Multi-slice combined magnetic resonance fingerprint tissue parameter mapping method and related device

Abstract

This application discloses a method and related device for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprints, which relates to the fields of magnetic resonance fingerprint imaging and deep learning. The method includes: preprocessing the magnetic resonance fingerprint data of different slices to obtain noise-free magnetic resonance fingerprint data of different slices; inputting the noise-free magnetic resonance fingerprint data of three adjacent slices into a multi-slice feature fusion network, and outputting a fused magnetic resonance fingerprint feature map; inputting the fused magnetic resonance fingerprint feature map into a parameter reconstruction network, and outputting the tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice. This application uses the features of adjacent slices to enrich the features of the middle slice, strengthens the learning of local features through cross and attention operations, uses a fully connected network to reduce feature dimensionality, and uses a DRD-Net network to increase the global receptive field, enhance the learning of global features, and improve the quality of tissue parameter reconstruction of magnetic resonance fingerprints.

Description

Technical Field

[0001] The present application relates to the fields of magnetic resonance fingerprint imaging and deep learning, and in particular to a method and related apparatus for mapping tissue parameters based on multi-slice combined magnetic resonance fingerprints. Background Art

[0002] Magnetic Resonance Fingerprinting (MRF) was first published in the journal Nature in 2013. MRF is a novel, rapid quantitative imaging method that can simultaneously quantify multiple tissue characteristics in a relatively short period of time. The traditional MRF implementation process is as follows: first, magnetic resonance fingerprint data is quickly acquired through a special data acquisition method, then all possible tissue characteristics are simulated and measured using the Bloch equation to construct a dictionary, and finally, the dot product of each fingerprint signal is calculated with all signal vectors in the dictionary. The larger the dot product, the higher the match between the dictionary and the fingerprint signal. However, under the traditional MRF method, as the dictionary size increases, the difficulty of storage and calculation also increases.

[0003] To address these challenges, deep learning-based MRI fingerprint imaging has been proposed. This technology can directly map the measured MRI fingerprint signal to the underlying parameters using fewer time series points. However, this technology still suffers from the problem of low reconstruction quality of MRI fingerprint tissue parameters. Summary of the Invention

[0004] The applicant's research found that the magnetic resonance fingerprint imaging technology based on deep learning mainly uses a fully convolutional network, which lacks the extraction of global features. In addition, the technology does not consider the auxiliary role of the magnetic resonance fingerprints of adjacent slices in the reconstruction of the magnetic resonance fingerprint parameter map of the intermediate slice, resulting in low quality of magnetic resonance fingerprint tissue parameter reconstruction. In view of this, the purpose of this application is to provide a method and related device for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprints, aiming to improve the quality of magnetic resonance fingerprint tissue parameter reconstruction.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint, comprising:

[0007] Acquiring magnetic resonance fingerprint data of different slices; the magnetic resonance fingerprint data is time series data;

[0008] Preprocessing the magnetic resonance fingerprint data of the different slices to obtain noise-free magnetic resonance fingerprint data of the different slices;

[0009] Determining the noise-free magnetic resonance fingerprint data of three adjacent slices in the noise-free magnetic resonance fingerprint data of different slices as a group, and inputting the noise-free magnetic resonance fingerprint data of the different slices into a multi-slice feature fusion network in groups, and outputting a fused magnetic resonance fingerprint feature map; the multi-slice feature fusion network is used to perform cross and attention operations on the noise-free magnetic resonance fingerprint data of the three adjacent slices, to obtain a magnetic resonance fingerprint feature map after feature fusion of the noise-free magnetic resonance fingerprint data of the three adjacent slices;

[0010] The fused magnetic resonance fingerprint feature map is input into a parameter reconstruction network, and a tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice is output.

[0011] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-slice joint magnetic resonance fingerprint tissue parameter mapping method described in the first aspect.

[0012] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-slice joint magnetic resonance fingerprint tissue parameter mapping method described in the first aspect.

