at random and comply with certain rule characteristics; secondly, in order to simplify the direction characteristic of the spots, the problem characteristic is extracted by using a rotation-invariant Gabor characteristic; in order to use the size characteristic of the spots, the TRUS image is segmented into a plurality of bar-shaped narrow bands and image pixels are processed differently in different narrow bands; and finally, the correlation among different pixels is considered and the image classification precision is improved by further using the graph cut method. For this reason, the segmentation result of the prostate image in the TRUS image is obtained finally by using a level set segmentation model.
Description
Technical field
The present invention relates to a kind of image Segmentation Technology, particularly relate to prostate Ultrasound Image Segmentation technology, belong to Computer Image Processing field.
Background technology
At home, the segmentation of some seminar to TRUS image has been had to be studied.As far back as 2003, Southeast China University professor Luo Limin just proposed a kind of multiple dimensioned nonlinear adaptive boundary testing method of ultrasound wave.First the method carries out multi-resolution decomposition to ultrasonoscopy, then suppresses finally to use the image of linear barrier's detection method to noise reduction based on " fillet " to process with reconstruction with wavelet image after speckle noise by non-linear soft thresholding method.But due to the time overhead of region propagation process, its calculated amount is larger.Beijing Jiaotong University professor Ruan Qiuqi proposes one and proposes a kind of Ultrasonic Image Denoising and edge enhancement algorithm: while removing noise, can keep important edge, local detail and ultrasonic echo bright wisp.Although strengthen border to restraint speckle have good effect, do not apply it in the segmentation of image and rim detection.Because ultrasonoscopy signal to noise ratio (S/N ratio) is low, speckle noise strong, above-mentioned simple image enhaucament and edge detection method are difficult to obtain good effect.
Researchist is devoted to combining image feature and domain knowledge, to reaching ideal segmentation effect.The texture features of ultrasonoscopy is one of priori of using under study for action of people.People have also attempted multiple method to utilize this feature to help prostate auto Segmentation.In addition, active performance model (Active Appearance Model) has the advantages that texture information to be embedded in shape, has also shown the potentiality in TRUS Iamge Segmentation.In addition, local binary patterns (Local Binary Pattern) texture feature extraction has also been widely applied in Ultrasound Image Segmentation.2010, the method that the active contour model that Harbin Institute of Technology professor Tang Jianglong proposes a kind of combination overall situation probability density difference and local gray level matching is split ultrasonoscopy.The method make use of the overall situation and the local message of image respectively on raw ultrasound image and pretreatment image.On the original image, utilize the intensity profile in each region, and in conjunction with the background knowledge of ultrasonoscopy to the global information modeling of image.Said method all obtains certain effect, but they all well do not utilize ultrasonoscopy to be different from some inherent characteristicses of natural image, and therefore in actual applications, its performance is all subject to a definite limitation.First, in prostate ultrasonoscopy, the size of speckle noise is not unalterable, but changes along with the distance between itself and ultrasonic probe and change.This characteristic is not embodied and is used in the feature extraction algorithm of current main flow.Secondly, the spot in ultrasonoscopy also has self angle, and angle Non-random distribution, and present certain regularity, therefore, directly use Gabor filtering to be inappropriate for carrying out extraction.The method of the people such as Shen take into account the rotation variation characteristic of Gabor filtering, but does not combine with the rotation of spot further, and therefore its performance is also subject to a definite limitation.In 2006, the people such as Zhan also using Gabor filtering, and be extend in the segmentation of 3D ultrasonoscopy.
