CN107330949A - A kind of artifact correction method and system - Google Patents
A kind of artifact correction method and system Download PDFInfo
- Publication number
- CN107330949A CN107330949A CN201710508431.2A CN201710508431A CN107330949A CN 107330949 A CN107330949 A CN 107330949A CN 201710508431 A CN201710508431 A CN 201710508431A CN 107330949 A CN107330949 A CN 107330949A
- Authority
- CN
- China
- Prior art keywords
- data
- projection
- die body
- correction
- corrected
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of artifact correction method and system, methods described includes：Die body is scanned, the first data for projection of the die body is obtained by detector；Handled the first data for projection of the die body or obtained by simulation the second data for projection of the die body；Neural network model is built, the neural network model is trained using the first data for projection and the second data for projection of the die body, correction coefficient is obtained；And the 3rd data for projection of detection target to be corrected is corrected using the correction coefficient, and rebuild the medical image for detecting target using the data for projection after correction.Method of the invention based on neutral net deep learning, using the training of big data, improves the accuracy of correction, obtains preferable ring artifact calibration result.
Description
Technical field
The present invention relates to medical imaging process field, more particularly to a kind of artifact correction method and system.
Background technology
Since CT comes out from the seventies in last century, constantly improved, from the first generation to the 5th generation, when constantly shortening scanning
Between and improve picture quality.CT is widely used to the numerous areas of medical diagnosis.CT is by measuring X-ray in all directions
Accumulation attenuation coefficient (or projection) during through human body tomography, then X-ray attenuation system on whole section is calculated by computer
Several distributions, is finally shown with image format, to assist the clinical diagnosis for carrying out disease.Because it can be provided than general
The higher soft tissue resolving power of logical x-ray imaging, and the problem of three-dimensional structure is overlapping is solved, it is had on medical imaging
Epoch-making significance.And in CT evolution, artifact is always the key factor for restricting CT.So-called artifact, refers to actual thing
When body is scanned, non-existent composition in the material object occurred in reconstruction image, it is to cause CT reconstruction images to lose diagnostic significance
One of key factor.Ring artifact is to influence CT reconstruction images as the important artifact of a class of matter, is reflected on pattern, is with weight
A series of donuts that center is different from surrounding pixel for the center of circle and gray scale are built, one kind typical case belonged in CT reconstruction images is pseudo-
Shadow.Ring artifact reduces the quality of CT reconstruction images, the processing of influence pictures subsequent and quantitative analysis.The formation of ring artifact,
Detector spy member is primarily due to cause the inconsistency of ray field strength response.The formation of ring artifact, it is also possible to be because
The influence come for the installation site deviation of detector to scanning strip, and the influence that signal cross-talk comes to scanning strip.Clinically
Easily caused because artifact and pathological tissue are overlapping it is image blurring cause mistaken diagnosis or fail to pinpoint a disease in diagnosis, therefore correction or farthest reduce
This artifact is very necessary.Research ring artifact and its corresponding solution are the emphasis and focus of current CT imaging problems
One of.
At present, most of ring artifact bearing calibration is all based on some factor of artifact generation, in imaging image
It is corrected using correcting algorithm, the influence that the factor carrys out artifact to picture strip is reduced or eliminated.
But it is corrected using correcting algorithm that there are the following problems to the ring artifact of image in imaging image：
1. all factors of pair influence image ring artifact can not be corrected simultaneously or major part is corrected；
2. correction accuracy depends on the applicability of correcting algorithm, different correcting algorithms can cause calibration result and accurate
Degree is not high.
The content of the invention
The problem of existing is corrected to the ring artifact of image using correcting algorithm in imaging image for above-mentioned, this
The purpose of invention is the calibration result and accuracy for improving ring artifact correction, solves to scheme influence in prior art CT equipment
The problem of can not being simultaneously corrected or largely be corrected as all factors of ring artifact.
To achieve the above object of the invention, the technical scheme that the present invention is provided is as follows：
A kind of artifact correction method, it is characterised in that this method includes：Die body is scanned, the mould is obtained by detector
First data for projection of body；First data for projection of the die body is handled or the second of the die body is obtained by simulation
Data for projection；Neural network model is built, using the first data for projection and the second data for projection of the die body to the nerve
Network model is trained, and obtains correction coefficient；And thrown using the correction coefficient the 3rd of detection target to be corrected the
Shadow data are corrected, and rebuild the medical image for detecting target using the data for projection after correction.
In the present invention, first data for projection and the second data for projection using the die body is to the neutral net
Model is trained, and obtains correction coefficient, including：By the first data for projection of the die body under the conditions of multigroup different scanning and
Second data for projection is input in the neural network model, obtains the corresponding correction coefficient of multigroup different scanning condition.
In the present invention, first data for projection to the die body is handled, and obtain the die body second is thrown
Shadow data, including：First data for projection of the die body is rebuild, the first image of the die body is obtained；To the mould
First image of body is smoothed, and obtains the second image of the die body；And the second image of the die body is carried out
Orthographic projection, obtains the second data for projection of the die body
In the present invention, the condition of scanning includes：The size of the die body, scan protocols and/or the die body it is inclined
The heart is set.
In the present invention, the optimization aim of the deep learning neutral net is the correction coefficient of desired output close to
The correction coefficient known.