[0013] In a fourth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the multi-slice joint magnetic resonance fingerprint tissue parameter mapping method described in the first aspect.

[0014] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0015] The present application provides a method and related device for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint. The present application obtains magnetic resonance fingerprint data of different slices; preprocesses the magnetic resonance fingerprint data of different slices to obtain noise-free magnetic resonance fingerprint data of different slices; determines the noise-free magnetic resonance fingerprint data of three adjacent slices in the noise-free magnetic resonance fingerprint data of different slices as a group, and inputs the noise-free magnetic resonance fingerprint data of the different slices into a multi-slice feature fusion network in groups, outputs the fused magnetic resonance fingerprint feature map; inputs the fused magnetic resonance fingerprint feature map into a parameter reconstruction network, and outputs the tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice. The present application uses the features of adjacent slices to enrich the features of the middle slice, strengthens the learning of local features through cross and attention operations, uses a fully connected network to reduce the feature dimension, and improves the global receptive field through the DRD-Net network, enhances the learning of global features, and improves the quality of tissue parameter reconstruction of magnetic resonance fingerprints. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 This is a diagram of the application environment of the multi-slice combined magnetic resonance fingerprint tissue parameter mapping method in one embodiment of the present application;

[0018] Figure 2 A schematic diagram of a process for a multi-slice combined magnetic resonance fingerprint tissue parameter mapping method according to an embodiment of the present application;

[0019] Figure 3 A schematic diagram of the network structure of a multi-slice joint magnetic resonance fingerprint tissue parameter mapping method provided in one embodiment of the present application;

[0020] Figure 4 A schematic diagram of each Temporal SwinTransformer block in a multi-slice fusion network provided in an embodiment of the present application;

[0021] Figure 5 A schematic diagram of each dense block in a parameter reconstruction network provided in an embodiment of the present application;

[0022] Figure 6 This is a diagram showing the multi-slice joint magnetic resonance fingerprint tissue parameter mapping result provided in one embodiment of the present application;

[0023] Figure 7 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0026] The method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the magnetic resonance fingerprint data of different slices to the server 104. The server 104 receives the magnetic resonance fingerprint data of different slices. For the magnetic resonance fingerprint data of different slices, the server 104 pre-processes the magnetic resonance fingerprint data of the different slices to obtain noise-free magnetic resonance fingerprint data of different slices; selects the magnetic resonance fingerprint data of three adjacent slices from the noise-free magnetic resonance fingerprint data of the different slices, inputs them into the multi-slice feature fusion network, and outputs the fused magnetic resonance fingerprint feature map; the multi-slice feature fusion network is used to perform cross and attention operations on the magnetic resonance fingerprint data of the three adjacent slices to achieve feature fusion of the magnetic resonance fingerprint data of the three adjacent slices; the fused magnetic resonance fingerprint feature map is input into the parameter reconstruction network, and the tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice is output. The server 104 may feed back the parameter mapping result of the noise-free magnetic resonance fingerprint data of the intermediate slice to the terminal 102. In addition, in some embodiments, the tissue parameter mapping method based on multi-slice joint magnetic resonance fingerprint may also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 may directly process the magnetic resonance fingerprint data of different slices, or the server 104 may obtain the magnetic resonance fingerprint data of different slices from a data storage system and process the magnetic resonance fingerprint data of different slices.

[0027] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0028] In an exemplary embodiment, Figure 2 As shown, a method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the process, including the following steps 201 to 204.

[0029] Step 201: Acquire magnetic resonance fingerprint data of different slices.

[0030] Among them, the magnetic resonance fingerprint data of different slices are all time series data.

[0031] When obtaining magnetic resonance fingerprint data of different slices, it is necessary to design a sampling sequence with continuously changing parameters such as RF pulse, repetition time, and sampling trajectory, so that each different tissue type has a unique signal response under the sampling sequence.

