CN110095742A - A kind of echo planar imaging neural network based and device - Google Patents
A kind of echo planar imaging neural network based and device Download PDFInfo
- Publication number
- CN110095742A CN110095742A CN201910395169.4A CN201910395169A CN110095742A CN 110095742 A CN110095742 A CN 110095742A CN 201910395169 A CN201910395169 A CN 201910395169A CN 110095742 A CN110095742 A CN 110095742A
- Authority
- CN
- China
- Prior art keywords
- magnetic resonance
- network
- echo
- artifact
- sample
- 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
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R33/00—Arrangements or instruments for measuring magnetic variables
- G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
- G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
- G01R33/48—NMR imaging systems
- G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
- G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution
- G01R33/5608—Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R33/00—Arrangements or instruments for measuring magnetic variables
- G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
- G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
- G01R33/48—NMR imaging systems
- G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
- G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution
- G01R33/561—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution by reduction of the scanning time, i.e. fast acquiring systems, e.g. using echo-planar pulse sequences
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Radiology & Medical Imaging (AREA)
- Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Signal Processing (AREA)
- General Health & Medical Sciences (AREA)
- High Energy & Nuclear Physics (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The application provides a kind of echo planar imaging neural network based and device.According to an example, which comprises according to the magnetic resonance parameters of selection, be scanned using Echo-plane imaging sequence to subject, obtain the data of k-space；According to the data of the k-space, the initial magnetic resonance images of the subject are obtained；By initial magnetic resonance images input trained in advance first nerves network and nervus opticus network, obtain the final magnetic resonance image of the subject, wherein, the first nerves network is configured as eliminating the neural network of chemical shift artifact, and the nervus opticus network is configured as the neural network of remedial frames deformation.Echo planar imaging neural network based and device provided by the present application can provide the function of deformation correction while effectively eliminating chemical shift artifact, so that rebuilding the better quality of obtained magnetic resonance image.
Description
Technical field
This application involves mr imaging technique field more particularly to a kind of Echo-plane imagings neural network based
Method and device.
Background technique
Magnetic resonance imaging (Magnetic Resonance Imaging, MRI) as a kind of multi-parameter, more contrasts at
It is one of main imaging mode in modern medical service iconography as technology.MRI can reflect out the longitudinal of subject inner tissue and relax
The multifrequency natures such as Henan time T1, lateral relaxation time T2 and proton density, so that the detection and diagnosis for disease provide information.
The basic functional principle of MRI system be according to electromagnetic induction phenomenon, using radio-frequency sending coil excitation subject in Hydrogen Proton, and
It is spatially encoded with gradient magnetic, the electromagnetic signal with location information, final weight is then received using RF receiving coil
Build out the magnetic resonance image of subject.
Echo-plane imaging (Echo Planar Imaging, EPI) is a kind of very quick MRI technique, can be several
Entire magnetic resonance image is obtained in/mono- second.EPI is current widely used fast imaging sequences, can be adapted for body
Many positions, including brain, abdomen and heart.
However the shortcoming of the main field stability due to MRI system, caused by the positive and negative gradient magnetic of EPI sequence is switched fast
Vortex, the factors such as chemical shift have been easy chemical shift artifact, image deformation etc. using the magnetic resonance image that EPI sequence obtains
Problem.
Summary of the invention
In view of this, the application provides a kind of method and device of Echo-plane imaging neural network based.
In a first aspect, a kind of method of Echo-plane imaging neural network based provided by the present application is by following skill
What art scheme was realized:
According to the magnetic resonance parameters of selection, subject is scanned using Echo-plane imaging sequence, obtains k-space
Data；
According to the data of the k-space, the initial magnetic resonance images of the subject are obtained；
By initial magnetic resonance images input trained in advance first nerves network and nervus opticus network, obtain described
The final magnetic resonance image of subject, wherein the first nerves network is configured as eliminating the nerve net of chemical shift artifact
Network, the nervus opticus network are configured as the neural network of remedial frames deformation.
Second aspect, the application provide a kind of Echo-plane imaging device neural network based, comprising:
Scan module sweeps subject using Echo-plane imaging sequence for the magnetic resonance parameters according to selection
It retouches, obtains the data of k-space；
First reconstruction module obtains the initial magnetic resonance images of the subject for the data according to the k-space；
Second rebuilds module, by initial magnetic resonance images input trained in advance first nerves network and nervus opticus
Network obtains the final magnetic resonance image of the subject, wherein the first nerves network is configured as eliminating chemical shift
The neural network of artifact, the nervus opticus network are configured as the neural network of remedial frames deformation.
The third aspect, the application provide a kind of Echo-plane imaging device neural network based, including processor and machine
Device readable storage medium storing program for executing, the machine readable storage medium are stored with the executable finger of the machine that can be executed by the processor
It enables, the processor is promoted to realize echo planar imaging neural network based described in first aspect by the machine-executable instruction
Imaging method.
Echo planar imaging neural network based and device provided by the present application, can effectively eliminate chemical potential
While moving artifact, the function of deformation correction is provided, so that rebuilding the better quality of obtained magnetic resonance image.
Detailed description of the invention
Fig. 1 is a kind of composition schematic diagram of MRI system；
Fig. 2A is the time diagram of EPI sequence；
Fig. 2 B is the track schematic diagram of EPI Sequence Filling k-space；
Fig. 3 is the flow chart for the echo planar imaging that the application one embodiment provides；
Fig. 4 A is the flow chart for the training first nerves network that the application one embodiment provides；
Fig. 4 B is the flow chart for the training nervus opticus network that the application one embodiment provides；
The timing signal for the scanning sequence that Fig. 5 is used when being the training nervus opticus network of the application one embodiment offer
Figure；
Fig. 6 is the structural schematic diagram for the Echo-plane imaging device that the application one embodiment provides；
Fig. 7 is the hardware structure diagram for the Echo-plane imaging device that the application one embodiment provides.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the application.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the application.
