Reconstruction method and system for magnetic resonance parameter quantitative imaging

Abstract

The present application relates to the field of magnetic resonance, and in particular to a reconstruction method and system for magnetic resonance parameter quantitative imaging, comprising the following steps: obtaining full-sampled K-space data; obtaining a corresponding reference full-sampled weighted image; obtaining training input data; obtaining a first weighted image based on the training input data input into a reconstruction module; optimizing the reconstruction module based on the calculation operation of the first weighted image and the reference full-sampled weighted image; obtaining a second weighted image based on the first weighted data input into a generation module; optimizing the generation module based on the calculation operation of the second weighted image and the reference full-sampled weighted image; obtaining a target neural network based on the simultaneous optimization of the operations of the reconstruction module and the generation module; obtaining target under-sampled K-space data and defining it as target input data; obtaining a target weighted image based on the target input data; obtaining a target quantitative parameter image corresponding to the target weighted image. The present application has the effect of realizing fast and accurate magnetic resonance parameter imaging.

Description

Technical Field

[0001] The present application relates to the field of magnetic resonance, and in particular to a reconstruction method and system for magnetic resonance parameter quantitative imaging. Background Art

[0002] As a complex non-invasive detection technology, magnetic resonance imaging technology can provide rich information about various tissues inside the human body, and display cross-sectional images of the anatomy and pathology of different human structures in different grayscales. Compared with CT imaging technology, magnetic resonance imaging technology has the advantages of no ionizing radiation and high soft tissue contrast. Different tissues of the human body can be distinguished by magnetic resonance imaging. The difference in image signals comes from different parameters of the tissue itself, such as longitudinal relaxation time (T1), transverse relaxation time (T2), and proton density (PD). In clinical applications, the parameter settings of magnetic resonance imaging include echo time (TE), repetition time (TR) and flip angle (FA). By setting different parameters, different image weightings can be obtained, such as T1-weighted or T2-weighted images. These weighted images can usually reflect the characteristics of many common diseases and can be used for disease diagnosis.

[0003] However, since quantitative imaging requires the acquisition of multiple weighted images, its imaging time is equal to the imaging time of a single scan multiplied by the number of contrasts that need to be acquired. A long single scan imaging time will cause motion artifacts in the image, and the patient's limb movement between multiple scans will cause displacement between different contrast-weighted images, resulting in errors in the fitted parameter values, interfering with the doctor's diagnosis, and doubling the total scanning time of parameter imaging. Summary of the invention

[0004] In order to achieve fast and accurate magnetic resonance parameter imaging, the present application provides a reconstruction method and system for magnetic resonance parameter quantitative imaging.

[0005] In a first aspect, the present application provides a reconstruction method for magnetic resonance parameter quantitative imaging, which adopts the following technical solution:

[0006] A reconstruction method for magnetic resonance parameter quantitative imaging, comprising the following steps:

[0007] Acquire full-sampled K-space data; acquire a reference full-sampled weighted image corresponding to the full-sampled K-space data, wherein the reference full-sampled weighted image includes a full-sampled weighted image corresponding to all contrasts; acquire partial under-sampled K-space data and define it as training input data, wherein the training input data is under-sampled K-space data corresponding to partial contrasts; acquire a first weighted image corresponding to the training input data based on the operation of inputting the training input data into a reconstruction module; optimize the reconstruction module based on the calculation operation of the first weighted image and the reference full-sampled weighted image; acquire a second weighted image corresponding to the first weighted image based on the operation of inputting the first weighted image into a generation module; optimize the generation module based on the calculation operation of the second weighted image and the reference full-sampled weighted image; acquire the corresponding target neural network based on the simultaneous optimization of the reconstruction module and the generation module, wherein the target neural network includes a reconstruction module and a generation module; acquire target under-sampled K-space data and define it as target input data, wherein the target under-sampled K-space data is partial contrast under-sampled K-space data of an image that needs to undergo magnetic resonance imaging; acquire a target weighted image based on the operation of inputting the target input data into a target neural network;

[0008] Obtain the target quantitative parameter image corresponding to the target weighted image.

[0009] By adopting the above technical solution, by adopting the above technical solution, the full-sampled K-space data on the magnetic resonance instrument is first obtained, and then the obtained full-sampled K-space data is inverse Fourier transformed to obtain the corresponding reference full-sampled weighted image. It should be noted that the reference full-sampled weighted image is a full-sampled weighted image corresponding to all contrasts; then, some under-sampled K-space data on the magnetic resonance instrument is obtained and defined as training input data. In addition, the training input data is under-sampled K-space data corresponding to some contrasts; then, the training input data is input into the reconstruction module of the convolutional neural network, and the reconstruction module outputs a first weighted image corresponding to the training input data, and calculates the first weighted image and the reference full-sampled weighted image, and optimizes the reconstruction module based on the calculation result; after the reconstruction module is optimized, the first weighted image is input into the generation module of the convolutional neural network, and then the generation module outputs a first weighted image corresponding to the first weighted image. A corresponding second weighted image is obtained, and the second weighted image and the reference fully sampled weighted image are calculated, and the generation module is optimized according to the calculation result; after the generation module is optimized separately, the reconstruction module and the generation module of the convolutional neural network are optimized simultaneously, and the corresponding neural network is obtained based on the completion of the simultaneous optimization of the reconstruction module and the generation module, wherein the neural network includes the reconstruction module and the generation module; after the target neural network is trained, the target under-sampled K-space data required for imaging on the magnetic resonance instrument can be obtained, and the target under-sampled K-space data is defined as the target input data, wherein the target input data is the partial contrast under-sampled K-space data of the image required for magnetic resonance imaging, and the target input data is input into the target neural network, and then the target neural network outputs the corresponding target weighted image, and then the target quantitative parameter image corresponding to the target weighted image can be obtained by nonlinear fitting.

[0010] In the entire above process, the reconstruction module and generation module of the convolutional neural network are optimized and trained, and then the target input data is input into the target neural network, and the required target quantitative parameter image is obtained through the output target weighted image. The reconstruction module based on the target neural network is used to restore the information loss caused by undersampling of K space. At the same time, the generation module of the convolutional neural network generates corresponding uncollected weighted images through part of the collected weighted images to deal with the information loss problem caused by undersampling in the time (contrast) dimension. By undersampling in both K space and time (contrast) dimensions, a higher acceleration factor can be achieved while ensuring the reconstruction quality, thereby realizing fast and accurate magnetic resonance parameter imaging.

[0011] Optionally, the step of calculating and optimizing the reconstruction module based on the first weighted image and the fully sampled weighted image further includes the following steps:

[0012] Retrieve a contrast fully sampled weighted image from a reference fully sampled weighted image, wherein the contrast fully sampled weighted image is a fully sampled weighted image corresponding to the contrast of the partially undersampled K-space data; calculate first MSE loss data corresponding to the contrast fully sampled weighted image and the first weighted image; if the first MSE loss data is greater than a first loss threshold, optimize the reconstruction module, otherwise execute the operation of the first weighted image data input generation module.

