Water-fat separation method, system and terminal for magnetic resonance imaging

Abstract

The water-fat separation method, system and terminal for magnetic resonance imaging of the present application obtain the corresponding water-fat separation result according to the echo complex image data through the water-fat separation model based on the deep learning and the traditional model segmentation combination. The present application optimizes the precision of the field map extraction, improves the stability and calculation speed of the water-fat separation, at the same time, retains the constraint force of the traditional algorithm on the output result, and maximizes the consistency of the output result and the original signal.

Description

Technical Field

[0001] This invention relates to the field of magnetic resonance imaging technology, and in particular to a method, system and terminal for separating water and lipids for magnetic resonance imaging. Background Technology

[0002] Magnetic Resonance Imaging (MRI) is an imaging technique that uses radio frequency pulses to excite specific atomic nuclei (usually hydrogen nuclei) in a magnetic field to generate magnetic resonance signals. These signals are then acquired and reconstructed to obtain tomographic images of anatomical structures. As a non-invasive, multi-directional, and multi-contrast imaging method, MRI has been widely used in the medical field and has become an important tool for clinical diagnosis. Human tissues contain both water and fat, and the bright signals from fat can interfere with the identification of some potential lesions. Separating water and fat for visualization can not only enhance the contrast of water images but also provide doctors with more auxiliary diagnostic information using fat images. Water-fat separation methods, as a highly efficient chemical shift imaging technique, are less sensitive to inhomogeneities in the main magnetic field and radio frequency, and can simultaneously obtain multiple contrast images with high signal-to-noise ratios (water images, fat images, in-phase images, and out-of-phase images). They are increasingly widely used in the field of MRI fat suppression. Traditional water-fat separation techniques include iterative smoothing, region growing, graph cutting, and information tree transfer methods, among other globally optimized methods. These traditional methods are computationally complex and time-consuming, and are highly dependent on the selection of high-confidence initial estimates and the ability to optimize convergence. In some challenging applications, such as those with low SNR or rapid changes in the main magnetic field, it is difficult to guarantee a globally optimal solution, which will reduce the accuracy of water-fat separation.

[0003] In recent years, with breakthroughs in big data and computing power, deep learning has developed rapidly, and more and more MRI studies have incorporated deep learning modules. The precise and efficient learning ability of neural networks to learn the nonlinear mapping relationship between input and output has also been applied to the water-fat separation task in MRI. Neural networks can be pre-trained with large datasets to improve the reconstruction speed and robustness of water-fat separation algorithms. However, most current deep learning-based water-fat separation methods rely on general network architectures such as U-net and CNN for end-to-end one-stop separation from raw acquisition signals to water and fat images. This not only increases the difficulty of network training, reduces the network's generalization ability, and exacerbates tissue water-fat exchange problems, but also loses some high-frequency detail information, making the separated water-fat images blurry. Although adding GAN networks can improve image sharpness, it further increases the difficulty of network training. Currently, the combination of deep learning-based water-fat separation methods with traditional methods is limited to the initialization module of the field map (P), and has not yet been applied to the accurate calculation of the major chemical components (B) and minor chemical components (S) of the enhanced pixels. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a water-lipid separation method, system and terminal for magnetic resonance imaging, so as to solve the above technical problems in the prior art.

[0005] To achieve the above and other related objectives, the present invention provides a method for separating water and lipids in magnetic resonance imaging. The method includes: acquiring complex echo image data to be separated; wherein the complex echo image data includes: complex echo images at least two different echo times; and obtaining corresponding water and lipid separation results based on a water and lipid separation model that combines deep learning with traditional segmented models, according to the complex echo image data.

[0006] In one embodiment of the present invention, the water-lipid separation model includes: a data input module, used to acquire at least two complex echo images with different echo times and extract the corresponding real and imaginary data for input; a candidate phase extraction optimization module, connected to the data input module, used to calculate the distribution map of major chemical components and the distribution map of minor chemical components based on the extracted real and imaginary data using a cascaded multi-scale convolutional network, and to calculate two candidate phase values; a field map extraction optimization module, connected to the candidate phase extraction optimization module, used to perform iterative mutual feedback field map extraction based on the two candidate phase values ​​and the initial estimated field map to obtain the final extracted field map; a water-lipid decomposition module, connected to the field map extraction optimization module, used to decompose the final extracted field map to obtain the corresponding water image and lipid image; and an output module, connected to the water-lipid decomposition module, used to output the water image and lipid image.

