CN113470128A - Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device - Google Patents

Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device Download PDF

Info

Publication number
CN113470128A
CN113470128A CN202010239184.2A CN202010239184A CN113470128A CN 113470128 A CN113470128 A CN 113470128A CN 202010239184 A CN202010239184 A CN 202010239184A CN 113470128 A CN113470128 A CN 113470128A
Authority
CN
China
Prior art keywords
data
image
training
space
slice
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010239184.2A
Other languages
Chinese (zh)
Inventor
王海峰
梁栋
王婉婷
苏适
刘新
郑海荣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Priority to CN202010239184.2A priority Critical patent/CN113470128A/en
Publication of CN113470128A publication Critical patent/CN113470128A/en
Pending legal-status Critical Current

Links

Images

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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves 
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/003—Reconstruction from projections, e.g. tomography
    • G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2211/00—Image generation
    • G06T2211/40—Computed tomography

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Evolutionary Biology (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Radiology & Medical Imaging (AREA)
  • Pathology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

The invention discloses a method for reconstructing simultaneous multilayer imaging signals. The reconstruction method comprises the following steps: generating a multi-slice virtual automatic calibration signal according to the acquired multi-slice automatic calibration signal, wherein the multi-slice automatic calibration signal and the virtual automatic calibration signal are complex data of k space; generating a training data set according to the multi-slice automatic calibration signal and the virtual automatic calibration signal; training the neural network model by using a training data set; inputting the obtained original aliasing data of the multiple layers into a trained neural network model to obtain corresponding network output data; according to the fusion of the original aliasing data and the network output data, complete reconstruction data of a k space are obtained; a reconstructed image is generated from the complete reconstruction data. The phase information of the virtual automatic calibration signal is fully utilized, the neural network model can be trained more fully, the virtual signal is used as another training data set, and a reconstructed image with higher quality is obtained under the condition that extra scanning time is not increased.