[0032] Step 202 : pre-processing the magnetic resonance fingerprint data of the different slices to obtain noise-free magnetic resonance fingerprint data of the different slices.

[0033] In an exemplary embodiment, step 202 specifically includes:

[0034] Step 2021: Select some segments in the magnetic resonance fingerprint data of different slices, and determine the some segments in the magnetic resonance fingerprint data of different slices as data to be processed.

[0035] The purpose of step 2021 is to reduce the amount of data and shorten the data processing time, thereby improving efficiency. For example, the MRI fingerprint data of each slice originally includes 2000 frames of images. After selecting some fragments to obtain the data to be processed, the data to be processed only includes 200 frames of images.

[0036] Step 2022: extracting the foreground area of ​​the data to be processed, separating the target area of ​​interest from the background area, and obtaining noise-free magnetic resonance fingerprint data.

[0037] As an optional implementation, step 2022 specifically includes:

[0038] (1) According to a preset global threshold, the proton density parameter map of the data to be processed is filtered and binarized to obtain a first proton density parameter map.

[0039] (2) extracting the maximum connected domain of the first proton density parameter map based on a connected domain extraction algorithm, removing noise in the first proton density parameter map, and obtaining a second proton density parameter map.

[0040] (3) Filling holes in the second proton density parameter map to repair missing parts in the second proton density parameter map to obtain a third proton density parameter map; the third proton density parameter map is used as a reference map for extracting the foreground area.

[0041] (4) Performing a matrix multiplication operation on each frame of the data to be processed and the reference image of the extracted foreground area to obtain noise-free magnetic resonance fingerprint data.

[0042] like Figure 3As shown, the overall network of the multi-slice joint magnetic resonance fingerprint tissue parameter mapping method provided by the present application includes two parts: a multi-slice feature fusion network and a parameter reconstruction network. Before actual application, it is necessary to set a suitable learning rate, optimizer and loss function, and use the noise-free magnetic resonance fingerprint data of multiple slices to train the overall network. In actual application, the magnetic resonance fingerprint data of different slices are first preprocessed to obtain noise-free magnetic resonance fingerprint data of different slices; then, the magnetic resonance fingerprint data of three adjacent slices are selected from the noise-free magnetic resonance fingerprint data of the different slices and input into the overall network. The magnetic resonance fingerprint data of the three adjacent slices are sequentially subjected to the multi-slice feature fusion and parameter reconstruction process, and the tissue parameter mapping result of the magnetic resonance fingerprint data of the middle slice is output.

[0043] Step 203: The noise-free magnetic resonance fingerprint data of three adjacent slices in the noise-free magnetic resonance fingerprint data of different slices are determined as a group, and the noise-free magnetic resonance fingerprint data of the different slices are input into a multi-slice feature fusion network in groups, and a fused magnetic resonance fingerprint feature map is output; the multi-slice feature fusion network is used to perform cross and attention operations on the noise-free magnetic resonance fingerprint data of three adjacent slices to obtain a magnetic resonance fingerprint feature map after feature fusion of the noise-free magnetic resonance fingerprint data of the three adjacent slices.

[0044] In an exemplary embodiment, Figure 4 As shown in Figure 1, the multi-slice feature fusion network includes three attention branches, namely the upper attention branch, the middle attention branch and the lower attention branch; each attention branch is composed of an even number of Temporal SwinTransformer blocks.

[0045] The Temporal SwinTransformer block includes a block embedding unit, a channel merging unit, a first linear mapping unit, a multi-head self-attention unit, a second linear mapping unit, a first residual connection unit, a layer normalization unit, a multi-layer perceptron, a second residual connection unit, a block de-embedding unit and a 3×3 convolution unit.

