Magnetic resonance image reconstruction method and device, computer equipment and storage medium

Abstract

The invention relates to a magnetic resonance image reconstruction method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a K space data set of a plurality of phases of at least one layer of a detection object, wherein the K space data set of the at least one layer is under-sampled; performing weighted combination on the K space data sets of the plurality of phase phases to generate a synthesized K space data set, and generating a coil sensitivity distribution diagram according to the synthesized K space data set; performing initialization reconstruction on the K space data sets of the plurality of phases to obtain initial dynamic magnetic resonance images; performing machine learning-based iterative reconstruction on the initial dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detected object; an iterative reconstruction of machine learning includes a coincidence term generated based on a K-space dataset of multiple phase phases and a coil sensitivity profile. By adopting the method, the reconstruction quality of the magnetic resonance image can be improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a magnetic resonance image reconstruction method, device, computer equipment and storage medium. Background Art

[0002] During the magnetic resonance imaging process, in order to ensure the speed of magnetic resonance imaging, multiple frames of magnetic resonance images are quickly acquired by undersampling, and then the undersampled multiple frames of magnetic resonance images are reconstructed to obtain reconstructed magnetic resonance images.

[0003] In the related art, data is mainly filled in other positions in the multiple-frame magnetic resonance images based on under-sampled data in the multiple-frame magnetic resonance images, and the multiple-frame magnetic resonance images after data filling are used as reconstructed magnetic resonance images.

[0004] However, the magnetic resonance images reconstructed by the related art methods have the problem of poor quality. Summary of the invention

[0005] Based on this, it is necessary to provide a magnetic resonance image reconstruction method, apparatus, computer equipment and storage medium to improve the reconstruction quality of magnetic resonance images in response to the above technical problems.

[0006] In a first aspect, the present application provides a magnetic resonance image reconstruction method, the method comprising:

[0007] Acquiring a K-space data set of multiple phases of at least one layer of the detection object, wherein the K-space data set of at least one layer is undersampled;

[0008] weightedly merging the K-space data sets of multiple phases to generate a synthetic K-space data set, and generating a coil sensitivity distribution map according to the synthetic K-space data set;

[0009] Initializing and reconstructing the K-space data sets of multiple phases respectively to obtain initial dynamic magnetic resonance images;

[0010] The initial dynamic magnetic resonance image is iteratively reconstructed based on machine learning to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction of machine learning includes consistent items generated based on K-space data sets of multiple phases and coil sensitivity distribution maps.

[0011] In one embodiment, at least two adjacent phases of K-space data sets are interleaved and collected. In one embodiment, initializing and reconstructing the K-space data sets of multiple phases to obtain the initial dynamic magnetic resonance image includes:

[0012] A non-machine learning method is used to restore the K-space data sets of multiple phases to obtain an intermediate K-space data set;

[0013] The coil sensitivity distribution map is used to reconstruct the intermediate K-space data set to obtain the initial dynamic magnetic resonance image.

[0014] In one of the embodiments, the machine learning based iterative reconstruction includes a proximal gradient descent algorithm.

[0015] In one of the embodiments, the iterative reconstruction based on machine learning includes a 3D convolution module and a 2D convolution module.

[0016] In one of the embodiments, the iterative reconstruction based on machine learning includes a three-dimensional reconstruction model and a two-dimensional reconstruction model;

[0017] The initial dynamic magnetic resonance image is iteratively reconstructed based on machine learning to obtain a dynamic magnetic resonance image of the detection object, including:

[0018] Reconstructing the initial dynamic magnetic resonance image using the three-dimensional reconstruction model to obtain an intermediate dynamic magnetic resonance image;

[0019] The two-dimensional reconstruction model is used to correct each phase magnetic resonance image in the intermediate dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object.

