CN110517249A - Imaging method, device, equipment and the medium of ultrasonic elastic image - Google Patents
Imaging method, device, equipment and the medium of ultrasonic elastic image Download PDFInfo
- Publication number
- CN110517249A CN110517249A CN201910797347.6A CN201910797347A CN110517249A CN 110517249 A CN110517249 A CN 110517249A CN 201910797347 A CN201910797347 A CN 201910797347A CN 110517249 A CN110517249 A CN 110517249A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- lesion characteristics
- current
- training
- 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
- 238000003384 imaging method Methods 0.000 title claims abstract description 26
- 230000003902 lesion Effects 0.000 claims abstract description 149
- 238000012549 training Methods 0.000 claims abstract description 71
- 238000013528 artificial neural network Methods 0.000 claims abstract description 53
- 238000000034 method Methods 0.000 claims abstract description 44
- 238000012360 testing method Methods 0.000 claims description 66
- 238000003860 storage Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 201000010099 disease Diseases 0.000 claims description 5
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 5
- 230000003862 health status Effects 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims 1
- 238000006073 displacement reaction Methods 0.000 abstract description 12
- 238000004088 simulation Methods 0.000 abstract description 3
- 239000000523 sample Substances 0.000 description 48
- 210000001519 tissue Anatomy 0.000 description 21
- 230000008569 process Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 8
- 238000009826 distribution Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 6
- 230000002779 inactivation Effects 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 230000005291 magnetic effect Effects 0.000 description 5
- 238000012423 maintenance Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 241001269238 Data Species 0.000 description 3
- 238000013480 data collection Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 238000002604 ultrasonography Methods 0.000 description 3
- 206010028980 Neoplasm Diseases 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000004218 nerve net Anatomy 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000000149 penetrating effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004660 morphological change Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000000505 pernicious effect Effects 0.000 description 1
- 238000003825 pressing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 210000004872 soft tissue Anatomy 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Radiology & Medical Imaging (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Pathology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
This application provides a kind of imaging method of ultrasonic elastic image, device, equipment and media, comprising: using the self-learning capability of artificial neural network, establishes the corresponding relationship between rf data and lesion characteristics in ultrasonic elastic image；Obtain the current radio frequency data of the current ultrasonic elastic image of patient；By the corresponding relationship, current lesion characteristics corresponding with the current radio frequency data are determined；Specifically, it is determined that current lesion characteristics corresponding with the rf data, comprising: by lesion characteristics corresponding to rf data identical with the current radio frequency data in the corresponding relationship, be determined as the current lesion characteristics.Ultrasonic elastograph imaging directly can be rebuild by radiofrequency signal by this method；And it only can be inferred to true elastogram from true rf data as training data using calculating simulation to be distributed；The neural network framework of this method directly can generate displacement field and strain field from ultrasonic radio frequency data.
Description
Technical field
This application involves the imaging method of medical science, especially ultrasonic elastic image, device, equipment and media.
Background technique
Ultrasonic elastograph imaging is to be proposed first by Ophir for 1991, and as a kind of completely new imaging technique, it is compensated for
The deficiency of conventional Ultrasound, more can more vivo show, lesion and identify lesion nature.Ultrasonic elastograph imaging can study biography
The tumour and disseminated disease imaging that system ultrasound can not detect.The basis of this technology is the hardness or elastic and lesion group of tissue
The pathology knitted is closely related.Its basic principle are as follows: the coefficient of elasticity according to different target tissues (normal and lesion) is different, outer adding
Its difference for straining (predominantly morphologic change) after power or alternation vibration collects target tissue at the appointed time each in section
Break signal is shown by processing, then in a manner of black and white, pseudo- color or coloud coding, eventually by the interpretation to elastic image
Diagnose the good pernicious matter of target tissue or the characteristic of tissue.Under identical external force, coefficient of elasticity is big, caused to become smaller；
Conversely, coefficient of elasticity is small, should become larger accordingly.That is, normal tissue soft under same pressure condition deform more than it is hard
Tumor tissues.After applying an external force, compare the super of pressurization (pressing lesion with ultrasonic probe) front and back target tissue elastic information
The hardness of acoustic image, front and back lesion strained to illustrate target tissue, the latter say the important parameter for identifying pathological tissues property.Ultrasound
Elastic image is the diagnosis that disease is helped using the elastic information of biological tissue.
The technology of existing elastogram includes normalized crosscorrelation etc..Normalized crosscorrelation is the matching based on grayscale information
Method.The basic theory of NCC algorithm is that the similitude of image is attributed to the similitude of 2 vectors.For normalizing mesh to be measured
Target degree of correlation.It can be used for extracting the displacement of tissue characteristic information in two-dimentional rf data, and then speculate the strain of tissue.
However, the signal-to-noise ratio and RMSE of the methods of normalization correlation be not high；The methods of normalized crosscorrelation and light stream hold
Vulnerable to noise or the insufficient interference of characteristic information；The methods of normalized crosscorrelation can only provide related with two-dimentional rf data
Localized variation low-level information.
Summary of the invention
In view of described problem, the application is proposed in order to provide overcoming described problem or at least being partially solved described ask
Imaging method, device, equipment and the medium of the ultrasonic elastic image of topic, comprising:
A kind of imaging method of ultrasonic elastic image, comprising:
Using the self-learning capability of artificial neural network, establish rf data in ultrasonic elastic image and lesion characteristics it
Between corresponding relationship；
Obtain the current radio frequency data of the current ultrasonic elastic image of patient；
By the corresponding relationship, current lesion characteristics corresponding with the current radio frequency data are determined；Specifically, it is determined that
Current lesion characteristics corresponding with the rf data, comprising: will be identical as the current radio frequency data in the corresponding relationship
Rf data corresponding to lesion characteristics, be determined as the current lesion characteristics.
Further,
The rf data, comprising: one-dimensional rf data or two-dimentional rf data；
And/or
The corresponding relationship, comprising: functional relation；The rf data is the input parameter of the functional relation, described
Lesion characteristics are the output parameter of the functional relation；
Determine current lesion characteristics corresponding with the current radio frequency data, further includes:
When the corresponding relationship includes functional relation, the current radio frequency data are inputted in the functional relation, really
The output parameter of the fixed functional relation is current lesion characteristics.
Further, the step for establishing the corresponding relationship between rf data and lesion characteristics in ultrasonic elastic image
Suddenly, comprising:
Obtain the sample data of the corresponding relationship for establishing between the rf data and the lesion characteristics；
The characteristic and its rule for analyzing the rf data determine the artificial neuron according to the characteristic and its rule
The network structure and its network parameter of network；
Using the sample data, the network structure and the network parameter are trained and are tested, described in determination
The corresponding relationship of rf data and the lesion characteristics.