Another important priori is prostatic style characteristic.At first, prostatic shape is regarded as hyperelliptic, carries out various conversion by alignment, comes and objective contour matching.The people such as UBC Badiei proposed the artificial initialization of a kind of needs in 2006, use hyperelliptic matching to find the method for prostate margin.This kind of method is simply efficient, has good effect to the prostate of general shape, and comparatively serious for deformation ratio, and target in irregular shape, be then difficult to reach desirable effect.Therefore, a kind of statistical shape model more flexibly obtains the larger concern of researchist, and gives good segmentation result.But, most methods hypothesis prostate shape Gaussian distributed.2007, Institutes Of Technology Of Nanjing professor Xia Deshen proposed a kind of prior shape parametric active contour model.This model is by introducing a kind of prior shape field of force of non-distance property, and build a kind of parametric active contour model that can reflect prior shape newly, new prior shape movable contour model avoids.The calculating of the spacing of curve, decreases the complicacy of model, can well utilize the prior shape information of segmentation object simultaneously.But this model still needed the initial curve of Manual definition's movable contour model before process image.The people such as North America, Philip research institute Yang proposed a kind of new thinking in 2010, use piecemeal activity skeleton pattern (Partial Active Shape Models) process border lose problem, be used in conjunction with discrete can deformation model (Discrete Deformable Model) further to segmentation result refinement.The method uses not pretreated image, does not excavate the information (as texture features) contained in image further.In addition; the method to the modeling of target shape based on principal component analysis (PCA) (Principal Component Analysis); the linear dependence of different dimensions in training set can only be removed; the generally relevant correlativity existed for real data is difficult to thorough removal, thus for and the prostate shape not meeting Gaussian distribution be difficult to obtain good effect.Also similar problem is faced in conjunction with the method for shape under level-set segmentation framework.As the people such as Tsai in directly principal component analysis (PCA) is applied on the surface evolution implicit function of vectorization, although achieve good effect, because the symbolic distance norm dimension of vectorization is too high, the limitation of model is larger.The nonparametric probability model that can approach any distribution is also introduced in medical image segmentation field [36], but these class methods are not due to well in conjunction with the domain knowledge of ultrasonoscopy and clinical practice itself at present, therefore mostly need manual intervention initialization just can meet the requirements of precision, greatly limit its application and expansion further.
Although the shape of non-gaussian distribution achieves more superior achievement in many medical image segmentation fields; Meanwhile, be done much research, non-gaussian shape prior be embedded in the display parted pattern (as movable contour model) and implicit expression parted pattern (as level set) be widely adopted and go.But few people use these non-gaussian shape priors for ultrasonoscopy specially.
Summary of the invention
Technical matters to be solved by this invention is to provide a kind of new prostate Ultrasound Image Segmentation technology, at TRUS image, there is signal to noise ratio (S/N ratio) (signal-to-noise ratio, SNR) low, contrast is low, speckle noise is strong, when artifact interference and same histoorgan intensity profile isomerism are by force and the distribution of different tissues organ intensity has the features such as homoorganicity, solve TRUS Image Automatic Segmentation problem.
Solution of the present invention is: we propose a comparatively comprehensively segmentation framework, incorporate (1) prostate shape prior, the gray scale roll-off characteristic near (2) border and (3) prostate mottled grain characteristic.Traditional view is thought, the spot of TRUS image can cause strong spinoff to Target Segmentation.Unlike this, we utilize the inherent attribute of spot to simplify segmentation problem.First, according to our observed result, the size and Orientation of spot is not completely random distribution, but obeys certain rule characteristic, and these spots are not necessarily harmful to, and even may offer help to prostate segmentation.In order to utilize the directivity characteristics of spot, we used the Gabor characteristic of invariable rotary to extract problem characteristic; In order to utilize the size dimensional properties of spot, we divide into many strip arrowbands TRUS image, in different arrowband, carry out different process to image pixel.Finally, we also contemplate the correlativity between different pixels, and improve pixel classifications precision further by this characteristic use figure (graph cut) method of cutting.Based on this, we, by level-set segmentation model, finally obtain prostatic segmentation result in TRUS image.
