MRI image reconstruction method and device based on feature mapping
Through the feature mapping-based method, using shallow feature extraction and nonlinear mapping, the problem of structural details loss in MRI image reconstruction is solved, and high-quality image reconstruction is achieved.
Patent Information
- Application Number
- CN202510124846.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-26
- Publication Date
- 2025-06-03
AI Technical Summary
MRI image reconstruction method can easily lead to loss of structural details.
A feature mapping-based method is used to construct reconstruction models through shallow feature extraction and nonlinear mapping to eliminate the problem of structural details loss. The specific steps include acquiring the MRI image, performing shallow feature extraction, nonlinear mapping, and finally performing image reconstruction.
It effectively eliminates the problem of structural details loss during MRI image reconstruction, maintains image texture details and structural details, and improves the quality of the reconstruction image.
Smart Images
Figure CN120088356A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of medical image recognition. Specifically, it relates to a method and device for MRI image reconstruction based on feature mapping. Background Art
[0002] Registering and fusing CT images and MRI images is crucial for computer-aided diagnosis or surgical navigation and positioning. However, the inconsistent slice thicknesses of CT images and MRI images are one of the main difficulties in registering and fusing CT images and MRI images.
[0003] Generally, an interpolation algorithm is used to insert intermediate slices between adjacent MRI images. However, this method of reconstructing MRI images is prone to losing structural details. Summary of the Invention
[0004] The problem solved by this application is that the method of reconstructing MRI images is prone to losing details.
[0005] To solve the above problem, a first aspect of this application provides a method for MRI image reconstruction based on feature mapping, including:
[0006] Obtain the MRI medical image of any object;
[0007] Input the MRI medical image into a shallow feature extraction structure to obtain a feature extraction map;
[0008] Input the feature extraction map into a non-linear mapping structure to obtain a mapped feature map;
[0009] Input the mapped feature map into an image reconstruction structure to obtain a reconstructed MRI image; the slice thickness of the MRI medical image is greater than that of the reconstructed MRI image.
[0010] A second aspect of this application provides a device for MRI image reconstruction based on feature mapping, which includes:
[0011] An image acquisition module, which is used to obtain the MRI medical image of any object;
[0012] A feature extraction module, which is used to input the MRI medical image into a shallow feature extraction structure to obtain a feature extraction map;
[0013] A feature mapping module, which is used to input the feature extraction map into a non-linear mapping structure to obtain a mapped feature map;
[0014] An image reconstruction module, which is used to input the mapped feature map into an image reconstruction structure to obtain a reconstructed MRI image; the slice thickness of the MRI medical image is greater than that of the reconstructed MRI image.
[0015] A third aspect of the present application provides an electronic device, which includes: a memory and a processor;
[0016] The memory is used for storing a program;
[0017] The processor, coupled to the memory, is used for executing the program to:
[0018] Obtain the MRI medical image of any object;
[0019] Input the MRI medical image into a shallow feature extraction structure to obtain a feature extraction map;
[0020] Input the feature extraction map into a non-linear mapping structure to obtain a mapped feature map;
[0021] Input the mapped feature map into an image reconstruction structure to obtain a reconstructed MRI image; the slice thickness of the MRI medical image is greater than that of the reconstructed MRI image.
[0022] A fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored, and the program is executed by a processor to implement the above-mentioned MRI image reconstruction method based on feature mapping.
