Low-field magnetic resonance image enhancement method and device, electronic equipment and storage medium

Abstract

The invention provides an enhancement method and device for a low-field magnetic resonance image, electronic equipment and a storage medium, and the method comprises the steps: obtaining a to-be-enhanced low-field magnetic resonance image, and carrying out the partitioning of the to-be-enhanced low-field magnetic resonance image, and obtaining a plurality of image blocks; inputting the plurality of image blocks into an image enhancement model to obtain a plurality of enhanced image blocks output by the image enhancement model; splicing the plurality of enhanced image blocks to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; wherein the image enhancement model is obtained based on training of a high-field magnetic resonance image sample and a synthetic low-field magnetic resonance image corresponding to the high-field magnetic resonance image sample. By adopting the technical scheme, the problem that a high-field magnetic resonance image corresponding to a low-field magnetic resonance image is difficult to obtain in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method, device, electronic equipment and storage medium for enhancing a low-field magnetic resonance image. Background Art

[0002] Magnetic resonance imaging is an indispensable imaging technology in clinical practice and is used in various medical scenarios. High-field magnetic resonance images can not only provide accurate diagnostic reference images, but are also widely used in surgical navigation, pathological analysis and other imaging analysis. Figure 1 However, in actual applications, due to site size limitations, insufficient budget, unstable power supply, etc., many underdeveloped areas can only use low-magnetic field MRI scanners, such as Figure 2 FIG. 1 is a schematic diagram of a low-field image obtained by scanning with a low-magnetic field magnetic resonance scanner in the prior art. Compared with a high-field magnetic resonance image obtained by scanning with a high magnetic field, a low-field magnetic resonance image usually has a lower signal-to-noise ratio and contrast, and is difficult to provide sufficient information assistance.

[0003] In the prior art, low-field magnetic resonance enhancement technology is used to improve the quality of magnetic resonance images, and SynthSR synthesis and image quality transfer (IQT) methods are mainly used for image enhancement.

[0004] However, the existing methods usually require a large number of paired high-field and low-field data pairs when performing image enhancement, and it is difficult to obtain high-field magnetic resonance images corresponding to low-field magnetic resonance images in scenarios suitable for low-field magnetic resonance applications. Summary of the invention

[0005] The present invention provides a method, device, electronic equipment and storage medium for enhancing a low-field magnetic resonance image, which are used to solve the problem in the prior art that it is difficult to obtain a high-field magnetic resonance image corresponding to a low-field magnetic resonance image.

[0006] In a first aspect, the present invention provides a method for enhancing a low-field magnetic resonance image, comprising: acquiring a low-field magnetic resonance image to be enhanced and performing block processing to obtain a plurality of image blocks; inputting the plurality of image blocks into an image enhancement model to obtain a plurality of enhanced image blocks output by the image enhancement model; splicing the plurality of enhanced image blocks to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; wherein the image enhancement model is trained based on a high-field magnetic resonance image sample and a synthetic low-field magnetic resonance image corresponding to the high-field magnetic resonance image sample.

[0007] Optionally, the method for acquiring synthetic low-field magnetic resonance image samples includes: acquiring high-field magnetic resonance image samples and low-field magnetic resonance image samples; based on the layer thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field magnetic resonance image samples, performing low-field processing on the high-field magnetic resonance image samples to obtain a synthetic low-field magnetic resonance image corresponding to the high-field magnetic resonance image samples.

[0008] Optionally, based on the slice thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field magnetic resonance image samples, the high-field magnetic resonance image samples are subjected to low-field processing, including: Gaussian filtering and blurring the high-field magnetic resonance image samples using the slice thickness as the half-maximum full width; down-sampling the high-field magnetic resonance image samples according to the spatial resolution; and adjusting the tissue contrast of the high-field magnetic resonance image samples according to the gray matter average signal-to-noise ratio and corresponding covariance matrix and the white matter average signal-to-noise ratio and corresponding covariance matrix.

[0009] Optionally, the image enhancement model is a Mamba-based model obtained by improving the Mamba model. The Mamba-based model includes multiple Mamba encoders, each of which is connected to a downsampling module to receive features output by the downsampling module and output the extracted features to the next downsampling module.