[0007] In one embodiment of the present invention, the candidate phase extraction optimization module includes: a chemical composition map extraction unit, comprising: a cascaded multi-scale convolutional network, used to calculate the distribution map of the main chemical components and the distribution map of the minor chemical components based on the extracted real part data and imaginary part data; and a traditional candidate phase calculation unit, connected to the chemical composition map extraction unit, used to calculate two candidate phase values ​​based on the distribution map of the main chemical components and the distribution map of the minor chemical components according to the traditional candidate phase calculation model.

[0008] In one embodiment of the present invention, the cascaded multi-scale convolutional network includes: multiple cascaded supervised networks; wherein, the first supervised network is used to input two candidate phase values ​​and output the corresponding primary chemical component distribution map and secondary chemical component distribution map; each non-first supervised network is used to input the primary chemical component distribution map, secondary chemical component distribution map, two candidate phase values ​​and their respective adjacent region information output by the previous supervised network, and output the generated primary chemical component distribution map and secondary chemical component distribution map.

[0009] In one embodiment of the present invention, the supervised network is a convolutional network with a corresponding convolutional kernel.

[0010] In one embodiment of the present invention, the replacement processing submodule includes: a deletion editing unit, configured to modify the Mel spectrogram based on a deletion indicator corresponding to a target deletion location region, and obtain first mean sequence data and first variance sequence data based on two one-dimensional convolution modules; and an addition editing unit, connected to the deletion editing unit, configured to modify the first mean sequence data and first variance sequence data based on an addition indicator corresponding to a target addition location region, to obtain mean sequence processed data and variance sequence processed data.

[0011] In one embodiment of the present invention, the field map extraction and optimization module includes: multiple cascaded field map extraction sub-modules, used to perform iterative mutual feedback field map extraction based on two candidate phase values ​​and an initial estimated field map, and output an initial field map; and a field map optimization sub-module, used to obtain a final extracted field map based on the initial field map and in combination with two candidate phase values, input real part data and imaginary part data, and reference water-lipid image information.

[0012] In one embodiment of the present invention, each field map extraction submodule includes: a first deep learning network model and a conventional field map extraction model connected to the first deep learning network model; wherein, the types of the field map extraction submodules include: a first field map extraction submodule, used to output a corresponding extracted first initial field map based on the first deep learning network model according to two input candidate phase values ​​and an initial estimated field map, and then input the conventional field map extraction model to extract and output the corresponding first field map; and a non-first field map extraction submodule, used to output a corresponding first initial field map based on the first deep learning network model according to two input candidate phase values ​​and the first field map extracted by the previous field map extraction submodule, and then input the conventional field map extraction model to extract and output the corresponding first field map, so that the final field map extraction submodule can output the initial field map.

[0013] In one embodiment of the present invention, the field map optimization submodule includes: a second deep learning network model, used to output a corresponding initial extracted field map based on the input initial field map; and an optimization weighting unit, connected to the second deep learning network model, used to perform weighted optimization calculations based on the initial extracted field map, two candidate phase values, input real part data and imaginary part data, and reference water-lipid image information to obtain the final extracted field map.

[0014] To achieve the above and other related objectives, the present invention provides a water-lipid separation system for magnetic resonance imaging. The system includes: an acquisition module for acquiring echo complex image data to be separated; wherein the echo complex image data includes: at least two echo complex images with different echo times; and a water-lipid separation module connected to the acquisition module for obtaining corresponding water-lipid separation results based on a water-lipid separation model that combines deep learning with traditional segmented models, according to the echo complex image data.

[0015] To achieve the above and other related objectives, the present invention provides a water-lipid separation terminal for magnetic resonance imaging, comprising: one or more memories and one or more processors; the one or more memories are used to store a computer program; the one or more processors are connected to the memories and are used to run the computer program to execute the water-lipid separation method for magnetic resonance imaging.

[0016] As described above, this invention provides a method, system, and terminal for water-lipid separation in magnetic resonance imaging (MRI), offering the following advantages: This invention utilizes a water-lipid separation model that combines deep learning with a segmented approach based on traditional models to obtain the corresponding water-lipid separation results from the echo complex image data. This invention optimizes the accuracy of field image extraction and improves the stability and computational speed of water-lipid separation, while retaining the constraints of traditional algorithms on the output results, maximizing the consistency between the output results and the original signal. Attached Figure Description

[0017] Figure 1 The diagram shown is a flowchart illustrating a method for separating water and lipids for magnetic resonance imaging according to an embodiment of the present invention.