In the present invention, first data for projection and the second data for projection using the die body is to the neutral net
Model is trained, and obtains correction coefficient, in addition to：Initial calibration parameter is obtained, and it is initial using the initial calibration parameter
Change the neural network model.
In the present invention, first data for projection to the die body is handled, and obtain the die body second is thrown
Shadow data, including：First data for projection of the die body is smoothed, the second data for projection of the die body is obtained.
In the present invention, it is described that school is carried out to the 3rd data for projection of detection target to be corrected using the correction coefficient
Just, including：According to the correction coefficient, calibration model is built；And the 3rd data for projection is inputted into the calibration model,
Data for projection after being corrected.
In the present invention, single order of the calibration model comprising explorer response is led and/or second order is led.
To reach above-mentioned purpose, present invention also offers another artifact correction method, methods described includes：Obtain at least
One training sample, each training sample includes the first data for projection and the second data for projection of die body, the first projection number
Obtained according to by scanning the die body, first data for projection of second data for projection based on the die body is obtained；Build god
Through network model, the neural network model is trained using at least one described training sample, correction parameter is obtained；Profit
The data for projection of sweep object to be corrected is corrected with the correction parameter, the data for projection after being corrected；And
The medical image of the sweep object is rebuild using the data for projection after the correction.
To reach above-mentioned purpose, present invention also offers a kind of artifact correction system, described device includes：At least one bag
Equipment containing detector, the equipment is used to scan die body, and the first data for projection of the die body is obtained by detector；One
Processor, is connected, the processor can be used for by network with least one described equipment：Thrown according to the first of the die body
Shadow data obtain the second data for projection of the die body；Neural network model is built, the first data for projection of the die body is utilized
The neural network model is trained with the second data for projection, correction coefficient is obtained；And utilize the correction coefficient pair
3rd data for projection of detection target to be corrected is corrected, and rebuilds the detection target using the data for projection after correction
Medical image.
Compared with prior art, beneficial effects of the present invention performance is as follows：
1. considering the Multiple factors of influence image ring artifact, and these influence factors are corrected simultaneously.
2. based on the method for neutral net deep learning, using the training of big data, improve the accuracy of correction, obtain compared with
Good ring artifact calibration result.
Brief description of the drawings
Fig. 1 is a kind of schematic diagram of the artifact correction system provided according to the present invention；
Fig. 2 is a kind of schematic diagram of the processor provided according to the present invention；
Fig. 3 is a kind of exemplary process diagram of the model that gets parms provided according to the present invention；
Fig. 4 is a kind of schematic diagram being trained to neural network model provided according to the present invention；
Fig. 5 is a kind of schematic diagram of the neural network model provided according to the present invention；
Fig. 6 is the exemplary flow of data after the ring artifact correction according to a kind of acquisition detection target of the invention provided
Figure；
Fig. 7 is the water modular ring shape artifact correction effect that the method provided according to the present invention is obtained；And
Fig. 8 is the human body image artifact correction effect that the method provided according to the present invention is obtained.
Fig. 1 is marked：110 be processor, and 120 be network, and 130 be equipment；
Fig. 2 is marked：210 be data acquisition module, and 220 be data processing module, and 230 be training module, and 240 be output mould
Block.
Embodiment
It is understandable to enable the above objects, features and advantages of the present invention to become apparent, with reference to the accompanying drawings and examples
Embodiment to the present invention is described in detail.
Fig. 1 is a kind of schematic diagram of the artifact correction system 100 provided according to the present invention.Artifact correction system 100 can be with
Including a processor 110, a network 120 and an equipment 130.Processor 110 and equipment 130 can pass through network 120
Connection communicates.In certain embodiments, artifact correction system 100 can include an imaging system.Artifact correction system 100
Artifact correction can be carried out to the imaging data of the imaging system, the artifact includes but is not limited to ring artifact.Imaging system
System can be single mode imaging system, including：DSA systems, magnetic resonance imaging (MRI) system, computed tomography angiography
System (CTA), positron emission tomography (PET) system, single photon emission computerized tomography (SPECT) system, meter
Calculation machine fault imaging (CT) system, digital radiography (DR) system etc..In certain embodiments, imaging system can be multi-modal
Imaging system, including：Positron emission tomography-computer tomography (PET-CT) system, positron emission fault into
Picture-magnetic resonance imaging (PET-MRI) system, single photon emission computerized tomography-positron emission tomography (SPECT-
PET) system, digital subtraction angiography-magnetic resonance imaging (DSA-MRI) system etc..It should be noted that puppet described below
Shadow correction system 100 is for illustration purposes only, and is not limited the scope of the invention.
Processor 110 is one and handled receiving data, and the equipment for exporting result.Processor 110 can be with
The data of receiving device 130, and based on the data, obtain device configuration information and scanning information.For example, processor 110 can
With the information for the detector scanning for obtaining equipment 130, and it is further processed.Rebuild for scan data, school
Just etc..In certain embodiments, processor 110 can be a computer, a smart mobile phone, a notebook computer, one
Individual intelligent medical instrument etc. possesses the electronic equipment of cpu function.
Network 120 can be the connected mode of any two or more equipment of connection.For example, network 120 can be wired
Network or wireless network.In certain embodiments, network 120 can be the combination of single network or multiple network.