Description

Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device
Technical Field
The present invention belongs to the field of image reconstruction technology of magnetic resonance imaging signals, and in particular, to a reconstruction method of simultaneous multi-slice imaging signals, a computer readable storage medium, and a computer device.
Background
Meanwhile, the Multi-Slice imaging (SMS) is a fast Magnetic Resonance (MR) scanning imaging method, which can simultaneously excite and acquire aliasing signals of a plurality of slices within the acquisition time length of a traditional Slice image, and reconstruct each Slice by using redundant information contained in a Multi-channel acquisition coil, thereby realizing the accelerated acquisition and reconstruction of Magnetic Resonance. The SMS reconstruction needs to separate a plurality of slice signals from an aliasing signal by using redundant information contained in a multi-channel coil, so that the higher the utilization rate of the redundant information by the SMS reconstruction algorithm is, the higher the acceleration factor can be obtained. For example, a Controlled Aliasing in Parallel Imaging resources in high elevation acquisition (CAIPIRINHA) technology is used to improve the Acceleration performance of SMS, and different linear phases are added to a plurality of slices excited simultaneously by modulating the phase of a multiband (Multiple Band) radio frequency pulse or applying an extra gradient field, so that the plurality of slices acquired simultaneously form different spatial offsets in an Aliasing image, thereby more fully utilizing the multi-channel coil information difference in an Imaging field and realizing SMS Acceleration of Higher multiples.
In the SMS reconstruction method using the CAIPIRINHA technique, Slice-GRAPPA is the most commonly used conventional reconstruction method, and it uses the acquired Auto Calibration Signal (ACS) data with low resolution to estimate and reconstruct the linear kernel, which has wide application convenience. Such SMS reconstruction methods are based on linear kernel methods, which only weight the selected regions linearly, cannot be adaptively changed, and have a low utilization of redundant information contained in the multi-channel receive coil.
Disclosure of Invention
(I) technical problems to be solved by the invention
The technical problem solved by the invention is as follows: how to better utilize the phase information of the magnetic resonance data to improve the image reconstruction quality.
(II) the technical scheme adopted by the invention
A method of simultaneous multi-slice imaging signal reconstruction, the method comprising:
generating a multi-slice virtual automatic calibration signal according to the acquired multi-slice automatic calibration signal, wherein the multi-slice automatic calibration signal and the virtual automatic calibration signal are complex data of k space;
generating a training data set according to the multi-slice automatic calibration signal and the virtual automatic calibration signal;
training the neural network model by using a training data set;
inputting the obtained original aliasing data of the multiple layers into a trained neural network model to obtain corresponding network output data;
according to the fusion of the original aliasing data and the network output data, complete reconstruction data of a k space are obtained;
a reconstructed image is generated from the complete reconstruction data.
Preferably, the specific method for generating the training data set according to the multi-slice auto-calibration signal and the virtual auto-calibration signal comprises:
and generating a first training sample according to the multi-slice automatic calibration signal, and generating a second training sample according to the multi-slice virtual automatic calibration signal, wherein the first training sample and the second training sample form two training data sets.
Preferably, the method for training the neural network model by using the training data set comprises the following steps:
training a neural network model by using a first training sample and a second training sample, wherein the first training sample comprises first input training data and first output training data, the first input training data is used for training an input part of the neural network model, and the first output training data is used for training an output part of the neural network model; the second training samples include second input training data for training an input portion of the neural network model and second output training data for training an output portion of the neural network model.
Preferably, the specific method for generating the first training sample according to the multi-slice auto-calibration signal comprises:
respectively performing inverse Fourier transform on each layer of automatic calibration signals of the k space to generate each layer of low-resolution images in an image space;
connecting the low-resolution images of the layers in the image space in a readout direction to generate a multi-layer connection image;
carrying out Fourier transform on the multilayer connection images to generate k-space data corresponding to the multilayer connection images;
normalizing k-space data corresponding to the multilayer connected image to obtain k-space data corresponding to the normalized multilayer connected image;
and performing down-sampling processing of an acceleration factor MB on the k-space data corresponding to the normalized multi-layer connection image to obtain first input training data, and taking the residual data of the k-space data corresponding to the multi-layer connection image after the down-sampling processing as first output training data, wherein the MB is equal to the number of the multi-layer sheets.
Preferably, the specific method for generating the second training sample according to the multi-slice virtual automatic calibration signal includes:
respectively performing inverse Fourier transform on each layer of virtual automatic calibration signals of the k space to generate each layer of virtual low-resolution images in an image space;
connecting each layer of virtual low-resolution images in an image space in a reading direction to generate a multilayer connected virtual image;
carrying out Fourier transformation on the multilayer connection virtual image to generate k space data corresponding to the multilayer connection virtual image;
normalizing k-space data corresponding to the multilayer connection virtual image to obtain k-space data corresponding to the normalized multilayer connection virtual image;
and performing down-sampling processing of an acceleration factor MB on the k-space data corresponding to the multilayer connection virtual image to obtain second input training data, and taking the residual data of the k-space data corresponding to the multilayer connection virtual image after the down-sampling processing as second output training data, wherein the MB is equal to the number of slices of the multilayer connection virtual image.
Preferably, the specific method for obtaining the complete reconstruction data according to the fusion of the original aliasing data and the corresponding network output data comprises:
constructing a zero padding data space in which the multiple slices are connected in the reading direction;
interpolating original aliased data in k-space into the zero-padded data space with MB times upsampling in a readout direction;
and interpolating the network output data corresponding to the original aliasing data to the residual position of the zero padding data space to obtain complete reconstruction data connected in the reading direction.