[0013] By adopting the above technical solution, the first MSE loss data between the contrast full-sampling weighted image and the first weighted image can be calculated by calling up the contrast full-sampling weighted image, and through the pre-set first loss threshold, when the first MSE loss data is greater than the first loss threshold, it means that the parameters of the reconstruction module need to continue to be adjusted, then the reconstruction module is optimized to adjust the parameters of the reconstruction module, otherwise it means that the parameters of the reconstruction module are appropriate; through the above process, the parameters of the reconstruction module of the convolutional neural network can be adjusted, and the reconstruction module parameters suitable for the corresponding magnetic resonance instrument can be optimized, which can more accurately restore the information loss caused by K-space undersampling.

[0014] Optionally, the step of optimizing the generation module based on the calculation operation of the second weighted image and the fully sampled weighted image further includes the following steps:

[0015] Retrieve the remaining fully sampled weighted image in the reference fully sampled weighted image, wherein the remaining fully sampled weighted image is the fully sampled weighted image corresponding to the remaining contrast after excluding the contrast fully sampled weighted image in the reference fully sampled weighted image; calculate the second MSE loss data corresponding to the remaining fully sampled weighted image and the second weighted image; if the second MSE loss data is greater than the second loss threshold, optimize the generation module, otherwise perform the operation of optimizing the reconstruction module and the generation module at the same time.

[0016] By adopting the above technical solution, the second MSE loss data between the remaining fully sampled weighted image and the second weighted image can be calculated by calling the remaining fully sampled weighted image, and through the pre-set second loss threshold, when the second MSE loss data is greater than the second loss threshold, it means that the parameters of the generation module need to be further adjusted, then the generation module is optimized to adjust the parameters of the generation module, otherwise it means that the parameters of the generation module are appropriate; through the above process, the parameters of the generation module of the convolutional neural network can be adjusted, and the generation module parameters suitable for the corresponding magnetic resonance instrument can be optimized, which can better deal with the information loss problem caused by undersampling in the time (contrast) dimension.

[0017] Optionally, the operation steps based on simultaneously optimizing the reconstruction module and the generation module further include the following steps:

[0018] Based on the operation of inputting the training input data into the reconstruction module, a corresponding first weighted image is obtained; based on the operation of inputting the first weighted image into the generation module, a corresponding second weighted image is obtained; based on the first weighted image and the second weighted image, corresponding first MSE loss data and corresponding second MSE loss data are obtained; if both the first MSE loss data is greater than the first loss threshold and the second MSE loss data is greater than the second loss threshold, the target neural network is determined, otherwise, the operation of simultaneously optimizing the reconstruction module and the generation module is performed.

[0019] By adopting the above technical solution, after the reconstruction module and the generation module are optimized separately, when the target neural network is applied, the reconstruction module and the generation module process the input data at the same time, so it is necessary to optimize the reconstruction module and the generation module at the same time, compare the first MSE loss data with the first loss threshold, and compare the second MSE loss data with the second loss threshold. When the first MSE loss data is greater than the first loss threshold and the second MSE loss data is greater than the second loss threshold, both comparisons are not valid. At this time, it means that the parameters of the reconstruction module and the generation module are appropriate. Otherwise, the reconstruction module and the generation module are optimized at the same time. Through the above process, the parameters that are adapted to the reconstruction module and the generation module can be adjusted at the same time, so as to achieve more accurate recovery of lost data in the two dimensions of K space and time (contrast).

[0020] Optionally, the operation steps based on simultaneously optimizing the reconstruction module and the generation module further include the following steps:

[0021] Based on the operation of inputting the training input data into the reconstruction module, the corresponding first weighted image is obtained; based on the operation of inputting the first weighted image into the generation module, the corresponding second weighted image is obtained; the total MSE loss data is obtained according to the first MSE loss data corresponding to the first weighted image and the corresponding second MSE loss data of the second weighted image; if the total MSE loss data is greater than the third loss threshold, the operation of simultaneously optimizing the reconstruction module and the generation module is performed, otherwise the neural network is determined.

[0022] By adopting the above technical solution, after the reconstruction module and the generation module are optimized separately, the first MSE loss data and the second MSE loss data are superimposed to obtain the total MSE loss data. If the total MSE loss data is greater than the third loss threshold, it means that the parameters of the reconstruction module and the generation module are appropriate, otherwise the reconstruction module and the generation module are optimized simultaneously. Through the above process, the parameters adapted to the reconstruction module and the generation module can be adjusted simultaneously, so as to achieve more accurate recovery of the lost data in the two dimensions of K space and time (contrast).

[0023] Optionally, after the step of obtaining the corresponding target neural network based on the simultaneous optimization of the reconstruction module and the generation module, the following steps are further performed:

[0024] Based on the simultaneous optimization of the reconstruction module and the generation module, a training neural network is obtained; based on the training input data input operation of the training neural network, a corresponding training weighted image is obtained; the training weighted image is subjected to nonlinear fitting to obtain a corresponding training quantitative parameter image; the reference full sampling weighted image is retrieved and subjected to nonlinear fitting to obtain a corresponding reference quantitative parameter image; the error data between the training quantitative parameter image and the reference quantitative parameter image is calculated; if the error data is greater than a preset threshold, the simultaneous optimization of the reconstruction module and the generation module is performed; otherwise, the corresponding training neural network is determined as a target neural network.

[0025] By adopting the above technical solution, a training neural network is first obtained, and then the training input data is input into the training neural network, and the training neural network outputs the corresponding training weighted image, and then the training weighted image at this time is matched with the corresponding training quantitative parameter image by nonlinear fitting, and then the error data between the training quantitative parameter image and the reference quantitative parameter image is calculated, and if the error data is greater than a preset threshold, it means that the parameters of the reconstruction module and the generation module of the training neural network need to be adjusted, and then the parameters are fine-tuned to optimize the reconstruction module and the generation module, otherwise the corresponding training neural network is determined as the target neural network; through the above process, after the error comparison of the corresponding weighted image, the training quantitative parameter image corresponding to the training weighted image is again compared, and the neural network can be further fine-tuned and optimized, so that the quantitative parameter image can be obtained more accurately.

[0026] Optionally, the step of obtaining a corresponding training weighted image based on the operation of inputting training input data into the training neural network further includes:

[0027] The training input data is input into the training neural network to obtain a first weighted image corresponding to the reconstruction module and a second weighted image corresponding to the generation module; the first weighted image and the second weighted image are superimposed to obtain a training weighted image corresponding to all contrasts.

