A method and system for spin echo type EPI image distortion correction

Abstract

The application discloses a spin echo type EPI image distortion correction method and system, comprising the following steps: designing a multi-echo chain bipolar phase encoding spin echo type EPI sequence; generating training samples of a deep neural network, wherein the training samples comprise paired distorted simulation samples and target samples without distortion; training the deep neural network by using the training samples to obtain a trained deep neural network; collecting data of an actual imaging object by using the multi-echo chain bipolar phase encoding spin echo type EPI sequence to obtain a distorted image of the actual imaging object; and correcting the distortion of the distorted image of the actual imaging object by using the trained deep neural network to obtain an image without distortion. The application can quickly and efficiently correct the spin echo type EPI image distortion without additional reference scanning assistance.

Description

Technical Field

[0001] This invention relates to the field of magnetic resonance imaging (MRI), and in particular to a method and system for correcting image distortion based on spin echo EPI. Background Technology

[0002] Echo planar imaging (EPI) allows for the acquisition of all data from a two-dimensional image in a single radiofrequency excitation, making it one of the fastest magnetic resonance imaging (MRI) methods to date. EPI-based MRI is widely used in important medical imaging fields such as functional magnetic resonance imaging (fMRI), quantitative magnetic resonance imaging (QMRI), diffusion-weighted imaging (DWI), and diffusion tensor imaging (DTI). However, during EPI readout, phase accumulation due to inhomogeneity of the main magnetic field and continuous gradient switching leads to image distortion. Current methods for correcting EPI image distortion still require two scans, increasing scan time and making image quality susceptible to motion effects. They also typically require time-consuming post-processing, limiting their clinical application. Furthermore, existing EPI image distortion correction methods rely on the quality of the predicted offset field, which may not reliably correct distortion in regions with severe magnetic field inhomogeneity. Therefore, there is an urgent need for an image distortion correction method that does not require additional reference scans and is independent of the predicted offset field. Summary of the Invention

[0003] This invention provides a spin-echo type EPI image distortion correction method and system, which can quickly acquire distortion-free high-quality images by using a deep neural network to correct the image distortion in the opposite direction caused by bipolar phase encoding without additional distortion-free reference scans or repeated scans.

[0004] The present invention adopts the following technical solution:

[0005] On the one hand, a spin-echo type EPI image distortion correction method includes:

[0006] S1: Design a multi-echo chain bipolar phase-coded spin echo EPI sequence; the spin echo EPI sequence is an arbitrary magnetic resonance imaging sequence that uses radio frequency pulses to generate spin echo signals and uses EPI technology for signal readout.

[0007] S2: Generate training samples for a deep neural network; the training samples include paired distorted simulation samples and undistorted target samples; the distorted simulation samples serve as input to the deep neural network training samples, and the undistorted target samples serve as labels for the deep neural network training samples; specifically including:

[0008] S21: Obtain the registered magnetic resonance T2-weighted image (T2w) and proton density-weighted image (PDw) from the public dataset, obtain the magnetic resonance parameter map using the Bloch formula, and use the magnetic resonance parameter map as a virtual imaging object;

[0009] S22: Model the inhomogeneity of the main magnetic field to obtain the model of the inhomogeneity of the main magnetic field;

[0010] S23: Generate randomly distributed two-dimensional Gaussian noise to simulate the noise present in the actual sampling process of magnetic resonance images;

[0011] S24: In the Bloch simulation, under the presence of the main magnetic field inhomogeneity model and the two-dimensional Gaussian noise, the multi-echo chain bipolar phase-coded spin echo EPI sequence is used to acquire data of the virtual imaging object to obtain a distorted simulation sample.

[0012] S25: If the desired distortion-free image is a magnetic resonance weighted image, then in the Bloch simulation, the multi-echo chain bipolar phase-coded spin echo EPI sequence is used to acquire data from the virtual imaging object to obtain a distortion-free simulation sample as the distortion-free target sample; if the desired distortion-free image is a magnetic resonance parameter map, then the virtual imaging object is used as the distortion-free target sample.

[0013] S26: Repeat the process from S21 to S25 until a set number of deep neural network training samples are generated;

[0014] S3: Use the training samples to train a deep neural network to obtain a trained deep neural network;

[0015] S4: Data acquisition of the actual imaging object is performed using the multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain a distorted image of the actual imaging object; wherein, the distortion direction of the odd-numbered echo chain image is opposite to that of the even-numbered echo chain image.

[0016] S5: The trained deep neural network is used to correct the distortion of the actual image of the object to obtain a distortion-free image.

