Magnetic resonance image reconstruction method and apparatus, electronic device, and readable storage medium

Abstract

The application provides a magnetic resonance image reconstruction method and device, electronic equipment and readable storage medium, and relates to the field of magnetic resonance data processing. By acquiring the collected k-space data, the sampling probability of the points in the k-space data is allocated to obtain the probability value corresponding to each point in the k-space data; the k-space data is interpolated according to the probability value to obtain interpolation data, and the k-space data is sampled by using a preset importance map to obtain sampling data; the interpolation data and the sampling data are fused to obtain fusion data, the image is reconstructed by using the fusion data to obtain a reconstructed image; the reconstructed image is evaluated to obtain an evaluation score, when the evaluation score is greater than a preset evaluation threshold, the probability value corresponding to each point in the k-space data is adjusted, the reconstructed image is Fourier transformed to obtain reconstructed data, the importance map is updated by using the reconstructed data, and the interpolation and sampling steps are executed, so that the accuracy of image reconstruction can be improved.

Description

Technical Field

[0001] This application relates to the field of magnetic resonance data processing technology, specifically to a magnetic resonance image reconstruction method, apparatus, electronic device, and readable storage medium. Background Technology

[0002] Magnetic Resonance Imaging (MRI), a non-invasive and radiation-free imaging technique, has wide applications in medical diagnosis and biological research. It works by applying radio frequency pulses of a specific frequency to the human body, exciting the hydrogen nuclei within the body to resonate and release electromagnetic signals. MRI scans then acquire k-space data, which is used for image reconstruction. Image reconstruction is a core component of MRI technology, and its quality directly affects the accuracy of the MRI examination. Common techniques employ undersampling to acquire k-space data, followed by compressed sensing algorithms for image reconstruction. However, compressed sensing algorithms rely on strict data assumptions, which the actual acquired k-space data often fails to meet, resulting in lower accuracy in image reconstruction. Summary of the Invention

[0003] This application provides a magnetic resonance image reconstruction method, apparatus, electronic device, and readable storage medium that can improve the accuracy of image reconstruction.

[0004] The technical solution of this application embodiment is as follows:

[0005] In a first aspect, embodiments of this application provide a magnetic resonance image reconstruction method, the method comprising:

[0006] Acquire the collected k-space data, assign sampling probabilities to the points in the k-space data, and obtain the probability value corresponding to each point in the k-space data;

[0007] The k-space data is interpolated according to the probability values ​​to obtain interpolated data, and the k-space data is sampled using a preset importance map to obtain sampled data;

[0008] The interpolated data and the sampled data are fused to obtain fused data, and the fused data is used to reconstruct the image to obtain the reconstructed image;

[0009] The reconstructed image is evaluated to obtain an evaluation score. If the evaluation score is greater than a preset evaluation threshold, the probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. The adjusted probability values ​​are used as the probability values ​​to perform a Fourier transform on the reconstructed image to obtain reconstructed data. The importance map is updated using the reconstructed data to obtain an updated importance map. The updated importance map is used as the importance map, and the k-space data is interpolated according to the probability values ​​to obtain interpolated data. The k-space data is sampled using the preset importance map to obtain sampled data.

[0010] In the above technical solution, firstly, the collected k-space data is acquired to provide data support for subsequent processing. Sampling probabilities are assigned to points in the k-space data to obtain the probability value corresponding to each point. This probability value reflects the importance of each point in the k-space data. Interpolation is then performed on the k-space data according to the probability values ​​to obtain interpolated data. This interpolation expands the k-space data, increasing its density and improving the resolution of the reconstructed image without increasing the k-space data size, thus avoiding the deficiencies caused by undersampled k-space data. Next, a preset importance map is used to sample the k-space data, obtaining sampled data. Sampling using the importance map reflects the importance of each location, facilitating edge processing and preventing artifacts in the reconstructed image. Finally, the interpolated data and the sampled data are fused to obtain fused data, which improves the overall image quality. The system utilizes fused data to reconstruct images, obtaining a reconstructed image for subsequent evaluation to determine its accuracy. The reconstructed image is then assessed, and an evaluation score is obtained. If the score exceeds a preset threshold, the reconstructed image has not yet met the requirements. The probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. These adjusted probability values ​​are used as the final probability values ​​to generate a more accurate reconstructed image. A Fourier transform is performed on the reconstructed image to obtain reconstructed data. The importance map is updated using this reconstructed data, and this updated importance map is used as the final importance map. Continuous adjustment of the importance map further improves accuracy. Interpolation is performed on the k-space data according to the probability values ​​to obtain interpolated data. Finally, the k-space data is sampled using the preset importance map to obtain sampled data. By evaluating the reconstructed image and continuously adjusting the probability values ​​and importance map, the system learns autonomously based on the above process, thereby improving the accuracy of the reconstructed image.

[0011] In some embodiments of this application, the step of interpolating the k-space data according to the probability value to obtain interpolated data includes:

[0012] Key data sampling is performed on the k-space data according to the probability values ​​to obtain key sampling data;

[0013] The probability distribution corresponding to the probability value is used to generate a mask center point for the key sampling data. The mask center point is filled out from the surrounding area according to the preset point interval and the probability distribution to obtain a sampling mask.

[0014] Based on the sampling mask, interpolation calculations are performed on the key sampling data to obtain the interpolated data.

[0015] In some embodiments of this application, the step of interpolating the key sampling data according to the sampling mask to obtain the interpolated data includes:

[0016] The sampling mask and the key sampling data are cross-fused to obtain the data to be interpolated. The cross-fusion process involves shifting the sampling mask by a preset number of bits, multiplying it with the key sampling data, and then integrating the data.

[0017] The interpolated data is obtained by interpolating the data to be interpolated using a preset sliding window algorithm.