[0018] Figure 2 The diagram shown is a structural schematic of a water-lipid separation model according to an embodiment of the present invention.

[0019] Figure 3 The diagram shown is a structural schematic of the candidate phase extraction optimization module in one embodiment of the present invention.

[0020] Figure 4 The diagram shown is a schematic representation of a cascaded multi-scale convolutional network in one embodiment of the present invention.

[0021] Figure 5 The diagram shown is a schematic representation of a cascaded multi-scale convolutional network in one embodiment of the present invention.

[0022] Figure 6 The diagram shown is a structural schematic of the field map extraction and optimization module in one embodiment of the present invention.

[0023] Figure 7 The diagram shown is a structural schematic of a water-lipid separation model according to an embodiment of the present invention.

[0024] Figure 8 The diagram shown is a schematic representation of a conventional algorithm flow according to an embodiment of the present invention.

[0025] Figure 9 The diagram shown is a schematic representation of a water-lipid separation system for magnetic resonance imaging according to an embodiment of the present invention.

[0026] Figure 10 The diagram shown is a structural schematic of a water-fat separation terminal for magnetic resonance imaging according to an embodiment of the present invention. Detailed Implementation

[0027] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0028] It should be noted that in the following description, reference is made to the accompanying drawings, which illustrate several embodiments of the invention. It should be understood that other embodiments may also be used, and changes in mechanical composition, structure, electrical system, and operation may be made without departing from the spirit and scope of the invention. The following detailed description should not be considered limiting, and the scope of the embodiments of the invention is defined only by the claims of the published patents. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. Spatially related terms, such as “upper,” “lower,” “left,” “right,” “below,” “below,” “lower part,” “above,” “upper part,” etc., may be used herein to illustrate the relationship between one element or feature shown in the figures and another element or feature.

[0029] Throughout this specification, when it is said that a part is "connected" to another part, this includes not only "direct connection" but also "indirect connection" by placing other elements in between. Furthermore, when it is said that a part "includes" a certain constituent element, unless otherwise stated otherwise, this does not exclude other constituent elements, but rather means that other constituent elements may also be included.

[0030] The terms "first," "second," and "third," etc., used herein are for the purpose of describing various parts, components, regions, layers, and / or segments, but are not limiting. These terms are used only to distinguish one part, component, region, layer, or segment from others. Therefore, the "first part," "component," "region," "layer," or "segment" described below may refer to a "second part," "component," "region," "layer," or "segment" without departing from the scope of this invention.

[0031] Furthermore, as used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms “comprising,” “including,” indicate the presence of the stated feature, operation, element, component, item, kind, and / or group, but do not preclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms “or” and “and / or” as used herein are interpreted as inclusive, or mean any one or any combination thereof. Thus, “A, B, or C” or “A, B, and / or C” means “any one of: A; B; C; A and B; A and C; B and C; A, B, and C.” Exceptions to this definition arise only when combinations of elements, functions, or operations are inherently mutually exclusive in some manner.

[0032] This invention provides a method, system, and terminal for water-lipid separation in magnetic resonance imaging. It utilizes a water-lipid separation model that combines deep learning with a segmented approach based on traditional models to obtain the corresponding water-lipid separation results from the echo complex image data. This invention optimizes the accuracy of field image extraction, improves the stability and computational speed of water-lipid separation, while retaining the constraints of traditional algorithms on the output results, maximizing the consistency between the output results and the original signal.

[0033] The present invention will now be described in detail with reference to the accompanying drawings, so that those skilled in the art can readily implement it. The present invention can be embodied in many different forms and is not limited to the embodiments described herein.

[0034] like Figure 1 This invention presents a schematic flowchart of a water-lipid separation method for magnetic resonance imaging according to an embodiment of the present invention.

[0035] Step S1: Obtain the complex echo image data to be separated.

[0036] Specifically, the echo complex image data includes: at least two echo complex images with different echo times.

[0037] Step S2: Based on the water-fat separation model that combines deep learning with traditional segmented models, obtain the corresponding water-fat separation results according to the echo complex image data.

[0038] In one embodiment, such as Figure 2 As shown, the water-lipid separation model includes:

[0039] Data input module 21 is used to acquire complex echo images at least two different echo times and extract the corresponding real part data I_r and imaginary part data I_i.

