System optimization method for k-space sampling and filling based on miniature nuclear magnetic equipment

Abstract

The invention discloses a system optimization method for k-space sampling and filling based on miniature nuclear magnetic equipment, and the method comprises the steps: S1, carrying out the k-space sampling of magnetic resonance data collected by the miniature nuclear magnetic equipment according to a pre-configured adaptive sampling template, and generating a sampling data set; s2, calling a corresponding sampling optimization set from a pre-configured feedback adjustment library based on the characteristics of the sampling data set; and S3, dynamically adjusting parameters of the adaptive sampling template based on the sampling optimization set to obtain an optimized k-space sampling data set. Through adaptive sampling and dynamic adjustment of a sampling template, a k-space sampling filling strategy is optimized, so that the acquisition data volume required by imaging is reduced, the image reconstruction process is accelerated, and the overall imaging time is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of k-space sampling, and particularly to a system optimization method for k-space sampling filling based on a micro nuclear magnetic device. Background Art

[0002] Magnetic Resonance Imaging (MRI) technology is an important part of contemporary medical imaging. Images are reconstructed by collecting magnetic resonance data and performing k-space positioning encoding, filling, and Fourier transform decoding. With the progress of related technologies such as superconducting technology, cryogenic technology, magnet technology, electronic technology, and computers, the speed and spatial resolution of magnetic resonance imaging have been continuously improved, and its clinical applications have become increasingly widespread.

[0003] However, the relatively long data acquisition time has always been the biggest drawback of magnetic resonance imaging technology. Shortening the imaging time can not only improve the examination efficiency and patient comfort, reduce motion artifacts, but also is the key to realizing dynamic imaging such as cardiovascular examinations, functional imaging, and real-time interventional surgeries. Researchers have been committed to developing high-switching-rate gradient magnetic fields and fast scanning sequences to improve the imaging speed, but this method has basically reached its limit due to problems such as high manufacturing costs, easy stimulation of patients' nerves and muscles, and generation of eddy current interference.

[0004] How to optimize the k-space sampling filling system, achieve fast and high-quality magnetic resonance imaging, and at the same time make full use of the characteristics of the micro nuclear magnetic device has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The present invention provides a system optimization method for k-space sampling filling based on a micro nuclear magnetic device to solve the above problems existing in the prior art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A system optimization method for k-space sampling filling based on a micro nuclear magnetic device, comprising:

[0008] S1: Performing k-space sampling on the magnetic resonance data collected by the micro nuclear magnetic device according to a pre-configured adaptive sampling template to generate a sampling data set;

[0009] S2: Retrieving a corresponding sampling optimization set from a pre-configured feedback adjustment library based on the characteristics of the sampling data set;

[0010] S3: Dynamically adjusting the parameters of the adaptive sampling template based on the sampling optimization set to obtain an optimized k-space sampling data set.

[0011] Among them, step S1 includes:

[0012] S11: Perform k-space sampling on magnetic resonance data according to a pre-configured adaptive sampling template, and perform grouped sampling on the magnetic resonance data at a preset time interval to generate multiple groups of sampling data;

[0013] S12: Arrange the multiple groups of sampling data in the acquisition order to generate a sampling data set.

[0014] Among them, step S2 includes:

[0015] S21: Based on the characteristics of the sampling data set, including the signal intensity at the k-space center, the edge signal distribution, and the sampling point density, retrieve the corresponding sampling optimization set from a pre-configured feedback adjustment library;

[0016] The feedback adjustment library is constructed by analyzing the acquisition and reconstruction effects of historical magnetic resonance data and contains optimization values based on device parameters and imaging targets.

[0017] Among them, step S3 includes:

[0018] S31: Dynamically adjust the parameters of the adaptive sampling template according to the optimization values in the sampling optimization set;

[0019] The parameters of the adaptive sampling template include the sampling density at the k-space center, the edge sampling strategy, and the sampling point distribution.

[0020] Among them, step S11 includes:

[0021] S111: Generate each group of sampling data;

[0022] S112: Extract the characteristics of each group of sampling data, including the signal intensity at the k-space center, the edge signal distribution, and the sampling point density.

[0023] Among them, step S22 includes:

[0024] [[ID=३६]]S221: Group the historical magnetic resonance sampling data according to the device parameter set and the imaging target;

[0025] S222: Perform segmented association on the sampling data within the group and compare and analyze it with the standard reconstructed image quality;

[0026] S223: Based on the mean difference between the sampling density and the image quality, query the corresponding table of density difference and sampling parameters to generate optimization values.

[0027] Among them, it also includes:

[0028] S4: Perform dictionary learning on the optimized k-space sampling data set to generate a dictionary atom set;

[0029] S5: Based on the dictionary atom set, retrieve the corresponding dictionary optimization set from a pre-configured optimization library;

[0030] S6: Adjust the hyperparameters of dictionary learning based on the dictionary optimization set to optimize sparse representation.

[0031] Among them, step S4 includes:

[0032] S41: Perform dictionary learning on the optimized k-space sampling data set according to a pre-configured deep learning model, and learn a set of dictionary atoms.

[0033] S42: Use the learned set of dictionary atoms as input for subsequent sparse representation and image reconstruction.

[0034] Among them, step S5 includes:

[0035] S51: Retrieve the corresponding dictionary optimization set from a pre-configured optimization library based on the characteristics of the set of dictionary atoms.

[0036] The optimization library is constructed through the training data analysis of the deep learning model in magnetic resonance reconstruction and contains optimization values based on model parameters.

[0037] Among them, step S6 includes:

[0038] S61: Adjust the hyperparameters of dictionary learning according to the optimization values in the dictionary optimization set.

[0039] The hyperparameters include the learning rate, sparse regularization parameter, and number of iterations to optimize the quality of sparse representation and the signal-to-noise ratio of the reconstructed image.

[0040] Compared with the prior art, the present invention has the following advantages:

[0041] A system optimization method for k-space sampling filling based on a micro nuclear magnetic device, including: S1: Perform k-space sampling on the magnetic resonance data collected by the micro nuclear magnetic device according to a pre-configured adaptive sampling template to generate a sampling data set; S2: Retrieve the corresponding sampling optimization set from a pre-configured feedback adjustment library based on the characteristics of the sampling data set; S3: Dynamically adjust the parameters of the adaptive sampling template based on the sampling optimization set to obtain an optimized k-space sampling data set. By means of adaptive sampling and dynamic adjustment of the sampling template, the k-space sampling filling strategy is optimized, thereby reducing the amount of collected data required for imaging, accelerating the image reconstruction process, and further shortening the overall imaging time.

[0042] Other features and advantages of the present invention will be described in the subsequent specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification, claims, and drawings.

[0043] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0044] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. In the accompanying drawings:

[0045] Figure 1 is a flowchart of a system optimization method for k-space sampling filling based on a micro nuclear magnetic resonance device in an embodiment of the present invention;

[0046] Figure 2 is a flowchart of generating a sampling data set in an embodiment of the present invention;

[0047] Figure 3 is a flowchart of retrieving a corresponding sampling optimization set in an embodiment of the present invention. Detailed Embodiments

[0048] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0049] An embodiment of the present invention provides as Figure 1 shown, a system optimization method for k-space sampling filling based on a micro nuclear magnetic resonance device, including:

[0050] S1: Perform k-space sampling on the magnetic resonance data collected by the micro nuclear magnetic resonance device according to a pre-configured adaptive sampling template to generate a sampling data set;

[0051] S2: Retrieve a corresponding sampling optimization set from a pre-configured feedback adjustment library based on the characteristics of the sampling data set;

[0052] S3: Dynamically adjust the parameters of the adaptive sampling template based on the sampling optimization set to obtain an optimized k-space sampling data set.