For example, the network 120 can include LAN, wide area network, common network, dedicated network, WLAN, virtual network,
It is one or more of in public phone network, Intranet, Zigbee network, near field communication network, fiber optic network, internet etc.
Combination.Each module or unit in artifact correction system 100 can realize the interaction of information by connecting network 120.For example,
In ring artifact correction system 100, processor 110 can be connected to multiple Medical Devices, and pair reception simultaneously by network 120
Scan data from multiple equipment.
Equipment 130 can be the equipment for carrying out data acquisition to detection target.The equipment can include scanning means, into
As device, hospital bed device etc..In certain embodiments, equipment 130 can be with Scanning Detction target, and scan data is passed through into network
120 are sent to processor 110.Equipment 130 can include CT devices, MRI device, SPECT devices, PET device, CTA devices, DR
One or more of device, PET-CT devices, PET-MRI devices, SPECT-PET devices etc..In certain embodiments, scan
Data can detect the Raw projection data of target, and be sent to processor 110, carry out next step data processing and behaviour
Make.
In certain embodiments, processor 110 directly can be communicated between equipment 130, without passing through net
Network 120.In certain embodiments, the artifact correction system 100 may further include a display device and/or monitoring is set
It is standby.
Fig. 2 is a kind of schematic diagram of the processor 110 provided according to the present invention.Processor 110 can include data acquisition
Module 210, data processing module 220, training module 230, output module 240.Connection in system between each module can be
It is wired, it is wireless, or both combination.Any one module can be it is local, it is long-range, or both combination.Mould
Corresponding relation between block can be man-to-man, or one-to-many.
Data acquisition module 210 can be with acquisition scans information.The scanning information can include the original throwing of detection target
Shadow data or raw image data.In certain embodiments, data acquisition module 210 can pass through one in equipment 130
Or multiple detectors, detection target is scanned, and obtain its scanning information.The information of scanning is in the detector scanning visual field
The data obtained after the scanning of all targets, including sick bed, patient and other equipment in scan vision etc..In some implementations
In example, data acquisition module 210 can gather the Raw projection data of water mould by detector.And by the original of the water mould collected
Beginning data for projection is sent to data processing module 220, is further processed.
Data processing module 220 can be handled the information collected.For example, data processing module 220 can be by
The multiple data collected are classified or merged.And for example, data processing module 220 can be divided data or signal
Analysis, screening, classification, filtering, denoising etc. are operated.In certain embodiments, data processing module 220 can be to the water mould that collects
Raw projection data handled, obtain data for projection of the water mould through processing.Obtained water mould Raw projection data and through place
The data for projection of reason can as neural network model input, and be sent to training module 230 carry out next step operation.One
In a little embodiments, data processing module 220 can include data message on user interface.In certain embodiments,
Data processing module 220 can include the processing unit of one or more interconnections.Wherein, one or more of processing are single
Member, can be communicated or be connected with part or all of module or equipment in artifact correction device.
Training module 230 can carry out learning training to neural network model, and training result is exported.In some implementations
In example, training module 230 carries out learning training using the data for projection of multigroup water mould to neural network model, by neutral net
The learning training of model, obtains training result.The training result can apply to scanning and its ring-type of new detection target
The correction of artifact.In certain embodiments, neural network model can include but are not limited to BP neural network, feedback neural net
Network, radial base neural net, self organizing neural network, convolutional neural networks, perceptron neural network, linear neural network etc..
Output module 240 can export the training result of training module 230 from processor, or by artifact correction system
The information of 100 generations is sent to user.In certain embodiments, the information that output module 240 is exported can include text, table
Lattice, image, sound, code etc..For example, output module 240 can export the training result of artifact correction system 100, rebuild knot
Really, instruction etc..In certain embodiments, output module 240 can also include one or more physical components or equipment, such as touch
Display screen, LED light, loudspeaker, microphone etc..
It will be understood by those skilled in the art that above-mentioned module map is only the exemplary illustration being corrected to projection value,
There can also be other changes.For example, in certain embodiments, a memory module can be added.In addition, between modules
Information interchange and transfer mode be not unique.
Fig. 3 is a kind of exemplary process diagram 300 of the acquisition correction coefficient provided according to the present invention.
In the step 310, processor can gather the first data for projection of water mould by the detector in equipment.The water
First data for projection of mould can be the Raw projection data of water mould.In certain embodiments, processor can choose it is a variety of not
Water mould with size is scanned, and the water mould that can also choose a kind of size sets different bias to be scanned.When water mould
Raw projection data has gathered rear, it is necessary to handle it.
In step 320, processor can handle the first data for projection progress of multigroup water mould of collection and obtain water mould
The second data for projection.Second data for projection of the water mould can be the preferred view data of water mould.In certain embodiments,
The preferred view data of water mould can be corrected acquisition to the first data for projection of water mould by conventional correction methods, also may be used
To be obtained by physical analogy.Because the structure and form of water mould are fixed, and it is known, by calculating or can simulates,
Artifact in the Raw projection data that correct scan is obtained, obtains its preferable water mould data for projection.