[0040] The candidate phase extraction optimization module 22 is connected to the data input module 21 and is used to calculate the distribution map of the main chemical components B and the distribution map of the minor chemical components S based on the extracted real part data I_r and imaginary part data I_i using a cascaded multi-scale convolutional network, and to calculate two candidate phase values.

[0041] The field map extraction optimization module 23 is connected to the candidate phase extraction optimization module 22 and is used to perform iterative mutual feedback field map extraction based on two candidate phase values ​​and the initial estimated field map to obtain the final extracted field map.

[0042] The water-lipid decomposition module 24 is connected to the field image extraction and optimization module 23 and is used to decompose the final extracted field image to obtain the corresponding water image and lipid image.

[0043] Output module 25 is connected to the water-lipid decomposition module 24 and is used to output the water image and the lipid image.

[0044] In one embodiment, the candidate phase extraction optimization module 22 includes:

[0045] The chemical composition map extraction unit includes a cascaded multi-scale convolutional network used to calculate the distribution maps of major and minor chemical components based on the extracted real and imaginary data; such as... Figure 3 As shown, the extracted real and imaginary data are input into the cascaded multi-scale convolutional network to obtain the estimated distribution map B of the main chemical components and the distribution map S of the minor chemical components. In this invention, the cascaded multi-scale convolutional network replaces the traditional algorithm that uses the formula...

[0046] The process of calculating B and S using Equations 1 and 2:

[0047]

[0048]

[0049] As can be seen from Equations 1 and 2, the calculation of B and S in the traditional algorithm is only related to the information of the current point, and some values ​​cannot be directly solved, requiring estimation based on multiple parameters and experience. When affected by high noise or extremely inhomogeneous fields, the accuracy of B and S decreases significantly.

[0050] A traditional candidate phase calculation unit, connected to the chemical composition map extraction unit, is used to calculate two candidate phase values ​​based on the distribution maps of the primary and secondary chemical components using a traditional candidate phase calculation model. For example... Figure 3 As shown, candidate phase 1 and candidate phase 2 are obtained based on the distribution map B of the main chemical components and the distribution map S of the minor chemical components obtained by the cascaded multi-scale convolutional network.

[0051] It should be noted that the algorithm used in the traditional candidate phase calculation model to calculate the candidate phase value Z1 and the second candidate phase value Z2 is as follows:

[0052]

[0053]

[0054] It should be noted that formulas 1, 2, 3, and 4 in this example use a single-peak fat model, but can also be extended to a multi-peak fat model.

[0055] In one embodiment, such as Figure 4As shown, the cascaded multi-scale convolutional network includes: multiple cascaded supervised networks; and each supervised network outputs a distribution map of major chemical components and a distribution map of minor chemical components, and uses them as the input of the next supervised network for output.

[0056] The first supervised network takes two candidate phase values ​​as input and outputs the corresponding distribution maps of primary and secondary chemical components. Each subsequent supervised network takes the primary and secondary chemical component distribution maps output by the previous supervised network, two candidate phase values, the primary and secondary chemical component distribution maps, and the adjacent region information of each of the two candidate phase values ​​as input, and outputs the generated primary and secondary chemical component distribution maps. The size of the adjacent region corresponding to the information taken from the adjacent region is related to the corresponding supervised network.

[0057] In this invention, the collected data is directly input into a pre-trained cascaded multi-scale convolutional network. This network combines information between input echoes and information about the adjacent regions of each voxel, expanding the detection range scale by scale. This allows for the extraction of richer and more accurate feature representations, reducing the influence of noise and low signal-to-noise ratio regions, and obtaining more precise distribution maps B and S of the main and secondary chemical components at each point. These are then combined with traditional models to calculate two candidate phase values.

[0058] In one embodiment, the supervised network employs a convolutional network with one convolutional kernel per convolutional layer. For example... Figure 5 As shown, for example, the real data I1 and imaginary data I2 are input into supervised network 1, which uses a convolutional network with a 1×1 kernel, and outputs the corresponding main chemical component distribution map B1 and minor chemical component distribution map S1. Then, the main chemical component distribution map B1, the minor chemical component distribution map S1, the real data I1, the imaginary data I2, and their respective neighborhood region information are input into supervised network 2, which uses a convolutional network with a 3×3 kernel. Then, the main chemical component distribution map and the minor chemical component distribution map output by the previous supervised network, the real data I1, the imaginary data I2, and the adjacent region information of each voxel are sequentially input into the next supervised network, until they are input into the final supervised network N, which uses a convolutional network with a 3×3 kernel, and outputs the final main chemical component distribution map B and the minor chemical component distribution map S.