[0017] Preferably, in step S1, the design of a multi-echo chain bipolar phase-coded spin-echo type EPI sequence specifically includes:

[0018] Phase coding design: Modify the phase coding of the multi-echo chain spin echo EPI sequence so that the phase coding gradient directions of the odd and even echo chains are opposite, resulting in a multi-echo chain bipolar phase-coded spin echo EPI sequence.

[0019] The phase-encoded wraparound design adds a phase-encoded wraparound gradient before the phase encoding of the echo chains except the first echo chain, ensuring that the echo signals of each echo chain in the multi-echo chain bipolar phase-encoded spin echo EPI sequence fill the same position in their respective k-space.

[0020] Preferably, in step S21, obtaining the magnetic resonance parameter map using the Bloch formula and using the magnetic resonance parameter map as a virtual imaging object specifically includes:

[0021] The analytical form of the Bloch formula is expressed as follows:

[0022]

[0023] Where S represents the signal intensity of the magnetic resonance image, x and y represent the coordinates of the two-dimensional plane, 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; and T1 represents the longitudinal relaxation time.

[0024] PDw images are magnetic resonance images acquired under short TE and long TR conditions. In this case, the analytical form of the Bloch formula simplifies to:

[0025] S(x,y)=PD(x,y)

[0026] Therefore, the image obtained after normalizing the signal intensity of the PDw image is used as the PD map of the virtual imaging object;

[0027] T2w images are magnetic resonance images acquired under appropriate TE and long TR conditions. In this case, the analytical form of the Bloch formula simplifies to:

[0028]

[0029] Therefore, the corresponding T2 can be obtained from the signal strength, TE and PD of the T2w image. All T2 values ​​on the two-dimensional plane constitute the T2 map. The T2 values ​​of the T2 map are scaled to a range that is suitable for the actual imaging object. The resulting T2 map is used as the T2 map of the virtual imaging object.

[0030] The T2 image and PD image are combined to form a virtual imaging object.

[0031] Preferably, in step S22, the inhomogeneity of the main magnetic field is modeled to obtain the main magnetic field inhomogeneity model, as follows:

[0032]

[0033] Among them, the main magnetic field inhomogeneity model dB0 is expressed as the sum of a two-dimensional polynomial function and a two-dimensional Gaussian function; The expression represents a two-dimensional polynomial function; x and y represent coordinates in the two-dimensional plane; n x and n y Let r represent the orders of x and y, respectively; p N represents the coefficients of a randomly generated two-dimensional polynomial function; p G represents the highest order of a two-dimensional polynomial function; G represents a two-dimensional Gaussian function; N g The number of two-dimensional Gaussian functions is represented by r; m and r s Let represent the mean and standard deviation of a randomly generated two-dimensional Gaussian function, respectively.

[0034] On the other hand, a spin-echo EPI image distortion correction system includes:

[0035] The pulse sequence design module is used to design multi-echo chain bipolar phase-coded spin echo EPI sequences; the spin echo EPI sequence is an arbitrary magnetic resonance imaging sequence that uses radio frequency pulses to generate spin echo signals and uses EPI technology for signal readout.

[0036] A training sample generation module is used to generate training samples for a deep neural network. The training samples include paired distorted simulation samples and undistorted target samples. The distorted simulation samples serve as input to the deep neural network training samples, and the undistorted target samples serve as labels for the deep neural network training samples. Specifically, it includes:

[0037] The virtual imaging object determination unit is used to obtain the registered magnetic resonance T2-weighted image (T2w) and proton density-weighted image (PDw) from the public dataset, obtain the magnetic resonance parameter map through the Bloch formula, and use the magnetic resonance parameter map as the virtual imaging object.

[0038] The main magnetic field inhomogeneity modeling unit is used to model the main magnetic field inhomogeneity and obtain the main magnetic field inhomogeneity model.

[0039] A Gaussian noise generation unit is used to generate randomly distributed two-dimensional Gaussian noise to simulate the noise present in the actual sampling process of magnetic resonance images;

[0040] A distortion sample generation unit is provided to generate distorted simulation samples. In the Bloch simulation, the virtual imaging object is acquired using the multi-echo chain bipolar phase-coded spin echo EPI sequence in the presence of the main magnetic field inhomogeneity model and the two-dimensional Gaussian noise, thereby obtaining distorted simulation samples.

