Multi-parametric simultaneous quantitative imaging method and system based on multiple overlapping echoes

Abstract

A multi-parameter simultaneous quantitative imaging method and system based on multiple overlapping echoes, relating to magnetic resonance imaging. Method: Design of a multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes, including T1, T2, and T3 echoes. * Simultaneous quantitative imaging sequence with multiple parameters; determining the T1, T2, and T3 parameters based on multiple overlapping echoes. * The method involves: setting sampling parameters for a multi-parameter quantitative imaging sequence; adding the sequence and setting the sampling parameters in an MRI scanner that meets the performance requirements of the sequence and sampling parameters; acquiring k-space data; generating training samples for a deep neural network; training the deep neural network using the training samples to obtain a trained deep neural network; and reconstructing a multi-parameter quantitative MRI image using the trained deep neural network and the k-space data acquired by the MRI sequence. This method enables simultaneous multi-parameter quantitative MRI imaging, utilizing the complementary information from different echoes and / or different k-spaces within k-space to correct for imperfections.

Description

Technical Field

[0001] This invention relates to the field of magnetic resonance imaging, and in particular to a multi-parameter simultaneous quantitative imaging method and system based on multiple overlapping echoes. Background Technology

[0002] Magnetic resonance imaging (MRI) is an imaging technique that utilizes the signals generated by the resonance of atomic nuclei within a strong magnetic field to reconstruct images. MRI can be divided into qualitative parametric-weighted imaging and quantitative parametric imaging. Quantitative parametric MRI can directly quantify the characteristic physiological parameters of tissues, which is of great significance for evaluating treatment efficacy and classifying diseases, and it currently has wide clinical applications. [1,2] The most significant advantage of quantitative parametric imaging is that it can reduce the influence of factors other than tissue properties on imaging results, such as operator dependence, differences in scanning parameters, spatial variations in magnetic fields, and image scaling. [3] Therefore, it has important application value.

[0003] Magnetic resonance multiparameter quantitative imaging provides a more comprehensive understanding of tissue characteristics for quantitative research, and is beneficial in medical diagnostics. [4] It shows great potential in scientific research. However, the long acquisition time required for this process still limits its widespread application, especially for functional / dynamic imaging or motion scenarios. Currently, a series of T2 fast quantitative magnetic resonance imaging methods based on overlapping echoes and employing EPI readout have been developed. [5,6] However, it can only quantitatively measure one type of magnetic resonance parameter and does not take into account the influence of non-uniform magnetic fields on quantitative imaging of parameters. Therefore, the images obtained under non-uniform magnetic fields will be distorted, which increases the difficulty of reconstructing the quantitative magnetic resonance parameter map.

[0004] Multiparameter quantitative magnetic resonance imaging based on multiple overlapping echoes [7] It can non-invasively characterize a variety of tissue properties and shows great potential in improving the sensitivity of MRI measurements. The multi-parameter quantitative magnetic resonance imaging method based on multiple overlapping echoes enables simultaneous quantification of T2 and T2. * While multiple parameters can be simultaneously and quantitatively imaged, quantitative measurement of the T1 parameter has not yet been achieved.

[0005] References:

[0006] 1.Zhao B,Lam F,Liang Z P.Model-Based MR Parameter Mapping WithSparsity Constraints:Parameter Estimation and Performance Bounds[J].IEEETransactions on Medical Imaging,2014,33(9):1832-1844.

[0007] 2.[1]B S B A,Lucas Nürnberger a,Udo Rüb c,et al.Quantitative mappingof T1 and T2*discloses nigral and brainstem pathology in early Parkinson'sdisease-ScienceDirect[J].NeuroImage,2010,51(2):512-520.

[0008] 3.MargaretCheng,H.-L,Stikov,et al.Practical medical applications ofquantitative MR relaxometry[J].Journal of Magnetic Resonance Imaging Jmri,2012,36:805-824.

[0009] 4.Wang L,Gaddam S,Wang N,et al.Multiparametric Mapping MagneticResonance Imaging of Pancreatic Disease[J].Frontiers in Physiology,2020,11.

[0010] 5.Cai C,Wang C,Zeng Y,et al.Single-shot T2 mapping using overlapping-echo detachment planar imaging and a deep convolutional neural network[J].Magnetic Resonance in Medicine,2018,80(5).

[0011] 6. Zhang J, Wu J, Chen S, et al. Robust Single-shot T 2Mapping via Multiple Overlapping-Echo Acquisition and Deep Neural Network[J]. IEEE transactions on medical imaging, 2019:1801.