[0028] By adopting the above technical solution, by superimposing the first weighted image output by the reconstruction module and the second weighted image output by the generation module, a training weighted image corresponding to all required contrasts can be obtained, so as to obtain a training quantitative parameter image corresponding to the training weighted image.

[0029] In a second aspect, the present application provides a reconstruction system for magnetic resonance parameter quantitative imaging, which adopts the following technical solution:

[0030] A reconstruction system for magnetic resonance parameter quantitative imaging, comprising:

[0031] A full sampling acquisition module is used to acquire full sampling K space data; a reference full sampling acquisition module is used to acquire a reference full sampling weighted image corresponding to the full sampling K space data, wherein the reference full sampling weighted image includes a full sampling weighted image corresponding to all contrasts; a training input acquisition module is used to acquire part of the under-sampled K space data and define it as training input data, wherein the training input data is the under-sampled K space data corresponding to part of the contrast; a first acquisition module is used to acquire a first weighted image corresponding to the training input data based on the operation of inputting the training input data into the reconstruction module; a first optimization module is used to optimize the reconstruction module based on the calculation operation of the first weighted image and the full sampling weighted image; a second acquisition module is used to acquire a first weighted image based on the operation of inputting the first weighted image into the generation module a second weighted image corresponding to the weighted image; a second optimization module, which is used to optimize the generation module based on the calculation operation of the second weighted image and the full-sampled weighted image; a network acquisition module, which is used to acquire the corresponding target neural network based on the operation of simultaneously optimizing the reconstruction module and the generation module, wherein the target neural network includes a reconstruction module and a generation module; a target acquisition module, which is used to acquire target under-sampled K-space data and define it as target input data, wherein the target under-sampled K-space data is partial contrast under-sampled K-space data of an image that needs to undergo magnetic resonance imaging; a target weighted image acquisition module, which is used to acquire a target weighted image based on the operation of inputting the target input data into the target neural network; a target quantitative parameter image acquisition module, which is used to acquire a quantitative parameter image corresponding to the target weighted image.

[0032] By adopting the above technical solution, the full sampling acquisition module first acquires the full sampling K space data on the magnetic resonance instrument, and then performs inverse Fourier transform on the acquired full sampling K space data with reference to the full sampling acquisition module to obtain the corresponding reference full sampling weighted image. It should also be pointed out that the reference full sampling weighted image is the full sampling weighted image corresponding to all contrasts; then the training input acquisition module acquires part of the under-sampled K space data on the magnetic resonance instrument and defines it as training input data. In addition, the training input data is the under-sampled K space data corresponding to part of the contrast; then the training input data is input into the reconstruction module of the convolutional neural network, the first acquisition module acquires the first weighted image output by the reconstruction module corresponding to the training input data, the first optimization module calculates the first weighted image with the reference full sampling weighted image, and the first optimization module optimizes the reconstruction module according to the calculation result; after the reconstruction module is optimized, the first weighted image is input into the generation module of the convolutional neural network, and then the second acquisition module acquires the output of the generation module corresponding to the first weighted image A second weighted image, a second optimization module calculates the second weighted image and a reference fully sampled weighted image, and the second optimization module optimizes the generation module according to the calculation result; after the generation module is optimized separately, the reconstruction module and the generation module of the convolutional neural network are optimized simultaneously, and based on the completion of the simultaneous optimization of the reconstruction module and the generation module, the network acquisition module acquires the corresponding neural network, wherein the neural network includes the reconstruction module and the generation module; after the neural network is trained, the target acquisition module can acquire the target under-sampled K-space data required for imaging on the magnetic resonance instrument, and define the target under-sampled K-space data as target input data, wherein the target input data is the partial contrast under-sampled K-space data of the image required for magnetic resonance imaging, and the target input data is input into the target neural network, and then the target weighted image acquisition module acquires the target weighted image corresponding to the output of the target neural network, and then the target quantitative parameter image corresponding to the target weighted image can be acquired by the target weighted image acquisition module in a nonlinear fitting manner.

[0033] Through the optimization training of the reconstruction module and generation module of the convolutional neural network, the target input data is then input into the target neural network, and the required target quantitative parameter image is obtained through the output target weighted image. The reconstruction module based on the target neural network is used to restore the information loss caused by undersampling of K space. At the same time, the generation module of the convolutional neural network generates corresponding uncollected weighted images through part of the collected weighted images to deal with the information loss problem caused by undersampling in the time (contrast) dimension. By undersampling in both K space and time (contrast) dimensions, a higher acceleration factor can be achieved while ensuring the reconstruction quality, thereby realizing fast and accurate magnetic resonance parameter imaging.

[0034] In a third aspect, the present application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and execute any one of the reconstruction methods for magnetic resonance parameter quantitative imaging.

[0035] In a fourth aspect, the present application provides an intelligent terminal, comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes any one of the reconstruction methods for magnetic resonance parameter quantitative imaging.

[0036] In summary, the present application includes at least one of the following beneficial technical effects:

[0037] 1. Through the optimization training of the reconstruction module and generation module of the convolutional neural network, the target input data is input into the target neural network, and the required target quantitative parameter image is obtained through the output target weighted image. The reconstruction module based on the target neural network is used to restore the information loss caused by undersampling of K space. At the same time, the generation module of the convolutional neural network generates the corresponding uncollected weighted images through part of the collected weighted images to deal with the information loss problem caused by undersampling in the time (contrast) dimension. By undersampling in both K space and time (contrast) dimensions, a higher acceleration factor can be achieved while ensuring the reconstruction quality, thereby realizing fast and accurate magnetic resonance parameter imaging.

[0038] 2. The parameters of the reconstruction module of the convolutional neural network can be adjusted to optimize the reconstruction module parameters suitable for the corresponding magnetic resonance instrument, which can more accurately restore the information loss caused by undersampling of the K space; the parameters of the generation module of the convolutional neural network can be adjusted to optimize the generation module parameters suitable for the corresponding magnetic resonance instrument, which can better deal with the information loss problem caused by undersampling in the time (contrast) dimension.