[0046] The multi-head self-attention unit is a windowed multi-head self-attention unit or a sliding window multi-head self-attention unit. Within the same attention branch, the multi-head self-attention units in adjacent Temporal Swin Transformer blocks are different. For example, in the middle attention branch, if the first Temporal Swin Transformer block uses a W-MSA unit, the second Temporal Swin Transformer block uses a SW-MSA unit. Furthermore, all three Temporal Swin Transformer blocks in the same column use the same type of multi-head self-attention unit.

[0047] The first end of the block embedding unit is the input end of the Temporal SwinTransformer block; the second end of the block embedding unit is connected to the first end of the channel merging unit; the second end of the channel merging unit is connected to one end of the first linear mapping unit; the other end of the first linear mapping unit is connected to one end of the first residual connection unit; the other end of the first residual connection unit is connected to one end of the layer normalization unit; the other end of the layer normalization unit is connected to one end of the multi-layer perceptron; the other end of the multi-layer perceptron is connected to one end of the second residual connection unit; the other end of the second residual connection unit is connected to one end of the block de-embedding unit; the other end of the block de-embedding unit is connected to one end of the 3×3 convolution unit; the other end of the 3×3 convolution unit is the output end of the Temporal SwinTransformer block.

[0048] Since each attention branch is composed of an even number of Temporal SwinTransformer blocks, this ensures that W-MSA (Windows Multi-Head Self-Attention, W-MSA) and SW-MSA (Shifted Windows Multi-Head Self-Attention, SW-MSA) can be used alternately. Among them, W-MSA stands for the window multi-head self-attention mechanism, which constrains the calculation range of the self-attention mechanism by setting a fixed-size window in the Transformer architecture. SW-MSA stands for the sliding window multi-head self-attention mechanism, which introduces a sliding window and covers the entire sequence by calculating W-MSA multiple times. Each time the sliding window is passed, the self-attention weight within the window is recalculated. The number of windows is calculated as follows:

[0049]

[0050] Where window_size represents the set window size, W and H represent the length and height of the input feature map respectively, and the window size setting needs to satisfy the calculated number of windows N is an integer value.

[0051] The block embedding unit of the upper attention branch, the block embedding unit of the lower attention branch and the block embedding unit of the intermediate attention branch perform channel merging in the channel merging unit of the intermediate attention branch; the block embedding unit of the intermediate attention branch and the block embedding unit of the upper attention branch perform channel merging in the channel merging unit of the upper attention branch; the block embedding unit of the intermediate attention branch and the block embedding unit of the lower attention branch perform channel merging in the channel merging unit of the lower attention branch.

[0052] Furthermore, the third end of the block embedding unit of the upper attention branch is connected to the third end of the channel merging unit of the intermediate attention branch; the third end of the block embedding unit of the lower attention branch is connected to the fourth end of the channel merging unit of the intermediate attention branch; the third end of the block embedding unit of the intermediate attention branch is connected to the third end of the channel merging unit of the upper attention branch; the fourth end of the block embedding unit of the intermediate attention branch is connected to the third segment of the channel merging unit of the lower attention branch.

[0053] It should be noted that when the magnetic resonance fingerprint data of three adjacent slices are selected from the noise-free magnetic resonance fingerprint data of the multiple slices and input into the multi-slice feature fusion network, the input into the upper attention branch, the middle attention branch, and the lower attention branch is the magnetic resonance fingerprint data of the three adjacent slices that have been shortened. For example, assuming that the number of time frames of the original magnetic resonance fingerprint data is 2000, the inner product calculation is performed using a preset dictionary and each voxel signal of the magnetic resonance fingerprint data to obtain an accurate tissue parameter mapping result, which will serve as the prediction target of the multi-slice feature fusion network and the parameter reconstruction network. When training the network, in order to achieve acceleration, only a portion of the data is selected, and fingerprint data of, for example, 200 time frames are obtained for each slice, and normalization and data enhancement are performed. The acceleration factor in this case is 10.