[0041] A distortion-free target sample generation unit is used to generate distortion-free target samples. If the desired distortion-free image is a magnetic resonance weighted image, the virtual imaging object is acquired using the multi-echo chain bipolar phase-coded spin echo EPI sequence in the Bloch simulation to obtain a distortion-free simulation sample as the distortion-free target sample. If the desired distortion-free image is a magnetic resonance parameter map, the virtual imaging object is used as the distortion-free target sample.

[0042] The repetitive processing unit repeatedly executes the virtual imaging object determination unit, the main magnetic field inhomogeneity modeling unit, the Gaussian noise generation unit, the distorted sample generation unit, and the distortion-free target sample generation unit until a set amount of deep neural network training samples are generated.

[0043] The network training module is used to train a deep neural network using the training samples to obtain a trained deep neural network.

[0044] The data acquisition module is used to acquire data of the actual imaging object using the multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain a distorted image of the actual imaging object; wherein the distortion direction of the odd-numbered echo chain image is opposite to that of the even-numbered echo chain image.

[0045] The distortion correction module is used to correct the distortion of the actual image of the object by using the trained deep neural network to obtain a distortion-free image.

[0046] The beneficial effects of this invention are as follows:

[0047] (1) The present invention can realize distortion correction of images acquired by any magnetic resonance imaging sequence that uses radio frequency pulses to generate spin echo signals and uses EPI technology for signal readout;

[0048] (2) This invention designs a multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain images with opposite distortion directions in a single scan.

[0049] (3) Existing image distortion correction methods cannot be used due to the existence of phase-encoded wraparound gradients; This invention is based on deep neural networks and uses complementary distortion information to correct image distortion.

[0050] (4) The present invention does not require additional reference scans to provide distortion-free information, nor does it require repeated scans to obtain different distortion information, thus improving scanning efficiency;

[0051] (5) By using simulated samples to train a deep neural network, the present invention does not require the collection of a large amount of MRI data as a training set, thus saving resources; the distortion correction of spin echo EPI images is achieved through the trained deep neural network, with faster correction speed and higher image quality.

[0052] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments, but the spin echo type EPI image distortion correction method and system of the present invention are not limited to the embodiments. Attached Figure Description

[0053] Figure 1 This is a flowchart of the spin echo EPI image distortion correction method according to an embodiment of the present invention;

[0054] Figure 2 This is a flowchart illustrating the generation of deep neural network training samples in an embodiment of the present invention.

[0055] Figure 3 This is a diagram of a multi-echo chain bipolar phase-coded spin-echo type EPI sequence according to an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of the simulation training samples generated in the embodiments of the present invention and the distortion correction results obtained by the trained deep neural network.

[0057] Figure 5 The images are those acquired by the multi-echo chain bipolar phase-coded spin echo EPI sequence in this embodiment of the invention, and the images after distortion correction by a trained deep neural network.

[0058] Figure 6 This is a structural block diagram of the spin-echo EPI image distortion correction system according to an embodiment of the present invention;

[0059] Figure 7 This is a structural block diagram of the training sample generation module in an embodiment of the present invention. Detailed Implementation

[0060] The present invention will be further described below through specific embodiments. It should be noted that the specific embodiments described herein are only for the convenience of illustrating and explaining the specific implementation of the present invention, and are not intended to limit the present invention.

[0061] To make the objectives and technical solutions of this invention clearer, the invention will be further described below with reference to the accompanying drawings and examples. It should be understood that the examples described herein are for illustrative purposes only and are not intended to limit the invention.

[0062] See Figure 1 As shown in this embodiment, a spin-echo type EPI image distortion correction method includes:

[0063] S1: Design a multi-echo chain bipolar phase-coded spin echo EPI sequence; the spin echo EPI sequence is an arbitrary magnetic resonance imaging sequence that uses radio frequency pulses to generate spin echo signals and uses EPI technology for signal readout.

[0064] S2: Generate training samples for a deep neural network; the training samples include paired distorted simulation samples and undistorted target samples; the distorted simulation samples serve as input to the deep neural network training samples, and the undistorted target samples serve as labels for the deep neural network training samples, which are simulation data of the desired undistorted image, and can be magnetic resonance weighted images or magnetic resonance parameter maps.

[0065] S3: Use the training samples to train a deep neural network to obtain a trained deep neural network;

[0066] S4: Data acquisition of the actual imaging object is performed using the multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain a distorted image of the actual imaging object; wherein, the distortion direction of the odd-numbered echo chain image is opposite to that of the even-numbered echo chain image.

[0067] S5: The trained deep neural network is used to correct the distortion of the actual image of the object to obtain a distortion-free image.