In certain embodiments, the second data for projection of water mould can be obtained by a variety of methods.For example：Obtain original projection
Data, are smoothed after reconstruction to image, and last orthographic projection obtains preferred view data；Because die body structure typically compares
Simply, processing can be modeled to die body, by introducing the analytic equation of x-ray transmission process, calculating obtains preferred view
Data；The structure of die body can also be quantized, by simulating the transmission path of x-ray photon, such as Monte Carlo simulation is obtained
To preferred view data；Even directly Raw projection data can be smoothed.It should be noted that based on water mould
The methods of the Raw projection data preferred view data that obtain water mould be not limited only to method in the example.
In a step 330, neural network model is entered using the first data for projection and the second data for projection of multigroup water mould
Row training study, by the training of big data, exports training result.In step 340, output parametric model.In some implementations
In example, the initial model of neural network model is the neural network model of parameter initialization, for example, the parameter can be random
Initialization.The initial model of neutral net is trained, is that training sample is input in neutral net initial model, nerve
Network initial model is exported after computing.Then the parameter current of model is reversely adjusted using majorized function.
In certain embodiments, the reverse adjustment process to model can be the process of an iteration.Use training sample every time
After training, the parameter in neural network model can all change, as input training sample next time be trained it is " initial
Change parameter ".For example, in certain embodiments, majorized function can be used reversely to adjust the parameter of model.Majorized function
Target be the parameter for adjusting model, make the real output value of model minimum with desired output difference.When the multiple training of use
After sample is trained, the parameter value in model can be optimal, i.e. real output value and desired output difference is minimum, instruction
Practice and complete.After the completion of neutral net initial model training, parameter model is obtained, parameter value therein is fixed.
The artifact correction process that the present invention is described, is directly carried out, it is assumed that original projection number on Raw projection data
It can be obtained according to for x, then the data for projection y after correcting by following formula：
Wherein ai, bi, ciFor correction coefficient, dx is the first derivative of explorer response, d2X is the second order of explorer response
Derivative, M, N, P is correction exponent number, is confirmed according to calibration result, is usually no more than three ranks.As shown in above formula, as long as being corrected
Coefficient, it is possible to the data for projection after being corrected according to the formula.By the training of big data, neural network model can lead to
The preferred view data and actual measurement data for projection for crossing water mould obtain correction coefficient, so as to obtain preferable ring artifact correction effect
Really.
In certain embodiments, formula 1. in explorer response first derivative dx can be understood as based on detector position
The Data correction put, the second dervative d of explorer response2X can be understood as the Data correction based on interference signal crosstalk.Formula
1. in, the data for projection y after obtained correction is to have considered detector property factor, detector position factor, train of signal
Disturb factor etc. and the influence caused is produced to ring artifact, and these factors are corrected, the number after the correction finally obtained
According to.Its multivariate calibration method can more preferably obtain ring artifact calibration result.
Fig. 4 is a kind of schematic diagram being trained to neural network model provided according to the present invention.In 401, scanning
The water mould of a certain or several sizes, obtains the data for projection of water mould in 405.The data for projection is rebuild in 402,406
Middle acquisition water mould original image.In 403, original image is smoothed, the reason without ring artifact is obtained in 407
Think water mould image.In 404, then to preferable water mould image orthographic projection, preferable water mould data for projection is obtained 408.409
In, the Raw projection data and preferred view data of water mould are input in parameter model neutral net jointly, pass through neutral net
The training of study, in 410, the related parameter (that is, correction coefficient) of output detector.In order to ensure the accurate of correction coefficient
Property, can choose a variety of various sizes of water moulds is carried out, and a kind of water mould of size can also be selected to set different eccentric acquisitions.
The acquisition of correction coefficient can be shown below：
[ai,bi,ci...] and=f (rawm,rawideal, kV ...) formula is 2.
Wherein ai, bi, ciFor correction coefficient, function f is trained neural network model, rawmScanning is obtained
Initial condition mould data for projection, rawidealIt is to calculate obtained preferable water mould data for projection, kV is the condition of scanning.Swept in multigroup difference
The water mould Raw projection data obtained under the conditions of retouching, and calculate the obtained multigroup preferable water mould data for projection of correspondence, input god
Through in network model, being learnt by the training of neural network model, output calibration coefficient.Obtained correction coefficient is applied to again
The scanning correction of new detection target, Scanning Detction is completed in same CT.Obtained using the new scanning for detecting target
Data for projection, calculates the data for projection obtained after ring artifact correction based on formula 1..
Fig. 5 is a kind of schematic diagram of model 500 of the neutral net provided according to the present invention.As illustrated, input is inputted
Content is Raw projection data rawm, preferred view data rawidealAnd condition of scanning kV etc., by multiple interconnections
The mutual weighting of neuron node, and by a function (present invention in the function be formula 2. in function) conversion, finally
In output end output calibration coefficient sets { coefi}.The training set of neural network model is the preferable school obtained under multiple agreements
Positive coefficient.Optimum correction coefficient can be obtained by conventional correction methods, can also be obtained by physical analogy.By big data
Training, neural network model can by preferred view data and actual measurement projection value obtain correction coefficient so that obtain preferably
Ring artifact calibration result.The optimization aim of the neural network model is min ‖ { coefi}-{coefi}ideal‖, wherein
{coefiIt is the correction coefficient set that neural network model is exported, { coefi}idealIt is known optimum correction coefficient.At some
In embodiment, the optimization aim of the neural network model can be most common least square method, ∑ | coefi-coefideal|
Or ∑ (coefi-coefideal)2；Optimization aim can also carry regular terms, such as while above formula is met, it is desirable to all
Coefficient all have to be between certain number range；Optimization aim can also carry weight, to ensure that specific coefficient is preferentially full
Foot, such as ∑ w* (coefi-coefideal)2.It should be noted that the optimization aim is not limited only to carry in the above embodiments
Arrive.The correction coefficient of the neural network model desired output is closer to preferable correction coefficient, then calibration result is better.