[0053] The working principle of the above technical solution is: S1. Perform k-space sampling on the magnetic resonance data collected by the micro nuclear magnetic resonance device according to a pre-configured adaptive sampling template to generate a sampling data set

[0054] In magnetic resonance imaging technology, the generation of images depends on the acquisition and processing of magnetic resonance data, and the core of this process is the k-space. The k-space is not a real physical space, but a virtual space used to store magnetic resonance signal data. These signal data are derived from the responses generated by the nuclei in human tissues excited by a magnetic field and radiofrequency pulses in a micro nuclear magnetic device. By performing mathematical processing, such as Fourier transform, on the data in the k-space, we can reconstruct clear magnetic resonance images to display the internal structure of the human body. A micro nuclear magnetic device is a miniaturized magnetic resonance imaging device, usually designed for portable diagnosis or specific research scenarios. Due to its limited hardware resources, how to efficiently acquire k-space data becomes a key issue. To solve this problem, we first use a pre-configured adaptive sampling template to guide the sampling process of the k-space. The adaptive sampling template is a flexible sampling strategy that is pre-designed according to the performance of the device and the imaging target, and determines which positions in the k-space need to collect data and the sampling method. This template is not fixed, but can adjust the positions and densities of the sampling points according to the task requirements to find a balance between image quality and acquisition efficiency.

[0055] In specific operations, the micro nuclear magnetic device will select specific points in the k-space for data acquisition according to the instructions of the adaptive sampling template. These points are concentrated in the central region of the k-space because the data there has the greatest impact on the overall contrast and basic structure of the image, while the points in the edge region are sampled less to save time. In this way, the device completes a complete k-space sampling and generates a sampling data set. This sampling data set contains all the acquired magnetic resonance signal data, laying a foundation for subsequent optimization and image reconstruction.

[0056] S2. Retrieve the corresponding sampling optimization set from the pre-configured feedback adjustment library based on the characteristics of the sampling data set

[0057] After obtaining the preliminary sampling data set, we need to conduct an in-depth analysis to determine whether the current sampling strategy can meet the imaging requirements. This process begins with the extraction of the characteristics of the sampling data set. These characteristics include the distribution uniformity of the data, the intensity variation of the signals, and the presence of obvious missing areas, etc. By analyzing these characteristics, we can understand the problems existing in the sampling process. For example, insufficient data in certain areas leads to blurred image details, or redundant data in some areas wastes the acquisition resources. To solve these problems targeted, we introduce a pre-configured feedback adjustment library. This library is like an intelligent reference manual, which stores a variety of preset sampling optimization strategies, and each strategy corresponds to a specific data feature pattern. These strategies are obtained through long-term experimental accumulation or theoretical derivation, aiming to deal with different sampling situations. For example, if the data points in a certain area are too sparse, the feedback adjustment library will recommend increasing the sampling density; if the data in some areas is already sufficient, the feedback adjustment library may recommend reducing the sampling to improve efficiency.

[0058] In this step, we first extract the characteristics of the sampling data set. This involves signal processing techniques to quantify the current sampling effect by calculating and comparing data characteristics. Then, the extracted characteristics are compared with the records in the feedback adjustment library to find the most matching pattern and retrieve the corresponding sampling optimization set from it. This optimization set contains specific adjustment suggestions for the sampling strategy, such as adding sampling points in which areas, or adjusting the sampling order and range. In this way, we obtain an optimization plan for the current sampling data set and prepare for the next adjustment.

[0059] [[ID=~6]]S3 Based on the sampling optimization set, dynamically adjust the parameters of the adaptive sampling template to obtain an optimized k-space sampling data set

[0060] After obtaining the sampling optimization set, we enter the crucial stage of optimization. The goal of this stage is to dynamically adjust the parameters of the adaptive sampling template according to the guidance of the optimization set, thereby improving the overall effect of k-space sampling. The parameters of the adaptive sampling template include the specific positions of the sampling points, the density of sampling, and the data acquisition path, etc. These parameters determine how the device collects data in k-space and directly affect the quality of the final image. During the specific implementation, we modify the template according to the suggestions of the sampling optimization set. If the optimization set indicates insufficient sampling in a certain area, we will adjust the template to increase the sampling points in that area to ensure clearer image details; if there is excessive sampling in some areas, we will reduce the corresponding sampling points to avoid waste of resources. In addition, the sampling path is also optimized according to the guidance of the optimization set, for example, by adjusting the acquisition order to reduce the device switching time or improve data consistency. After completing these adjustments, we use the optimized adaptive sampling template to resample k-space to generate a new sampling data set. This optimized data set has been significantly improved compared to the previous one, manifested as stronger signals, more uniform coverage, or more efficient acquisition processes. Finally, these improvements will be reflected during image reconstruction, helping the micro nuclear magnetic resonance device to generate higher-quality magnetic resonance images to meet the needs of diagnosis or research.

[0061] The beneficial effects of the above technical solution are: Through the adaptive sampling template and the dynamic adjustment mechanism, the efficiency of k-space sampling and the quality of image reconstruction are significantly improved. The system can optimize the sampling strategy according to the characteristics of the actually acquired magnetic resonance data. Especially in the micro nuclear magnetic resonance device, this optimization effectively reduces the scanning time and energy consumption, while maintaining or improving the image quality, providing an efficient solution for resource-constrained application scenarios.

[0062] In another embodiment, as Figure 2 shown, step S1 includes:

[0063] S11: Perform k-space sampling on the magnetic resonance data according to the pre-configured adaptive sampling template, and sample the magnetic resonance data at a preset time interval to generate multiple groups of sampling data;

[0064] S12: Arrange the multiple groups of sampling data in the acquisition order to generate a sampling data set.

[0065] The working principle of the above technical solution is: S11, adaptive k-space sampling

[0066] The k-space sampling of magnetic resonance data is performed through a pre-configured adaptive sampling template, which differentially samples the k-space according to different structural features. Under the combined action of radiofrequency pulses and slice selection gradients, nuclear magnetic resonance phenomena occur simultaneously for all the acquisition information on the imaging plane. At this time, although the imaging voxels at different spatial positions have the same magnetization intensity vector, spatial encoding is still required to distinguish the signals at different positions. The adaptive sampling template densely samples the central region of the k-space by dynamically adjusting the sampling density, while sparsely sampling the edge region of the k-space to balance the data acquisition efficiency and image quality. During the sampling process, the system groups and samples the magnetic resonance data at a preset time interval, and each group of sampling corresponds to a specific time point. This grouped sampling method can capture the characteristics of magnetic resonance signals that change over time, especially suitable for dynamic imaging scenarios. The system first applies a phase encoding gradient to make the atomic nuclei have different initial phases along a specific direction, and then applies a frequency encoding gradient to make the atomic nuclei show different precession frequencies along another direction, thereby achieving two-dimensional spatial positioning. Each time sampling is performed, the system adjusts the parameters of the phase encoding gradient and the frequency encoding gradient according to the guidance of the sampling template to ensure that different regions of the k-space are appropriately sampled. When the sampling reaches the preset time interval, the system will automatically switch to the next group of sampling and adjust the sampling trajectory according to the adaptive template. This time-segmented sampling mechanism can reflect the magnetic resonance characteristics at different time points and provide information in the time dimension for subsequent image reconstruction. During the sampling process, the radiofrequency receiving coil receives the magnetic resonance signals from different voxels, and these signals are pre-amplified and analog-to-digital converted to form the original k-space data points. The system performs preliminary processing on each group of sampling data, including phase correction and amplitude normalization, to eliminate the influence of system errors on subsequent processing.