It is only to be not intended to be limiting the application merely for for the purpose of describing particular embodiments in term used in this application.
It is also intended in the application and the "an" of singular used in the attached claims, " described " and "the" including majority
Form, unless the context clearly indicates other meaning.It is also understood that term "and/or" used herein refers to and wraps
It may be combined containing one or more associated any or all of project listed.
It will be appreciated that though various information, but this may be described using term first, second, third, etc. in the application
A little information should not necessarily be limited by these terms.These terms are only used to for same type of information being distinguished from each other out.For example, not departing from
In the case where the application range, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as
One information.Depending on context, word as used in this " if " can be construed to " ... when " or " when ...
When " or " in response to determination ".
In order to better understand the application, MRI system is introduced first.As shown in Figure 1, simply illustrating the group of MRI system
At mainly including examination couch 110, magnet 120, gradient coil 131-133, radio-frequency coil 140, master computer 150, gradient amplification
Device 160, rf control unit 170 and console 180.Magnet 120 is the device for generating main field.Gradient coil 131-133, gradient
The composition gradients systems such as amplifier 160 are compiled mainly for generation of gradient magnetic with the space orientation for being able to carry out magnetic resonance signal
Code.Wherein, gradient coil is made of three groups of independent coils, respectively X-axis gradient coil 131, Y-axis gradient coil 132 and Z axis
Gradient coil 133.In general, the space encoding of X-direction is frequency coding, and the space encoding of Y-direction is phase code, the side Z
To space encoding be select layer.Radio-frequency coil 140, rf control unit 170 etc. form radio frequency system, are mainly used for sending out to subject
Radiofrequency signal is penetrated, then receives radiofrequency signal from subject, therefore radio-frequency coil 140 has radio-frequency sending coil and RF receiving coil
Point, radio-frequency sending coil and RF receiving coil can be the same coil, can also be divided into different coils.In order to simple
For the sake of, a coil 140 has only been drawn in Fig. 1.Master computer 150 is responsible for the radio frequency number of the transmission of MR imaging sequences, acquisition
According to operation, magnetic resonance image reconstruction and display etc..
When MRI system is scanned subject, the magnetic resonance signal that RF receiving coil samples becomes through modulus
Alternatively it is filled into k-space afterwards.Wherein, k-space is matrix form.After k-space is filled up, Fu is carried out to the data of k-space
In leaf transformation, the magnetic resonance image of a width subject can be reconstructed.For traditional magnetic resonance sequences, such as spin echo
Sequence, after a radio-frequency drive, multiple data that RF receiving coil obtains can fill a line of k-space.Every row of k-space
A specific phase code is all corresponded to, each data in every row correspond to different frequency codings.Wherein, the pulse of radio-frequency drive
It is to be issued by radio-frequency sending coil.After undergoing n times radio-frequency drive, k-space can be filled up, wherein N is k-space
Line number.Therefore sweep time needed for traditional magnetic resonance sequences is longer.
For EPI, after single radio frequency excitation, multirow k-space data can be obtained, it might even be possible to obtain all k
Spatial data.Therefore, the time needed for can greatly shortening magnetic resonance imaging using EPI sequence.As shown in Figure 2 A, Tu2AZhong
Abscissa indicate the time.After the excitation pulse that radio-frequency sending coil issues that angle is α, X-axis gradient coil is from positive amplitude
Quick oscillation forms a series of gtadient echos to negative amplitude, referring to fig. 2 the waveform of the Gx in A.By Y-axis gradient coil to this
A series of each echo in echelon's echoes carries out different phase codes, referring to fig. 2 the waveform of the Gy in A.X-axis gradient coil
Each oscillation correspond to the data line in k-space, and carried out frequency coding for each data in the row data.This
Outside, the Gz in Fig. 2A indicates the setting of Z axis gradient coil.It should be noted that the waveform only schematical description in Fig. 2A,
The setting of each gradient coil can be more complicated in the EPI sequence that actual clinical uses.
Since multiple echoes of EPI are generated by the continuous forward and reverse switching of X-axis gradient coil, the signal generated exists
Filling in k-space is a kind of track of detour, as shown in Figure 2 B.
But biggish vortex can be generated by carrying out being switched fast between positive antigradient, simultaneously because the unevenness of main field
Even property can to generate phase difference between the parity rows of the k-space of EPI.So as to cause a series of problems, for example, artifact, image
Deformation etc..
The principle of magnetic resonance is it is found that the resonant frequency of fat proton is lower than the resonant frequency of water proton, and in magnetic resonance,
Generally using the resonant frequency of water proton as center frequency, the frequency of fat proton resonance can fall into frequency in water proton space encoding
The frequency range of coding.The signal of adipose tissue can generate dislocation in magnetic resonance image after rebuilding in this way.Chemical shift artifact
In the boundary for appearing in adipose tissue and its hetero-organization.For EPI, chemical shift artifact appears in phase-encoding direction
On.And main field is stronger, and chemical shift artifact is more obvious.For chemical shift artifact, Fat-suppression technique, example can be used
Such as, inhibit the signal of normal-fat tissue using the sequence of rouge module with pressure.However, using the sequence of additional rouge module with pressure
Column would generally reduce the signal-to-noise ratio of the magnetic resonance image of reconstruction.