In certain embodiments, the acquisition of neural network model can be obtained by following manner.Training sample is obtained first
This collection, it will be understood by those skilled in the art that being trained a large amount of training samples of needs to neural network model.Training sample
Concentrate and include multiple training samples, in the present invention, training sample should include the Raw projection data of water mould and calculating is obtained
Preferable water mould preferred view data.Then concentrated from training sample and obtain a training sample, its desired output is
Know.The input value of the training sample is input to neutral net initial model, neutral net initial model is to the training sample
Input value carries out the processes such as calculation process, including feature extraction and feature recognition, finally obtains one group of output valve.In training process
In, the parameter of model is reversely adjusted using majorized function.For example, in certain embodiments, majorized function can be used
min‖{coefi}-{coefi}ideal‖ is reversely adjusted to the parameter of model, wherein { coefiBe neutral net reality it is defeated
Go out value, { coefi}idealIt is the desired output of training sample.The target of the majorized function is the parameter of reverse adjustment model, is made
The real output value of neural network model is minimum with desired output difference.Every time after training, the parameter in neural network model
It will change, be used as " initiation parameter " that input training sample is trained next time.Finally, the god after training of judgement
Whether preparatory condition is met through network model.Wherein preparatory condition can be determined by user.For example, in certain embodiments, in advance
If condition can be that housebroken training samples number reaches preset value；In further embodiments, preparatory condition can be
Neural network model after training is tested, test result is qualified.If it is judged that being "Yes", then parameter is directly obtained
Model, and parameter model is exported to user or is stored in processor.If it is judged that being "No", then repeat to obtain instruction
Practice sample to be trained, step is as above-mentioned afterwards.
Fig. 6 is the exemplary flow of data after the ring artifact correction according to a kind of acquisition detection target of the invention provided
Figure 60 0.In step 610, detector obtains the 3rd data for projection of detection target, and the 3rd data for projection can be detection
The Raw projection data of target, and it is sent to processor.In step 620, processor is obtained based on above-mentioned neural network model
Correction coefficient, by formula 1. to detect target the 3rd data for projection calculate.In act 630, output detection mesh
The 3rd data for projection after target correction.In certain embodiments, processor can be further to the school of obtained detection target
Data for projection after just is rebuild, and obtains detecting the image after the correction of target, and feeds back to user, facilitates doctor to patient
Diagnosed.
Fig. 7 is the water modular ring shape artifact correction effect that the method provided according to the present invention is obtained；Fig. 8 is carried according to the present invention
The human body image artifact correction effect that the method for confession is obtained.In figures 7 and 8, (a) image is original with ring artifact
Original image；(b) image is the image after above-mentioned model is corrected to original image., it is apparent that the image in (b)
Ring artifact be corrected.As can be seen from Figures 7 and 8, the ring artifact calibration result and the ring of human body image of water mould
Shape artifact correction effect is closely.Therefore deduce that, by the Raw projection data and preferred view data of die body, by
Neural metwork training goes out the related all correction coefficient of artifact, and the processing of human projection's data is carried out using these correction coefficient,
The ring artifact in human body image can effectively be corrected.
Being preferable to carry out for the present invention is the foregoing is only, is not intended to limit the invention, for the technology of this area
For personnel, the present invention can have various modifications and variations.Within the spirit and principles of the invention, that is made any repaiies
Change, equivalent substitution, improvement etc., should be included in the scope of the protection.
Claims (10)
1. a kind of artifact correction method, it is characterised in that this method includes：
Die body is scanned, the first data for projection of the die body is obtained by detector；
Handled the first data for projection of the die body or obtained by simulation the second data for projection of the die body；
Neural network model is built, using the first data for projection and the second data for projection of the die body to the neutral net mould
Type is trained, and obtains correction coefficient；And
The 3rd data for projection of detection target to be corrected is corrected using the correction coefficient, and utilizes the throwing after correction
The medical image of target is detected described in shadow data reconstruction.
2. the method as described in claim 1, it is characterised in that described to be thrown using the first data for projection of the die body and second
Shadow data are trained to the neural network model, obtain correction coefficient, including：
The first data for projection and the second data for projection of the die body under the conditions of multigroup different scanning are input to the nerve
In network model, the corresponding correction coefficient of multigroup different scanning condition is obtained.
3. method as claimed in claim 2, it is characterised in that first data for projection to the die body is handled,
The second data for projection of the die body is obtained, including：
First data for projection of the die body is rebuild, the first image of the die body is obtained；
First image of the die body is smoothed, the second image of the die body is obtained；And
Orthographic projection is carried out to the second image of the die body, the second data for projection of the die body is obtained.
4. method as claimed in claim 2, it is characterised in that the condition of scanning includes the size of the die body, scanning association
The eccentric setting of view and/or the die body.