[0018] In some embodiments of this application, the step of interpolating the data to be interpolated using a preset sliding window algorithm to obtain the interpolated data includes:

[0019] The data to be interpolated is divided into multiple sub-blocks according to the sliding window algorithm.

[0020] Interpolate each of the sub-block data to obtain multiple sub-interpolated data, and generate data for each of the sub-block data to obtain the generated data corresponding to each of the sub-block data;

[0021] The sub-interpolation data corresponding to each sub-block data is weighted and fused with the generated data to obtain interpolated fused data. The interpolated fused data is then low-pass filtered to obtain the interpolated data.

[0022] In some embodiments of this application, the importance graph is obtained through the following steps:

[0023] Multiple training samples are obtained, including part names, k-space samples corresponding to the part names, and important point marker maps corresponding to the k-space samples;

[0024] The part name and the k-space sample are input into a preset importance learning network to generate a prediction map;

[0025] The loss function is calculated using the predicted map and the important point marker map. The parameters of the importance learning network are adjusted using the loss function value until the preset training conditions are met, thus obtaining the importance map.

[0026] In some embodiments of this application, updating the importance map using the reconstructed data to obtain an updated importance map includes:

[0027] The reconstructed data is labeled, and the labeled reconstructed data is added to the training samples. The importance learning network is then retrained using the training samples to obtain the updated importance map.

[0028] In some embodiments of this application, the step of using the fused data to perform image reconstruction to obtain a reconstructed image includes:

[0029] The fused data is divided into regions according to the probability values ​​to obtain a central region and an edge region. The central region is subjected to a first normalization process to obtain first normalized data, and the edge region is subjected to a second normalization process to obtain second normalized data. The first normalized data and the second normalized data are concatenated to obtain third normalized data.

[0030] The reconstructed image is obtained by reconstructing the third normalized data using a preset variational autoencoder.

[0031] Secondly, embodiments of this application provide a magnetic resonance image reconstruction apparatus, the apparatus comprising:

[0032] The data acquisition and processing module is used to acquire the collected k-space data, assign sampling probabilities to the points in the k-space data, and obtain the probability value corresponding to each point in the k-space data.

[0033] The data sampling module is used to interpolate the k-space data according to the probability value to obtain interpolated data, and to sample the k-space data using a preset importance map to obtain sampled data;

[0034] The data reconstruction module is used to fuse the interpolated data and the sampled data to obtain fused data, and to use the fused data to reconstruct the image to obtain the reconstructed image;

[0035] The data adjustment module is used to evaluate the reconstructed image to obtain an evaluation score. If the evaluation score is greater than a preset evaluation threshold, the probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. The adjusted probability values ​​are used as the probability values ​​to perform a Fourier transform on the reconstructed image to obtain reconstructed data. The importance map is updated using the reconstructed data to obtain an updated importance map. The updated importance map is used as the importance map. The steps of interpolating the k-space data according to the probability values ​​to obtain interpolated data and sampling the k-space data using the preset importance map to obtain sampled data are performed.

[0036] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, a user interface, a communication bus, and a network interface. The processor, the memory, the user interface, and the network interface are respectively connected to the communication bus. The memory is used to store instructions. The user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method described in any one of the first aspects.

[0037] Fourthly, embodiments of this application provide a computer-readable storage medium storing instructions that, when executed, perform the method described in any one of the methods provided in the first aspect above.

[0038] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0039] 1. Firstly, k-space data is acquired to provide data support for subsequent processing. Sampling probabilities are assigned to points in the k-space data, resulting in probability values ​​for each point. These probability values ​​reflect the importance of each point in the k-space data. Interpolation is then performed on the k-space data according to these probability values ​​to obtain interpolated data. This interpolation expands the k-space data, increasing its density and improving the resolution of the reconstructed image without increasing the amount of k-space data. This avoids the deficiencies caused by undersampled k-space data. A preset importance map is then used to sample the k-space data, resulting in sampled data. This sampled data reflects the importance of each location, facilitating edge processing and preventing artifacts in the reconstructed image. Finally, the interpolated data and sampled data are fused to obtain fused data, which improves the overall image quality. The system utilizes fused data for image reconstruction to obtain a reconstructed image, which is then used to determine if it meets requirements and achieves greater accuracy. The reconstructed image is evaluated to obtain an evaluation score. If the score exceeds a preset threshold, it indicates that the reconstructed image has not yet met the requirements. The probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. These adjusted probability values ​​are used as the final probability values ​​to generate a more accurate reconstructed image. A Fourier transform is performed on the reconstructed image to obtain reconstructed data. The importance map is updated using this reconstructed data, and this updated importance map is used as the final importance map. Continuous adjustment of the importance map further improves accuracy. Interpolation is performed on the k-space data according to the probability values ​​to obtain interpolated data. Sampling is then performed on the k-space data using the preset importance map to obtain sampled data. By evaluating the reconstructed image and continuously adjusting the probability values ​​and importance map, the system learns autonomously based on the above process, thus improving the accuracy of the reconstructed image. Therefore, this method effectively solves the problem in related technologies where the actual acquired k-space data cannot meet the strict data assumptions of compressed sensing algorithms, resulting in low image reconstruction accuracy.

[0040] 2. Important point locations are marked by generating a sampling mask to facilitate accurate interpolation processing and generate a more accurate reconstructed image.

[0041] 3. The sampling mask and the key sampling data are cross-fused to make the fusion result have a stronger data expression capability.