[0068] Specifically, in S1, the design of the multi-echo chain bipolar phase-coded spin-echo type EPI sequence includes:

[0069] The phase coding design modifies the phase coding of the multi-echo chain spin echo EPI sequence, making the phase coding gradient directions of the odd and even echo chains opposite, resulting in a multi-echo chain bipolar phase-coded spin echo EPI sequence. The purpose is to make the distortion directions of the odd-echo chain image and the even-echo chain image opposite, providing complementary information.

[0070] The phase-encoded wraparound design, by adding a phase-encoded wraparound gradient before the phase encoding of all echo chains except the first echo chain, ensures that the echo signals of each echo chain in the multi-echo chain bipolar phase-encoded spin-echo EPI sequence fill the same position in its respective k-space. Specifically, the area of ​​the phase-encoded wraparound gradient is the sum of the phase-encoded gradients, and its direction is opposite to the phase-encoded direction. Due to the presence of the additional phase-encoded wraparound gradient, existing image reconstruction methods cannot be used; therefore, a deep neural network is introduced for image reconstruction.

[0071] See Figure 2 As shown, in step S2, the training samples for generating the deep neural network are specifically as follows.

[0072] S21: Obtain the registered magnetic resonance T2-weighted image (T2w) and proton density-weighted image (PDw) from the public dataset, obtain the magnetic resonance parameter map using the Bloch formula, and use the magnetic resonance parameter map as a virtual imaging object.

[0073] Specifically, the analytical form of the Bloch formula can be described as follows:

[0074]

[0075] Where S represents the signal intensity of the magnetic resonance image, x and y represent the coordinates of the two-dimensional plane, 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; and T1 represents the longitudinal relaxation time.

[0076] PDw images are magnetic resonance images acquired under short TE (TE≤10ms) and long TR (TR≥8s). In this case, the analytical form of the Bloch formula can be simplified to:

[0077] S(x,y)=PD(x,y)

[0078] Therefore, the image obtained after normalizing the signal intensity of the PDw image is used as the PD map of the virtual imaging object;

[0079] T2w images are magnetic resonance images acquired under appropriate TE (60ms ≤ TE ≤ 100ms) and long TR (TR ≥ 8s). In this case, the analytical form of the Bloch formula can be simplified to:

[0080]

[0081] Therefore, the corresponding T2 can be obtained from the signal strength, TE and PD of the T2w image. All T2 values ​​on the two-dimensional plane constitute the T2 map. The T2 values ​​of the T2 map are scaled to a range that is suitable for the actual imaging object. The resulting T2 map is used as the T2 map of the virtual imaging object.

[0082] The T2 image and PD image are combined to form a virtual imaging object.

[0083] S22: Model the inhomogeneity of the main magnetic field to obtain the model of the inhomogeneity of the main magnetic field.

[0084] Specifically, the inhomogeneity of the main magnetic field is modeled as follows:

[0085]

[0086] The main magnetic field inhomogeneity model dB0 is represented as the sum of a two-dimensional polynomial function and a two-dimensional Gaussian function. This represents a two-dimensional polynomial function, where x and y represent coordinates in the two-dimensional plane, and n... x and n y Let r represent the orders of x and y, respectively. p N represents the coefficients of a randomly generated two-dimensional polynomial function. p Let G denote the highest order of the two-dimensional polynomial function. Let G denote the two-dimensional Gaussian function, and N... g r represents the number of two-dimensional Gaussian functions. m and r s This represents the mean and standard deviation of a randomly generated two-dimensional Gaussian function.

[0087] S23: Generate randomly distributed two-dimensional Gaussian noise to simulate the noise present in the actual sampling process of magnetic resonance images.

[0088] S24: In the Bloch simulation, under the conditions of the main magnetic field inhomogeneity model and the two-dimensional Gaussian noise, the virtual imaging object is acquired using the multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain a distorted simulation sample.

[0089] S25: If the desired distortion-free image is a magnetic resonance weighted image, then in the Bloch simulation, the multi-echo chain bipolar phase-coded spin echo EPI sequence is used to acquire data from the virtual imaging object to obtain a distortion-free simulation sample as the distortion-free target sample; if the desired distortion-free image is a magnetic resonance parameter map, then the virtual imaging object is used as the distortion-free target sample.

[0090] S26: Repeat the process from S21 to S25 until a set number of deep neural network training samples are generated.

[0091] The process of the spin echo EPI image distortion correction method will be described in detail below through a specific embodiment, including the following steps.

