A method and system for magnetic resonance liquid attenuated T2 quantification imaging

Abstract

The application relates to a magnetic resonance liquid attenuation T2 quantitative imaging method and system, and relates to magnetic resonance imaging. The application aims to quickly obtain a FLA-T2 image without cerebrospinal fluid signal interference. The method adopts a single-scan inversion recovery multi-echo chain multi-overlapping echo sequence IR-METMOLED, sets sampling parameters, collects IR-METMOLED signals of an actual imaging object in a magnetic resonance imaging instrument, and obtains a real sampling image after pretreatment; a neural network is trained through paired training samples of simulated IR-METMOLED images and simulated FLA-T2 images; the real part and the imaginary part of the pretreated real sampling image are input into the trained neural network for reconstruction, and a FLA-T2 image of the actual imaging object is obtained. The method can eliminate the echo time limitation of the echo chain length, eliminate the interference of cerebrospinal fluid signals, and quickly obtain the FLA-T2 image without increasing the signal complexity.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic resonance imaging, and in particular to a magnetic resonance fluid attenuation T2 (FLA-T2) quantitative imaging method and system. Background Art

[0002] Fluid-attenuated inversion recovery (FLAIR) imaging is indispensable in conventional MRI because it suppresses cerebrospinal fluid signals, thereby highlighting lesions adjacent to the cerebrospinal fluid, and has important application value in clinical disease diagnosis. However, due to the influence of cerebrospinal fluid dynamics and the long acquisition time of a series of weighted images, quantitative FLAIR imaging is difficult to apply clinically.

[0003] Recently, a study proposed a single-scan inversion recovery multiple overlapped echo split-plane imaging (IR-MOLED) method, which can obtain reliable fluid-attenuated inversion recovery T2 quantitative images in a very short time with a single scan, eliminating interference from cerebrospinal fluid signals. However, due to the limited echo train length, IR-MOLED's accuracy is reduced when quantifying large T2 values ​​due to its short echo time (TE). Some common clinical tumors are often accompanied by tumor infiltration and edema, resulting in large T2 values. This limitation reduces the clinical value of IR-MOLED in this area. Summary of the Invention

[0004] The present invention aims to provide a method and system for quantitative magnetic resonance imaging (FLA-T2). By designing a single-scan inversion recovery multiple echo chain with multiple overlapping echoes (IR-METMOLED) sequence, the limitation of echo chain length on TE can be eliminated without increasing signal complexity, enabling accurate quantification of large T2 values ​​while effectively suppressing cerebrospinal fluid signals and eliminating interference from these signals.

[0005] To achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows.

[0006] In one aspect, the present invention provides a method for magnetic resonance FLA-T2 quantitative imaging, comprising:

[0007] S1: Design of a magnetic resonance FLA-T2 quantitative imaging sequence called single scan inversion recovery multiple echo chain multiple overlap echo sequence (IR-METMOLED);

[0008] S2: Determine IR-METMOLED sequence sampling parameters;

[0009] S3: importing the IR-METMOLED sequence into the magnetic resonance imager and collecting data of the actual imaging object according to the determined sampling parameters to obtain the IR-METMOLED signal of the actual imaging object;

[0010] S4: Preprocessing the IR-METMOLED signal of the actual imaging object to obtain a preprocessed real-sampled IR-METMOLED image;

[0011] S5: Generate training samples for a neural network; the training samples include paired simulated IR-METMOLED images and simulated FLA-T2 images without cerebrospinal fluid signals; the simulated IR-METMOLED images without cerebrospinal fluid signals are used as inputs to the neural network, and the simulated FLA-T2 images are used as labels for the neural network; specifically, the steps include:

[0012] S51: generating a virtual imaging object;

[0013] S52: Writing the IR-METMOLED sequence on a simulation platform and performing simulated data acquisition on the virtual imaging object according to the determined sampling parameters to obtain a simulated IR-METMOLED signal of the virtual imaging object;

[0014] S53: Processing the simulated IR-METMOLED signal to obtain a simulated IR-METMOLED image;

[0015] S54: Repeat the process from S51 to S53 until a set amount of training samples are generated;

[0016] S6: Using the training samples to train the neural network to obtain a trained neural network;

[0017] S7: Inputting the real part and the imaginary part of the pre-processed real-sampled IR-METMOLED image into the trained neural network for reconstruction to obtain a FLA-T2 image of the actual imaged object.

[0018] In step S1, the IR-METMOLED sequence includes:

[0019] A liquid attenuation module, a MOLED module and n (n>1) readout modules. The liquid attenuation module contains an inversion recovery pulse and a layer selection gradient; the MOLED module contains a radio frequency excitation pulse with a flip angle of α and a corresponding echo shift gradient G; each readout module consists of a refocusing pulse with a flip angle of β, a destruction gradient G crand a sampling echo chain; each RF excitation pulse is applied synchronously with the slice-selective gradient of the slice selection dimension; the echo shift gradient is applied in the frequency encoding dimension and the phase encoding dimension; the destruction gradient is applied in the frequency encoding dimension, the phase encoding dimension and the slice selection dimension before and after the refocusing pulse; the sampling echo chain is composed of gradient chains acting in the frequency encoding dimension and the phase encoding dimension respectively, the gradient chain in the frequency encoding dimension is composed of a series of positive and negative gradients, and the gradient chain in the phase encoding dimension is composed of a series of gradients of equal area;

[0020] In step S2, the IR-METMOLED sequence sampling parameters are determined, specifically including:

[0021] Determine the number of RF excitation pulses; determine the flip angle and pulse shape of each RF excitation pulse; determine the flip angle and pulse shape of the refocusing pulse; determine the overlapping mode of the echoes generated by each RF excitation pulse, that is, the position of each echo in K space, so as to determine the time interval between each RF excitation pulse and the size ratio of the echo shift gradient after each RF excitation pulse; determine the imaging field of view, imaging matrix, number of acquisition layers, layer thickness, inter-layer spacing, acquisition bandwidth and each echo time.