[0010] Optionally, the Mamba encoder is an improved Mamba encoder obtained by replacing the multilayer perceptron module with a new module including a convolutional layer and a channel attention layer, and in the improved Mamba encoder, the layer normalization module, the convolutional layer, the channel attention layer, and the output layer are connected in sequence.

[0011] Optionally, the Mamba encoder uses a spatially continuous input sequence scanning method.

[0012] In a second aspect, the present invention provides an enhancement device for a low-field magnetic resonance image, comprising: a first processing module, used to acquire a low-field magnetic resonance image to be enhanced and perform block processing to obtain a plurality of image blocks; a second processing module, used to input the plurality of image blocks into an image enhancement model, and obtain a plurality of enhanced image blocks output by the image enhancement model; a splicing module, used to perform splicing processing on the plurality of enhanced image blocks, and obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; wherein the image enhancement model is trained based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

[0013] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the above-mentioned methods for enhancing a low-field magnetic resonance image is implemented.

[0014] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned methods for enhancing low-field magnetic resonance images.

[0015] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which implements any of the above-mentioned methods for enhancing low-field magnetic resonance images when executed by a processor.

[0016] The method, device, electronic device and storage medium for enhancing low-field magnetic resonance images provided by the present invention first perform block processing to generate multiple image blocks after obtaining the low-field magnetic resonance image to be enhanced. Then, these image blocks are input into a trained image enhancement model to obtain multiple enhanced image blocks output by the model. Finally, these enhanced image blocks are synthesized through splicing processing to obtain an enhanced magnetic resonance image corresponding to the original low-field magnetic resonance image. The image enhancement model is trained based on high-field magnetic resonance image samples and their corresponding synthetic low-field magnetic resonance images. The present invention obtains low-field magnetic resonance images corresponding to high-field magnetic resonance image samples by synthesis, which reduces the difficulty of obtaining high-field magnetic resonance images corresponding to low-field magnetic resonance images, and the registration relationship between high-field magnetic resonance image samples and synthetic low-field magnetic resonance images is better. The high-field magnetic resonance image samples and their corresponding synthetic low-field magnetic resonance images are used for model training, which further improves the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a schematic diagram of a high-field magnetic resonance image obtained by scanning with a high-magnetic field magnetic resonance scanner in the prior art.

[0019] Figure 2 It is a schematic diagram of a low-field image obtained by scanning with a magnetic resonance scanner with a low magnetic field in the prior art.

[0020] Figure 3 It is an enhanced image obtained by enhancing a low-field image using an image enhancement method in the prior art.

[0021] Figure 4 It is a schematic flow chart of the method for enhancing low-field magnetic resonance images provided by the present invention.

[0022] Figure 5 It is a schematic diagram of the framework of the image enhancement model provided by the present invention.

[0023] Figure 6 It is a processing flow diagram of the low-field magnetic resonance image enhancement method based on Mamba and image quality transfer provided by the present invention.

[0024] Figure 7 The enhanced magnetic resonance image is obtained based on the low-field magnetic resonance image enhancement method provided by the present invention.

[0025] Figure 8 It is a schematic diagram of the effect of performing hippocampal segmentation on a low-field image and an enhanced image provided by the present invention.

[0026] Fig. 9 It is a schematic structural diagram of the low-field magnetic resonance image enhancement device provided by the present invention.

[0027] Fig.10 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Figure 4 FIG. 1 is a flow chart of a method for enhancing a low-field magnetic resonance image provided by the present invention. Figure 4 As shown, the method includes the following steps S401 to S403.

[0030] Step S401: acquiring a low-field magnetic resonance image to be enhanced and performing block processing to obtain a plurality of image blocks.

[0031] As an optional embodiment, the low-field magnetic resonance image to be enhanced can be obtained from a hospital image management system. The present invention does not further limit the method for obtaining the low-field magnetic resonance image.

[0032] As an optional embodiment, the low-field magnetic resonance image is a medical image acquired by a low-field magnetic resonance imaging technique, which utilizes a relatively low magnetic field strength, usually between 0.2 Tesla (T) and 0.5 Tesla (T).

[0033] It should be noted that the size of the image block is not further limited in the present invention and can be flexibly set according to the size of the low-field magnetic resonance image to be enhanced.