[0092] Step 1: Design a multi-echo chain bipolar phase-coded spin-echo EPI sequence. The design of a multi-echo chain bipolar phase-coded spin-echo EPI sequence can be referenced... Figure 3The figure shows two multi-echo chain EPI sequences with different radio frequency pulse excitation forms: (a) represents a multi-echo chain bipolar phase-coded spin echo EPI sequence, and (b) represents a multi-echo chain bipolar phase-coded multi-overlapping echo EPI sequence. Both (a) and (b) generate spin echo signals after the radio frequency pulse is applied, and EPI technology is used for signal readout, which constitutes the spin echo EPI sequence described in this invention. To achieve image distortion correction using only a single scan of data, bipolar phase coding and phase coding wraparound gradients are introduced into the spin echo EPI sequence, thus obtaining images with different distortion directions in the phase coding direction. Deep neural networks can utilize these characteristics to correct EPI images.

[0093] Step 2: Generate training samples for the deep neural network. For spin-echo EPI sequences with different RF pulse excitation modes, corresponding training samples need to be generated. The training samples for the deep neural network can be found in [reference needed]. Figure 4 . Figure 4 Examples of training samples required for the deep neural network in this invention are provided below. For multi-echo chain bipolar phase-coded spin echo (EPI) sequences: Distorted simulation samples are obtained by using this sequence to acquire data from virtual imaging objects in Bloch simulations with a main magnetic field inhomogeneity model and two-dimensional Gaussian noise. Distorted simulation samples are then obtained by using this sequence to acquire data from virtual imaging objects in Bloch simulations, serving as distortion-free target samples. The distorted simulation samples are used as input to the deep neural network, and the distortion-free target samples are used as labels for the deep neural network. For multi-echo chain bipolar phase-coded multi-overlapping echo (EPI) sequences: Distorted simulation samples are obtained by using this sequence to acquire data from virtual imaging objects in Bloch simulations with a main magnetic field inhomogeneity model and two-dimensional Gaussian noise. The distorted simulation samples are used as input to the deep neural network, and the T2 map of the virtual imaging object is used as labels for the deep neural network.

[0094] Step 3: Train the deep neural network. Use the training samples generated in Step 2 to train the deep neural network.

[0095] Step 4: Use the multi-echo chain bipolar phase-coded spin echo EPI sequence designed in Step 1 to acquire data of the actual imaging object and obtain a distorted EPI image of the actual imaging object.

[0096] Step 5: Use the deep neural network trained in Step 3 to correct the distortion of the EPI image acquired in Step 4.

[0097] Images acquired using actual multi-echo chain bipolar phase-coded spin-echo EPI sequences and images after distortion correction using the spin-echo EPI image distortion correction method described in this embodiment can be referenced. Figure 5 .

[0098] In this embodiment, in step 1, the number of echo chains in the multi-echo chain bipolar phase-coded spin echo EPI sequence is 2.

[0099] In step 2, the preferred acquisition parameter values ​​for the T2w and PDw images are as follows: the TE range for the T2w image is [60ms, 100ms], and the TR range is [8s, 10s]; the TE range for the PDw image is [4ms, 10ms], and the TR range is [8s, 10s].

[0100] In step 2, the magnetic resonance parameters are designed to closely match the parameter distribution of the actual imaged object. Preferred parameter values ​​are as follows: T2 values ​​are uniformly distributed between [20ms, 300ms], and PD values ​​are uniformly distributed between [0, 1].

[0101] In step 2, the range of the magnitude of the main magnetic field non-uniformity is [-200Hz, +200Hz]; the range of the two-dimensional Gaussian noise is [0dB, 60dB].

[0102] In step 3, the deep neural network used is U-Net.

[0103] In step 3, the training samples need to be normalized. Since there are two echo chains, each training sample has two image data points. Therefore, the maximum value of the first echo chain image in all samples is selected, and the two echo chain image data points of all samples are divided by this maximum value to achieve normalization.

[0104] In step 3, for multi-echo chain bipolar phase-coded spin echo (EPI) sequences, the input to the deep neural network is in the form of amplitude, i.e., the input contains two images; the output is two distortion-free EPI images. For multi-echo chain bipolar phase-coded multi-overlapping spin echo (EPI) sequences, the input to the deep neural network is in the form of separate real and imaginary parts, i.e., the input contains four images; the output is one T2 image.

[0105] In step 3, the L1 norm is used to constrain the training error of the network, the number of training samples is 3000, no blocks are used, and the number of iterations is 200000.