Preferably, the specific method for generating the reconstructed image according to the complete reconstruction data comprises the following steps:
performing inverse Fourier transform on the complete reconstruction data connected in the readout direction to generate a complete multi-slice image connected in the readout direction;
cutting the complete multi-slice image connected in the readout direction into MB slices to obtain reconstructed MB images;
and fusing the multi-coil data of each image by using a coil fusion square sum method to obtain MB layer image data.
The invention also discloses a computer readable storage medium, which stores a reconstruction program of the simultaneous multi-slice imaging signals, and the reconstruction program of the simultaneous multi-slice imaging signals realizes the reconstruction method of the simultaneous multi-slice imaging signals when being executed by a processor.
The invention also discloses a computer device, which comprises a computer readable storage medium, a processor and a reconstruction program of the simultaneous multi-slice imaging signal stored in the computer readable storage medium, wherein the reconstruction program of the simultaneous multi-slice imaging signal realizes the reconstruction method of the simultaneous multi-slice imaging signal when being executed by the processor.
(III) advantageous effects
The invention discloses a method for reconstructing simultaneous multilayer imaging signals, which has the following technical effects compared with the traditional calculation method:
the method has the advantages that the virtual automatic calibration signal doubled according to the expansion of the actually acquired automatic calibration signal is generated, the data characteristics and information of the SMS aliasing data on the virtual conjugate coil are fully utilized, the neural network model can be trained more fully, the reconstructed image with higher quality can be obtained under the condition of not increasing extra scanning time, and the artifact and noise of the reconstructed image are reduced.
Drawings
FIG. 1 is a flow chart of a method of simultaneous multi-slice imaging signal reconstruction in accordance with an embodiment of the present invention;
FIG. 2 is a flow chart of constructing a first training sample according to an embodiment of the present invention;
FIG. 3 is a flow chart of constructing a second training sample according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a process for constructing training data according to an embodiment of the present invention;
FIG. 5 is a flow chart of fusing raw aliased data and network output data according to an embodiment of the invention;
FIG. 6 is a schematic diagram of a process for fusing raw aliased data and network output data according to an embodiment of the invention;
fig. 7 is a fully sampled reference image, an image obtained by the conventional method and the reconstruction method of the present application when the acceleration factor is MB 2, according to an embodiment of the present invention.
Fig. 8 is an error image ratio obtained by the conventional method and the reconstruction method of the present application when the acceleration factor is MB-2 according to the embodiment of the present invention;
fig. 9 is a fully sampled reference image, an image obtained by the conventional method and the reconstruction method of the present application when the acceleration factor is MB-3, according to an embodiment of the present invention.
Fig. 10 is an error image ratio obtained by the conventional method and the reconstruction method of the present application when the acceleration factor is MB-3 according to the embodiment of the present invention;
fig. 11 is a fully sampled reference image, an image obtained by the conventional method and the reconstruction method of the present application when the acceleration factor is MB-4, according to an embodiment of the present invention.
Fig. 12 is an error image ratio obtained by the conventional method and the reconstruction method of the present application when the acceleration factor is MB-4 according to the embodiment of the present invention;
FIG. 13 is a functional block diagram of a computer device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Before describing in detail the various embodiments of the present application, the inventive concepts of the present application are first briefly described: in order to solve the technical problem that the existing SMS reconstruction method cannot effectively utilize redundant information contained in a multi-channel receiving coil, virtual data which is twice as much as the complex conjugate symmetry characteristic of real data acquired by a magnetic resonance multi-channel coil is generated, and the real data and the virtual data generated correspondingly are used as training samples of a convolutional neural network, so that the channel redundant information of the magnetic resonance receiving coil is fully utilized, and a reconstructed image with higher quality is obtained.
Specifically, as shown in fig. 1, the method for reconstructing a simultaneous multi-slice imaging signal of the present application includes the following steps:
step S10: and acquiring a fully sampled multi-slice automatic calibration signal, and generating a multi-slice virtual automatic calibration signal according to the multi-slice automatic calibration signal, wherein the multi-slice automatic calibration signal and the virtual automatic calibration signal are complex data of k space.
Specifically, the data to be acquired by the application includes original aliasing data acquired by a simultaneous multi-slice imaging technology and an automatic calibration signal, wherein the automatic calibration signal is a low-frequency signal which is equivalent to prior information and is used for training a neural network. The raw aliased data includes low frequency signals and high frequency signals, which are used to acquire missing data through a neural network for final image reconstruction. First, the SMS technique of the present application is preferably an SMSiES technique, and is specifically implemented by simultaneously exciting imaging in multiple planes in combination with a multiple Echo shift technique and an accelerated Gradient Echo (GRE) sequence. The technology simultaneously excites a plurality of lamella in an effective repetition time, applies a staggered lamella acquisition scheme and combines an echo shift technology, thereby avoiding signal loss, simultaneously not influencing the acquisition speed, and simultaneously applying the CAIPIRINHA technology to increase the difference of coil sensitivity, thereby being convenient for more accurately separating each lamella signal.
In the magnetic resonance field, data in the fourier domain, referred to as k-space data, are acquired directly by the above-described techniques. A slice of raw aliased data containing multiple slices is acquired simultaneously in magnetic resonance. In order to find the mapping relation between the original aliasing data and the missing data by using the neural network, ACS data of a plurality of corresponding slices, namely low-resolution automatic calibration signals, are acquired slice by slice, the acquisition parameters of the automatic calibration signals are the same as the acquisition parameters of the original aliasing data, and 15 central phase encoding lines are acquired as the automatic calibration signals.