[0039] 3. After the error comparison of the corresponding weighted image, the training quantitative parameter image corresponding to the training weighted image can be further fine-tuned and optimized for the neural network, so that the quantitative parameter image can be obtained more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is an overall flow chart of the reconstruction method for magnetic resonance parameter quantitative imaging in an embodiment of the present application;

[0041] Figure 2 It is a specific flow chart of the optimization reconstruction module in the reconstruction method for magnetic resonance parameter quantitative imaging according to the embodiment of the present application;

[0042] Figure 3 is a function diagram corresponding to the first MSE loss data in the reconstruction method for magnetic resonance parameter quantitative imaging according to an embodiment of the present application;

[0043] Figure 4 It is a specific flow chart of the optimization generation module in the reconstruction method for magnetic resonance parameter quantitative imaging according to the embodiment of the present application;

[0044] Figure 5 is a function diagram corresponding to the second MSE loss data in the reconstruction method for magnetic resonance parameter quantitative imaging according to an embodiment of the present application;

[0045] Figure 6 It is a specific flow chart of the first method for simultaneously optimizing the reconstruction module and the production module in the reconstruction method for magnetic resonance parameter quantitative imaging in an embodiment of the present application;

[0046] Figure 7 It is a specific flow chart of the second method for simultaneously optimizing the reconstruction module and the production module in the reconstruction method for magnetic resonance parameter quantitative imaging in an embodiment of the present application;

[0047] Figure 8 is a corresponding function diagram of total MSE loss data in the reconstruction method for magnetic resonance parameter quantitative imaging according to an embodiment of the present application;

[0048] Fig. 9 It is a specific flowchart of the reconstruction method for magnetic resonance parameter quantitative imaging according to an embodiment of the present application;

[0049] Fig.10 It is an overall flow chart of the reconstruction system for magnetic resonance parameter quantitative imaging according to an embodiment of the present application.

[0050] Description of reference numerals:

[0051] 1. Full sampling acquisition module; 2. Reference full sampling acquisition module; 3. Training input acquisition module; 4. First acquisition module; 5. First optimization module; 6. Second acquisition module; 7. Second optimization module; 8. Network acquisition module; 9. Target acquisition module; 10. Target weighted image acquisition module; 11. Target quantitative parameter image acquisition module. DETAILED DESCRIPTION

[0052] The following is a further detailed description of this application.

[0053] Magnetic resonance quantitative imaging technology can directly obtain the quantitative relaxation parameter value of voxels by changing specific imaging parameters, collecting a series of images with different contrast weightings, and combining nonlinear fitting algorithms. Quantitative imaging can reduce the differences in the results of different devices and provide a basis for disease classification and prognosis assessment, so it has very important clinical and research value; the existing mainstream magnetic resonance rapid imaging technology can be divided into the following four types in principle:

[0054] (1) Fast imaging sequence: Fast magnetic resonance imaging is performed by designing pulse sequences with shorter acquisition times, such as fast spin echo sequence, gradient echo sequence, and planar echo imaging. Although such fast imaging sequences improve the scanning speed to a certain extent, they still cannot meet the speed requirements of magnetic resonance parameter imaging.

[0055] (2) Parallel imaging technology. The principle of this technology is to collect data in parallel through multiple coil arrays and decode them according to information such as coil sensitivity, so as to achieve the purpose of rapid acquisition. Although parallel imaging can significantly shorten the imaging time of magnetic resonance imaging, as the number of coil arrays increases, the signal-to-noise ratio of magnetic resonance images will also decrease. How to balance the signal-to-noise ratio and imaging time of the reconstructed image is also a difficult problem that needs to be solved urgently. Therefore, parallel imaging technology cannot effectively solve the problem of very long magnetic resonance parameter imaging time.

[0056] (3) Fast imaging technology based on compressed sensing. The principle of this method is to achieve sparse sampling of the magnetic resonance K space by designing a highly random undersampling matrix, and then use nonlinear reconstruction under certain constraints to obtain high-quality magnetic resonance images. However, this method requires the use of sparse transformations such as wavelet functions and total variation, and the expression ability of this sparse basis is limited. It has poor expression ability for some small structures in magnetic resonance images, and image blurring and detail loss are more serious at higher acceleration factors. In addition, there are disadvantages in the compressed sensing fast reconstruction method, such as difficult parameter optimization and long reconstruction time, which also hinders its large-scale clinical application.

[0057] (4) Rapid imaging technology based on deep learning. This method uses a training set to learn the mapping of a K-space zero-filled under-sampled image to the corresponding full-sampled image or the corresponding fully sampled image fitted to obtain a parameter image, thereby achieving faster magnetic resonance reconstruction. This method has a faster reconstruction speed and avoids the problem of difficult parameter optimization in compressed sensing methods. In recent years, it has gradually become a popular method for rapid magnetic resonance imaging.

[0058] In magnetic resonance parametric imaging, the total scanning time is equal to the product of a single scanning time and the number of contrasts to be collected, which causes the total scanning time to increase exponentially. Therefore, in magnetic resonance parametric imaging applications, a higher acceleration factor is required to achieve an acceptable scanning time. However, under the existing fast reconstruction algorithm, an excessively high acceleration factor will cause more serious artifacts and reduce image quality. In order to achieve fast and accurate magnetic resonance parametric imaging, the present application embodiment discloses a reconstruction method for magnetic resonance parameter quantitative imaging, referring to Figure 1 , including the following steps:

[0059] S1, acquiring full-sample K-space data on the magnetic resonance instrument;

[0060] The full sampling space data includes K-space data of all contrasts.

[0061] S2, based on the fully sampled K-space data, a corresponding reference fully sampled weighted image is obtained by inverse Fourier transform;

[0062] The reference fully sampled weighted image includes a fully sampled weighted image corresponding to all contrasts of magnetic resonance imaging performed on a magnetic resonance instrument.

[0063] S3, obtaining part of the under-sampled K-space data and defining it as training input data;

[0064] The acquired K-space data is subjected to variable-density Poisson undersampling to obtain undersampled K-space data, and the training input data is the undersampled K-space data corresponding to a portion of the contrast.

[0065] S4, inputting the training input data into the reconstruction module, so as to obtain a first weighted image corresponding to the training input data output by the reconstruction module;

[0066] The reconstruction module may be any deep learning magnetic resonance reconstruction network, the first weighted image is a portion of a reference fully sampled weighted image, and the contrast of the first weighted image is the same as the contrast of the training input data.

[0067] S41, optimizing the reconstruction module based on the calculation operation of the first weighted image and the reference fully sampled weighted image, referring to Figure 1 The specific process includes the following steps:

[0068] S411, retrieve a comparison fully sampled weighted image from the reference fully sampled weighted image;

[0069] Wherein, the contrast fully sampled weighted image is a fully sampled weighted image corresponding to the contrast of the partially undersampled K-space data;

[0070] S412, calculating and comparing first MSE loss data corresponding to the fully sampled weighted image and the first weighted image;

[0071] Among them, MSE loss is the mean square error function. In this embodiment, the function image of the first MSE loss data is as follows: Figure 3 As shown, Loss1 is the first MSE loss data, and the image of the first MSE loss function gradually tends to be flat, which means that the error between the fully sampled weighted image and the first weighted image is small.

[0072] S413, calling a preset first loss threshold, and comparing the first MSE loss data with the first loss threshold, if the first MSE loss data is greater than the first loss threshold, optimizing the reconstruction module and executing step S4, otherwise executing step S5;

[0073] Among them, in the process of optimizing the reconstruction module, the Adam optimizer is used for optimization, but it is not limited to the Adam optimizer, and the stochastic gradient descent method and the AdaGrad algorithm can also be used for optimization; in this embodiment, the optimization of the reconstruction module is achieved by adjusting the parameters of the reconstruction module.