The present invention is for realizing above-mentioned solution, and its method step is as follows:
1.
feature extractiongabor filtering is the common method of ultrasonoscopy being carried out to texture feature extraction.Gabor filtering can be counted as angle and the adjustable edge of yardstick and line segment detector, and can extract the local feature information of different frequency range by arranging different Rotation and Zoom parameters.The Gabor filter equation of two dimension
(1)
Wherein
(2)
λ represents the wavelength factor of sine function, and θ represents the sense of rotation of Gabor function, and Ψ is difference, and σ represents the standard deviation of Gaussian convolution, and γ is spatial aspect ratio.The proper vector that tradition Gabor filtering is extracted can use column vector
represent, each element in this vector is
, it is
with scale parameter
.Gabor function and the convolution of original image at position (x, y) place obtain.General, the Selecting parameter of Gabor filtering bunch follows the half adjacent principle between two of the frequency response of each adjacent filter function in frequency domain.
Although Gabor filtering has the ability of stronger texture feature extraction, we also see, Gabor characteristic is extracted and be adapted with the sense of rotation of spot.Notice that the people such as Shen proposed the Gabor characteristic extracting method of invariable rotary in 2003, although its starting point is different from us, but its rotation center is chosen as ultrasonic probe, consistent with our demand, therefore the method can be adopted and improve a little, to reach the object solving above-mentioned first speckle characteristics.Strict, if the circular arc centered by ultrasonic probe, the normal orientation at (x, y) place is that Φ, Φ then can be calculated by following formula
(3)
be wherein the center of ultrasonic probe, function
for tape symbol arctan function, its symbol determines according to four of its place different quadrants.By when the arc method vector when definition, in position
the invariable rotary Gabor characteristic at place
can be calculated by following formula
(4)
Wherein
.Notice
for lower bracket function,
, wherein n is the rotation parameter number altogether used.Finally, invariable rotary proper vector F (x, y) namely by
for its element is formed
2.
tagsortextract according to pixel characteristic, pixels all in image will doubly be divided into three classes.First we can set up the banded arrowband of many arcs in ultrasonoscopy, and in different arrowband, train different characteristics dictionaries to carry out linear expression to pixel, obtain and represent residual error.Then, based on this residual error, we use figure to cut (graph cut) method, the correlativity between each pixel is considered in the lump, obtains final pixel classifications result.
According to the theory in compressed sensing (Compressive Sensing), for each proper vector F and a group of our dictionary D of obtaining in advance, as long as use enough sparse of the expression factor alpha of dictionary D to vectorial F, then finding can by additional on original function to the process of the sparse expression of F
norm constraint comes that substitution problem itself proposes
minimization problem under norm, now, this sparse expression problem can be expressed as
(5)
Wherein λ Lagrange's multiplier (Lagrange multiplier).We use the expression residual error of above-mentioned sparse expression as classification foundation.
We obtain second feature of residual error strategy based on ultrasonoscopy.This feature is thought, the size of ultrasonic speckle changes and changes along with the distance at itself and supersonic sounding center.Based on this characteristic, we think and only use Global Dictionary as expression foundation, and what be not sufficient to fully to contain ultrasonic speckle enriches texture features.In order to address this problem, we establish the banded fillet of many arcs in ultrasonoscopy, in these arrowbands, all train separately and have respective expression dictionary.Strict, for each pixel in region interested on image (Region Of Interest, ROI)
Ρ, the fillet belonging to it can be calculated by following formula
(6)
Wherein i is the index of arrowband, and c is the center of detector bar, and h is the bandwidth of arrowband.In each arrowband, not in training set, all pixels all bring training wordbook, but only have those pixels in the picture near prostate contours to be just used for training dictionary.For three pixel classes, prostate border class, prostate border inner class and prostate border outer class, we are respectively it in each arrowband and trained respective dictionary.Here once why we will be divided three classes pixel instead of two classes (outside in border and border) intuitively for our first brief explanation.First, in our experiment, we find, although it is larger that pixel is divided into the difficulty of three classes, our method obtains the experimental result consistent with two classes of classifying.Secondly, when being divided into three classes, parted pattern more directly can locate prostate contours (and not needing to be located by the border of two classes in two class situations), thus helps whole parted pattern to obtain better overall segmentation effect.