[0054] In an exemplary embodiment, the noise-free MRI fingerprint data of different slices are input into the multi-slice feature fusion network in groups as follows: first, 200 frames of noise-free MRI fingerprint data corresponding to the first, second, and third slices are selected, respectively, and input into the upper attention branch, the middle attention branch, and the lower attention branch, and the tissue parameter mapping result of the second slice is output; second, 200 frames of noise-free MRI fingerprint data corresponding to the second, third, and fourth slices are selected, respectively, and input into the upper attention branch, the middle attention branch, and the lower attention branch, and the tissue parameter mapping result of the third slice is output, and so on. The optimization goal of the network of the present application is to make the parameter mapping result of each slice using 200 frames of MRI fingerprint data close to the tissue parameter mapping result obtained by dictionary matching using 2000 frames of MRI fingerprint data.

[0055] Dictionary matching is performed on longer sequence lengths, and the result obtained from dictionary matching will serve as the gold standard graph to supervise the learning process of the network. The optimization goal of the method provided in this application is to still be able to approach this gold standard graph when the sequence length is shorter.

[0056] Furthermore, Figure 4 The details of the multi-slice network are shown. First, the block embedding unit is used to complete the block embedding process, where the block size is P. The block embedding process can transform the feature dimensions of the three slice fingerprints of the input from [L, H, W] to [(L×H) / P 2 ,W], thereby adjusting the original image size and adding the absolute position encoding information to each block. Among them, W and H represent the length and height of the input feature map respectively, L represents the sequence length of the fingerprint map, and the size of P is 1. It can be seen that Figure 4 Each block embedding unit draws 8 rounded rectangles, and every 2 rounded rectangles are drawn together, one is used to represent the block information, and the other is used to represent the position coding information added to the block, that is, Figure 4 A total of four blocks are shown in the block embedding unit in FIG. However, those skilled in the art will appreciate that the four blocks shown in the figure are only for the purpose of illustrating the structure of the multi-slice feature network, and that the number of blocks is not limited to four in actual applications and may be any other number of blocks.

[0057] After the block embedding process is completed, the feature cross-merging process needs to be performed. Taking the intermediate attention branch as an example, C in the figure represents the channel merging unit, and the cross-merging of features is completed in the channel merging unit. After merging, the dimension of the image information will increase by 3 times. Therefore, it is necessary to set the first linear mapping unit ( Figure 4 The linear mapping 1 in the figure is used to reduce the dimension of the cross-merged information. Then, the multi-head self-attention unit is used to perform attention operation, and the second linear mapping unit ( Figure 4 The linear mapping in 2) adjusts the dimension of the image.

[0058] Next, a residual connection is performed on the output of the second linear mapping unit to enhance feature learning. The features are then normalized using a layer normalization unit, scaling the data to a smaller range to stabilize the training process. A multilayer perceptron is then used to further enhance feature learning, and a second residual connection unit is used to complete the residual connection again.

[0059] Finally, the de-embedding process is completed by the block de-embedding unit, and batch normalization and activation are performed through a 3x3 convolutional layer with a stride of 1 to further learn the features, resulting in the output of the Temporal Swin Transformer block. The block de-embedding unit restores the feature map of size [L×H,W] to [L,H,W]. The output of the Temporal Swin Transformer block is input to the next Temporal Swin Transformer block (if any) or to the parameter reconstruction network.

[0060] Step 204: Input the fused MRI fingerprint feature map into a parameter reconstruction network to output a tissue parameter mapping result of the noise-free MRI fingerprint data of the middle slice, where the middle slice refers to the middle slice between three adjacent slices.

[0061] In an exemplary embodiment, Figure 3 As shown, the parameter reconstruction network is a DRD-Net network; the DRD-Net network includes a fully connected layer, n dense blocks, a global feature fusion layer, a global residual connection layer, and an upsampling convolution layer. One end of the fully connected layer is connected to the output end of the multi-slice feature fusion network; the other end of the fully connected layer is connected to one end of the n dense blocks; the other end of the n dense blocks is connected to one end of the global feature fusion layer; the other end of the global feature fusion layer is connected to one end of the global residual connection layer; the other end of the global residual connection layer is connected to one end of the upsampling convolution layer; and the other end of the upsampling convolution layer is used to output the parameter mapping result of the noise-free magnetic resonance fingerprint data of the intermediate slice.