[0067] S12. Generation of Sampling Data Set

[0068] The processing of multiple groups of sampling data first requires sorting them in the order of acquisition time to maintain the temporal continuity and consistency of the data. The system marks each group of sampling data according to the acquisition timestamp and constructs a time-series index table to ensure the accuracy of data sorting. The sorting process takes into account the echo time and repetition time parameters in magnetic resonance imaging, enabling the sampling data to correctly reflect the relaxation characteristics of the tissue structure. After the data arrangement is completed, the system combines these ordered sampling data into a complete sampling data set, which contains the complete k-space information and its temporal evolution characteristics. During the data set generation process, the system performs a data integrity check to ensure that each region of the k-space is covered by sufficient sampling points. For regions with insufficient sampling, the system uses an interpolation algorithm to supplement the data to ensure the uniformity of k-space sampling. The sampling data set also contains sampling trajectory information, recording the exact position of each data point in the k-space, which is particularly important for non-Cartesian sampling. To meet different tissue contrast requirements, the system marks the data with different echo times in the sampling data set for subsequent multi-echo image reconstruction. At the same time, the sampling data set also records gradient field information, including the time-varying curves of the gradient fields in each direction, which is crucial for correcting gradient nonlinearity and eddy current effects. Before the final formation of the data set, the system performs a data quality assessment, calculating metrics such as signal-to-noise ratio and artifact level to ensure that the sampling data meets the quality requirements for image reconstruction. The finally generated sampling data set will serve as the data basis for magnetic resonance image reconstruction and be converted into a spatial-domain image through algorithms such as Fourier transform, thus realizing the visual expression of the tissue.

[0069] The beneficial effects of the above technical solution are as follows: Through the technical features of grouped sampling and sorting according to the acquisition order, the system can capture the temporal dynamic changes in the magnetic resonance signal. Especially in dynamic imaging and functional magnetic resonance imaging, this mechanism improves the temporal resolution and accuracy of the image, providing reliable support for applications that require high temporal precision.

[0070] In another embodiment, as Figure 3 shown, step S2 includes:

[0071] S21: Based on the characteristics of the sampling data set, including the signal intensity at the k-space center, the edge signal distribution, and the sampling point density, retrieve the corresponding sampling optimization set from the pre-configured feedback adjustment library;

[0072] S22: The feedback adjustment library is constructed through the analysis of the acquisition and reconstruction effects of historical magnetic resonance data and contains optimization values based on device parameters and imaging targets.

[0073] The working principle of the above technical solution is as follows: S21: Feedback adjustment based on the characteristics of the sampling data set is a key link in magnetic resonance imaging optimization. During this process, the system will perform real-time analysis on multi-dimensional characteristics such as the signal intensity at the center of the k-space, the edge signal distribution, and the sampling point density. When the magnetic resonance device acquires the original data, the system immediately evaluates the quality indicators of the sampling data, including parameters such as signal-to-noise ratio, uniformity, and coverage. The system extracts the signal intensity value in the central region of the k-space through a feature extraction algorithm. This region contains the low-frequency information of the image and has a decisive impact on the overall imaging quality. At the same time, it analyzes the signal distribution in the edge region of the k-space, and the edge region corresponds to the high-frequency detail part of the image. In addition, it also calculates the distribution density of the sampling points in the k-space to ensure the rationality of the sampling strategy. Based on the above feature analysis results, the system retrieves the sampling optimization set that best matches the current sampling characteristics from the pre-configured feedback adjustment library. The system uses a similarity calculation algorithm to match the feature vectors and selects the optimal combination of optimization parameters according to the matching degree. When it is detected that the signal intensity at the center of the k-space is lower than the preset threshold, the system will automatically retrieve the corresponding radio frequency pulse enhancement optimization set from the feedback adjustment library. When the edge signal distribution is uneven, the system will retrieve the gradient field equalization optimization set. When the sampling point density is insufficient, the system will retrieve the sampling trajectory optimization set. The feedback adjustment process is an adaptive mechanism that can dynamically adjust the optimization parameters according to different anatomical structures and imaging objectives to ensure the best image quality under various imaging conditions.

[0074] S22: The feedback adjustment library is a knowledge base system constructed through systematic analysis of the acquisition and reconstruction effects of historical magnetic resonance data. The construction process of this library is first based on a large amount of historical patient scan data, including the acquisition results of different anatomical regions, different physiological states, and different imaging sequences. The system uses machine learning algorithms to classify and organize the historical data, extracting the k-space features of each group of data and the corresponding reconstructed image quality scores. For each group of historical data, the system records the device parameter configurations during acquisition, including key information such as magnetic field strength, gradient system performance, coil configuration, and sampling sequence; at the same time, it also records the imaging target information, such as anatomical region, tissue type, and clinical diagnosis requirements. Based on the paired analysis of device parameters and imaging targets, the system constructs a parameter-effect association model, thereby deriving an optimal set of optimization values. To improve the adaptability of the feedback adjustment library, the system also records the changes in image quality before and after optimization, quantifies the optimization effect through comparative analysis, and continuously optimizes the optimization parameter set. The feedback adjustment library supports incremental learning functions and can continuously absorb new scanning cases and optimization experiences to keep the library content dynamically updated. When new acquisition devices are added or imaging technologies are updated, the system automatically activates the library expansion module to construct corresponding parameter optimization sets for the new scenarios. The feedback adjustment library also includes an abnormal situation handling mechanism. When encountering rare k-space feature patterns, the system generates temporary optimization parameters through interpolation or approximation algorithms and continuously improves them in subsequent use. Through the dynamic optimization mechanism of the feedback adjustment library, the magnetic resonance imaging system can continuously improve the quality of reconstructed images and meet the requirements of different clinical application scenarios.

[0075] The beneficial effects of the above technical solution are as follows: By using the feedback adjustment library constructed based on historical data, intelligent adjustment of the sampling strategy is achieved. The system can automatically retrieve the optimization set according to the characteristics of the sampling data set, reducing the need for manual intervention, while ensuring the optimized adaptability of the sampling strategy and improving the overall imaging performance and stability.

[0076] In another embodiment, step S3 includes:

[0077] S31: Dynamically adjust the parameters of the adaptive sampling template according to the optimization values in the sampling optimization set;

[0078] The parameters of the adaptive sampling template include the sampling density at the k-space center, the edge sampling strategy, and the sampling point distribution.

[0079] Among them, step S31 includes:

[0080] S311: Based on the data acquisition of the micro-nuclear magnetic device, obtain the sampling optimization set, which contains the k-space sampling positions and their corresponding optimization values; compare the optimization values with the preset threshold range. When the optimization values exceed the threshold range, trigger the parameter adjustment mechanism of the adaptive sampling template;

[0081] S312: Perform time-domain analysis on the data in the sampling optimization set to extract key features reflecting image quality changes. The key features include signal-to-noise ratio, contrast change index, and edge sharpness value. Input the key features into a pre-trained parameter mapping model. The parameter mapping model adopts a deep neural network structure and establishes a mapping relationship from feature values to sampling parameters through the training of prior sampling data to generate a sampling parameter adjustment scheme.

[0082] S313: Before performing parameter adjustment, perform artifact detection on the sampling optimization set. When artifacts are detected, reduce the amplitude of parameter adjustment to avoid unstable image quality caused by overcompensation.

[0083] S314: According to the sampling parameter adjustment scheme, adjust the parameters of the adaptive sampling template. Perform parameter verification by generating a predicted image through simulated sampling and rapid reconstruction, and calculate the similarity between the predicted image and the target image. When the similarity is lower than the preset threshold, repeat steps S312 to S314 until the similarity reaches the preset threshold or the maximum number of iterations is reached.

[0084] S315: According to the current imaging sequence type, select the corresponding parameter adjustment strategy from the pre-configured policy library. The imaging sequence types include T1-weighted, T2-weighted, or diffusion-weighted imaging. When a change in the patient's body position or physiological state is detected, increase the frequency of parameter adjustment to adapt to changes in sampling requirements. For long scanning sequences, perform parameter adjustment multiple times during the scanning process to compensate for time-related system drift and changes in the patient's state.