[0012] 7.Ma L,Wu J,YQ,et al.Single-shot multi-parametric mapping based on multiple overlapping-echo detachment(MOLED)imaging[J].NeuroImage,2022,263. Summary of the Invention

[0013] The purpose of this invention is to provide a method and system for simultaneous quantitative imaging of multiple parameters in magnetic resonance imaging (MRI) capable of achieving quantitative imaging of multiple parameters and correcting for imperfections, including but not limited to distortion, artifacts, radio frequency field inhomogeneity, and main magnetic field inhomogeneity. This method includes T1, T2, and T3. * And many other parameters.

[0014] To achieve the above objectives, the present invention provides the following solution:

[0015] This invention provides a method based on multiple overlapping echoes including T1, T2, and T3. * A method for simultaneous quantitative imaging of multiple parameters includes the following steps:

[0016] S1: Design based on multiple overlapping echoes including T1, T2, T3 * Simultaneous quantitative imaging sequence with multiple parameters;

[0017] S2: Determine the components T1, T2, and T2 based on multiple overlapping echoes. * Sampling parameters for simultaneous quantitative imaging of multiple parameters;

[0018] S3: Add the sequence and set the sampling parameters in an advanced magnetic resonance spectrometer that meets the sequence and sampling parameters, and complete the data acquisition to obtain magnetic resonance data;

[0019] S4: Generate training samples for the deep neural network based on the sequences and parameters described in steps S1-S3;

[0020] S5: The deep neural network is trained using the training samples to obtain a trained deep neural network;

[0021] S6: Using the trained deep neural network and the data collected in step S3, reconstruct a multi-parameter quantitative magnetic resonance image.

[0022] The present invention also provides a method based on multiple overlapping echoes including T1, T2, and T3. * A multi-parameter simultaneous quantitative imaging system, including:

[0023] The pulse sequence design module is used to design pulse sequences based on multiple overlapping echoes, including T1, T2, and T3. * Simultaneous quantitative imaging sequence with multiple parameters; determining the T1, T2, and T3 parameters based on multiple overlapping echoes. * Sampling parameters of a multi-parameter simultaneous quantitative imaging sequence;

[0024] The signal acquisition module is used to sample the test object in a high-performance magnetic resonance spectrometer using a pulse sequence under set sampling parameters to obtain the overlapping echo magnetic resonance signal of the test object.

[0025] The signal processing module is used to perform two-dimensional Fourier transform, normalization and other processing on the overlapping echo magnetic resonance signal of the object under test to obtain the magnetic resonance image to be reconstructed of the object under test.

[0026] The training sample set generation module is used to simulate and generate a set number of deep neural network training samples based on the characteristics of the object under test, thus forming the training sample set.

[0027] A deep neural network determination module is used to determine a deep neural network for multi-parameter quantitative image reconstruction; the deep neural network is trained using the training sample set to obtain a trained deep neural network;

[0028] The multi-parameter quantitative image reconstruction module is used to input the overlapping echo magnetic resonance image obtained from the magnetic resonance instrument into the trained deep neural network for reconstruction, thereby obtaining a multi-parameter quantitative image of the analyte.

[0029] This invention designs a method based on multiple overlapping echoes including T1, T2, and T3. *A method and system for simultaneous quantitative imaging of multiple parameters. Magnetic resonance imaging (mqMRI) based on multiple overlapping echoes can rapidly acquire quantitative information, thus exhibiting strong resistance to motion changes. Rapid quantification of multiple parameters can promote the application of mqMRI in treatment and metabolic / functional process analysis. This invention can simultaneously acquire multiple magnetic resonance k-space data. Different echoes within a single k-space or between different k-spaces have different weights for the quantitative magnetic resonance parameters. Therefore, this invention can achieve simultaneous quantitative imaging of multiple magnetic resonance parameters and can utilize the complementary information from different echoes within k-spaces and / or different k-spaces to correct for imperfections, including but not limited to artifacts, radiofrequency field inhomogeneities, and main magnetic field inhomogeneities.