[0074] S5, inputting the first weighted image into a generation module, so as to obtain a second weighted image corresponding to the first weighted image output by the generation module;

[0075] Among them, the first weighted image is the output of the reconstruction module after the training input data output is optimized separately, the generation module can be any deep learning magnetic resonance reconstruction network, the second weighted image is a part of the reference full-sampling weighted image, and the contrast of the second weighted image is the remaining contrast of all contrasts corresponding to the reference full-sampling weighted image excluding the contrast corresponding to the training input data; for example, if there are 5 contrasts corresponding to the reference full-sampling weighted image (the first, second, third, fourth, and fifth), and the contrast corresponding to the training input data includes the first contrast and the fifth contrast, then the contrast of the first weighted image includes the first contrast and the fifth contrast, and the contrast of the second weighted image includes the second contrast, the third contrast, and the fourth contrast.

[0076] S51, optimizing the generation module based on the calculation operation of the second weighted image and the reference fully sampled weighted image; Figure 4 The specific process includes the following steps:

[0077] S511, retrieve the remaining fully sampled weighted image in the reference fully sampled weighted image, wherein the remaining fully sampled weighted image is the fully sampled weighted image corresponding to the remaining contrast after excluding the contrast fully sampled weighted image in the reference fully sampled weighted image, that is, the remaining fully sampled weighted image includes the fully sampled weighted images corresponding to the second contrast, the third contrast, and the fourth contrast.

[0078] S512, calculating the second MSE loss data corresponding to the remaining fully sampled weighted image and the second weighted image;

[0079] The second MSE loss data is close to the first MSE loss data. The MSE loss is a mean square error function. In this embodiment, the function graph of the second MSE loss data is as follows: Figure 5As shown, Loss2 is the second MSE loss data, and the image of the second MSE loss function gradually tends to be flat, which means that the error between the remaining fully sampled weighted image and the second weighted image is small.

[0080] S513, retrieve the preset second loss threshold, and compare the second MSE loss data with the second loss threshold. If the second MSE loss data is greater than the second loss threshold, optimize the generation module and execute S5, otherwise execute S6.

[0081] Among them, in the process of optimizing the generation module, the Adam optimizer is also used for optimization. At the same time, it is not limited to the Adam optimizer, and the stochastic gradient descent method and the AdaGrad algorithm can also be used for optimization; in this embodiment, the optimization of the generation module is achieved by adjusting the parameters of the generation module.

[0082] S6, after the reconstruction module and the generation module are optimized separately, the reconstruction module and the generation module are optimized simultaneously, and a corresponding target neural network is obtained;

[0083] Among them, the target neural network includes a reconstruction module and a generation module; and the specific process of simultaneously optimizing the reconstruction module and the generation module includes the following two methods, and this method can select one or both of the methods.

[0084] Reference Figure 6 , the first method:

[0085] S61, based on the operation of inputting the training input data into the reconstruction module, a corresponding first weighted image is obtained; based on the operation of inputting the first weighted image into the generation module, a corresponding second weighted image is obtained.

[0086] S611, obtaining first MSE loss data corresponding to the first weighted image and second MSE loss data corresponding to the second weighted image;

[0087] Among them, the first MSE loss data and the second MSE loss data have the same meaning as described above.

[0088] S612, compare the first MSE loss data and the first loss threshold, and the second MSE loss data and the second loss threshold at the same time; if both the first MSE loss data is greater than the first loss threshold and the second MSE loss data is greater than the second loss threshold are not true, determine the target neural network and execute step S7; otherwise, execute step S6.

[0089] Reference Figure 7 , the second way:

[0090] S62, based on the operation of inputting the training input data into the reconstruction module, a corresponding first weighted image is obtained, and based on the operation of inputting the first weighted image into the generation module, a corresponding second weighted image is obtained.

[0091] S621, obtaining total MSE loss data according to first MSE loss data corresponding to the first weighted image and second MSE loss data corresponding to the second weighted image;

[0092] Among them, the total MSE loss data = the first MSE loss data + the second MSE loss data, among which the function graph of the total MSE loss data refers to Figure 8 , Loss1+ Loss2 is the total MSE loss data.

[0093] S622, compare the total MSE loss data with the third loss threshold. If the total MSE loss data is greater than the third loss threshold, it means that when the optimization reconstruction module and the generation module are used at the same time, it is also necessary to adjust the parameters of the reconstruction module and the generation module for optimization, then execute step S6; otherwise, determine the target neural network and execute step S7.

[0094] After obtaining the corresponding target neural network based on the operation of simultaneously optimizing the reconstruction module and the generation module, refer to Fig. 9 , and also perform the following steps:

[0095] S7, verifying the target neural network, based on the operation of simultaneously optimizing the reconstruction module and the generation module, first obtaining the training neural network;

[0096] Among them, training the neural network also includes optimized reconstruction modules and generation modules.

[0097] S71, acquiring a corresponding training weighted image based on the operation of inputting training input data into the training neural network;

[0098] The training weighted image includes a first weighted image output by the optimized reconstruction module and a second weighted image output by the optimized generation module.

[0099] S72, performing nonlinear fitting on the training weighted image to obtain a corresponding training quantitative parameter image.

[0100] S73, calling the reference fully sampled weighted image and performing nonlinear fitting, so as to obtain a reference quantitative parameter image corresponding to the reference fully sampled weighted image.

[0101] S74, performing error calculation on the training quantitative parameter image and the reference quantitative parameter image, so as to obtain error data between the training quantitative parameter image and the reference quantitative parameter image;

[0102] The calculation method of the error data is the same as that of the first MSE loss data and the second MSE loss data.

[0103] S75, compare the error data with a preset threshold. If the error data is greater than the threshold, it means that the parameters of the reconstruction module and the generation module of the training neural network need to be adjusted and optimized, then execute step S6; otherwise, the corresponding training neural network is determined as the target neural network.

[0104] In addition, step S71 also includes:

[0105] S711, inputting the training input data into the training neural network to obtain a first weighted image corresponding to the reconstruction module and a second weighted image corresponding to the generation module;

[0106] S712: Superimpose the first weighted image and the second weighted image to obtain a training weighted image corresponding to all contrasts.

[0107] S8, acquiring target under-sampled K-space data and defining it as target input data;

[0108] The target under-sampled K-space data is partial contrast under-sampled K-space data of an image that needs to be subjected to magnetic resonance imaging, and the target input data is also obtained by performing variable-density Poisson under-sampling.

[0109] S9, obtaining a target weighted image based on the operation of inputting the target input data into the target neural network;

[0110] The target weighted image is a fully sampled weighted image on a magnetic resonance instrument.