[0020] In a second aspect, the present application also provides a magnetic resonance image reconstruction device, the device comprising:

[0021] An acquisition module, used for acquiring a K-space data set of multiple phases of at least one layer of the detection object, wherein the K-space data set of at least one layer is undersampled;

[0022] A generating module, used for weightedly merging the K-space data sets of multiple phases to generate a synthetic K-space data set, and generating a coil sensitivity distribution map according to the synthetic K-space data set;

[0023] An initialization and reconstruction module is used to respectively initialize and reconstruct the K-space data sets of multiple phases to obtain an initial dynamic magnetic resonance image;

[0024] The iterative reconstruction module is used to perform iterative reconstruction based on machine learning on the initial dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction of machine learning includes consistent items generated based on K-space data sets of multiple phases and coil sensitivity distribution maps.

[0025] In a third aspect, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the content of any one of the embodiments of the magnetic resonance image reconstruction method in the first aspect is implemented.

[0026] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the content of any one of the embodiments of the magnetic resonance image reconstruction method in the first aspect when the computer program is executed by a processor.

[0027] In a fifth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the content of any one of the embodiments of the magnetic resonance image reconstruction method in the first aspect is implemented.

[0028] The magnetic resonance image reconstruction method, device, computer equipment and storage medium obtain K-space data sets of multiple phases of at least one layer of the detection object, and the K-space data set of at least one layer is undersampled; the K-space data sets of multiple phases are weighted and merged to generate a synthetic K-space data set, and a coil sensitivity distribution map is generated according to the synthetic K-space data set; the K-space data sets of multiple phases are respectively initialized and reconstructed based on the coil sensitivity distribution map to obtain an initial dynamic magnetic resonance image; the initial dynamic magnetic resonance image is iteratively reconstructed based on machine learning to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction of machine learning includes consistent items generated based on the K-space data sets of multiple phases and the coil sensitivity distribution map. The method reconstructs the K-space data sets of multiple phases in sequence through two reconstruction processes consisting of initialization and iterative reconstruction based on machine learning, which can more accurately complete the magnetic resonance image reconstruction process. At the same time, the iterative reconstruction based on machine learning includes consistent items generated based on the K-space data sets of multiple phases and the coil sensitivity distribution map. In the reconstruction process, it can ensure that the error between the reconstructed image and the K-space data sets of multiple phases is small, that is, it is closer to the real multi-phase magnetic resonance image of the detection object, which can improve the reconstruction quality of the magnetic resonance image. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A diagram showing an application environment of a magnetic resonance image reconstruction method in one embodiment;

[0030] Figure 2 is a schematic flow chart of a magnetic resonance image reconstruction method in one embodiment;

[0031] Figure 3 is a schematic flow chart of a magnetic resonance image reconstruction method in one embodiment;

[0032] Figure 4 is a schematic flow chart of a magnetic resonance image reconstruction method in one embodiment;

[0033] Figure 5 A schematic diagram of an iterative reconstruction model training process based on machine learning in one embodiment;

[0034] Figure 6 A schematic diagram of a network architecture of an iterative reconstruction model based on machine learning in one embodiment;

[0035] Figure 7 is a schematic flow chart of a magnetic resonance image reconstruction method in one embodiment;

[0036] Figure 8 is a schematic diagram of a K-space dataset mask for multiple phases in one embodiment;

[0037] Fig. 9 is a schematic diagram of a magnetic resonance image reconstruction result in one embodiment;

[0038] Fig.10 FIG. 4 is a structural block diagram of a magnetic resonance image reconstruction device in one embodiment. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0040] The image reconstruction method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. For example, the computer device may be a server, a personal computer, a laptop, a smart phone, a tablet computer, a smart mobile phone, etc. The computer device may include a processor, a memory and a network interface connected by a system bus or wirelessly. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device may include a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data during the magnetic resonance image reconstruction process. The network interface of the computer device is used to communicate with an external terminal through a network connection, and the computer program is executed by the processor to implement a magnetic resonance image reconstruction method. Among them, the computer device can be implemented by an independent computer device or a computer device cluster composed of multiple computer devices. It should be noted that the memory of the computer device is not limited to the above-mentioned memory, and may also include a high-speed random access memory, a volatile solid-state memory, etc. In addition, the composition architecture of the computer device is not limited to the above-mentioned situation, and some components may be added or omitted.