Further, the sample of corresponding relationship of the acquisition for establishing between the rf data and the lesion characteristics
The step of notebook data, comprising:
Collect the patient of different health status the rf data and the lesion characteristics；
The expertise information prestored is analyzed the rf data and combined, is chosen and the lesion characteristics phase
The data of pass are as the rf data；
The data pair that the lesion characteristics and the rf data chosen are constituted, as sample data.
Further,
The network structure, comprising: BP neural network, CNN neural network, RNN neural network, and, residual error nerve net
At least one of network；
And/or
The network parameter, comprising: intensive block number, the output number of plies, the convolution number of plies, the excessive number of plies, initial weight, and,
At least one of bias.
Further,
The network structure and the network parameter are trained, comprising:
A part of data in the sample data are chosen as training sample, by the radio frequency in the training sample
Data are input to the network structure, are trained, are obtained by the activation primitive and the network parameter of the network structure
Hands-on result；
Determine the hands-on error between the hands-on result and the corresponding lesion characteristics in the training sample
Whether satisfaction presets training error；
When the hands-on error meets the default training error, determine to the network structure and the network
The training of parameter is completed；
And/or
The network structure and the network parameter are tested, comprising:
Another part data in the sample data are chosen as test sample, will penetrated described in the test sample
Frequency is according to being input in the network structure that the training is completed, with the net of the activation primitive and the training completion
Network parameter is tested, and actual test result is obtained；
Determine the actual test error between the actual test result and the corresponding lesion characteristics in the test sample
Whether satisfaction sets test error；
When the actual test error meets the setting test error, determine to the network structure and the network
The test of parameter is completed.
Further,
The network structure and the network parameter are trained, further includes:
When the hands-on error is unsatisfactory for the setting training error, pass through the error energy of the network structure
Function updates the network parameter；
Re -training is carried out by the activation primitive and the updated network parameter of the network structure, until
Hands-on error after the re -training meets the setting training error；
And/or
The network structure and the network parameter are tested, further includes:
When the actual test error is unsatisfactory for the setting test error, the network structure and the network are joined
Number carries out re -trainings, until the actual test error setting test error at a slow speed after the re -training.
A kind of imaging device of ultrasonic elastic image, comprising:
Module is established, for the self-learning capability using artificial neural network, establishes the radio frequency number in ultrasonic elastic image
According to the corresponding relationship between lesion characteristics；
Obtain module, the current radio frequency data of the current ultrasonic elastic image for obtaining patient；
Determining module, for determining that current lesion corresponding with the current radio frequency data is special by the corresponding relationship
Sign；Specifically, it is determined that current lesion characteristics corresponding with the rf data, comprising: will work as in the corresponding relationship with described
Lesion characteristics corresponding to the identical rf data of preceding rf data, are determined as the current lesion characteristics.
A kind of equipment, including processor, memory and be stored on the memory and can transport on the processor
Capable computer program, the computer program realized when being executed by the processor ultrasonic elastic image as described above at
The step of image space method.
A kind of computer readable storage medium stores computer program, the meter on the computer readable storage medium
The step of calculation machine program realizes the imaging method of ultrasonic elastic image as described above when being executed by processor.
The application has the following advantages:
In embodiments herein, by the self-learning capability using artificial neural network, ultrasonic elastic image is established
In rf data and lesion characteristics between corresponding relationship；Obtain the current radio frequency number of the current ultrasonic elastic image of patient
According to；By the corresponding relationship, current lesion characteristics corresponding with the current radio frequency data are determined；Specifically, it is determined that with institute
State the corresponding current lesion characteristics of rf data, comprising: penetrate identical with the current radio frequency data in the corresponding relationship
Frequency is determined as the current lesion characteristics according to corresponding lesion characteristics, can directly pass through radiofrequency signal by this method
Rebuild ultrasonic elastograph imaging；And it can only be inferred to from true rf data really by training data of calculating simulation
Elastogram distribution；The neural network framework of this method directly can generate displacement field and strain field from ultrasonic radio frequency data.
Detailed description of the invention
It, below will be to attached needed in the description of the present application in order to illustrate more clearly of the technical solution of the application
Figure is briefly described, it should be apparent that, the drawings in the following description are only some examples of the present application, for this field
For those of ordinary skill, without any creative labor, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is a kind of step flow chart of the imaging method for ultrasonic elastic image that one embodiment of the application provides；
Fig. 2 is a kind of artificial network's structural representation of the imaging method for ultrasonic elastic image that one embodiment of the application provides
Figure；
Fig. 3 is the random inactivation schematic diagram that one embodiment of the application provides；
Fig. 4-a is the strain field estimation result schematic diagram for the model data that one specific embodiment of the application provides；
Fig. 4-b is the strain field estimation result schematic diagram for the model data that one specific embodiment of the application provides；
Fig. 5-a is the strain field estimation result schematic diagram for the model data that one specific embodiment of the application provides；
Fig. 5-b is the strain field estimation result schematic diagram for the model data that one specific embodiment of the application provides；
Fig. 6 is the strain field estimation result schematic diagram for the patient data that one specific embodiment of the application provides；
Fig. 7 is a kind of structural block diagram of the imaging device for ultrasonic elastic image that one embodiment of the application provides；
Fig. 8 is a kind of structural schematic diagram of computer equipment of one embodiment of the invention.
Specific embodiment
It is with reference to the accompanying drawing and specific real to keep the objects, features and advantages of the application more obvious and easy to understand
Applying mode, the present application will be further described in detail.Obviously, described embodiment is some embodiments of the present application, without
It is whole embodiments.Based on the embodiment in the application, those of ordinary skill in the art are not before making creative work
Every other embodiment obtained is put, shall fall in the protection scope of this application.
As shown in Figure 1, showing a kind of imaging method of ultrasonic elastic image of one embodiment of the application offer, comprising:
S110, the self-learning capability using artificial neural network establish rf data and lesion in ultrasonic elastic image
Corresponding relationship between feature；
S120, obtain patient current ultrasonic elastic image current radio frequency data；
S130, pass through the corresponding relationship, determining current lesion characteristics corresponding with the current radio frequency data；Specifically
Ground determines corresponding with the rf data current lesion characteristics, comprising: by the corresponding relationship with the current radio frequency number
According to lesion characteristics corresponding to identical rf data, it is determined as the current lesion characteristics.
In embodiments herein, by the self-learning capability using artificial neural network, ultrasonic elastic image is established
In rf data and lesion characteristics between corresponding relationship；Obtain the current radio frequency number of the current ultrasonic elastic image of patient
According to；By the corresponding relationship, current lesion characteristics corresponding with the current radio frequency data are determined；Specifically, it is determined that with institute
State the corresponding current lesion characteristics of rf data, comprising: penetrate identical with the current radio frequency data in the corresponding relationship
Frequency is determined as the current lesion characteristics according to corresponding lesion characteristics, can directly pass through radiofrequency signal by this method
Rebuild ultrasonic elastograph imaging；And it can only be inferred to from true rf data really by training data of calculating simulation
Elastogram distribution；The neural network framework of this method directly can generate displacement field and strain field from ultrasonic radio frequency data.