5. the method as described in claim 1, it is characterised in that described to be thrown using the first data for projection of the die body and second
Shadow data are trained to the neural network model, obtain correction coefficient, in addition to：Initial calibration parameter is obtained, and is utilized
Neural network model described in the initial calibration parameter initialization.
6. the method as described in claim 1, it is characterised in that first data for projection to the die body is handled,
The second data for projection of the die body is obtained, including：
First data for projection of the die body is smoothed, the second data for projection of the die body is obtained.
7. the method as described in claim 1, it is characterised in that described to utilize the correction coefficient to detection target to be corrected
The 3rd data for projection be corrected, including：
According to the correction coefficient, calibration model is built；And
3rd data for projection is inputted into the calibration model, the data for projection after being corrected.
8. method as claimed in claim 7, it is characterised in that single order of the calibration model comprising explorer response is led or two
Rank is led.
9. a kind of artifact correction method, it is characterised in that methods described includes：
At least one training sample is obtained, each training sample includes the first data for projection and the second data for projection of die body, institute
State the first data for projection and obtained by scanning the die body, first data for projection of second data for projection based on the die body
Obtain；
Neural network model is built, the neural network model is trained using at least one described training sample, obtained
Correction parameter；
The data for projection of sweep object to be corrected is corrected using the correction parameter, the projection number after being corrected
According to；And
The medical image of the sweep object is rebuild using the data for projection after the correction.
10. a kind of artifact correction system, it is characterised in that described device includes：
At least one includes the equipment of detector, and the equipment is used to scan die body, and the of the die body is obtained by detector
One data for projection；
One processor, is connected, the processor can be used for by network with least one described equipment：
The second data for projection of the die body is obtained according to the first data for projection of the die body；
Neural network model is built, using the first data for projection and the second data for projection of the die body to the neutral net mould
Type is trained, and obtains correction coefficient；And
The 3rd data for projection of detection target to be corrected is corrected using the correction coefficient, and utilizes the throwing after correction
The medical image of target is detected described in shadow data reconstruction.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710508431.2A CN107330949B (en) | 2017-06-28 | 2017-06-28 | Artifact correction method and system |
| US15/954,953 US10977843B2 (en) | 2017-06-28 | 2018-04-17 | Systems and methods for determining parameters for medical image processing |
| US17/228,690 US11908046B2 (en) | 2017-06-28 | 2021-04-12 | Systems and methods for determining processing parameter for medical image processing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710508431.2A CN107330949B (en) | 2017-06-28 | 2017-06-28 | Artifact correction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107330949A true CN107330949A (en) | 2017-11-07 |
| CN107330949B CN107330949B (en) | 2020-11-03 |
Family
ID=60198180
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710508431.2A Active CN107330949B (en) | 2017-06-28 | 2017-06-28 | Artifact correction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107330949B (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053456A (en) * | 2017-11-13 | 2018-05-18 | 深圳先进技术研究院 | A kind of PET reconstruction images optimization method and system |
| CN108122265A (en) * | 2017-11-13 | 2018-06-05 | 深圳先进技术研究院 | A kind of CT reconstruction images optimization method and system |
| CN108670282A (en) * | 2018-03-28 | 2018-10-19 | 上海联影医疗科技有限公司 | A kind of osteosclerosis artifact correction method |
| CN109171792A (en) * | 2018-09-29 | 2019-01-11 | 江苏影医疗设备有限公司 | Imaging method and the CT imaging system for using the imaging method |
| CN109697741A (en) * | 2018-12-28 | 2019-04-30 | 上海联影智能医疗科技有限公司 | A kind of PET image reconstruction method, device, equipment and medium |
| CN109978769A (en) * | 2019-04-04 | 2019-07-05 | 深圳安科高技术股份有限公司 | A kind of CT scan image data interpolation method and its system |
| CN110378982A (en) * | 2019-07-23 | 2019-10-25 | 上海联影医疗科技有限公司 | Reconstruction image processing method, device, equipment and storage medium |
| CN110390701A (en) * | 2019-07-08 | 2019-10-29 | 东软医疗***股份有限公司 | A kind of bearing calibration of artifact, artifact correction coefficient emulation mode and device |
| CN110660111A (en) * | 2019-09-18 | 2020-01-07 | 东软医疗***股份有限公司 | PET scattering correction and image reconstruction method, device and equipment |
