Nuclear magnetic resonance image quality evaluation method and device, equipment and storage medium

Abstract

The invention relates to a nuclear magnetic resonance image quality evaluation method and device, equipment and a storage medium. The method comprises the following steps: collecting nuclear magnetic resonance image data, wherein the nuclear magnetic resonance image data comprises image domain data with artifacts and K space domain data with artifacts; performing artifact mode extraction on the nuclear magnetic resonance image data by using a first neural network, and generating paired artifact data and artifact-free data according to the artifact mode; utilizing a second neural network to carry out image domain and K space domain artifact feature extraction on the pairwise data with and without artifacts to obtain artifact feature parameters; inputting the artifact characteristic parameters into an adaptive hybrid expert network model for quality evaluation, and outputting artifact measurement indexes of the nuclear magnetic resonance image data; wherein the adaptive hybrid expert network model is constructed by using a hybrid expert system and dynamic convolution. According to the invention, the capability of processing artifact images of different grades by a network is improved, so that the image quality evaluation result is more accurate.

Description

Technical Field

[0001] The present application belongs to the field of magnetic resonance imaging technology, and in particular relates to a method, device, equipment and storage medium for evaluating the quality of nuclear magnetic resonance images. Background Art

[0002] Magnetic resonance imaging (MRI) is a widely used medical imaging technology that can depict anatomical structures with excellent soft tissue contrast and high spatial resolution while avoiding radiation. However, there are artifacts on MRI images that are not present in natural images, causing image blur and loss of details, thereby limiting doctors' detection of subtle pathologies and affecting patient treatment. Artifacts in MRI can generally be divided into hardware-related artifacts and human-related artifacts. Hardware-related artifacts are related to wave field inhomogeneity, radio frequency noise or irregularity, chemical shift, ghosting, electromagnetic interference, etc. of the imaging system, and are manifested as spots, ghosts, alternating light and dark bands, folds, zipper-shaped artifacts, etc. on MRI images. Human-related artifacts are mainly due to the fact that most patients find it difficult to remain still during long-term scanning, and the patient's breathing, heartbeat, blood flow and other physiological processes can cause artifacts. Most of these artifacts appear as streak artifacts along the phase encoding direction, which are called motion artifacts.

[0003] Image Quality Assessment (IQA) is one of the basic techniques in image processing. It mainly analyzes the characteristics of the image and then evaluates the quality of the image (the degree of image distortion). Image quality assessment can be divided into subjective assessment and objective assessment in terms of methods. Subjective assessment refers to the use of human subjective perception to evaluate the quality of the image. Objective assessment uses mathematical models to give quantitative values. Image quality assessment is generally divided into three categories according to the amount of information provided by the original reference image: Full Reference-IQA (FR-IQA), Reduced Reference-IQA (RR-IQA) and No Reference-IQA (NR-IQA). FR-IQA has both the original (undistorted, reference) image and the distorted image. The core is to compare the amount of information or feature similarity of the two images. It is less difficult and is a relatively mature research direction. NR-IQA only has distorted images and is more difficult. It has been a research hotspot in recent years and is also the most challenging problem in image quality assessment. RR-IQA only has partial information of the original image or partial features extracted from the reference image. This type of method is between the first two methods. Generally, reference images cannot be provided in practical applications, so NR-IQA is the most practical and has a wide range of applications. Existing traditional FR-IQA methods include Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM). In recent years, many NR-IQA algorithms based on deep learning have been proposed, which can better extract features and keep the results consistent with human subjective judgment, showing better performance than traditional algorithms. However, due to the complex causes of artifacts in nuclear magnetic resonance images, various types of artifacts are shown in the image domain, which is fundamentally different from natural images. Most of the above image quality assessment methods achieve quality assessment by comparing the amount of information or feature similarity of the image. They only consider image domain information and cannot consider k-space information. In addition, the values ​​of some image quality evaluation criteria are not proportional to human subjective feelings, and cannot be applied to the quality assessment of nuclear magnetic resonance imaging. Specifically, motion artifacts are the most common artifacts in nuclear magnetic resonance imaging. In structural imaging studies, motion artifacts manifest themselves in two types: ringing and blurring. Ringing appears as incoherent ghosting, which appears as dark and light ripples in the phase encoding direction. Blurring occurs when data is acquired at multiple locations due to subject motion, resulting in voxels with mixed signals from different tissues. These motions cause the k-space signal to be perturbed, and signals from different tissues that appear at the same location in the scanner at different times during the acquisition process are averaged together in the resulting image, so the extracted value is the average of the signals from the different tissues and does not accurately reflect any one underlying tissue.It can be seen that the fundamental cause of artifacts is signal errors in K space, and it is impossible to use natural image quality assessment methods to evaluate the image quality of MRI. Summary of the invention

[0004] The present application provides a method, device, equipment and storage medium for evaluating the quality of a nuclear magnetic resonance image, aiming to solve at least one of the above-mentioned technical problems in the prior art to a certain extent.