[0034] As an optional embodiment, it is assumed that the three-dimensional low-field magnetic resonance image array to be enhanced is 256×256×32 voxels, the size of each image block is 64×64×8 voxels, and the extraction step size is 32×32×4. Therefore, the low-field magnetic resonance image to be enhanced is divided into blocks to obtain 648 image blocks.

[0035] Step S402: input multiple image blocks into an image enhancement model to obtain multiple enhanced image blocks output by the image enhancement model.

[0036] The image enhancement model is trained based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

[0037] It should be noted that high magnetic field resonance images are medical images obtained through high-field magnetic resonance imaging technology. High-field magnetic resonance imaging generally uses a magnetic field of 1.5 Tesla (T) or 3 Tesla (T).

[0038] In an exemplary embodiment, the method of acquiring a synthetic low-field magnetic resonance image sample includes: A high-field magnetic resonance image sample and a low-field magnetic resonance image sample are acquired.

[0039] As an optional embodiment, the high-field magnetic resonance image samples and the low-field magnetic resonance image samples may be acquired in a manner known in the prior art, and the present invention does not impose any further limitation thereto.

[0040] Based on the slice thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field MRI image samples, the high-field MRI image samples are processed to obtain the synthetic low-field MRI images corresponding to the high-field MRI image samples.

[0041] As an optional embodiment, the slice thickness and spatial resolution of the low-field magnetic resonance image sample can be directly acquired from the low-field magnetic resonance image.

[0042] As an optional embodiment, if the high-field MRI image and the low-field MRI image are brain images, it is necessary to remove the skull and perform brain segmentation on the low-field MRI image and the high-field MRI image through statistical parameter mapping, and segment the brain into gray matter, white matter and cerebrospinal fluid to obtain a gray matter segmentation map, a white matter segmentation map and a cerebrospinal fluid segmentation map.

[0043] As an optional embodiment, after obtaining the segmented image, the signal-to-noise ratio of the gray matter segmentation map and the white matter segmentation map, as well as the corresponding covariance matrix, are calculated. Then, the gray matter average signal-to-noise ratio and the corresponding covariance matrix and the white matter average signal-to-noise ratio and the corresponding covariance matrix are used as parameters for synthesizing the low-field magnetic resonance image, and the high-field magnetic resonance image samples are subjected to low-field processing.

[0044] In an exemplary embodiment, based on the slice thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field magnetic resonance image samples, the high-field magnetic resonance image samples are subjected to low-field processing, including: The high-field MRI samples were Gaussian-filtered and blurred using the slice thickness as the full width at half maximum. The high-field MRI samples were downsampled according to the spatial resolution. The tissue contrast of the high-field MRI samples was adjusted according to the average signal-to-noise ratio of the gray matter and the corresponding covariance matrix and the average signal-to-noise ratio of the white matter and the corresponding covariance matrix.

[0045] As an optional embodiment, the low-field processing of the high-field magnetic resonance image sample is mainly divided into two steps.

[0046] In the first step, the high-field MRI image samples are Gaussian-filtered and blurred using the slice thickness of the low-field MRI image samples as the half-width at half maximum. Each voxel of the high-field MRI image samples is smoothed by the Gaussian filter, destroying the image details but maintaining the overall structure to simulate the blurring effect of the low-field MRI image. Then, the high-field MRI image samples are downsampled according to the spatial resolution of the low-field MRI image. For example, the size of a single voxel of the high-field MRI image sample is 1 mm × 1 mm × 1 mm, assuming that it has 256 × 256 × 256 voxels; the voxel size of the target low-field MRI image is 1 mm × 1 mm × 4 mm, assuming that it has 256 × 256 × 64 voxels. Then the original image is downsampled 4 times along the cross-sectional direction, and the image can be reduced from 256 × 256 × 256 voxels to 256 × 256 × 32 voxels using affine transformation and spline interpolation methods, keeping the overall structure unchanged and reducing the accuracy of details.

[0047] In the second step, the tissue contrast of the high-field MRI image samples is adjusted according to the average signal-to-noise ratio of the gray matter and the corresponding covariance matrix and the average signal-to-noise ratio of the white matter and the corresponding covariance matrix, so that the tissue contrast of the high-field MRI image after low-field processing is close to that of the low-field MRI image.

[0048] Step S403: splicing the multiple enhanced image blocks to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image.

[0049] As an optional embodiment, the size of the enhanced image block is consistent with the size of the image block, and the present invention does not further limit the splicing method.