[0085] The working principle of the above technical solution is as follows: S31: The parameter dynamic adjustment technology of the adaptive sampling template is applied to the magnetic resonance imaging process. When the system obtains the sampling optimization set, it compares the optimization value corresponding to each position recorded in the optimization set with the pre-configured threshold range. When the optimization value exceeds the threshold range, the system will automatically trigger the parameter adjustment mechanism of the sampling template. During the parameter adjustment process, first, perform time-domain analysis on the data in the optimization set to extract the key features reflecting the change in image quality in the optimization set. These features usually include signal-to-noise ratio, contrast change index, and edge sharpness value, etc. The system inputs these feature values into the pre-trained parameter mapping model, which can automatically generate the optimal sampling template parameter adjustment scheme based on the current image features. The parameter mapping model adopts a deep neural network structure and establishes the mapping relationship between the feature values and the sampling parameters through the training of a large number of prior sampling data; due to the existence of varying degrees of motion artifacts or noise interference in the magnetic resonance imaging process, before performing parameter adjustment, the system will perform artifact detection on the optimization set. When obvious artifacts are detected, the system will correspondingly reduce the amplitude of parameter adjustment to avoid unstable image quality caused by overcompensation. After the parameter dynamic adjustment is completed, the system will perform a parameter verification step, generate a predicted image through simulated sampling and rapid reconstruction, compare the predicted image with the target image, and calculate the similarity index; when the similarity is lower than the preset threshold, the system will re-perform the parameter adjustment step until a satisfactory similarity is achieved or the maximum number of iterations is reached. For different types of imaging sequences, such as T1-weighted, T2-weighted, or diffusion-weighted imaging, the system uses different parameter adjustment strategies, which are stored in the pre-configured strategy library, and the system automatically selects the most suitable strategy according to the current imaging protocol. When it is detected that the patient's body position changes or the physiological state changes, the system will increase the frequency of parameter adjustment to adapt to the change in sampling requirements caused by these changes; for long scanning sequences, the system will perform parameter adjustment multiple times during the scanning process to compensate for time-related system drift and patient state changes.

[0086] S32: The parameters of the adaptive sampling template include the sampling density at the k-space center, the edge sampling strategy, and the sampling point distribution. These parameters directly affect the image quality and acquisition efficiency during magnetic resonance imaging. The sampling density at the k-space center determines the sampling degree of low-frequency information, and this parameter is automatically configured by analyzing the tissue contrast requirements of the target region. When a higher contrast is required in the target region, the system increases the sampling density in the k-space center region to achieve more refined low-frequency information acquisition by increasing the number of sampling lines or reducing the sampling interval. The definition of the k-space center region is based on the frequency response characteristics. Usually, the central region where the energy concentration reaches a certain percentage of the total energy is defined as the k-space center, and this percentage can be dynamically adjusted according to different clinical applications. The edge sampling strategy is responsible for the acquisition of high-frequency information and directly affects the edge sharpness and detail performance of the image. The system analyzes the structural complexity of the target region. For regions containing fine structures, it increases the sampling coverage rate in the edge region. Edge sampling uses a variable density sampling method, that is, the sampling density gradually decreases from the k-space center outward, and the decreasing rate is adaptively adjusted according to the spectral distribution characteristics of the image features. The system also considers the relaxation characteristics of tissues. For tissues with a short T2, it increases the sampling speed in the edge region to reduce the influence of signal attenuation. For regions where motion may occur, the system applies a motion compensation strategy to adjust the sampling order in the edge region and preferentially acquire data in the motion-sensitive direction. The sampling point distribution refers to the geometric arrangement of the k-space sampling points and directly determines the shape and efficiency of the sampling trajectory. The system supports multiple distribution modes, including Cartesian sampling, radial sampling, and spiral sampling, etc., and automatically selects the optimal distribution mode according to the imaging object and the characteristics of the imaging sequence. The sampling point distribution also considers the physical limitations of the gradient system, such as the maximum gradient intensity and slope, and avoids overloading the gradient system by optimizing the sampling trajectory. For dynamic imaging that requires high temporal resolution, the system adopts a time-interleaved sampling strategy to improve the sampling efficiency by interleaving the sampling of different regions of the k-space between consecutive time frames. The system also applies specially optimized sampling point distribution templates according to the characteristics of different anatomical regions, such as the brain, heart, or liver, etc. These templates integrate a large amount of clinical experience and data-driven optimization results. For complex regions containing multi-tissue interfaces, the system increases the sampling density along the interface direction to improve the imaging quality of the interface. At the same time, the system also monitors hardware performance parameters, such as gradient temperature and stability, and adjusts the sampling point distribution when necessary to protect the device and maintain the image quality.

[0087] The beneficial effects of the above technical solutions are as follows: By dynamically adjusting parameters such as the sampling density at the k-space center, the edge sampling strategy, and the sampling point distribution, the system can optimize the sampling strategy according to specific imaging requirements. This flexibility significantly improves the image quality in scenarios that require high contrast or high resolution and enhances the application adaptability of the micro nuclear magnetic device.

[0088] In another embodiment, step S11 includes:

[0089] S111: Generate each group of sampling data;

[0090] S112: Extract the features of each group of sampling data, including the signal intensity at the center of k-space, the edge signal distribution, and the sampling point density.

[0091] The working principle of the above technical solution is as follows: In the data sampling process, it is first necessary to generate each group of sampling data. This process is carried out in k-space, where k-space refers to the frequency domain space after Fourier transform and is used to represent the distribution characteristics of signals in the frequency domain. The data sampling process adopts a grouping strategy, dividing data with different frequency ranges and different spatial positions into multiple sampling groups, and each sampling group has its unique sampling distribution characteristics. The system determines the number of sampling points for each group of sampling through a pre-configured sampling density threshold (any value generally in the range of 30% to 70%). When the sampling density is lower than the threshold, the system will automatically increase the number of sampling points to ensure data quality; when the sampling density exceeds the upper threshold, the number of sampling points will be appropriately reduced to improve sampling efficiency. The generation of sampling points also takes into account the characteristics of the gradient waveform, and the k-space trajectory is controlled by adjusting the amplitude and duration of the gradient waveform, so as to achieve targeted sampling of different regions. The sampling order of the data can be radial, Cartesian, or spiral, and different sampling orders correspond to different k-space coverage patterns. The system will dynamically select the optimal sampling order according to actual needs. During the process of generating each group of sampling data, the sampling trajectory will also be smoothed to ensure the continuity of the transition between sampling points and avoid artifacts caused by sudden changes in the sampling trajectory. The noise encountered during the sampling process is processed by a pre-configured noise suppression algorithm. When the noise intensity exceeds a certain proportion of the signal intensity, the system will automatically trigger the resampling mechanism to re-collect the data in this area.

[0092] After the sampling data generation is completed, it is necessary to extract features from each group of sampling data, including the signal intensity at the center of k-space, the edge signal distribution, and the sampling point density. The signal intensity at the center of k-space reflects the energy distribution of the overall signal and is obtained by calculating the average value of the signal amplitudes of the sampling points in the central region. The edge signal distribution reflects the distribution of high-frequency information, and the system characterizes it by analyzing the signal intensity changes in the edge region of k-space. The sampling point density refers to the number of sampling points in a unit k-space region, and a spatial distribution map is obtained by dividing k-space into multiple sub-regions and calculating the number of sampling points in each sub-region. The feature extraction process adopts an adaptive feature selection mechanism. When a significant signal feature is detected in a certain region, the system will automatically increase the feature weight of that region; at the same time, by comparing the feature differences between different groups, a discriminative feature set is identified. Feature extraction also includes phase consistency analysis. By calculating the phase changes between adjacent sampling points, the continuity and consistency of the signal are judged. For the extracted features, the system will establish a feature dictionary and encode the features using the sparse representation method in the compressed sensing theory to achieve dimensionality reduction and compression of the feature space. The data after feature extraction will be used in subsequent image reconstruction or signal analysis processes. The system automatically adjusts the parameters of the reconstruction algorithm according to the quality of the feature data. When the feature data is incomplete or abnormal, a feature re-extraction mechanism is triggered to ensure the accuracy and reliability of subsequent processing. By deeply analyzing the features of the sampling data, the essential characteristics of the signal can be obtained, providing a solid data basis for subsequent applications.