[0005] In order to solve the above problems, this application provides the following technical solutions:

[0006] A method for assessing the quality of a nuclear magnetic resonance image, comprising:

[0007] Collecting nuclear magnetic resonance image data, wherein the nuclear magnetic resonance image data includes image domain data with artifacts and K-space domain data with artifacts;

[0008] Extracting artifact patterns from the nuclear magnetic resonance image data using a first neural network, and generating pairs of artifact-free data according to the artifact patterns;

[0009] Using a second neural network to extract artifact features in the image domain and the K-space domain from the paired artifact-free data to obtain artifact feature parameters;

[0010] The artifact feature parameters are input into an adaptive hybrid expert network model for quality assessment, and an artifact metric index of the nuclear magnetic resonance image data is output; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution.

[0011] The technical solution adopted by the embodiment of the present application also includes: after collecting the nuclear magnetic resonance image data, it also includes:

[0012] An artifact severity assessment is performed on the nuclear magnetic resonance image data, and the nuclear magnetic resonance image data is divided into four artifact levels: no artifact, mild artifact, moderate artifact, and severe artifact.

[0013] The technical solution adopted by the embodiment of the present application also includes: extracting the artifact pattern of the nuclear magnetic resonance image data using the first neural network, specifically:

[0014] The magnetic resonance imaging data after the artifact level classification is cut into a set number of image blocks;

[0015] Using an unsupervised clustering method to classify the image blocks, and obtaining a first type of image blocks without structural boundaries and a second type of image blocks with structural boundaries;

[0016] Artifact pattern extraction is performed according to the classified first-category image blocks and second-category image blocks.

[0017] The technical solution adopted by the embodiment of the present application also includes: using the second neural network to extract artifact features in the image domain and the K-space domain from the paired artifact-free data, specifically:

[0018] The paired artifact data and the artifact-free data are input into a second neural network, and the second neural network uses a network P to extract artifact features from the paired artifact data and output artifact feature parameters. According to the artifact characteristic parameters Calculate the adaptive weighting factor a∈R through network A 1×N , the adaptive weighting factor a can adaptively mix N expert networks,

[0019] The technical solution adopted by the embodiment of the present application also includes: the adaptive hybrid expert network model is constructed in the following manner:

[0020] Through N expert networks E=[E 1 ,E 2 ,...,E N ] and adaptive weighting factor a to construct an adaptive hybrid expert network model in a nonlinear way, where each expert network E i is a lightweight neural network with independent parameters. All E i They share the same network topology and are jointly optimized under the same loss supervision. For each convolution layer C of the adaptive hybrid expert network model, dynamic convolution is used to parameterize the convolution kernel:

[0021]

[0022] Among them, f input and f output Represents input and output respectively, a i represents the i-th value of a, W i C Denotes the expert network E i The C layer parameters, λ is the activation function.

[0023] Another technical solution adopted by the embodiment of the present application is: a nuclear magnetic resonance image quality assessment device, comprising:

[0024] Data acquisition module: used for collecting nuclear magnetic resonance image data, wherein the nuclear magnetic resonance image data includes image domain data with artifacts and K-space domain data with artifacts;

[0025] An artifact pattern extraction module is used to extract the artifact pattern of the nuclear magnetic resonance image data using the first neural network, and generate paired artifact-free data according to the artifact pattern;

[0026] An artifact feature extraction module is used to extract artifact features in the image domain and the K-space domain from the paired artifact-free data using a second neural network to obtain artifact feature parameters;

[0027] Quality assessment module: used for inputting the artifact feature parameters into an adaptive hybrid expert network model for quality assessment, and outputting artifact measurement indicators of the nuclear magnetic resonance image data; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution.