[0050] Through the above steps S401-S403, after obtaining the low-field magnetic resonance image to be enhanced, firstly block processing is performed to generate multiple image blocks. Then, these image blocks are input into the trained image enhancement model to obtain multiple enhanced image blocks output by the model. Finally, these enhanced image blocks are synthesized through splicing processing to obtain an enhanced magnetic resonance image corresponding to the original low-field magnetic resonance image, and the image enhancement model is trained based on the high-field magnetic resonance image sample and its corresponding synthetic low-field magnetic resonance image. The present invention obtains the low-field magnetic resonance image corresponding to the high-field magnetic resonance image sample by synthesis, which reduces the difficulty of obtaining the high-field magnetic resonance image corresponding to the low-field magnetic resonance image, and the registration relationship between the high-field magnetic resonance image sample and the synthetic low-field magnetic resonance image is better. The high-field magnetic resonance image sample and its corresponding synthetic low-field magnetic resonance image are used for model training, which further improves the generalization ability of the model.

[0051] In an exemplary embodiment, the image enhancement model is a Mamba-based model obtained by improving the Mamba model. The Mamba-based model includes multiple Mamba encoders, each of which is connected to a downsampling module to receive features output by the downsampling module and output the extracted features to the next downsampling module.

[0052] As an optional embodiment, Figure 5 , which is a schematic diagram of the framework of the image enhancement model provided by the present invention. The present invention improves the Mamba model to obtain a Mamba-based model, and maintains feature transfer through jump connections. The Mamba-based model includes multiple Mamba encoders, each of which is connected to a downsampling module, and the Mamba encoder is mainly used for feature extraction. On this basis, the present invention replaces the multi-layer perceptron (MLP) of the Mamba model with a convolutional layer and channel attention.

[0053] As an optional embodiment, the downsampling module reduces the spatial resolution of the input image through pooling operations and extracts important feature information. The downsampled features are input to the next Mamba encoder, which further extracts and processes these features. Through this gradual downsampling and feature extraction, the image enhancement model can capture richer image information at different scales.

[0054] In an exemplary embodiment, the Mamba encoder is an improved Mamba encoder obtained by replacing the multilayer perceptron module with a new module including a convolutional layer and a channel attention layer, and in the improved Mamba encoder, the layer normalization module, the convolutional layer, the channel attention layer, and the output layer are connected in sequence.

[0055] As an optional embodiment, Figure 5 The lower right corner shows the schematic diagram of the structure of the Mamba encoder. In the improved Mamba encoder, the layer normalization module, the convolution layer, the channel attention layer, and the output layer are connected in sequence. Among them, the layer normalization module performs normalization processing on the input image features. Then, the Mamba encoder extracts features through the convolution layer to capture the local features of the image. After the convolution layer, the present invention introduces a channel attention layer. The channel attention layer dynamically adjusts the weights of each channel according to the different channels of the image features, highlights important features, suppresses information redundancy in the channel, and optimizes the expression ability of the features. Finally, the processed feature map is passed to the next downsampling module or decoder through the output layer.

[0056] As an optional embodiment, after the encoder part completes the extraction of low-field magnetic resonance image features, the image enhancement model enters the decoding stage. The low-resolution feature map is restored to a higher resolution through the upsampling module. The features of the decoding stage are jump-connected with the features of the encoding stage.

[0057] In the above embodiment, by introducing multiple Mamba encoders, the improved model can extract and retain the detail information of the low-field magnetic resonance image at different resolution levels. Compared with the traditional convolutional model, the improved Mamba-based model has stronger feature extraction and recovery capabilities, thereby improving the image enhancement effect. The enhanced magnetic resonance image presents a higher signal-to-noise ratio and clearer structural details.

[0058] In an exemplary embodiment, the Mamba encoder adopts a spatially continuous input sequence scanning method.

[0059] As an optional embodiment, most of the image scanning in the prior art adopts Z-shaped scanning, for example, the image is partitioned according to a nine-square grid, and each row is 0-1-2, 3-4-5, 6-7-8 from left to right. The scanning method in the prior art is 0-1-2-3-4-5-6-7-8. The present invention adopts a spatially continuous input sequence scanning method, that is, a 0-1-2-5-4-3-6-7-8 scanning method. The connection relationship of the voxels in the edge area is improved without increasing the computational complexity within the block.