[0030] The present invention has the following outstanding technical effects:

[0031] 1. This invention achieves fast magnetic resonance parameters T1, M0, T2, T2 * Simultaneous quantitative imaging of B1 and ΔB0;

[0032] 2. The magnetic resonance quantitative parameter images obtained by this invention have high accuracy;

[0033] 3. The present invention has a high signal-to-noise ratio and can correct image distortion and other undesirable factors;

[0034] 4. This invention can be applied to different parts and organs of the human body and small animals, and has good stability. Attached Figure Description

[0035] Figure 1 This is a flowchart of the magnetic resonance multi-parameter simultaneous quantitative imaging method of the present invention;

[0036] Figure 2 This invention is based on multiple overlapping echoes including T1, T2, and T3. * Simultaneous quantitative imaging sequence with multiple parameters;

[0037] Figure 3 This is a flowchart of the magnetic resonance multi-parameter simultaneous quantitative imaging system of the present invention;

[0038] Figure 4 This is a magnetic resonance imaging sequence diagram from an embodiment of the present invention;

[0039] Figure 5 The k-space data and magnetic resonance images of the simulated samples in this embodiment of the invention;

[0040] Figure 6 This is a multi-parameter quantitative magnetic resonance image of a simulated sample in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention, including but not limited to: adding common preparation modules such as reversal, saturation, and compression before using the sequence and during the intermediate process of cyclic use of the sequence. These should all be considered as not involving significant creative effort based on the present invention, and are therefore within the scope of protection of this patent.

[0042] Figure 1 This is a flowchart of the magnetic resonance multi-parameter simultaneous quantitative imaging method of the present invention. Figure 1 As shown, a method based on multiple overlapping echoes including T1, T2, and T3 is described. * Simultaneous quantitative imaging methods with multiple parameters include:

[0043] S1: Design based on multiple overlapping echoes including T1, T2, T2 * Simultaneous quantitative imaging sequence with multiple parameters;

[0044] S2: Determine the components T1, T2, and T2 based on multiple overlapping echoes. * Sampling parameters of a multi-parameter simultaneous quantitative imaging sequence;

[0045] S3: Add the sequence and set the sampling parameters in an advanced magnetic resonance spectrometer that meets the sequence and sampling parameters, and complete the data acquisition to obtain magnetic resonance data;

[0046] The magnetic resonance imaging sequence is used to acquire data from the actual imaging object to obtain the k-space data of the actual imaging object;

[0047] S4: Generate training samples for the deep neural network based on the sequences and parameters described in S1-S3;

[0048] S5: Use the training samples to train the deep neural network to obtain a trained deep neural network;

[0049] S6: Using the trained deep neural network and the k-space data (or amplitude and / or phase) acquired by the magnetic resonance imaging sequence, reconstruct a multi-parameter quantitative image of magnetic resonance.

[0050] like Figure 2 As shown, this invention is based on multiple overlapping echoes including T1, T2, and T3. * Simultaneous quantitative imaging sequence diagram of multiple parameters, Figure 2 (a) is a schematic diagram of the overall sequence, where the large and small trapezoids represent the dual-echo-chain acquisition module and the single-echo-chain acquisition module, respectively. Figure 2 (b) Figure 2 (c) are the specific component diagrams of the dual-echo chain acquisition module and the single-echo chain acquisition module, respectively.

[0051] Where, ΔT i This represents the interval time, which is the time interval between the i-th acquisition module and the (i+1)-th acquisition module. In particular, the number of acquisition modules can be set as needed, and different interval times can be selected.

[0052] Preferably, considering signal attenuation, dual-echo chains are acquired in the early acquisition modules, while only single-echo chains are acquired in the later acquisition modules s. i = 1, 2, ..., n, where n is a positive integer, and a total of n acquisition modules are applied, of which m are dual-echo chain acquisitions and nm are single-echo chain acquisitions;

[0053] β i This represents the i-th refocusing pulse, and there are a total of n refocusing pulses; α j α represents the flip angle of the j-th radio frequency excitation pulse, and in particular, α j Different sizes, phases, and pulse shapes can be selected. δ j Represents α j With α j+1 The time interval between;

[0054] G ROj Indicates the application of α j The subsequent shift gradient on the frequency encoding dimension, G PEj Indicates the application of α j The shift gradients on the subsequent phase encoding dimension, j = 1, 2, ..., k, where k is a positive integer, are applied with a total of k radio frequency excitation pulses, resulting in k frequency shift gradients and k phase shift gradients respectively;

[0055] G sp This represents the destructive gradient applied at the end of each acquisition module to ensure that the residual transverse magnetization vector is completely dephased after each acquisition, so as not to interfere with the next acquisition.

[0056] G SS The layer / segment gradient of the layer selection dimension is represented by the application of G simultaneously with the application of the radio frequency excitation pulse. ss .