[0022] In step S51, a virtual imaging object is generated, which specifically includes:

[0023] The virtual imaging objects include simulated FLA-T2 images, simulated T1 images, and simulated PD images, which are calculated using the magnetic resonance signal formula from the registered magnetic resonance T1-weighted images (T1WI), magnetic resonance T2-weighted images (T2WI), and magnetic resonance PD-weighted images (PDWI) in the public dataset. The analytical form of the magnetic resonance signal formula is expressed as follows:

[0024]

[0025] Wherein, SI represents the signal intensity of the magnetic resonance image; PD represents the proton density; TE represents the echo time of the magnetic resonance image acquisition; T2 represents the transverse relaxation time; TR represents the repetition time of the magnetic resonance image acquisition; T1 represents the longitudinal relaxation time;

[0026] For magnetic resonance PD-weighted images, according to the following magnetic resonance signal formula:

[0027]

[0028] Obtain the PD map, normalize it and use it as the simulated PD map of the virtual imaging object; wherein, SI PDWI represents the signal intensity of PD-weighted magnetic resonance images;

[0029] For magnetic resonance T1-weighted images, according to the following magnetic resonance signal formula:

[0030]

[0031] Obtain a T1 map, and scale it within a certain fluctuation range according to the T1 value range of the actual imaging object as a simulated T1 map of the virtual imaging object; wherein, SI T1WI It represents the signal intensity of T1-weighted magnetic resonance images.

[0032] For magnetic resonance T2-weighted images, according to the following magnetic resonance signal formula:

[0033]

[0034] Obtain a T2 map, scale it within a certain fluctuation range according to the T2 value range of the actual imaging object, and set the area where T1 is greater than 1800 milliseconds to zero according to the simulated T1 map, and obtain a FLA-T2 map without cerebrospinal fluid signal as the simulated FLA-T2 map of the virtual imaging object; wherein, SI T2WI It represents the signal intensity of T2-weighted magnetic resonance images.

[0035] In step S53, the simulated IR-METMOLED signal is processed, specifically including:

[0036] performing inverse Fourier transform, normalization, noise addition, Fourier transform, zero padding, and inverse Fourier transform on a first echo chain signal of the simulated IR-METMOLED signal to obtain an IR-METMOLED image of the first echo chain of the virtual imaging object;

[0037] The nth (n>1)th echo chain signal of the simulated IR-METMOLED signal is subjected to inverse Fourier transform, normalization, noise addition, Fourier transform, zero filling and inverse Fourier transform to obtain an uncorrected IR-METMOLED image of the nth echo chain of the virtual imaging object, and then the uncorrected IR-METMOLED image is point-multiplied by a two-dimensional random surface B βn Correction is performed to obtain the IR-METMOLED image of the nth echo chain of the virtual imaging object, B βn It is generated by polynomial interpolation and then dot-multiplied with the Boolean matrix of the FLA-T2 graph obtained in S51.

[0038] In step S6, the neural network is trained using the training samples to obtain a trained neural network, which specifically includes:

[0039] Determine the network structure, parameters and loss function of the neural network; when training the neural network, input the training samples into the neural network in batches for iterative training, calculate the value of the loss function, and automatically adjust the parameter values ​​of the neural network based on the value to reduce the value of the loss function. Repeat the above training until the loss function converges and save the neural network parameters.

[0040] In another aspect, the present invention provides a magnetic resonance FLA-T2 quantitative imaging system, comprising:

[0041] Sequence design module, used to determine the IR-METMOLED sequence and its sampling parameters;

[0042] a signal acquisition module, configured to import the IR-METMOLED sequence into a magnetic resonance imager and acquire data of an actual imaging object according to determined sampling parameters to obtain an IR-METMOLED signal of the actual imaging object;

[0043] The real-sampled signal preprocessing module is used to preprocess the IR-METMOLED signal of the actual imaging object to obtain the preprocessed real-sampled IR-METMOLED image;

[0044] A training sample generation module is used to generate training samples for a neural network; the training samples include paired simulated IR-METMOLED images and simulated FLA-T2 images without cerebrospinal fluid signals; the simulated IR-METMOLED images without cerebrospinal fluid signals are used as inputs to the neural network, and the simulated FLA-T2 images are used as labels for the neural network; specifically, the module includes:

[0045] A virtual imaging object generating unit is configured to generate a virtual imaging object; the virtual imaging object comprises a simulated PD map, a simulated T1 map, and a simulated FLA-T2 map, which are obtained from registered magnetic resonance PD-weighted images, magnetic resonance T1-weighted images, and magnetic resonance T2-weighted images in a public dataset using a magnetic resonance signal formula;

[0046] a simulation signal generating unit, configured to program the IR-METMOLED sequence on a simulation platform and perform simulation data acquisition on the virtual imaging object according to the determined sampling parameters to obtain a simulated IR-METMOLED signal of the virtual imaging object;