For this purpose, can be improved this application provides a kind of echo planar imaging neural network based using EPI skill
The quality of the magnetic resonance image of art.Referring to Fig. 3, the described method comprises the following steps.
Step S301 is scanned subject using Echo-plane imaging sequence, obtains according to the magnetic resonance parameters of selection
To the data of k-space.
The setting of the magnetic resonance parameters of subject, can be according to physical condition, diagnostic requirements, lesion of subject etc. by grasping
Work person determines.Specific magnetic resonance imaging parameter value can be each magnetic resonance imaging parameter customary in the art, such as echo
Time, repetition time, slice thickness etc..It should be noted that difference of the EPI sequence according to excitation pulse, and can be divided into
Gtadient echo EPI sequence, spin echo EPI sequence, inversion recovery EPI sequence.Method provided by the present application is suitable for these types
Imaging sequence.In addition, echo planar imaging provided by the present application, the sequence used does not include pressure rouge module.
Step S302 obtains the initial magnetic resonance images of the subject according to the data of the k-space.
Fourier transformation is carried out to the data of the k-space, obtains initial magnetic resonance images.It should be noted that this is first
The problems such as beginning magnetic resonance image is converted there may be chemical shift artifact and image.
Step S303, by initial magnetic resonance images input trained in advance first nerves network and nervus opticus net
Network obtains the final magnetic resonance image of the subject.
The extensive in parallel network that neural network is made of adaptable simple unit, it can simulate life
Object nervous system cross reaction to made by real world objects.Neural network may include input layer, output layer and several
Hidden layer can use every layer of complicated function of several neural unit fittings.The application can use various common nerve nets
Network, such as full convolutional neural networks (Fully Convolutional Network), U-Net, generation confrontation network
(Generative Adversarial Net, GAN) etc., the application is not intended to limit the concrete form of neural network.Wherein, institute
It states first nerves network and identical neural network can be used in the nervus opticus network, different nerve nets also can be used
Network, the application to this with no restriction.
Neural network used in this application includes first nerves network and nervus opticus network.Wherein, first nerves net
Network is the neural network for eliminating chemical shift artifact, and nervus opticus network is the neural network of remedial frames deformation.It can will be first
Beginning magnetic resonance image inputs first nerves network, and first nerves network carries out chemical shift artifact to the initial magnetic resonance images
It eliminates, obtains the magnetic resonance image of chemical shift artifact, then the magnetic resonance image of chemical shift artifact will be gone to input second
Neural network, nervus opticus network correct the deformation of the image, to obtain final magnetic resonance image.It is of course also possible to first
Initial magnetic resonance images are inputted into nervus opticus network, the magnetic resonance image after deformation is corrected are obtained, after then correcting deformation
Magnetic resonance image input first nerves network, to obtain final magnetic resonance image.
It should be noted that for different EPI sequences, for example, gtadient echo EPI sequence, spin echo EPI sequence,
Inversion recovery EPI sequence, the initial magnetic resonance images that these sequences generate can be used same neural network trained in advance and come
Eliminate chemical shift artifact and anamorphose.
Echo planar imaging neural network based provided by the present application, first according to the magnetic resonance parameters of selection,
Subject is scanned using Echo-plane imaging sequence, obtains the data of k-space.Then according to the data of the k-space,
Obtain the initial magnetic resonance images of the subject.Finally that initial magnetic resonance images input is trained in advance first nerves
Network and nervus opticus network obtain the final magnetic resonance image of the subject.The application uses nerve net trained in advance
Network can provide the function of deformation correction while effectively eliminating chemical shift artifact, so that it is total to rebuild obtained magnetic
The better quality of vibration image.
It is stated before neural network rebuilds initial magnetic resonance images in use, needs to train the nerve net in advance
Network.As shown in Figure 4 A, the training process of the first nerves network includes the following steps.
S401 obtains the first sample set of the first nerves network to be trained.Wherein, the sample of the first sample set
This includes fatty input magnetic resonance image and the first standard picture with chemical shift artifact.
A certain subject can be scanned according to EPI sequence, obtained k-space data, then by the k-space data
Fourier transformation is carried out, width input magnetic resonance image is obtained.It should be noted that this input magnetic resonance image is fatty
Magnetic resonance image with chemical shift artifact.Then scanning is re-started to the subject using the sequence of rouge module with pressure, obtained
To the magnetic resonance image of no chemical shift artifact, as the first standard picture.The input magnetic resonance image and of one subject
One standard picture constitutes a sample and can be made of multiple such samples for first sample set.In order to make sample
The coverage area of collection is as wide as possible, and subject can have different lesions, different statures, different ages, different genders
Etc..
S402, according to the first sample set, the training first nerves network.
Following step can be used in the training first nerves network.
Firstly, the parameter of each layer of initialization first nerves network.The parameter of its each layer can be random initial value, example
Each layer parameter of Gaussian Profile random initializtion such as can be used.The initial method of each layer parameter of neural network is according to being used
Neural network type different from, can be initialized according to method well known to those skilled in the art, the application couple
This is with no restriction.
Then, the input magnetic resonance image in the sample of first sample set is inputted into first nerves network, obtains the sample
First optimization image.The optimization direction of the optimization image be do not include chemical shift artifact.In the training first nerves net
When network, single sample can be inputted into first nerves network, first nerves network is trained, it can also be same by multiple samples
When input first nerves network, first nerves network is trained.All samples can also be inputted first nerves net simultaneously
Network is trained first nerves network.The application is not construed as limiting this.
Then, according to the first standard picture of sample and the first optimization image, the loss function of first nerves network is calculated,
And optimize the parameter value of each layer of first nerves network.