In the following, by being further described to the imaging method of ultrasonic elastic image in the present exemplary embodiment.
As described in above-mentioned steps S110, using the self-learning capability of artificial neural network, establish in ultrasonic elastic image
Corresponding relationship between rf data and lesion characteristics.
Such as: the display state in the corresponding ultrasonic elastic image of lesion characteristics is analyzed using artificial neural network algorithm
Rule finds rf data and disease in patient's ultrasonic elastic image by the self study of artificial neural network, adaptive characteristic
Become the mapping principle between feature.
Such as: it can use artificial neural network algorithm, by (including but not limited to following to a large amount of different volunteers
It is one or more: the age, if to have the state of an illness, gender, patient's condition etc.) ultrasonic elastic image in rf data summarize collection, select
Take rf data in the ultrasonic elastic image of several volunteers and lesion characteristics as sample data, to neural network
It practises and trains, by adjusting the weight between network structure and network node, make penetrating in neural network fitting ultrasonic elastic image
Relationship between frequency evidence and lesion characteristics, finally enables neural network accurately fit in the ultrasonic elastic image of different patients
Rf data and lesion characteristics corresponding relationship.
In one embodiment, the rf data, comprising: one-dimensional rf data or two-dimentional rf data；
Optionally, the corresponding relationship, comprising: functional relation；The rf data is that the input of the functional relation is joined
Number, the lesion characteristics are the output parameter of the functional relation；
Preferably, the rf data is the input parameter of the functional relation, and the lesion characteristics are function pass
The output parameter of system；
As a result, by the corresponding relationship of diversified forms, the flexibility and convenient determined to current lesion characteristics can be promoted
Property.
In one embodiment, it can further illustrate in step S110 and " establish in ultrasonic elastic image in conjunction with following description
Rf data and lesion characteristics between corresponding relationship " detailed process.
As described in the following steps: obtaining the corresponding relationship for establishing between the rf data and the lesion characteristics
Sample data；
In an advanced embodiment, it can further illustrate and " obtain for establishing the rf data in conjunction with following description
The detailed process of the sample data of corresponding relationship between the lesion characteristics ".
As described in the following steps: collect the patient of different health status the rf data and the lesion characteristics；
Such as: data collection: collect the patient of different health status rf data and corresponding lesion characteristics；And
The rf data of the patient of collection all ages and classes and corresponding lesion characteristics；And collect the radio frequency number of the patient of different sexes
According to and corresponding lesion characteristics.
It collects operation data through a variety of ways as a result, is conducive to the amount for increasing operation data, promotes artificial neural network
Learning ability, and then promote the accuracy and reliability of determining corresponding relationship.
As described in the following steps: analyzed the rf data and combined the expertise information prestored, choose with
The relevant data of the lesion characteristics as the rf data (such as: choose rf data influential on lesion characteristics and make
To input parameter, using specified parameter as output parameter)；
Such as: by using the rf data in the related data of the volunteer made a definite diagnosis as input parameter, by its correlation
Lesion characteristics in data are as output parameter.
As described in the following steps: the data pair that the lesion characteristics and the rf data chosen are constituted, as
Sample data.
Such as: by obtained input, output parameter pair, a part is used as training book sample data, and a part is used as test specimens
Notebook data.
As a result, by the way that the rf data being collected into is analyzed and handled, and then sample data is obtained, operating process letter
It is single, operating result high reliablity.
As described in the following steps: analyzing the characteristic and its rule of the rf data, according to the characteristic and its rule, really
The network structure and its network parameter of the fixed artificial neural network；
Such as: on the influential data characteristic of health condition tool and its contained according to different ages, the state of an illness, gender etc.
Rule, can primarily determine the basic structure of network, the input of network, output node number, network hidden layer number, Hidden nodes, net
Network initial weight etc..
As shown in Fig. 2, one in the specific implementation, the framework of neural network is made of two stages.First stage, from
Displacement of tissue is speculated in two-dimentional rf data.The second stage of neural network goes prediction tissue strain by displacement field.This side
Method can directly obtain the displacement field and strain field of tissue.
In the first phase, soft tissue is collected in the forward and backward two-dimentional rf data collection that is under pressure, and is denoted as I respectively1And I2。
The phase time interval of 2D rf data collection and the displacement of tissue, which have, twice greatly contacts.To compressed rf data into
After the lesser global extension of row, we have obtained as input extracting and connect I first with a separable convolution1With
I2Composite character.
Then one five layers of convolutional neural networks are created to extract I1And I2Between difference language ambience information.In order to remove
Spatial correlation improves the ability of convolution nuclear expression and the related high-level information in spatial position, using part connection convolution come generation
For common full-mesh convolution.Facilitate the phase time interval of the different tissue of differentiation elasticity.
The component (from low to high) for being further processed all layers of output characteristic pattern is respectively to inactivate at random, Batch regularization,
Batch regularization+random inactivation, Batch regularization, Batch regularization+random inactivation.
It should be noted that as shown in figure 3, random inactivation is then led to by traversing each layer of neural network of node
It crosses and a Keep_prob (node reservation probability) is arranged to the neural network of this layer, is i.e. the node of this layer has that Keep_prob's is general
Rate is retained.The value range of Keep_prob is between 0 to 1.By the way that the reservation probability of the neural network node layer is arranged, make
Obtaining neural network will not go to be biased to some node, so that the weight of each node is not too big, mitigate mind with this
Over-fitting through network.
Specifically, the course of work inactivated at random is as follows:
Each node layer of neural network is traversed, setting node retains probability Keep_prob；Delete the section of neural network
Point, and delete network and remove the connection between node；Input sample is trained using simplified network.
Preferably, above procedure is repeated in each input sample.
It should be noted that the neural network formula that above-mentioned training uses includes:
Without the neural network formula inactivated at random:
In the presence of the neural network formula inactivated at random:
It should be noted that Batch regularization: activation input value of the deep-neural-network before doing nonlinear transformation with
Network depth deepens its distribution in perhaps training process and gradually shifts or change, and why trains convergence slow, usually
Overall distribution is gradually close toward the bound both ends of the value interval of nonlinear function, so leading to bottom nerve when backpropagation
The gradient of network disappears.For each hidden neuron, gradually to value interval limit saturation region after nonlinear function mapping
The input distribution drawn close is forced to be withdrawn into the normal distribution that mean value is the standard of comparison that 0 variance is 1, so that non-linear transform function
Input value, fall into input than more sensitive region, gradient disappearance problem is avoided with this.