[0041] In one embodiment, Figure 2 As shown, a magnetic resonance image reconstruction method is provided, which is applied to Figure 1The computer device in the example is used to illustrate, including the following steps:

[0042] S201, obtaining a K-space data set of multiple phases of at least one layer of a detection object, wherein the K-space data set of at least one layer is under-sampled.

[0043] K space refers to the space filled with the original data of the magnetic resonance signal with spatial positioning coding information. It is an abstract three-dimensional space or a flat two-dimensional space. The magnetic resonance image is obtained by arranging it at a specific K space position according to different spatial frequencies. It should be noted that the sampling density of the K space area is greater than the sampling density of the K space peripheral area.

[0044] Undersampling refers to the method in which the magnetic resonance device uses a sampling frequency lower than the original signal bandwidth for sampling during the sampling process. The multi-phase 2D K-space data set is a K-space data set in a time series.

[0045] Optionally, the computer device can obtain a K-space dataset of multiple phases of at least one layer of the detection object from a picture archiving and communication system (PACS). For example, the computer device can search for a K-space dataset of multiple phases continuously collected at at least one layer of the detection object from the PACS based on the identification information of the detection object. Alternatively, the magnetic resonance imaging device can continuously collect a K-space dataset of multiple phases continuously collected at at least one layer of the detection object according to a preset sampling trajectory, and during the continuous collection process, send the collected K-space dataset of multiple phases continuously collected at at least one layer to the computer device. This embodiment does not limit the method for obtaining a K-space dataset of multiple phases of at least one layer of the detection object.

[0046] It should be noted that, among the above-mentioned multiple phases of K-space data sets, at least two adjacent phases of K-space data sets are interleaved, that is, the encoding positions of these two adjacent data are different, and the data of different phases represent the sampling data of different positions of the detection object. If the K-space data of adjacent phases are all collected in an interleaved manner, then after the K-space data sets of multiple phases are fused, the sampling data of most positions of the detection object are obtained, even if these data are not collected in the same phase.

[0047] S202 , weightedly merging the K-space data sets of multiple phases to generate a synthetic K-space data set, and generating a coil sensitivity distribution map according to the synthetic K-space data set.

[0048] In this embodiment, since the K-space data sets of multiple phases are all for the same part, the computer device can calculate the product of the weight of each phase and the corresponding K-space data according to the weight of each phase, and add the products of all phases to obtain the weighted combination result of the K-space data sets of multiple phases, and determine the weighted combination result as the synthetic K-space data set.

[0049] The coil sensitivity distribution diagram refers to a schematic diagram of the sensitivity of the receiving coil in the magnetic resonance device to magnetic resonance signals at different spatial positions. The sensitivity of the magnetic resonance image close to the receiving coil is high, and the sensitivity of the magnetic resonance image far from the receiving coil is low.

[0050] In this embodiment, the computer device can calculate the synthetic K-space data set according to the relevant calculation formula to obtain the coil sensitivity distribution map. The relevant calculation formula can be expressed as:

[0051] CSM=A / SOS(A)

[0052] Wherein, CSM represents the coil sensitivity distribution map; A represents the synthetic K-space data set; SOS represents the square sum root of the synthetic K-space data set.

[0053] The coil sensitivity distribution map can also be obtained in another way. When the K-space acquisition method is discrete acquisition, the phase reference line (Auto Calibration Signals, ACS) is first used to calculate the average value of the K-space data set of multiple phases, and the K-space center of the average value is used to calculate the corresponding coil sensitivity distribution map of the K-space data set of multiple phases.

[0054] S203, initializing and reconstructing the K-space data sets of multiple phases respectively to obtain an initial dynamic magnetic resonance image.

[0055] In this embodiment, for the K space data of any phase in the K space data set, the computer device can fill the K space data of the phase based on the K space data of other phases to obtain the filled K space data of the phase. According to the above method, the K space data of each phase is reconstructed to obtain an initial dynamic magnetic resonance image. Alternatively, the computer device can also initialize and reconstruct the K space data set of each phase in multiple phases according to a preset initialization reconstruction rule to obtain an initial dynamic magnetic resonance image. This embodiment does not limit the method of initialization reconstruction.