[0023] In the present application, a reconstruction model is constructed through shallow extraction and non-linear mapping, so as to eliminate the problem of loss of structural details in the process of MRI image reconstruction. Description of the Drawings
[0024] Figure 1 It is a flowchart of the MRI image reconstruction method based on feature mapping according to an embodiment of the present application;
[0025] Figure 2 It is a model architecture diagram of the MRI image reconstruction method based on feature mapping according to an embodiment of the present application;
[0026] Figure 3 It is an architecture diagram of the deep feature extraction module of the MRI image reconstruction method based on feature mapping according to an embodiment of the present application;
[0027] Figure 4 It is an architecture diagram of the residual bottleneck processing of the MRI image reconstruction method based on feature mapping according to an embodiment of the present application;
[0028] Figure 5 It is an architecture diagram of the side residual convolution processing of the MRI image reconstruction method based on feature mapping according to an embodiment of the present application;
[0029] Figure 6 It is a structural block diagram of the MRI image reconstruction device based on feature mapping according to an embodiment of the present application;
[0030] Figure 7 Block diagram of an electronic device according to an embodiment of the present application. Detailed implementation manners
[0031] To make the above objects, features, and advantages of the present application more obvious and understandable, the following will describe in detail the specific embodiments of the present application with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.
[0032] It should be noted that unless otherwise specified, the technical terms or scientific terms used in the present application should have the ordinary meaning understood by those skilled in the art to which the present application belongs.
[0033] To address the above problems, the present application provides a new MRI image reconstruction scheme based on feature mapping, which can construct a reconstruction model through shallow extraction and non-linear mapping to eliminate the problem of loss of structural details during MRI image reconstruction.
[0034] An embodiment of the present application provides an MRI image reconstruction method based on feature mapping. The specific scheme of this method is Figures 1 - 5 as shown. This method can be executed by an MRI image reconstruction device based on feature mapping, and this MRI image reconstruction device based on feature mapping can be integrated into electronic devices such as computers, servers, computers, server clusters, and data centers. Combining Figure 1 、 Figure 2 as shown, it is a flowchart of an MRI image reconstruction method based on feature mapping according to an embodiment of the present application; wherein, the MRI image reconstruction method based on feature mapping includes:
[0035] S101, obtain the MRI medical image of any object;
[0036] Among them, the MRI medical image (Magnetic Resonance Imaging) is a medical imaging technology based on the nuclear magnetic resonance phenomenon, which uses a strong magnetic field and radio frequency pulses to acquire images of internal tissues and organs of the human body. S102, input the MRI medical image into the shallow feature extraction structure to obtain a feature extraction map;
[0037] S103, input the feature extraction map into the non-linear mapping structure to obtain a mapped feature map;
[0038] S104, input the mapped feature map into the image reconstruction structure to obtain a reconstructed MRI image; the slice thickness of the MRI medical image is greater than that of the reconstructed MRI image.
[0039] In this application, a reconstruction model is constructed through shallow extraction and non-linear mapping, thereby eliminating the problem of loss of structural details during the MRI image reconstruction process.
[0040] Combined with Figure 2 As shown, the designed network mainly consists of three parts: a shallow feature extraction structure (Shallow Feature Extraction, SFE), a non-linear mapping structure (Encoder-decoder Feature Extraction, EFE), and an image reconstruction structure, namely a sub-pixel convolution layer (Subpixel convolution upsampling, SCU).
[0041] Preferably, the slice thickness of the reconstructed MRI image is 1 mm, and the slice thickness of the MRI medical image is 5 mm.
[0042] In this application, by reconstructing the MRI image, the large-slice-thickness image is reconstructed into a small-slice-thickness image, and structural details are maintained during the reconstruction process, avoiding the problem of loss of structural details.
[0043] In this application, through non-linear mapping, deep feature extraction is performed to obtain deep features such as image texture details and structural details, so that image texture details, structural details, etc. are retained during the entire model processing process.
[0044] In this application, the reconstruction structure is a sub-pixel convolution layer.
[0045] It should be noted that the sub-pixel convolution layer converts a low-resolution feature map into a high-resolution output image without the need for traditional interpolation (such as bilinear interpolation) for upsampling. Therefore, in order to maintain the size consistency between the input MRI medical image and the reconstructed MRI image, downsampling can be added during the previous convolution processing (or directly set downsampling before the reconstruction structure).