[0047] A simulation signal processing unit is used to perform Fourier inverse transform, normalization, noise addition, Fourier transform, zero filling, Fourier inverse transform and point multiplication of the simulated IR-METMOLED signal to a two-dimensional random surface B βn And other operations to obtain the simulated IR-METMOLED image;

[0048] a repetitive processing unit, which repeatedly executes the virtual imaging object generating unit, the simulation signal generating unit, and the simulation signal processing unit until a set amount of training samples are generated;

[0049] A network training module is used to train the neural network using the training samples to obtain a trained neural network;

[0050] The FLA-T2 image reconstruction module is used to input the real and imaginary parts of the preprocessed real-sampled IR-METMOLED image into the trained neural network for reconstruction to obtain a FLA-T2 image of the actual imaged object.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] (1) Compared with IR-MOLED imaging, the IR-METMOLED sequence designed in this invention eliminates the limitation of echo chain length on TE by introducing multiple echo chains. Each echo chain acquires spin echoes with different TE evolution times, thereby conveying T2 information with different TE weightings. This effectively improves the problem of limited signal tissue structure information obtained by only one echo chain and the difficulty in accurately quantifying large T2 values ​​due to short TE.

[0053] (2) In the neural network training sample generation stage, the nth (n>1) echo chain signal of the simulated IR-METMOLED signal is processed by point multiplication with a two-dimensional random surface B βn , improving the signal intensity degradation problem caused by B1 field inhomogeneity. Therefore, this imaging method can eliminate the limitation of echo train length on TE, achieve accurate quantification of large T2 values, and effectively suppress cerebrospinal fluid signals to eliminate interference from cerebrospinal fluid signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 FIG. 4 is a flow chart of a magnetic resonance FLA-T2 quantitative imaging method according to an embodiment of the present invention.

[0055] Figure 2 This is a sequence diagram of an IR-METMOLED according to an embodiment of the present invention.

[0056] Figure 3 1 and 2 are simulated IR-METMOLED images in an embodiment of the present invention, wherein (a) is the image of the first echo chain signal, and (b) is the image of the second echo chain signal.

[0057] Figure 4The FLA-T2 images and difference images reconstructed by the trained neural network for the simulated FLA-T2 images of a certain virtual imaging object in the test set in the embodiment of the present application and the corresponding simulated IR-METMOLED images. Among them, (a) is the simulated FLA-T2 image of the virtual imaging object, (b) is the FLA-T2 image reconstructed using only the first echo train signal of the simulated IR-METMOLED signal, (c) is the FLA-T2 image reconstructed using the first and second echo train signals of the simulated IR-METMOLED signal, (d) is the difference image of the FLA-T2 image reconstructed using only the first echo train signal of the simulated IR-METMOLED signal and the label image, and (e) is the difference image of the FLA-T2 image reconstructed using the first and second echo train signals of the simulated IR-METMOLED signal and the label image.

[0058] Figure 5 The FLA-T2 images reconstructed by the trained neural network for a certain real IR-METMOLED image in the example of the present application. Among them, (a) is the FLA-T2 image reconstructed using only the first echo train signal of the real IR-METMOLED signal, and (b) is the FLA-T2 image reconstructed using the first and second echo train signals of the real IR-METMOLED signal.

[0059] Figure 6 The structure block diagram of the magnetic resonance FLA-T2 quantitative imaging system of the embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0061] Reference Figure 1 The embodiment discloses a magnetic resonance FLA-T2 quantitative imaging method, comprising:

[0062] S1: design a magnetic resonance FLA-T2 quantitative imaging sequence, called IR-METMOLED sequence;

[0063] S2: determine the sampling parameters of the IR-METMOLED sequence;

[0064] S3: importing the IR-METMOLED sequence into the magnetic resonance imager and collecting data of the actual imaging object according to the determined sampling parameters to obtain the IR-METMOLED signal of the actual imaging object;

[0065] S4: Preprocessing the IR-METMOLED signal of the actual imaging object to obtain a preprocessed real-sampled IR-METMOLED image;

[0066] S5: generating training samples for a neural network; the training samples include paired simulated IR-METMOLED images and simulated FLA-T2 images without cerebrospinal fluid signals; using the simulated IR-METMOLED images without cerebrospinal fluid signals as inputs to the neural network, and using the simulated FLA-T2 images as labels for the neural network;

[0067] S6: Using the training samples to train the neural network to obtain a trained neural network;

[0068] S7: Inputting the real part and the imaginary part of the pre-processed real-sampled IR-METMOLED image into the trained neural network for reconstruction to obtain a FLA-T2 image of the actual imaged object.

