CN110264456A - A kind of plantar pressure method for registering images based on deep learning - Google Patents
A kind of plantar pressure method for registering images based on deep learning Download PDFInfo
- Publication number
- CN110264456A CN110264456A CN201910533477.9A CN201910533477A CN110264456A CN 110264456 A CN110264456 A CN 110264456A CN 201910533477 A CN201910533477 A CN 201910533477A CN 110264456 A CN110264456 A CN 110264456A
- Authority
- CN
- China
- Prior art keywords
- plantar pressure
- pressure image
- registration
- parameter
- fine tuning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of plantar pressure method for registering images based on deep learning, the present invention is the registration parameter for estimating to obtain plantar pressure image using concatenated convolutional neural net regression model, and plantar pressure image subject to registration is then estimated to obtain initial registration parameter using main shaft algorithm, so that the rotation angle of image subject to registration narrows down to a certain range, so as to subsequent secondary registration, and the concatenated convolutional neural network framework of design, it is divided into coarse adjustment network and fine tuning network, by different transformation parameters come generating source plantar pressure image, generated data set is used for model training, image subject to registration is sequentially input into the coarse adjustment network model and fine tuning network model trained later, the result that coarse adjustment network model and fine tuning network model export finally is overlapped combination, to obtain final plantar pressure figure As registration parameter, so that the efficiency of optimization registration parameter is significantly improved.
Description
Technical field
The present invention relates to image registration techniques field, specially a kind of plantar pressure image registration side based on deep learning
Method.
Background technique
Plantar pressure image registration plays an important role the analysis of statistics and biomethanics, in addition to human gait's
Except General Mechanics, plantar nervous arch is also provided for the researcher of medical domain and expert about foot structure and function
Important information.Therefore, all there is very much side in terms of the useful information of diagnosis foot discomfort, exploitation shoes and acquisition gait analysis
It helps.Plantar nervous arch can also compare before injured and non-wounded patient, wound and after wound or the four limbs load of surgical state.
In addition, it, which is good at, compares patient and control group, and provide the details for each contact area.
Nowadays it is available to the distribution of relevant pressure that there are many different technologies, for most in these technologies
Number, plantar pressure data can be converted discrete rectangular array.Therefore, image procossing and analytical technology are used directly for helping
Researcher and medical expert is helped to obtain relevant information from the numerical data of acquisition.
For plantar pressure data, it is also very desirable to which the method for registering images of precise operation, i.e., optimal alignment is by image
The method of the homologous structure of expression.For example, the registration for the researcher of medical domain and expert, between same subject
For accurately comparing damage or surgical state before and after patient's plantar pressure over time, wound, or establishes and accurately take off
It is extremely valuable for the model for the foot pressure distribution leted others have a look at.Registration between same body is for establishing foot database
It is essential, the map for the foot data being correctly aligned with common reference system can be stored, which can will be specific
Foot be compared with the foot of previous research, to help to be partitioned into from plantar pressure image the main region of foot, and energy
Automatic foot is enough supported to classify.
In addition to the advantages described above, plantar pressure image registration also supports Pixel-level to count, this makes from plantar pressure figure
It is more more efficient than obtaining information from traditional area technology currently in use that biomethanics relevant information is obtained as in, therefore
Need to develop a kind of full-automatic, plantar pressure image registration techniques accurately and quickly, it can provide speed but also provide essence
Degree, and enough plantar pressure image corresponding relationships can also be kept according to various similitudes or measure of dissimilarity standard,
So that researcher and medical expert can it is cumbersome needed for the traditional manual or semi-automatic registration solution used now and
It is freed in time-consuming task, so that plantar pressure image registration is more practical near real-time laboratory and clinical application；
And the existing method for registering based on plantar pressure image is mostly by iterative optimization procedure, to minimize registration error function
Method estimate registration parameter, and in order to obtain satisfactory registration accuracy, the calculation amount of these methods is very big and disappears
Time-consuming is very long.
In order to solve drawbacks described above, a kind of technical solution is now provided.
Summary of the invention
The purpose of the present invention is to provide a kind of plantar pressure method for registering images based on deep learning, the present invention be by
Convolutional neural networks are introduced into image registration, its object is to simulate the neural network of practical biology, and according to excellent spy
Learning ability is levied to learn to be embedded in robust features in the picture, for updating image registration parameter, to reach the foot of near real-time
Thus bottom pressure image registration solves the technologies such as existing plantar pressure method for registering images accuracy is low, elapsed time is long and asks
Topic.