[0106] In step 4, the imaging parameters for the multi-echo chain bipolar phase-coded spin echo EPI sequence are: imaging field of view of 220×220mm. 2The slice thickness was 3 mm, the sampling matrix size was 96×96, and GRAPPA acceleration was used with a speedup factor of 2. The imaging parameters for the multi-echo chain bipolar phase-coded multi-overlap spin-echo EPI sequence were: imaging field of view 220×220 mm. 2 The layer thickness is 3 mm, the sampling matrix size is 128×128, and GRAPPA technology is used for acceleration, with a speedup factor of 2. Data acquisition of the imaging object is performed using a multi-echo chain bipolar phase-coded spin echo EPI sequence. The acquired images are then input into the deep neural network trained in step 3 to obtain distortion-free images.

[0107] To evaluate a spin-echo EPI image distortion correction method, this embodiment provides simulated samples of dual-echo chain bipolar phase-coded spin-echo EPI sequences and the distortion correction results after training a deep neural network, which are shown in [the provided text]. Figure 4 In the diagram, (a) represents the distortion-free target sample of the first echo chain generated in step 2, (b) represents the distorted simulated sample of the first echo chain generated in step 2, (c) represents the distortion-corrected image of the distorted simulated sample of the first echo chain obtained in step 5, (d) represents the distortion-free target sample of the second echo chain generated in step 2, (e) represents the distorted simulated sample of the second echo chain generated in step 2, and (f) represents the distortion-corrected image of the distorted simulated sample of the second echo chain obtained in step 5. A comparison of (a) and (c) with (d) and (e) shows that the present invention effectively corrects the distortion of EPI images and clearly restores image details. Quantitative evaluation revealed that, compared to the undistorted target sample, the uncorrected first echo chain image had a peak signal-to-noise ratio (PSNR) of 21.65 and a structural similarity (SSIM) of 0.65, while the corrected first echo chain image had a PSNR of 33.31 and a SSIM of 0.98. Similarly, compared to the undistorted target sample, the uncorrected second echo chain image had a PSNR of 24.32 and a SSIM of 0.54, while the corrected second echo chain image had a PSNR of 33.71 and a SSIM of 0.97. These data demonstrate the powerful distortion correction capability of this invention for spin-echo EPI images.

[0108] Figure 5The images shown are those acquired using a multi-echo chain bipolar phase-coded spin echo EPI sequence and those corrected by a deep neural network. (a) and (b) are multi-echo chain bipolar phase-coded spin echo EPI images; (e) and (f) are multi-echo chain bipolar phase-coded multi-overlapping spin echo EPI images; (c) and (d) are multi-echo chain spin echo EPI images after distortion correction using this method; and (g) is a T2 quantitative image obtained after distortion correction using the method of this embodiment. The white outlines represent the edge contours of the distortion-free magnetic resonance image; overlaying them on the image allows for comparison of image distortion. The results show that this invention can effectively correct distortion in spin echo and multi-overlapping spin echo EPI images. Each distortion correction operation takes only a few seconds, far faster than the 5-10 minutes reconstruction time of traditional algorithms.

[0109] See Figure 6 and Figure 7 As shown, this embodiment also discloses a spin-echo EPI image distortion correction system, comprising:

[0110] The pulse sequence design module 61 is used to design a multi-echo chain bipolar phase-coded spin echo EPI sequence; the spin echo EPI sequence is an arbitrary magnetic resonance imaging sequence that uses radio frequency pulses to generate spin echo signals and uses EPI technology for signal readout.

[0111] Training sample generation module 62 is used to generate training samples for a deep neural network; the training samples include paired distorted simulation samples and undistorted target samples; the distorted simulation samples serve as input to the deep neural network training samples, and the undistorted target samples serve as labels for the deep neural network training samples; specifically including:

[0112] The virtual imaging object determination unit 621 is used to obtain the registered magnetic resonance T2-weighted image T2w and proton density-weighted image PDw from the public dataset, obtain the magnetic resonance parameter map through the Bloch formula, and use the magnetic resonance parameter map as the virtual imaging object.

[0113] The main magnetic field inhomogeneity modeling unit 622 is used to model the main magnetic field inhomogeneity to obtain the main magnetic field inhomogeneity model.

[0114] The Gaussian noise generation unit 623 is used to generate randomly distributed two-dimensional Gaussian noise to simulate the noise present in the actual sampling process of magnetic resonance images;

[0115] A distortion sample generation unit 624 is used to generate distorted simulation samples. In the Bloch simulation, under the presence of the main magnetic field inhomogeneity model and the two-dimensional Gaussian noise, the multi-echo chain bipolar phase-coded spin echo EPI sequence is used to acquire data from the virtual imaging object to obtain distorted simulation samples.