[0069] Specifically, in step S1, the IR-METMOLED sequence specifically includes:

[0070] A liquid attenuation module, a MOLED module and n (n>1) readout modules. The liquid attenuation module contains an inversion recovery pulse and a layer selection gradient; the MOLED module contains a radio frequency excitation pulse with a flip angle of α and a corresponding echo shift gradient G; each readout module consists of a refocusing pulse with a flip angle of β, a destruction gradient G cr and a sampling echo chain; each RF excitation pulse is applied synchronously with the slice-selective gradient of the slice selection dimension; the echo shift gradient is applied in the frequency encoding dimension and the phase encoding dimension; the destruction gradient is applied in the frequency encoding dimension, the phase encoding dimension and the slice selection dimension before and after the refocusing pulse; the sampling echo chain is composed of gradient chains acting in the frequency encoding dimension and the phase encoding dimension respectively, the gradient chain in the frequency encoding dimension is composed of a series of positive and negative gradients, and the gradient chain in the phase encoding dimension is composed of a series of gradients of equal area;

[0071] Specifically, in step S2, determining the IR-METMOLED sequence sampling parameters specifically includes:

[0072] determining the number of radio frequency excitation pulses; determining the flip angle size and pulse shape of each radio frequency excitation pulse; determining the flip angle size and pulse shape of the refocusing pulse; determining the overlapping mode of the echoes generated by each radio frequency excitation pulse, i.e. the position of each echo in K-space, thereby determining the time interval between each radio frequency excitation pulse and the size proportion of the echo shift gradient after each radio frequency excitation pulse; determining the imaging field of view, the imaging matrix, the number of acquisition layers, the layer thickness, the interlayer spacing, the acquisition bandwidth and the echo time;

[0073] Specifically, in step S5, a training sample of the neural network is generated, and the specific process is as follows:

[0074] S51: generating a virtual imaging object;

[0075] Specifically, the virtual imaging object includes a simulated FLA-T2 map, a simulated T1 map and a simulated PD map, which are calculated from the registered magnetic resonance T1 weighted image T1WI, the magnetic resonance T2 weighted image T2WI and the magnetic resonance PD weighted image PDWI in the public data set by a magnetic resonance signal formula, and the analytical form of the magnetic resonance signal formula is expressed as follows:

[0076]

[0077] Wherein, SI represents the signal intensity of the magnetic resonance image; PD represents the proton density; TE represents the echo time of the magnetic resonance image acquisition; T2 represents the transverse relaxation time; TR represents the repetition time of the magnetic resonance image acquisition; T1 represents the longitudinal relaxation time;

[0078] For the magnetic resonance PD weighted image, according to the following magnetic resonance signal formula:

[0079]

[0080] The PD map is obtained, and after normalization, it is used as the simulated PD map of the virtual imaging object; wherein, SI PDWI represents the signal intensity of the magnetic resonance PD weighted image;

[0081] For the magnetic resonance T1 weighted image, according to the following magnetic resonance signal formula:

[0082]

[0083] The T1 map is obtained, and after scaling within a certain fluctuation range according to the T1 value range of the actual imaging object, it is used as the simulated T1 map of the virtual imaging object; wherein, SI T1WI represents the signal intensity of the magnetic resonance T1 weighted image. TR represents the repetition time of the magnetic resonance image acquisition; PD represents the proton density.

[0084] For magnetic resonance T2-weighted images, according to the following magnetic resonance signal formula:

[0085]

[0086] Obtain a T2 map, scale it within a certain fluctuation range according to the T2 value range of the actual imaging object, and set the area where T1 is greater than 1800 milliseconds to zero according to the simulated T1 map, and obtain a FLA-T2 map without cerebrospinal fluid signal as the simulated FLA-T2 map of the virtual imaging object; wherein, SI T2WI TE represents the echo time of magnetic resonance image acquisition; PD represents the proton density.

[0087] S52: Writing the IR-METMOLED sequence on a simulation platform and performing simulated data acquisition on the virtual imaging object according to the determined sampling parameters to obtain a simulated IR-METMOLED signal of the virtual imaging object;

[0088] S53: Processing the simulated IR-METMOLED signal to obtain a simulated IR-METMOLED image;

[0089] Specifically, the first echo chain signal of the simulated IR-METMOLED signal is subjected to inverse Fourier transform, normalization, noise addition, Fourier transform, zero filling, and inverse Fourier transform to obtain an IR-METMOLED image of the first echo chain of the virtual imaging object;

[0090] The nth (n>1)th echo chain signal of the simulated IR-METMOLED signal is subjected to inverse Fourier transform, normalization, noise addition, Fourier transform, zero filling and inverse Fourier transform to obtain an uncorrected IR-METMOLED image of the nth echo chain of the virtual imaging object, and then the uncorrected IR-METMOLED image is point-multiplied by a two-dimensional random surface B βn Correction is performed to obtain the IR-METMOLED image of the nth echo chain of the virtual imaging object, B βn The result is generated by polynomial interpolation and then dot-multiplied by the Boolean matrix of the FLA-T2 graph obtained in S51;

[0091] S54: Repeat steps S51 to S53 until a set amount of training samples are generated;

[0092] Specifically, in step S6, the neural network is trained using the training samples to obtain a trained neural network, which specifically includes:

[0093] Determine the network structure, parameters and loss function of the neural network; when training the neural network, input the training samples into the neural network in batches for iterative training, calculate the value of the loss function, and automatically adjust the parameter values ​​of the neural network based on the value to reduce the value of the loss function. Repeat the above training until the loss function converges and save the neural network parameters.

[0094] The following is a detailed description of the magnetic resonance FLA-T2 quantitative imaging process through a specific embodiment, which includes the following steps.

[0095] Step 1: Design the IR-METMOLED sequence and sampling parameters.

[0096] Step 2: Using the IR-METMOLED sequence and its parameters determined in step 1, GRAPPA technology is used to collect data on the actual imaging object to obtain the IR-METMOLED signal of the actual imaging object.

[0097] Step 3: Preprocess the IR-METMOLED signal of the actual imaging object to obtain a preprocessed real-sampled IR-METMOLED image.