[0062] As an optional implementation, step 204 specifically includes:

[0063] Step 2041: input the fused magnetic resonance fingerprint feature map into the fully connected layer of the parameter reconstruction network to output a low-dimensional feature F0; the fully connected layer is a 1x1 convolutional layer.

[0064] Step 2042: Input the low-dimensional feature F0 into the n dense blocks for processing, and output n sub-features F1, F2, ..., F n In this embodiment, the value of n is 4, that is, the number of dense blocks is 4. Of course, the number of n can also be other values.

[0065] Step 2043: Input the n sub-features into the global feature fusion layer for global feature fusion and output the global feature. Among them, the feature fusion is completed through the 1x1 convolution layer and the 3x3 convolution layer to complete the global feature fusion and output the feature F GThe global feature fusion layer includes n sub-features F1, F2, ..., F n The combined feature pooling layer, a 1x1 convolutional layer, and a 3x3 convolutional layer.

[0066] Step 2044: Combine the low-dimensional feature F0 and the global feature F G Input the global residual connection layer and output the features after global residual connection;

[0067] Step 2045: Input the features after global residual connection into the upsampling convolution layer, and output the tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice. The upsampling convolution layer includes two 3x3 convolution layers.

[0068] As an optional implementation method, Figure 5 As shown, the specific details of each dense block in the parameter reconstruction network are as follows:

[0069] First, feature F n After inputting the dense block, it passes through a 3x3 convolution layer with a dilation rate of 1 to obtain feature 1. Then, feature 1 is combined with feature F n Combine the combined features as a whole, perform a 3x3 convolution with an expansion rate of 2, and output feature 2. Then, feature 2, feature 1, and feature F n Combine the combined features as a whole, perform a 3x3 convolution with an expansion rate of 4, and output feature 3. Then, feature 3, feature 2, feature 1 and feature F n Combine the combined features as a whole, perform a 3x3 convolution with an expansion rate of 4, and output feature 4. Then, feature 4, feature 3, feature 2, feature 1 and feature F n Combine the combined features as a whole and perform a 3x3 convolution (with an expansion rate of 1) to achieve local feature fusion and output feature 5. Finally, the fused feature 5 of the local features is input into the channel attention network to calculate the channel attention. The calculation of the channel attention includes the following process:

[0070] (1) Through global average pooling, the data of each channel is directly converted into one value, realizing data compression in the spatial dimension.

[0071] (2) Assign weights to each feature channel. The number of output weight parameters is the same as the number of channels in the input feature map. Two fully connected layers are used to assign weights. The first fully connected layer is followed by a ReLU function, and the second fully connected layer is followed by a Sigmoid function, thereby effectively controlling the data scale.

[0072] (3) Assign the normalized weight of each channel to each channel of the feature map.

[0073] After completing the calculation of channel attention, the features after channel attention are residually connected with the input features of the dense block to obtain the output features of a single dense block.

[0074] By implementing the above steps 201 to 204, the present application first uses the features of the adjacent slices to enrich the features of the middle slice, then uses a fully connected network to reduce the feature dimension, and then uses the DRD-Net network to improve the receptive field and enhance the learning of global features; finally, the mapping of tissue parameters is achieved through two 3×3 convolutions, thereby improving the tissue parameter reconstruction quality of the magnetic resonance fingerprint.

