CN101238993A - Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis - Google Patents
Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis Download PDFInfo
- Publication number
- CN101238993A CN101238993A CNA2008100639679A CN200810063967A CN101238993A CN 101238993 A CN101238993 A CN 101238993A CN A2008100639679 A CNA2008100639679 A CN A2008100639679A CN 200810063967 A CN200810063967 A CN 200810063967A CN 101238993 A CN101238993 A CN 101238993A
- Authority
- CN
- China
- Prior art keywords
- image
- registration
- wavelet
- integer
- rigid body
- 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
Images
Landscapes
- Image Processing (AREA)
Abstract
A non-rigid body image registering method, belongs to medicine ultrasonic image processing field. The invention provides a ultrasonic image registering method implemented by multi-resolution analysis and lamination based on integer lifting wavelet, directing towards the difficulty that the previous registering computation of non rigid body image is complicated and the requirement for memory is large. The registration image is subjected to two-layer decomposition using integer lifting wavelet and then the decomposed approximate image is subjected to overall rigid body registration of low resolution, furthermore the registered image is subjected to image reconstruction of original resolution using integer lifting wavelet, finally the local non-rigid body registration is performed at the original image resolution, therefore the final registered image is obtained. The method of multi-resolution analysis based on integer lifting wavelet can reduce the amount of calculation and the computation complexity and registration accuracy are unified using the method of multi-layer decomposition and fine and rough combination, therefore the amount of calculation is reduced without reducing the registration accuracy. The non-rigid body image registering method has wide application prospect in ultrasound imaging field.
Description
Technical field
The present invention relates to the medical ultrasound image technical field, relate in particular to a kind of method for registering of medical ultrasonic image.
Background technology
Medical ultrasound image is the very wide diagnostic method of a kind of scope of application, has not damaged, no ionizing radiation, advantage such as easy to use.Image registration has important clinical application value, not only can be used for the diagnosis of disease character, can also guide therapeutic process and therapeutic effect is made evaluation by lesions position is followed the tracks of.In image co-registration, 3D image reconstruction and surgical navigation, the extensive use method for registering comes the spatial relation between the positioning image.
Medical figure registration is meant for a width of cloth medical image (being called floating image), seek a kind of or a series of spatial alternation, make it and corresponding point on another width of cloth medical image (being called reference picture) reach consistent on the space, what require here to reach consistently is meant all anatomic points of human body or is that the locus of anatomic points in two width of cloth images with diagnostic significance is identical at least.
Image registration is generally by reference picture and floating image are carried out feature space (comprising characteristic point or characteristic curve/curved surface or pixel intensity etc.) extraction (its objective is the feature of using in the selection registration) respectively, selecting the search volume (is the image transformation type, comprise rigid body translation, affine transformation, projective transformation and nonlinear transformation etc.) with transformation parameter and calculate the changing image of floating image under given conversion, similarity between the floating image after calculating reference picture and the conversion is (as the gray scale mean deviation, correlation coefficient, mutual information etc.), according to similarity transformation parameter is carried out optimized search, the similarity registration that meets the demands is then finished, similarity does not meet the demands and reselects transformation parameter, till iteration arrival similarity meets the demands.
Because ultrasonoscopy, detects Shi doctor simultaneously to detect elastomers such as soft tissue, blood flow and often applies certain external force according to own judgement to image vision, so ultrasonoscopy is typical non-rigid body image, and the general difficulty of its registration is bigger.
Non-rigid registration can be applied in the registration of the same imaging pattern of same target, is used for disease is followed the tracks of or the sequence image that different time obtains is compared research.Follow the tracks of as radiotherapy,, when moving into or shifting out scanner, can produce non-rigid shape deformations for non-rigid structure or organs such as chest, abdominal part or breast to the cancer patient; Non-rigid registration can also be applied between the different objects, non-rigid registration is not a physical deformation of considering anatomical structure in this case, but consider the anatomical structure difference of Different Individual, and the anatomical structure between this individuality need utilize non-rigid registration to realize in size, in shape difference usually.