The technical problems to be solved by the invention are as follows:
How a kind of effective mode is provided, and to solve, existing plantar pressure method for registering images accuracy is low, consumes
The situation of time length.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of plantar pressure method for registering images based on deep learning, includes the following steps:
A1: to acquire the continuous plantar pressure image of each frame when user's walking in real time according to pressure sensing mattress system,
And the plantar pressure image at per moment is to be acquired one by one, is not a complete plantar pressure image,
And plantar pressure image data base is established accordingly；
A2: appoint from plantar pressure image data base and take a plantar pressure image as template image, water is carried out to it
Translation is dynamic, vertical shift, rotates angle and zoom operations, and obtains plantar pressure picture number in source used in training pattern accordingly
According to collection；
A3: being first divided into training set and test set for source plantar pressure image data set, and training set accounts for data set total amount
90%, test set accounts for the 10% of data set total amount, and is to randomly select, then source plantar pressure image data set is input to grade
In the coarse adjustment network for joining convolutional neural networks frame, to train coarse adjustment network model, it is input to concatenated convolutional mind later
In fine tuning network through network frame, to train fine tuning network model, and the coarse adjustment network model trained and fine tuning are saved
Network model；
A4: it estimates plantar pressure image to be registered to obtain initial registration parameter using main shaft algorithm；
A5: the plantar pressure image for estimating to obtain initial registration parameter through main shaft algorithm is first input to the coarse adjustment trained
In network model, then it is input in fine tuning network model, and export registration parameter.
Further, the establishment process of the plantar pressure image data base are as follows:
S1: the continuous plantar pressure image of each frame when getting user's walking；
S2: the maximum pressure that each sensor in the plantar pressure image is born in sole is superimposed, and obtain one it is complete
Whole peak value plantar pressure image；
S3: repeat the above steps S1 and S2, and by all complete peak value plantar pressure image collections, to obtain vola
Pressure image data base.
Further, described move horizontally, vertical shift, rotate angle and zoom operations transformation parameter range are as follows: water
Translation is moved and the transformation parameter range of vertical shift is arranged between -5 to 5 pixels, and therein-number expression is put down to the left or downwards
It moves；The transformation parameter range for rotating angle is arranged at -5 ° to 5 °, and therein-number expression rotates clockwise；The transformation of zoom operations
Parameter area is arranged between 0.5 to 2.
Further, the coarse adjustment network includes nine sequentially connected layers, respectively three convolutional layers, three maximum ponds
Change layer and three full articulamentums；The fine tuning network includes seven sequentially connected layers, respectively two convolutional layers, two maximums
Pond layer and three full articulamentums.
Further, the main shaft algorithmic notation is to rotate plantar pressure image to be registered to make its spindle alignment template foot
The main shaft of bottom pressure image, and main shaft is calculated as the principal eigenvector of pressure weighting covariance matrix.
Further, the export registration parameter is expressed as coarse adjustment network model and fine tuning network model exports the folded of result
Add, and including moving horizontally, vertical shift, rotation angle and zoom operations totally four parameters.
Beneficial effects of the present invention:
The present invention is the registration parameter for estimating to obtain plantar pressure image using concatenated convolutional neural net regression model,
And plantar pressure image subject to registration is then estimated to obtain initial registration parameter using main shaft algorithm, so that image subject to registration
Rotation angle narrow down to a certain range, so as to subsequent secondary registration, and the concatenated convolutional neural network framework designed, point
For coarse adjustment network and fine tuning network, and the data set comprising 100,000 width source plantar pressure images obtained is used for model training,
Image subject to registration is sequentially input into the coarse adjustment network model and fine tuning network model trained later, finally by coarse adjustment network
Model and the result of fine tuning network model output are overlapped combination, to obtain final plantar pressure image registration parameter；This
The concatenated convolutional neural network method for registering of invention optimizes the accuracy rate of registration parameter since model is off-line training
It is higher, while real-time plantar pressure image registration is developed based on the model of convolutional neural networks using pre-training, also more
Has practicability.
Detailed description of the invention
In order to facilitate the understanding of those skilled in the art, the present invention will be further described below with reference to the drawings.
Fig. 1 be it is of the invention move horizontally, the effect signal of vertical shift, rotation angle and zoom operations transformation parameter
Figure；
Fig. 2 is the acquisition methods schematic diagram of coarse adjustment network model and fine tuning network model of the invention.
Specific embodiment
Technical solution of the present invention is clearly and completely described below in conjunction with embodiment, it is clear that described reality
Applying example is only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, this field is general
Logical technical staff all other embodiment obtained without creative efforts belongs to what the present invention protected
Range.