[0060] In the above embodiment, the spatial relationship between adjacent voxels can be retained to the greatest extent when extracting features through a spatially continuous input sequence scanning method, thereby improving the effectiveness of feature extraction.

[0061] Obviously, the above described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. In order to better understand the above method, the above process is described below in conjunction with the embodiments, but the technical solutions of the embodiments of the present invention are not limited.

[0062] The present invention introduces the state-space model Mamba into the field of low-field magnetic resonance image enhancement, uses the IQT method to simulate clinical low-field image patches from high-quality high-field data sets, does not rely on clinically collected high- and low-field data pairs, and ensures that the model can learn the contrast features in low-field images.

[0063] The present invention realizes isotropy of various anisotropic voxels by adopting a partial downsampling feature method in an encoder, and improves the resolution of a low-field image through a decoder.

[0064] Furthermore, the Mamba architecture was first proposed to process one-dimensional natural language sequences and is not suitable for three-dimensional magnetic resonance imaging. To address this difficulty, we adopted a spatially continuous sequence scanning method and alternately changed different scanning directions to make the model as adaptable to three-dimensional spatial input as possible.

[0065] In addition, we added a convolution block to learn neighboring voxels to enhance the model's perception of local features. Considering that the number of channels in the hidden layer of the Mamba architecture is doubled, we used a channel attention module to suppress information redundancy within the channel, replacing the previous multi-layer perceptron module and optimizing the expressiveness of features.

[0066] Specifically, Figure 6 FIG. 1 is a schematic diagram of a processing flow of a low-field magnetic resonance image enhancement method based on Mamba and image quality transfer provided by the present invention. The processing flow includes the following steps.

[0067] 1. Magnetic resonance scanning and data acquisition: First, obtain public high-quality magnetic resonance data sets and low-field magnetic resonance images of clinical subjects. The low-field magnetic resonance images are stored in the computer system in Nifti format.

[0068] 2. Data preprocessing: First, record the key information of the low-field MRI images, such as spatial resolution, slice thickness, and slice spacing. Use statistical parameter mapping to remove the skull and segment the brain from all images, and segment them into gray matter, white matter, cerebrospinal fluid, and other tissues. Calculate the average signal-to-noise ratio and corresponding covariance matrix of gray matter and white matter in the low-field MRI images, and use them as parameters for synthesizing low-field MRI images. The method of synthesizing low-field MRI images is mainly divided into two steps: (1) Use the low-field slice thickness as the half-peak full width for Gaussian filtering blur, and downsample the high-field image according to the spatial resolution of the low-field MRI image. (2) Adjust the tissue contrast of the high-field MRI image to a low field according to the calculated tissue signal-to-noise ratio and covariance of the low-field MRI image, and add random noise according to the signal-to-noise ratio.

[0069] 3. Extract small blocks: Extract corresponding small blocks from the synthetic low-field MRI and high-field MRI respectively, with a size of 32×32×4 and a step size of 16×16×2. There are overlapping parts between small blocks, and multiple small blocks will be combined for output when inferring the results. The small blocks are normalized and saved in npz format. The npz format is a commonly used array storage format for Numpy. In this way, data pairs that are easy to input into model training are obtained.

[0070] 4. Input data to the model: Read small data pairs from the npz file, divide the data into training and validation parts according to the proportion of the original image to avoid overfitting and loss of generalization ability. Then input it into the model.

[0071] 5. Reconstruct quality image blocks: In order to reconstruct high-quality magnetic resonance images more effectively, we proposed a low-field magnetic resonance image enhancement method based on Mamba and image quality transfer. The input data undergoes a series of feature extraction and data calculations according to the architecture we designed, and finally uses the convolutional layer to output the enhanced magnetic resonance image blocks. During training, the reconstructed results will be compared with the real samples in the constructed data pair, the mean square error between voxels will be calculated, the error will be derived and the various parameters in the model will be reversely updated. The parameters of the model will be iteratively optimized during training. During inference, data that has not been learned by the input model will be reconstructed according to the iterative parameters, and the reconstructed high-quality magnetic resonance image blocks will be mixed and spliced ​​to output high-quality magnetic resonance images, such as Figure 7 As shown, an enhanced magnetic resonance image is obtained based on the low-field magnetic resonance image enhancement method provided by the present invention.