Illustratively, the present application acquires aliased data of a subject's head with a simultaneous excitation slice number MB of 5, where slices 1, 3, 5 have no intra-slice offset and slices 2, 4 have a field-of-view offset of 1/2FOV in the phase-encoding direction, with the same acquisition parameters for each data. The resulting data matrix is [ N ]pe，Nfe，Ncoil，Nsample]＝[192，192，32，3]Wherein N ispe、Nfe、Ncoil、NsampleRespectively representing the number of phase encoding lines, the number of readout points, the number of coils and the number of samples. The auto-calibration signals required to train the neural network are acquired slice by slice under the same parameters as the raw alias data, so that the final data matrix obtained is [ N [ ]acs，Nfe，Ncoil，Nsl，Nsample]＝[15，192，32，5，3]，NslNumber of layers of the sheet, NacsIndicating the number of phase encoding lines acquired. Meanwhile, the comparison is convenient, and the full sampling reference data of the slice layer is also obtained, and the matrix size is [ N ]pe，Nfe，Ncoil，Nsl，Nsample]＝[192，192，32，5，3]。
The present invention uses the above-described fully-sampled reference data to construct raw aliasing data with an acceleration factor of 3 (MB-3) and a corresponding auto-calibration signal by retrospective downsampling. There is a cyclic equidistant shift offset of 2 pi/3 between the slices, which is manifested in the image as an offset of 2/3FOV in the phase encoding direction. After reconstruction is complete, only the offset needs to be restored in image space.
Further, the fully sampled multi-slice auto-calibration signal is expanded by using a multi-channel Virtual coil (VCC) technique to generate one more Virtual auto-calibration signal. Specifically, the virtual auto-calibration signal is expanded by using the complex conjugate symmetry relation for the auto-calibration signal of each slice acquired by magnetic resonance. The expansion is performed according to the formula (1),
Figure BDA0002431987000000072
a virtual auto-calibration signal representing coil j at Fourier space position k, denoted by Sj(k) Complex conjugation of
Figure BDA0002431987000000073
Thus obtaining the product. Where L denotes the actual number of coils, j is the coil index, and x denotes the complex conjugate operation. In the present application, L is equal to 32, and after applying the VCC method, L remains unchanged, which is an important point different from the application of the VCC method in the conventional reestablishment. The inventionThe expanded automatic calibration signal is used as another sample for training, which is equivalent to increase of training data volume without modifying specific network parameters, and the feasible reason is that the artificial neural network method fuses the phase information of the automatic calibration signal and the virtual automatic calibration signal in the model training process, so that the training effect is stably improved.
Figure BDA0002431987000000071
Step S20: a training data set is generated from the multi-slice auto-calibration signal and the virtual auto-calibration signal.
Specifically, a first training sample is generated according to the multi-slice automatic calibration signal, a second training sample is generated according to the multi-slice virtual automatic calibration signal, and the first training sample and the second training sample form two training data sets.
As a preferred embodiment, the first training sample comprises first input training data for training an input portion of the neural network model and first output training data for training an output portion of the neural network model. As shown in fig. 2, a specific method for generating a first training sample according to an auto-calibration signal of multiple slices includes the following steps:
step S21: the respective layers of the auto-calibration signals of k-space are inverse fourier transformed to generate respective layers of low resolution images in image space.
Specifically, taking 3-layer data as an example, two-dimensional inverse fourier transform is performed on the acquired 3-layer automatic calibration signals of the k space respectively to obtain 3-layer low-resolution images in the image space, where a method of the two-dimensional inverse fourier transform is the prior art and is not described herein again.
Step S22: layers of low resolution images in image space are connected in a readout direction to generate a multi-layer connected image.
As a preferred embodiment, the 3-layer low resolution images are concatenated in the readout direction (X direction) to form a single multi-layer concatenated image, which is an image space image.
Step S23: fourier transform is performed on the multi-layer connected images to generate k-space data corresponding to the multi-layer connected images.
Specifically, a multi-layered connected image in an image space is transformed into a k-space by fourier transform to generate k-space data corresponding to the multi-layered connected image, wherein each data point in the k-space contains information of the whole image.
Step S24: and carrying out normalization processing on k-space data corresponding to the multilayer connection images to obtain k-space data corresponding to the normalized multilayer connection images.
Step S25: and performing down-sampling processing of an acceleration factor MB on the k-space data corresponding to the normalized multi-layer connection image to obtain first input training data, and taking the residual data of the k-space data corresponding to the multi-layer connection image after the down-sampling processing as first output training data, wherein the MB is equal to the number of the multi-layer sheets.
As shown in fig. 4, the arrangement of k-space data in the present application is in a cartesian form, phase lines are in a y-direction, frequency-encoded data are in an x-direction, which are simply referred to as encoding lines, each encoding line is arranged in sequence, each encoding line is filled with data, the data of different encoding lines represent data at different positions, and there is a nonlinear relationship between the data on the encoding lines.
For visual description, in the present application, taking an acceleration factor MB equal to 3 as an example, performing down-sampling processing of an acceleration factor 3 on k-space data corresponding to a normalized multi-layer connection image to obtain first input training data, where data on 1 st, 4 th, 7 th, and 10 th encoding lines … … may be used as the first input training data; the remaining data of the k-space data corresponding to the multi-layer connected image after the down-sampling process is used as the first output training data, and in fig. 4, the data on the 2 nd, 3 rd, 5 th, and 6 th … … encoding lines may be used as the first output training data. By the construction method, the neural network can be trained by the down-sampled data, so that the nonlinear relation between the down-sampled data and the adjacent position data is obtained.
As a preferred embodiment, the second training sample includes second input training data and second output training data, the second input training data is used for training the input part of the neural network model, the second output training data is used for training the output part of the neural network model, and the specific method for generating the second training sample according to the multi-slice virtual automatic calibration signal includes:
step S21': and respectively carrying out inverse Fourier transform on each layer of the virtual automatic calibration signals of the k space to generate each layer of virtual low-resolution images in the image space.