Following numerical value can be used, the first optimization image is compared with corresponding standard picture.Such as calculating first is excellent
Change the mean square error (Mean Square Error, MSE) between image and corresponding first standard picture, is also possible to calculate
Structural similarity (Structural Similarity, SSIM) index between two images, can also be between two images
Mean absolute error (Mean Absolute Error, MAE).MSE can be used in loss function, and the value of SSIM, MAE etc. are made
For the benchmark of calculating, loss function, the application can also be calculated using other numerical value well known to those skilled in the art certainly
With no restriction to this.
Gradient optimization algorithm can be used to optimize each layer parameter value of neural network.Gradient decline is common
Optimization neural network method.In the algorithm of gradient decline, from certain initial solutions, iteration finds optimal parameter
Value.In each iteration, error amount is first calculated in the gradient of current point, searches for optimal solution then along negative gradient direction.Learning rate
It determines the step-length for being optimal solution, the size of learning rate can be set based on experience value.
According to the difference of the number of samples of same primary input first nerves network, different gradient decline can be used and calculate
Method.Such as when merely entering a sample every time, can be used stochastic gradient descent (Stochastic Gradient Descent,
SGD) algorithm.When inputting multiple samples every time, small lot gradient descent method (Mini-Batch Gradient can be used
Descent, MBGD) algorithm.When inputting all samples simultaneously, batch gradient descent method (Batch Gradient can be used
Descent, BGD) algorithm.
It is, of course, also possible to the parameter of neural network is optimized using other algorithms well known to those skilled in the art,
The application is not construed as limiting this.The specific implementation of the algorithm of the optimization of neural network parameter may refer to the description of related algorithm,
Details are not described herein.
Until first sample concentrates all samples to be all used to optimize the parameter of first nerves network.When all samples
Originally it is involved in first nerves network after training, completes a trained bout (epoch).For the instruction of first nerves network
Practice, can only carry out the training of a trained bout, the training of multiple trained bouts can also be carried out.This is by first sample set
The factors such as size, the convergence rate of loss function determine.
After the loss function for judging first nerves network has been restrained, training to first nerves network is completed, obtains the
The Optimal Parameters of each layer of one neural network.Method well known to those skilled in the art can be used and judge whether loss function is received
It holds back, details are not described herein again.
As shown in Figure 4 B, the training process of the nervus opticus network includes the following steps.
S411 obtains second, sample collection of the nervus opticus network to be trained.Wherein, the sample of second sample set
This includes at least the artifact figure of a width deformation.
In embodiment 1, the sample in the second sample set may include two width along out of phase coding direction deformation
Artifact figure.
To obtain two width along the artifact figure of out of phase coding direction deformation, firstly, it is necessary to which corresponding scanning sequence is arranged.
Referring to Fig. 5, scanning sequence includes positive sequence and reverse sequence.Wherein, the phase code setting of positive sequence is positive, and reversed
Phase code setting is negative in sequence.Positive sequence is consistent with the other parameters of reverse sequence.Using positive sequence to subject
It is scanned to obtain the k-space data of a width forward phase coding direction deformation, which is rebuild to obtain forward direction
The artifact figure of phase-encoding direction deformation.Same subject is scanned using reverse sequence to obtain reverse phase coding direction
The k-space data of deformation rebuilds the k-space data to obtain the artifact figure of reverse phase coding direction deformation.This two width
Artifact figure constitutes the sample of the second sample set.It does not include standard picture for embodiment one, a sample only includes this
Two width artifact figures.
In the second embodiment, the sample in the second sample set may include two width along out of phase coding direction deformation
Artifact figure and the second standard picture.
In this embodiment, the sample of the second sample set can also include standard picture.It is can be used first as implemented
Method described in mode one obtains two width along the artifact figure of out of phase coding direction deformation.Then by traditional algorithm to this
Two width artifact figures are finely registrated, and obtain corresponding second standard picture, or can be by other imaging sequences to same
One subject acquires undeformed image as the second standard picture, using the second standard picture and this two width artifact figure as one
A sample.
In the third embodiment, the sample in the second sample set may include the artifact figure and third standard drawing of a width deformation
Picture.
For the sample of embodiment three, the deformation direction of artifact figure is mainly the shape along phase-encoding direction
Become.For the artifact figure of each sample, there is a corresponding third standard drawing, which can be to same
Subject is collected by other imaging sequences.
S412, according to second sample set, the training nervus opticus network.
Following step can be used in the training nervus opticus network.
Firstly, the parameter of each layer of initialization nervus opticus network.Its initial method is similar with first nerves network,
This is repeated no more.
Then, the artifacts in the sample of the second sample set are inputted into nervus opticus network, obtains the second of the sample
Optimize image.The optimization direction of the optimization image be do not include deformation.Likewise, in the training nervus opticus network, it can
A sample is inputted nervus opticus network, nervus opticus network is trained, multiple samples can also be inputted simultaneously
Nervus opticus network is trained nervus opticus network.All samples can also be inputted into nervus opticus network simultaneously, to the
Two neural networks are trained.The application is not construed as limiting this.
For above-mentioned embodiment one, two width artifact figures in sample are inputted into nervus opticus network, it is available
Two second optimization images.That is, the artifact figure of the forward phase coding direction deformation for sample, an available width
Second optimization image, for the artifact figure of the reverse phase coding direction deformation of sample, available another second optimization figure
Picture.Since the artifact figure of the forward and reverse phase-encoding direction deformation of sample is that same subject is scanned, only
The phase-encoding direction that is arranged when scanning is different, therefore, this two second optimization images obtained by nervus opticus network
Similarity is higher.