[0046] In one implementation, combined with Figure 2 As shown, for S102, inputting the MRI medical image into the shallow feature extraction structure to obtain a feature extraction map includes:
[0047] Performing dilated convolution processing on the MRI medical image to obtain a first dilated feature map;
[0048] Performing activation processing and dilated convolution processing on the first dilated feature map to obtain a second dilated feature map;
[0049] Performing activation processing on the second dilated feature map to obtain the feature extraction map.
[0050] Combined withFigure 2 As shown in Figure 2 , the MRI medical image extracts the shallow features of the image through two convolutional layers of size 3×3; the convolutional layer is a dilated convolution, and each dilated convolution is followed by an activation function.
[0051] In this application, dilated convolution is also known as atrous convolution or expansive convolution, which can enlarge the receptive field without losing information.
[0052] In one embodiment, in combination with Figure 2 As shown in Figure 2 , in step S103, the feature extraction map is input into the non-linear mapping structure to obtain a mapped feature map, including:
[0053] The feature extraction map is input into a cascaded deep extraction module to obtain a deep feature map;
[0054] The feature extraction map and the deep feature map are concatenated to obtain a concatenated feature map;
[0055] The concatenated feature map is processed by continuous dilated convolution to obtain a continuous feature map;
[0056] After adding the continuous feature map and the MRI medical image, the mapped feature map is obtained.
[0057] In one embodiment, in combination with Figure 2 As shown in Figure 2 , there are four cascaded deep extraction modules, and the output map of the previous deep extraction module is used as the input map of the next deep extraction module.
[0058] Among them, in combination with Figure 2 As shown in Figure 2 , there are four cascaded deep extraction modules (EDB). The feature extraction map is input into the first deep extraction module to obtain a first map, the first map is input into the second deep extraction module to obtain a second map, then the second map is input into the third deep extraction module to obtain a third map, and then the third map is input into the fourth deep extraction module to obtain a deep feature map.
[0059] Among them, in combination with Figure 2 As shown in Figure 2 , in the step of concatenating the feature extraction map and the deep feature map to obtain a concatenated feature map, the feature extraction map is concatenated with the first map, the second map, the third map, and the deep feature map to obtain a concatenated feature map.
[0060] Among them, in the step of processing the concatenated feature map by continuous dilated convolution to obtain a continuous feature map, the concatenated feature map is processed by two consecutive dilated convolutions to obtain a continuous feature map.
[0061] Among them, the concatenation process is to merge multiple feature maps along a certain dimension (usually the channel dimension). The dimension of the output feature map after concatenation will increase.
[0062] In this application, the splicing will not lose the information of the original feature map, and all input features will be completely retained; multiple features can be utilized simultaneously to make the network more expressive.
[0063] Among them, the addition operation is to add multiple feature maps pixel by pixel (it is necessary to ensure that the shapes of the feature maps are the same). The dimension of the added feature map is the same as that of the input feature map.
[0064] In this application, the addition process will not increase the number of channels of the feature map, so the computational cost is relatively small; the problem of gradient disappearance is alleviated through residual learning.
[0065] In this application, through cascaded deep extraction, the shallow features extracted are represented at a deeper level to obtain deep features such as image texture details.
[0066] In this application, the cascaded deep extraction module constitutes the restoration of the texture detail information of the image. Through splicing and addition, the problems of gradient disappearance and network degradation are improved.
[0067] In one implementation, combined with Figure 3 as shown, the processing process of the deep extraction module includes:
[0068] After performing convolution processing on the input feature map, continuous pooling processing is carried out to obtain the first pooling map, the second pooling map, and the third pooling map in sequence;
[0069] The first pooling map, the second pooling map, and the third pooling map are spliced and convolved to obtain a convolutional pooling map;
[0070] Residual bottleneck processing is performed on the convolutional pooling map to obtain the output feature map.
[0071] It should be noted that the deep extraction module in this application is the specific content of any one of the four cascaded deep extraction modules; the structures of the four deep extraction modules are the same, only the specific parameters are different.