[0028] The technical solution adopted by the embodiment of the present application also includes: the artifact feature extraction module uses a second neural network to extract artifact features of the paired artifact-free data in the image domain and the K-space domain, specifically:

[0029] The paired artifact data and the artifact-free data are input into a second neural network, and the second neural network uses a network P to extract artifact features from the paired artifact data and output artifact feature parameters. According to the artifact characteristic parameters Calculate the adaptive weighting factor a∈R through network A 1×N , the adaptive weighting factor a can adaptively mix N expert networks,

[0030] The technical solution adopted by the embodiment of the present application also includes: the adaptive hybrid expert network model is constructed in the following manner:

[0031] Through N expert networks E = [E 1 ,E 2 ,...,E N ] and adaptive weighting factor a to construct an adaptive hybrid expert network model in a nonlinear way, where each expert network E i is a lightweight neural network with independent parameters. All E i They share the same network topology and are jointly optimized under the same loss supervision. For each convolution layer C of the adaptive hybrid expert network model, dynamic convolution is used to parameterize the convolution kernel:

[0032]

[0033] Among them, f input and f output Represents input and output respectively, a i represents the i-th value of a, W i C Denotes the expert network E i The C layer parameters, λ is the activation function.

[0034] Another technical solution adopted by the embodiment of the present application is: a device, the device includes a processor and a memory coupled to the processor, wherein:

[0035] The memory stores program instructions for implementing the nuclear magnetic resonance image quality assessment method;

[0036] The processor is used to execute the program instructions stored in the memory to control the nuclear magnetic resonance image quality assessment method.

[0037] Another technical solution adopted by the embodiment of the present application is: a storage medium storing program instructions executable by a processor, wherein the program instructions are used to execute the magnetic resonance image quality assessment method.

[0038] Compared with the prior art, the beneficial effects of the embodiments of the present application are as follows: the nuclear magnetic resonance image quality assessment method, device, equipment and storage medium of the embodiments of the present application collect nuclear magnetic resonance image data, the nuclear magnetic resonance image data includes image domain data with artifacts and K space domain data with artifacts, and extracts artifact patterns from the nuclear magnetic resonance image data to generate paired artifact-free data, and then uses a neural network to extract artifact feature parameters from the artifact-free data, and finally, outputs the quality assessment result of the nuclear magnetic resonance image according to the artifact feature parameters through an adaptive hybrid expert network model. The embodiments of the present application establish an adaptive network including a hybrid expert system based on deep learning, which significantly improves the ability of the network to process artifact images of different levels, making the final image quality evaluation result more accurate, more targeted and interpretable. The embodiments of the present application use image domain and K space domain data information at the same time, which can provide richer detail information than a single domain and increase the evaluation accuracy of the model. The embodiments of the present application fit the physical principles of nuclear magnetic resonance imaging and artifact generation, ensure the feasibility, effectiveness and authenticity of the artifact quantification standard, and have very important scientific significance and application prospects for auxiliary medical and diagnostic fields and medical imaging fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of a method for evaluating the quality of a nuclear magnetic resonance image according to an embodiment of the present application;

[0040] Figure 2 is a schematic diagram of the structure of an adaptive hybrid expert network model in an embodiment of the present application;

[0041] Figure 3 This is a schematic diagram of the structure of a nuclear magnetic resonance image quality assessment device according to an embodiment of the present application;

[0042] Figure 4 A schematic diagram of the device structure of an embodiment of the present application;

[0043] Figure 5A schematic diagram of the structure of a storage medium according to an embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] The terms "first", "second", "third" in this application are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first", "second", "third" can expressly or implicitly include at least one of the features. In the description of this application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In the embodiments of this application, all directional indications (such as up, down, left, right, front, back...) are only used to explain the relative position relationship, movement, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication also changes accordingly. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the steps or units listed, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0046] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0047] Specifically, see Figure 1 , is a flow chart of the method for evaluating the quality of a nuclear magnetic resonance image according to an embodiment of the present application. The method for evaluating the quality of a nuclear magnetic resonance image according to an embodiment of the present application comprises the following steps:

[0048] S100: Collecting nuclear magnetic resonance image data; wherein the nuclear magnetic resonance image data includes image domain data with artifacts and K-space domain data with artifacts;

[0049] In this step, the image domain data with artifacts and the K-space domain data with artifacts do not need to correspond one to one. According to the parameters of the magnetic resonance imaging device, the K-space domain data with artifacts can be obtained by inverse Fourier transforming the image domain data with artifacts. It can be understood that the embodiment of the present application uses both K-space domain data and image domain data to ensure the accuracy of artifact image quality assessment.