[0098] Step 4: Generate training samples for the neural network; the training samples include paired simulated IR-METMOLED images without cerebrospinal fluid signals and simulated FLA-T2 images; the simulated IR-METMOLED images without cerebrospinal fluid signals are used as inputs to the neural network, and the simulated FLA-T2 images are used as labels for the neural network; specifically, the steps include:

[0099] Step 41: Generate a virtual imaging object;

[0100] Step 42: Program the IR-METMOLED sequence on a simulation platform and perform simulated data acquisition on the virtual imaging object according to the determined sampling parameters to obtain a simulated IR-METMOLED signal of the virtual imaging object. The experimental parameters in the simulation should be consistent with the parameters of the actual acquired data as much as possible. The sampling matrix size is 128×128, and small random perturbations such as random gradient offset and non-uniform RF field are added during the simulation.

[0101] Step 43: processing the analog IR-METMOLED signal to obtain an analog IR-METMOLED image; first, performing inverse Fourier transform on the analog IR-METMOLED signal to obtain an analog IR-METMOLED image; second, normalizing the analog IR-METMOLED image by dividing the maximum value of the first echo chain image, and then performing Fourier transform to convert it into a K-space signal; then, adding noise close to the signal-to-noise ratio of the real IR-METMOLED signal to reduce the gap between the synthetic domain and the real domain; then, performing zero padding on the K-space of the synthetic data to expand the original data matrix from 128x128 to 256x256; then, performing inverse Fourier transform to convert it to the image domain. The IR-METMOLED image of the first echo chain corresponding to the virtual imaging object is obtained, which is referred to as (a) in Figure 3

[0102] For the analog IR-METMOLED signal of the second echo chain, after the inverse Fourier transform, normalization, noise addition, Fourier transform, zero padding, and inverse Fourier transform, the uncorrected IR-METMOLED image of the second echo chain of the virtual imaging object is obtained, and then the uncorrected IR-METMOLED image is point multiplied by a two-dimensional random curved surface B βn to correct it to obtain the IR-METMOLED image of the second echo chain of the virtual imaging object, which is referred to as (b) in Figure 3 B βn which is generated by polynomial interpolation and then point multiplied by the Boolean matrix of the FLA-T2 map obtained in the S51 to obtain B βn The range of B

[0103] Step 44: repeating the steps 41-43 until a certain amount of training samples are generated.

[0104] Step 5: training the neural network using the training samples to obtain a trained neural network.

[0105] Step 6: inputting the real part and the imaginary part of the preprocessed real IR-METMOLED image into the trained neural network for reconstruction to obtain the FLA-T2 map of the actual imaging object.

[0106] In this embodiment, in step 1, the designed IR-METMOLED sequence diagram is as shown in Figure 2 ​As shown, the sequence includes a liquid attenuation module, a MOLED module and n (n>1) readout modules. In this embodiment, n is 2. The liquid attenuation module includes an inversion recovery pulse and a slice selection gradient; the MOLED module includes a radio frequency excitation pulse with a flip angle of α and a corresponding echo shift gradient G; each readout module consists of a refocusing pulse with a flip angle of β, a destruction gradient G cr and a sampling echo chain; each RF excitation pulse is applied synchronously with the slice selection gradient of the slice selection dimension; the echo shift gradient is applied in the frequency encoding dimension and the phase encoding dimension; the destruction gradient is applied in the frequency encoding dimension, the phase encoding dimension and the slice selection dimension before and after the refocusing pulse; the sampling echo chain is composed of gradient chains acting on the frequency encoding dimension and the phase encoding dimension respectively, the gradient chain of the frequency encoding dimension is composed of a series of positive and negative gradients, and the gradient chain of the phase encoding dimension is composed of a series of gradients of equal area; l1 + l2 + l3 + l4 + l5 + 8 = L, where L is the number of phase encoding steps, l i express Figure 2 The dotted brackets in the figure indicate the number of phase encoding steps; echo indicates the center position of echo refocusing;

[0107] Determine the sampling parameters of the IR-METMOLED sequence, including:

[0108] The number of RF excitation pulses was 4, and the RF excitation pulses were Sinc pulses with a flip angle of 30°. The time intervals between the four RF excitation pulses were 17.35 ms, 17.35 ms, 17.35 ms, and 12.19 ms, respectively. The refocusing pulse was a Sinc pulse with a flip angle of 180°. G1, G2, G3, and G4 were four shift gradients. The ratios of the four shift gradient areas to the first readout gradient area in the frequency encoding dimension were -74 / 256, 74 / 256, -74 / 256, and -93 / 256, respectively. The ratios of the four shift gradient areas to the total blip gradient area in the phase encoding dimension were -64 / 260, -88 / 260, -24 / 260, and -11 / 260, respectively. The TI was 2.5 s. The imaging matrix was 128×128. The slice thickness was 3.5 mm. The imaging field of view was 220 mm×220 mm. mm; the number of acquisition layers is 21; the interval between layers is 4.5 mm; the acquisition bandwidth is 1300 Hz / Pixel; TE1, TE2, TE3, and TE4 are 22 ms, 52 ms, 82 ms, and 110 ms, respectively.