As shown in Figs. 1-2, a kind of plantar pressure method for registering images based on deep learning, wherein vola to be registered is pressed
Try hard to as from 30 young students, including 25 males and 5 women, male's average age therein is 20.8 ± 4.3
Year, women average age are 25.6 ± 2.3 years old；Each subject is required in pressure sensing mattress system, i.e. Germany Zebris
Six autonomous in total are carried out with normal speed in FDM-S system, sensing area is 54.2 × 33.9cm, physical resolution is
1.4 sensors/cm2, sample frequency 100Hz；
Template plantar pressure image is that the plantar pressure that an age is 24 years old male is had chosen from this 30 subjects
Image, and establish comprising 100,000 width source plantar pressure images data set and to pass through the transformation institute of template plantar pressure image
, and the transformation of two-dimentional plantar pressure image is indicated by four parameters, respectively zoom operations parameter tα, move horizontally parameter tx、
Vertical shift parameter tyAnd rotation angle parameter tθ, horizontal and vertical moving parameter tx,ty- 5 pixels are arranged on to 5 pixels
In the range of, therein-number expression translates to the left or downwards, zoom operations parameter tαSize be arranged between 0.5 to 2,
Rotation angle parameter tθFrom -5 ° to 5 °, therein-number expression rotates clockwise range；Last every width source plantar pressure image is all
It is cut into the image that unified size is 71 × 61, and randomly selects from data set 90% sample as training data,
Remaining sample is used as verify data；
And a kind of plantar pressure method for registering images based on deep learning, it is made of two stages being linked in sequence, the
One stage obtains initial registration parameter for estimating, and executes preliminary alignment, and second stage is used based on concatenated convolutional mind
It estimates to obtain specific registration parameter through net regression model；And " C ", " MP " and " FC " in Fig. 2 respectively indicate convolutional layer, maximum
Pond layer and full articulamentum；The output dimension of the characteristic pattern quantity or full articulamentum in digital representation convolutional layer before "@",
The size of digital representation characteristic pattern after "@", the kernel parameter of the corresponding operation of digital representation in bracket；
Plantar pressure image subject to registration moves horizontally parameter tx, vertical shift parameter tyWith zoom operations parameter tαIt can be
A small range, which is estimated, to be obtained, and rotation angle parameter tθCoverage but may be very big, in order to by rotation angle parameter tθLimit
For system in a small range, the first stage of the invention is first to rotate plantar pressure image to be registered, makes its spindle alignment
The main shaft of template plantar pressure image, main shaft are calculated as the principal eigenvector of pressure weighting covariance matrix, then pass through this
The small parameter space of a little individually range spans can guarantee quickly recurrence and accurately registration；
And the regression model of second stage of the invention is the two-stage cascade neural network based on convolutional neural networks, this time
Returning model includes coarse adjustment network and fine tuning network；The approximation that coarse adjustment network can provide correct plantar pressure image registration parameter is estimated
It calculates, fine tuning network is used to finely tune the plantar pressure image registration parameter obtained in previous stage, final plantar pressure image registration
Parameter is the stack combinations of coarse adjustment network and the output of this two-level network of fine tuning network；
Coarse adjustment network is also known as the first order network of concatenated convolutional neural network, it will be by the 100000 of template image transformation
Input of the width source plantar pressure image as network, and being made of nine sequentially connected layers, including three convolutional layers, three most
Great Chiization layer and three full articulamentums；And assume x ∈ RhIndicate that plantar pressure image has h pixel, T (x) ∈ R4Indicate four foots
The registration parameter of bottom pressure image, then the target of registration parameter estimation problem is that study is empty from plantar pressure image space to parameter
Between Nonlinear Mapping relationship F:F:x → T (x) minimum mean square error can be translated into and according to nonlinear function F model
Difference function is defined as:G indicates the quantity of training sample, y in formulaiIt is i-th of training
The label of sample, F (xi；W) output of i-th of training sample in the first order network of concatenated convolutional neural network is indicated,
The label of training sample is the true value of plantar pressure image registration parameter, table in the first order network of concatenated convolutional neural network
It is shown as
Due to search step limited in parameter space, coarse adjustment network model can only provide plantar pressure image registration parameter
Initial estimation, in order to further adjust plantar pressure image registration parameter, the registration parameter of previous stage network estimation and ground
True value tgtBetween residual error may be used as fine tuning model in recurrence label；
Fine tuning network is also known as the second level network of concatenated convolutional neural network, it is by two convolutional layers, two maximum ponds