[0111] S10, acquiring a corresponding target quantitative parameter image based on the target weighted image;

[0112] The target quantitative parameter image is obtained by nonlinear fitting of the target weighted image.

[0113] The implementation principle of a reconstruction method for magnetic resonance parameter quantitative imaging in the embodiment of the present application is:

[0114] First, the full-sampled K-space data on the magnetic resonance instrument is acquired, and then the acquired full-sampled K-space data is inversely Fourier transformed to obtain the corresponding reference full-sampled weighted image, which is a full-sampled weighted image corresponding to all contrasts.

[0115] Subsequently, some under-sampled K-space data on the magnetic resonance apparatus is obtained and defined as training input data. In addition, the training input data is under-sampled K-space data corresponding to some contrasts. In this embodiment, an example is given that there are 5 total contrasts (the 1st, 2nd, 3rd, 4th, and 5th) corresponding to the fully sampled weighted image, and the contrasts corresponding to the training input data include the 1st contrast and the 5th contrast. Then, the contrast of the first weighted image includes the 1st contrast and the 5th contrast, and the contrast of the second weighted image proposed later includes the 2nd contrast, the 3rd contrast, and the 4th contrast. It should be pointed out that the scope of protection of this scheme is not limited to the 5 total contrasts corresponding to the fully sampled weighted image, and the situation where there are multiple contrasts corresponding to the fully sampled weighted image is within the protection scope.

[0116] Then the training input data is input into the reconstruction module of the convolutional neural network, the reconstruction module outputs the first weighted image corresponding to the training input data, and the first weighted image is calculated with the reference full sampling weighted image, and the reconstruction module is optimized according to the calculation result; and the reconstruction module is optimized for the calculation operation based on the first weighted image and the reference full sampling weighted image, and the specific process includes the following steps: calling the contrast full sampling weighted image in the reference full sampling weighted image; then calculating the first MSE loss data corresponding to the contrast full sampling weighted image and the first weighted image, when the image of the first MSE loss function gradually tends to be flat, it means that the error between the contrast full sampling weighted image and the first weighted image is small; then calling the preset first loss threshold, and comparing the first MSE loss data with the first loss threshold, if the first MSE loss data is greater than the first loss threshold, then optimizing the reconstruction module, and continuing to execute the step of inputting the training input data into the reconstruction module of the convolutional neural network, by repeating the above steps related to optimizing the reconstruction module, after the reconstruction module is completed, the reconstruction module parameters are suitable for the corresponding magnetic resonance instrument, and the information loss caused by K space undersampling can be more accurately restored; otherwise, the reconstruction module is preliminarily optimized, and the optimization generation module step is executed.

[0117] After the reconstruction module is optimized, the first weighted image is input into the generation module of the convolutional neural network, and then the generation module outputs a second weighted image corresponding to the first weighted image, and the second weighted image is calculated with the reference fully sampled weighted image, and the generation module is optimized according to the calculation result; and the specific process includes the following steps: calling the remaining fully sampled weighted image in the reference fully sampled weighted image, and the remaining fully sampled weighted image includes fully sampled weighted images corresponding to the second contrast, the third contrast, and the fourth contrast, and then calculating the second MSE loss data corresponding to the remaining fully sampled weighted image and the second weighted image, when the images of the second MSE loss function are gradually Gradually becoming flat, that is, it means that the error between the remaining fully sampled weighted image and the second weighted image is small; firstly, the pre-set second loss threshold is retrieved, and the second MSE loss data is compared with the second loss threshold. If the second MSE loss data is greater than the second loss threshold, the generation module is optimized, and the step of retrieving the remaining fully sampled weighted image in the reference fully sampled weighted image is continued, otherwise, the steps of optimizing the reconstruction module and the generation module are performed at the same time; by optimizing the generation module, the parameters of the generation module of the convolutional neural network can be adjusted, and the generation module parameters suitable for the corresponding magnetic resonance instrument can be optimized, which can more quickly recover the information loss problem caused by undersampling in the time (contrast) dimension.

[0118] After the generation module is optimized separately, the reconstruction module and the generation module of the convolutional neural network are optimized simultaneously, and the corresponding neural network is obtained based on the completion of the simultaneous optimization of the reconstruction module and the generation module; and the specific process includes the following two methods:

[0119] In the first method, the training input data is input into the reconstruction module to obtain the first weighted image, and then the first weighted image is input into the generation module to obtain the second weighted image; then the first MSE loss data and the second MSE loss data can be obtained at the same time; then the first MSE loss data and the first loss threshold, the second MSE loss data and the second loss threshold are compared at the same time; and if both the first MSE loss data is greater than the first loss threshold and the second MSE loss data is greater than the second loss threshold are not true, then the target neural network is determined and the step of verifying the target neural network is executed; otherwise, the step of simultaneously optimizing the input reconstruction module and the generation module is continued.

[0120] The second method is different from the first method in that the total MSE loss data is obtained based on the first MSE loss data and the second MSE loss data, and the total MSE loss data = the first MSE loss data + the second MSE loss data; then the third loss threshold is called, and the total MSE loss data is compared with the third loss threshold. If the total MSE loss data is greater than the third loss threshold, it means that when the optimized reconstruction module and the generation module are used at the same time, it is also necessary to adjust the parameters of the reconstruction module and the generation module at the same time, then the step of simultaneously optimizing the input reconstruction module and the generation module is executed; otherwise, the target neural network is determined, and the step of verifying the target neural network is executed.

[0121] By performing synchronous optimization operations on the reconstruction module and the generation module in the above two ways, the parameters adapted to the reconstruction module and the generation module can be adjusted simultaneously, thereby achieving more accurate recovery of lost data in the two dimensions of K space and time (contrast).

[0122] After the target neural network is trained, the step of verifying the target neural network is performed, and the specific process includes the following steps: based on the operation of simultaneously optimizing the reconstruction module and the generation module, the training neural network is first obtained; then based on the training input data, the operation of the training neural network is input, and then the first weighted image and the second weighted image are superimposed to obtain the training weighted image corresponding to all contrasts; then the training weighted image is subjected to nonlinear fitting to obtain the corresponding training quantitative parameter image; similarly, the reference full sampling weighted image is retrieved and the reference quantitative parameter image is obtained by nonlinear fitting; the error between the training quantitative parameter image and the reference quantitative parameter image is calculated to obtain the error data between the training quantitative parameter image and the reference quantitative parameter image; the error data is compared with a preset threshold value, if the error data is greater than the threshold value, it means that the parameters of the reconstruction module and the generation module of the training neural network need to be adjusted and optimized, and the step of simultaneously optimizing the reconstruction module and the generation module is performed; otherwise, the corresponding training neural network is determined as the target neural network.

[0123] Through the above verification process of the target neural network, after the error comparison of the corresponding weighted image, the training quantitative parameter image corresponding to the training weighted image is again used to further fine-tune and optimize the neural network, so that the obtained quantitative parameter image can be more accurate.