[0042] 4. Using a sliding window to process images in blocks before integrating them can avoid artifacts in the reconstructed image. Attached Figure Description

[0043] Figure 1This is a schematic flowchart of a magnetic resonance image reconstruction method provided in one embodiment of this application;

[0044] Figure 2 This is a schematic diagram of a transceiver-separated three-combination ultra-high field dual-tuned head coil for a magnetic resonance image reconstruction method provided in one embodiment of this application;

[0045] Figure 3 yes Figure 1 A flowchart illustrating a sub-step of step S200;

[0046] Figure 4 yes Figure 3 A flowchart illustrating a sub-step of step S230;

[0047] Figure 5 This is a schematic diagram of the structure of a magnetic resonance image reconstruction apparatus provided in one embodiment of this application;

[0048] Figure 6 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0049] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0050] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0051] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0052] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0053] This application provides a magnetic resonance image reconstruction method, apparatus, electronic device, and readable storage medium. The method first acquires collected k-space data to provide data support for subsequent processing. Sampling probabilities are assigned to points in the k-space data to obtain probability values ​​corresponding to each point, reflecting the importance of each point. Interpolation is then performed on the k-space data according to the probability values ​​to obtain interpolated data. This interpolation expands the k-space data, increasing its density and improving the resolution of the reconstructed image without increasing the amount of k-space data, thus avoiding the deficiencies caused by undersampled k-space data. A preset importance map is used to sample the k-space data, obtaining sampled data. This sampled data reflects the importance of each location, facilitating edge processing and preventing artifacts in the reconstructed image. Finally, the interpolated data and the sampled data are fused. The process involves obtaining fused data, which enhances the data's expressive power. Image reconstruction is then performed using this fused data to obtain a reconstructed image, which is then used to determine if it meets requirements and achieves greater accuracy. The reconstructed image is evaluated to obtain an evaluation score. If the score exceeds a preset evaluation threshold, the reconstructed image has not yet met the requirements. The probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. These adjusted probability values ​​are used as the final probability values ​​to generate a more accurate reconstructed image. A Fourier transform is performed on the reconstructed image to obtain reconstructed data. The importance map is updated using this reconstructed data, and this updated importance map is used as the final importance map. Continuous adjustment of the importance map further improves accuracy. Interpolation is then performed on the k-space data according to the probability values ​​to obtain interpolated data. Finally, the k-space data is sampled using the preset importance map to obtain sampled data. The reconstructed image is evaluated, and the probability values ​​and importance map are continuously adjusted. Through this process of self-learning, the accuracy of the reconstructed image can be improved.

[0054] It should be noted that this magnetic resonance image reconstruction method is mainly used in medical diagnosis using magnetic resonance imaging (MRI) for imaging the anatomical structures of the brain, spine, joints, and other parts of the body. It can reconstruct relatively accurate MRI images, improving the accuracy and efficiency of MRI examinations.

[0055] The technical solutions provided in the embodiments of this application will be further described below with reference to the accompanying drawings.

[0056] Reference Figure 1 , Figure 1 This is a schematic flowchart of the magnetic resonance image reconstruction method provided in the embodiments of this application. The magnetic resonance image reconstruction method is applied to a magnetic resonance image reconstruction device and is executed by a processor in an electronic device or a readable storage medium. The magnetic resonance image reconstruction method includes steps S100, S200, S300, and S400.

[0057] Step S100: Obtain the collected k-space data, assign sampling probabilities to the points in the k-space data, and obtain the probability value corresponding to each point in the k-space data.

[0058] In one embodiment, a three-combination ultra-high field dual-tuned head coil with separate transmit and receive functions is used to perform head MRI detection under an ultra-high field of ≥7T to acquire k-space data. Figure 2 As shown, the transceiver-separated three-combination ultra-high field dual-tuning head coil includes two external birdcage coils and an internal multi-channel RF array coil. Birdcage coil one, birdcage coil two, and the multi-channel RF array coil are all independent of each other. Birdcage coil one is responsible for transmitting and receiving signals from hydrogen nuclei, birdcage coil two is responsible for transmitting signals from non-hydrogen nuclei, and the multi-channel RF array coil is responsible for receiving signals from non-hydrogen nuclei. On the multi-channel RF array coil, each channel has two notch circuits arranged in parallel between its active and passive protection circuits. Each notch circuit consists of a capacitor and an inductor connected in parallel, used to shield the hydrogen nuclei signals from birdcage coil one, preventing them from affecting the multi-channel RF array coil. Birdcage coil one, birdcage coil two, and the multi-channel RF array coil are independent of each other and connected by a mounting bracket. The transceiver-separated three-combination ultra-high field dual-tuning head coil also includes a detachable inner detection layer 120 and an outer detection layer 110. The multi-channel RF array coil is located inside the inner detection layer 120, and birdcage coil one and birdcage coil two are located inside the outer detection layer 110. Each of the birdcage coil one, birdcage coil two, and multi-channel radio frequency array coil has a signal line connected to it. Each signal line is used to connect to the magnetic resonance imaging host for data acquisition.

[0059] In one embodiment, the above-described coil structure, under ultra-high field conditions, exhibits strong non-hydrogen nucleus signals and good imaging in the birdcage coil, which is beneficial for generating clear reconstructed images. Using this coil structure, during magnetic resonance imaging (MRI), dual-tuned imaging of hydrogen and non-hydrogen nuclei under ultra-high field conditions generates k-space data from multiple coils. This k-space data is undersampled, acquired, and stored. A preset readout function is used to retrieve the acquired k-space data, providing data support for subsequent processing. The preset readout function can be either `read()` or `open()`. The k-space data can be data from different body parts, including k-space data from the heart and brain, etc., which will not be elaborated upon here.