[0093] The beneficial effects of the above technical solution are as follows: By extracting the features of each group of sampling data, including the signal intensity at the center of k-space, the edge signal distribution, and the sampling point density, an accurate basis is provided for subsequent sampling adjustment. The system can more accurately understand the data characteristics, thereby enhancing the pertinence and effectiveness of the sampling strategy and improving the overall effect of image reconstruction.

[0094] In another embodiment, step S22 includes:

[0095] S221: Group the historical magnetic resonance sampling data according to the device parameter set and the imaging target;

[0096] S222: Perform segmented association on the sampling data within the group and conduct a comparative analysis with the standard reconstructed image quality;

[0097] S223: Based on the mean difference between the sampling density and the image quality, query the corresponding table of density difference and sampling parameters to generate an optimized value.

[0098] The working principle of the above technical solution is as follows: Historical magnetic resonance sampling data is grouped according to device parameter sets and imaging targets. During this grouping process, the system automatically identifies the characteristic parameters of different magnetic resonance devices, including device parameters such as field strength, gradient performance, coil configuration, etc., and at the same time classifies and organizes them in combination with the characteristic requirements of different tissue regions such as the brain, abdomen, heart, etc. as imaging targets. Through this grouping method, it can be ensured that the magnetic resonance data collected under similar conditions is grouped into the same group, providing a basis for subsequent data analysis. The system will establish a correspondence matrix between device parameters and imaging targets. Each element in the matrix represents the optimal sampling configuration for a specific imaging target under specific device parameters, and this matrix can be dynamically updated with the continuously accumulated data during the system usage process.

[0099] Segmentally correlate the sampling data within the group and conduct comparative analysis with the standard reconstructed image quality. In the process of segmental correlation, the K-space data is first divided into multiple segments according to the sampling trajectory and density, and each segment represents the information within a specific frequency range. The system evaluates the data integrity and signal intensity of each segment, and calculates the contribution degree of each segment to the final reconstructed image quality. At the same time, the system retrieves the pre-stored standard reconstructed images, including key image quality indicators such as contrast, resolution, signal-to-noise ratio, etc. as reference benchmarks. By calculating the difference between each segment and the corresponding part of the standard image, the system generates a segment quality mapping diagram, clearly showing which K-space regions have insufficient sampling resulting in image quality degradation. This comparative analysis uses an adaptive algorithm that can automatically adjust the quality evaluation weights according to the clinical requirements of different imaging parts.

[0100] Based on the mean difference between the sampling density and the image quality, query the corresponding table of density difference and sampling parameters to generate optimized values. The system first calculates the difference value between the actual sampling density and the ideal density of each K-space segment, and establishes a mathematical correlation model with the degree of image quality degradation corresponding to this segment. Through this correlation model, the system can quantify the impact degree of insufficient sampling in different regions on the overall image quality. The system maintains a dynamically updated corresponding table of density difference and sampling parameters, which is established based on a large amount of historical data statistics and records the optimal compensation parameter settings under different sampling density difference conditions. When querying this corresponding table, the system comprehensively considers the scanning time limit, hardware capabilities, and clinical requirements to generate an optimal combination of optimized values. These optimized values are applied to the subsequent magnetic resonance sampling process under the same conditions, which can optimize the K-space sampling trajectory, adjust the number of phase encoding steps, or change the sampling order, thereby improving the image quality while ensuring the scanning efficiency. When the system detects that the image quality of a certain type of scanning task continuously falls below the expected threshold, it will automatically trigger the retraining process of the deep learning model, update the sampling strategy using the newly accumulated high-quality dataset, and achieve the continuous optimization and iteration of the sampling scheme.

[0101] Perform quality assessment on the execution results of the configured magnetic resonance sampling scheme to determine image clarity and contrast. The system automatically analyzes the reconstructed images through an image quality assessment algorithm and quantitatively evaluates key indicators such as image clarity, contrast, and signal-to-noise ratio. When the image clarity is less than or equal to the preset clarity threshold, the system performs a stepped increase in sampling density based on the pre-configured density enhancement parameter set. When the contrast is less than or equal to the preset contrast threshold, the system performs a stepped contrast enhancement based on the pre-configured sequence optimization parameter set. Adjustment of image clarity judgment is given priority because clarity is usually the basic guarantee for magnetic resonance imaging quality. Both the density enhancement parameter set and the sequence optimization parameter set are stored in association with the current working parameter set, that is, the density enhancement parameter set and the sequence optimization parameter set corresponding to different working parameter sets also vary, so as to improve image quality as much as possible with the least extension of the scanning time.

[0102] When it is determined through clarity that density enhancement is required and after the adjustment is stable, if the contrast is less than or equal to the preset warning threshold, the system outputs an abnormal warning to remind the technician to check the status of the radiofrequency coil or the magnetic field homogeneity. The system will automatically record the working parameters and environmental conditions under abnormal circumstances to form a problem log, which is convenient for equipment maintenance personnel to quickly locate the cause of the failure. Since parameter optimization is carried out on the basis of the magnetic resonance equipment running for a period of time, when a new model of magnetic resonance equipment is first put into use, a reference migration control mode is proposed. The reference migration control mode extracts experience from the historical data of similar equipment to provide initial parameter settings for the new equipment and shortens the commissioning cycle of the new equipment.

[0103] Based on the device characteristic parameter set and the environmental condition parameter set, other magnetic resonance devices within the medical institution are screened to obtain reference objects. The system will analyze device characteristic parameters such as field strength, gradient performance, coil configuration, etc., and combine environmental conditions such as room temperature stability, radiofrequency interference level, etc. to find devices with similar working conditions as optimization references. According to the image clarity and contrast, calculate the imaging quality scores of each reference object. According to the device usage duration and stability, calculate the reliability scores of each reference object. The system will retrieve the working parameter set of the reference object with the highest weighted sum of the imaging quality score and the reliability score as the initial parameter setting of the device to be optimized.

[0104] The initial application stage is when the time span of the historical data obtained meets the minimum period required for constructing the analysis dataset. During the initial application stage, fine-tuning can be carried out according to the density enhancement parameter set and the sequence optimization parameter set to adapt to individual device differences and specific patient needs. The system continuously monitors the image quality differences of the same scanning protocol among different devices, identifies the parameter combinations with the best performance through statistical analysis, and extracts the common features of these combinations to form the best practice templates. These templates are classified according to clinical feedback and case characteristics to construct a dedicated parameter library for different disease diagnoses. The imaging quality score is the sum of the products of clarity and contrast with the pre-configured weight coefficients respectively, and the weight coefficients reflect the dependence of different clinical applications on specific image attributes. The reliability score is the score value determined by querying the pre-configured reliability assessment table based on the device usage duration, and this assessment table is established based on the device aging curve and maintenance records, which can accurately reflect the stability level of the device at different usage stages. Through this multi-device collaborative optimization mechanism, medical institutions can make full use of the usage experience of all devices to achieve the continuous improvement and standardized application of magnetic resonance imaging technology.

[0105] The beneficial effects of the above technical solution are as follows: Based on the grouped analysis of historical data, the relationship between sampling density and image quality is identified, and scientific optimization values are generated. This mechanism provides a reliable reference for the current sampling strategy, improves the accuracy and consistency of sampling optimization, and ensures high-quality imaging results.

[0106] In another embodiment, it further includes:

[0107] S4: Perform dictionary learning on the optimized k-space sampling dataset to generate a dictionary atom set;

[0108] S5: Based on the dictionary atom set, retrieve the corresponding dictionary optimization set from the pre-configured optimization library;

[0109] S6: Based on the dictionary optimization set, adjust the hyperparameters of the dictionary learning to optimize the sparse representation.