[0056] S204, performing iterative reconstruction based on machine learning on the initial dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction based on machine learning includes consistent items generated based on K-space data sets of multiple phases and coil sensitivity distribution maps.

[0057] The iterative reconstruction based on machine learning includes a 3D convolution module and a 2D convolution module. For example, the iterative reconstruction based on machine learning can be a proximal gradient descent algorithm.

[0058] In this embodiment, the computer device can use the initial dynamic magnetic resonance image as an input signal, input it into the iterative reconstruction model based on machine learning, and use the iterative reconstruction model based on machine learning to reconstruct the initial dynamic magnetic resonance image. In addition, since the iterative reconstruction of machine learning includes consistent items generated based on the K-space data set of multiple phases and the coil sensitivity distribution map. Therefore, during the reconstruction process, the computer device will compare the reconstructed magnetic resonance image with the corresponding K-space data set of multiple phases to ensure that the non-under-sampled part of the reconstructed image is consistent with the 2D magnetic resonance under-sampled image. In this way, the quality of the reconstructed dynamic magnetic resonance image is better.

[0059] Considering the problems of the direct inversion method of deep network, such as unstable training, focusing on post-processing in the image domain, ignoring the fidelity of the K-space domain, and easily generating unnecessary artifacts, etc. To address this problem, the iterative reconstruction method based on machine learning in this embodiment is used to gradually realize rapid imaging of magnetic resonance. The iterative reconstruction method based on machine learning can be expressed as:

[0060]

[0061] Wherein, x represents the magnetic resonance image obtained by iterative reconstruction based on machine learning at each update; represents the magnetic resonance image after iterative reconstruction; η represents the fidelity term coefficient; A represents the forward process of magnetic resonance undersampling; b represents the undersampling K-space data; R(x) is the prior constraint term or regular term for magnetic resonance rapid imaging.

[0062] The iterative reconstruction process based on machine learning is implemented based on the Projected Gradient Descent (PGD) algorithm. Expanding the above formula, we get:

[0063]

[0064] Next, the content of the PGD algorithm is introduced. The PGD algorithm can be divided into two steps: (1) data fidelity of the reconstructed magnetic resonance; (2) prior constraints on the reconstructed magnetic resonance. Among them, data fidelity is to strengthen the consistency of the data through methods such as approximation operators or data backfilling, so as to enhance the stability and robustness of the model framework. The purpose of the prior constraint is to accelerate the iterative convergence of the iterative reconstruction model based on machine learning and make the reconstruction result more consistent with the cognition of the detected object. This prior can be either specific or abstract. For example, the prior constraint can be represented by a convolutional neural network (CNN) to be trained.

[0065] In order to train and update the CNN part of the iterative reconstruction model based on machine learning, this embodiment adopts a model-driven approach to expand the traditional iterative loop process, namely the solution loop. The specific method is to encapsulate the above-mentioned PGD algorithm solution process into a solution loop module, which includes two parts: data fidelity and CNN calculation. The PGD algorithm needs to iterate k times during the solution process, that is, the solution loop model framework contains k solution loop modules. However, traditional model-driven algorithms are faced with the problem of selecting the number of iterative loops. If the number of iterations is selected too small, the solution loop will not be sufficient, and the image quality of the magnetic resonance reconstruction will be poor; if the number of iterations is selected too large, it will lead to slow training and memory explosion, affecting the efficiency of the magnetic resonance reconstruction process.

[0066] To solve the above problems, the Deep Equilibrium Model (DEQ) and the Find Fixed Point (FFP) algorithm are introduced into the iterative reconstruction model based on machine learning to realize the adaptive adjustment of the number of de-loop modules by the model framework. Among them, the DEQ model greatly increases the range of selection of the number of de-loop modules. The DEQ model allows the model framework to select and use an infinite number of de-loop modules, especially with the help of the "Jacobian-free" method, while also solving the problem of video memory explosion during training. To a certain extent, the introduction of the DEQ model improves the reconstruction capability of the model framework of the iterative reconstruction model based on machine learning.