Correspondingly, the automatic calibration signal is 3-layer data, the generated virtual automatic calibration signal is also 3-layer data, and the acquired 3-layer virtual automatic calibration signal of the k space is respectively subjected to two-dimensional inverse fourier transform to obtain 3-layer virtual low-resolution images in the image space, wherein the method of the two-dimensional inverse fourier transform is the prior art and is not described herein again.
Step S22': layers of the virtual low resolution images in the image space are connected in the readout direction to generate a multilayer connected virtual image.
As a preferred embodiment, the 3-layer virtual low resolution images are connected in the readout direction (X direction) to form one multi-layer connected virtual image, which is an image in image space.
Step S23': and carrying out Fourier transformation on the multilayer connection virtual image to generate k-space data corresponding to the multilayer connection virtual image.
Specifically, a multi-layer connected virtual image of the image space is transformed into k-space through Fourier transform to generate k-space data corresponding to the multi-layer connected virtual image, wherein each data point in the k-space contains information of the whole virtual image.
Step S24': and carrying out normalization processing on the k-space data corresponding to the multilayer connection virtual image to obtain the k-space data corresponding to the normalized multilayer connection virtual image.
Step S25': and performing down-sampling processing of an acceleration factor MB on the k-space data corresponding to the multilayer connection virtual image to obtain second input training data, and taking the residual data of the k-space data corresponding to the multilayer connection virtual image after the down-sampling processing as second output training data, wherein the MB is equal to the number of slices of the multilayer connection virtual image.
As shown in fig. 4, the arrangement of k-space data in the present application is in a cartesian form, that is, the k-space data are sequentially arranged according to each encoding line, different encoding lines represent data at different positions, and there is a nonlinear relationship between the data on the encoding lines.
For the purpose of visual description, in the present application, taking an acceleration factor MB equal to 3 as an example, performing down-sampling processing of an acceleration factor 3 on k-space data corresponding to a normalized multi-layer connected virtual image to obtain second input training data, and in fig. 4, data on the 1 st, 4 th, 7 th, and 10 th encoding lines … … may be used as the second input training data; the residual data of the k-space data corresponding to the multi-layer connected image after the down-sampling process is used as second output training data, and the data on the 2 nd, 3 rd, 5 th and 6 th encoding lines … … in the image can be used as second output training data. By the construction method, the neural network can be trained by the down-sampled data, so that the nonlinear relation between the down-sampled data and the adjacent position data is obtained.
Step S30: training the neural network model using the training data set.
Specifically, the first training sample and the second training sample constructed in step S20 are used to train the neural network model, that is, the first input training data and the second input training data are used to train the input part of the neural network model, and the first output training data and the second output training data are used to train the output part of the neural network model. It should be noted that, in each round of training, the training data of the input part and the output part of the neural network model are from the same training sample, i.e. the first training sample or the second training sample.
As a preferred embodiment, the neural network model of the present application is a convolutional neural network with two layers, and the training parameters are set as follows: the input part is [ N ]s，Npe，Nfe，Nc*2]The output part is [ N ]s，Npe,，Nfe，Nc*2*(MB-1)]The loss function is the Mean Absolute Error (MAE), and Adam is selected for the optimizer. Initializing network weights with normal distribution, the first layer network weight being [5, 3, Nc*2，Nc*2*(MB-1)]The layer-two network weight is [2, 2, Nc*2*(MB-1)，Nc*2*(MB-1)]The initial learning rate was 0.001, decaying exponentially at 0.98, for a total of 1500 rounds of training.
Further, the fitting equation of the k-space signal is as follows (2),
Sj,1..L(kx,ky)≈fj({Sl,collapse(kx-bxΔkx,ky-byΔky)}bx∈[-Bx,Bx],by∈[-By,By],l∈[1,L]) (2)
the equation indicates a non-linear mapping between the input and output signals of the network, the learned parameters being embodied in the variable b, i.e. the weight. Wherein j ∈ 1.. L represents a coil index; k is a radical ofx，kyPosition indexes respectively representing a reading direction and a phase encoding direction, wherein z belongs to 1. bxAnd byEach representing an index of a fitted core of size 2BxAnd 2. By。Sj,1..LFull acquisition k-space data, S, representing all slices of coil jl,collapseAn aliased k-space signal, f, representing the coil/jRepresenting a non-linear mapping function from the aliased signal to the reconstructed signal. This formulation expresses the signal S from aliasingl,collapseNon-linear mapping f using network weightsjRecovering all the slices of the coil j at kx，kyA reconstructed signal of the location.
Step S40: and inputting the acquired original aliasing data of the multiple layers into the trained neural network model to obtain corresponding network output data.
Specifically, after training in step S30, the neural network model obtains a nonlinear relationship between aliasing data and missing k-space data adjacent in the readout direction, that is, a mapping relationship between input data and output data. And inputting the original aliasing data of the multi-slice layer obtained in the step S10 into the neural network model, outputting the corresponding network output data by the neural network model, wherein the original aliasing data can be regarded as undersampled data in the readout direction, the obtained network output data is equivalent to the other part of missing data of the image to be reconstructed in the readout direction, and the original aliasing data and the network output data are used together for reconstructing the complete image.
Step S50: and obtaining complete reconstruction data of the k space according to the fusion of the original aliasing data and the network output data.
The network output data predicted by the neural network model in step S40 cannot be directly used for reconstructing an image, and needs to be fused with the original aliasing data to obtain complete reconstruction data, specifically, as shown in fig. 5, the step S50 specifically includes the following steps:
step S51: a zero-padded data space is constructed in which multiple slices are connected in the readout direction.
Step S52: the original aliased data in k-space is interpolated in the readout direction in MB times upsampling into the zero-padded data space.
As shown in fig. 6, the filling manner of the zero padding data space is a cartesian form, each encoding line is sequentially arranged along the readout direction, and when interpolation is not performed, the value of each encoding line of the zero padding data space is zero. Taking MB equal to 3 as an example, each layer of original aliasing data in k-space is interpolated into zero-padding data space by 3 times of upsampling, which corresponds to the original aliasing data being sequentially padded onto the 1 st, 4 th, 7 th and 10 th … … encoding lines in fig. 6.