[0075] In an exemplary embodiment, the multi-slice joint magnetic resonance fingerprint tissue parameter mapping result is as follows: Figure 6 shown. Figure 6 It shows the T1 truth map of the synthetic data of five people, the T2 truth map of the synthetic data of five people, the tissue parameter mapping results of the traditional Drone method, and the parameter mapping results of the tissue parameter mapping method based on multi-slice joint magnetic resonance fingerprint of this application. Figure 6 The T1 and T2 ground truth images in the figure are obtained by the traditional dictionary matching method. Dictionary matching is performed under the condition of longer sequence length and takes a long time to complete. The result obtained by dictionary matching will be used as the gold standard image.

[0076] from Figure 6 It can be seen from the 4-fold residual map and the 5-fold residual map that, whether it is the T1 true value map or the T2 true value map, the error of the tissue parameter mapping result of the present application is significantly lower than the error of the tissue parameter mapping result of the traditional Drone method, which indicates that the method provided by the present application significantly improves the tissue parameter reconstruction quality of the magnetic resonance fingerprint.

[0077] The present application also provides an application scenario. Specifically, the method for tissue parameter mapping based on multi-slice combined magnetic resonance fingerprint provided in this embodiment can be applied to the display of magnetic resonance fingerprint data tissue parameter mapping results in the medical field, thereby assisting doctors in conducting research on diseases. For example, research on neurological diseases, cardiovascular diseases, musculoskeletal diseases, and abdominal and pelvic organ diseases. The method for tissue parameter mapping based on multi-slice combined magnetic resonance fingerprint provided in this application can still have high parameter reconstruction quality even when the amount of magnetic resonance fingerprint data is small. Due to the small amount of data required, the time required to collect magnetic resonance fingerprint data is also significantly reduced, thereby improving the efficiency of disease diagnosis and detection.

[0078] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store magnetic resonance fingerprint data of different slices. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for mapping tissue parameters based on multi-slice joint magnetic resonance fingerprint is implemented.

[0079] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0080] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0081] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0082] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0083] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0084] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0085] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0086] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A tissue parameter mapping method based on multi-slice joint magnetic resonance fingerprint, characterized in that: The method for tissue parameter mapping based on multi-slice combined magnetic resonance fingerprint includes: Acquiring magnetic resonance fingerprint data of different slices; the magnetic resonance fingerprint data is time series data; Preprocessing the magnetic resonance fingerprint data of the different slices to obtain noise-free magnetic resonance fingerprint data of the different slices; Determining the noise-free magnetic resonance fingerprint data of three adjacent slices in the noise-free magnetic resonance fingerprint data of different slices as a group, and inputting the noise-free magnetic resonance fingerprint data of the different slices into a multi-slice feature fusion network in groups, and outputting a fused magnetic resonance fingerprint feature map; the multi-slice feature fusion network is used to perform cross and attention operations on the noise-free magnetic resonance fingerprint data of the three adjacent slices, to obtain a magnetic resonance fingerprint feature map after feature fusion of the noise-free magnetic resonance fingerprint data of the three adjacent slices; Inputting the fused magnetic resonance fingerprint feature map into a parameter reconstruction network, and outputting a tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice; The multi-slice feature fusion network includes three attention branches, namely the upper attention branch, the middle attention branch and the lower attention branch; each attention branch is composed of an even number of Temporal SwinTransformer blocks; The Temporal SwinTransformer block includes a block embedding unit, a channel merging unit, a first linear mapping unit, a multi-head self-attention unit, a second linear mapping unit, a first residual connection unit, a layer normalization unit, a multi-layer perceptron, a second residual connection unit, a block de-embedding unit and a 3×3 convolution unit; The multi-head self-attention unit is a window multi-head self-attention unit or a sliding window multi-head self-attention unit; in the same attention branch, the multi-head self-attention units in adjacent Temporal Swin Transformer blocks are different; The first end of the block embedding unit is the input end of the Temporal SwinTransformer block; the second end of the block embedding unit is connected to the first end of the channel merging unit; the second end of the channel merging unit is connected to one end of the first linear mapping unit; the other end of the first linear mapping unit is connected to one end of the first residual connection unit; the other end of the first residual connection unit is connected to one end of the layer normalization unit; the other end of the layer normalization unit is connected to one end of the multi-layer perceptron; the other end of the multi-layer perceptron is connected to one end of the second residual connection unit; the other end of the second residual connection unit is connected to one end of the block de-embedding unit; the other end of the block de-embedding unit is connected to one end of the 3×3 convolution unit; the other end of the 3×3 convolution unit is the output end of the Temporal SwinTransformer block; The block embedding unit of the upper attention branch, the block embedding unit of the lower attention branch and the block embedding unit of the intermediate attention branch perform channel merging in the channel merging unit of the intermediate attention branch; the block embedding unit of the intermediate attention branch and the block embedding unit of the upper attention branch perform channel merging in the channel merging unit of the upper attention branch; the block embedding unit of the intermediate attention branch and the block embedding unit of the lower attention branch perform channel merging in the channel merging unit of the lower attention branch.