[0116] The distortion-free target sample generation unit 625 is used to generate distortion-free target samples. If the desired distortion-free image is a magnetic resonance weighted image, the virtual imaging object is acquired using the multi-echo chain bipolar phase-coded spin echo EPI sequence in the Bloch simulation to obtain a distortion-free simulation sample as the distortion-free target sample. If the desired distortion-free image is a magnetic resonance parameter map, the virtual imaging object is used as the distortion-free target sample.

[0117] The repetitive processing unit 626 repeatedly executes the virtual imaging object determination unit, the main magnetic field inhomogeneity modeling unit, the Gaussian noise generation unit, the distorted sample generation unit, and the distortion-free target sample generation unit until a set amount of deep neural network training samples are generated.

[0118] Network training module 63 is used to train a deep neural network using the training samples to obtain a trained deep neural network;

[0119] The data acquisition module 64 is used to acquire data of the actual imaging object using the multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain a distorted image of the actual imaging object; wherein the distortion direction of the odd-numbered echo chain image is opposite to that of the even-numbered echo chain image.

[0120] The distortion correction module 65 is used to correct the distortion of the actual image of the object by using the trained deep neural network to obtain a distortion-free image.

[0121] A specific implementation of a spin-echo EPI image distortion correction system is described in this embodiment, which is the same as the spin-echo EPI image distortion correction method.

[0122] It should be understood that those skilled in the art can make improvements and modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for correcting distortion in spin-echo type EPI images, characterized in that, include: S1: Design a multi-echo chain bipolar phase-coded spin echo EPI sequence; the spin echo EPI sequence is an arbitrary magnetic resonance imaging sequence that uses radio frequency pulses to generate spin echo signals and uses EPI technology for signal readout. S2: Generate training samples for a deep neural network; the training samples include paired distorted simulation samples and undistorted target samples; the distorted simulation samples serve as input to the deep neural network training samples, and the undistorted target samples serve as labels for the deep neural network training samples; specifically including: S21: Obtain the registered magnetic resonance T2-weighted image (T2w) and proton density-weighted image (PDw) from the public dataset, obtain the magnetic resonance parameter map using the Bloch formula, and use the magnetic resonance parameter map as a virtual imaging object; S22: Model the inhomogeneity of the main magnetic field to obtain the model of the inhomogeneity of the main magnetic field; S23: Generate randomly distributed two-dimensional Gaussian noise to simulate the noise present in the actual sampling process of magnetic resonance images; S24: In the Bloch simulation, under the presence of the main magnetic field inhomogeneity model and the two-dimensional Gaussian noise, the multi-echo chain bipolar phase-coded spin echo EPI sequence is used to acquire data of the virtual imaging object to obtain a distorted simulation sample. S25: If the desired distortion-free image is a magnetic resonance weighted image, then in the Bloch simulation, the multi-echo chain bipolar phase-coded spin echo EPI sequence is used to acquire data from the virtual imaging object to obtain a distortion-free simulation sample as the distortion-free target sample; if the desired distortion-free image is a magnetic resonance parameter map, then the virtual imaging object is used as the distortion-free target sample. S26: Repeat the process from S21 to S25 until a set number of deep neural network training samples are generated; S3: Use the training samples to train a deep neural network to obtain a trained deep neural network; S4: Data acquisition of the actual imaging object is performed using the multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain a distorted image of the actual imaging object; wherein, the distortion direction of the odd-numbered echo chain image is opposite to that of the even-numbered echo chain image. S5: The trained deep neural network is used to correct the distortion of the actual image of the object to obtain a distortion-free image.

2. The spin-echo EPI image distortion correction method according to claim 1, characterized in that, In S1, a multi-echo chain bipolar phase-coded spin-echo type EPI sequence is designed, specifically including: Phase coding design: Modify the phase coding of the multi-echo chain spin echo EPI sequence so that the phase coding gradient directions of the odd and even echo chains are opposite, resulting in a multi-echo chain bipolar phase-coded spin echo EPI sequence. The phase-encoded wraparound design adds a phase-encoded wraparound gradient before the phase encoding of the echo chains except the first echo chain, ensuring that the echo signals of each echo chain in the multi-echo chain bipolar phase-encoded spin echo EPI sequence fill the same position in their respective k-space.