[0110] The working principle of the above technical solution is as follows: perform dictionary learning on the optimized k-space sampling data set. First, divide the collected k-space data into blocks to form a training sample matrix. The system automatically sets the initial parameters of dictionary learning, including key parameters such as the number of atoms, sparsity constraint, and convergence threshold. After the training samples are prepared, an online dictionary learning algorithm is used to iteratively analyze the data, and the dictionary is optimized by minimizing the weighted sum of the reconstruction error and sparsity. As the number of iterations increases, the system continuously updates the dictionary atoms until the preset convergence condition or the maximum number of iterations is reached. After optimization, the system generates a dictionary atom set containing multiple atomic elements, and each atomic element represents the basic structural characteristics of the k-space data. During the generation of dictionary atoms, the system will record the usage frequency and reconstruction contribution degree of each atom simultaneously for subsequent fine-tuning of the dictionary. After the dictionary atom set is generated, the system will perform regularization on it to ensure that the correlation between atoms is maintained within a reasonable range and avoid computational waste caused by redundant representations.

[0111] Based on the generated dictionary atom set, the system automatically retrieves the corresponding dictionary optimization set from the pre-configured optimization library. The retrieval process first calculates the feature vector by analyzing the feature distribution of the current dictionary atom set. The system matches the similarity of this feature vector with the feature vectors corresponding to each optimization set stored in the optimization library to find several candidate optimization sets that are the closest. For these candidate optimization sets, the system further evaluates their performance on the current data set, including multiple indicators such as reconstruction accuracy, computational complexity, and convergence speed. On the basis of comprehensive evaluation, the system selects the optimal dictionary optimization set, which contains the optimization parameters and supplementary atoms for the current dictionary atoms. When environmental factors cause changes in sampling conditions, the system will re-evaluate the applicability of the current dictionary optimization set and trigger a new round of optimization set retrieval if necessary. The retrieval process of the dictionary optimization set fully considers the tissue characteristic differences of different anatomical parts, and adopts different optimization strategies for strong signal regions and weak signal regions respectively to ensure the balanced filling of k-space data.

[0112] Based on the retrieved dictionary optimization set, the system adaptively adjusts the hyperparameters of dictionary learning to achieve optimal sparse representation. First, the system adjusts the regularization parameters of sparse coding. When the reconstruction error is large, the sparsity constraint is appropriately reduced; when the reconstruction error is small, the sparsity constraint is strengthened to improve representation efficiency. The system dynamically adjusts the learning rate parameters based on the recommended values in the dictionary optimization set, using a larger learning rate to accelerate convergence in the initial stage, and then gradually reducing the learning rate for fine-tuning. For the correlation constraint parameters between dictionary atoms, the system optimizes based on the correlation matrix in the optimization set to ensure that the atoms maintain appropriate orthogonality and avoid representation redundancy. After the hyperparameter adjustment is completed, the system re-performs the sparse coding calculation and uses the optimized dictionary and parameter settings to sparsely represent the k-space data. During the representation process, the system adopts an adaptive threshold strategy, applying different screening criteria to coefficients in different frequency regions to ensure that both high-frequency details and low-frequency structure are reasonably preserved. The system regularly evaluates the quality of the sparse representation, including calculations of reconstruction error, sparsity, and perceptual quality metrics. When the evaluation results fall below a preset threshold, a new round of hyperparameter adjustments is triggered, forming a closed-loop optimization mechanism. Through multiple rounds of iterative optimization, the system ultimately achieves an efficient sparse representation of k-space data, maximizing the compression of sampled data while ensuring reconstruction quality.

[0113] The beneficial effects of this technical solution include: introducing dictionary learning and optimization libraries, optimizing k-space sampling datasets by generating dictionary atomic sets, and providing an efficient approach for sparse representation and image reconstruction. This technology significantly improves image quality and reconstruction speed, and is particularly suitable for micro-MRI devices with large data volumes and limited computing resources.

[0114] In another embodiment, step S4 includes:

[0115] S41: performing dictionary learning on the optimized k-space sampling data set according to a pre-configured deep learning model to obtain a dictionary atom set;

[0116] S42: The learned dictionary atom set is used as input for subsequent sparse representation and image reconstruction.

[0117] The working principle of the above technical solution is as follows: Preprocess the optimized k-space sampling data set to ensure data quality and integrity; when the data quality index is greater than the preset quality threshold, directly input the data into the pre-configured deep learning model for a complete dictionary learning process; when the data quality index is less than or equal to the preset quality threshold, based on the pre-configured data augmentation strategy, supplement and enhance the k-space data, and then input it into the deep learning model. An optimization algorithm is adopted in the dictionary learning process, which includes two stages: sparse coding and dictionary update, and the optimal dictionary atom set is obtained by minimizing the weighted sum of the reconstruction error and sparsity. The dictionary learning model has pre-set multiple groups of learning parameter sets according to different data characteristics and task objectives, and each group of parameter sets corresponds to a different learning strategy. When processing high-resolution k-space data, a more refined atom division strategy is adopted to generate dictionary atoms with more detailed characterization capabilities; when processing low-resolution or noisy data, a more robust learning strategy is adopted to generate dictionary atoms with stronger generalization capabilities. Regularization constraints are introduced in the dictionary learning process, and the original optimization problem is transformed into a constrained optimization problem through the Lagrange multiplier method, so that the generated dictionary atoms can accurately represent the characteristics of the input data and avoid overfitting. The learned dictionary atom set will be evaluated and screened. When the atom characterization ability is lower than the preset threshold, it will be eliminated; when the redundancy of the atom set is higher than the preset threshold, a merging operation will be performed to reduce redundancy; when the atom set characterization ability does not cover the key features, new basic atoms will be introduced to ensure the completeness of the dictionary atom set. After the dictionary learning is completed, the dictionary atom set will be normalized for subsequent sparse representation calculations.

[0118] Import the learned dictionary atom set into the sparse representation module and perform sparse decomposition on the newly input k-space data. First, calculate the correlation between the input data and the dictionary atoms to generate the initial sparse coefficients. Then, optimize the initial coefficients through the iterative shrinkage algorithm, retaining the coefficients with the strongest correlation and setting the remaining coefficients to zero in each iteration. When the number of iterations reaches the preset value or the reconstruction error is less than the preset threshold, the iteration stops and the final sparse coefficients are output. An adaptive threshold selection strategy is adopted during the sparse representation process to dynamically adjust the sparsity parameter according to the characteristics of the input data. When the signal-to-noise ratio of the input data is high, a larger sparsity parameter is adopted to generate a more accurate representation. When the signal-to-noise ratio of the input data is low, a smaller sparsity parameter is adopted to improve the robustness of the representation. To improve the computational efficiency, a block processing strategy is adopted during the sparse representation process, dividing the entire data space into multiple subspaces, calculating the sparse representations of each subspace in parallel, and finally merging them to obtain the complete representation. The image reconstruction process is based on the obtained sparse coefficients and the dictionary atom set, and the original signal is reconstructed through linear combination. First, the sparse coefficients are weighted and summed with the corresponding dictionary atoms to generate the reconstructed basic image. Then, the reconstructed image is mapped back from the transform domain to the spatial domain through inverse transformation. For the artifacts generated during the reconstruction process, an adaptive filtering technique is used to suppress them. Finally, post-processing is performed on the reconstructed image to enhance the image edge and detail features. A quality assessment mechanism is introduced during the reconstruction process. When the reconstruction quality is lower than the preset threshold, the sparse representation parameters are automatically adjusted and the sparse decomposition and reconstruction processes are re-executed until the reconstruction quality meets the requirements or the maximum number of iterations is reached. A multi-scale analysis strategy is integrated into the reconstruction algorithm, adopting different reconstruction strategies for different frequency components. For the low-frequency components, more sparse coefficients are retained to ensure the accurate reconstruction of the basic structure. For the high-frequency components, stricter sparse constraints are adopted, and at the same time, edge-preserving filtering is combined to improve the detail recovery ability. The finally output reconstructed image is normalized and contrast-adjusted to generate a high-quality image that meets the requirements of subsequent analysis.