For the pixel in i-th arrowband bar
pif, its proper vector
fp, we can use dictionary matrix
rarefaction representation is carried out to it, wherein
,
with
represent for prostate border class in i-th arrowband respectively, the dictionary of prostate border inner class and three the class training of prostate border outer class.According to these three dictionaries, pixel
pexpression residual error can be calculated by following formula
(7)
Wherein
(8)
be such function, it follows k not to be that a dvielement is all set to zero all in α.As for the method for dictionary training, academia has proposed many solutions, has the algorithm of many maturations.
After acquisition represents residual error, we directly do not use the class with minimum expression residual error as the class of this pixel, but get the negative value representing residual error, and all residual errors are normalized to [0,1] interval tolerance belonging to the possibility of a certain class as this pixel.Consider that neighbor has higher possibility and can belong to same class, we use following energy function to cut by figure the classification that (graph cut) finally realizes pixel
μ represents weight parameter,
for image area,
for the set of neighbor,
for pixel p belongs to k class punishment value, B is used for maintaining the flatness of classifying between neighbor, and its formula is:
by residual error
specification is interval to [0,1].
3.
level-set segmentation modelthis model is the summation of feature extracting method and arrowband dividing method.Our level-set segmentation model is based on energy minimization models, and its energy definition is
(10)
Wherein α is weight parameter,
be the arrowband contrast energy of definition, and
it is then the texture energy according to the textural characteristics design in this chapter.This two parts energy can be defined rigorously for
(11)
(12)
Wherein
for image area, I is gradation of image, Φ symbolic measurement (SDF).Symbol
,
with
represent pixel classifications result, be respectively prostate border class, prostate border inner class and prostate border outer class.If a pixel is classified a certain class, then the l of such correspondence is set to 1, and other l is all set as 0.Three formula define the arrowband of edge of model below, for information extraction, help segmentation:
(13)
(14)
(15)
Wherein h and c is two different bandwidth, and h>c.These two different narrow band modes use respectively in formula 11 and formula 12, and we can see contrast energy, i.e. formula 11.It is the expansion of normal vector profile (NVP) on Level Set Models, simultaneously, it has also merged the region-competitive feature of Chan-Vese model, and in conjunction with the characteristic of ultrasonoscopy itself, only considered the information near prostate contours.On the other hand, we see texture features energy, i.e. formula 12, be then used to measure the goodness of fit of final segmentation result and pixel classifications result before.
energy minimizationwe realize the parametrization method for expressing for Level Set Models.Therefore, we can directly use gradient descent method to minimize formula 10 to solve.The gradient of ENERGY E is
(16)
Wherein three different H (
,
, and
) gradient can be obtained by chain rule:
(17)
Wherein δ (.) is dirac thunder function (Dirac Delta function).Symbol
,
h, 0, c} representative at formula 13, the bound used in formula 14 and formula 14, and
<
.Formula 17 is substituted in formula 16, and we can obtain
(18)
By the evolution of this model, (namely iteration upgrades w and p), can realize the segmentation to ultrasonoscopy.Model evolution process is
(19)
(20)
Wherein
with
for step parameter, t represents iterations.
Because our energy equation is based on local message, therefore we need the process of a model initialization to be absorbed in Local Minimum to avoid model.Here, we improve the model initialization method in a upper chapter, to obtaining better effect.Initial method is the exhaustive search on image area.Our initial method is at image
an interior exhaustive search process, we can test repeatedly
pdifferent value, fix w=0 constant simultaneously, roughly find prostate position in the picture.Each is tested
pvalue, we investigate it by three steps.First, because form parameter is set as fixed value, model cannot accurately with the prostate matching in image.Therefore, we are attempted prostatic matching by the energy minimization step of carrying out a few step little to master mould, and in this step,
with
the ratio that these two step parameters are arranged is larger under normal circumstances.Then, in each step, we calculate contrast energy
.Finally, we save
corresponding
, parameter w and
p.After all situations is all tested and terminated, we get n (in our realization n=5) and have minimum
situation.In this several situation, have minimum
that be regarded as initialization model.We directly do not use formula 10 to carry out initialization is because our energy function has local characteristics, directly use in initial search process formula 10 easily affect by other pixel misclassification result.