[0072] Among them, combined with Figure 3 as shown, convolution processing and three pooling processes are performed on the input feature map to obtain the convolved convolutional map and the first pooling map, the second pooling map, and the third pooling map after pooling; the first pooling map, the second pooling map, and the third pooling map, as well as the convolutional map, are spliced and then convolved, and finally residual bottleneck processing is performed to obtain the output feature map.
[0073] Among them, the three pooling processes are respectively three maximum pooling processes with 5×5, 9×9, and 13×13 pooling kernels; the feature map processed by the convolutional block is spliced and fused with the three feature maps processed by the maximum pooling to extract the final feature map.
[0074] In this application, through residual bottleneck processing, the pooled features are further feature-extracted in a cascaded manner to obtain semantically abstract features, enhancing the feature extraction ability of the model.
[0075] In this application, by serially cascading pooling kernels, feature information at different receptive field levels can be obtained, and each pooling layer directly outputs to the next layer, making the information transfer more direct and efficient.
[0076] In this application, the feature information of different scales after pooling and the feature information before pooling are concatenated and stacked together in a residual connection manner, which can not only integrate multi-scale local feature information but also enable the network to have a global perspective, helping the model better understand the internal structure and rules of the features.
[0077] In one implementation, combined with Figure 4 as shown, performing residual bottleneck processing on the convolutional pooling graph to obtain the output feature map includes:
[0078] Performing residual convolution processing on the convolutional pooling graph to obtain a residual convolution graph;
[0079] Performing normalization processing, activation processing, and channel convolution processing on the residual convolution graph to obtain a channel convolution graph;
[0080] Performing normalization processing, activation processing, residual convolution processing, and normalization processing on the channel convolution graph to obtain a normalized feature map;
[0081] Adding the convolutional pooling graph and the normalized feature map to obtain the output feature map.
[0082] In this application, combined with Figure 4 as shown, performing two residual convolution processes on the convolutional pooling graph, the first is used to increase the feature dimension to generate rich feature information, and the second is used to reduce the feature dimension to make it consistent with the content of the skip connection.
[0083] Among them, the shortcut path is connected between the input and output of the two residual convolution processes (the input of the first and the output of the second). Batch Normalization (BN) and ReLU non-linear activation are used after each layer.
[0084] In this application, through residual bottleneck processing, the number of model parameters is greatly reduced.
[0085] In this application, a channel convolution process is added between the two residual convolution processes.
[0086] Among them, DW convolution is also known as channel convolution, and its characteristic is that each convolution kernel only processes one channel of the input feature map. Specifically, DW convolution performs convolution operations independently on each input channel, rather than processing all channels simultaneously like ordinary convolution. This characteristic enables DW convolution to maintain spatial information while reducing the computational complexity and the number of parameters when processing images.
[0087] In this application, normalization adjusts the distribution of data to improve the training effect of the network and the generalization ability of the model.
[0088] In this application, the activation function introduces non-linearity to enable the neural network to learn complex non-linear features.
[0089] In one implementation, in combination with Figure 5 as shown, performing residual convolution processing on the convolutional pooling map to obtain a residual convolution map includes:
[0090] Performing convolution processing on the convolutional pooling map to obtain a feature map;
[0091] Performing a linear transformation on each channel of the feature map to generate a shadow feature map;
[0092] Concatenating the feature map and the shadow feature map to obtain a residual convolution map.
[0093] In this application, in combination with Figure 5 as shown, the residual convolution processing is divided into three parts: convolution part, linear transformation operation, and feature concatenation. First, use traditional convolution to obtain feature map I (feature map); then process the inherent feature map of each channel through the linear transformation operation Φ (similar to 3×3 convolution) to generate a shadow feature map; finally, connect the feature map obtained in the first step and the shadow feature map obtained in the second step to obtain the final output result. In this way, it is possible to obtain feature information similar to that of ordinary convolution under the premise of a relatively small computational cost and a relatively small number of parameters.