[0059] It should be noted that the convolution kernels used in each supervised network are set according to specific needs, and the same or different convolution kernels can be set.

[0060] In one embodiment, the field map extraction and optimization module includes:

[0061] Multiple cascaded field map extraction submodules are used to perform iterative mutual feedback field map extraction based on two candidate phase values ​​and an initial predicted field map, and output the initial field map;

[0062] The field map optimization submodule is used to obtain the final extracted field map based on the initial field map and in combination with two candidate phase values, the input real part data and imaginary part data, and the reference water-lipid image information.

[0063] In one embodiment, such as Figure 6 As shown, each field map extraction submodule includes: a first deep learning network model and a conventional field map extraction model connected to the first deep learning network model; wherein,

[0064] The types of field map extraction submodules include: a first field map extraction submodule, which is used to output the corresponding extracted first initial field map based on the first deep learning network model according to the two input candidate phase values ​​and the initial estimated field map, and then input the traditional field map extraction model to extract and output the corresponding first field map; and a non-first field map extraction submodule, which is used to output the corresponding first initial field map based on the first deep learning network model according to the two input candidate phase values ​​and the first field map extracted by the previous field map extraction submodule, and then input the traditional field map extraction model to extract and output the corresponding first field map, so that the final field map extraction submodule can output the initial field map.

[0065] It should be noted that the field map extraction submodule can employ RIPE, water phase analysis, region growth, TRWs, graph cut method, projected power method, etc.

[0066] In one embodiment, such as Figure 6 As shown, the field map optimization submodule includes: a second deep learning network model, used to output a corresponding initial extracted field map based on the input initial field map; and an optimization weighting unit, connected to the second deep learning network model, used to perform weighted optimization calculations based on the initial extracted field map, two candidate phase values, input real and imaginary data, and reference water-lipid image information to obtain the final extracted field map.

[0067] It should be noted that the first deep learning network model and the second deep learning network model are not limited to CNN, U-net, V-net, GAN, and their variations.

[0068] To better illustrate the above-mentioned water-lipid separation method for magnetic resonance imaging, the present invention provides the following specific embodiments.

[0069] Example 1: A water-lipid separation method based on a combination of deep learning and traditional model segmentation.

[0070] The method is based on a water-fat separation model that combines deep learning with traditional segmented models, and obtains the corresponding water-fat separation results based on the complex echo images at at least two different echo times.

[0071] Among them, such as Figure 7 As shown, the water-lipid separation model includes:

[0072] The data input module acquires complex images at at least two different TE times and extracts the input according to the real part I_r and the imaginary part I_i.

[0073] This module, based on a cascaded multi-scale convolutional network, extracts B and S values, replacing the traditional process of calculating B and S (see flowchart of the traditional algorithm). Figure 8 In traditional algorithms, the calculation of B and S depends only on the information of the current point, and some values ​​cannot be directly solved, requiring estimation based on multiple parameters and experience. When affected by high noise or extremely inhomogeneous fields, the accuracy of B and S decreases significantly.

[0074] The field map extraction and optimization module based on a deep learning network model inputs candidate phase values ​​and the estimated field map into a pre-trained deep learning network to obtain a globally high-confidence initial field map P0. P0 is then input into a traditional field map extraction module to obtain a smoothed field map Pns. The candidate phase values ​​and Pns are then fed back into the mutual feedback field map extraction and optimization module, formed by the deep learning network model and the traditional field map extraction algorithm, to obtain the output field map Ps. Finally, the candidate phase values ​​and Ps are input into a pre-trained convolutional network, and the field map is weighted and optimized by integrating the original signal and water map / fat image information to obtain the final field map.

[0075] The water-lipid decomposition module uses a traditional model to decompose and obtain the final water and lipid images W and F.

[0076] Similar in principle to the above embodiments, the present invention provides a water-lipid separation system for magnetic resonance imaging.

[0077] The following specific embodiments are provided in conjunction with the accompanying drawings:

[0078] like Figure 9 A schematic diagram of a water-fat separation system for magnetic resonance imaging is shown in an embodiment of the present invention.

[0079] The system includes:

[0080] The acquisition module 91 is used to acquire the echo complex image data to be separated; wherein, the echo complex image data includes: at least two echo complex images with different echo times;

[0081] The water-fat separation module 92 is connected to the acquisition module 91 and is used to obtain the corresponding water-fat separation result based on the echo complex image data using a water-fat separation model that combines deep learning with traditional segmented models.