[0124] After determining the target neural network, the neural network can be applied to magnetic resonance imaging. First, the target under-sampled K-space data that needs to be imaged on the magnetic resonance instrument is obtained, and the target under-sampled K-space data is defined as the target input data. The target input data is input into the target neural network. The target neural network then outputs the corresponding target weighted image. Then, the target quantitative parameter image corresponding to the target weighted image can be obtained by nonlinear fitting.

[0125] In the entire above process, the reconstruction module and generation module of the convolutional neural network are optimized and trained, and then the target input data is input into the target neural network, and the required target quantitative parameter image is obtained through the output target weighted image. The reconstruction module based on the target neural network is used to restore the information loss caused by undersampling of K space. At the same time, the generation module of the convolutional neural network generates corresponding uncollected weighted images through part of the collected weighted images to deal with the information loss problem caused by undersampling in the time (contrast) dimension. By undersampling in both K space and time (contrast) dimensions, a higher acceleration factor can be achieved while ensuring the reconstruction quality, thereby realizing fast and accurate magnetic resonance parameter imaging.

[0126] In addition, the numbers of the steps in this embodiment are only for convenience of explanation and do not limit the execution order of the steps. In practical applications, the execution order of the steps can be adjusted as needed, or performed simultaneously. These adjustments or replacements are all within the scope of protection of the present invention.

[0127] In a second aspect, the present application provides a reconstruction system for magnetic resonance parameter quantitative imaging, referring to Fig.10 ,include:

[0128] Full sampling acquisition module 1, used for acquiring full sampling K-space data;

[0129] A reference full sampling acquisition module 2 is used to acquire a reference full sampling weighted image corresponding to the full sampling K-space data, wherein the reference full sampling weighted image includes a full sampling weighted image corresponding to all contrasts;

[0130] A training input acquisition module 3 is used to acquire part of the under-sampled K-space data and define it as training input data, wherein the training input data is the under-sampled K-space data corresponding to part of the contrast;

[0131] A first acquisition module 4 is used to acquire a first weighted image corresponding to the training input data based on the operation of the reconstruction module inputted with the training input data;

[0132] A first optimization module 5, for optimizing the reconstruction module based on the calculation operation of the first weighted image and the fully sampled weighted image;

[0133] A second acquisition module 6, used for acquiring a second weighted image corresponding to the first weighted image based on the operation of the first weighted image input generation module;

[0134] A second optimization module 7, for optimizing the generation module based on the calculation operation of the second weighted image and the fully sampled weighted image;

[0135] A network acquisition module 8 is used to acquire a corresponding target neural network based on the operation of simultaneously optimizing the reconstruction module and the generation module, wherein the target neural network includes the reconstruction module and the generation module;

[0136] A target acquisition module 9 is used to acquire target under-sampled K-space data and define it as target input data, wherein the target under-sampled K-space data is partial contrast under-sampled K-space data of an image to be subjected to magnetic resonance imaging;

[0137] A target weighted image acquisition module 10, which is used to acquire a target weighted image based on the operation of inputting target input data into a target neural network;

[0138] The target quantitative parameter image acquisition module 11 is used to acquire a quantitative parameter image corresponding to the target weighted image.

[0139] Although the steps in the flowcharts of the drawings in the specification of the present invention are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows; unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be performed in other orders.

[0140] The implementation principle of a reconstruction system for magnetic resonance parameter quantitative imaging in an embodiment of the present application is as follows: the full sampling acquisition module 1 first acquires the full sampling K space data on the magnetic resonance instrument, and then performs an inverse Fourier transform on the acquired full sampling K space data with reference to the full sampling acquisition module 2 to obtain the corresponding reference full sampling weighted image. It should also be noted that the reference full sampling weighted image is a full sampling weighted image corresponding to all contrasts; then the training input acquisition module 3 acquires part of the under-sampled K space data on the magnetic resonance instrument and defines it as training input data. In addition, the training input data is the under-sampled K space data corresponding to part of the contrast; then the training input data is input into the reconstruction module of the convolutional neural network, the first acquisition module 4 acquires the first weighted image output by the reconstruction module corresponding to the training input data, the first optimization module 5 calculates the first weighted image with the reference full sampling weighted image, and the first optimization module 5 optimizes the reconstruction module according to the calculation result; after the reconstruction module is optimized, the first weighted image is input into the generation module of the convolutional neural network, and then the second acquisition module 6 acquires the generation module output with reference to the weighted image. The second weighted image corresponding to the first weighted image, the second optimization module 7 calculates the second weighted image and the reference full-sampled weighted image, and the second optimization module 7 optimizes the generation module according to the calculation result; after the generation module is optimized separately, the reconstruction module and the generation module of the convolutional neural network are optimized simultaneously, and based on the completion of the simultaneous optimization of the reconstruction module and the generation module, the network acquisition module 8 acquires the corresponding neural network, wherein the neural network includes the reconstruction module and the generation module; after the neural network is trained, the target acquisition module 9 can acquire the target under-sampled K-space data required for imaging on the magnetic resonance instrument, and define the target under-sampled K-space data as target input data, wherein the target input data is the partial contrast under-sampled K-space data of the image required for magnetic resonance imaging, and the target input data is input into the target neural network, and then the target weighted image acquisition module 10 acquires the target weighted image corresponding to the output of the target neural network, and then the target quantitative parameter image corresponding to the target weighted image can be acquired by the target weighted image acquisition module 10 by nonlinear fitting.

[0141] In the entire above process, the reconstruction module and generation module of the convolutional neural network are optimized and trained, and then the target input data is input into the target neural network, and the required target quantitative parameter image is obtained through the output target weighted image. The reconstruction module based on the target neural network is used to restore the information loss caused by undersampling of K space. At the same time, the generation module of the convolutional neural network generates corresponding uncollected weighted images through part of the collected weighted images to deal with the information loss problem caused by undersampling in the time (contrast) dimension. By undersampling in both K space and time (contrast) dimensions, a higher acceleration factor can be achieved while ensuring the reconstruction quality, thereby realizing fast and accurate magnetic resonance parameter imaging.

[0142] In a third aspect, the present application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and execute any one of the reconstruction methods for magnetic resonance parameter quantitative imaging.

[0143] In a fourth aspect, the present application provides an intelligent terminal, comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed by any method in the reconstruction method for magnetic resonance parameter quantitative imaging.