[0109] In step 3, the IR-METMOLED signal of the actual imaged object is preprocessed to obtain a preprocessed, actually acquired IR-METMOLED image. This specifically includes: performing GRAPPA reconstruction on the acquired IR-METMOLED signal, combining signals from different coils, and then performing an inverse Fourier transform on the signal, transferring it to the image domain, and normalizing it. The normalization is based on the maximum value of the first echo train image of the actually acquired IR-METMOLED. The normalized image is Fourier transformed into a K-space signal, which is then zero-filled. The original data matrix (128×128) is expanded to 256×256, and then subjected to an inverse Fourier transform to the image domain.

[0110] In step 5, the neural network adopts U-Net with a residual block and an attention mechanism; the Adam optimizer is used, the initial learning rate is 0.0001, and the learning rate is reduced by 20% after every 80,000 iterations; the network training is performed in blocks, with a batch number of 8 and a total of 1,000 rounds of training; the input of the neural network is an IR-METMOLED image, which is separated into real and imaginary parts, and the output is a FLA-T2 quantitative image; the loss function is a pixel-based mean square error function, and the formula is as follows:

[0111]

[0112] Among them, N represents the number of training samples, which is 5400, f is the nonlinear mapping represented by the network, and x i Represents the input of the i-th training sample, y i Represents the label of the i-th training sample.

[0113] In order to evaluate the magnetic resonance FLA-T2 quantitative imaging method of the present invention, this example Figure 4 The simulated FLA-T2 image of a virtual imaging object and the corresponding simulated IR-METMOLED image are displayed, and the FLA-T2 image and difference image reconstructed by the trained neural network are shown; Figure 4 (a) is the simulated FLA-T2 image of the virtual imaging object. Figure 4 (b) is the FLA-T2 image reconstructed using only the first echo chain signal of the simulated IR-METMOLED signal. Figure 4 (c) is the FLA-T2 image reconstructed using the first and second echo chain signals of the simulated IR-METMOLED signal. Figure 4 (d) is the difference image between the FLA-T2 image reconstructed using only the first echo chain signal of the simulated IR-METMOLED signal and the label image. Figure 4(e) is the difference image between the FLA-T2 image and the label image reconstructed using the first and second echo chain signals of the simulated IR-METMOLED signal; Figure 4 (d) in Figure 4 Compared with (e) in the figure, it can be seen that the present invention can quantify the T2 value more accurately, especially in the area with higher T2 values. The FLA-T2 image reconstructed using two echo chain signals is closer to the label image, and the difference between the two is smaller.

[0114] Figure 5 This is a FLA-T2 image reconstructed from a real IR-METMOLED image using a trained neural network in an embodiment of the present invention; Figure 5 (a) is the FLA-T2 image reconstructed using only the first echo chain signal of the real-sampled IR-METMOLED signal. Figure 5 (b) is the FLA-T2 image reconstructed using the first and second echo chain signals of the real IR-METMOLED signal; Figure 5 (a) and Figure 5 Comparing the results with those in Figure 2 (b) shows that the present invention can effectively reconstruct FLA-T2 images from real-world IR-METMOLED signals. Furthermore, the reconstruction using two echo train signals better restores object details and more accurately quantifies T2 values ​​compared to reconstruction using only a single echo train signal. This demonstrates that, compared with IR-MOLED, IR-METMOLED improves quantitative accuracy and tissue detail. These improvements are more pronounced in applications with large T2 tissues and high temporal resolution.

[0115] See also Figure 6 As shown, this embodiment also discloses a magnetic resonance FLA-T2 quantitative imaging system, comprising:

[0116] A sequence design module 71 is used to determine the IR-METMOLED sequence and its sampling parameters;

[0117] A signal acquisition module 72 is configured to import the IR-METMOLED sequence into the magnetic resonance imager and acquire data of the actual imaging object according to the determined sampling parameters to obtain the IR-METMOLED signal of the actual imaging object;

[0118] The real-acquired signal preprocessing module 73 is used to preprocess the IR-METMOLED signal of the actual imaging object to obtain a preprocessed real-acquired IR-METMOLED image;

[0119] The training sample generation module 74 is configured to generate training samples for the neural network; the training samples include paired simulated IR-METMOLED images and simulated FLA-T2 images that do not contain cerebrospinal fluid signals; the simulated IR-METMOLED images that do not contain cerebrospinal fluid signals are used as inputs to the neural network, and the simulated FLA-T2 images are used as labels for the neural network; specifically, the following steps are performed:

[0120] A virtual imaging object generating unit 741 is configured to generate a virtual imaging object; the virtual imaging object includes a simulated PD map, a simulated T1 map, and a simulated FLA-T2 map, which are obtained from the registered MRI PD-weighted image, MRI T1-weighted image, and MRI T2-weighted image in the public dataset using a MRI signal formula;

[0121] A simulation signal generating unit 742 is configured to program the IR-METMOLED sequence on a simulation platform and perform simulation data acquisition on the virtual imaging object according to the determined sampling parameters to obtain a simulated IR-METMOLED signal of the virtual imaging object;

[0122] The simulation signal processing unit 743 is used to perform inverse Fourier transform, normalization, noise addition, Fourier transform, zero filling, inverse Fourier transform and point multiplication of the simulated IR-METMOLED signal to a two-dimensional random surface B. βn And other operations to obtain the simulated IR-METMOLED image;

[0123] a repetitive processing unit 744, which repeatedly executes the virtual imaging object generating unit, the simulation signal generating unit, and the simulation signal processing unit until a set amount of training samples are generated;

[0124] A network training module 75 is used to train the neural network using the training samples to obtain a trained neural network;

[0125] The FLA-T2 image reconstruction module 76 is used to input the real part and the imaginary part of the pre-processed real-sampled IR-METMOLED image into the trained neural network for reconstruction to obtain a FLA-T2 image of the actual imaged object.