Change layer and three full articulamentum compositions；It can learn the mapping function from plantar pressure image space to registration parameter difference space
Lk, the following loss function of fine tuning network minimum:G indicates the number of training sample in formula
Amount, Δ yi (1)It is label, indicates the difference between the output and its true value of i-th of training sample first order, L (xi；It w) is i-th
Output of a training sample in fine tuning network, the label of fine tuning network are ground true value tgtWith matching for previous stage network estimation
Residual error between quasi- parameter is defined as follows shown: Δ y(1)=tgt-tini, tiniIt is the output of first order network, two-stage returns net
The final output of network is as follows: test=tini+Δt1, wherein Δ t1It is the output of second level network；
Training for two above different stage network carries out in sequence, first passes through minimum equationIn objective function train coarse adjustment network, fine tuning network later is according to formulaIn target be trained to, reuse Adam algorithm optimize network parameter with minimize
Square error loss function；
From everyone 6 effective samples of acquisition, randomly selected in the plantar pressure image data of total 180 effective samples
Partial data compares test, in order to evaluate registration performance, is assessed using four indexs, respectively mean square error MSE, different
Or XOR, mutual information MI and speed, different method for registering images is respectively adopted and carries out 5 retests, to show that the present invention can
The Quasi velosity of matching of high registration accuracy and near real-time is provided, and returns device by training convolutional neural networks to disclose image space
To the mapping relations of parameter space, to efficiently solve image registration problem.
Above content is only to structure of the invention example and explanation, affiliated those skilled in the art couple
Described specific embodiment does various modifications or additions or is substituted in a similar manner, without departing from invention
Structure or beyond the scope defined by this claim, is within the scope of protection of the invention.
Claims (6)
1. a kind of plantar pressure method for registering images based on deep learning, which comprises the steps of:
A1: carrying out the continuous plantar pressure image of each frame when acquiring user's walking in real time according to pressure sensing mattress system, and according to
This establishes plantar pressure image data base；
A2: appoint from plantar pressure image data base and take a plantar pressure image as template image, horizontal shifting is carried out to it
Dynamic, vertical shift, rotation angle and zoom operations, and plantar pressure image data in source used in training pattern is obtained accordingly
Collection；
A3: being first divided into training set and test set for source plantar pressure image data set, and training set account for data set total amount 90%,
Test set accounts for the 10% of data set total amount, and is to randomly select, then source plantar pressure image data set is input to cascade volume
In the coarse adjustment network of product neural network framework, to train coarse adjustment network model, it is input to concatenated convolutional nerve net later
In the fine tuning network of network frame, to train fine tuning network model, and the coarse adjustment network model and fine tuning network trained is saved
Model；
A4: it estimates plantar pressure image to be registered to obtain initial registration parameter using main shaft algorithm；
A5: the plantar pressure image for estimating to obtain initial registration parameter through main shaft algorithm is first input to the coarse adjustment network trained
In model, then it is input in fine tuning network model, and export registration parameter.
2. a kind of plantar pressure method for registering images based on deep learning according to claim 1, which is characterized in that institute
State the establishment process of plantar pressure image data base are as follows:
S1: the continuous plantar pressure image of each frame when getting user's walking；
S2: each sensor in the plantar pressure image is superimposed in the maximum pressure that sole is born, and obtains one completely
Peak value plantar pressure image；
S3: repeat the above steps S1 and S2, and by all complete peak value plantar pressure image collections, to obtain plantar pressure
Image data base.
3. a kind of plantar pressure method for registering images based on deep learning according to claim 1, which is characterized in that institute
State move horizontally, vertical shift, rotate angle and zoom operations transformation parameter range are as follows: move horizontally the change with vertical shift
It changes parameter area to be arranged between -5 to 5 pixels, therein-number expression translates to the left or downwards；Rotate the transformation parameter of angle
Range is arranged at -5 ° to 5 °, and therein-number expression rotates clockwise；The transformation parameter range of zoom operations is arranged 0.5 to 2
Between.
4. a kind of plantar pressure method for registering images based on deep learning according to claim 1, which is characterized in that institute
Stating coarse adjustment network includes nine sequentially connected layers, respectively three convolutional layers, three maximum pond layers and three full articulamentums；
The fine tuning network includes seven sequentially connected layers, respectively two convolutional layers, two maximum pond layers and three full connections
Layer.