2. The method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint according to claim 1, characterized in that: Preprocessing the magnetic resonance fingerprint data of the different slices to obtain noise-free magnetic resonance fingerprint data of the different slices specifically includes: selecting some segments in the magnetic resonance fingerprint data of different slices, and determining the some segments in the magnetic resonance fingerprint data of different slices as data to be processed; The foreground area of ​​the data to be processed is extracted, and the target area of ​​interest is separated from the background area to obtain noise-free magnetic resonance fingerprint data.

3. The method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint according to claim 2, characterized in that: Extracting the foreground area of ​​the data to be processed, separating the target area of ​​interest from the background area, and obtaining noise-free magnetic resonance fingerprint data, specifically includes: filtering and binarizing the proton density parameter map of the data to be processed according to a preset global threshold to obtain a first proton density parameter map; extracting the maximum connected domain of the first proton density parameter map based on a connected domain extraction algorithm, removing noise from the first proton density parameter map, and obtaining a second proton density parameter map; Filling holes in the second proton density parameter map to repair missing parts in the second proton density parameter map to obtain a third proton density parameter map; the third proton density parameter map is used as a reference map for extracting a foreground area; A matrix multiplication operation is performed on each frame image in the data to be processed and the reference image for extracting the foreground area to obtain noise-free magnetic resonance fingerprint data.

4. The method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint according to claim 1, characterized in that: The parameter reconstruction network is a DRD-Net network; the DRD-Net network includes a fully connected layer, n dense blocks, a global feature fusion layer, a global residual connection layer and an upsampling convolution layer; One end of the fully connected layer is connected to the output end of the multi-slice feature fusion network; the other end of the fully connected layer is connected to one end of the n dense blocks; the other end of the n dense blocks is connected to one end of the global feature fusion layer; the other end of the global feature fusion layer is connected to one end of the global residual connection layer; the other end of the global residual connection layer is connected to one end of the upsampling convolution layer; the other end of the upsampling convolution layer is used to output the tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the intermediate slice.

5. The method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint according to claim 4, characterized in that: Inputting the fused magnetic resonance fingerprint feature map into a parameter reconstruction network and outputting a tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice specifically includes: Inputting the fused magnetic resonance fingerprint feature map into the fully connected layer of the parameter reconstruction network to output low-dimensional features; Input the low-dimensional features into the n dense blocks for processing, and output n sub-features; Input n sub-features into the global feature fusion layer for global feature fusion and output the global feature; Inputting the low-dimensional features and the global features into a global residual connection layer, and outputting the features after global residual connection; The features after global residual connection are input into the upsampling convolution layer, and the tissue parameter mapping result of the noise-free magnetic resonance fingerprint data of the middle slice is output.

6. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multi-slice joint magnetic resonance fingerprint tissue parameter mapping method according to any one of claims 1 to 5.

7. A 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 tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint according to any one of claims 1 to 5 is implemented.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for tissue parameter mapping based on multi-slice joint magnetic resonance fingerprint according to any one of claims 1 to 5 is implemented.

Images