[0060] Then, the sampling probability is calculated by fitting a large amount of historical k-space data from different body parts to obtain the sampling probability of each point in the historical k-space data for different body parts. The sampling probability of the corresponding body part is obtained from the collected k-space data. Using the structure of the historical k-space data, the sampling probability is assigned to the collected k-space data in a point-to-point manner to obtain the probability value corresponding to each point in the k-space data. It should be noted that the k-space data is stored in a matrix format, similar to the format of images, and is represented by a two-dimensional or three-dimensional matrix. Different body parts can be the heart, brain, etc. For example, a large amount of historical heart k-space data is fitted to obtain a probability distribution, and the sampling probability of each point is statistically analyzed. When the heart k-space data is scanned again, the sampling probability statistically obtained from the historical data is assigned to the k-space data to obtain the probability value corresponding to each point.

[0061] Step S200: Interpolate the k-space data according to the probability values ​​to obtain interpolated data, and sample the k-space data using a preset importance map to obtain sampled data.

[0062] like Figure 3 As shown, interpolation is performed on the k-space data according to the probability values ​​to obtain interpolated data, including but not limited to the following steps:

[0063] Step S210: Perform key data sampling on the k-space data according to the probability value to obtain key sampled data.

[0064] In some possible embodiments of this application, based on the probability value obtained in step S100, which reflects the probability of each point in the k-space data, key data sampling is performed on the k-space data according to this probability value. Specifically, a sampling threshold is set. If the probability value of a point in the k-space data is greater than the sampling threshold, it indicates that the point is key data. If the probability value of a point in the k-space data is less than or equal to the sampling threshold, it indicates that the point data belongs to the high-frequency data of the reconstructed image. All points greater than the sampling threshold are selected, and the positions of each selected point are recorded to constitute key sampling data, so that the sampling mask and interpolation data can be obtained subsequently using the key sampling data. The sampling threshold can be 0.6, 0.8, etc. The larger the value of the sampling threshold, the closer it is to the central region of the k-space data. The central low-frequency region is selected to provide the main structure of the image for subsequent image reconstruction.

[0065] Step S220: Generate mask center points for key sampling data using the probability distribution corresponding to the probability values, and fill the mask from the center points outwards according to the preset point intervals and probability distribution to obtain the sampling mask.

[0066] In some possible embodiments of this application, the key sampling data obtained in step S210 includes key sampling points and their corresponding positions. Since the center point of the k-space data can delineate the main structure of the image, selecting the point with the highest probability value and located at the center among the key sampling points as the mask center point can capture the main structure of the image. Then, the mask is filled outwards from the center point according to a preset interval and probability distribution, where the preset interval can be 1 point, 2 points, or 4 points apart. For example, with the mask center point as the center, 0 is filled outwards at 2 points. After filling outwards at 2 points, the data after filling is adjusted using a probability distribution. Specifically, according to the probability value and position of each point in the k-space data, points with a probability value greater than a preset threshold are selected, and the position of the point is obtained. The selected point is compared with the position of the data after filling. If there is a filling value of 0 at the position corresponding to the selected probability value, the filling value is changed to 1 to obtain the sampling mask. This sampling mask not only reflects the outline of the main subject but also fills in details in high-frequency regions (the surrounding boundary areas) to obtain a more accurate reconstructed image. The preset threshold is 0.2, but can also be 0.3. A higher preset threshold means more zeros are filled in the surrounding boundary areas, which is detrimental to image reconstruction in high-frequency regions. The sampling mask marks the locations of important points for accurate interpolation processing, resulting in a more accurate reconstructed image.

[0067] Step S230: Based on the sampling mask, perform interpolation calculations on the key sampling data to obtain interpolated data.

[0068] like Figure 4 As shown, based on the sampling mask, interpolation calculations are performed on the key sampling data to obtain interpolated data, including but not limited to the following steps:

[0069] Step S231: The sampling mask and key sampling data are cross-fused to obtain the data to be interpolated. The cross-fusion process involves shifting the sampling mask by a preset number of bits, multiplying it with the key sampling data, and then integrating the data.

[0070] In some possible embodiments of this application, the preset number of bits can be 1 bit, 2 bits, etc. The cross-fusion process involves shifting the sampling mask according to the preset number of bits, multiplying it with the key sampling data, and then integrating the data. Specifically, the sampling mask is first multiplied with the key sampling data to obtain the first fused data. Then, the sampling mask is shifted one bit to the right and multiplied with the key sampling data again to obtain the second fused data. It should be noted that after shifting one bit to the right, the leftmost column is filled to the rightmost column. Following the above process, one fused data is obtained for each shift, until the rightmost end of the original data is reached, resulting in multiple fused data. The first fused data, the second fused data, and the subsequent multiple fused data are weighted and fused to obtain the data to be interpolated. Through the above cross-fusion process, the data dimensions can be enriched, the expressive power of undersampled k-space data can be improved, data limitations can be reduced, and support can be provided for subsequent reconstruction of more accurate images.

[0071] Step S232: Use a preset sliding window algorithm to interpolate the data to be interpolated to obtain the interpolated data.

[0072] In one embodiment, a preset sliding window algorithm is used to interpolate the data to be interpolated to obtain interpolated data, including but not limited to: dividing the data to be interpolated into multiple sub-blocks according to the sliding window algorithm; interpolating each sub-block to obtain multiple sub-interpolated data; generating data for each sub-block to obtain generated data corresponding to each sub-block; weightedly fusing the sub-interpolated data corresponding to each sub-block with the generated data to obtain interpolated fused data; and low-pass filtering the interpolated fused data to obtain interpolated data.