| CN110755100A (en) * | 2019-10-17 | 2020-02-07 | 东软医疗***股份有限公司 | Correction method, correction device, console equipment and PET system |
| CN110866880A (en) * | 2019-11-14 | 2020-03-06 | 上海联影智能医疗科技有限公司 | Image artifact detection method, device, equipment and storage medium |
| CN110916708A (en) * | 2019-12-26 | 2020-03-27 | 南京安科医疗科技有限公司 | CT scanning projection data artifact correction method and CT image reconstruction method |
| CN111062890A (en) * | 2019-12-17 | 2020-04-24 | 散裂中子源科学中心 | Projection data processing method and device, electronic equipment and storage medium |
| CN111325678A (en) * | 2018-12-17 | 2020-06-23 | 西门子医疗有限公司 | Artifact correction method for three-dimensional image data |
| CN111443318A (en) * | 2019-01-16 | 2020-07-24 | 上海联影医疗科技有限公司 | Magnetic resonance image processing method, magnetic resonance image processing device, storage medium and magnetic resonance imaging system |
| CN111462168A (en) * | 2020-04-22 | 2020-07-28 | 上海联影医疗科技有限公司 | Motion parameter estimation method and motion artifact correction method |
| CN111627083A (en) * | 2020-05-26 | 2020-09-04 | 上海联影医疗科技有限公司 | Bone sclerosis artifact correction method and device, computer equipment and readable storage medium |
| CN112014870A (en) * | 2019-05-31 | 2020-12-01 | 佳能医疗***株式会社 | Radiation detection device, energy correction method, and program |
| CN112734877A (en) * | 2021-01-13 | 2021-04-30 | 上海联影医疗科技股份有限公司 | Method and system for correcting artifacts |
| CN112991228A (en) * | 2021-04-16 | 2021-06-18 | 上海联影医疗科技股份有限公司 | Method and system for correcting crosstalk |
| CN114638907A (en) * | 2022-02-23 | 2022-06-17 | 赛诺威盛科技(北京)股份有限公司 | Deep learning-based bone sclerosis artifact correction method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6682491B2 (en) * | 2000-02-11 | 2004-01-27 | Kci Licensing, Inc. | Method for artifact reduction in intracranial pressure measurements |
| US20050089215A1 (en) * | 2003-10-25 | 2005-04-28 | Carl Staelin | Image artifact reduction using a neural network |
| CN106725569A (en) * | 2016-01-30 | 2017-05-31 | 上海联影医疗科技有限公司 | Osteosclerosis artifact correction method and device |
-
2017
- 2017-06-28 CN CN201710508431.2A patent/CN107330949B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6682491B2 (en) * | 2000-02-11 | 2004-01-27 | Kci Licensing, Inc. | Method for artifact reduction in intracranial pressure measurements |
| US20050089215A1 (en) * | 2003-10-25 | 2005-04-28 | Carl Staelin | Image artifact reduction using a neural network |
| CN106725569A (en) * | 2016-01-30 | 2017-05-31 | 上海联影医疗科技有限公司 | Osteosclerosis artifact correction method and device |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108122265A (en) * | 2017-11-13 | 2018-06-05 | 深圳先进技术研究院 | A kind of CT reconstruction images optimization method and system |
| CN108053456A (en) * | 2017-11-13 | 2018-05-18 | 深圳先进技术研究院 | A kind of PET reconstruction images optimization method and system |
| CN108670282B (en) * | 2018-03-28 | 2021-12-17 | 上海联影医疗科技股份有限公司 | Method for correcting bone hardening artifact |
| CN108670282A (en) * | 2018-03-28 | 2018-10-19 | 上海联影医疗科技有限公司 | A kind of osteosclerosis artifact correction method |
| CN109171792A (en) * | 2018-09-29 | 2019-01-11 | 江苏影医疗设备有限公司 | Imaging method and the CT imaging system for using the imaging method |
| CN111325678A (en) * | 2018-12-17 | 2020-06-23 | 西门子医疗有限公司 | Artifact correction method for three-dimensional image data |
| CN111325678B (en) * | 2018-12-17 | 2023-11-17 | 西门子医疗有限公司 | Artifact correction method for three-dimensional volume image data |
| US11887221B2 (en) | 2018-12-28 | 2024-01-30 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for image correction in positron emission tomography |
| CN109697741A (en) * | 2018-12-28 | 2019-04-30 | 上海联影智能医疗科技有限公司 | A kind of PET image reconstruction method, device, equipment and medium |
| CN111443318B (en) * | 2019-01-16 | 2022-08-02 | 上海联影智能医疗科技有限公司 | Magnetic resonance image processing method, magnetic resonance image processing device, storage medium and magnetic resonance imaging system |
| CN111443318A (en) * | 2019-01-16 | 2020-07-24 | 上海联影医疗科技有限公司 | Magnetic resonance image processing method, magnetic resonance image processing device, storage medium and magnetic resonance imaging system |
| CN109978769B (en) * | 2019-04-04 | 2023-06-20 | 深圳安科高技术股份有限公司 | CT scanning image data interpolation method and system thereof |
| CN109978769A (en) * | 2019-04-04 | 2019-07-05 | 深圳安科高技术股份有限公司 | A kind of CT scan image data interpolation method and its system |
| CN112014870A (en) * | 2019-05-31 | 2020-12-01 | 佳能医疗***株式会社 | Radiation detection device, energy correction method, and program |
| CN110390701B (en) * | 2019-07-08 | 2023-04-25 | 东软医疗***股份有限公司 | Artifact correction method, artifact correction coefficient simulation method and device |
| CN110390701A (en) * | 2019-07-08 | 2019-10-29 | 东软医疗***股份有限公司 | A kind of bearing calibration of artifact, artifact correction coefficient emulation mode and device |
| CN110378982A (en) * | 2019-07-23 | 2019-10-25 | 上海联影医疗科技有限公司 | Reconstruction image processing method, device, equipment and storage medium |
| CN110378982B (en) * | 2019-07-23 | 2023-09-12 | 上海联影医疗科技股份有限公司 | Reconstructed image processing method, device, equipment and storage medium |