The registration of present non-rigid body image mainly contains two class methods, one class is based on the method for physical model, as strain physical model, viscous fluid physical model, light stream physical model etc., these class methods are mainly passed through the distortion of partial differential equation image, find the solution by iterative approximation and seek transformation parameter; Another kind ofly be based on parameterized spatial transform method, as the polynomial transformation model of high order (secondary, three times, four times, five inferior), spline function transformation model, RBF transformation model etc., these class methods mainly come match to be registered difference between image by function approximation, adopt to optimize algorithm and seek transformation parameter.At present because the difficulty that partial differential equation are found the solution is the main means of non-rigid image registration based on parameterized spatial transform method.
For parameterized non-rigid image registration, the amount of calculation of registration Algorithm generally is very big, in order to improve registration efficient, can adopt the layering registration strategies of multiresolution analysis to carry out image registration.For given image, along with the reduction of resolution, the details of this image is removed gradually, and remaining approximate image has just constituted the multi-resolution representation of image, and it is carried out conversion and obvious reduction of searching transformation parameter amount of calculation meeting.The first generation wavelet transformation that is based on Fourier analysis that existing multiresolution analysis method is used is promptly analyzed problem from frequency domain.In actual applications, this wavelet transformation realizes by convolution, so calculation of complex, arithmetic speed are slow, bigger to the demand of internal memory; And these wavelet basiss all are floating numbers, and computing certainly leads in various degree distortion because the floating-point of computer rounds off in the quantizing process, is unfavorable for the harmless reconstruct of image.
Summary of the invention
The present invention is directed to non-in the past rigid body ultrasound image registration calculation of complex, difficulty that memory demand is big, a kind of ultrasound image registration method of realizing based on the multiresolution analysis layering of integer lifting wavelet is provided, this method is utilized integer lifting wavelet to treat registering images to carry out two-layer decomposition, then the approximate image after decomposing is carried out the overall rigid body registration of low resolution, image after utilizing integer lifting wavelet to the decomposition registration again carries out the former resolution reconstruct of image, under the former stage resolution ratio of image, carry out local elasticity's registration at last, thereby obtain final registering images.
Integer lifting wavelet is a second filial generation small echo, it has inherited the multiresolution characteristic of first generation small echo, but the definition of its wavelet basis is very flexible, not necessarily obtain by yardstick and translation, and any traditional small echo (also claiming first generation small echo) can be generated by second filial generation small echo structure by a certain definite basic small echo.The integer lifting algorithm directly the time (sky) territory in problem analysis, broken away from dependence to frequency domain, coefficient behind the wavelet transformation is an integer, adopt which kind of boundary extension mode irrelevant when the recovery quality of image and conversion, can carry out conversion to the arbitrary dimension image, lifting scheme allows original position calculating completely, promptly do not need add-in memories, the primary signal data can directly be replaced by wavelet coefficient, and its computation complexity is about half of original convolution algorithm, so the operation efficiency height; Lifting scheme can be mapped to integer from integer with the wave filter output of wavelet transformation simultaneously, is reducing computation complexity, is being easy to the hard-wired while, realizes the undistorted reconstruct of image, thereby improves the accuracy of image registration.
Utilize the multiresolution analysis of small echo can reduce size of images, reduce iterations, and then reach the purpose that reduces computation time.Utilize the integer lifting form of small echo that image is carried out two-layer decomposition, its reason be on computational efficiency and computational accuracy, do one compromise.If only do one deck wavelet decomposition, then size of images can not reduce too much, and the time that reduces also needs to do the decomposition and the reconstruct of multiresolution analysis, and the computational efficiency of whole algorithm is not had too big effect; And if do more multi-layered analysis owing to only the approximate image after decomposing is applied overall rigid body translation, front and back multilamellar conversion also can not produce bigger benefit to registration results, therefore selects directly floating image and reference picture to be done two-layer wavelet decomposition.
For the low-resolution image of doing wavelet decomposition, it is carried out overall rigid body registration, can obtain the approximation of the rigid body registration under two width of cloth image coarse resolutions.Because the conversion unknown parameter that comprises of rigid body registration is few, thereby the optimization algorithm of localization can search optimal value quickly, and the optimal value that searches simultaneously will be as the initial value of back elastic registrating.The computing cost of doing like this can be very little.
To the image behind the overall rigid body registration, utilize the integer lifting form of small echo that it is carried out the former resolution reconstruct of image, because the former image in different resolution of this moment has had the registration than rough grade, thereby this former stage resolution ratio image can be used as the initial pictures of next step elastic image registration, its objective is the extensive search that reduces parameter, to reduce computing cost.