[0073] In some possible embodiments of this application, the sliding window algorithm specifies a sliding step size and a sliding window size. Starting from the initial position of the data to be interpolated, the sliding window moves according to the sliding step size, dividing the data into multiple sub-blocks according to the size of the sliding window. The sliding window is represented by H×W, where H can be one-third or one-quarter of the data to be interpolated, and W can be one-third or one-quarter of the data to be interpolated. The sliding step size can be the size of H and W. For example, when the sliding step size is H and W, and the sliding window size H×W is set to one-third each, sliding horizontally three times for each column can divide the data into 9 sub-blocks. It should be noted that when dividing the data according to the sliding window and sliding step size, if the boundaries of the data to be interpolated do not meet the corresponding sizes, the excess portion is padded with 0s to ensure that the sliding window can accurately divide the boundaries.

[0074] Then, interpolation is performed on each sub-block of data, using methods such as linear interpolation or multinomial interpolation, to obtain multiple sub-interpolated data sets. These sub-interpolated data sets are then used to generate the final interpolated data. A deep learning model is then used to generate data for each sub-block, resulting in corresponding generated data. This deep learning model is a Generative Adversarial Network (GAN), a pre-trained network. The parameters of the trained GAN are further adjusted using k-space data to improve its generalization ability to k-space data. The generated data, obtained using the GAN, exhibits high accuracy.

[0075] The process involves obtaining the first interpolation weights of the sub-interpolation data corresponding to each sub-block and the second interpolation weights of the generated data corresponding to each sub-block. The sub-interpolation data is multiplied by the first interpolation weight, and the generated data is multiplied by the second interpolation weight. These results are then summed to obtain the interpolated fused data, which possesses strong data expressive power and is beneficial for reconstructing accurate magnetic resonance images. Weighted fusion is used to combine the data from each sub-block, and fusing the generated data with the sub-interpolation data avoids artifacts. The sum of the first and second interpolation weights is 1. Since the data generated using a generative adversarial network (GAN) enhances data expressiveness, the first interpolation weight is set to be less than the second interpolation weight, resulting in a more expressive interpolated fused data. The interpolated fused data is then low-pass filtered to obtain the interpolated data, which ensures phase consistency. Using a sliding window for block processing before integration helps avoid artifacts in the reconstructed image.

[0076] In one embodiment, a preset importance map reflects which regions in the k-space data have a greater impact on reconstruction quality (i.e., contain more information), and data from these regions are prioritized. Sampling of the k-space data using the importance map specifically involves either fixed-pattern sampling or hierarchical sampling, used to indicate the sampling trajectory. For example, fixed-pattern sampling involves sampling according to importance, with higher importance data receiving denser sampling and lower importance data receiving sparser sampling. The sampling density and sparsity can also be adjusted for sampling control. Hierarchical sampling divides the k-space data into multiple levels based on importance, assigns sampling ratios to different levels, and performs sampling according to these ratios. Based on the importance map, the k-space data is sampled using a hierarchical sampling strategy to obtain sampled data, which is then used for subsequent data fusion to generate a more accurate reconstructed image.

[0077] In one embodiment, the importance map is obtained through the following steps: acquiring multiple training samples, including part names, k-space samples corresponding to the part names, and importance point marker maps corresponding to the k-space samples; inputting the part names and k-space samples into a preset importance learning network to generate a prediction map; calculating the value of the loss function using the prediction map and the importance point marker map; adjusting the parameters of the importance learning network using the value of the loss function until the preset training conditions are met to obtain the importance map.

[0078] In some possible embodiments of this application, the training samples include a body part name, a k-space sample corresponding to the body part name, and an important point marker map corresponding to the k-space sample. In actual scanning, scanning is performed according to the submitted details, which contain the body part names to be scanned; these body part names are then stored. After scanning the body part name, a k-space sample corresponding to the body part name is obtained. Important points are manually marked on the k-space sample to form an important point marker map. The k-space sample and its corresponding important point marker map are stored in the same file as the body part name. A preset file reading function is used to obtain the body part name, the corresponding k-space sample, and the corresponding important point marker map, providing a data foundation for subsequent training. Alternatively, training samples can be downloaded online, and key content can be extracted from the training samples to obtain the body part name, the corresponding k-space sample, and the corresponding important point marker map, providing a data foundation for subsequent training.

[0079] According to a preset batch, the body part name and k-space samples are input into a preset importance learning network to generate a prediction map. The preset batch is a hyperparameter of the model, which can be adjusted and can take values ​​of 1, 2, 4, etc. The preset importance learning network is a deep neural network model, such as the U-net model or its variants. For example, if the preset batch value is 1, only one body part name and one k-space sample are input at a time in each training iteration. Then, the body part name and k-space samples undergo feature transformation, which can be done using embedding methods to convert them into a body part feature vector corresponding to the body part name and a k-space feature vector corresponding to the k-space sample. The body part feature vector and the k-space feature vector are then fused using vector concatenation to obtain the input features. The U-net model is then used to perform feature extraction, non-linear processing, and normalization on the input features to output the prediction map.

[0080] The logarithmic loss function is then used to calculate the loss function between the predicted image and the importance map, obtaining the value of the loss function. This loss function value is then used to perform backpropagation on the U-net model to update its parameters until the preset training conditions are met, resulting in a trained U-net model and outputting a trained importance map. This importance map is then used for subsequent data sampling to reconstruct accurate images.

[0081] The preset training conditions can be: the loss function value tends to stabilize, and training ends after stabilization; or a set number of training iterations is set, and training ends after the set number of iterations is reached. Alternatively, a combination of the loss function stabilization and a set number of training iterations can be used, ending training when either of these conditions is met.

[0082] Step S300: The interpolated data and the sampled data are fused to obtain fused data. The fused data is then used to reconstruct the image to obtain the reconstructed image.