Detailed process is as follows:
In above-mentioned formula step, γ and β are the adjustment parameter obtained by study, μBFor the average value of mini-batch,
σB 2For the variance of mini-batch.
The characteristic pattern that above-mentioned convolutional neural networks generate is adjusted to a vector, is used for subsequent three layers of full-mesh net
Network.The units of this network number of plies is respectively 64,32 and 1.This three layers network is available because organizing position caused by pressure
The distribution of shifting.
In second stage, the high-layer semantic information of displacement of tissue is extracted with another three layers convolutional network.By this
The characteristic pattern of three layers of generation can criticize standardization and random inactivation, criticize standardization, random inactivation is further located respectively by batch standardization
Reason.Similar with prediction displacement of tissue, the network of a full-mesh is pre- by the vector characteristics figure that be used to release from local displacement field
Survey the strain of tissue.
As an example, it is 2608 × 128 that the size of all pictures as input, which is adjusted by,.The displacement prediction stage
Patch-sized with strain forecast period is respectively 71 × 9 and 61 × 9.Made using the ADAM optimizer that a momentum is set as 0.9
For optimization algorithm.Learning rate is set as 10e-4, and loss equation is set as absolute average error.
Pseudocode is as follows:
In foregoing description, m0Represent initial first torque vector, v0Initial second torque vector is represented, t represents initial time
Step, θ0Initial parameter vector is represented, α represents learning rate.
Calculate the gradient of t time step:
Firstly, calculating the index moving average of gradient, m0It is initialized as 0.Similar to Momentum algorithm, comprehensively consider
The gradient momentum of time step before.β1Coefficient is exponential decay rate, controls weight distribution, usually takes the value close to 1.
mt=β1mt-1+(1-β1)gt
Then, the index moving average of gradient square, v are calculated0It is initialized as 0.β2For exponential decay rate, before control
Gradient squared impact situation.Similar to RMSprop algorithm, mean value is weighted to gradient square.
Third, due to m0It is initialized as 0, will lead to mtIt is partial to 0, especially training stage in the early stage.So equal to gradient
Value mtBias correction is carried out, influence of the deviation to training initial stage is reduced.
4th and m0It is similar, because of v0Being initialized as 0 causes to train initial stage vtIt is biased to 0, it is corrected.
Finally, undated parameter, initial learning rate α multiplied by gradient mean value and gradient variance the ratio between square root.By expressing
Formula can be seen that the step size computation to update, can be adaptively adjusted from two angles of gradient mean value and gradient square,
Rather than determined by current gradient.
Preferably, the network structure, comprising: BP neural network, CNN neural network, RNN neural network, and, residual error
At least one of neural network.
Preferably, the network parameter, comprising: intensive block number exports the number of plies, the convolution number of plies, the excessive number of plies, initial power
Value, and, at least one of bias.
As described in the following steps: using the sample data, be trained to the network structure and the network parameter
And test, determine the corresponding relationship of the rf data Yu the lesion characteristics.
Such as: after the completion of network design, it need to be trained with the neural network that training sample data complete design.Training
Method can be adjusted according to the problem of discovery in actual network structure and training.
As a result, by collection image data, sample data is therefrom chosen, and is trained and tests based on sample data,
It determines the corresponding relationship between rf data and lesion characteristics, is conducive to promote the accuracy for generating specified parameter.
It is alternatively possible to further illustrate that step " uses the sample data, to the network structure in conjunction with following description
It is trained and tests with the network parameter, determines the corresponding relationship of the rf data Yu the lesion characteristics " in
The detailed process that the network structure and the network parameter are trained.
As described in the following steps, a part of data in the sample data are chosen as training sample, by the training
The rf data in sample is input to the network structure, is joined by the activation primitive of the network structure and the network
Number is trained, and obtains hands-on result；Determine that the hands-on result and the corresponding lesion in the training sample are special
Whether the hands-on error between sign meets default training error；When the hands-on error meets the default training accidentally
When poor, determine that the training to the network structure and the network parameter is completed；
More optionally, the network structure and the network parameter are trained, further includes:
When the hands-on error is unsatisfactory for the setting training error, pass through the error energy of the network structure
Function updates the network parameter；It is carried out by the activation primitive and the updated network parameter of the network structure
Re -training, until the hands-on error after the re -training meets the setting training error；
Such as: if test error is met the requirements, network training test is completed.
As a result, by being used to test sample obtained network structure and network parameter be trained to test, with further
Verify the reliability of network structure and network parameter.
It is alternatively possible to further illustrate that step " uses the sample data, to the network structure in conjunction with following description
It is trained and tests with the network parameter, determines the corresponding relationship of the rf data Yu the lesion characteristics " in
The detailed process that the network structure and the network parameter are tested.
As described in the following steps, another part data in the sample data are chosen as test sample, by the survey
The rf data in sample sheet is input in the network structure that the training is completed, with the activation primitive and described
The network parameter that training is completed is tested, and actual test result is obtained；Determine the actual test result and the survey
Whether the actual test error between corresponding lesion characteristics in sample sheet meets setting test error；When the actual test is missed
When difference meets the setting test error, determine that the test to the network structure and the network parameter is completed.
At step S120, the current radio frequency data of the current ultrasonic elastic image of patient are obtained.
At step S130, by the corresponding relationship, determine that current lesion corresponding with the current radio frequency data is special
Sign.
Such as: the lesion characteristics in the ultrasonic elastic image of patient are identified in real time.
As a result, by being based on corresponding relationship, efficiently identified out according to current radio frequency data current in ultrasonic elastic image
Lesion characteristics, so that the diagnosis for doctor provides accurate judgment basis, and judging result accuracy is good.
In an optional example, corresponding with the rf data current lesion characteristics are determined in step S130, it can be with
It include: to be determined as lesion characteristics corresponding to rf data identical with the current radio frequency data in the corresponding relationship
The current lesion characteristics.
In an optional example, current lesion characteristics corresponding with the rf data are determined in step S130, may be used also
To include: to input the current radio frequency data in the functional relation when the corresponding relationship may include functional relation,
The output parameter for determining the functional relation is current lesion characteristics.
As a result, by being based on corresponding relationship or functional relation, current lesion characteristics are determined according to current radio frequency data, are determined
Mode is easy, definitive result high reliablity.
It can also include: that the verifying current lesion characteristics are characterized in actual lesion in an optional embodiment
The no process being consistent.
It is alternatively possible to receive the current lesion characteristics and verification result that actual lesion feature is not inconsistent, and/or really
In the fixed corresponding relationship rf data identical with the current radio frequency data when, more to corresponding relationship progress
Newly, it corrects, at least one of learn attended operation again.