Step S53: and interpolating the network output data corresponding to the original aliasing data to the residual position of the zero padding data space to obtain complete reconstruction data of the MB layer connected in the reading direction.
Taking MB equal to 3 as an example, interpolating the network output data corresponding to the original aliasing data to the remaining positions of the multi-slice zero padding data, which corresponds to sequentially padding the network output data to the 2 nd, 3 rd, 5 th, 6 th, 8 th, and 9 th … … encoding lines in fig. 6, so that the zero padding data space is filled with the original aliasing data and the network output data together, thereby obtaining the multi-slice connected complete k-space reconstruction data.
Step S60: a reconstructed image is generated from the multi-slice concatenated complete k-space reconstruction data.
Specifically, the complete reconstructed data connected in the readout direction is subjected to inverse fourier transform to generate a reconstructed image with multiple slices connected in the readout direction, the reconstructed image with multiple slices connected in the readout direction is further cut into MB slices to obtain reconstructed MB images, and then the multi-coil data of each image is fused by using a coil fusion method to obtain final MB-layer image data, so that an MB-layer reconstructed image is obtained. In the invention, the coil fusion method is a square sum method, namely, the images of all coils are subjected to square sum processing. Due to the adoption of the CAIPIRINHA acquisition method, each layer of the reconstructed multilayer image has image displacement in the phase encoding direction, and only the motion recovery is needed.
According to the method for reconstructing the simultaneous multilayer imaging signals, the virtual automatic calibration signals which are doubled are expanded and generated according to the actually acquired automatic calibration signals, the data characteristics and information of the down-sampled data in the slice direction on the virtual conjugate coils are fully utilized, a neural network model can be trained more fully, a reconstructed image with higher quality can be obtained under the condition that extra scanning time is not increased, and artifacts and noises of the reconstructed image are reduced.
In order to prove the advantages of the image reconstruction method, the full sampling image and the reconstructed image obtained by the reconstruction method are compared through experiments, the reconstructed image obtained by the reconstruction method is compared with the reconstructed image obtained based on the convolutional neural network method through experiments, and evaluation indexes comprise a peak signal to noise ratio (PSNR), Structural Similarity (SSIM) and Root Mean Square Error (RMSE), wherein the higher the PSNR and the SSIM, the higher the image quality is, and the smaller the RMSE is, the higher the image quality is.
As shown in fig. 7, the full-sampling reference image, the image reconstructed by the conventional method when the acceleration factor is MB is 2, and the image obtained by the reconstruction method of the present application are sequentially provided from left to right, as shown in fig. 8, it can be known by comparing the image reconstructed by the conventional method and the image error obtained by the reconstruction method of the present application, and the numerical value at the lower left corner of each image sequentially represents the PSNR value, the SSIM value, and the RMSE value, and it can be found that both the PSNR value and the SSIM value are improved, and the RMSE value is reduced, which indicates that the image quality is improved.
As shown in fig. 7, the full-sampling reference image, the image reconstructed by the conventional method when the acceleration factor is MB is 2, and the image obtained by the reconstruction method of the present application are sequentially provided from left to right, as shown in fig. 8, it can be known by comparing the image reconstructed by the conventional method and the image error obtained by the reconstruction method of the present application, and the numerical value at the lower left corner of each image sequentially represents the PSNR value, the SSIM value, and the RMSE value, and it can be found that both the PSNR value and the SSIM value are improved, and the RMSE value is reduced, which indicates that the image quality is improved.
As shown in fig. 9, the full sampling reference image, the image reconstructed by the conventional method when the acceleration factor is MB is 3, and the image obtained by the reconstruction method of the present application are sequentially provided from left to right, as shown in fig. 10, it can be known by comparing the image reconstructed by the conventional method and the image error obtained by the reconstruction method of the present application, and the numerical value at the lower left corner of each image sequentially represents the PSNR value, the SSIM value, and the RMSE value, and it can be found that both the PSNR value and the SSIM value are improved, and the RMSE value is reduced, which indicates that the image quality is improved.
As shown in fig. 11, the full sampling reference image, the image reconstructed by the conventional method when the acceleration factor is MB is 4, and the image obtained by the reconstruction method of the present application are sequentially provided from left to right, as shown in fig. 12, it can be known by comparing the image reconstructed by the conventional method and the image error obtained by the reconstruction method of the present application, and the numerical value at the lower left corner of each image sequentially represents the PSNR value, the SSIM value, and the RMSE value, and it can be found that both the PSNR value and the SSIM value are improved, and the RMSE value is reduced, which indicates that the image quality is improved.
The application also discloses a computer readable storage medium, which stores a reconstruction program of the simultaneous multi-slice imaging signals, and the reconstruction program of the simultaneous multi-slice imaging signals realizes the reconstruction method of the simultaneous multi-slice imaging signals when being executed by a processor.
The present application also discloses a computer device, which comprises a processor 12, an internal bus 13, a network interface 14, and a computer-readable storage medium 11, as shown in fig. 13, on a hardware level. The processor 12 reads a corresponding computer program from the computer-readable storage medium and then runs, forming a request processing apparatus on a logical level. Of course, besides software implementation, the one or more embodiments in this specification do not exclude other implementations, such as logic devices or combinations of software and hardware, and so on, that is, the execution subject of the following processing flow is not limited to each logic unit, and may also be hardware or logic devices. The computer-readable storage medium 11 has stored thereon a reconstruction program of simultaneous multi-slice imaging signals which, when executed by a processor, implements the above-described reconstruction method of simultaneous multi-slice imaging signals.
Computer-readable storage media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
Although a few embodiments of the present invention have been shown and described, it would be appreciated by those skilled in the art that changes may be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the claims and their equivalents, and that such changes and modifications are intended to be within the scope of the invention.