Method for registering images provided by the invention is realized by following steps:
The first step utilizes the integer lifting form of small echo that it is carried out wavelet decomposition respectively floating image and reference picture
Wavelet arithmetic mainly contains division, predicts and upgrades three steps.Division is that data are divided into even number sequence and odd number sequence two parts; Prediction is that the forecast error that obtains is the high fdrequency component of conversion with even number sequence prediction odd number sequence, and the prediction link is called as antithesis and promotes; Renewal is to upgrade the even number sequence by forecast error, obtains the low frequency component of conversion, upgrades link and be called original lifting in promoting term.Prediction and renewal can repeat repeatedly, at last also may be again through a convergent-divergent step, and whole lifting scheme all is reversible.
Integer wavelet transformation is based upon on the basis of lifting scheme, carries out the round computing by the numerical value to prediction and renewal and realizes.Though this computing is equivalent to original wavelet filter coefficient is changed, but still the multiple dimensioned characteristic and the area coherence of small echo have been kept.Coefficient behind the integer wavelet transformation all is an integer, and this makes all coefficient amplitudes can not have the error of quantification and changes into binary representation.Numerical value before and after the integer wavelet transformation all is integer, nondestructively reconstruct, and need not consider the continuation mode of data.
(1) integer lifting wavelet decomposes
Use s
iThe expression primary signal, d
i 0Odd samples after the expression division, s
i 0Even samples after the expression division, s
I+1Expression is through the low frequency signal of integer lifting, d
I+1Detail signal behind the expression integer lifting, P and U are represented prediction and are upgraded operator.
Splitted purpose is that this two-part dependency is stronger with signal segmentation two parts that become to be mutually related, and the effect of cutting apart is good more.Concrete grammar is primary signal s
iOdd even order according to arranging is divided into two groups of sample set d by the mode that extracts at interval
i 0And s
i 0, its extraction formula is:
This extraction has fully taken into account the local dependency of signal, for subsequently prediction and renewal process provides the foundation.Prediction keeps even samples s
i 0Constant, utilize interpolation subdividing to predict odd samples d
i 0, odd samples d
i 0And the difference between the predictive value is called detail coefficients, promptly
P[n] be interpolation coefficient,
Be operating as rounding operation, its value is for being not more than the maximum integer of x, and " 1/2 " in the rounding operation is as correction value, and purpose is to eliminate owing to rounding the error of introducing, to guarantee the reconstruction fully of data, the d that obtains at last
I+1 0[n] is wavelet coefficient.
Renewal is that the high frequency coefficient after keeping predicting is constant, utilizes interpolation subdividing to upgrade even samples s
i 0, keep the integer characteristic (average, vanishing moment etc.) of raw data set constant by introducing a linear operator that upgrades.Its more new formula be:
From the angle of frequency analysis, wavelet coefficient d
I+1Be called as high frequency coefficient, the radio-frequency component of expression initial data is promptly from s
I+1Return to s
iThe time needed details component; And corresponding s
I+1Be called as the lifting coefficient, the low-frequency component of expression initial data.As seen, make the data after the conversion not have energy to disappear, can be resumed fully, and the calculating of each link all can carry out on former data are, and then save Installed System Memory based on the integer wavelet transformation that promotes structure.
(2) integer lifting wavelet reconstruct
For the Lifting Wavelet algorithm, if obtained forward transform, just can obtain transformation by reciprocal direction immediately, that need make just changes the prediction and the plus-minus symbol of new portion more, and this is a good characteristic of boosting algorithm.
The reconstruction formula of integer lifting wavelet is:
(3) prediction and renewal operator
Prediction is to utilize the even number sequence samples to predict the odd number sequence samples, in boosting algorithm, prediction has the effect of two aspects: the first, can represent data with the form of compactness, in general, because signal s[n] all have a local dependency, therefore, the numerical value of forecast error is always much smaller than odd samples, that is to say, represent signal s[n with even samples and error sample], than represent the many of compactness together with even samples and odd samples; It two is to isolate signal s[n in spatial domain] high fdrequency component, when prediction, predicted their intermediate point (odd samples sequence) owing to always use a smoothed curve of odd number sequence and even number sequence samples, here smoothly mean low frequency, forecast error then means signal s[n] in the error of a regional area and own low frequency component, so forecast error can be regarded signal s[n as] high fdrequency component.Exactly because also like this, in Via Lifting Scheme, forecast error is referred to as the wavelet coefficient of signal.