[0082] Since the implementation principle of the water-lipid separation system for magnetic resonance imaging has been described in the foregoing embodiments, it will not be repeated here.

[0083] In one embodiment, the water-lipid separation model includes: a data input module for acquiring at least two complex echo images at different echo times and extracting the corresponding real and imaginary data; a candidate phase extraction optimization module connected to the data input module for calculating the distribution maps of major and minor chemical components based on the extracted real and imaginary data using a cascaded multi-scale convolutional network, and calculating two candidate phase values; a field map extraction optimization module connected to the candidate phase extraction optimization module for iteratively extracting the field map based on the two candidate phase values ​​and the initial estimated field map to obtain the final extracted field map; a water-lipid decomposition module connected to the field map extraction optimization module for decomposing the final extracted field map to obtain corresponding water and lipid images; and an output module connected to the water-lipid decomposition module for outputting the water and lipid images.

[0084] In one embodiment, the candidate phase extraction optimization module includes: a chemical composition map extraction unit, comprising: a cascaded multi-scale convolutional network, used to calculate the distribution map of the main chemical components and the distribution map of the minor chemical components based on the extracted real and imaginary data; and a traditional candidate phase calculation unit, connected to the chemical composition map extraction unit, used to calculate two candidate phase values ​​based on the distribution map of the main chemical components and the distribution map of the minor chemical components according to the traditional candidate phase calculation model.

[0085] In one embodiment, the cascaded multi-scale convolutional network includes: multiple cascaded supervised networks; wherein, the first supervised network is used to input two candidate phase values ​​and output the corresponding primary chemical component distribution map and secondary chemical component distribution map; each non-first supervised network is used to input the primary chemical component distribution map, secondary chemical component distribution map, two candidate phase values ​​and their respective adjacent region information output by the previous supervised network, and output the generated primary chemical component distribution map and secondary chemical component distribution map.

[0086] In one embodiment, the supervised network is a convolutional network with one convolutional kernel.

[0087] In one embodiment, the field map extraction and optimization module includes: multiple cascaded field map extraction sub-modules, used for iterative mutual feedback field map extraction based on two candidate phase values ​​and an initial estimated field map, and outputting an initial field map; and a field map optimization sub-module, used for obtaining a final extracted field map based on the initial field map and in combination with two candidate phase values, input real and imaginary data, and reference water-lipid image information.

[0088] In one embodiment, each field map extraction submodule includes: a first deep learning network model and a conventional field map extraction model connected to the first deep learning network model; wherein, the types of the field map extraction submodules include: a first field map extraction submodule, used to output a corresponding extracted first initial field map based on the first deep learning network model according to two input candidate phase values ​​and an initial estimated field map, and then input the conventional field map extraction model to extract and output the corresponding first field map; and a non-first field map extraction submodule, used to output a corresponding first initial field map based on the first deep learning network model according to two input candidate phase values ​​and the first field map extracted by the previous field map extraction submodule, and then input the conventional field map extraction model to extract and output the corresponding first field map, so that the final field map extraction submodule can output the initial field map.

[0089] In one embodiment, the field map optimization submodule includes: a second deep learning network model, used to output a corresponding initial extracted field map based on the input initial field map; and an optimization weighting unit, connected to the second deep learning network model, used to perform weighted optimization calculations based on the initial extracted field map, two candidate phase values, input real and imaginary data, and reference water-lipid image information to obtain the final extracted field map.

[0090] like Figure 10 A schematic diagram of the structure of the water-lipid separation terminal 100 for magnetic resonance imaging in an embodiment of the present invention is shown.

[0091] The water-lipid separation terminal 100 for magnetic resonance imaging includes: a memory 101 and a processor 102. The memory 101 is used to store computer programs; the processor 102 runs the computer programs to implement, for example... Figure 1 The water-fat separation method for magnetic resonance imaging.

[0092] Optionally, the number of memories 101 can be one or more, and the number of processors 102 can be one or more. Figure 10 Each example is taken as an instance.

[0093] Optionally, the processor 102 in the water-lipid separation terminal 100 for magnetic resonance imaging will perform the following... Figure 1The steps described involve loading one or more instructions corresponding to the process of an application into memory 101, and having the processor 102 run the application stored in the first memory 101, thereby achieving the following: Figure 1 Various functions in the water-fat separation method for magnetic resonance imaging.