[0050] S110: Evaluate the severity of artifacts on the MRI image data according to the set evaluation criteria, and divide the MRI image data into four artifact levels: no artifacts, mild artifacts (no impact on diagnosis), moderate artifacts (affecting diagnosis), and severe artifacts (unable to diagnose);

[0051] S120: extracting artifact patterns from the magnetic resonance image data after artifact level classification using the first neural network, and generating pairs of artifact-free and artifact-without data according to the artifact patterns;

[0052] In this step, the artifact pattern extraction method of the first neural network is specifically as follows: first, the input magnetic resonance image data is cropped into a set number of image blocks; secondly, the image blocks are classified using an unsupervised clustering method to obtain a first type of image blocks without structural boundaries (that is, the background is relatively uniform) and a second type of image blocks with structural boundaries; then, artifact patterns are extracted based on the classified first type of image blocks and second type of image blocks; wherein, since the patches in the first type of image blocks all contain a relatively uniform background, the embodiment of the present application extracts a zero-mean artifact pattern by subtracting the average pixel value of each patch; finally, paired artifact data and non-artifact data are generated based on the extracted artifact pattern.

[0053] S130: using a second neural network to extract artifact features in the image domain and the K-space domain for paired artifact data and artifact-free data, and obtaining artifact feature parameters of different artifact levels;

[0054] In this step, the second neural network extracts artifact features in the following way: first, the paired artifact data and the artifact-free data are input into the second neural network, and the second neural network uses a tiny network P to extract artifact features from the paired artifact-free data and outputs artifact feature parameters. Then according to the artifact feature parameters The adaptive weighting factor a∈R is calculated by a tiny network A 1×N , a is adaptive to the input data and can adaptively mix N expert networks, that is,

[0055] S140: inputting the artifact feature parameters into an adaptive hybrid expert network model, wherein the adaptive hybrid expert network model performs quality assessment on the nuclear magnetic resonance image data according to the artifact feature parameters, and outputs artifact measurement indicators of the nuclear magnetic resonance image data; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution;

[0056] In this step, the Mixture of Experts (MoE) is a long-standing neural network that improves network performance by calculating the weighted sum of multiple expert networks. The hybrid expert system separately trains multiple neural networks, each of which is called an expert network (Expert). Each expert network is assigned to be applied to different parts of the data set. The gating module (Managing Neural Net) is used to determine which expert network should be used for an input. The actual output of the model is a weighted combination of the output of each model and the gating module. The embodiment of the present application can improve the model's ability to handle artifacts of varying degrees by using a hybrid expert system and dynamic convolution to construct an adaptive hybrid expert network model and perform quality assessment on magnetic resonance imaging data. Specifically, the adaptive hybrid expert network model is constructed as follows: using N expert networks E=[E 1 ,E 2 ,...,E N ] and adaptive weighting factor a to construct an adaptive hybrid expert network model in a nonlinear way. i is a lightweight neural network with independent parameters. All E i Share the same network topology and jointly optimize under the same loss supervision. For each convolution layer C of the adaptive hybrid expert network model, the embodiment of the present application uses dynamic convolution to parameterize the convolution kernel as follows:

[0057]

[0058] Among them, f input and f output Represent the input and output features respectively, a i represents the i-th value of a, W i C Denotes the expert network E i The C layer parameters, λ is the activation function. The embodiment of the present application adaptively fuses the parameters of each layer in all expert networks to form an adaptive hybrid expert network model E A , the adaptive hybrid expert network model structure is as follows Figure 2 shown.

[0059] Finally, the artifact feature parameters are input into the adaptive hybrid expert network model E A, outputs artifact metrics for MRI image data.

[0060] It can be understood that after appropriate transformation, the embodiment of the present application is also applicable to other types of medical image quality assessment such as CT. After further optimization, the embodiment of the present application can also be applied to artifact correction of MRI and CT images and artifact simulation tasks of magnetic resonance imaging.

[0061] Based on the above, the method for assessing the quality of nuclear magnetic resonance images in the embodiment of the present application collects nuclear magnetic resonance image data, wherein the nuclear magnetic resonance image data includes image domain data with artifacts and K-space domain data with artifacts, and extracts artifact patterns from the nuclear magnetic resonance image data to generate paired data with and without artifacts, and then uses a neural network to extract artifact feature parameters from the data with and without artifacts, and finally, outputs artifact measurement indicators of the nuclear magnetic resonance image according to the artifact feature parameters through an adaptive hybrid expert network model. The embodiment of the present application establishes an adaptive network including a hybrid expert system based on deep learning, which significantly improves the ability of the network to process artifact images of different levels, making the final image quality evaluation result more accurate, more targeted and interpretable. The embodiment of the present application utilizes image domain and K-space domain data information at the same time, which can provide richer detail information than a single domain and increase the evaluation accuracy of the model. The embodiment of the present application fits the physical principles of nuclear magnetic resonance imaging and artifact generation, ensures the feasibility, effectiveness and authenticity of the artifact quantification standard, and has very important scientific significance and application prospects for auxiliary medical and diagnostic fields and medical imaging fields.