[0094] The embodiment of this application provides an MRI image reconstruction device based on feature maps, which is used to execute an MRI image reconstruction method based on feature maps described above in this application. The following provides a detailed description of the MRI image reconstruction device based on feature maps.
[0095] As Figure 6 shown, the MRI image reconstruction device based on feature maps includes:
[0096] An image acquisition module 101, which is used to acquire the MRI medical image of any object;
[0097] A feature extraction module 102, which is used to input the MRI medical image into a shallow feature extraction structure to obtain a feature extraction map;
[0098] A feature mapping module 103, which is used to input the feature extraction map into a non-linear mapping structure to obtain a mapped feature map;
[0099] An image reconstruction module 104, which is used to input the mapped feature map into an image reconstruction structure to obtain a reconstructed MRI image; the slice thickness of the MRI medical image is greater than that of the reconstructed MRI image.
[0100] In one implementation, the feature extraction module 102 is further used for:
[0101] Performing dilated convolution processing on the MRI medical image to obtain a first dilated feature map; performing activation processing and dilated convolution processing on the first dilated feature map to obtain a second dilated feature map; performing activation processing on the second dilated feature map to obtain the feature extraction map.
[0102] In one implementation, the feature mapping module 103 is further used for:
[0103] Inputting the feature extraction map into a cascaded deep extraction module to obtain a deep feature map; splicing the feature extraction map and the deep feature map to obtain a spliced feature map; performing continuous dilated convolution processing on the spliced feature map to obtain a continuous feature map; adding the continuous feature map and the MRI medical image to obtain the mapped feature map.
[0104] In one implementation, there are four cascaded deep extraction modules, and the output map of the previous deep extraction module is used as the input map of the next deep extraction module.
[0105] In one implementation, the feature mapping module 103 is further used for:
[0106] After performing convolution processing on the input feature map, performing continuous pooling processing to obtain a first pooling map, a second pooling map, and a third pooling map in sequence; splicing and performing convolution processing on the first pooling map, the second pooling map, and the third pooling map to obtain a convolution pooling map; performing a residual bottleneck process on the convolution pooling map to obtain the output feature map.
[0107] In one implementation, the feature mapping module 103 is further used for:
[0108] Performing residual convolution processing on the convolution pooling map to obtain a residual convolution map; performing normalization processing, activation processing, and channel convolution processing on the residual convolution map to obtain a channel convolution map; performing normalization processing, activation processing, residual convolution processing, and normalization processing on the channel convolution map to obtain a normalized feature map; adding the convolution pooling map and the normalized feature map to obtain the output feature map.
[0109] In one embodiment, the feature mapping module 103 is further configured to:
[0110] Perform convolution processing on the convolution pooling graph to obtain a feature mapping graph; perform a linear transformation on each channel of the feature mapping graph to generate a shadow feature graph; splice the feature mapping graph and the shadow feature graph to obtain a residual convolution graph.
[0111] An MRI image reconstruction device based on feature mapping provided by the above embodiments of the present application and an MRI image reconstruction method based on feature mapping provided by the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.
[0112] The internal functions and structures of an MRI image reconstruction device based on feature mapping are described above. As Figure 7 shown, in practice, the MRI image reconstruction device based on feature mapping can be implemented as an electronic device, including: a memory 301 and a processor 303.
[0113] The memory 301 can be configured to store programs.
[0114] In addition, the memory 301 can also be configured to store various other data to support operations on the electronic device. Examples of these data include instructions for any application program or method for operating on the electronic device, contact data, phone book data, messages, pictures, videos, etc.
[0115] The memory 301 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disk.