Claims (4)
1. a new prostate Ultrasound Image Segmentation technology, its feature extraction and tagsort method comprise the following steps:
1) utilize the Gabor characteristic texture feature extraction of invariable rotary, if the circular arc centered by ultrasonic probe, the normal orientation at (x, y) place is that Φ, Φ then can be calculated by following formula
be wherein the center of ultrasonic probe, function
for tape symbol arctan function, its symbol determines according to four of its place different quadrants. by when the arc method vector when definition, in position
the invariable rotary Gabor characteristic at place
can be calculated by following formula
Wherein
. notice
for lower bracket function,
, wherein n is the rotation parameter number altogether used. last, invariable rotary proper vector F (x, y) namely by
for its element is formed.
2) according to the theory in compressed sensing (Compressive Sensing), for each proper vector F and a group of our dictionary D of obtaining in advance, as long as use enough sparse of the expression factor alpha of dictionary D to vectorial F, then finding can by additional on original function to the process of the sparse expression of F
norm constraint comes that substitution problem itself proposes
minimization problem under norm, now, this sparse expression problem can be expressed as
Wherein λ Lagrange's multiplier (Lagrange multiplier). we use the expression residual error of above-mentioned sparse expression as classification foundation;
3) second feature of residual error strategy based on ultrasonoscopy is obtained. this feature is thought, the size of ultrasonic speckle changes and changes along with the distance at itself and supersonic sounding center. based on this characteristic, we think that only use Global Dictionary is as expression foundation, what be not sufficient to fully to contain ultrasonic speckle enriches texture features. in order to address this problem, we establish the banded fillet of many arcs in ultrasonoscopy, in these arrowbands, all train separately and have respective expression dictionary. strict, for region interested on image (Region Of Interest, ROI) each pixel in
Ρ, the fillet belonging to it can be calculated by following formula
Wherein i is the index of arrowband, c is the center of detector bar, h is the bandwidth of arrowband. in each arrowband, not in training set, all pixels all bring training wordbook, but only have those pixels in the picture near prostate contours to be just used for training dictionary. and for the pixel in i-th arrowband bar
pif, its proper vector
fp, we can use dictionary matrix
rarefaction representation is carried out to it, wherein
,
with
represent for prostate border class in i-th arrowband respectively, the dictionary of prostate border inner class and three the class training of prostate border outer class. according to these three dictionaries, pixel
pexpression residual error can be calculated by following formula
Wherein
be such function, it follows k not to be that a dvielement is all set to zero all in α.
2. a kind of new prostate Ultrasound Image Segmentation technology according to right 1, it realizes sorting technique and comprises the following steps
1) after acquisition represents residual error, we directly do not use the class with minimum expression residual error as the class of this pixel, but get the negative value representing residual error, and all residual errors are normalized to [0,1] the interval tolerance belonging to the possibility of a certain class as this pixel. consider that neighbor has higher possibility and can belong to same class, we use following energy function to cut by figure the classification that (graph cut) finally realizes pixel
μ represents weight parameter,
for image area,
for the set of neighbor,
for pixel p belongs to k class punishment value, B is used for maintaining the flatness of classifying between neighbor, and its formula is:
by residual error
specification is interval to [0,1].
3. the prostate Ultrasound Image Segmentation new technology according to right 1, its characteristic level collection parted pattern comprises the following steps:
1) level-set segmentation model is based on energy minimization models, and its energy definition is
Wherein α is weight parameter,
be the arrowband contrast energy of definition, and
be then according in this chapter textural characteristics design texture energy. this two parts energy can be defined rigorously for
Wherein
for image area, I is gradation of image, Φ symbolic measurement (SDF). symbol
,
with
represent pixel classifications result, be respectively prostate border class, prostate border inner class and prostate border outer class. if a pixel is classified a certain class, then the l of such correspondence is set to 1, other l be all set as 0. below three formula define the arrowband of edge of model, for information extraction, help segmentation:
Wherein h and c is two different bandwidth, and h>c.