[0083] In one embodiment, confidence analysis is performed on the sampled data. First, the signal-to-noise ratio (SNR) of each sampling point in the sampled data is calculated to obtain the SNR value. Second, the phase difference between adjacent sampling points in the sampled data is evaluated to obtain the phase value. Then, the variance of the sampled data is calculated using a preset window to obtain the signal variance within the window. Finally, the confidence level is calculated using a confidence formula on the SNR value, phase value, and signal variance to obtain the sampling confidence level. The preset window is a 3×3 window, and the confidence formula is expressed as:

[0084] ZXD=tanh(SNR)×exp(-ph 2 )×(1 / (1+s))

[0085] Where ZXD represents the sampling confidence level, SNR represents the signal-to-noise ratio, ph represents the phase value, s represents the signal variance, and tanh() represents a mathematical function that constrains the signal-to-noise ratio to between 0 and 1.

[0086] The sampling confidence level is a value between 0 and 1. The higher the sampling confidence level, the higher the reliability. The smaller the signal variance, the higher the neighborhood consistency; the smaller the phase value, the more stable the phase, and thus the higher the sampling confidence level.

[0087] In one embodiment, a confidence analysis is performed on the interpolated data. The confidence level of the process of interpolating k-space data according to probability values ​​is evaluated. Based on the above interpolation process, the confidence level of the interpolation method according to probability values ​​is set to 0.8, which has a high confidence level. This value is obtained through prior knowledge. The interpolation confidence level is thus obtained.

[0088] The confidence ratio is calculated using the sampling confidence level and the interpolation confidence level. A weight adjustment factor is then calculated based on this confidence ratio. The adjusted weights are then used for weighted fusion calculations to obtain the fused data. Specifically, the ratio of the sampling confidence level to the interpolation confidence level is calculated. The tanh() function is then used to subtract 1 from the confidence ratio to obtain the adjustment factor. The base weights of the sampling data are obtained; these base weights are values ​​set by professionals based on experience. These base weights are directly read and then adjusted using the weight adjustment formula to obtain the sampling weights and interpolation weights. The weight adjustment formula is: Sampling weight = Base weight × (1 + 0.5 × Adjustment factor). Since the sum of the sampling weight and the interpolation weight is 1, the interpolation weight = 1 - Sampling weight. The result of multiplying the interpolation data by the interpolation weight is summed with the result of multiplying the sampling data by the sampling weight to obtain the fused data, which can then be used for subsequent image reconstruction.

[0089] In another embodiment, image reconstruction is performed using fused data to obtain a reconstructed image, including but not limited to: dividing the fused data into regions according to probability values ​​to obtain a central region and an edge region; performing a first normalization process on the central region to obtain first normalized data; performing a second normalization process on the edge region to obtain second normalized data; stitching the first normalized data and the second normalized data together to obtain third normalized data; and reconstructing the third normalized data using a preset variational autoencoder to obtain a reconstructed image.

[0090] In some possible embodiments of this application, the probability value obtained in step S100 reflects the probability of each point in the k-space data. Dividing the fused data into regions according to the probability values ​​allows for the identification of regions with higher probability values ​​and regions with lower probability values. Regions with higher probability values ​​correspond to the central region, and regions with lower probability values ​​correspond to the edge regions. Mean-variance normalization is applied to the central region to obtain first normalized data, and nonlinear normalization is applied to the edge regions to obtain second normalized data. The first and second normalized data are then concatenated to obtain third normalized data, which is subsequently used for image reconstruction. Specifically, the nonlinear normalization process involves applying a square root to the edge region data to enhance low-amplitude signals. A preset variational autoencoder is then used to perform feature extraction, feature normalization, and feature nonlinear processing on the third normalized data to output a reconstructed image. This variational autoencoder uses skip connections, which can fully extract features and improve the accuracy of the reconstructed image.

[0091] Step S400 involves evaluating the reconstructed image to obtain an evaluation score. If the evaluation score is greater than a preset evaluation threshold, the probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. The adjusted probability values ​​are used as probability values ​​to perform a Fourier transform on the reconstructed image to obtain reconstructed data. The importance map is updated using the reconstructed data to obtain an updated importance map. The updated importance map is used as the importance map to perform interpolation processing on the k-space data according to the probability values ​​to obtain interpolated data. The k-space data is sampled using the preset importance map to obtain sampled data.

[0092] In one embodiment, a preset no-reference spatial domain image quality assessment algorithm is used to evaluate the reconstructed image, obtaining an evaluation score. A lower evaluation score indicates better image quality. If the evaluation score exceeds a preset evaluation threshold, it indicates that the quality of the reconstructed image needs improvement. To reconstruct a high-quality image, the probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. Specifically, the adjustment involves adding the collected k-space data to historical k-space data with a fitted probability distribution, then refitting the updated k-space data to obtain a probability distribution, and obtaining the adjusted probability values ​​based on this distribution. Simultaneously, a Fourier transform is performed on the reconstructed image to obtain reconstructed data. This Fourier transform is an inverse transform process that maps the reconstructed image to k-space data, referred to here as reconstructed data. The importance map is updated using the reconstructed data to obtain an updated importance map.

[0093] In another embodiment, the importance map is updated using the reconstructed data to obtain an updated importance map, including but not limited to: labeling the reconstructed data, adding the labeled reconstructed data to the training samples, and retraining the importance learning network using the training samples to obtain an updated importance map.

[0094] In some possible embodiments of this application, the reconstructed data is k-space data. The reconstructed data is labeled manually or using a preset labeling algorithm to obtain labeled reconstructed data. The labeling algorithm can be an object detection labeling algorithm, with parameters adjusted before labeling. Using a labeling algorithm automates the labeling process and saves time. The labeled reconstructed data is added to the training samples, and the importance learning network is retrained using these samples to obtain an updated importance map. This process not only increases the training samples for the importance learning network but also incorporates input data into the training samples, improving the accuracy of the output importance map.