[0116] The processor 303 is coupled to the memory 301 and is configured to execute the programs in the memory 301 for:
[0117] Obtain an MRI medical image of any object;
[0118] Input the MRI medical image into a shallow feature extraction structure to obtain a feature extraction graph;
[0119] Input the feature extraction graph into a non-linear mapping structure to obtain a mapped feature graph;
[0120] Input the mapped feature graph into an image reconstruction structure to obtain a reconstructed MRI image; the slice thickness of the MRI medical image is greater than that of the reconstructed MRI image.
[0121] In one embodiment, the processor 303 is further configured to:
[0122] Perform dilated convolution processing on the MRI medical image to obtain a first dilated feature map; perform activation processing and dilated convolution processing on the first dilated feature map to obtain a second dilated feature map; perform activation processing on the second dilated feature map to obtain the feature extraction map.
[0123] In one embodiment, the processor 303 is further configured to:
[0124] Input the feature extraction map into a cascaded deep extraction module to obtain a deep feature map; splice the feature extraction map and the deep feature map to obtain a spliced feature map; perform continuous dilated convolution processing on the spliced feature map to obtain a continuous feature map; add the continuous feature map and the MRI medical image to obtain the mapped feature map.
[0125] In one embodiment, there are four cascaded deep extraction modules, and the output map of the previous deep extraction module serves as the input map of the subsequent deep extraction module.
[0126] In one embodiment, the processor 303 is further configured to:
[0127] After performing convolution processing on the input feature map, perform continuous pooling processing to obtain a first pooling map, a second pooling map, and a third pooling map in sequence; splice and perform convolution processing on the first pooling map, the second pooling map, and the third pooling map to obtain a convolution pooling map; perform residual bottleneck processing on the convolution pooling map to obtain the output feature map.
[0128] In one embodiment, the processor 303 is further configured to:
[0129] Perform residual convolution processing on the convolution pooling map to obtain a residual convolution map; perform normalization processing, activation processing, and channel convolution processing on the residual convolution map to obtain a channel convolution map; perform normalization processing, activation processing, residual convolution processing, and normalization processing on the channel convolution map to obtain a normalized feature map; add the convolution pooling map and the normalized feature map to obtain the output feature map.
[0130] In one embodiment, the processor 303 is further configured to:
[0131] Perform convolution processing on the convolution pooling map to obtain a feature mapping map; perform linear transformation on each channel of the feature mapping map to generate a shadow feature map; splice the feature mapping map and the shadow feature map to obtain a residual convolution map.
[0132] In this application, Figure 7 Only some components are schematically shown, and it does not mean that the electronic device only includes Figure 7 the components shown.
[0133] The electronic device provided in this embodiment is based on the same inventive concept as a method for reconstructing MRI images based on feature mapping provided in an embodiment of the present application, and has the same beneficial effects as the method adopted, run, or implemented by the application program stored therein.
[0134] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.
[0135] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.
[0136] These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.
[0137] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.
[0138] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.
[0139] The memory may include non-permanent memory in the form of computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (Flash RAM). The memory is an example of computer-readable media.
[0140] The present application also provides a computer-readable storage medium corresponding to a method for reconstructing an MRI image based on feature mapping provided in the foregoing embodiments. A computer program (i.e., a program product) is stored thereon. When the computer program is run by a processor, it will execute a method for reconstructing an MRI image based on feature mapping provided in any of the foregoing embodiments.
[0141] Computer-readable media includes permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology for storing information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.
[0142] The computer-readable storage medium provided in the above embodiments of the present application and a method for reconstructing an MRI image based on feature mapping provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the method adopted, run, or implemented by the application program stored thereon.
[0143] It should be noted that in the specification provided herein, a large number of specific details are set forth. However, it can be understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known structures and technologies are not shown in detail so as not to obscure the understanding of this specification.
[0144] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the said element.
[0145] The above are only the embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.