For above-mentioned embodiment two, as embodiment one, by the two width artifact figures input second in sample
Neural network, available two second optimization image.
For above-mentioned embodiment three, the width artifact figure in sample is inputted into nervus opticus network, it is available
One width third optimizes image.
Then, the loss function of nervus opticus network is calculated, and optimizes the parameter value of each layer of nervus opticus network.
When calculating loss function, whether the calculation method of loss function includes that standard picture is related to sample.Specifically such as
It is lower described.
For above embodiment two, each sample includes the second standard picture, similar to first nerves network, can be with
Calculate separately the value of MSE, SSIM, MAE between every second optimization image and corresponding second standard picture etc..To use
For MSE is as loss function is calculated, for a sample, available two second optimization image optimizes every second
Image seeks the value of itself and the MSE of the second standard picture of the sample respectively.Then the value of obtain two MSE is substituted into meter together
Calculate the formula of loss function.
For above embodiment three, each sample includes third standard picture, Ke Yiji similar to embodiment two
Calculate the value of MSE, SSIM, MAE between third optimization image and corresponding third standard picture etc..To use MSE as calculating
For loss function, for a sample, available width third optimizes image, asks itself and corresponding third standard picture
The value of MSE.Then the value of obtained MSE is substituted into the formula for calculating loss function.
For above-mentioned embodiment one, due to not having standard picture, every second optimization image is calculated first
Difference between corresponding input artifacts, thus available two width of method optimizes the corresponding difference of image,
By the formula for calculating loss function with substitution together of two differences.Still by taking MSE value as an example, the calculating of the sum of difference can pass through
Following formula (1) obtains:
MSE_total=| MSE_image1 |+| MSE_image2 | (1),
Wherein, MSE_image1 is the first width artifact figure between the second optimization figure for obtaining according to the first width artifact figure
MSE value, MSE_image2 are the MSE value between the second optimization figure that the second width artifact figure and the second width artifact figure obtain.It will
The calculation formula of MSE_total substitution loss function.Other than above-mentioned numerical value can be used, it can also calculate and calculate separately
Differential chart between second optimization image and the artifacts of input, thus the available two width differential chart of method, then calculates
Then the mutual information is substituted into the calculation formula of loss function by the mutual information (mutual information) of two width differential charts.
According to the calculation method of above-mentioned loss function, the gradient decline optimization as first nerves network can be used
Algorithm optimizes each layer parameter value of nervus opticus network, details are not described herein according to loss function.
Until sample all in the second sample set is all used to optimize the parameter of nervus opticus network.When all samples
Originally it is involved in nervus opticus network after training, completes a trained bout.Training for nervus opticus network, can be with
The training for only carrying out a trained bout, can also carry out the training of multiple trained bouts.This be by the second sample set size,
What the factors such as the convergence of loss function determined.
After the loss function for judging nervus opticus network has been restrained, training to nervus opticus network is completed, obtains the
The Optimal Parameters of each layer of two neural networks.Method well known to those skilled in the art can be used and judge whether loss function is received
It holds back, details are not described herein again.
It should be noted that, although in training nervus opticus network, the sample in used sample set may include
Artifact figure of two width along different phase-encoding direction deformation.But after its parameter optimization, in actual use, the figure of input
It seem single image.The coding of single image includes the deformation of the deformation of positive coding direction and/or the coding direction of negative sense,
By nervus opticus network, the deformation in the two directions can be corrected.
By the step of above-mentioned trained first nerves network and nervus opticus network it is found that for the two neural networks,
When training, it is trained respectively.And be when in use by a width initial magnetic resonance images input include this two
The neural network of a neural network, the initial magnetic resonance images successively pass through the two neural networks, carry out deformation correction and change
The removal of chemical shift artefacts.
Corresponding with the embodiment of aforementioned echo planar imaging neural network based, present invention also provides be based on
The embodiment of the Echo-plane imaging device of neural network.It is described in detail with reference to the accompanying drawing.
It is a kind of structure of Echo-plane imaging device neural network based provided by the embodiments of the present application referring to Fig. 6
Schematic diagram, wherein the imaging device can be applied in MRI system.As shown in fig. 6, the imaging device may include: scanning mould
Block 610, first rebuilds module 620 and second and rebuilds module 630.
Scan module 610 carries out subject using Echo-plane imaging sequence for the magnetic resonance parameters according to selection
Scanning, obtains the data of k-space.
First reconstruction module 620 obtains the initial magnetic resonance figure of the subject for the data according to the k-space
Picture.
Second rebuilds module 630, by initial magnetic resonance images input first nerves network and second trained in advance
Neural network obtains the final magnetic resonance image of the subject.Wherein, first nerves network is to eliminate chemical shift artifact
Neural network, the nervus opticus network are the neural network of remedial frames deformation.
In a kind of optional embodiment, the Echo-plane imaging sequence includes following any: gtadient echo plane
Echo imaging sequence；Spin echo Echo-plane imaging sequence；Inversion recovery Echo-plane imaging sequence.
In a kind of optional embodiment, described device further includes the first training module, is used for: obtaining first sample
Collection, wherein the sample of the first sample set includes fatty with the input magnetic resonance image of chemical shift artifact and the first mark
Quasi- image；According to the first sample set, the training first nerves network.
In a kind of optional embodiment, described device further includes the second training module, is used for: obtaining the second sample
Collection, wherein the sample of second sample set includes at least the artifact figure of a width deformation；According to second sample set, training
The nervus opticus network.