Renewal is to revise even samples with forecast error, so that revised even samples only comprises signal s[n] low-frequency component.The process of upgrading intuitively can be understood as with a level and smooth matched curve to come signal s[n] carry out envelope, from mathematics, to make updated sample and primary signal s[n exactly] have an identical low order vanishing moment.Like this, more new samples is the level and smooth match of original signal, and the sample after therefore upgrading is called as the low frequency component of original signal.
Method for improving design small echo has very big motility, can design the predictive operator of any linearity, non-linear, spatially-variable in principle or upgrade operator to satisfy the application of different occasions.
In order to reach the purpose of further reduction image registration amount of calculation, can adopt the form of multilamellar wavelet decomposition, promptly after once decomposing, implement again to decompose, consider computation time and precision compromise be generally two-layer.
In second step, the low-resolution image after decomposing is carried out the thick registration of overall rigid body
Result after floating image that the first step is obtained and reference picture decompose, i.e. the image of two width of cloth low resolution carries out the approximation of the approximate registration of overall rigid body, and with the initial value of this registration results as subsequent registration.Overall situation rigid body method for registering can adopt general rigid body method for registering, as utilizes nonlinear interaction coefficient to estimate as registration, and the formula simplex method is carried out unknown parameter optimization in conjunction with going down the hill.Because the conversion unknown parameter that the rigid body registration comprises is few, thereby the optimization algorithm of localization can search optimal value quickly.
In the 3rd step, the image behind the approximate registration of rigid body is carried out former resolution reconstruct
Behind the overall situation rigid body registration, the integer lifting form of utilizing small echo has been carried out the former resolution reconstruct of image to the image that decomposes behind the registration.At this moment former image in different resolution has had the registration than rough grade, thereby this former stage resolution ratio image can be used as the initial pictures of next step non-rigid image registration.The integer lifting wavelet restructing algorithm is with the relevant algorithm in the first step.
The 4th step, non-rigid image registration under the former stage resolution ratio of image
Under the former stage resolution ratio of image, be initial value with the approximate thick registration of rigid body image, floating image and reference picture are carried out the local smart registration of non-rigid body.Method for registering can adopt the non-rigid image registration method based on parameterized spatial alternation, as the parametrization method for registering based on RBF.Parametrization distortion based on RBF both can have been simulated overall non-rigid shape deformations, also can simulate partial non-rigid shape deformations, and therefore both having can be used for global registration also can be used for local registration.Quadratic function, contrary quadratic function, Gaussian function and thin plate spline function can be as the RBFs that uses in the image registration.
Four steps of registration of the present invention will be realized by the computer off-line that strong computing capability is arranged, be mainly used in the analysis and the diagnosis of image; Or, be mainly used in online auxiliary diagnosis by the auxiliary canbe used on line of embedded system.
Description of drawings
Fig. 1 is the theory diagram based on the medical ultrasound image registration method of integer lifting wavelet multiresolution analysis.
Fig. 2 decomposes thick smart coupling method for registering sketch map for layering.
Fig. 3 is the wavelet decomposition sketch map based on the integer lifting form.
Fig. 4 is the wavelet reconstruction sketch map based on the integer lifting form.
Fig. 5 is many resolution decomposition of the ultrasonoscopy flow chart based on Lifting Wavelet.
Fig. 6 is the thyroid ultrasonoscopy that three width of cloth need registration.
Fig. 7 is the iterations of the accurate image of two assembly under the different resolution rank.