[0126] The specific implementation of a magnetic resonance FLA-T2 quantitative imaging system is the same as that of the magnetic resonance FLA-T2 quantitative imaging method, and will not be repeated in this embodiment.

[0127] The present invention designs a single-scan inversion recovery multi-echo chain multi-overlap echo sequence IR-METMOLED and determines its sampling parameters. The IR-METMOLED sequence is imported into a magnetic resonance imaging device and data is collected from an actual imaging object according to the set sampling parameters to obtain an IR-METMOLED signal of the actual imaging object. The IR-METMOLED signal of the actual imaging object is preprocessed to obtain a preprocessed real-acquired IR-METMOLED image. A neural network training sample is generated, including paired simulated IR-METMOLED images and simulated FLA-T2 maps that do not contain cerebrospinal fluid signals. The neural network is trained using synthetic training samples. The real and imaginary parts of the preprocessed real-acquired IR-METMOLED images are input into the trained neural network for reconstruction to obtain a FLA-T2 map of the actual imaging object. This method eliminates the limitation of echo chain length on echo time without increasing signal complexity, while eliminating interference from cerebrospinal fluid signals and rapidly acquiring a FLA-T2 map.

[0128] The principles and operation of the present invention are illustrated through the above-mentioned specific implementation examples. These examples are intended to provide readers with a clear understanding framework to grasp the core ideas and key operational points of the present invention. However, it should be made clear that these examples are not intended to limit the scope of application of the present invention. For professionals in this field, various forms of improvements and innovations based on the core ideas of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for quantitative magnetic resonance liquid attenuation T2 imaging, characterized in that: include: S1: Design of a magnetic resonance FLA-T2 quantitative imaging sequence, called single-scan inversion recovery multiple echo chain multiple overlap echo sequence IR-METMOLED; IR-METMOLED includes: A liquid attenuation module, a MOLED module and n readout modules, n>1; the liquid attenuation module contains an inversion recovery pulse and a layer selection gradient; the MOLED module contains a radio frequency excitation pulse with a flip angle of α and a corresponding echo shift gradient G; each readout module consists of a refocusing pulse with a flip angle of β, a destruction gradient G cr and a sampling echo chain; each RF excitation pulse is applied synchronously with the slice-selective gradient of the slice selection dimension; the echo shift gradient is applied in the frequency encoding dimension and the phase encoding dimension; the destruction gradient is applied in the frequency encoding dimension, the phase encoding dimension and the slice selection dimension before and after the refocusing pulse; the sampling echo chain is composed of gradient chains acting in the frequency encoding dimension and the phase encoding dimension respectively, the gradient chain in the frequency encoding dimension is composed of a series of positive and negative gradients, and the gradient chain in the phase encoding dimension is composed of a series of gradients of equal area; S2: Determine the IR-METMOLED sequence sampling parameters; S3: importing the IR-METMOLED sequence into the magnetic resonance imager and collecting data of the actual imaging object according to the determined sampling parameters to obtain the IR-METMOLED signal of the actual imaging object; S4: Preprocessing the IR-METMOLED signal of the actual imaging object to obtain a preprocessed real-sampled IR-METMOLED image; S5: Generate training samples for a neural network; the training samples include paired simulated IR-METMOLED images and simulated FLA-T2 images without cerebrospinal fluid signals; the simulated IR-METMOLED images without cerebrospinal fluid signals are used as inputs to the neural network, and the simulated FLA-T2 images are used as labels for the neural network; specifically, the steps include: S51: generating a virtual imaging object; S52: Writing the IR-METMOLED sequence on a simulation platform and performing simulated data acquisition on the virtual imaging object according to the determined sampling parameters to obtain a simulated IR-METMOLED signal of the virtual imaging object; S53: Processing the simulated IR-METMOLED signal to obtain a simulated IR-METMOLED image; S54: Repeat the process from S51 to S53 until a set amount of training samples are generated; S6: Use the training samples to train the neural network to obtain a trained neural network; S7: The real part and the imaginary part of the pre-processed real-sampled IR-METMOLED image are input into the trained neural network for reconstruction to obtain a FLA-T2 image of the actual imaged object.

2. The method for quantitative magnetic resonance liquid attenuation T2 imaging according to claim 1, characterized in that: In step S2, the IR-METMOLED sequence sampling parameters are determined, specifically including: Determine the number of RF excitation pulses; determine the flip angle and pulse shape of each RF excitation pulse; determine the flip angle and pulse shape of the refocusing pulse; determine the overlapping mode of the echoes generated by each RF excitation pulse, that is, the position of each echo in K space, so as to determine the time interval between each RF excitation pulse and the size ratio of the echo shift gradient after each RF excitation pulse; determine the imaging field of view, imaging matrix, number of acquisition layers, layer thickness, inter-layer spacing, acquisition bandwidth and each echo time.