[0094] Optionally, the memory 101 may include, but is not limited to, high-speed random access memory and non-volatile memory. For example, one or more disk storage devices, flash memory devices, or other non-volatile solid-state storage devices; the processor 102 may include, but is not limited to, a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0095] Optionally, the processor 102 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0096] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed, implements as follows: Figure 1The illustrated method for separating water and lipids for magnetic resonance imaging. The computer-readable storage medium may include, but is not limited to, floppy disks, optical disks, CD-ROMs (Read-Only Optical Disk Memory), magneto-optical disks, ROMs (Read-Only Memory), RAMs (Random Access Memory), EPROMs (Erasable Programmable Read-Only Memory), EEPROMs (Electrically Erasable Programmable Read-Only Memory), magnetic cards or optical cards, flash memory, or other types of media / machine-readable media suitable for storing machine-executable instructions. The computer-readable storage medium may be a product not connected to a computer device or a component used in a computer device.

[0097] The water-lipid separation method, system, and terminal for magnetic resonance imaging of the present invention further improve the reconstruction speed and accuracy of water-lipid separation while ensuring that the separated water and fat maps are highly consistent with the original signal. Compared with existing technologies, it has the following advantages:

[0098] 1. This invention is the first to propose a water-lipid separation method that combines deep learning with traditional segmented models. This invention optimizes the accuracy of module-extracted values ​​by constructing different network structures and combining them with traditional algorithms, thereby improving the stability and computational speed of water-lipid separation. At the same time, it retains the constraints of traditional algorithms on the output results, maximizing the consistency between the output results and the original signal.

[0099] 2. The cascaded multi-scale convolutional network proposed in this invention expands the detection range scale by combining the information between input echoes and the information of adjacent regions of each voxel, so as to extract richer and more accurate feature representations, reduce the influence of noise and low signal-to-noise ratio regions, and further improve the accuracy of the first step of candidate error phase calculation in the traditional water-fat separation algorithm.

[0100] 3. This invention employs an iterative mutual feedback field map optimization extraction method combining a deep learning network model and a traditional field map extraction module to further improve the accuracy of the field map. Finally, the deep learning network model is used to further optimize the field map from a global perspective, integrating the original input information and information from reference water and fat images to improve the stability of water-fat separation.

[0101] In summary, the water-lipid separation method, system, and terminal for magnetic resonance imaging of the present invention obtain the corresponding water-lipid separation results based on the echo complex image data by using a water-lipid separation model that combines deep learning with traditional segmented models. The present invention optimizes the accuracy of field image extraction and improves the stability and computational speed of water-lipid separation, while retaining the constraints of traditional algorithms on the output results, maximizing the consistency between the output results and the original signal. Therefore, the present invention effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0102] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for separating water and lipids for magnetic resonance imaging, characterized in that, The method includes: Acquire the complex echo image data to be separated; wherein, the complex echo image data includes: complex echo images at least two different echo times; A water-fat separation model based on a combination of deep learning and traditional segmented models is used to obtain the corresponding water-fat separation results based on the echo complex image data. The water-lipid separation model includes: a data input module for acquiring complex echo images at at least two different echo times and extracting the corresponding real and imaginary data; a candidate phase extraction optimization module connected to the data input module for calculating the distribution maps of major and minor chemical components based on the extracted real and imaginary data using a cascaded multi-scale convolutional network, and calculating two candidate phase values; a field map extraction optimization module connected to the candidate phase extraction optimization module for iterative mutual feedback field map extraction based on the two candidate phase values ​​and the initial estimated field map to obtain the final extracted field map; a water-lipid decomposition module connected to the field map extraction optimization module for decomposing the final extracted field map to obtain the corresponding water and lipid images; and an output module connected to the water-lipid decomposition module for outputting the water and lipid images. The field map extraction and optimization module includes: multiple cascaded field map extraction sub-modules, used for iterative mutual feedback field map extraction based on two candidate phase values ​​and an initial estimated field map, and outputting an initial field map; and a field map optimization sub-module, used for obtaining the final extracted field map based on the initial field map, combined with two candidate phase values, input real and imaginary data, and reference water-lipid image information. Each field map extraction submodule includes: a first deep learning network model and a conventional field map extraction model connected to the first deep learning network model; wherein, the types of the field map extraction submodules include: a first field map extraction submodule, used to output a corresponding extracted first initial field map based on the first deep learning network model according to two input candidate phase values ​​and an initial estimated field map, and then input the conventional field map extraction model to extract and output the corresponding first field map; and a non-first field map extraction submodule, used to output a corresponding first initial field map based on the first deep learning network model according to two input candidate phase values ​​and the first field map extracted by the previous field map extraction submodule, and then input the conventional field map extraction model to extract and output the corresponding first field map, so that the final field map extraction submodule can output the initial field map.