[0095] In one embodiment, the adjusted probability value is used as the probability value, and the updated importance map is used as the importance map. Interpolation is performed on the k-space data according to the probability value to obtain interpolated data. The k-space data is then sampled using a preset importance map to obtain sampled data. The interpolated data and sampled data are fused to obtain fused data. Image reconstruction is performed using the fused data to obtain a reconstructed image. The reconstructed image is evaluated to obtain an evaluation score. If the evaluation score is greater than a preset evaluation threshold, the above adjustment steps are repeated, and calculations are performed based on the adjusted probability value and importance map. If the evaluation score is less than or equal to the preset evaluation threshold, the reconstructed image is obtained, which is the final presented image. By evaluating the reconstructed image and continuously adjusting the probability value and importance map, and through self-learning based on the above process, the accuracy of the reconstructed image can be improved.

[0096] like Figure 5As shown in the figure, this application provides a magnetic resonance image reconstruction device 100. The magnetic resonance image reconstruction device 100 acquires collected k-space data through a data acquisition and processing module 110, providing data support for subsequent processing of the k-space data. It assigns sampling probabilities to points in the k-space data, obtaining probability values ​​corresponding to each point. These probability values ​​reflect the importance of each point in the k-space data. A data sampling module 120 interpolates the k-space data according to the probability values ​​to obtain interpolated data. This interpolation expands the k-space data, increasing its density and improving the resolution of the reconstructed image without increasing the k-space data, thus avoiding the deficiencies caused by undersampled k-space data. A preset importance map is used to sample the k-space data, obtaining sampled data. This sampling data reflects the importance of each location, facilitating edge processing and preventing artifacts in the reconstructed image. Finally, a data reconstruction module 130 combines the interpolated data with... The sampled data is fused to obtain fused data, which improves the data's expressive power. Image reconstruction is performed using the fused data to obtain a reconstructed image, which is then used to determine if it meets the requirements for a more accurate image. The data adjustment module 140 evaluates the reconstructed image to obtain an evaluation score. If the evaluation score is greater than a preset evaluation threshold, it indicates that the reconstructed image has not yet met the image requirements. The probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. These adjusted probability values ​​are used as the final probability values ​​to generate a more accurate reconstructed image. A Fourier transform is performed on the reconstructed image to obtain reconstructed data. The importance map is updated using the reconstructed data to obtain an updated importance map. This updated importance map is used as the final importance map. Continuous adjustment of the importance map further improves accuracy. Interpolation processing is performed on the k-space data according to the probability values ​​to obtain interpolated data. Sampling is then performed on the k-space data using the preset importance map to obtain sampled data. The reconstructed image is evaluated, and the probability values ​​and importance map are continuously adjusted. Through this process of self-learning, the accuracy of the reconstructed image can be improved.

[0097] It should be noted that the data acquisition and processing module 110 is connected to the data sampling module 120, the data sampling module 120 is connected to the data reconstruction module 130, and the data reconstruction module 130 is connected to the data adjustment module 140. The aforementioned magnetic resonance image reconstruction method is applied to a magnetic resonance image reconstruction device 100. The device acquires collected k-space data to provide data support for subsequent processing. It assigns sampling probabilities to points in the k-space data, obtaining probability values ​​corresponding to each point. These probability values ​​reflect the importance of each point in the k-space data. Interpolation is then performed on the k-space data according to the probability values ​​to obtain interpolated data. This interpolation expands the k-space data, increasing its density and improving the resolution of the reconstructed image without increasing the amount of k-space data, thus avoiding the deficiencies caused by undersampled k-space data. A preset importance map is used to sample the k-space data, obtaining sampled data. This sampled data reflects the importance of each location, facilitating edge processing and preventing artifacts in the reconstructed image. Finally, the interpolated data and the sampled data are fused to obtain a fused image. The process involves fusing data to enhance its expressive power. This fused data is used for image reconstruction, resulting in a reconstructed image. This reconstructed image is then evaluated to determine its accuracy. An evaluation score is obtained; if the score exceeds a preset threshold, the reconstructed image does not yet meet the requirements. The probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. These adjusted probability values ​​are used as the final probability values ​​to generate a more accurate reconstructed image. A Fourier transform is performed on the reconstructed image to obtain reconstructed data. This reconstructed data is then used to update the importance map, which is used as the final importance map. Continuous adjustment of the importance map further improves accuracy. Interpolation is performed on the k-space data according to the probability values ​​to obtain interpolated data. Finally, the k-space data is sampled using the preset importance map to obtain sampled data. By evaluating the reconstructed image and continuously adjusting the probability values ​​and importance map, the process learns autonomously and improves the accuracy of the reconstructed image.

[0098] It should also be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0099] This application also discloses an electronic device. (See reference...) Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0100] The communication bus 502 is used to enable communication between these components.

[0101] The user interface 503 may include a display screen and a camera. Optionally, the user interface 503 may also include a standard wired interface and a wireless interface.

[0102] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0103] The processor 501 may include one or more processing cores. The processor 501 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 505, and by calling data stored in memory 505. Optionally, the processor 501 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 501 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 501 and may be implemented as a separate chip.

[0104] The memory 505 may include random access memory (RAM) or read-only memory. Optionally, the memory 505 may include a non-transitory computer-readable storage medium. The memory 505 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 505 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 505 may also be at least one storage device located remotely from the aforementioned processor 501. (Refer to...) Figure 6 The memory 505, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a magnetic resonance image reconstruction method.

[0105] exist Figure 6 In the illustrated electronic device 500, the user interface 503 is mainly used to provide an input interface for the user and acquire user input data; while the processor 501 can be used to call an application program of a magnetic resonance image reconstruction method stored in the memory 505. When executed by one or more processors 501, the electronic device 500 performs one or more methods as described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0106] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0107] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0108] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0109] Furthermore, the functional units in the various embodiments of this 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 integrated unit can be implemented in hardware or as a software functional unit.