[0057] This sequence is divided into two parts: dual-echo-chain and single-echo-chain acquisition. In the dual-echo-chain part, the first echo chain generates spin echoes through refocusing pulses, and the second echo chain generates echoes through gradient inversion. In the single-echo-chain part, all spin echoes are generated using refocusing pulses, and then read out using EPI via a readout module. The signal obtained from the spin echoes is predominantly T2-weighted, while the signal obtained from the gradient echoes is predominantly T2-weighted. * Weighted average is the primary factor.

[0058] The readout module can employ gradients of equal magnitude but opposite direction in both frequency encoding and phase encoding directions as needed. Due to the presence of inhomogeneous magnetic fields, the images read out by the EPI are distorted. Therefore, in this invention, preferably, in the dual-echo chain acquisition of m modules (typically m>=2), at least two sets of acquisition modules have their second echo chain's phase encoding gradients acquired in opposite directions, resulting in two images with opposite distortions in the phase encoding direction. Subsequently, using the phase information contained in the deep neural network and K-space, as well as the amplitude map information of the opposite distortions, the image distortion can be corrected.

[0059] For each module, when using dual-echo chain acquisition, the data will be filled into two k-spaces respectively, while when using single-echo chain acquisition, the data will be filled into one k-space.

[0060] In this invention, the radio frequency pulse is taken at a small angle (usually less than 90°). After the radio frequency pulse is excited, the transverse magnetization vector M is obtained. ⊥ and a larger longitudinal magnetization vector M z By acquiring data from various modules at different intervals, the acquired images record information with different T1 weights, thus enabling quantitative measurement of T1. It is worth noting that, due to the accurate estimation of B1 and ΔB0, the obtained quantitative T1 parameters possess high reliability and stability.

[0061] By applying a shift gradient after the radio frequency excitation pulse and after the readout gradient, several overlapping echoes are formed in each k-space. The distribution of each overlapping echo in the k-space is determined by the magnitude and direction of each shift gradient. Different echoes in each k-space have different weights for the quantitative magnetic resonance parameters. By utilizing the echo information in each k-space, this invention can achieve simultaneous quantitative imaging of multiple magnetic resonance parameters.

[0062] Determine the sampling parameters of the overlapping echo multi-parameter quantitative imaging sequence, specifically including:

[0063] Determine the number of acquisition modules n and the number of dual-echo chain acquisitions m;

[0064] Determine the time interval ΔT of the acquisition module i ;

[0065] Determine the number k of radio frequency excitation pulses in each acquisition module;

[0066] Determine the flip angle α of each radio frequency excitation pulse. j The magnitude, phase, and pulse shape;

[0067] Determine the time interval δ of each radio frequency excitation pulse. j ;

[0068] Determine the level selection dimension G SS Area and orientation;

[0069] Destroy gradient G after determining each acquisition module sp The area;

[0070] Determine each shift gradient G ROi G PEi The area, direction, and application time are used to determine the number of overlapping echoes in each k-space and the position of each overlapping echo in the k-space;

[0071] Determine other acquisition parameters such as imaging field of view, imaging matrix, and parallel imaging acceleration factor;

[0072] Before determining the deep neural network, this invention considers that deep neural networks require a large number of training samples, but real training samples are difficult to obtain. Therefore, this invention uses simulation methods to generate the training samples required for deep neural networks. The specific steps are as follows:

[0073] The pulse sequence is input into the magnetic resonance imaging simulation software, and corresponding non-ideal terms are added according to the non-idealities of the real experiment in order to simulate the real situation as much as possible.

[0074] Generate a set number of random templates based on the characteristics of the test object (or generate templates using an MRI dataset with the characteristics of the test object); the complexity of the templates should match that of the test object and should be able to include all the characteristics of the experimental samples; the number of templates needs to be sufficient to ensure good reconstruction quality.

[0075] The template was simulated and sampled using simulation software to obtain the template's magnetic resonance signal. During the simulation sampling process, considering the changes in the real experimental environment, unstable factors were added to improve the robustness of the network model to the unsatisfactory experimental environment. The unstable factors included excitation pulse angle deviation, displacement gradient deviation, non-uniform field, and noise.

[0076] The magnetic resonance echo signals of the templates are rearranged into two-dimensional k-space signals, and then two-dimensional Fourier transform is performed to obtain the magnetic resonance images of each template. The magnetic resonance images are then normalized and noise-added.

[0077] Based on the characteristics of different parameters, the corresponding quantitative parameters in the template are normalized so that they are distributed within the optimal range for network learning, and corresponding labels are created.