[0067] Furthermore, on the basis of ensuring the reliability of the magnetic resonance reconstruction process, in order to improve the efficiency of magnetic resonance reconstruction, a scale estimation method (Fast Filter Processor, FFP) can be added on the basis of the DEQ model. The FFP algorithm is used to select the appropriate number of de-loop modules. The FFP algorithm can be expressed as:

[0068]

[0069] When using an iterative reconstruction algorithm based on machine learning for image reconstruction, if the result of the current iteration is almost the same as the result of the previous iteration, it can be considered that the current iteration has converged, and the result of the current iteration is recorded as a fixed point. At the same time, it is assumed that subsequent iterative cycles have little effect on improving the reconstruction effect, and the iterative cycle is terminated early.

[0070] In addition, in order to improve the adaptability of the model framework, the selection of the fidelity coefficient η is also changed from artificial empirical setting to learning of the convolution layer. For example, the spatial attention convolution layer can be used to k and A H (Ax-b) tensor to learn and obtain an accurate fidelity coefficient η.

[0071] It should be noted that the magnetic resonance image reconstruction method provided in the present application is applicable to the image reconstruction process of all parts of the body. This process can obtain dynamic magnetic resonance images of various parts at high temporal resolution and can fully capture the movement state of various parts of the body.

[0072] In the above-mentioned magnetic resonance image reconstruction method, a K-space data set of multiple phases of at least one layer of the detection object is obtained, and the K-space data set of at least one layer is undersampled; the K-space data sets of multiple phases are weighted and merged to generate a synthetic K-space data set, and a coil sensitivity distribution map is generated according to the synthetic K-space data set; the K-space data sets of multiple phases are respectively initialized and reconstructed based on the coil sensitivity distribution map to obtain an initial dynamic magnetic resonance image; the initial dynamic magnetic resonance image is iteratively reconstructed based on machine learning to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction of machine learning includes consistent items generated based on the K-space data sets of multiple phases and the coil sensitivity distribution map. The method reconstructs the K-space data sets of multiple phases in sequence through two reconstruction processes consisting of initialization and iterative reconstruction based on machine learning, which can more accurately complete the magnetic resonance image reconstruction process. At the same time, the iterative reconstruction based on machine learning includes consistent items generated based on the K-space data sets of multiple phases and the coil sensitivity distribution map. In the reconstruction process, it can ensure that the error between the reconstructed image and the K-space data sets of multiple phases is small, that is, it is closer to the real multi-phase magnetic resonance image of the detection object, which can improve the reconstruction quality of the magnetic resonance image.

[0073] Based on the above embodiments, this embodiment is to Figure 2 The relevant contents of step S203 "initializing and reconstructing the K-space data sets of multiple phases respectively to obtain the initial dynamic magnetic resonance image" are introduced and explained. Figure 3 As shown, as a non-limiting example, the above step S203 may include the following content:

[0074] S301, using a non-machine learning method to restore the K-space data sets of multiple phases to obtain an intermediate K-space data set.

[0075] In this embodiment, for the K-space data of each phase in the K-space data sets of multiple phases, the computer device can use a non-machine learning method to restore the K-space data of the phase to complete the data filling of the under-sampled image and obtain the K-space data of the phase. The K-space data of each phase can be obtained by the above method, and all the restored K-space data are determined as intermediate K-space data. For example, the non-machine learning method can be Phase Correction and Conjugate Synthesis (PCCS) or Projection Onto Convex Sets (POCS).

[0076] S302, reconstructing the intermediate K-space data set using the coil sensitivity distribution map to obtain an initial dynamic magnetic resonance image.

[0077] In this embodiment, after data recovery is performed on each phase 2D MRI under-sampled image, the computer device can use the coil sensitivity distribution map to reconstruct the intermediate K-space data set, obtain multiple frames of two-dimensional images after reconstruction, and determine the reconstructed image as the initial dynamic MRI image.