[0110] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0111] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will readily conceive of those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0112] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A magnetic resonance image reconstruction method, characterized in that, The method includes: Acquire the collected k-space data, assign sampling probabilities to the points in the k-space data, and obtain the probability value corresponding to each point in the k-space data; The k-space data is interpolated according to the probability values ​​to obtain interpolated data, and the k-space data is sampled using a preset importance map to obtain sampled data; The importance graph is obtained through the following steps: Multiple training samples are obtained, including part names, k-space samples corresponding to the part names, and important point marker maps corresponding to the k-space samples; The part name and the k-space sample are input into a preset importance learning network to generate a prediction map; The loss function is calculated using the predicted map and the important point marker map. The parameters of the importance learning network are adjusted using the loss function value until the preset training conditions are met, and an importance map is obtained. The interpolated data and the sampled data are fused to obtain fused data, and the fused data is used to reconstruct the image to obtain the reconstructed image; The reconstructed image is evaluated to obtain an evaluation score. If the evaluation score is greater than a preset evaluation threshold, the probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. The adjusted probability values ​​are used as the probability values ​​to perform a Fourier transform on the reconstructed image to obtain reconstructed data. The importance map is updated using the reconstructed data to obtain an updated importance map. The updated importance map is used as the importance map, and the k-space data is interpolated according to the probability values ​​to obtain interpolated data. The k-space data is sampled using the preset importance map to obtain sampled data.

2. The method according to claim 1, characterized in that, The step of interpolating the k-space data according to the probability value to obtain interpolated data includes: Key data sampling is performed on the k-space data according to the probability values ​​to obtain key sampling data; The probability distribution corresponding to the probability value is used to generate a mask center point for the key sampling data. The mask center point is filled out from the surrounding area according to the preset point interval and the probability distribution to obtain a sampling mask. Based on the sampling mask, interpolation calculations are performed on the key sampling data to obtain the interpolated data.

3. The method according to claim 2, characterized in that, The step of interpolating the key sampling data according to the sampling mask to obtain the interpolated data includes: The sampling mask and the key sampling data are cross-fused to obtain the data to be interpolated. The cross-fusion process involves shifting the sampling mask by a preset number of bits, multiplying it with the key sampling data, and then integrating the data. The interpolated data is obtained by interpolating the data to be interpolated using a preset sliding window algorithm.

4. The method according to claim 3, characterized in that, The step of interpolating the data to be interpolated using a preset sliding window algorithm to obtain the interpolated data includes: The data to be interpolated is divided into multiple sub-blocks according to the sliding window algorithm. Interpolate each of the sub-block data to obtain multiple sub-interpolated data, and generate data for each of the sub-block data to obtain the generated data corresponding to each of the sub-block data; The sub-interpolation data corresponding to each sub-block data is weighted and fused with the generated data to obtain interpolated fused data. The interpolated fused data is then low-pass filtered to obtain the interpolated data.

5. The method according to claim 1, characterized in that, The step of updating the importance map using the reconstructed data to obtain an updated importance map includes: The reconstructed data is labeled, and the labeled reconstructed data is added to the training samples. The importance learning network is then retrained using the training samples to obtain the updated importance map.

6. The method according to claim 1, characterized in that, The process of using the fused data to reconstruct the image and obtain the reconstructed image includes: The fused data is divided into regions according to the probability values ​​to obtain a central region and an edge region. The central region is subjected to a first normalization process to obtain first normalized data, and the edge region is subjected to a second normalization process to obtain second normalized data. The first normalized data and the second normalized data are concatenated to obtain third normalized data. The reconstructed image is obtained by reconstructing the third normalized data using a preset variational autoencoder.

7. A magnetic resonance image reconstruction device, characterized in that, The device includes: The data acquisition and processing module (110) is used to acquire the collected k-space data, assign sampling probabilities to the points in the k-space data, and obtain the probability value corresponding to each point in the k-space data. The data sampling module (120) is used to interpolate the k-space data according to the probability value to obtain interpolated data, and to sample the k-space data using a preset importance map to obtain sampled data; The importance graph is obtained through the following steps: Multiple training samples are obtained, including part names, k-space samples corresponding to the part names, and important point marker maps corresponding to the k-space samples; The part name and the k-space sample are input into a preset importance learning network to generate a prediction map; The loss function is calculated using the predicted map and the important point marker map. The parameters of the importance learning network are adjusted using the loss function value until the preset training conditions are met, and an importance map is obtained. The data reconstruction module (130) is used to fuse the interpolated data and the sampled data to obtain fused data, and to use the fused data to reconstruct the image to obtain the reconstructed image; The data adjustment module (140) is used to evaluate the reconstructed image and obtain an evaluation score. If the evaluation score is greater than a preset evaluation threshold, the probability values ​​corresponding to each point in the k-space data are adjusted to obtain adjusted probability values. The adjusted probability values ​​are used as the probability values ​​to perform a Fourier transform on the reconstructed image to obtain reconstructed data. The reconstructed data is used to update the importance map to obtain an updated importance map. The updated importance map is used as the importance map. The steps of interpolating the k-space data according to the probability values ​​to obtain interpolated data and sampling the k-space data using the preset importance map to obtain sampled data are performed.

8. An electronic device, characterized in that, The device includes a processor (501), a memory (505), a user interface (503), a communication bus (502), and a network interface (504). The processor (501), the memory (505), the user interface (503), and the network interface (504) are respectively connected to the communication bus (502). The memory (505) is used to store instructions. The user interface (503) and the network interface (504) are used to communicate with other devices. The processor (501) is used to execute the instructions stored in the memory (505) so that the electronic device (500) performs the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-6.

Images