4. the prostate Ultrasound Image Segmentation new technology according to right 1, its characteristic energy minimizes and comprises the following steps:
1) directly using gradient descent method to minimize energy theorem to solve. the gradient of ENERGY E is
(16)
Wherein three different H (
,
, and
) gradient can be obtained by chain rule:
(17)
Wherein δ (.) is dirac thunder function (Dirac Delta function. symbol
,
h, 0, c} representative at formula 13, the bound used in formula 14 and formula 14, and
<
. formula 17 is substituted in formula 16, and we can obtain
by the evolution of this model (namely iteration upgrades w and p), can realize the segmentation to ultrasonoscopy. model evolution process is
Wherein
with
for step parameter, t represents iterations.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510006415.4A CN104504720B (en) | 2015-01-07 | 2015-01-07 | New prostate ultrasonoscopy segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510006415.4A CN104504720B (en) | 2015-01-07 | 2015-01-07 | New prostate ultrasonoscopy segmentation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104504720A true CN104504720A (en) | 2015-04-08 |
| CN104504720B CN104504720B (en) | 2017-04-19 |
Family
ID=52946114
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510006415.4A Expired - Fee Related CN104504720B (en) | 2015-01-07 | 2015-01-07 | New prostate ultrasonoscopy segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104504720B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105894496A (en) * | 2016-03-18 | 2016-08-24 | 常州大学 | Semi-local-texture-feature-based two-stage image segmentation method |
| CN108090953A (en) * | 2017-12-14 | 2018-05-29 | 上海联影医疗科技有限公司 | Area-of-interest method for reconstructing, system and computer readable storage medium |
| CN108596894A (en) * | 2018-04-25 | 2018-09-28 | 王成彦 | A kind of prostate automatic Mesh Partition Method for multi-parameter nuclear magnetic resonance image |
| CN108921860A (en) * | 2018-07-10 | 2018-11-30 | 北京大学 | A kind of prostate magnetic resonance image full-automatic partition method |
| CN113838001A (en) * | 2021-08-24 | 2021-12-24 | 内蒙古电力科学研究院 | Ultrasonic full-focus image defect processing method and device based on nuclear density estimation |
| CN116229189A (en) * | 2023-05-10 | 2023-06-06 | 深圳市博盛医疗科技有限公司 | Image processing method, device, equipment and storage medium based on fluorescence endoscope |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100081931A1 (en) * | 2007-03-15 | 2010-04-01 | Destrempes Francois | Image segmentation |
| CN102737382A (en) * | 2012-06-22 | 2012-10-17 | 刘怡光 | Automatic precise partition method for prostate ultrasonic image |
| CN103295224A (en) * | 2013-03-14 | 2013-09-11 | 北京工业大学 | Breast ultrasonoscopy automatic segmentation method based on mean shift and divide |
-
2015
- 2015-01-07 CN CN201510006415.4A patent/CN104504720B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100081931A1 (en) * | 2007-03-15 | 2010-04-01 | Destrempes Francois | Image segmentation |
| CN102737382A (en) * | 2012-06-22 | 2012-10-17 | 刘怡光 | Automatic precise partition method for prostate ultrasonic image |
| CN103295224A (en) * | 2013-03-14 | 2013-09-11 | 北京工业大学 | Breast ultrasonoscopy automatic segmentation method based on mean shift and divide |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105894496A (en) * | 2016-03-18 | 2016-08-24 | 常州大学 | Semi-local-texture-feature-based two-stage image segmentation method |
| CN108090953A (en) * | 2017-12-14 | 2018-05-29 | 上海联影医疗科技有限公司 | Area-of-interest method for reconstructing, system and computer readable storage medium |
| CN108090953B (en) * | 2017-12-14 | 2022-03-25 | 上海联影医疗科技股份有限公司 | Region-of-interest reconstruction method, system and computer-readable storage medium |
| CN108596894A (en) * | 2018-04-25 | 2018-09-28 | 王成彦 | A kind of prostate automatic Mesh Partition Method for multi-parameter nuclear magnetic resonance image |