[0078] In the above-mentioned magnetic resonance image reconstruction method, a non-machine learning method is used to restore the K-space data sets of multiple phases to obtain an intermediate K-space data set; the coil sensitivity distribution map is used to reconstruct the intermediate K-space data set to obtain an initial dynamic magnetic resonance image. This method restores the K-space data sets of multiple phases by a non-machine learning method, and the coil sensitivity distribution map can be used to accurately fill in the data, so that the obtained initial dynamic magnetic resonance image is more accurate.

[0079] On the basis of the above-mentioned embodiment, the iterative reconstruction based on machine learning includes a three-dimensional reconstruction model and a two-dimensional reconstruction model; this embodiment is Figure 2 The relevant contents of step S204 "iteratively reconstructing the initial dynamic magnetic resonance image based on machine learning to obtain a dynamic magnetic resonance image of the detection object" are introduced and explained. Figure 4 As shown, as a non-limiting example, the above step S204 may include the following content:

[0080] S401, reconstructing the initial dynamic magnetic resonance image using the three-dimensional reconstruction model to obtain an intermediate dynamic magnetic resonance image.

[0081] In this embodiment, the computer device can input the initial dynamic magnetic resonance image into the three-dimensional reconstruction model, and reconstruct the initial dynamic magnetic resonance image using the three-dimensional reconstruction model. After the reconstruction is completed, an intermediate dynamic magnetic resonance image is obtained. This process can be based on the overall process to ensure the coherence between the magnetic resonance images of adjacent phases in the initial dynamic magnetic resonance image and avoid motion artifacts in the dynamic magnetic resonance image.

[0082] S402, using the two-dimensional reconstruction model to correct each phase magnetic resonance image in the intermediate dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object.

[0083] In this embodiment, since the intermediate dynamic magnetic resonance image only considers the continuity between adjacent phase magnetic resonance images, and does not consider the correction of each phase magnetic resonance image, it is necessary to correct each phase magnetic resonance image. The computer device can input each phase magnetic resonance image into the two-dimensional reconstruction model, and correct each phase magnetic resonance image through the two-dimensional reconstruction model to ensure the consistency between each phase magnetic resonance image and the corresponding magnetic resonance under-sampling image, so that the obtained dynamic magnetic resonance image of the detected object is closer to the real magnetic resonance full-sampling image.

[0084] In the above image reconstruction method, the initial dynamic magnetic resonance image is reconstructed using a three-dimensional reconstruction model to obtain an intermediate dynamic magnetic resonance image, and each phase magnetic resonance image in the intermediate dynamic magnetic resonance image is corrected using a two-dimensional reconstruction model to obtain a dynamic magnetic resonance image of the detection object. This method reconstructs the initial dynamic magnetic resonance image from both the overall and local dimensions using a three-dimensional reconstruction model and a two-dimensional reconstruction model, which can not only ensure the continuity of the reconstructed image in the time series, but also ensure the accuracy of the 2D magnetic resonance under-sampling image of each phase, and can improve the reconstruction quality of the magnetic resonance image.

[0085] Figure 5 The schematic diagram of the training process of the iterative reconstruction model based on machine learning is as follows: S501, estimating the coil sensitivity distribution map of the K-space data set of multiple phases; S502, based on the coil sensitivity distribution map, initializing the K-space data set of multiple phases to obtain an initial dynamic magnetic resonance sample image; S503, using the training data of the initial dynamic magnetic resonance sample image to train the iterative reconstruction model based on machine learning, during the training process, using the K-space data set of multiple phases as constraints, updating the parameters of the iterative reconstruction model based on machine learning, and obtaining a trained iterative reconstruction model based on machine learning; S504, using the test data in the initial dynamic magnetic resonance sample image to test the trained iterative reconstruction model based on machine learning to ensure the accuracy of the model.

[0086] Figure 6 This is a schematic diagram of the network architecture of the iterative reconstruction model based on machine learning. The iterative reconstruction model based on machine learning includes a three-dimensional reconstruction model and a two-dimensional reconstruction model. The left side of the model is the model input, that is, the initial dynamic magnetic resonance image, and the right side of the model is the model output, that is, the dynamic magnetic resonance image of the target part.