Among Fig. 1:
The integer lifting wavelet decomposition algorithm of 101 reference pictures
The integer lifting wavelet decomposition algorithm of 102 floating images
103 decompose the overall rigid body registration Algorithm of back low-resolution reference image and floating image
104 usefulness integer lifting wavelet reconstruct realize former stage resolution ratio reference picture
105 usefulness integer lifting wavelet reconstruct realize former stage resolution ratio floating image
The local registration Algorithm of 106 high-definition pictures
Among Fig. 2:
The thick registration Algorithm of 201 overall rigid bodies
The smart registration Algorithm of 202 local non-rigid bodies
Among Fig. 3:
The branch split operator of 301 integer lifting wavelets
The predictive operator of 302 integer lifting wavelets
The renewal operator of 303 integer lifting wavelets
Among Fig. 4:
The renewal operator of 401 integer lifting wavelets
The predictive operator of 402 integer lifting wavelets
The merging operator of 403 integer lifting wavelets
Among Fig. 5:
Advanced every trade promotes, and is listed as lifting again
Among Fig. 6:
Be respectively the floating image of reference picture and two different distortion from left to right, size of images is 512 * 512.
The specific embodiment
Below in conjunction with accompanying drawing the specific embodiment of the present invention is described in further detail.
Whole registration process as shown in Figure 1.Reference picture subject to registration and floating image will pass through four processes of the local registration of non-rigid body under the thick registration of overall rigid body of wavelet decomposition, approximate image, the wavelet reconstruction of thick registration results image, the former stage resolution ratio.What adopt is two layers of thick smart coupling registration mode, and its method as shown in Figure 2.
The integer lifting wavelet of reference picture subject to registration and floating image decomposes, use be the integer lifting form of biorthogonal bior4.4 small echo, catabolic process is as shown in Figure 3.What image was carried out is two layers of wavelet decomposition, and Fig. 5 has provided the decomposition process figure of a width of cloth hepatic vein ultrasonoscopy.
The thick registration of overall rigid body of approximate image is finished under low resolution, the rigid body method for registering that is based on mutual information of use, and parameter search adopts the formula simplex method.
The wavelet reconstruction of thick registration results image, use be the integer lifting form of biorthogonal bior4.4 small echo, restructuring procedure is as shown in Figure 4.
The local registration of the non-rigid body of former image in different resolution is finished the parametrization non-rigid image registration method that is based on the RBF transformation model of employing under former stage resolution ratio.
Fig. 6 has provided three thyroid ultrasonoscopys, the floating image that be respectively reference picture from left to right, be out of shape less floating image, distortion is bigger, and size of images is 512 * 512.It is one group that the floating image that reference picture and distortion is less is compiled, and is called first group; It is one group that the floating image that reference picture and distortion is bigger is compiled, and is called second group.Carry out one deck and two layers of wavelet decomposition respectively to first group and second group, employing be the integer lifting form of biorthogonal bior4.4 small echo.Two groups of images are respectively under the situation that former stage resolution ratio, one deck decompose, under the situation of two layers of decomposition, carry out registration, the rigid body method for registering that is based on mutual information of employing, and parameter search adopts the formula simplex method.When Fig. 7 had provided two groups of image registrations respectively, under identical registration accuracy required, the needed iterations of parameter search during registration, this figure had illustrated along with the increase iterations that decomposes the number of plies reduces thereupon, that is the registration amount of calculation also will reduce.
Claims (4)
1. method that is used for medical ultrasound image registration, it is characterized in that: floating image subject to registration and reference picture are carried out the integer lifting wavelet multiresolution decomposition, image after decomposing is similar to the rigid body registration, image after utilizing integer lifting wavelet to the decomposition registration then carries out the former resolution reconstruct of image, under the former stage resolution ratio of image, carry out local non-rigid body registration at last, thereby obtain final registering images.
2. medical image registration method according to claim 1 is characterized in that: adopt the multilamellar integer lifting wavelet to decompose, with further minimizing amount of calculation, the number of plies of decomposition is that amount of calculation and registration accuracy one is compromise.
3. medical image registration method according to claim 1, it is characterized in that: utilize small echo integer lifting algorithm directly the time (sky) territory in problem analysis, coefficient behind the wavelet transformation is an integer, lifting scheme allows original position calculating completely, do not need add-in memories, the primary signal data can directly be replaced by wavelet coefficient, and its computation complexity greatly reduces, amount of calculation greatly reduces, and are easy to hardware simultaneously and realize.