[0072] In summary, the present invention realizes the enhancement and reconstruction of low-field magnetic resonance images through magnetic resonance scanning, data preprocessing technology and low-field magnetic resonance image enhancement algorithm. Low-field magnetic resonance image enhancement helps to improve the quality of magnetic resonance images and the accuracy of subsequent image analysis, such as Figure 8 FIG. 1 is a schematic diagram of the effect of performing hippocampal segmentation on low-field images and enhanced images provided by the present invention. The present invention helps to improve the coverage and screening efficiency of magnetic resonance scanning.

[0073] The low-field magnetic resonance image enhancement device provided by the present invention is described below. The low-field magnetic resonance image enhancement device described below and the low-field magnetic resonance image enhancement method described above can be referred to each other.

[0074] Fig. 9 Schematic diagram of the structure of the low-field magnetic resonance image enhancement device provided by the present invention. Fig. 9 As shown, the device comprises: A first processing module 910 is used to acquire a low-field magnetic resonance image to be enhanced and perform block processing to obtain a plurality of image blocks; A second processing module 920 is used to input multiple image blocks into the image enhancement model to obtain multiple enhanced image blocks output by the image enhancement model; A splicing module 930 is used to splice the multiple enhanced image blocks to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; The image enhancement model is trained based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

[0075] After acquiring the low-field magnetic resonance image to be enhanced, the above-mentioned device first performs block processing to generate multiple image blocks. Then, these image blocks are input into a trained image enhancement model to obtain multiple enhanced image blocks output by the model. Finally, these enhanced image blocks are synthesized through splicing processing to obtain an enhanced magnetic resonance image corresponding to the original low-field magnetic resonance image. The image enhancement model is trained based on high-field magnetic resonance image samples and their corresponding synthetic low-field magnetic resonance images. The present invention obtains low-field magnetic resonance images corresponding to high-field magnetic resonance image samples by synthesis, which reduces the difficulty of obtaining high-field magnetic resonance images corresponding to low-field magnetic resonance images, and the registration relationship between high-field magnetic resonance image samples and synthetic low-field magnetic resonance images is better. The high-field magnetic resonance image samples and their corresponding synthetic low-field magnetic resonance images are used for model training, which further improves the generalization ability of the model.

[0076] In some embodiments, the method for acquiring synthetic low-field magnetic resonance image samples includes: acquiring high-field magnetic resonance image samples and low-field magnetic resonance image samples; based on the layer thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field magnetic resonance image samples, performing low-field processing on the high-field magnetic resonance image samples to obtain a synthetic low-field magnetic resonance image corresponding to the high-field magnetic resonance image samples.

[0077] In some embodiments, based on the slice thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field magnetic resonance image samples, the high-field magnetic resonance image samples are subjected to low-field processing, including: Gaussian filtering and blurring the high-field magnetic resonance image samples using the slice thickness as the half-maximum full width; down-sampling the high-field magnetic resonance image samples according to the spatial resolution; and adjusting the tissue contrast of the high-field magnetic resonance image samples according to the gray matter average signal-to-noise ratio and corresponding covariance matrix and the white matter average signal-to-noise ratio and corresponding covariance matrix.

[0078] In some embodiments, the image enhancement model is a Mamba-based model obtained by improving the Mamba model. The Mamba-based model includes multiple Mamba encoders, each of which is connected to a downsampling module to receive features output by the downsampling module and output the extracted features to the next downsampling module.

[0079] In some embodiments, the Mamba encoder is an improved Mamba encoder obtained by replacing the multilayer perceptron module with a new module including a convolutional layer and a channel attention layer, and in the improved Mamba encoder, the layer normalization module, the convolutional layer, the channel attention layer, and the output layer are connected in sequence.

[0080] In some embodiments, the Mamba encoder uses a spatially continuous input sequence scanning method.