Further, the sample of second sample set includes following any: two width are encoded along out of phase
The artifact figure of direction deformation；Artifact figure and second standard picture of two width along out of phase coding direction deformation；One width deformation
Artifact figure and third standard picture.
The function of each unit and the realization process of effect are specifically detailed in the above method and correspond to step in above-mentioned apparatus
Realization process, details are not described herein.For device embodiment, since it corresponds essentially to embodiment of the method, so related
Place illustrates referring to the part of embodiment of the method.The apparatus embodiments described above are merely exemplary, wherein institute
Stating unit as illustrated by the separation member may or may not be physically separated, and component shown as a unit can
To be or may not be physical unit, it can it is in one place, or may be distributed over multiple network units.
Some or all of the modules therein can be selected to realize the purpose of application scheme according to the actual needs.This field is common
Technical staff can understand and implement without creative efforts.
Fig. 7 is referred to, is a kind of the hard of Echo-plane imaging device neural network based provided by the embodiments of the present application
Part structural schematic diagram.The imaging device may include processor 701, machine readable storage Jie for being stored with machine-executable instruction
Matter 702.Processor 701 can be communicated with machine readable storage medium 702 via system bus 703.Also, by reading and executing
Machine-executable instruction corresponding with the imaging logic of Echo-plane imaging in machine readable storage medium 702, processor 701 can
Execute above-described echo planar imaging neural network based.
Machine readable storage medium 702 referred to herein can be any electronics, magnetism, optics or other physical stores
Device may include or store information, such as executable instruction, data, etc..For example, machine readable storage medium may is that it is non-
Volatile memory, flash memory, memory driver (such as hard disk drive), solid state hard disk, any kind of storage dish (such as CD,
DVD etc.) perhaps similar storage medium or their combination.
Theme described in this specification and the embodiment of feature operation can be realized in the following: Fundamental Digital Circuit,
Computer software or firmware, the computer including structure disclosed in this specification and its structural equivalents of tangible embodiment are hard
The combination of part or one or more of which.The embodiment of theme described in this specification can be implemented as one or
Multiple computer programs, i.e. coding are executed by data processing equipment on tangible non-transitory program carrier or are controlled at data
Manage one or more modules in the computer program instructions of the operation of device.Alternatively, or in addition, program instruction can be with
It is coded on manually generated transmitting signal, such as electricity, light or electromagnetic signal that machine generates, the signal are generated will believe
Breath encodes and is transferred to suitable receiver apparatus to be executed by data processing equipment.Computer storage medium can be machine can
Read storage equipment, machine readable storage substrate, random or serial access memory equipment or one or more of which group
It closes.
Processing described in this specification and logic flow can by execute one of one or more computer programs or
Multiple programmable calculators execute, to execute corresponding function by the way that output is operated and generated according to input data.Institute
It states processing and logic flow can also be by dedicated logic circuit-such as FPGA (field programmable gate array) or ASIC (dedicated collection
At circuit) Lai Zhihang, and device also can be implemented as dedicated logic circuit.
The computer for being suitable for carrying out computer program includes, for example, general and/or special microprocessor or it is any its
The central processing unit of his type.In general, central processing unit will refer to from read-only memory and/or random access memory reception
Order and data.The basic module of computer includes central processing unit for being practiced or carried out instruction and for storing instruction
With one or more memory devices of data.In general, computer will also be including one or more great Rong for storing data
Amount storage equipment, such as disk, magneto-optic disk or CD etc. or computer will be coupled operationally with this mass-memory unit
To receive from it data or have both at the same time to its transmission data or two kinds of situations.However, computer is not required to have in this way
Equipment.In addition, computer can be embedded in another equipment, such as mobile phone, personal digital assistant (PDA), mobile sound
Frequency or video player, game console, global positioning system (GPS) receiver or such as universal serial bus (USB) flash memory
The portable memory apparatus of driver, names just a few.
It is suitable for storing computer program instructions and the computer-readable medium of data including the non-volatile of form of ownership
Memory, medium and memory devices, for example including semiconductor memory devices (such as EPROM, EEPROM and flash memory device),
Disk (such as internal hard drive or removable disk), magneto-optic disk and CD ROM and DVD-ROM disk.Processor and memory can be by special
It is supplemented or is incorporated in dedicated logic circuit with logic circuit.
Although this specification includes many specific implementation details, these are not necessarily to be construed as the model for limiting any invention
It encloses or range claimed, and is primarily used for describing the feature of the specific embodiment of specific invention.In this specification
Certain features described in multiple embodiments can also be combined implementation in a single embodiment.On the other hand, individually implementing
Various features described in example can also be performed separately in various embodiments or be implemented with any suitable sub-portfolio.This
Outside, although feature can work in certain combinations as described above and even initially so be claimed, institute is come from
One or more features in claimed combination can be removed from the combination in some cases, and claimed
Combination can be directed toward the modification of sub-portfolio or sub-portfolio.
Similarly, although depicting operation in the accompanying drawings with particular order, this is understood not to require these behaviour
Make the particular order shown in execute or sequentially carry out or require the operation of all illustrations to be performed, to realize desired knot
Fruit.In some cases, multitask and parallel processing may be advantageous.In addition, the various system modules in above-described embodiment
Separation with component is understood not to be required to such separation in all embodiments, and it is to be understood that described
Program assembly and system can be usually integrated in together in single software product, or be packaged into multiple software product.
The specific embodiment of theme has been described as a result,.Other embodiments are within the scope of the appended claims.?
In some cases, the movement recorded in claims can be executed in different order and still realize desired result.This
Outside, the processing described in attached drawing and it is nonessential shown in particular order or sequential order, to realize desired result.In certain realities
In existing, multitask and parallel processing be may be advantageous.