4. medical image registration method according to claim 1, it is characterized in that: adopt the integer lifting wavelet method, adopt which kind of boundary extension mode irrelevant when the recovery quality of image and conversion, can carry out conversion to the arbitrary dimension image, can realize the undistorted reconstruct of image, create condition for the thick smart coupling registration of layering, also improved the accuracy of image registration simultaneously.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2008100639679A CN101238993A (en) | 2008-02-01 | 2008-02-01 | Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2008100639679A CN101238993A (en) | 2008-02-01 | 2008-02-01 | Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101238993A true CN101238993A (en) | 2008-08-13 |
Family
ID=39931018
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2008100639679A Pending CN101238993A (en) | 2008-02-01 | 2008-02-01 | Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101238993A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102169578A (en) * | 2011-03-16 | 2011-08-31 | 内蒙古科技大学 | Non-rigid medical image registration method based on finite element model |
| CN103429157A (en) * | 2011-03-07 | 2013-12-04 | 爱克发医疗保健公司 | Radiographic imaging method and apparatus |
| CN103679720A (en) * | 2013-12-09 | 2014-03-26 | 北京理工大学 | Fast image registration method based on wavelet decomposition and Harris corner detection |
| CN108143611A (en) * | 2017-12-15 | 2018-06-12 | 泗洪县正心医疗技术有限公司 | It is a kind of that the method for realizing Shu Points orientation is projected based on vein |
| CN108198235A (en) * | 2017-12-25 | 2018-06-22 | 中国科学院深圳先进技术研究院 | A kind of three dimentional reconstruction method, apparatus, equipment and storage medium |
| CN108472015A (en) * | 2015-12-22 | 2018-08-31 | 皇家飞利浦有限公司 | The medical imaging apparatus and medical imaging procedure of volume for check object |
| CN108537723A (en) * | 2018-04-08 | 2018-09-14 | 华中科技大学苏州脑空间信息研究院 | The three dimensional non-linear method for registering and system of magnanimity brain image data collection |
| CN110223220A (en) * | 2019-06-14 | 2019-09-10 | 北京百度网讯科技有限公司 | A kind of method and apparatus handling image |
| CN110415279A (en) * | 2019-06-25 | 2019-11-05 | 北京全域医疗技术集团有限公司 | Method for registering images, device and equipment |
| CN110517299A (en) * | 2019-07-15 | 2019-11-29 | 温州医科大学附属眼视光医院 | Elastic image registration algorithm based on local feature entropy |
| CN112616058A (en) * | 2019-10-03 | 2021-04-06 | 腾讯美国有限责任公司 | Video encoding or decoding method, apparatus, computer device and storage medium |
| CN113448232A (en) * | 2021-07-06 | 2021-09-28 | 哈尔滨理工大学 | Measurement matrix dimension reduction method for three-dimensional layered target compression holography |
-
2008
- 2008-02-01 CN CNA2008100639679A patent/CN101238993A/en active Pending
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103429157A (en) * | 2011-03-07 | 2013-12-04 | 爱克发医疗保健公司 | Radiographic imaging method and apparatus |
| CN103429157B (en) * | 2011-03-07 | 2017-01-18 | 爱克发医疗保健公司 | radiographic imaging method and apparatus |
| CN102169578B (en) * | 2011-03-16 | 2013-02-13 | 内蒙古科技大学 | Non-rigid medical image registration method based on finite element model |
| CN102169578A (en) * | 2011-03-16 | 2011-08-31 | 内蒙古科技大学 | Non-rigid medical image registration method based on finite element model |
| CN103679720A (en) * | 2013-12-09 | 2014-03-26 | 北京理工大学 | Fast image registration method based on wavelet decomposition and Harris corner detection |
| CN108472015A (en) * | 2015-12-22 | 2018-08-31 | 皇家飞利浦有限公司 | The medical imaging apparatus and medical imaging procedure of volume for check object |
| CN108143611B (en) * | 2017-12-15 | 2021-09-17 | 苏州科灵医疗科技有限公司 | Manufacturing method of acupoint positioning model based on vein projection |
| CN108143611A (en) * | 2017-12-15 | 2018-06-12 | 泗洪县正心医疗技术有限公司 | It is a kind of that the method for realizing Shu Points orientation is projected based on vein |
| CN108198235A (en) * | 2017-12-25 | 2018-06-22 | 中国科学院深圳先进技术研究院 | A kind of three dimentional reconstruction method, apparatus, equipment and storage medium |
| CN108198235B (en) * | 2017-12-25 | 2022-03-04 | 中国科学院深圳先进技术研究院 | Three-dimensional ultrasonic reconstruction method, device, equipment and storage medium |