Such as: equipment itself can not learn actual lesion feature, need the feedback operation of doctor, i.e., if set
Standby intelligent decision goes out lesion characteristics, and by operational feedback, it is not inconsistent with actual state doctor, and equipment can just be known.
Verify the current lesion characteristics whether be consistent with actual lesion feature (such as: can be by AR display module pair
Actual lesion feature is shown whether be consistent with actual lesion feature with the current lesion characteristics for verifying determining).
When the current lesion characteristics and actual lesion feature be not inconsistent, and/or the corresponding relationship in do not work as with described
When the identical rf data of preceding rf data, maintenance at least one of is updated, corrected, learning again to the corresponding relationship
Operation.
Such as: current lesion characteristics can be determined according to the current radio frequency data according to the corresponding relationship after maintenance.Example
Such as: by the corresponding lesion characteristics of rf data identical with the current radio frequency data in the corresponding relationship after maintenance, really
It is set to current lesion characteristics.
As a result, by the maintenance to the corresponding relationship between determining rf data and lesion characteristics, be conducive to promotion pair
The accuracy and reliability that lesion characteristics determine.
As shown in-a~5 Fig. 4, by the comparison with existing scheme, it is as follows to compare condition:
Table 1
Table 1 is the comparison condition of the present invention program and currently existing scheme in analogue data.
Table 2
Table 2 is that the comparison of the present invention program and currently existing scheme in model data and patient data compares condition.
Fig. 4-a and Fig. 4-b is the strain field estimation result of 8 analogue datas, and in same group of supposition, left side figure be by showing
The ultrasonic elastic image for having technical solution to generate, right image are the ultrasonic elastic image that technical solution generates through the invention；
Fig. 5-a and Fig. 5-b is the strain field estimation result of other 8 model datas, and in same group of supposition, left side figure be to lead to
The ultrasonic elastic image of prior art generation is crossed, right image is the Ultrasonic elasticity figure that technical solution generates through the invention
Picture；
Fig. 6 is the strain field estimation result of 4 patient datas, and in same group of supposition, left side figure is by prior art side
The ultrasonic elastic image that case generates, right image are the ultrasonic elastic image that technical solution generates through the invention；
By the above experimental result, in analogue data, the present invention in strain prediction, has more preferable than currently existing scheme
Performance, and clinically, strain is used as important parameter, so of the invention will be more preferably.In model data and patient data
Experiment in, the solution of the present invention has higher signal-to-noise ratio and comparison noise ratio, and prediction effect is more preferable.Therefore, using this reality
The technical solution for applying example, pair by the self-learning function using neural network, between the rf data and lesion characteristics of foundation
It should be related to；According to current radio frequency data, current lesion characteristics can be determined by the corresponding relationship, method of determination is reliable, determines
As a result accuracy is good.
For device embodiment, since it is basically similar to the method embodiment, related so being described relatively simple
Place illustrates referring to the part of embodiment of the method.
As shown in fig. 7, a kind of detection device of lesion characteristics of one embodiment of the application offer is shown, applied to passing through
The lesion characteristics detection for the ultrasonic elastic image that contrast-agent-free obtains, comprising:
Module 310 is established, for the self-learning capability using artificial neural network, establishes the radio frequency in ultrasonic elastic image
Corresponding relationship between data and lesion characteristics；
Obtain module 320, the current radio frequency data of the current ultrasonic elastic image for obtaining patient；
Determining module 330, for determining current lesion corresponding with the current radio frequency data by the corresponding relationship
Feature；Specifically, it is determined that current lesion characteristics corresponding with the rf data, comprising: by the corresponding relationship with it is described
Lesion characteristics corresponding to the identical rf data of current radio frequency data are determined as the current lesion characteristics.
In one embodiment, the rf data, comprising: be used for table by what setting rule was extracted in region of interest image sequence
Show the motor pattern of each pixel；Wherein,
The rf data, comprising: color characteristic and/or textural characteristics, and/or by special from the color by setting rule
The array more than one-dimensional or bidimensional of the feature composition extracted in sign, the textural characteristics；Wherein,
The color characteristic, comprising: the color characteristic for extracting the ultrasonic elastic image by wavelet transformation to
Amount；
And/or
The textural characteristics, comprising: the co-occurrence matrix characteristic value of the ultrasonic elastic image is calculated by wavelet transformation, and
The texture feature vector constructed by the co-occurrence matrix characteristic value；
And/or
The corresponding relationship, comprising: functional relation；The rf data is the input parameter of the functional relation, described
Lesion characteristics are the output parameter of the functional relation；
Determine current lesion characteristics corresponding with the current radio frequency data, further includes:
When the corresponding relationship includes functional relation, the current radio frequency data are inputted in the functional relation, really
The output parameter of the fixed functional relation is current lesion characteristics.
It is in one embodiment, described to establish module 310, comprising:
Acquisition submodule, for obtaining the corresponding relationship for establishing between the rf data and the lesion characteristics
Sample data；
Submodule is analyzed, for analyzing the characteristic and its rule of the rf data, according to the characteristic and its rule, really
The network structure and its network parameter of the fixed artificial neural network；
Training submodule is trained the network structure and the network parameter for using the sample data
And test, determine the corresponding relationship of the rf data Yu the lesion characteristics.
In one embodiment, the acquisition submodule, comprising:
Collect submodule, for collect the patient of different health status the rf data and the lesion characteristics；
Analyze submodule, for being analyzed the rf data and combined the expertise information prestored, choose and
The relevant data of the lesion characteristics are as the rf data；
Sample data generates submodule, the number for constituting the lesion characteristics and the rf data chosen
According to right, as sample data.
In one embodiment,
The network structure, comprising: BP neural network, CNN neural network, RNN neural network, and, residual error nerve net
At least one of network；
And/or
The network parameter, comprising: intensive block number, the output number of plies, the convolution number of plies, the excessive number of plies, initial weight, and,
At least one of bias.
In one embodiment,
The trained submodule, comprising:
Training result generates submodule, for choosing a part of data in the sample data as training sample, incites somebody to action
The rf data in the training sample is input to the network structure, activation primitive and institute by the network structure
It states network parameter to be trained, obtains hands-on result；
Training result error judgment submodule, for determining that the hands-on result is corresponding in the training sample
Whether the hands-on error between lesion characteristics meets default training error；
Decision sub-module is completed in training, for determining when the hands-on error meets the default training error
The training of the network structure and the network parameter is completed；
And/or
Submodule is tested, for testing the network structure and the network parameter, the test submodule, packet
It includes:
Test result generates submodule, for choosing another part data in the sample data as test sample,
The rf data in the test sample is input in the network structure that the training is completed, with the activation letter
The network parameter that the several and described training is completed is tested, and actual test result is obtained；
Test result error judgment submodule, for determining that the actual test result is corresponding in the test sample
Whether the actual test error between lesion characteristics meets setting test error；
Decision sub-module is completed in test, for determining when the actual test error meets the setting test error
The test of the network structure and the network parameter is completed.