3. The spin-echo EPI image distortion correction method according to claim 1, characterized in that, In step S21, obtaining the magnetic resonance parameter map using the Bloch formula and using the magnetic resonance parameter map as a virtual imaging object specifically includes: The analytical form of the Bloch formula is expressed as follows: Where S represents the signal intensity of the magnetic resonance image, x and y represent the coordinates of the two-dimensional plane, 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; and T1 represents the longitudinal relaxation time. PDw images are magnetic resonance images acquired under short TE and long TR conditions. In this case, the analytical form of the Bloch formula simplifies to: S(x,y)=PD(x,y) Therefore, the image obtained after normalizing the signal intensity of the PDw image is used as the PD map of the virtual imaging object; T2w images are magnetic resonance images acquired under appropriate TE and long TR conditions. In this case, the analytical form of the Bloch formula simplifies to: Therefore, the corresponding T2 can be obtained from the signal strength, TE and PD of the T2w image. All T2 values ​​on the two-dimensional plane constitute the T2 map. The T2 values ​​of the T2 map are scaled to a range that is suitable for the actual imaging object. The resulting T2 map is used as the T2 map of the virtual imaging object. The T2 image and PD image are combined to form a virtual imaging object.

4. The spin-echo EPI image distortion correction method according to claim 1, characterized in that, In step S22, the inhomogeneity of the main magnetic field is modeled to obtain the main magnetic field inhomogeneity model, as follows: Among them, the main magnetic field inhomogeneity model dB0 is expressed as the sum of a two-dimensional polynomial function and a two-dimensional Gaussian function; The expression represents a two-dimensional polynomial function; x and y represent coordinates in the two-dimensional plane; n x and n y Let r represent the orders of x and y, respectively; p N represents the coefficients of a randomly generated two-dimensional polynomial function; p G represents the highest order of a two-dimensional polynomial function; G represents a two-dimensional Gaussian function; N g The number of two-dimensional Gaussian functions is represented by r; m and r s Let represent the mean and standard deviation of a randomly generated two-dimensional Gaussian function, respectively.

5. A spin-echo EPI image distortion correction system, characterized in that, include: The pulse sequence design module is used to design multi-echo chain bipolar phase-coded spin echo EPI sequences; the spin echo EPI sequence is an arbitrary magnetic resonance imaging sequence that uses radio frequency pulses to generate spin echo signals and uses EPI technology for signal readout. A training sample generation module is used to generate training samples for a deep neural network. The training samples include paired distorted simulation samples and undistorted target samples. The distorted simulation samples serve as input to the deep neural network training samples, and the undistorted target samples serve as labels for the deep neural network training samples. Specifically, it includes: The virtual imaging object determination unit is used to obtain the registered magnetic resonance T2-weighted image (T2w) and proton density-weighted image (PDw) from the public dataset, obtain the magnetic resonance parameter map through the Bloch formula, and use the magnetic resonance parameter map as the virtual imaging object. The main magnetic field inhomogeneity modeling unit is used to model the main magnetic field inhomogeneity and obtain the main magnetic field inhomogeneity model. A Gaussian noise generation unit is used to generate randomly distributed two-dimensional Gaussian noise to simulate the noise present in the actual sampling process of magnetic resonance images; A distortion sample generation unit is provided to generate distorted simulation samples. In the Bloch simulation, the virtual imaging object is acquired using the multi-echo chain bipolar phase-coded spin echo EPI sequence in the presence of the main magnetic field inhomogeneity model and the two-dimensional Gaussian noise, thereby obtaining distorted simulation samples. A distortion-free target sample generation unit is used to generate distortion-free target samples. If the desired distortion-free image is a magnetic resonance weighted image, the virtual imaging object is acquired using the multi-echo chain bipolar phase-coded spin echo EPI sequence in the Bloch simulation to obtain a distortion-free simulation sample as the distortion-free target sample. If the desired distortion-free image is a magnetic resonance parameter map, the virtual imaging object is used as the distortion-free target sample. The repetitive processing unit repeatedly executes the virtual imaging object determination unit, the main magnetic field inhomogeneity modeling unit, the Gaussian noise generation unit, the distorted sample generation unit, and the distortion-free target sample generation unit until a set amount of deep neural network training samples are generated. The network training module is used to train a deep neural network using the training samples to obtain a trained deep neural network. The data acquisition module is used to acquire data of the actual imaging object using the multi-echo chain bipolar phase-coded spin echo EPI sequence to obtain a distorted image of the actual imaging object; wherein the distortion direction of the odd-numbered echo chain image is opposite to that of the even-numbered echo chain image. The distortion correction module is used to correct the distortion of the actual image of the object by using the trained deep neural network to obtain a distortion-free image.

Images