[0087] As a specific embodiment of the present application, Figure 7 As shown, the magnetic resonance image reconstruction method includes:

[0088] S601, obtaining a K-space data set of multiple phases of at least one layer of the detection object;

[0089] S602, weightedly merging the K-space data sets of multiple phases to generate a synthetic K-space data set, and generating a coil sensitivity distribution map according to the synthetic K-space data set;

[0090] S603, determining a coil sensitivity map according to the merged multi-phase 2D magnetic resonance under-acquisition image;

[0091] S604, recovering the K-space data sets of multiple phases using a non-machine learning method to obtain an intermediate K-space data set;

[0092] S605, reconstructing the intermediate K-space data set using the coil sensitivity distribution map to obtain an initial dynamic magnetic resonance image;

[0093] S606, reconstructing the initial dynamic magnetic resonance image using the three-dimensional reconstruction model to obtain an intermediate dynamic magnetic resonance image;

[0094] S607, using the two-dimensional reconstruction model to correct each phase magnetic resonance image in the intermediate dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object.

[0095] Figure 8 It is a schematic diagram of the K-space data set masks of multiple phases. Each column in the figure represents a K-space data mask image of a phase, and the K-space data of each phase are obtained by under-sampling. It can be seen from the figure that the K-space data sets of at least two adjacent phases are collected interlaced.

[0096] Fig. 9Schematic diagram of magnetic resonance image reconstruction results, which uses five phases of K-space data sets as examples for explanation, including magnetic resonance under-sampling images before image reconstruction, magnetic resonance images reconstructed by initialization, reconstructed magnetic resonance images, and magnetic resonance full-sampling images. It can be seen from the figure that the image quality of the magnetic resonance under-sampling images before image reconstruction is poor, the quality of the magnetic resonance images after initialization processing is improved compared with the magnetic resonance under-sampling images, and the difference between the reconstructed magnetic resonance images and the magnetic resonance full-sampling images is small, that is, through the magnetic resonance image reconstruction method provided in the embodiment of the present application, the reconstructed magnetic resonance images obtained are closer to the real magnetic resonance full-sampling images, and the image reconstruction quality is better.

[0097] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0098] Based on the same inventive concept, the embodiment of the present application also provides a magnetic resonance image reconstruction device for implementing the magnetic resonance image reconstruction method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more magnetic resonance image reconstruction device embodiments provided below can refer to the limitations of the magnetic resonance image reconstruction method above, and will not be repeated here.

[0099] In one embodiment, Fig.10 As shown, a magnetic resonance image reconstruction device is provided, comprising: an acquisition module 11, a generation module 12, an initialization reconstruction module 13 and an iterative reconstruction module 14, wherein:

[0100] An acquisition module 11 is used to acquire a K-space data set of multiple phases of at least one layer of the detection object, wherein the K-space data set of at least one layer is undersampled;

[0101] Wherein, K-space data sets of at least two adjacent phases are acquired in an interlaced manner.

[0102] A generating module 12, configured to weight and merge the K-space data sets of multiple phases to generate a synthetic K-space data set, and to generate a coil sensitivity distribution map according to the synthetic K-space data set;

[0103] An initialization reconstruction module 13 is used to respectively initialize and reconstruct the K-space data sets of multiple phases to obtain an initial dynamic magnetic resonance image;

[0104] An iterative reconstruction module 14 is used to perform iterative reconstruction based on machine learning on the initial dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction of machine learning includes consistent items generated based on the K-space data set of multiple phases and the coil sensitivity distribution map;

[0105] Among them, the iterative reconstruction based on machine learning includes a proximal gradient descent algorithm; the iterative reconstruction based on machine learning includes a 3D convolution module and a 2D convolution module.

[0106] In one embodiment, the initialization and reconstruction module includes: a recovery unit and a reconstruction unit, wherein:

[0107] A restoration unit, used for restoring the K-space data sets of multiple phases by using a non-machine learning method to obtain an intermediate K-space data set;

[0108] The reconstruction unit is used to reconstruct an initial dynamic magnetic resonance image by using the coil sensitivity distribution map and the intermediate K-space data set.