In one embodiment,
The trained submodule, further includes:
Network parameter updates submodule, for leading to when the hands-on error is unsatisfactory for the setting training error
The error energy function for crossing the network structure updates the network parameter；
First retraining submodule, for the activation primitive and the updated network by the network structure
Parameter carries out re -training, until the hands-on error after the re -training meets the setting training error；
And/or
The test submodule, further includes:
Second retraining submodule, for when the actual test error is unsatisfactory for the setting test error, to institute
It states network structure and the network parameter carries out re -training, until the actual test error after the re -training is described at a slow speed
Set test error.
Referring to Fig. 8, a kind of computer equipment of the imaging method of ultrasonic elastic image of the invention is shown, it specifically can be with
Include the following:
Above-mentioned computer equipment 12 is showed in the form of universal computing device, the component of computer equipment 12 may include but
Be not limited to: one or more processor or processing unit 16, system storage 28, connecting different system components (including is
Unite memory 28 and processing unit 16) bus 18.
Bus 18 indicates one of a few 18 structures of class bus or a variety of, including memory bus 18 or memory control
Device, peripheral bus 18, graphics acceleration port, processor or the office using 18 structure of any bus in a variety of 18 structures of bus
Domain bus 18.For example, these architectures include but is not limited to industry standard architecture (ISA) bus 18, microchannel
Architecture (MAC) bus 18, enhanced isa bus 18, audio-video frequency electronic standard association (VESA) local bus 18 and outer
Enclose component interconnection (PCI) bus 18.
Computer equipment 12 typically comprises a variety of computer system readable media.These media can be it is any can be by
The usable medium that computer equipment 12 accesses, including volatile and non-volatile media, moveable and immovable medium.
System storage 28 may include the computer system readable media of form of volatile memory, such as arbitrary access
Memory (RAM) 30 and/or cache memory 32.Computer equipment 12 may further include other movement/it is not removable
Dynamic, volatile/non-volatile computer decorum storage medium.Only as an example, storage system 34 can be used for read and write can not
Mobile, non-volatile magnetic media (commonly referred to as " hard disk drive ").Although being not shown in Fig. 8, can provide for can
The disc driver of mobile non-volatile magnetic disk (such as " floppy disk ") read-write, and to removable anonvolatile optical disk (such as CD-
ROM, DVD-ROM or other optical mediums) read-write CD drive.In these cases, each driver can pass through one
A or multiple data mediums interface is connected with bus 18.Memory may include at least one program product, the program product
With one group of (for example, at least one) program module 42, these program modules 42 are configured to perform the function of various embodiments of the present invention
Energy.
Program/utility 40 with one group of (at least one) program module 42, can store in memory, for example,
Such program module 42 includes --- but being not limited to --- operating system, one or more application program, other program moulds
It may include the realization of network environment in block 42 and program data, each of these examples or certain combination.Program mould
Block 42 usually executes function and/or method in embodiment described in the invention.
Computer equipment 12 can also with one or more external equipments 14 (such as keyboard, sensing equipment, display 24,
Camera etc.) communication, the equipment interacted with the computer equipment 12 can be also enabled a user to one or more to be communicated, and/
Or with enable the computer equipment 12 and one or more other calculate any equipment that equipment are communicated (such as network interface card,
Modem etc.) communication.This communication can be carried out by interface input/output (I/O) 22.Also, computer equipment
12 can also by network adapter 20 and one or more network (such as local area network (LAN)), wide area network (WAN) and/or
Public network (such as internet) communication.As shown, network adapter 20 passes through other of bus 18 and computer equipment 12
Module communication.It should be understood that although being not shown in Fig. 8 other hardware and/or software mould can be used in conjunction with computer equipment 12
Block, including but not limited to: microcode, device driver, redundant processing unit 16, external disk drive array, RAID system, magnetic
Tape drive and data backup storage system 34 etc..
Processing unit 16 by the program that is stored in system storage 28 of operation, thereby executing various function application and
Data processing, such as realize the imaging method of ultrasonic elastic image provided by the embodiment of the present invention.
That is, above-mentioned processing unit 16 is realized when executing above procedure: using the self-learning capability of artificial neural network, building
The corresponding relationship between rf data and lesion characteristics in vertical ultrasonic elastic image；Obtain the current ultrasonic elastic image of patient
Current radio frequency data；By the corresponding relationship, current lesion characteristics corresponding with the current radio frequency data are determined；Specifically
Ground determines corresponding with the rf data current lesion characteristics, comprising: by the corresponding relationship with the current radio frequency number
According to lesion characteristics corresponding to identical rf data, it is determined as the current lesion characteristics.
In embodiments of the present invention, the present invention also provides a kind of computer readable storage medium, it is stored thereon with computer
Program realizes the imaging method of the ultrasonic elastic image provided such as all embodiments of the application when the program is executed by processor:
That is, realization when being executed by processor to program: using the self-learning capability of artificial neural network, establishing ultrasonic bullet
The corresponding relationship between rf data and lesion characteristics in property image；Obtain currently penetrating for the current ultrasonic elastic image of patient
Frequency evidence；By the corresponding relationship, current lesion characteristics corresponding with the current radio frequency data are determined；Specifically, it is determined that
Current lesion characteristics corresponding with the rf data, comprising: will be identical as the current radio frequency data in the corresponding relationship
Rf data corresponding to lesion characteristics, be determined as the current lesion characteristics.
It can be using any combination of one or more computer-readable media.Computer-readable medium can be calculating
Machine gram signal media or computer readable storage medium.Computer readable storage medium for example can be --- but it is unlimited
In system, device or the device of --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, or any above combination.Computer
The more specific example (non exhaustive list) of readable storage medium storing program for executing includes: electrical connection with one or more conducting wires, portable
Formula computer disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory
(EPOM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device or
Above-mentioned any appropriate combination.In this document, computer readable storage medium can be it is any include or storage program
Tangible medium, the program can be commanded execution system, device or device use or in connection.
Computer-readable signal media may include in a base band or as carrier wave a part propagate data-signal,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including --- but
It is not limited to --- electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be
Any computer-readable medium other than computer readable storage medium, which can send, propagate or
Transmission is for by the use of instruction execution system, device or device or program in connection.