[0078] The normalized magnetic resonance images of each template and their corresponding labels constitute a training sample. All simulation training samples are divided proportionally to form a simulation sample training set and a validation set.

[0079] It is worth noting that, in this invention, the training samples can be composed entirely of simulated samples containing all features of real experimental data, or entirely of samples made from real data, or a mixture of both.

[0080] The deep neural network is trained using the training samples to obtain a trained deep neural network, specifically including:

[0081] Determine the network structure of the deep neural network;

[0082] Determine the number of input and output channels for the deep neural network;

[0083] Determine the loss function and optimizer for training the deep neural network.

[0084] When training a deep neural network, the training sample set is input into the deep neural network in batches for iterative training. Each time the network is trained, the value of the loss function is calculated. Based on this value, the parameter values ​​of the neural network are automatically adjusted to reduce the value of the loss function. The above training is repeated until the value of the loss function no longer decreases, and then the parameters of the deep neural network are saved.

[0085] Using a trained deep neural network and k-space data acquired from the magnetic resonance imaging sequence, a multi-parameter magnetic resonance image is reconstructed, specifically including:

[0086] Data was acquired from the actual imaging object using magnetic resonance imaging sequences to obtain the k-space data of the actual imaging object;

[0087] A Fourier transform is performed on the k-space data of the actual imaging object to obtain the magnetic resonance image of the actual imaging object.

[0088] The magnetic resonance image of the actual imaging object is reconstructed using a trained deep neural network to obtain a quantitative image of magnetic resonance multi-parameters.

[0089] Figure 3 This is a flowchart of the magnetic resonance multi-parameter simultaneous quantitative imaging system of the present invention. Figure 3 As shown, a method based on multiple overlapping echoes including T1, T2, and T3 is described. * Simultaneous quantitative imaging systems with multiple parameters include:

[0090] The pulse sequence design module is used to design pulse sequences based on multiple overlapping echoes, including T1, T2, and T3. * Simultaneous quantitative imaging sequence with multiple parameters; determining the T1, T2, and T3 parameters based on multiple overlapping echoes. * Sampling parameters of a multi-parameter simultaneous quantitative imaging sequence;

[0091] The signal acquisition module is used to sample the test object in a high-performance magnetic resonance spectrometer using the pulse sequence under set sampling parameters to obtain the overlapping echo magnetic resonance signal of the test object.

[0092] The signal processing module is used to perform two-dimensional Fourier transform, normalization and other processing on the overlapping echo magnetic resonance signal of the object under test to obtain the magnetic resonance image to be reconstructed of the object under test.

[0093] The training sample set generation module is used to simulate and generate a set amount of deep neural network training samples based on the characteristics of the object under test, which constitute the training sample set or a part of the training set.

[0094] A deep neural network determination module is used to determine a deep neural network for multi-parameter quantitative image reconstruction; the deep neural network is trained using the training sample set to obtain a trained deep neural network;

[0095] The multi-parameter quantitative image reconstruction module is used to input the overlapping echo magnetic resonance image obtained from the magnetic resonance instrument into the trained deep neural network for reconstruction, thereby obtaining a multi-parameter quantitative image of the analyte.

[0096] Specific embodiments are given below.

[0097] This embodiment includes the following steps:

[0098] Step 1: Design a multi-overlap echo pattern including T1, T2, and T3. * Simultaneous quantitative analysis of multiple parameters in imaging sequences and determination of sequence sampling parameters. The design of magnetic resonance imaging sequences can refer to... Figure 4 , Figure 4 Figure (a) in the middle is a schematic diagram of the overall sequence. The large and small trapezoids in the figure represent the dual-echo-chain acquisition module and the single-echo-chain acquisition module, respectively. Figure 4 Figures (b) and (c) show the specific components of the dual-echo-chain acquisition module and the single-echo-chain acquisition module, respectively.

[0099] The entire sequence comprises six acquisition modules. The first two modules use a dual-echo train, while the last four modules use a single-echo train. The interval ΔT between the acquisition modules is... i (i=1,2,3,4,5) are 1750ms, 1410ms, 995ms, 745ms, and 495ms respectively.

[0100] Excitation pulse α j All pulses are Sinc pulses with α = 30°, and the refocusing pulse β i All pulses are sinc pulses with β = 180°.

[0101] Excitation pulse interval δ j The total time is set to δ1 = 10.0863 ms.