[0119] The beneficial effects of the above technical solution are as follows: Using a deep learning model for dictionary learning can extract complex patterns and features from the optimized k-space sampling dataset and generate a more effective dictionary atom set. This mechanism improves the accuracy of sparse representation, thereby significantly enhancing the quality and detail performance of image reconstruction.

[0120] In another embodiment, step S5 includes:

[0121] S51: Retrieve the corresponding dictionary optimization set from the pre-configured optimization library based on the characteristics of the dictionary atom set;

[0122] S52: The optimization library is constructed through the training data analysis of the deep learning model in magnetic resonance reconstruction and contains optimization values based on model parameters.

[0123] The working principle of the above technical solution is as follows: Before reconstructing the magnetic resonance image, the reconstruction parameter set is also optimized according to the characteristics of the dictionary atom set. The specific steps of parameter optimization are as follows: Retrieve the corresponding dictionary optimization set from the pre-configured optimization library according to the characteristics of the dictionary atom set. The optimization library is constructed by analyzing the training data in magnetic resonance reconstruction through a deep learning model and contains the optimized values based on the model parameters. The feature extraction of the dictionary atom set first performs discrete transformation processing on the collected k-space data to obtain the transformation coefficients of the scale and direction sub-bands corresponding to the nuclear magnetic resonance image. Analyze the transformation coefficients by setting the sampling rate to determine the characteristic parameter set of each sub-band. Form a feature vector according to the preset arrangement rule for the characteristic parameter set. Retrieve the matching dictionary optimization set from the pre-configured optimization library through the feature vector. Optimize the corresponding parameters in the reconstruction parameter set according to each optimization coefficient in the dictionary optimization set. Among them, the characteristic parameter set includes: parameters representing the spatial frequency distribution, parameters representing the sub-band energy distribution, parameters representing the direction characteristics, parameters representing the scale characteristics, etc. The grouping rule of the characteristic parameters is to divide the parameters at preset spatial frequency intervals to form groups. The optimization library is constructed in advance, and the dictionary optimization set in the optimization library is associated with the characteristic parameter set.

[0124] The construction of the optimization library is carried out by analyzing a large amount of data during the reconstruction of magnetic resonance images. First, the stored magnetic resonance reconstruction data is grouped according to the identification parameter set of the image type and the environmental parameter set of the acquisition environment. Among them, the parameters in the identification parameter set include: parameters representing tissue type, imaging sequence, resolution, etc.; the parameters in the environmental parameter set include: parameters representing magnetic field strength, parameters representing coil configuration, etc. Statistical analysis is performed on the reconstruction data of each magnetic resonance image within the same group, and different types of dictionary atom sets are corresponded to the reconstruction quality index. Determine the average value of the difference between the reconstruction quality and the standard reconstruction quality under the same acquisition conditions and the same reconstruction parameters, and query the pre-configured quality difference and dictionary parameter correspondence table according to the difference to determine the corresponding dictionary optimization value. In this way, an optimization item in the optimization library is obtained. The analysis of the dictionary atom set is carried out based on the frequency domain characteristics, for example, segmented using a pre-configured frequency interval (which can be configured as any interval value between low frequency and high frequency). In the specific operation link of parameter optimization, the frequency range corresponding to the characteristic parameters of the dictionary atom set needs to match the pre-configured frequency interval to ensure the accurate progress of optimization. In order to achieve more accurate extraction of the characteristics of the dictionary atom set, in practical applications, a multi-scale decomposition method is used to process the k-space data to determine the sub-band coefficients at each scale. Integrate the sub-band coefficients and the original characteristic parameters at each scale, and determine the final characteristic parameter set according to the pre-configured characteristic integration database. Among them, integrating the sub-band coefficients and the original characteristic parameters at each scale is to arrange the coefficients and the original characteristic parameters of each scale according to a preset rule to form an integrated characteristic matrix. Determine the final characteristic parameter set according to the pre-configured characteristic integration database, including: matching the integrated characteristic matrix with the standard matrix in the characteristic integration database, and retrieving the characteristic parameter set corresponding to the matching standard matrix as the final characteristic parameter set. The characteristic integration database is constructed in advance through deep learning model training and is obtained through corresponding association analysis based on a large amount of magnetic resonance image reconstruction data and the corresponding quality evaluation results. In the actual application process, monitor and analyze the optimized dictionary parameters to determine the stability and reconstruction quality of the reconstruction. When the stability is less than or equal to the preset stability threshold, perform a step-down based on the pre-configured down-regulation parameter set. When the reconstruction quality is less than or equal to the preset quality threshold, perform a step-up based on the pre-configured up-regulation parameter set. The adjustment based on the judgment of stability is given priority. The down-regulation parameter set and the up-regulation parameter set are both associated and stored with the current dictionary parameter set, that is, the down-regulation parameter set and the up-regulation parameter set corresponding to different dictionary parameter sets are also different, so as to make the adjustment with the least impact while ensuring the reconstruction quality as much as possible. When it is judged as a down-regulation through stability and the adjustment is stable, and the reconstruction quality is less than or equal to the preset alarm threshold, output an alarm to remind the operator to re-examine the quality of the acquisition data or adjust the acquisition parameters.Since the dictionary parameter optimization is carried out based on a certain amount of feature data, when a new image type is reconstructed for the first time, a joint optimization mode is proposed for how to optimize. In practical applications, the optimization method based on the characteristics of the dictionary atom set also includes: screening the existing image reconstruction data based on the identification parameter set and the environment parameter set to obtain reference objects; calculating the first score of each reference object according to the reconstruction stability and reconstruction quality; calculating the second score of each reference object according to the usage frequency; retrieving the dictionary parameter set of the reference object with the largest sum of the first score and the second score as the initial dictionary parameter set for the reconstruction of the new image type; the initial reconstruction stage is for the number of obtained feature data to meet the data required for constructing the analysis data set; fine-tuning can be performed according to the down-regulation parameter set and the up-regulation parameter set in the initial reconstruction stage. Among them, the first score is the sum of the products of the reconstruction stability, the reconstruction quality and the pre-configured weight coefficients respectively; the second score is the score value determined by querying the pre-configured second score determination table according to the usage frequency.

[0125] The beneficial effects of the above technical solutions are as follows: The optimization library retrieves the targeted optimization set according to the characteristics of the dictionary atom set, enabling the system to dynamically adjust the dictionary learning parameters. This adaptive adjustment enhances the applicability and effect of dictionary learning, and further improves the performance and stability of image reconstruction.

[0126] In another embodiment, step S6 includes:

[0127] S61: Adjust the hyperparameters of dictionary learning according to the optimization values in the dictionary optimization set;

[0128] S62: The hyperparameters include the learning rate, the sparse regularization parameter, and the number of iterations to optimize the quality of sparse representation and the signal-to-noise ratio of the reconstructed image.