The computer for executing operation of the present invention can be write with one or more programming languages or combinations thereof
Program code, above procedure design language include object oriented program language --- such as Java, Smalltalk, C+
+, further include conventional procedural programming language --- such as " C " language or similar programming language.Program code
It can fully execute on the user computer, partly execute, held as an independent software package on the user computer
Part executes on the remote computer or holds on a remote computer or server completely on the user computer for row, part
Row.In situations involving remote computers, remote computer can pass through the network of any kind --- including local area network
(LAN) or wide area network (WAN) --- it is connected to subscriber computer, or, it may be connected to outer computer (such as using because of spy
Service provider is netted to connect by internet).All the embodiments in this specification are described in a progressive manner, each
What embodiment stressed is the difference from other embodiments, the mutual coherent in same and similar part between each embodiment
See.
Although preferred embodiments of the embodiments of the present application have been described, once a person skilled in the art knows bases
This creative concept, then additional changes and modifications can be made to these embodiments.So the following claims are intended to be interpreted as
Including preferred embodiment and all change and modification within the scope of the embodiments of the present application.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that process, method, article or terminal device including a series of elements not only wrap
Those elements are included, but also including other elements that are not explicitly listed, or further includes for this process, method, article
Or the element that terminal device is intrinsic.In the absence of more restrictions, being wanted by what sentence "including a ..." limited
Element, it is not excluded that there is also other identical elements in process, method, article or the terminal device for including the element.
Above to the imaging method of ultrasonic elastic image provided herein, device, equipment and medium, carry out in detail
It introduces, specific examples are used herein to illustrate the principle and implementation manner of the present application, the explanation of above embodiments
It is merely used to help understand the present processes and its core concept；At the same time, for those skilled in the art, according to this
The thought of application, there will be changes in the specific implementation manner and application range, in conclusion the content of the present specification is not answered
It is interpreted as the limitation to the application.
Claims (10)
1. a kind of imaging method of ultrasonic elastic image characterized by comprising
Using the self-learning capability of artificial neural network, establish between rf data and lesion characteristics in ultrasonic elastic image
Corresponding relationship；
Obtain the current radio frequency data of the current ultrasonic elastic image of patient；
By the corresponding relationship, current lesion characteristics corresponding with the current radio frequency data are determined；Specifically, it is determined that with institute
State the corresponding current lesion characteristics of rf data, comprising: penetrate identical with the current radio frequency data in the corresponding relationship
Frequency is determined as the current lesion characteristics according to corresponding lesion characteristics.
2. the method according to claim 1, wherein
The rf data, comprising: one-dimensional rf data or two-dimentional rf data；
And/or
The textural characteristics, comprising: calculate the co-occurrence matrix characteristic value of the ultrasonic elastic image by wavelet transformation, and pass through
The texture feature vector of the co-occurrence matrix characteristic value construction；
And/or
The corresponding relationship, comprising: functional relation；The rf data is the input parameter of the functional relation, the lesion
Feature is the output parameter of the functional relation；
Determine current lesion characteristics corresponding with the current radio frequency data, further includes:
When the corresponding relationship includes functional relation, the current radio frequency data are inputted in the functional relation, determine institute
The output parameter for stating functional relation is current lesion characteristics.
3. the method according to claim 1, wherein the rf data established in ultrasonic elastic image and disease
The step of becoming the corresponding relationship between feature, comprising:
Obtain the sample data of the corresponding relationship for establishing between the rf data and the lesion characteristics；
The characteristic and its rule for analyzing the rf data determine the artificial neural network according to the characteristic and its rule
Network structure and its network parameter；
Using the sample data, the network structure and the network parameter are trained and are tested, determines the radio frequency
The corresponding relationship of data and the lesion characteristics.
4. according to the method described in claim 3, it is characterized in that, the acquisition is for establishing the rf data and the disease
The step of becoming the sample data of the corresponding relationship between feature, comprising:
Collect the patient of different health status the rf data and the lesion characteristics；
The expertise information prestored is analyzed the rf data and combined, is chosen relevant to the lesion characteristics
Data are as the rf data；
The data pair that the lesion characteristics and the rf data chosen are constituted, as sample data.
5. according to the method described in claim 4, it is characterized in that,
The network structure, comprising: BP neural network, CNN neural network, RNN neural network, and, in residual error neural network
At least one of；
And/or
The network parameter, comprising: intensive block number, the output number of plies, the convolution number of plies, the excessive number of plies, initial weight, and, biasing
At least one of value.
6. according to the described in any item methods of claim 3-5, which is characterized in that
The network structure and the network parameter are trained, comprising:
A part of data in the sample data are chosen as training sample, by the rf data in the training sample
It is input to the network structure, is trained by the activation primitive and the network parameter of the network structure, obtains reality
Training result；
Determine whether is hands-on error between the hands-on result and the corresponding lesion characteristics in the training sample
Meet default training error；
When the hands-on error meets the default training error, determine to the network structure and the network parameter
It is described training complete；
And/or
The network structure and the network parameter are tested, comprising:
Another part data in the sample data are chosen as test sample, by the radio frequency number in the test sample
According to being input in the network structure that the training is completed, joined with the network that the activation primitive and the training are completed
Number is tested, and actual test result is obtained；
Determine whether is actual test error between the actual test result and the corresponding lesion characteristics in the test sample
Meet setting test error；
When the actual test error meets the setting test error, determine to the network structure and the network parameter
The test complete.
7. according to the method described in claim 6, it is characterized in that,
The network structure and the network parameter are trained, further includes:
When the hands-on error is unsatisfactory for the setting training error, pass through the error energy function of the network structure
Update the network parameter；
Re -training is carried out by the activation primitive and the updated network parameter of the network structure, until described
Hands-on error after re -training meets the setting training error；
And/or
The network structure and the network parameter are tested, further includes:
When the actual test error is unsatisfactory for the setting test error, to the network structure and the network parameter into
Row re -training, until the actual test error setting test error at a slow speed after the re -training.
8. a kind of imaging device of ultrasonic elastic image characterized by comprising
Establish module, for the self-learning capability using artificial neural network, establish rf data in ultrasonic elastic image with
Corresponding relationship between lesion characteristics；
Obtain module, the current radio frequency data of the current ultrasonic elastic image for obtaining patient；
Determining module, for determining current lesion characteristics corresponding with the current radio frequency data by the corresponding relationship；Tool
Body determines current lesion characteristics corresponding with the rf data, comprising: by the corresponding relationship with the current radio frequency
Lesion characteristics corresponding to the identical rf data of data are determined as the current lesion characteristics.
9. a kind of equipment, which is characterized in that including processor, memory and be stored on the memory and can be at the place
The computer program run on reason device is realized when the computer program is executed by the processor as appointed in claim 1 to 7
Method described in one.