| CN108537723A (en) * | 2018-04-08 | 2018-09-14 | 华中科技大学苏州脑空间信息研究院 | The three dimensional non-linear method for registering and system of magnanimity brain image data collection |
| CN108537723B (en) * | 2018-04-08 | 2021-09-28 | 华中科技大学苏州脑空间信息研究院 | Three-dimensional nonlinear registration method and system for massive brain image data sets |
| CN110223220A (en) * | 2019-06-14 | 2019-09-10 | 北京百度网讯科技有限公司 | A kind of method and apparatus handling image |
| CN110415279A (en) * | 2019-06-25 | 2019-11-05 | 北京全域医疗技术集团有限公司 | Method for registering images, device and equipment |
| CN110517299A (en) * | 2019-07-15 | 2019-11-29 | 温州医科大学附属眼视光医院 | Elastic image registration algorithm based on local feature entropy |
| CN110517299B (en) * | 2019-07-15 | 2021-10-26 | 温州医科大学附属眼视光医院 | Elastic image registration algorithm based on local feature entropy |
| CN112616058A (en) * | 2019-10-03 | 2021-04-06 | 腾讯美国有限责任公司 | Video encoding or decoding method, apparatus, computer device and storage medium |
| CN113448232A (en) * | 2021-07-06 | 2021-09-28 | 哈尔滨理工大学 | Measurement matrix dimension reduction method for three-dimensional layered target compression holography |
| CN113448232B (en) * | 2021-07-06 | 2022-06-21 | 哈尔滨理工大学 | Measurement matrix dimension reduction method for three-dimensional layered target compression holography |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101238993A (en) | Medical ultrasound image registration method based on integer lifting wavelet multiresolution analysis | |
| Kazerouni et al. | Diffusion models in medical imaging: A comprehensive survey | |
| Tang et al. | High-resolution 3D abdominal segmentation with random patch network fusion | |
| Iqbal et al. | MDA-Net: Multiscale dual attention-based network for breast lesion segmentation using ultrasound images | |
| CN100456323C (en) | Registration method of three dimension image | |
| CN111260705B (en) | Prostate MR image multi-task registration method based on deep convolutional neural network | |
| CN103295234B (en) | Based on the medical image segmentation system and method for deformation surface model | |
| US20090324041A1 (en) | Apparatus for real-time 3d biopsy | |
| CN114119515B (en) | Brain tumor detection method based on attention mechanism and MRI multi-mode fusion | |
| EP4030385A1 (en) | Devices and process for synthesizing images from a source nature to a target nature | |
| CN112634265B (en) | Method and system for constructing and segmenting fully-automatic pancreas segmentation model based on DNN (deep neural network) | |
| CN111063441A (en) | Liver deformation prediction method and system and electronic equipment | |
| CN116645380A (en) | Automatic segmentation method for esophageal cancer CT image tumor area based on two-stage progressive information fusion | |
| CN104166994B (en) | A kind of bone suppressing method optimized based on training sample | |
| Wu et al. | Image synthesis in contrast MRI based on super resolution reconstruction with multi-refinement cycle-consistent generative adversarial networks | |
| CN117274147A (en) | Lung CT image segmentation method based on mixed Swin Transformer U-Net | |
| CN118037791A (en) | Construction method and application of multi-mode three-dimensional medical image segmentation registration model | |
| Singh et al. | The role of geometry in convolutional neural networks for medical imaging | |
| Wang et al. | TT-Net: Tensorized Transformer Network for 3D medical image segmentation | |
| Ayu et al. | U-Net Tuning Hyperparameter for Segmentation in Amniotic Fluid Ultrasonography Image | |
| CN116433734A (en) | Registration method for multi-mode image guided radiotherapy | |
| CN114511602A (en) | Medical image registration method based on graph convolution Transformer | |
| CN116092643A (en) | Interactive semi-automatic labeling method based on medical image | |
| Song | Algorithm and hardware co-optimization for image segmentation in wearable ultrasound devices: Continuous bladder monitoring | |
| Huang et al. | Si-MSPDNet: A multiscale Siamese network with parallel partial decoders for the 3-D measurement of spines in 3D ultrasonic images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |