Nuclear magnetic resonance image sequence processing method and device, equipment and medium
By constructing and training inter-layer resolution reconstruction neural networks, high-layer resolution reconstruction is achieved while reducing the NMR scanning time, improving the user experience.
Patent Information
- Application Number
- CN202410206749.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-26
- Publication Date
- 2025-08-26
AI Technical Summary
The existing MRI scanning technology takes a long time, and patients stay still for a long time may lead to discomfort and hearing damage, and it is difficult to achieve high-level resolution reconstruction while reducing scanning time.
The initial inter-layer resolution reconstruction neural network is constructed, and the training data is used for training is used to obtain the target inter-layer resolution reconstruction neural network. Through this network, the nuclear magnetic resonance image sequence with low inter-layer resolution is processed, and the high inter-layer resolution image sequence is automatically output.
It reduces the MRI scan time, improves the user experience, and avoids the discomfort and noise effects caused by patients staying still for a long time.
Smart Images
Figure CN120543376A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing technology, and in particular to a method, device, equipment and medium for processing nuclear magnetic resonance image sequences. Background Art
[0002] Magnetic resonance imaging (MRI) is a safe and noninvasive medical imaging technique widely used in modern hospitals and clinics as a vital diagnostic tool. Given the diversity and complexity of lesions, especially in functionally critical organs, low inter-slice resolution makes lesion visualization challenging. This has led to the development of MRI inter-slice resolution reconstruction, which produces high-quality images (high inter-slice resolution sequences) in the sagittal, coronal, and axial planes. This provides medical experts with rich and accurate information, helping to prevent misdiagnosis and missed diagnoses.
[0003] To obtain high-resolution MRI sequences between layers, current technologies are divided into two types: 3D MRI scanning and multi-layer 2D MRI scanning. In 3D MRI scanning, a hard pulse (with a relatively large bandwidth) is first used to excite the entire imaging range, and then spatial encoding is performed using gradients in three directions. The essence of 3D MRI scanning is to first excite all imaging areas with radio frequency pulses, and then perform three-dimensional separation between and within layers. In multi-layer 2D MRI scanning, the human body is first "cut" into multiple sections using layer selection gradients, first stimulating a specific layer of interest. Spatial positioning is then performed using gradients in two other directions (frequency encoding gradient and phase encoding gradient).
[0004] However, this type of scanning process takes a long time, and the patient has to remain still for a long time, which may cause some discomfort. In addition, due to the noise generated during the MRI operation, the hearing of some patients may also be damaged.
[0005] In summary, how to reduce the MRI scanning time while completing inter-layer resolution reconstruction and improving the user experience is a problem to be solved in this field. Summary of the Invention
[0006] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for processing MRI image sequences, which can reduce MRI scan time while completing inter-slice resolution reconstruction and improving user experience. The specific solution is as follows:
[0007] In a first aspect, the present application discloses a method for processing a nuclear magnetic resonance image sequence, comprising:
[0008] Constructing an initial inter-layer resolution reconstruction neural network;
[0009] Training the initial inter-layer resolution reconstruction neural network using the training data to obtain a target inter-layer resolution reconstruction neural network;
[0010] Outputting the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category;
[0011] The inter-layer resolution of the current nuclear magnetic resonance image sequence is reconstructed through the target inter-layer resolution reconstruction neural network, and a processed nuclear magnetic resonance image sequence is output; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high-layer inter-resolution category.
[0012] Optionally, the using the training data to train the initial inter-layer resolution reconstruction neural network to obtain a target inter-layer resolution reconstruction neural network includes:
[0013] Acquiring an initial nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the initial nuclear magnetic resonance image sequence is a preset high-layer inter-layer resolution category;
[0014] Acquire first target training data based on the initial nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the first target training data is a preset low inter-layer resolution category;
[0015] performing a sharpening process on the initial nuclear magnetic resonance image sequence to obtain second target training data;
[0016] The initial inter-layer resolution reconstruction neural network is trained using the first target training data and the second target training data to obtain a target inter-layer resolution reconstruction neural network.
[0017] Optionally, acquiring an initial nuclear magnetic resonance image sequence includes:
[0018] Performing 2D MRI scans on different planes on each subject to obtain a first initial MRI image sequence;
[0019] Accordingly, the acquiring first target training data based on the initial nuclear magnetic resonance image sequence includes:
[0020] performing normalization processing on the first initial nuclear magnetic resonance image sequence to obtain a first normalized image sequence;
[0021] The first normalized image sequence is processed using a downsampling algorithm to obtain first target training data; wherein the downsampling algorithm is any one or more of a pixel degradation algorithm, a blur processing algorithm, an image resolution reduction algorithm, a noise addition algorithm, a ringing effect addition algorithm, and an image compression algorithm.