[0062] See also Figure 3 , is a schematic diagram of the structure of a nuclear magnetic resonance image quality assessment device according to an embodiment of the present application. The nuclear magnetic resonance image quality assessment device 40 according to an embodiment of the present application comprises:

[0063] Data acquisition module 41: used for collecting nuclear magnetic resonance image data, wherein the nuclear magnetic resonance image data includes image domain data with artifacts and K-space domain data with artifacts;

[0064] An artifact pattern extraction module 42 is used to extract the artifact pattern of the nuclear magnetic resonance image data using a first neural network, and generate pairs of artifact-free data according to the artifact pattern;

[0065] An artifact feature extraction module 43 is used to extract artifact features in the image domain and the K-space domain from the paired artifact-free data using a second neural network to obtain artifact feature parameters;

[0066] The quality assessment module 44 is used to input the artifact feature parameters into the adaptive hybrid expert network model for quality assessment, and output artifact measurement indicators of the nuclear magnetic resonance image data; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution.

[0067] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0068] The device provided in the embodiment of the present application can be applied in the aforementioned method embodiment. For details, please refer to the description of the aforementioned method embodiment, which will not be repeated here.

[0069] See also Figure 4 , is a schematic diagram of the device structure of an embodiment of the present application. The device 50 includes:

[0070] A memory 51 storing executable program instructions;

[0071] A processor 52 connected to the memory 51;

[0072] The processor 52 is used to call the executable program instructions stored in the memory 51 and perform the following steps: collect nuclear magnetic resonance image data, the nuclear magnetic resonance image data including image domain data with artifacts and K-space domain data with artifacts; use a first neural network to extract artifact patterns from the nuclear magnetic resonance image data, and generate paired artifact-free data based on the artifact patterns; use a second neural network to extract artifact features in the image domain and K-space domain from the paired artifact-free data to obtain artifact feature parameters; input the artifact feature parameters into an adaptive hybrid expert network model for quality assessment, and output artifact measurement indicators for the nuclear magnetic resonance image data; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution.

[0073] The processor 52 may also be referred to as a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip having signal processing capabilities. The processor 52 may also be a general-purpose processor, 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, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0074] See also Figure 5, is a schematic diagram of the structure of the storage medium of the embodiment of the present application. The storage medium of the embodiment of the present application stores a program instruction 61 capable of implementing the following steps: collecting nuclear magnetic resonance image data, the nuclear magnetic resonance image data including image domain data with artifacts and K-space domain data with artifacts; using a first neural network to extract artifact patterns from the nuclear magnetic resonance image data, and generating paired artifact-free data according to the artifact patterns; using a second neural network to extract artifact features in the image domain and K-space domain from the paired artifact-free data to obtain artifact feature parameters; inputting the artifact feature parameters into an adaptive hybrid expert network model for quality assessment, and outputting artifact measurement indicators for the nuclear magnetic resonance image data; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution. Among them, the program instruction 61 can be stored in the above-mentioned storage medium in the form of a software product, including a number of instructions for enabling a device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform all or part of the steps of the various implementation methods of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program instructions, or terminal devices such as computers, servers, mobile phones, tablets, etc. Among them, the server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0075] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0076] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. The above is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the description and drawings of this application, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A method for assessing the quality of magnetic resonance images. It is characterized in that include: Collecting nuclear magnetic resonance image data, wherein the nuclear magnetic resonance image data includes image domain data with artifacts and K-space domain data with artifacts; Extracting artifact patterns from the nuclear magnetic resonance image data using a first neural network, and generating pairs of artifact-free data according to the artifact patterns; Using a second neural network to extract artifact features in the image domain and the K-space domain from the paired artifact-free data to obtain artifact feature parameters; The artifact feature parameters are input into an adaptive hybrid expert network model for quality assessment, and an artifact metric index of the nuclear magnetic resonance image data is output; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution.

2. The method for evaluating the quality of a nuclear magnetic resonance image according to claim 1, It is characterized in that After collecting the nuclear magnetic resonance image data, the method further comprises: An artifact severity assessment is performed on the nuclear magnetic resonance image data, and the nuclear magnetic resonance image data is divided into four artifact levels: no artifact, mild artifact, moderate artifact, and severe artifact.