[0081] Fig.10 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.10 As shown, the electronic device may include: a processor 1010, a communication interface 1020, a memory 1030 and a communication bus 1040, wherein the processor 1010, the communication interface 1020 and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute the low-field magnetic resonance image enhancement method, which includes: Acquiring a low-field magnetic resonance image to be enhanced and performing block processing to obtain a plurality of image blocks; Inputting a plurality of image blocks into an image enhancement model to obtain a plurality of enhanced image blocks output by the image enhancement model; The plurality of enhanced image blocks are spliced ​​to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; The image enhancement model is trained based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

[0082] In addition, the logic instructions in the above-mentioned memory 1030 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0083] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can perform the low-field magnetic resonance image enhancement method provided by the above methods, the method comprising: Acquiring a low-field magnetic resonance image to be enhanced and performing block processing to obtain a plurality of image blocks; Inputting a plurality of image blocks into an image enhancement model to obtain a plurality of enhanced image blocks output by the image enhancement model; The plurality of enhanced image blocks are spliced ​​to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; The image enhancement model is trained based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

[0084] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the method for enhancing the low-field magnetic resonance image provided by the above methods, the method comprising: Acquiring a low-field magnetic resonance image to be enhanced and performing block processing to obtain a plurality of image blocks; Inputting a plurality of image blocks into an image enhancement model to obtain a plurality of enhanced image blocks output by the image enhancement model; The plurality of enhanced image blocks are spliced ​​to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; The image enhancement model is trained based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

[0085] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0086] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for enhancing a low-field magnetic resonance image, characterized in that: include: Acquiring a low-field magnetic resonance image to be enhanced and performing block processing to obtain a plurality of image blocks; Inputting the multiple image blocks into an image enhancement model to obtain multiple enhanced image blocks output by the image enhancement model; Splicing the multiple enhanced image blocks to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; The image enhancement model is obtained by training based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

2. The method for enhancing low-field magnetic resonance images according to claim 1, characterized in that: The method for acquiring the synthetic low-field magnetic resonance image sample includes: acquiring a high-field magnetic resonance image sample and a low-field magnetic resonance image sample; Based on the slice thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field magnetic resonance image samples, the high-field magnetic resonance image samples are subjected to low-field processing to obtain a synthetic low-field magnetic resonance image corresponding to the high-field magnetic resonance image sample.

3. The method for enhancing low-field magnetic resonance images according to claim 2, characterized in that: The low-field processing of the high-field magnetic resonance image samples based on the slice thickness, spatial resolution, gray matter average signal-to-noise ratio and corresponding covariance matrix and white matter average signal-to-noise ratio and corresponding covariance matrix of the low-field magnetic resonance image samples includes: The high-field magnetic resonance image samples are Gaussian-filtered and blurred using the layer thickness as the full width at half maximum; the high-field magnetic resonance image samples are downsampled according to the spatial resolution; and the tissue contrast of the high-field magnetic resonance image samples is adjusted according to the gray matter average signal-to-noise ratio and the corresponding covariance matrix and the white matter average signal-to-noise ratio and the corresponding covariance matrix.

4. The method for enhancing low-field magnetic resonance images according to any one of claims 1 to 3, characterized in that: The image enhancement model is a Mamba-based model obtained by improving the Mamba model. The Mamba-based model includes multiple Mamba encoders, each of which is connected to a downsampling module to receive the features output by the downsampling module and output the extracted features to the next downsampling module.

5. The method for enhancing low-field magnetic resonance images according to claim 4, characterized in that: The Mamba encoder is an improved Mamba encoder obtained by replacing a multilayer perceptron module with a new module including a convolutional layer and a channel attention layer, and in the improved Mamba encoder, a layer normalization module, the convolutional layer, the channel attention layer, and an output layer are connected in sequence.

6. The method for enhancing low-field magnetic resonance images according to claim 4, characterized in that: The Mamba encoder adopts a spatially continuous input sequence scanning method.

7. A low-field magnetic resonance image enhancement device, characterized in that: include: A first processing module is used to acquire a low-field magnetic resonance image to be enhanced and perform block processing to obtain a plurality of image blocks; A second processing module, used for inputting the multiple image blocks into an image enhancement model to obtain multiple enhanced image blocks output by the image enhancement model; A splicing module, used for splicing the multiple enhanced image blocks to obtain an enhanced magnetic resonance image corresponding to the low-field magnetic resonance image; The image enhancement model is obtained by training based on high-field magnetic resonance image samples and synthetic low-field magnetic resonance images corresponding to the high-field magnetic resonance image samples.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for enhancing a low-field magnetic resonance image according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for enhancing a low-field magnetic resonance image according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for enhancing a low-field magnetic resonance image according to any one of claims 1 to 6 is implemented.

Images