Claims (9)

1. A method of simultaneous multi-slice imaging signal reconstruction, the method comprising:
generating a multi-slice virtual automatic calibration signal according to the acquired multi-slice automatic calibration signal, wherein the multi-slice automatic calibration signal and the virtual automatic calibration signal are complex data of k space;
generating a training data set according to the multi-slice automatic calibration signal and the virtual automatic calibration signal;
training the neural network model by using a training data set;
inputting the obtained original aliasing data of the multiple layers into a trained neural network model to obtain corresponding network output data;
according to the fusion of the original aliasing data and the network output data, complete reconstruction data of a k space are obtained;
a reconstructed image is generated from the complete reconstruction data.
2. The method for reconstructing simultaneous multi-slice imaging signal according to claim 1, wherein the specific method for generating the training data set according to the auto-calibration signal and the virtual auto-calibration signal of the multi-slice comprises:
and generating a first training sample according to the multi-slice automatic calibration signal, and generating a second training sample according to the multi-slice virtual automatic calibration signal, wherein the first training sample and the second training sample form two training data sets.
3. The method for reconstructing simultaneous multi-slice imaging signals according to claim 2, wherein the method for training the neural network model using the training data comprises:
training a neural network model by using a first training sample and a second training sample, wherein the first training sample comprises first input training data and first output training data, the first input training data is used for training an input part of the neural network model, and the first output training data is used for training an output part of the neural network model; the second training samples include second input training data for training an input portion of the neural network model and second output training data for training an output portion of the neural network model.
4. The method for reconstructing simultaneous multi-slice imaging signal according to claim 3, wherein the specific method for generating the first training sample according to the auto-calibration signal of the multi-slice comprises:
respectively performing inverse Fourier transform on each layer of automatic calibration signals of the k space to generate each layer of low-resolution images in an image space;
connecting the low-resolution images of the layers in the image space in a readout direction to generate a multi-layer connection image;
carrying out Fourier transform on the multilayer connection images to generate k-space data corresponding to the multilayer connection images;
normalizing k-space data corresponding to the multilayer connected image to obtain k-space data corresponding to the normalized multilayer connected image;
and performing down-sampling processing of an acceleration factor MB on the k-space data corresponding to the normalized multi-layer connection image to obtain first input training data, and taking the residual data of the k-space data corresponding to the multi-layer connection image after the down-sampling processing as first output training data, wherein the MB is equal to the number of the multi-layer sheets.
5. The method for reconstructing simultaneous multi-slice imaging signal according to claim 3, wherein the specific method for generating the second training sample according to the virtual auto-calibration signal of the multi-slice comprises:
respectively performing inverse Fourier transform on each layer of virtual automatic calibration signals of the k space to generate each layer of virtual low-resolution images in an image space;
connecting each layer of virtual low-resolution images in an image space in a reading direction to generate a multilayer connected virtual image;
carrying out Fourier transformation on the multilayer connection virtual image to generate k space data corresponding to the multilayer connection virtual image;
normalizing k-space data corresponding to the multilayer connection virtual image to obtain k-space data corresponding to the normalized multilayer connection virtual image;
and performing down-sampling processing of an acceleration factor MB on the k-space data corresponding to the multilayer connection virtual image to obtain second input training data, and taking the residual data of the k-space data corresponding to the multilayer connection virtual image after the down-sampling processing as second output training data, wherein the MB is equal to the number of slices of the multilayer connection virtual image.
6. The method for reconstructing simultaneous multi-slice imaging signals according to claim 1, wherein the specific method for obtaining the complete reconstruction data according to the fusion of the original aliasing data and the corresponding network output data comprises:
constructing a zero padding data space in which the multiple slices are connected in the reading direction;
interpolating original aliased data in k-space into the zero-padded data space with MB times upsampling in a readout direction;
and interpolating the network output data corresponding to the original aliasing data to the residual position of the zero padding data space to obtain complete reconstruction data connected in the reading direction.
7. The method for reconstructing simultaneous multi-slice imaging signals according to claim 6, wherein the specific method for generating the reconstructed image from the complete reconstruction data comprises:
performing inverse Fourier transform on the complete reconstruction data connected in the readout direction to generate a complete multi-slice image connected in the readout direction;
cutting the complete multi-slice image connected in the readout direction into MB slices to obtain reconstructed MB images;
and fusing the multi-coil data of each image by using a coil fusion square sum method to obtain MB layer image data.
8. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a reconstruction program of simultaneous multi-slice imaging signals, which when executed by a processor implements the reconstruction method of simultaneous multi-slice imaging signals of any one of claims 1 to 7.
9. A computer device, characterized in that the computer device comprises a computer readable storage medium, a processor and a reconstruction program of simultaneous multi-slice imaging signals stored in the computer readable storage medium, which when executed by the processor implements the reconstruction method of simultaneous multi-slice imaging signals of any one of claims 1 to 7.
CN202010239184.2A 2020-03-30 2020-03-30 Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device Pending CN113470128A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010239184.2A CN113470128A (en) 2020-03-30 2020-03-30 Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010239184.2A CN113470128A (en) 2020-03-30 2020-03-30 Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device