3. The method for evaluating the quality of a nuclear magnetic resonance image according to claim 2, It is characterized in that The extracting of artifact patterns from the nuclear magnetic resonance image data using the first neural network is specifically: The magnetic resonance imaging data after the artifact level classification is cut into a set number of image blocks; Using an unsupervised clustering method to classify the image blocks, and obtaining a first type of image blocks without structural boundaries and a second type of image blocks with structural boundaries; Artifact pattern extraction is performed according to the classified first-category image blocks and second-category image blocks.

4. The method for evaluating the quality of a nuclear magnetic resonance image according to any one of claims 1 to 3, It is characterized in that The use of the second neural network to extract artifact features in the image domain and the K-space domain from the paired artifact-free data is specifically as follows: The paired artifact data and the artifact-free data are input into a second neural network, and the second neural network uses a network P to extract artifact features from the paired artifact data and output artifact feature parameters. According to the artifact characteristic parameters Calculate the adaptive weighting factor a∈R through network A 1×N , the adaptive weighting factor a can adaptively mix N expert networks, 5. The method for evaluating the quality of a nuclear magnetic resonance image according to claim 4, It is characterized in that The adaptive hybrid expert network model is constructed in the following way: Through N expert networks E = [E 1 ,E 2 ,...,E N ] and adaptive weighting factor a to construct an adaptive hybrid expert network model in a nonlinear way, where each expert network E i is a lightweight neural network with independent parameters. All E i They share the same network topology and are jointly optimized under the same loss supervision. For each convolution layer C of the adaptive hybrid expert network model, dynamic convolution is used to parameterize the convolution kernel: Among them, f input and f output Represents input and output respectively, a i represents the i-th value of a, W i C Denotes the expert network E i The C layer parameters, λ is the activation function.

6. A nuclear magnetic resonance image quality assessment device, It is characterized in that include: Data acquisition module: used for collecting nuclear magnetic resonance image data, wherein the nuclear magnetic resonance image data includes image domain data with artifacts and K-space domain data with artifacts; An artifact pattern extraction module is used to extract the artifact pattern of the nuclear magnetic resonance image data using the first neural network, and generate paired artifact-free data according to the artifact pattern; An artifact feature extraction module is used to extract artifact features in the image domain and the K-space domain from the paired artifact-free data using a second neural network to obtain artifact feature parameters; Quality assessment module: used for inputting the artifact feature parameters into an adaptive hybrid expert network model for quality assessment, and outputting artifact measurement indicators of the nuclear magnetic resonance image data; wherein the adaptive hybrid expert network model is constructed using a hybrid expert system and dynamic convolution.

7. The apparatus for evaluating the quality of a nuclear magnetic resonance image according to claim 6, It is characterized in that The artifact feature extraction module uses a second neural network to extract artifact features in the image domain and the K-space domain from the paired artifact-free data, specifically: The paired artifact data and the artifact-free data are input into a second neural network, and the second neural network uses a network P to extract artifact features from the paired artifact data and output artifact feature parameters. According to the artifact characteristic parameters Calculate the adaptive weighting factor a∈R through network A 1×N , the adaptive weighting factor a can adaptively mix N expert networks, 8. The apparatus for evaluating the quality of a nuclear magnetic resonance image according to claim 7, It is characterized in that The adaptive hybrid expert network model is constructed in the following way: Through N expert networks E=[E 1 ,E 2 ,...,E N ] and adaptive weighting factor a to construct an adaptive hybrid expert network model in a nonlinear way, where each expert network E i is a lightweight neural network with independent parameters. All E i They share the same network topology and are jointly optimized under the same loss supervision. For each convolution layer C of the adaptive hybrid expert network model, dynamic convolution is used to parameterize the convolution kernel: Among them, f input and f output Represents input and output respectively, a i represents the i-th value of a, W i C Denotes the expert network E i The C layer parameters, λ is the activation function.

9. A device, It is characterized in that The device comprises a processor and a memory coupled to the processor, wherein: The memory stores program instructions for implementing the method for evaluating the quality of a nuclear magnetic resonance image according to any one of claims 1 to 5; The processor is used to execute the program instructions stored in the memory to control the nuclear magnetic resonance image quality assessment method.

10. A storage medium, It is characterized in that The device stores program instructions executable by a processor, wherein the program instructions are used to execute the method for evaluating the quality of a nuclear magnetic resonance image according to any one of claims 1 to 5.

Images