[0102] The field of view (FOV) is 0.22m x 0.22m.

[0103] The imaging matrix size is 128*126.

[0104] Assume the readout gradient area in the frequency encoding direction during sampling is 1 unit; assume the area in the phase encoding direction during sampling is 1 unit. Shift gradients are applied in the frequency encoding and phase encoding dimensions after each RF excitation pulse, with the relative areas of each shift gradient being: (GRO1, GPE1) = (-22 / 128, 32 / 128), (GRO2, GPE2) = (11 / 128, 32 / 128), (GRO3, GPE3) = (-22 / 128, 32 / 128), (GRO4, GPE4) = (-48 / 128, 16 / 128).

[0105] G sp This represents the destructive gradient applied at the end of each acquisition module.

[0106] G SS The layer / segment gradient of the layer selection dimension is represented by the application of G simultaneously with the application of the radio frequency excitation pulse. SS .

[0107] Step 2: Generate training samples for the deep neural network;

[0108] The pulse sequence is input into the magnetic resonance imaging simulation software, and corresponding non-ideal terms are added according to the non-ideal nature of the real experiment in order to simulate the real situation as much as possible.

[0109] A set number of random templates are generated based on the characteristics of the test object; the complexity of the templates is matched with that of the test object and should be able to contain all the characteristics of the experimental sample; the number of templates is sufficient to ensure good reconstruction quality.

[0110] The template was simulated and sampled using simulation software to obtain the template's magnetic resonance signal. During the simulation sampling process, considering the changes in the real experimental environment, unstable factors were added to improve the robustness of the network model to the unsatisfactory experimental environment. The unstable factors included excitation pulse angle deviation, shift gradient deviation, and noise.

[0111] The magnetic resonance echo signals of the templates are rearranged into two-dimensional k-space signals, and then two-dimensional Fourier transform is performed to obtain the magnetic resonance images of each template. The magnetic resonance images are then normalized and noise-added.

[0112] Based on the characteristics of different parameters, the corresponding quantitative parameters in the template are normalized so that they are distributed within the optimal range for network learning, and corresponding labels are created.

[0113] The normalized magnetic resonance images of each template and their corresponding labels constitute a training sample. All simulation training samples are divided proportionally to form a simulation sample training set and a validation set.

[0114] Step 3: Use the training samples to train the deep neural network to obtain a trained deep neural network;

[0115] The network structure of the deep neural network is determined. Since U-Net has good performance in nonlinear fitting and feature extraction for image-based tasks, this embodiment selects the U-Net deep neural network, but other deep neural network structures can also be selected.

[0116] Determine the number of input and output channels for the deep neural network; such as Figure 4 As shown, this embodiment acquires a total of 8 k-space data points, corresponding to 8 magnetic resonance images. The real and imaginary parts of each magnetic resonance image are used as inputs to a deep neural network, therefore the deep neural network has 16 input channels. This embodiment reconstructs M0, T1, T2, and T2... * The graphs of the six quantitative magnetic resonance parameters, B1, and ΔB0, show that the deep neural network has six output channels.

[0117] Determine the loss function for training the deep neural network. Use the pixel-wise mean squared error as the loss function, as described below:

[0118]

[0119] Where n(·) is the output of the deep neural network, M represents the batch size, Xi represents the network input data in the i-th training sample, and Yi represents the network label data in the i-th training sample. W and b represent the parameters of the deep neural network.

[0120] Using the Adam optimizer, the learning rate is adjusted exponentially: the initial learning rate is set to 0.00002, and the learning rate is set to decrease once every 20,000 iterations, with a decay rate of 0.8.

[0121] The training sample set is randomly divided into two parts: 80% for training and 20% for validation.

[0122] When training a deep neural network, the training sample set is input into the deep neural network in batches for iterative training. Each time the network is trained, the value of the loss function is calculated. Based on this value, the parameter values ​​of the neural network are automatically adjusted to reduce the value of the loss function. The above training is repeated until the value of the loss function tends to stabilize, and the parameters W and b of the deep neural network are saved.

[0123] Step 4: Reconstruct the multi-parameter magnetic resonance image using the trained deep neural network and the k-space data acquired by the magnetic resonance imaging sequence.

[0124] The magnetic resonance imaging sequence is used to acquire data of the imaging object to obtain the k-space data of the imaging object;

[0125] A two-dimensional Fourier transform is performed on the k-space data of the imaging object to obtain the magnetic resonance image of the imaging object.