Claims
1. A method for MRI image reconstruction based on feature mapping, characterized in that: include: Acquiring MRI medical images of any subject; Inputting the MRI medical image into a shallow feature extraction structure to obtain a feature extraction graph; Input the feature extraction map into the nonlinear mapping structure to obtain a mapping feature map; The mapping feature map is input into an image reconstruction structure to obtain a reconstructed MRI image; the layer thickness of the MRI medical image is greater than that of the reconstructed MRI image.
2. The MRI image reconstruction method based on feature mapping according to claim 1, characterized in that: The step of inputting the MRI medical image into a shallow feature extraction structure to obtain a feature extraction graph comprises: Performing dilated convolution processing on the MRI medical image to obtain a first dilated feature map; Performing activation processing and hole convolution processing on the first hole feature map to obtain a second hole feature map; The second hole feature map is activated to obtain the feature extraction map.
3. The MRI image reconstruction method based on feature mapping according to claim 1 or 2, characterized in that: The step of inputting the feature extraction graph into a nonlinear mapping structure to obtain a mapping feature graph comprises: Input the feature extraction map into the cascaded deep extraction module to obtain a deep feature map; Splice the feature extraction map with the deep feature map to obtain a spliced feature map; Perform continuous dilated convolution on the concatenated feature map to obtain a continuous feature map; After the continuous feature map is added to the MRI medical image, the mapping feature map is obtained.
4. The MRI image reconstruction method based on feature mapping according to claim 3, characterized in that: There are four cascaded deep extraction modules, and the output graph of the previous deep extraction module serves as the input graph of the next deep extraction module.
5. The MRI image reconstruction method based on feature mapping according to claim 3, characterized in that: The processing process of the deep extraction module includes: After convolution processing of the input feature map, continuous pooling processing is performed to obtain the first pooling map, the second pooling map and the third pooling map in sequence; The first pooling map, the second pooling map and the third pooling map are concatenated and convolved to obtain a convolution pooling map; Perform residual bottleneck processing on the convolution pooling map to obtain the output feature map.
6. The MRI image reconstruction method based on feature mapping according to claim 5, characterized in that: The convolution pooling graph is subjected to residual bottleneck processing to obtain an output feature graph, including: Perform residual convolution on the convolution pooling map to obtain a residual convolution map; The residual convolution map is normalized, activated and channel convolved to obtain a channel convolution map; The channel convolution map is normalized, activated, and subjected to residual convolution and normalization to obtain a normalized feature map; Add the convolution pooling map and the normalized feature map to get the output feature map.
7. The MRI image reconstruction method based on feature mapping according to claim 6, characterized in that: The convolution pooling map is subjected to residual convolution processing to obtain a residual convolution map, including: Perform convolution processing on the convolution pooling map to obtain a feature map; Perform a linear transformation on each channel of the feature map to generate a shadow feature map; The feature map and the shadow feature map are concatenated to obtain the residual convolution map.
8. An MRI image reconstruction device based on feature mapping, characterized in that: include: An image acquisition module, which is used to acquire an MRI medical image of any object; A feature extraction module, which is used to input the MRI medical image into a shallow feature extraction structure to obtain a feature extraction graph; A feature mapping module, which is used to input the feature extraction map into a nonlinear mapping structure to obtain a mapping feature map; An image reconstruction module is used to input the mapping feature map into the image reconstruction structure to obtain a reconstructed MRI image; the layer thickness of the MRI medical image is greater than that of the reconstructed MRI image.
9. An electronic device, characterized in that: include: Memory and processor; The memory is used to store programs; The processor, coupled to the memory, is configured to execute the program to: Acquiring MRI medical images of any subject; Inputting the MRI medical image into a shallow feature extraction structure to obtain a feature extraction graph; Input the feature extraction map into the nonlinear mapping structure to obtain a mapping feature map; The mapping feature map is input into an image reconstruction structure to obtain a reconstructed MRI image; the layer thickness of the MRI medical image is greater than that of the reconstructed MRI image.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the MRI image reconstruction method based on feature mapping as described in any one of claims 1 to 7.