| CN110660111A (en) * | 2019-09-18 | 2020-01-07 | 东软医疗***股份有限公司 | PET scattering correction and image reconstruction method, device and equipment |
| CN110660111B (en) * | 2019-09-18 | 2023-05-30 | 东软医疗***股份有限公司 | PET scattering correction and image reconstruction method, device and equipment |
| CN110755100A (en) * | 2019-10-17 | 2020-02-07 | 东软医疗***股份有限公司 | Correction method, correction device, console equipment and PET system |
| CN110755100B (en) * | 2019-10-17 | 2023-07-04 | 沈阳智核医疗科技有限公司 | Correction method, correction device, console device and PET system |
| CN110866880A (en) * | 2019-11-14 | 2020-03-06 | 上海联影智能医疗科技有限公司 | Image artifact detection method, device, equipment and storage medium |
| CN111062890A (en) * | 2019-12-17 | 2020-04-24 | 散裂中子源科学中心 | Projection data processing method and device, electronic equipment and storage medium |
| CN110916708A (en) * | 2019-12-26 | 2020-03-27 | 南京安科医疗科技有限公司 | CT scanning projection data artifact correction method and CT image reconstruction method |
| CN111462168A (en) * | 2020-04-22 | 2020-07-28 | 上海联影医疗科技有限公司 | Motion parameter estimation method and motion artifact correction method |
| CN111462168B (en) * | 2020-04-22 | 2023-09-19 | 上海联影医疗科技股份有限公司 | Motion parameter estimation method and motion artifact correction method |
| CN111627083A (en) * | 2020-05-26 | 2020-09-04 | 上海联影医疗科技有限公司 | Bone sclerosis artifact correction method and device, computer equipment and readable storage medium |
| CN111627083B (en) * | 2020-05-26 | 2023-11-21 | 上海联影医疗科技股份有限公司 | Bone hardening artifact correction method, device, computer equipment and readable storage medium |
| WO2022152218A1 (en) * | 2021-01-13 | 2022-07-21 | Shanghai United Imaging Healthcare Co., Ltd. | Methods and systems for correcting projection data |
| CN112734877A (en) * | 2021-01-13 | 2021-04-30 | 上海联影医疗科技股份有限公司 | Method and system for correcting artifacts |
| CN112991228B (en) * | 2021-04-16 | 2023-02-07 | 上海联影医疗科技股份有限公司 | Method and system for correcting crosstalk |
| CN112991228A (en) * | 2021-04-16 | 2021-06-18 | 上海联影医疗科技股份有限公司 | Method and system for correcting crosstalk |
| CN114638907A (en) * | 2022-02-23 | 2022-06-17 | 赛诺威盛科技(北京)股份有限公司 | Deep learning-based bone sclerosis artifact correction method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107330949B (en) | 2020-11-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107330949A (en) | A kind of artifact correction method and system | |
| Torosdagli et al. | Deep geodesic learning for segmentation and anatomical landmarking | |
| EP3226766B1 (en) | System and method for image calibration | |
| El-Baz et al. | Appearance analysis for diagnosing malignant lung nodules | |
| CN110390351A (en) | A kind of Epileptic focus three-dimensional automatic station-keeping system based on deep learning | |
| CN110047128A (en) | The method and system of X ray CT volume and segmentation mask is rebuild from several X-ray radiogram 3D | |
| CN106062755A (en) | Systems and methods for identifying medical image acquisition parameters | |
| CN109255354B (en) | Medical CT-oriented computer image processing method and device | |
| CN107133549A (en) | ECT motion gates signal acquiring method and ECT image rebuilding methods | |
| CN111080584A (en) | Quality control method for medical image, computer device and readable storage medium | |
| CN107837090A (en) | Scattered ray correction based on sinogram in computer tomography | |
| US10068318B2 (en) | Enhancing the detectability of objects in medical images | |
| CN106999138A (en) | Diagnosing image method and apparatus and its recording medium | |
| CN109215040B (en) | Breast tumor segmentation method based on multi-scale weighted learning | |
| CN103229210A (en) | Image registration apparatus | |
| US11605162B2 (en) | Systems and methods for determining a fluid and tissue volume estimations using electrical property tomography | |
| CN110021022A (en) | A kind of thyroid gland nuclear medical image diagnostic method based on deep learning | |
| WO2019157435A1 (en) | Ct big data from simulation, emulation and transfer learning | |
| WO2022212010A2 (en) | Artificial intelligence training with multiple pulsed x-ray source-in-motion tomosynthesis imaging system | |
| CN108553121A (en) | A kind of method and apparatus of PET delayed sweeps | |
| CN110575178B (en) | Diagnosis and monitoring integrated medical system for judging motion state and judging method thereof | |
| da Cruz et al. | Enabling autonomous medical image data annotation: A human-in-the-loop reinforcement learning approach | |
| CN104000618B (en) | The true number of photons gate control method of one ring carries out the bearing calibration of respiratory movement gate | |
| Bagher-Ebadian et al. | Neural network and fuzzy clustering approach for automatic diagnosis of coronary artery disease in nuclear medicine | |
| CN112001979A (en) | Motion artifact processing method, system, readable storage medium and device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant after: Shanghai Lianying Medical Technology Co., Ltd Address before: 201807 Shanghai City, north of the city of Jiading District Road No. 2258 Applicant before: SHANGHAI UNITED IMAGING HEALTHCARE Co.,Ltd. |