The foregoing is merely the preferred embodiments of the application, not to limit the application, all essences in the application
Within mind and principle, any modification, equivalent substitution, improvement and etc. done be should be included within the scope of the application protection.
Claims (10)
1. a kind of echo planar imaging neural network based is applied to magnetic resonance imaging system, which is characterized in that described
Method includes:
According to the magnetic resonance parameters of selection, subject is scanned using Echo-plane imaging sequence, obtains the number of k-space
According to；
According to the data of the k-space, the initial magnetic resonance images of the subject are obtained；
By initial magnetic resonance images input trained in advance first nerves network and nervus opticus network, obtain described tested
The final magnetic resonance image of body, wherein the first nerves network is configured as eliminating the neural network of chemical shift artifact, institute
State the neural network that nervus opticus network is configured as remedial frames deformation.
2. the method according to claim 1, wherein the Echo-plane imaging sequence includes following any:
Gradient-echo echo-planar imaging sequence；
Spin echo Echo-plane imaging sequence；
Inversion recovery Echo-plane imaging sequence.
3. the method according to claim 1, wherein the initial magnetic resonance images are inputted training in advance
Before first nerves network, the method also includes:
Obtain first sample set, wherein the sample that the first sample is concentrated includes the fatty input with chemical shift artifact
Magnetic resonance image and the first standard picture；
According to the first sample set, the training first nerves network.
4. the method according to claim 1, wherein the initial magnetic resonance images are inputted training in advance
Before nervus opticus network, the method also includes:
Obtain the second sample set, wherein the sample of second sample set includes at least the artifact figure of a width deformation；
According to second sample set, the training nervus opticus network.
5. according to the method described in claim 4, it is characterized in that, the sample of second sample set includes following any
Kind:
Artifact figure of two width along out of phase coding direction deformation；
Artifact figure and second standard picture of two width along out of phase coding direction deformation；
The artifact figure and third standard picture of one width deformation.
6. a kind of Echo-plane imaging device neural network based is applied to magnetic resonance imaging system, which is characterized in that described
Device includes:
Scan module is scanned subject using Echo-plane imaging sequence, obtains for the magnetic resonance parameters according to selection
To the data of k-space；
First reconstruction module obtains the initial magnetic resonance images of the subject for the data according to the k-space；
Second rebuilds module, by initial magnetic resonance images input trained in advance first nerves network and nervus opticus net
Network obtains the final magnetic resonance image of the subject, wherein the first nerves network is configured as eliminating chemical shift puppet
The neural network of shadow, the nervus opticus network are configured as the neural network of remedial frames deformation.
7. device according to claim 6, which is characterized in that described device further includes the first training module, is used for:
Obtain first sample set, wherein the sample of the first sample set includes the fatty input magnetic with chemical shift artifact
Resonance image and the first standard picture；
According to the first sample set, the training first nerves network.
8. device according to claim 6, which is characterized in that described device further includes the second training module, is used for:
Obtain the second sample set, wherein the sample of second sample set includes at least the artifact figure of a width deformation；
According to second sample set, the training nervus opticus network.
9. device according to claim 8, which is characterized in that the sample of second sample set includes following any
Kind:
Artifact figure of two width along out of phase coding direction deformation；
Artifact figure and second standard picture of two width along out of phase coding direction deformation；
The artifact figure and third standard picture of one width deformation.
10. a kind of Echo-plane imaging device neural network based is applied to magnetic resonance imaging system, which is characterized in that institute
State device include: memory, processor and storage on a memory and the computer program that can run on a processor, the place
It manages device and executes method of the described program to realize any Echo-plane imaging neural network based in claim 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910395169.4A CN110095742B (en) | 2019-05-13 | 2019-05-13 | Plane echo imaging method and device based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910395169.4A CN110095742B (en) | 2019-05-13 | 2019-05-13 | Plane echo imaging method and device based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110095742A true CN110095742A (en) | 2019-08-06 |
| CN110095742B CN110095742B (en) | 2022-02-08 |
Family
ID=67447920
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910395169.4A Active CN110095742B (en) | 2019-05-13 | 2019-05-13 | Plane echo imaging method and device based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110095742B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110895320A (en) * | 2019-10-31 | 2020-03-20 | 清华大学 | Deformation-free single-shot planar echo imaging method and device based on deep learning |
| CN112763958A (en) * | 2020-12-10 | 2021-05-07 | 复旦大学 | Multi-excitation plane echo magnetic resonance imaging method based on neural network |
| CN112890798A (en) * | 2019-12-04 | 2021-06-04 | 上海联影医疗科技股份有限公司 | Neural network training method, magnetic resonance imaging method, apparatus, device, and medium |