3. The method for quantitative magnetic resonance liquid attenuation T2 imaging according to claim 1, characterized in that: Generating a virtual imaging object in step S51 specifically includes: The virtual imaging objects include simulated FLA-T2 images, simulated T1 images, and simulated PD images, which are calculated using the magnetic resonance signal formula from the registered magnetic resonance T1-weighted images (T1WI), magnetic resonance T2-weighted images (T2WI), and magnetic resonance PD-weighted images (PDWI) in the public dataset. The analytical form of the magnetic resonance signal formula is expressed as follows: Where SI is the signal intensity of the magnetic resonance image; PD is the proton density; TE is the echo time of the magnetic resonance image acquisition; T2 is the transverse relaxation time; TR is the repetition time of the magnetic resonance image acquisition; T1 is the longitudinal relaxation time; e is the base of the natural logarithm; For magnetic resonance PD-weighted images, according to the following magnetic resonance signal formula: PD=YES PDWI Obtain the PD map, normalize it and use it as the simulated PD map of the virtual imaging object; wherein, SI PDWI represents the signal intensity of PD-weighted magnetic resonance images; For magnetic resonance T1-weighted images, according to the following magnetic resonance signal formula: Obtain a T1 map, and scale it within a certain fluctuation range according to the T1 value range of the actual imaging object as a simulated T1 map of the virtual imaging object; wherein, SI T1WI represents the signal intensity of the T1-weighted magnetic resonance image; TR represents the repetition time of magnetic resonance image acquisition; PD represents the proton density; For magnetic resonance T2-weighted images, according to the following magnetic resonance signal formula: Obtain a T2 map, scale it within a certain fluctuation range according to the T2 value range of the actual imaging object, and set the area where T1 is greater than 1800 milliseconds to zero according to the simulated T1 map, and obtain a FLA-T2 map without cerebrospinal fluid signal as the simulated FLA-T2 map of the virtual imaging object; wherein, SI T2WI TE represents the echo time of magnetic resonance image acquisition; PD represents the proton density.

4. The method for quantitative magnetic resonance liquid attenuation T2 imaging according to claim 3, characterized in that: In step S53, the analog IR-METMOLED signal is processed, specifically including: Performing inverse Fourier transform, normalization, noise addition, Fourier transform, zero filling, and inverse Fourier transform on the first echo chain signal of the simulated IR-METMOLED signal to obtain an IR-METMOLED image of the first echo chain of the virtual imaging object; The nth echo chain signal of the simulated IR-METMOLED signal is subjected to inverse Fourier transform, normalization, noise addition, Fourier transform, zero filling and inverse Fourier transform, where n>1, to obtain the uncorrected IR-METMOLED image of the nth echo chain of the virtual imaging object. The uncorrected IR-METMOLED image is then point-multiplied by a two-dimensional random surface B. βn Correction is performed to obtain the IR-METMOLED image of the nth echo chain of the virtual imaging object, B βn It is generated by polynomial interpolation and then dot-multiplied with the Boolean matrix of the simulated FLA-T2 map obtained in S51.

5. The method for quantitative magnetic resonance liquid attenuation T2 imaging according to claim 1, characterized in that: In step S6, the neural network is trained using the training samples to obtain a trained neural network, which specifically includes: Determine the network structure, parameters and loss function of the neural network; when training the neural network, input the training samples into the neural network in batches for iterative training, calculate the value of the loss function, and automatically adjust the parameter values ​​of the neural network based on the value to reduce the value of the loss function. Repeat the above training until the loss function converges and save the neural network parameters.

6. A magnetic resonance liquid attenuation T2 quantitative imaging system, characterized in that The method for quantitative magnetic resonance liquid attenuation T2 imaging according to claim 1 is used, wherein the system comprises: Sequence design module, used to determine the IR-METMOLED sequence and its sampling parameters; a signal acquisition module, configured to import the IR-METMOLED sequence into a magnetic resonance imager and acquire data of an actual imaging object according to determined sampling parameters to obtain an IR-METMOLED signal of the actual imaging object; The real-sampled signal preprocessing module is used to preprocess the IR-METMOLED signal of the actual imaging object to obtain the preprocessed real-sampled IR-METMOLED image; A training sample generation module is used to generate training samples for a neural network; the training samples include paired simulated IR-METMOLED images and simulated FLA-T2 images without cerebrospinal fluid signals; the simulated IR-METMOLED images without cerebrospinal fluid signals are used as inputs to the neural network, and the simulated FLA-T2 images are used as labels for the neural network; specifically, the module includes: A virtual imaging object generating unit is configured to generate a virtual imaging object; the virtual imaging object comprises a simulated PD map, a simulated T1 map, and a simulated FLA-T2 map, which are obtained from registered magnetic resonance PD-weighted images, magnetic resonance T1-weighted images, and magnetic resonance T2-weighted images in a public dataset using a magnetic resonance signal formula; a simulation signal generating unit, configured to program an IR-METMOLED sequence on a simulation platform and perform simulation data acquisition on the virtual imaging object according to determined sampling parameters to obtain a simulated IR-METMOLED signal of the virtual imaging object; The simulation signal processing unit is used to perform Fourier inverse transform, normalization, noise addition, Fourier transform, zero filling, Fourier inverse transform and point multiplication of a two-dimensional random surface B on the simulated IR-METMOLED signal. βn , get the simulated IR-METMOLED image; a repetitive processing unit, which repeatedly executes the virtual imaging object generating unit, the simulation signal generating unit, and the simulation signal processing unit until a set amount of training samples are generated; A network training module is used to train the neural network using the training samples to obtain a trained neural network; The FLA-T2 image reconstruction module is used to input the real and imaginary parts of the preprocessed real-sampled IR-METMOLED image into the trained neural network for reconstruction to obtain a FLA-T2 image of the actual imaged object.

Images