[0126] Figure 5 The data includes k-space data and magnetic resonance images of a human brain simulation sample; each k-space contains several overlapping echoes, the positions of which are determined by... Figure 4 The shift gradients in the magnetic resonance imaging sequence shown are determined by these gradients. Figure 5 As shown, the magnetic resonance image corresponding to the k-space containing overlapping echoes is a striped aliased image.

[0127] The magnetic resonance image of the simulated human brain sample is input into the deep neural network trained in step 3 for reconstruction, resulting in... Figure 6 The image shown is a multi-parameter quantitative image from magnetic resonance imaging. Figure 6 The simulated human brain sample is shown at M0, T1, T2, and T2. * Images of six quantitative magnetic resonance imaging (MRI) parameters (B1, ΔB0, ΔB0, etc.) are shown. The first row of images represents the MRI quantitative parameter images reconstructed by the deep neural network; the second row shows the corresponding label images; and the third row shows the absolute error values ​​between the deep neural network reconstruction results and the corresponding label images. Specifically, the absolute error value image for B1 is magnified 10 times. The error graphs show that the MRI quantitative parameter images reconstructed by the deep neural network have accurate quantitative values ​​and high imaging precision.

[0128] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes, characterized in that... The multiple parameters include T1, T2, The method includes the following steps: S1: Design of a multi-parameter simultaneous quantitative imaging sequence based on multiple overlapping echoes; S2: Determine the sampling parameters of the multi-parameter simultaneous quantitative imaging sequence based on multiple overlapping echoes; S3: Add the sequence and set the sampling parameters in an advanced magnetic resonance spectrometer that meets the sequence and sampling parameters, and acquire K-space data; The acquired K-space data sequence is divided into two parts: dual-echo-chain acquisition and single-echo-chain acquisition. In the dual-echo-chain part, the first echo chain generates spin echoes through refocusing pulses, and the second echo chain generates echoes through gradient inversion. In the single-echo-chain part, all spin echoes are generated using refocusing pulses, and then read out using EPI by the readout module. The signal obtained from the spin echoes is predominantly T2-weighted, while the signal obtained from the gradient echoes is... Weighted average is the primary method; considering signal attenuation, dual echo chains are acquired in the early acquisition modules, while only single echo chains are acquired in the later acquisition modules. In the dual-echo chain acquisition of m modules, m>=2, the phase encoding gradients of the second echo chain of at least two acquisition modules can be acquired in opposite directions, so that the two acquired images will produce opposite distortions in the phase encoding direction; the image distortion is corrected by using the phase information contained in the deep neural network and K space, as well as the amplitude map information of opposite distortions. For each module, during dual-echo chain acquisition, the data is filled into two k-spaces respectively, while during single-echo chain acquisition, the data is filled into one k-space. S4: Generate training samples for the deep neural network based on the sequences and parameters described in steps S1-S3; S5: Train the deep neural network using training samples to obtain a trained deep neural network; S6: Using the trained deep neural network and the K-space data collected in step S3, reconstruct the multi-parameter quantitative image of magnetic resonance.

2. The multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes as described in claim 1, characterized in that... In step S2, determining the sampling parameters of the multi-parameter simultaneous quantitative imaging sequence based on multiple overlapping echoes specifically includes: Determine the number of acquisition modules n and the number of dual-echo chain acquisitions m; Determine the time interval ΔT of the acquisition module i ; Determine the number k of radio frequency excitation pulses in each acquisition module; Determine the flip angle α of each radio frequency excitation pulse. j The magnitude, phase, and pulse shape; Determine the time interval δ of each radio frequency excitation pulse. j ; Determine the level selection dimension G SS Area and orientation; Destroy gradient G after determining each acquisition module sp The area; Determine each shift gradient G ROi G PEi The area, direction, and application time are used to determine the number of overlapping echoes in each k-space and the position of each overlapping echo in the k-space; Determine other acquisition parameters: imaging field of view, imaging matrix, and parallel imaging acceleration factor.

3. The multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes as described in claim 2, characterized in that... The radio frequency excitation pulse has a small angle, less than 90°. After the radio frequency pulse excitation, the transverse magnetization vector M is obtained. ⊥ and a larger longitudinal magnetization vector M z By collecting data from each module at different intervals, the collected images record information with different T1 weights, thus enabling quantitative measurement of T1.

4. The multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes as described in claim 1, characterized in that... In step S4, the training samples are composed of at least one of the following: simulation samples containing all features of real experimental data and real samples made from real data.