[0144] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A reconstruction method for magnetic resonance parameter quantitative imaging, characterized in that: The following steps are involved: Acquire full-sampled K-space data; Acquire a reference fully sampled weighted image corresponding to the fully sampled K-space data, wherein the reference fully sampled weighted image includes a fully sampled weighted image corresponding to all contrasts; Acquire partial under-sampled K-space data and define it as training input data, wherein the training input data is the under-sampled K-space data corresponding to the partial contrast; Acquire a first weighted image corresponding to the training input data based on the operation of inputting the training input data into the reconstruction module; optimizing the reconstruction module based on computational operations of the first weighted image and the reference fully sampled weighted image; Acquire a second weighted image corresponding to the first weighted image based on the operation of the first weighted image input generation module; A computational operation optimization generation module based on the second weighted image and the reference fully sampled weighted image; Acquire a corresponding target neural network based on the operation of simultaneously optimizing the reconstruction module and the generation module, wherein the target neural network includes the reconstruction module and the generation module; Acquire target under-sampled K-space data and define it as target input data, wherein the target under-sampled K-space data is partial contrast under-sampled K-space data of an image to be subjected to magnetic resonance imaging; Obtaining a target weighted image based on an operation of inputting target input data into a target neural network; Obtain the target quantitative parameter image corresponding to the target weighted image.

2. The reconstruction method for magnetic resonance parameter quantitative imaging according to claim 1, characterized in that: The step of calculating and optimizing the reconstruction module based on the first weighted image and the fully sampled weighted image further includes the following steps: Retrieving a contrast fully sampled weighted image in the reference fully sampled weighted image, wherein the contrast fully sampled weighted image is a fully sampled weighted image corresponding to the contrast of the partial undersampled K-space data; Calculate and compare the first MSE loss data corresponding to the fully sampled weighted image and the first weighted image; If the first MSE loss data is greater than the first loss threshold, the reconstruction module is optimized, otherwise the operation of the first weighted image input generation module is executed.

3. The reconstruction method for magnetic resonance parameter quantitative imaging according to claim 1, characterized in that: In the step of optimizing the generation module based on the calculation operation of the second weighted image and the fully sampled weighted image, the following steps are also included: Retrieving a remaining fully sampled weighted image from the reference fully sampled weighted image, wherein the remaining fully sampled weighted image is a fully sampled weighted image corresponding to the remaining contrast after excluding the contrast fully sampled weighted image from the reference fully sampled weighted image; Calculate the second MSE loss data corresponding to the remaining fully sampled weighted image and the second weighted image; If the second MSE loss data is greater than the second loss threshold, the generation module is optimized; otherwise, the operations of optimizing the reconstruction module and the generation module are performed simultaneously.

4. The reconstruction method for magnetic resonance parameter quantitative imaging according to claim 3, characterized in that: The operation steps based on simultaneously optimizing the reconstruction module and the generation module further include the following steps: Based on the operation of inputting the training input data into the reconstruction module, a corresponding first weighted image is obtained; based on the operation of inputting the first weighted image into the generation module, a corresponding second weighted image is obtained; Acquire corresponding first MSE loss data and corresponding second MSE loss data according to the first weighted image and the second weighted image; If the first MSE loss data is greater than the first loss threshold and the second MSE loss data is greater than the second loss threshold, the target neural network is determined, otherwise the simultaneous optimization of the reconstruction module and the generation module operation is performed.

5. The reconstruction method for magnetic resonance parameter quantitative imaging according to claim 3, characterized in that: The operation steps based on simultaneously optimizing the reconstruction module and the generation module further include the following steps: Based on the operation of inputting the training input data into the reconstruction module, a corresponding first weighted image is obtained; based on the operation of inputting the first weighted image into the generation module, a corresponding second weighted image is obtained; Obtaining total MSE loss data based on first MSE loss data corresponding to the first weighted image and second MSE loss data corresponding to the second weighted image; If the total MSE loss data is greater than the third loss threshold, the reconstruction module and the generation module are optimized simultaneously, otherwise the target neural network is determined.

6. The reconstruction method for magnetic resonance parameter quantitative imaging according to claim 4 or 5, characterized in that: After the step of obtaining the corresponding target neural network based on the simultaneous optimization of the reconstruction module and the generation module, the following steps are further performed: A training neural network is obtained based on the operation of simultaneously optimizing the reconstruction module and the generation module; Based on the operation of inputting the training input data into the training neural network, a corresponding training weighted image is obtained; Perform nonlinear fitting on the training weighted image to obtain the corresponding training quantitative parameter image; Retrieving the reference full sampling weighted image and performing nonlinear fitting to obtain the corresponding reference quantitative parameter image; Calculating error data between the training quantitative parameter image and the reference quantitative parameter image; If the error data is greater than a preset threshold, the operation of optimizing the reconstruction module and the generation module is performed simultaneously; Otherwise, the corresponding training neural network is determined as the target neural network.

7. The reconstruction method for magnetic resonance parameter quantitative imaging according to claim 6, characterized in that: In the step of obtaining a corresponding training weighted image based on the operation of inputting training input data into the training neural network, the step further includes: Input the training input data into the training neural network to obtain a first weighted image corresponding to the reconstruction module and a second weighted image corresponding to the generation module; The first weighted image and the second weighted image are superimposed to obtain a training weighted image corresponding to all contrasts.

8. A reconstruction system for magnetic resonance parameter quantitative imaging, characterized in that: include: A full sampling acquisition module (1) is used to acquire full sampling K-space data; A reference full sampling acquisition module (2) is used to acquire a reference full sampling weighted image corresponding to the full sampling K-space data, wherein the reference full sampling weighted image includes a full sampling weighted image corresponding to all contrasts; A training input acquisition module (3) is used to acquire part of the under-sampled K-space data and define it as training input data, wherein the training input data is the under-sampled K-space data corresponding to part of the contrast; A first acquisition module (4) is used to acquire a first weighted image corresponding to the training input data based on the operation of the reconstruction module inputted with the training input data; A first optimization module (5) is used to optimize the reconstruction module based on the calculation operation of the first weighted image and the reference fully sampled weighted image; A second acquisition module (6) is used to acquire a second weighted image corresponding to the first weighted image data based on the operation of the first weighted image input generation module; A second optimization module (7) is used to optimize the generation module based on the calculation operation of the second weighted image and the reference fully sampled weighted image; A network acquisition module (8) is used to acquire a corresponding target neural network based on the operation of simultaneously optimizing the reconstruction module and the generation module, wherein the target neural network includes the reconstruction module and the generation module; A target acquisition module (9) is used to acquire target under-sampled K-space data and define it as target input data, wherein the target under-sampled K-space data is partial contrast under-sampled K-space data of an image to be subjected to magnetic resonance imaging; A target weighted image acquisition module (10) is used to acquire a target weighted image based on the operation of inputting target input data into a target neural network; The target quantitative parameter image acquisition module (11) is used to acquire a quantitative parameter image corresponding to the target weighted image.

9. A computer-readable storage medium storing a computer program capable of being loaded by a processor and executing the method according to any one of claims 1 to 7.

10. An intelligent terminal comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes any one of the methods of claims 1 to 7.

Images