| CN108921860A (en) * | 2018-07-10 | 2018-11-30 | 北京大学 | A kind of prostate magnetic resonance image full-automatic partition method |
| CN108921860B (en) * | 2018-07-10 | 2021-09-10 | 北京大学 | Full-automatic segmentation method for prostate magnetic resonance image |
| CN113838001A (en) * | 2021-08-24 | 2021-12-24 | 内蒙古电力科学研究院 | Ultrasonic full-focus image defect processing method and device based on nuclear density estimation |
| CN113838001B (en) * | 2021-08-24 | 2024-02-13 | 内蒙古电力科学研究院 | Ultrasonic wave full focusing image defect processing method and device based on nuclear density estimation |
| CN116229189A (en) * | 2023-05-10 | 2023-06-06 | 深圳市博盛医疗科技有限公司 | Image processing method, device, equipment and storage medium based on fluorescence endoscope |
| CN116229189B (en) * | 2023-05-10 | 2023-07-04 | 深圳市博盛医疗科技有限公司 | Image processing method, device, equipment and storage medium based on fluorescence endoscope |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104504720B (en) | 2017-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108053417B (en) | lung segmentation device of 3D U-Net network based on mixed rough segmentation characteristics | |
| CN104504720A (en) | New prostate ultrasonoscopy segmentation technique | |
| Liu et al. | A feature-preserving framework for point cloud denoising | |
| Zheng et al. | Unsupervised saliency-guided SAR image change detection | |
| Park et al. | Multi-scale tensor voting for feature extraction from unstructured point clouds | |
| Liu et al. | Hierarchical semantic model and scattering mechanism based PolSAR image classification | |
| CN102737382B (en) | Automatic precise partition method for prostate ultrasonic image | |
| Nitzken et al. | Improving full-cardiac cycle strain estimation from tagged CMR by accurate modeling of 3D image appearance characteristics | |
| Wang et al. | Surface feature based mesh segmentation | |
| Monteiro et al. | Watershed framework to region-based image segmentation | |
| CN104200471A (en) | SAR image change detection method based on adaptive weight image fusion | |
| Chauhan et al. | An improved image contrast enhancement based on histogram equalization and brightness preserving weight clustering histogram equalization | |
| Chen et al. | A novel infrared small target detection method based on BEMD and local inverse entropy | |
| CN104008386A (en) | Method and system for identifying type of tumor | |
| Arakeri et al. | An effective and efficient approach to 3D reconstruction and quantification of brain tumor on magnetic resonance images | |
| CN103065299A (en) | Ultrasonic image edge extraction method and ultrasonic image edge extraction device | |
| Huang et al. | Medical image fusion algorithm based on nonlinear approximation of contourlet transform and regional features | |
| CN106408580A (en) | Liver region extraction method based on fuzzy C mean and mean shift | |
| Chen et al. | A novel spatial–temporal detection method of dim infrared moving small target | |
| CN106504239A (en) | A kind of method of liver area in extraction ultrasonoscopy | |
| Potapov | Textures, fractals, scaling effects, and fractional operators as a basis of new methods for data processing and fractal radio system design | |
| CN109102503A (en) | It is a kind of based on color space smoothly and improve the significant model of frequency tuning high score image change detection method | |
| Xiaolan et al. | Texture Feature Extraction Method Combining Nonsubsampled Contour Transformation with Gray Level Co-occurrence Matrix. | |
| CN114066816B (en) | SAR image unsupervised change detection method based on static wavelet transformation extraction | |
| CN102693530B (en) | Synthetic aperture radar (SAR) image despeckle method based on target extraction and speckle reducing anisotropic diffusion (SRAD) algorithm |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170419 Termination date: 20180107 |