5. The multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes as described in claim 4, characterized in that... The simulated samples are training samples needed to generate a deep neural network using simulation methods. The specific steps are as follows: The sequence is input into the magnetic resonance imaging simulation software, and corresponding non-ideal terms are added based on the non-idealities of the real experiment; A set amount of random templates are generated based on the characteristics of the test analyte, or templates are generated using an MRI dataset with the characteristics of the test analyte; the complexity of the templates is matched to the test analyte and should be able to contain all the characteristics of the experimental samples. The template was simulated and sampled using simulation software to obtain the template's magnetic resonance signal. During the simulation sampling process, considering changes in the real experimental environment, unstable factors were added to improve the robustness of the network model to imperfect experimental environments. The unstable factors included excitation pulse angle deviation, displacement gradient deviation, non-uniform field, and noise. The magnetic resonance echo signals of the templates are rearranged into two-dimensional k-space signals, and then two-dimensional Fourier transform is performed to obtain the magnetic resonance images of each template. The magnetic resonance images are then normalized and noise-added. Based on the characteristics of different parameters, the corresponding quantitative parameters in the template are normalized so that they are distributed within the optimal range for network learning, and corresponding labels are created. The normalized magnetic resonance images of each template and their corresponding labels constitute a training sample. All simulation training samples are divided proportionally to form a simulation sample training set and a validation set.

6. The multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes as described in claim 1, characterized in that... In step S5, the deep neural network is trained using training samples, and the specific steps include: Determine the network structure of the deep neural network; Determine the number of input and output channels for the deep neural network; Determine the loss function and optimizer for training the deep neural network; When training a deep neural network, the training sample set is input into the deep neural network in batches for iterative training. Each time the network is trained, the value of the loss function is calculated. Based on this value, the parameter values ​​of the neural network are automatically adjusted to reduce the value of the loss function. The above training is repeated until the value of the loss function no longer decreases, and then the parameters of the deep neural network are saved.

7. The multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes as described in claim 1, characterized in that... In step S6, the trained deep neural network and the K-space data acquired in step S3 are used to reconstruct a multi-parameter quantitative magnetic resonance image. Specific steps include: the following: Data was acquired from the actual imaging object using magnetic resonance imaging sequences to obtain the k-space data of the actual imaging object; A Fourier transform is performed on the k-space data of the actual imaging object to obtain the magnetic resonance image of the actual imaging object. The magnetic resonance image of the actual imaging object is reconstructed using a trained deep neural network to obtain a quantitative image of magnetic resonance multi-parameters.

8. The multi-parameter simultaneous quantitative imaging method based on multiple overlapping echoes as described in claim 1, characterized in that... In step S6, the reconstructed multi-parameter quantitative magnetic resonance image is formed by applying a shift gradient after the radio frequency excitation pulse and after the readout gradient. Several overlapping echoes are formed in each k-space. The distribution of each overlapping echo in the k-space is determined by the magnitude and direction of each shift gradient. Different echoes in each k-space have different weights of the magnetic resonance quantitative parameters. By utilizing the echo information in each k-space, simultaneous quantitative imaging of multiple magnetic resonance parameters is achieved.

9. A multi-parameter simultaneous quantitative imaging system based on multiple overlapping echoes, characterized in that... include: The pulse sequence design module is used to design pulse sequences based on multiple overlapping echoes, including T1, T2, and... Simultaneous quantitative imaging sequence of multiple parameters; determining the T1, T2, and T3 parameters based on multiple overlapping echoes. Sampling parameters of multi-parameter simultaneous quantitative imaging sequences; The signal acquisition module is used to sample the test object in a high-performance magnetic resonance spectrometer using the pulse sequence under set sampling parameters to obtain the overlapping echo magnetic resonance signal of the test object. The signal processing module is used to perform two-dimensional Fourier transform and normalization processing on the overlapping echo magnetic resonance signal of the object under test to obtain the magnetic resonance image to be reconstructed of the object under test. The training sample set generation module is used to simulate and generate a set number of deep neural network training samples based on the characteristics of the object under test, thus forming the training sample set. A deep neural network determination module is used to determine the deep neural network for multi-parameter quantitative image reconstruction. The deep neural network is trained using the training sample set to obtain a trained deep neural network; The multi-parameter quantitative image reconstruction module is used to input the overlapping echo magnetic resonance image obtained from the magnetic resonance instrument into the trained deep neural network for reconstruction, thereby obtaining a multi-parameter quantitative image of the analyte.

Images