2. The water-lipid separation method for magnetic resonance imaging according to claim 1, characterized in that, The candidate phase extraction optimization module includes: The chemical composition map extraction unit includes a cascaded multi-scale convolutional network, which is used to calculate the distribution maps of major and minor chemical components based on the extracted real and imaginary data. The traditional candidate phase calculation unit is connected to the chemical composition map extraction unit and is used to calculate two candidate phase values ​​based on the distribution map of the main chemical composition and the distribution map of the secondary chemical composition according to the traditional candidate phase calculation model.

3. The water-lipid separation method for magnetic resonance imaging according to claim 2, characterized in that, The cascaded multi-scale convolutional network includes: Multiple cascaded supervised networks are used. The first supervised network is used to input real part data and imaginary part data and output the corresponding distribution maps of major and minor chemical components. Each subsequent supervised network is used to input the distribution maps of major and minor chemical components, real part data, imaginary part data and their respective adjacent region information output by the previous supervised network, and output the generated distribution maps of major and minor chemical components.

4. The water-lipid separation method for magnetic resonance imaging according to claim 3, characterized in that, The supervised network is a convolutional network with one convolutional kernel.

5. The water-lipid separation method for magnetic resonance imaging according to claim 1, characterized in that, The field map optimization submodule includes: The second deep learning network model is used to output the corresponding initial extracted field map based on the input initial field map; The optimization weighting unit, connected to the second deep learning network model, is used to perform weighted optimization calculations based on the initial extracted field map, two candidate phase values, input real and imaginary data, and reference water-lipid image information to obtain the final extracted field map.

6. A water-fat separation system for magnetic resonance imaging, characterized in that, The system includes: An acquisition module is used to acquire the echo complex image data to be separated; wherein, the echo complex image data includes: at least two echo complex images with different echo times; The water-fat separation module, connected to the acquisition module, is used to obtain the corresponding water-fat separation result based on the echo complex image data using a water-fat separation model that combines deep learning with a traditional segmented model. The water-lipid separation model includes: a data input module for acquiring complex echo images at at least two different echo times and extracting the corresponding real and imaginary data; a candidate phase extraction optimization module connected to the data input module for calculating the distribution maps of major and minor chemical components based on the extracted real and imaginary data using a cascaded multi-scale convolutional network, and calculating two candidate phase values; a field map extraction optimization module connected to the candidate phase extraction optimization module for iterative mutual feedback field map extraction based on the two candidate phase values ​​and the initial estimated field map to obtain the final extracted field map; a water-lipid decomposition module connected to the field map extraction optimization module for decomposing the final extracted field map to obtain the corresponding water and lipid images; and an output module connected to the water-lipid decomposition module for outputting the water and lipid images. The field map extraction and optimization module includes: multiple cascaded field map extraction sub-modules, used for iterative mutual feedback field map extraction based on two candidate phase values ​​and an initial estimated field map, and outputting an initial field map; and a field map optimization sub-module, used for obtaining the final extracted field map based on the initial field map, combined with two candidate phase values, input real and imaginary data, and reference water-lipid image information. Each field map extraction submodule includes: a first deep learning network model and a conventional field map extraction model connected to the first deep learning network model; wherein, the types of the field map extraction submodules include: a first field map extraction submodule, used to output a corresponding extracted first initial field map based on the first deep learning network model according to two input candidate phase values ​​and an initial estimated field map, and then input the conventional field map extraction model to extract and output the corresponding first field map; and a non-first field map extraction submodule, used to output a corresponding first initial field map based on the first deep learning network model according to two input candidate phase values ​​and the first field map extracted by the previous field map extraction submodule, and then input the conventional field map extraction model to extract and output the corresponding first field map, so that the final field map extraction submodule can output the initial field map.

7. A water-fat separation terminal for magnetic resonance imaging, characterized in that, include: One or more memories and one or more processors; The one or more memories are used to store computer programs; The one or more processors are connected to the memory and are used to run the computer program to perform the method as described in any one of claims 1-5.

Images