Publications (1)

Publication Number Publication Date
CN113470128A true CN113470128A (en) 2021-10-01

Family

ID=77865061

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010239184.2A Pending CN113470128A (en) 2020-03-30 2020-03-30 Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device

Country Status (1)

Country Link
CN (1) CN113470128A (en)

Similar Documents

Publication Publication Date Title
US10489943B2 (en) System and method for sparse image reconstruction
Chang et al. Nonlinear GRAPPA: A kernel approach to parallel MRI reconstruction
Samsonov et al. POCSENSE: POCS‐based reconstruction for sensitivity encoded magnetic resonance imaging
US10274567B2 (en) Magnetic resonance imaging apparatus and image processing apparatus
CN106772167B (en) Magnetic resonance imaging method employing and device
CN107205686B (en) Method for removing ghost artifacts in magnetic resonance imaging and magnetic resonance device therefor
CN103142228A (en) Compressed sensing magnetic resonance fast imaging method
CN110133556A (en) A kind of magnetic resonance image processing method, device, equipment and storage medium
WO2022236851A1 (en) Method for reconstructing wrap-around field-of-view magnetic resonance image, computer device, and storage medium
Samsonov et al. Advances in locally constrained k‐space‐based parallel MRI
Chen et al. An improved non‐Cartesian partially parallel imaging by exploiting artificial sparsity
CN105678822A (en) Three-regular magnetic resonance image reconstruction method based on Split Bregman iteration
CN113298901B (en) Method for reconstructing magnetic resonance image in convoluted field of view, computer device and storage medium
CN113030813B (en) Magnetic resonance T2Quantitative imaging method and system
CN107942271B (en) SPEED rapid magnetic resonance imaging method based on iteration
CN112329920A (en) Unsupervised training method and unsupervised training device for magnetic resonance parameter imaging model
Dawood et al. Iterative training of robust k‐space interpolation networks for improved image reconstruction with limited scan specific training samples
Jin et al. Improving accelerated MRI by deep learning with sparsified complex data
Park et al. Strategies for rapid reconstruction in 3D MRI with radial data acquisition: 3D fast Fourier transform vs two-step 2D filtered back-projection
KR102163337B1 (en) Method for accelerating multiple-acquisition magnetic resonance imaging by varying undersampling-dimension and device for the same
Ryu et al. Accelerated multicontrast reconstruction for synthetic MRI using joint parallel imaging and variable splitting networks
CN113470128A (en) Method for reconstructing simultaneous multi-slice imaging signals, storage medium and computer device
CN108333545B (en) Magnetic resonance image reconstruction method based on high-pass filtering
Wang et al. 2D partially parallel imaging with k‐space surrounding neighbors‐based data reconstruction
KR102163220B1 (en) Method and apparatus for processing MR angiography image using neural network

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