[0129] The working principle of the above technical solution is as follows: The dictionary optimization set is a data set formed by adaptive learning based on the collected data, which contains multiple reference optimization values. These optimization values provide a guiding direction for optimizing the hyperparameters of dictionary learning. The hyperparameter adjustment process is a dynamic optimization process aimed at improving the sparse representation efficiency and quality of the reconstructed image. The formation of the dictionary optimization set is achieved by analyzing and comparing the results of multiple reconstructions, extracting the parameter variation rules that can improve the image quality, and quantifying these rules into specific optimization values. The process of adjusting the hyperparameters of dictionary learning according to the optimization values in the dictionary optimization set includes key steps such as adaptive parameter estimation, gradient calculation, and step size update. First, parameter sensitivity analysis is carried out. By quantitatively evaluating the impact of different hyperparameter changes on the reconstruction results, the sensitivity coefficient of each hyperparameter is determined. The sensitivity coefficient measures the degree to which the hyperparameter change affects the optimization objective. The higher the sensitivity coefficient, the more significant the impact of the hyperparameter on the reconstruction quality. After determining the sensitivity coefficient, according to the optimization values in the dictionary optimization set, the hyperparameters with higher sensitivity are adjusted in order of priority. For the learning rate parameter, a dynamic adaptive method is used for adjustment. When the reconstruction error shows a downward trend, the learning rate is appropriately increased to accelerate convergence, while when the error fluctuates greatly, the learning rate is decreased to ensure optimization stability. For the sparse regularization parameter, by analyzing the sparsity of the current reconstructed image, the sparse regularization strength is adaptively adjusted on the premise of ensuring the reconstruction quality. When the complexity of the image region is high, the sparse constraint is appropriately reduced, while for the simple-structured region, the sparse constraint is strengthened. For the number of iterations, it is dynamically adjusted based on the change slope of the convergence curve. When the convergence curve tends to be flat, the number of iterations is appropriately reduced to improve the calculation efficiency, while when the error still shows an obvious downward trend, the number of iterations is appropriately increased to further improve the reconstruction quality.

[0130] Hyperparameters include the learning rate, the sparse regularization parameter, and the number of iterations. These parameters jointly affect the quality of sparse representation and the signal-to-noise ratio of the reconstructed image. The process of hyperparameter optimization needs to consider the coupling relationship between parameters and their respective contributions to the optimization objective. First, initialize the hyperparameters, and select a reasonable initial value range based on prior knowledge, including the initial value of the learning rate, the benchmark value of the sparse regularization parameter, and the upper and lower limits of the number of iterations. Then construct a parameter evaluation function, which comprehensively considers multiple factors such as the peak signal-to-noise ratio, structural similarity, and computational complexity of the reconstructed image to form a comprehensive evaluation index. During the optimization process, adopt an adaptive grid search method to dynamically divide the hyperparameter space. Initially, use a larger search step size to roughly scan the entire parameter space to find the sub-region containing the optimal solution. Subsequently, use a finer step size to gradually approach these sub-regions until the preset convergence condition or accuracy requirement is reached. To improve the search efficiency, introduce a gradient-based local search strategy. Based on each grid search, perform a small-range parameter fine-tuning along the gradient direction to find the local optimum. At the same time, to avoid falling into the local optimum, combine the simulated annealing algorithm, and accept the search in the non-optimal direction with a certain probability during the search process. This randomness gradually decreases as the number of iterations increases, ensuring that the algorithm can finally converge to the global optimum or a solution close to the global optimum. For the learning rate parameter, in addition to static adjustment, an adaptive learning rate adjustment mechanism is introduced. By monitoring the change trend of the reconstruction error in several consecutive iterations, dynamically adjust the size of the learning rate. When the error continuously decreases, appropriately increase the learning rate to accelerate convergence; when the error shows fluctuations or an upward trend, then appropriately decrease the learning rate to stabilize the optimization process. For the sparse regularization parameter, by analyzing the frequency domain characteristics and spatial texture complexity of the reconstructed image, adopt a differentiated regularization intensity for different types of image regions. Appropriately reduce the sparse constraint for regions with rich details, while enhance the sparse constraint for smooth regions. This region-adaptive regularization strategy can effectively suppress noise and artifacts while maintaining image details. For the optimization of the number of iterations, design an automatic termination criterion based on the change rate of the error convergence curve. When the change rate of the error in several consecutive iterations is lower than the preset threshold, determine that the algorithm has reached the convergence state and terminate the iteration process in a timely manner to avoid wasting computing resources and improve the overall efficiency. Through this hyperparameter optimization method that combines multiple strategies, the high efficiency and high quality performance of the dictionary learning algorithm in the image reconstruction process are achieved.

[0131] The beneficial effects of the above technical solution are as follows: By adjusting hyperparameters such as the learning rate, the sparse regularization parameter, and the number of iterations, the convergence speed of dictionary learning and the quality of sparse representation are optimized. This refined adjustment significantly improves the signal-to-noise ratio and overall quality of image reconstruction, providing a strong guarantee for high-quality imaging.

[0132] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention.

Claims

1. A system optimization method for k-space sampling filling based on a micro nuclear magnetic device, characterized in that Including: S1: Perform k-space sampling on the magnetic resonance data collected by the micro nuclear magnetic device according to a pre-configured adaptive sampling template to generate a sampling data set. S2: Based on the characteristics of the sampling data set, retrieve the corresponding sampling optimization set from a pre-configured feedback adjustment library. S3: Based on the sampling optimization set, dynamically adjust the parameters of the adaptive sampling template to obtain an optimized k-space sampling data set.

2. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 1, wherein Step S1 includes: S11: Perform k-space sampling on the magnetic resonance data according to a pre-configured adaptive sampling template, and perform grouped sampling on the magnetic resonance data at a preset time interval to generate multiple groups of sampling data. S12: Arrange the multiple groups of sampling data in the acquisition order to generate a sampling data set.

3. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 1, characterized in that, Step S2 includes: S21: Based on the characteristics of the sampling data set, including the signal intensity at the k-space center, the edge signal distribution, and the sampling point density, retrieve the corresponding sampling optimization set from a pre-configured feedback adjustment library. S22: The feedback adjustment library is constructed through the analysis of the acquisition and reconstruction effects of historical magnetic resonance data, and contains optimization values based on device parameters and imaging targets.

4. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 1, characterized in that Step S3 includes: S31: Dynamically adjust the parameters of the adaptive sampling template according to the optimization values in the sampling optimization set. S32: The parameters of the adaptive sampling template include the sampling density at the k-space center, the edge sampling strategy, and the sampling point distribution.

5. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 2, characterized in that Step S11 includes: S111: Generate each group of sampling data. S112: Extract the characteristics of each group of sampling data, including the signal intensity at the k-space center, the edge signal distribution, and the sampling point density.

6. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 3, wherein Step S22 includes: S221: Group the historical magnetic resonance sampling data according to the device parameter set and the imaging target. S222: Perform segmented correlation on the sampling data within the group and compare and analyze it with the standard reconstructed image quality. S223: Based on the mean difference between the sampling density and the image quality, query the corresponding table of density difference and sampling parameters to generate an optimization value.

7. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 1, wherein Also included: S4: Perform dictionary learning on the optimized k-space sampling data set to generate a dictionary atom set. S5: Based on the dictionary atom set, retrieve the corresponding dictionary optimization set from a pre-configured optimization library. S6: Based on the dictionary optimization set, adjust the hyperparameters of the dictionary learning to optimize the sparse representation.

8. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 7, wherein Step S4 includes: S41: Perform dictionary learning on the optimized k-space sampling data set according to a pre-configured deep learning model to learn a dictionary atom set. S42: Use the learned dictionary atom set as input for subsequent sparse representation and image reconstruction.

9. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 7, wherein Step S5 includes: S51: Based on the characteristics of the dictionary atom set, retrieve the corresponding dictionary optimization set from a pre-configured optimization library. S52: The optimization library is constructed through the analysis of the training data of the deep learning model in magnetic resonance reconstruction, and contains optimization values based on model parameters.

10. The system optimization method for k-space sampling filling based on a micro nuclear magnetic device according to claim 7, wherein Step S6 includes: S61: Adjust the hyperparameters of the dictionary learning according to the optimization values in the dictionary optimization set. S62: The hyperparameters include the learning rate, the sparse regularization parameter, and the number of iterations to optimize the quality of the sparse representation and the signal-to-noise ratio of the reconstructed image.

Images