10. a kind of computer readable storage medium, which is characterized in that store computer journey on the computer readable storage medium
Sequence realizes the method as described in any one of claims 1 to 7 when the computer program is executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910797347.6A CN110517249A (en) | 2019-08-27 | 2019-08-27 | Imaging method, device, equipment and the medium of ultrasonic elastic image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910797347.6A CN110517249A (en) | 2019-08-27 | 2019-08-27 | Imaging method, device, equipment and the medium of ultrasonic elastic image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110517249A true CN110517249A (en) | 2019-11-29 |
Family
ID=68627282
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910797347.6A Pending CN110517249A (en) | 2019-08-27 | 2019-08-27 | Imaging method, device, equipment and the medium of ultrasonic elastic image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110517249A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127400A (en) * | 2019-11-29 | 2020-05-08 | 深圳蓝韵医学影像有限公司 | Method and device for detecting breast lesions |
| CN111275686A (en) * | 2020-01-20 | 2020-06-12 | 中山大学 | Method and device for generating medical image data for artificial neural network training |
| CN111598966A (en) * | 2020-05-18 | 2020-08-28 | 中山大学 | Magnetic resonance imaging method and device based on generation countermeasure network |
| CN111612865A (en) * | 2020-05-18 | 2020-09-01 | 中山大学 | MRI (magnetic resonance imaging) method and device for generating countermeasure network based on conditions |
| CN111932507A (en) * | 2020-07-31 | 2020-11-13 | 苏州慧维智能医疗科技有限公司 | Method for identifying lesion in real time based on digestive endoscopy |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106470613A (en) * | 2014-07-02 | 2017-03-01 | 皇家飞利浦有限公司 | It is used for characterizing the pathological changes signature of pathology for special object |
| CN109616195A (en) * | 2018-11-28 | 2019-04-12 | 武汉大学人民医院（湖北省人民医院） | The real-time assistant diagnosis system of mediastinum endoscopic ultrasonography image and method based on deep learning |
| CN110021014A (en) * | 2019-03-29 | 2019-07-16 | 无锡祥生医疗科技股份有限公司 | Nerve fiber recognition methods, system and storage medium neural network based |
| CN110074813A (en) * | 2019-04-26 | 2019-08-02 | 深圳大学 | A kind of ultrasonic image reconstruction method and system |
-
2019
- 2019-08-27 CN CN201910797347.6A patent/CN110517249A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106470613A (en) * | 2014-07-02 | 2017-03-01 | 皇家飞利浦有限公司 | It is used for characterizing the pathological changes signature of pathology for special object |
| CN109616195A (en) * | 2018-11-28 | 2019-04-12 | 武汉大学人民医院（湖北省人民医院） | The real-time assistant diagnosis system of mediastinum endoscopic ultrasonography image and method based on deep learning |
| CN110021014A (en) * | 2019-03-29 | 2019-07-16 | 无锡祥生医疗科技股份有限公司 | Nerve fiber recognition methods, system and storage medium neural network based |
| CN110074813A (en) * | 2019-04-26 | 2019-08-02 | 深圳大学 | A kind of ultrasonic image reconstruction method and system |
Non-Patent Citations (3)
| Title |
|---|
| 刘志东: ""基于超声射频RF 信号的脂肪肝分级量化方法"", 《四川大学学报》 * |
| 彭国华: ""基于超声射频信号的脂肪肝特征信息分析与辅助诊断系统设计"", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 * |
| 杨鑫: ""基于B超脉冲回波RF信号鉴别甲状腺结节良恶性的研究"", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127400A (en) * | 2019-11-29 | 2020-05-08 | 深圳蓝韵医学影像有限公司 | Method and device for detecting breast lesions |
| CN111275686A (en) * | 2020-01-20 | 2020-06-12 | 中山大学 | Method and device for generating medical image data for artificial neural network training |
| CN111275686B (en) * | 2020-01-20 | 2023-05-26 | 中山大学 | Method and device for generating medical image data for artificial neural network training |
| CN111598966A (en) * | 2020-05-18 | 2020-08-28 | 中山大学 | Magnetic resonance imaging method and device based on generation countermeasure network |
| CN111612865A (en) * | 2020-05-18 | 2020-09-01 | 中山大学 | MRI (magnetic resonance imaging) method and device for generating countermeasure network based on conditions |
| CN111612865B (en) * | 2020-05-18 | 2023-04-18 | 中山大学 | MRI (magnetic resonance imaging) method and device for generating countermeasure network based on conditions |
| CN111598966B (en) * | 2020-05-18 | 2023-04-18 | 中山大学 | Magnetic resonance imaging method and device based on generation countermeasure network |
| CN111932507A (en) * | 2020-07-31 | 2020-11-13 | 苏州慧维智能医疗科技有限公司 | Method for identifying lesion in real time based on digestive endoscopy |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110517249A (en) | Imaging method, device, equipment and the medium of ultrasonic elastic image | |
| WO2021036695A1 (en) | Method and apparatus for determining image to be marked, and method and apparatus for training model | |
| WO2020215984A1 (en) | Medical image detection method based on deep learning, and related device | |
| WO2021159742A1 (en) | Image segmentation method and apparatus, and storage medium | |
| CN110074813B (en) | Ultrasonic image reconstruction method and system | |
| Keshavan et al. | Combining citizen science and deep learning to amplify expertise in neuroimaging | |
| Cenggoro et al. | A systematic literature review of machine learning application in COVID-19 medical image classification | |
| CN106777953A (en) | The analysis method and system of medical image data | |
| CN110414607A (en) | Classification method, device, equipment and the medium of capsule endoscope image | |
| KR102239163B1 (en) | Method and apparatus for predicting the presence or absence of diseases using artificial neural networks | |
| KR102460257B1 (en) | Method or apparatus for providing diagnostic results | |
| JP7434537B2 (en) | Bidirectional training of machine learning models for tissue segmentation | |
| CN110046707B (en) | Evaluation optimization method and system of neural network model | |
| CN110400298A (en) | Detection method, device, equipment and the medium of heart clinical indices | |
| CN111932529A (en) | Image segmentation method, device and system | |
| CN112529149B (en) | Data processing method and related device | |
| CN111523593B (en) | Method and device for analyzing medical images | |
| CN112307947A (en) | Method and apparatus for generating information | |
| CN114359629A (en) | Pneumonia X chest radiography classification and identification method based on deep migration learning | |
| CN115798711A (en) | Chronic nephropathy diagnosis and treatment decision support system based on counterfactual contrast learning | |
| Lin et al. | Deep learning for emergency ascites diagnosis using ultrasonography images | |
| JP7173482B2 (en) | Health care data analysis system, health care data analysis method and health care data analysis program | |
| US20210145389A1 (en) | Standardizing breast density assessments | |
| CN110472239A (en) | Training method, device and the electronic equipment of entity link model | |
| CN115662510A (en) | Method, device and equipment for determining causal parameters and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191129 |
|
| RJ01 | Rejection of invention patent application after publication |