| CN113534031A (en) * | 2020-04-21 | 2021-10-22 | 上海联影医疗科技股份有限公司 | Image domain data generating method, computer device and readable storage medium |
| WO2024083112A1 (en) * | 2022-10-19 | 2024-04-25 | 之江实验室 | Method and system for simulating artifact in magnetic resonance echo-planar imaging |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104035059A (en) * | 2013-03-06 | 2014-09-10 | 上海联影医疗科技有限公司 | Echo planar imaging sequence image reconstruction method |
| CN108132274A (en) * | 2017-12-21 | 2018-06-08 | 厦门大学 | Echo-planar imaging is without reference scan pattern distortion antidote under non-uniform magnetic field |
| CN109544534A (en) * | 2018-11-26 | 2019-03-29 | 上海联影智能医疗科技有限公司 | A kind of lesion image detection device, method and computer readable storage medium |
| CN109727663A (en) * | 2018-09-30 | 2019-05-07 | 上海联影医疗科技有限公司 | The system and method for generating the neural network model for image procossing |
| WO2019086337A1 (en) * | 2017-10-31 | 2019-05-09 | Koninklijke Philips N.V. | Motion artifact prediction during data acquisition |
| CN109741409A (en) * | 2018-11-30 | 2019-05-10 | 厦门大学 | Echo-planar imaging eddy current artifacts without reference scan bearing calibration |
-
2019
- 2019-05-13 CN CN201910395169.4A patent/CN110095742B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104035059A (en) * | 2013-03-06 | 2014-09-10 | 上海联影医疗科技有限公司 | Echo planar imaging sequence image reconstruction method |
| WO2019086337A1 (en) * | 2017-10-31 | 2019-05-09 | Koninklijke Philips N.V. | Motion artifact prediction during data acquisition |
| CN108132274A (en) * | 2017-12-21 | 2018-06-08 | 厦门大学 | Echo-planar imaging is without reference scan pattern distortion antidote under non-uniform magnetic field |
| CN109727663A (en) * | 2018-09-30 | 2019-05-07 | 上海联影医疗科技有限公司 | The system and method for generating the neural network model for image procossing |
| CN109544534A (en) * | 2018-11-26 | 2019-03-29 | 上海联影智能医疗科技有限公司 | A kind of lesion image detection device, method and computer readable storage medium |
| CN109741409A (en) * | 2018-11-30 | 2019-05-10 | 厦门大学 | Echo-planar imaging eddy current artifacts without reference scan bearing calibration |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110895320A (en) * | 2019-10-31 | 2020-03-20 | 清华大学 | Deformation-free single-shot planar echo imaging method and device based on deep learning |
| WO2021082103A1 (en) * | 2019-10-31 | 2021-05-06 | 清华大学 | Distortionless single-shot echo planar imaging method and apparatus employing deep learning |
| CN110895320B (en) * | 2019-10-31 | 2021-12-24 | 清华大学 | Deformation-free single-shot planar echo imaging method and device based on deep learning |
| CN112890798A (en) * | 2019-12-04 | 2021-06-04 | 上海联影医疗科技股份有限公司 | Neural network training method, magnetic resonance imaging method, apparatus, device, and medium |
| CN112890798B (en) * | 2019-12-04 | 2023-04-25 | 上海联影医疗科技股份有限公司 | Neural network training method, magnetic resonance imaging method, device, equipment and medium |
| CN113534031A (en) * | 2020-04-21 | 2021-10-22 | 上海联影医疗科技股份有限公司 | Image domain data generating method, computer device and readable storage medium |
| CN112763958A (en) * | 2020-12-10 | 2021-05-07 | 复旦大学 | Multi-excitation plane echo magnetic resonance imaging method based on neural network |
| CN112763958B (en) * | 2020-12-10 | 2022-06-21 | 复旦大学 | Multi-excitation plane echo magnetic resonance imaging method based on neural network |
| WO2024083112A1 (en) * | 2022-10-19 | 2024-04-25 | 之江实验室 | Method and system for simulating artifact in magnetic resonance echo-planar imaging |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110095742B (en) | 2022-02-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6786670B2 (en) | Magnetic resonance imaging device | |
| CN103211596B (en) | system for motion corrected MR diffusion imaging | |
| CN103608693B (en) | Contrast with the chemical shift coding for fat suppression strengthens magnetic resonance angiography | |
| US8116541B2 (en) | Method and apparatus for multi-coil magnetic resonance imaging | |
| CN110095742A (en) | A kind of echo planar imaging neural network based and device | |
| JP5221570B2 (en) | Magnetic resonance imaging apparatus and multi-contrast image acquisition method | |
| CN105005011B (en) | The MR imaging method of mixing acquisition in 3-dimensional fast spin echo | |
| US10324153B2 (en) | System and method for flip angle determination in magnetic resonance imaging | |
| US8482279B2 (en) | System and method of parallel imaging for magnetic resonance imaging near metallic implants | |
| JP2018519909A (en) | MR imaging using motion detection | |
| US9746539B2 (en) | MR imaging with suppresion of flow artifacts | |
| JP5848713B2 (en) | Magnetic resonance imaging apparatus and contrast-enhanced image acquisition method | |
| EP2730935A1 (en) | Motional error correction in functional magnetic resonance imaging | |
| US10302729B2 (en) | Method and magnetic resonance apparatus for speed-compensated diffusion-based diffusion imaging | |
| CN109907759B (en) | Magnetic resonance imaging method and system | |
| JP7272818B2 (en) | Data processing device, method and program | |
| CN102597795A (en) | Mr imaging using navigators | |
| JP6510273B2 (en) | Magnetic resonance imaging apparatus, magnetic resonance imaging method and magnetic resonance imaging program | |
| CN103505214A (en) | A system for simultaneous dual-slab acquisition of MR images | |
| CN108459289A (en) | A kind of multiple excitation Diffusion weighted MR imaging method based on data consistency | |
| CN105829906A (en) | Phase-Sensitive Inversion Recovery Mri With Water/Fat Separation | |
| WO2014154544A1 (en) | Real-time motion correction for mri using fat navigators | |
| CN107076819A (en) | The Dixon MR imagings of suppression with convection current artifact | |
| JP2017529960A (en) | Propeller MR imaging with artifact suppression | |
| US10976397B2 (en) | MRI apparatus utilizing non-ultrashort TE(UTE) imaging to generate a mask image for performance of mask processing |
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 |