[0109] In one embodiment, the iterative reconstruction module comprises: a reconstruction unit and a correction unit, wherein:

[0110] A reconstruction unit, used for reconstructing the initial dynamic magnetic resonance image using the three-dimensional reconstruction model to obtain an intermediate dynamic magnetic resonance image;

[0111] The correction unit is used to correct each phase magnetic resonance image in the intermediate dynamic magnetic resonance image by using the two-dimensional reconstruction model to obtain a dynamic magnetic resonance image of the target part.

[0112] Each module in the above-mentioned image reconstruction device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module above.

[0113] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the content of any one of the above-mentioned magnetic resonance image reconstruction method embodiments is implemented.

[0114] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the content of any one of the above-mentioned magnetic resonance image reconstruction method embodiments is implemented.

[0115] In one embodiment, a computer program product is provided, comprising a computer program, which implements the content of any one of the above-mentioned magnetic resonance image reconstruction method embodiments when executed by a processor.

[0116] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0117] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0118] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A magnetic resonance image reconstruction method, characterized in that: The method comprises: Acquiring a K-space data set of multiple phases of at least one layer of the detection object, wherein the K-space data set of at least one layer is undersampled; weightedly merging the K-space data sets of multiple phases to generate a synthetic K-space data set, and generating a coil sensitivity distribution map according to the synthetic K-space data set; Initializing and reconstructing the K-space data sets of the multiple phases respectively to obtain an initial dynamic magnetic resonance image; The initial dynamic magnetic resonance image is iteratively reconstructed based on machine learning to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction of machine learning includes consistent items generated based on the K-space data sets of the multiple phases and the coil sensitivity distribution map.

2. The method according to claim 1, characterized in that The K-space data sets of at least two adjacent phases are acquired in an interleaved manner.

3. The method according to claim 2, characterized in that The step of respectively initializing and reconstructing the K-space data sets of the multiple phases to obtain an initial dynamic magnetic resonance image includes: Recovering the K-space data sets of the multiple phases using a non-machine learning method to obtain an intermediate K-space data set; The intermediate K-space data set is reconstructed using the coil sensitivity distribution map to obtain the initial dynamic magnetic resonance image.

4. The method according to claim 1, characterized in that: The machine learning based iterative reconstruction includes a proximal gradient descent algorithm.

5. The method according to any one of claims 1 to 4, characterized in that: The iterative reconstruction based on machine learning includes a 3D convolution module and a 2D convolution module.

6. The method according to any one of claims 1 to 4, characterized in that: The iterative reconstruction based on machine learning includes a three-dimensional reconstruction model and a two-dimensional reconstruction model; The iterative reconstruction based on machine learning on the initial dynamic magnetic resonance image to obtain the dynamic magnetic resonance image of the detection object includes: Reconstructing the initial dynamic magnetic resonance image using the three-dimensional reconstruction model to obtain an intermediate dynamic magnetic resonance image; The two-dimensional reconstruction model is used to correct each phase magnetic resonance image in the intermediate dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object.

7. A magnetic resonance image reconstruction device, characterized in that: The device comprises: An acquisition module, used for acquiring a K-space data set of multiple phases of at least one layer of the detection object, wherein the K-space data set of at least one layer is undersampled; A generating module, configured to weight and merge the K-space data sets of multiple phases to generate a synthetic K-space data set, and to generate a coil sensitivity distribution map according to the synthetic K-space data set; An initialization and reconstruction module, used to respectively initialize and reconstruct the K-space data sets of the multiple phases to obtain an initial dynamic magnetic resonance image; An iterative reconstruction module is used to perform iterative reconstruction based on machine learning on the initial dynamic magnetic resonance image to obtain a dynamic magnetic resonance image of the detection object; the iterative reconstruction of the machine learning includes consistent items generated based on the K-space data sets of the multiple phases and the coil sensitivity distribution map.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images