5. a kind of plantar pressure method for registering images based on deep learning according to claim 1, which is characterized in that institute
Stating main shaft algorithmic notation is the main shaft that rotation plantar pressure image to be registered makes its spindle alignment template plantar pressure image, and
Main shaft is calculated as the principal eigenvector of pressure weighting covariance matrix.
6. a kind of plantar pressure method for registering images based on deep learning according to claim 1, which is characterized in that institute
The superposition that export registration parameter is expressed as coarse adjustment network model and fine tuning network model output result is stated, and including moving horizontally,
Vertical shift rotates angle and zoom operations totally four parameters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910533477.9A CN110264456A (en) | 2019-06-19 | 2019-06-19 | A kind of plantar pressure method for registering images based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910533477.9A CN110264456A (en) | 2019-06-19 | 2019-06-19 | A kind of plantar pressure method for registering images based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110264456A true CN110264456A (en) | 2019-09-20 |
Family
ID=67919535
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910533477.9A Pending CN110264456A (en) | 2019-06-19 | 2019-06-19 | A kind of plantar pressure method for registering images based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110264456A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112070695A (en) * | 2020-09-03 | 2020-12-11 | 深圳大学 | Correction method of registration matrix and computer equipment |
| CN112396014A (en) * | 2020-11-26 | 2021-02-23 | 河北工业大学 | Visual-touch fusion gait recognition method based on feature fusion |
| CN113222875A (en) * | 2021-06-01 | 2021-08-06 | 浙江大学 | Image harmonious synthesis method based on color constancy |
-
2019
- 2019-06-19 CN CN201910533477.9A patent/CN110264456A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| YI XIA等: "A convolutional neural network Cascade for plantar pressure images registration", 《GAIT & POSTURE》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112070695A (en) * | 2020-09-03 | 2020-12-11 | 深圳大学 | Correction method of registration matrix and computer equipment |
| CN112396014A (en) * | 2020-11-26 | 2021-02-23 | 河北工业大学 | Visual-touch fusion gait recognition method based on feature fusion |
| CN113222875A (en) * | 2021-06-01 | 2021-08-06 | 浙江大学 | Image harmonious synthesis method based on color constancy |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110264456A (en) | A kind of plantar pressure method for registering images based on deep learning | |
| CN111145170B (en) | Medical image segmentation method based on deep learning | |
| CN111047594B (en) | Tumor MRI weak supervised learning analysis modeling method and model thereof | |
| CN112017198B (en) | Right ventricle segmentation method and device based on self-attention mechanism multi-scale features | |
| Fu et al. | StoHisNet: A hybrid multi-classification model with CNN and Transformer for gastric pathology images | |
| CN109886946B (en) | Deep learning-based early senile maculopathy weakening supervision and classification method | |
| CN112766142B (en) | Sole pressure image processing method, sole pressure image identification method and gait analysis system | |
| CN103020935A (en) | Self-adaption online dictionary learning super-resolution method | |
| CN110265142A (en) | A kind of assistant diagnosis system and method for lesion region restored map | |
| CN108734108A (en) | A kind of fissured tongue recognition methods based on SSD networks | |
| CN115375711A (en) | Image segmentation method of global context attention network based on multi-scale fusion | |
| CN114445339A (en) | Urinary calculus detection and classification method based on improved fast RCNN algorithm | |
| CN115272231A (en) | Non-proliferative diabetic retinopathy classification method | |
| CN113345578A (en) | Orthopedics rehabilitation information processing system based on CEST orthopedics rehabilitation clinical system | |
| CN113643261B (en) | Lung disease diagnosis method based on frequency attention network | |
| Yu et al. | HCT-net: hybrid CNN-transformer model based on a neural architecture search network for medical image segmentation | |
| Pouch et al. | Modeling the myxomatous mitral valve with three-dimensional echocardiography | |
| CN113870327A (en) | Medical image registration method based on multi-level deformation field prediction | |
| CN108898601A (en) | Femoral head image segmentation device and dividing method based on random forest | |
| Li et al. | 3D-Yoga: a 3D yoga dataset for visual-based hierarchical sports action analysis | |
| CN110660477A (en) | System and method for automatically screening and labeling helicobacter pylori | |
| CN109886988B (en) | Method, system, device and medium for measuring positioning error of microwave imager | |
| CN112200810A (en) | Multi-modal automated ventricular segmentation system and method of use thereof | |
| CN116309754A (en) | Brain medical image registration method and system based on local-global information collaboration | |
| CN111582330A (en) | Integrated ResNet-NRC method for dividing sample space based on lung tumor image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190920 |