[0022] Optionally, acquiring an initial nuclear magnetic resonance image sequence includes:
[0023] performing a multi-slice 2D MRI scan or a 3D MRI scan on each subject to obtain a second initial MRI image sequence;
[0024] Accordingly, the acquiring first target training data based on the initial nuclear magnetic resonance image sequence includes:
[0025] performing normalization processing on the second initial nuclear magnetic resonance image sequence to obtain a second normalized image sequence;
[0026] The second normalized image sequence is processed using a downsampling algorithm to obtain first target training data; wherein the downsampling algorithm is any one or more of a voxel degradation algorithm, a voxel blurring algorithm, a voxel noise addition algorithm, a mosaic algorithm, a sequence resampling algorithm, and a voxel resolution reduction algorithm.
[0027] Optionally, constructing an initial inter-layer resolution reconstruction neural network includes:
[0028] The backbone generation network and the bypass generation network are used to construct the initial inter-layer resolution reconstruction neural network.
[0029] Optionally, the using the training data to train the initial inter-layer resolution reconstruction neural network to obtain a target inter-layer resolution reconstruction neural network includes:
[0030] Determining the initial inter-layer resolution reconstruction neural network as the current inter-layer resolution reconstruction neural network;
[0031] Inputting the first target training data into the current inter-layer resolution reconstruction neural network, so that the current inter-layer resolution reconstruction neural network uses the backbone generation network to extract features from the training data to obtain feature images, and uses the bypass generation network to process the feature images to obtain a reconstructed image sequence;
[0032] The parameters of the current inter-layer resolution reconstruction neural network are updated based on the reconstructed image sequence to obtain the next inter-layer resolution reconstruction neural network.
[0033] Optionally, updating the parameters of the current inter-layer resolution reconstruction neural network based on the reconstructed image sequence to obtain a next inter-layer resolution reconstruction neural network includes:
[0034] Calculating a loss value between the reconstructed image sequence and the second target training data using a target loss function; wherein the target loss function is any one or more of a pixel feature loss function, an image feature loss function, an adversarial generation loss function, and a K-space feature loss function;
[0035] The parameters of the current inter-layer resolution reconstruction neural network are updated based on the loss value to obtain the next inter-layer resolution reconstruction neural network.
[0036] In a second aspect, the present application discloses a nuclear magnetic resonance image sequence processing device, comprising:
[0037] Network building module, used to build the initial inter-layer resolution reconstruction neural network;
[0038] A network training module is used to train the initial inter-layer resolution reconstruction neural network using training data to obtain a target inter-layer resolution reconstruction neural network;
[0039] A sequence input module is used to output the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category;
[0040] A sequence reconstruction module is used to reconstruct the inter-layer resolution of the current nuclear magnetic resonance image sequence through the target inter-layer resolution reconstruction neural network, and output a processed nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high-layer inter-resolution category.
[0041] In a third aspect, the present application discloses an electronic device, comprising:
[0042] Memory, used to store computer programs;
[0043] The processor is used to execute the computer program to implement the steps of the aforementioned method for processing nuclear magnetic resonance images.
[0044] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the aforementioned disclosed method for processing nuclear magnetic resonance image sequences are implemented.
[0045] The beneficial effects of the present application are as follows: the present application constructs an initial inter-layer resolution reconstruction neural network; uses training data to train the initial inter-layer resolution reconstruction neural network to obtain a target inter-layer resolution reconstruction neural network; outputs the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category; reconstructs the inter-layer resolution of the current nuclear magnetic resonance image sequence through the target inter-layer resolution reconstruction neural network, and outputs a processed nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high inter-layer resolution category. It can be seen that the present application uses training data to train the initial inter-layer resolution reconstruction neural network to obtain the target inter-layer resolution reconstruction neural network. In this way, the current MRI image sequence of the preset low inter-layer resolution category is input into the target inter-layer resolution reconstruction neural network, and the target inter-layer resolution reconstruction neural network can automatically output the processed MRI image sequence of the preset high inter-layer resolution category. That is to say, the present application can use the target inter-layer resolution reconstruction neural network to process the current MRI image sequence of the preset low inter-layer resolution category with a shorter MRI scanning time to obtain the processed MRI image sequence of the preset high inter-layer resolution category, thereby reducing the MRI scanning time and the discomfort brought to the patient during the scan, and improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] 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 or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.
[0047] Figure 1 This is a flow chart of a method for processing nuclear magnetic resonance image sequences disclosed in this application;
[0048] Figure 2 A specific NMR sequence comparison diagram disclosed in this application;
[0049] Figure 3 This is a schematic structural diagram of a nuclear magnetic resonance image sequence processing device disclosed in this application;
[0050] Figure 4 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION
[0051] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0052] To obtain high-resolution MRI sequences between layers, current technologies are divided into two types: 3D MRI scanning and multi-layer 2D MRI scanning. In 3D MRI scanning, a hard pulse (with a relatively large bandwidth) is first used to excite the entire imaging range, and then spatial encoding is performed using gradients in three directions. The essence of 3D MRI scanning is to first excite all imaging areas with radio frequency pulses, and then perform three-dimensional separation between and within layers. In multi-layer 2D MRI scanning, the human body is first "cut" into multiple sections using layer selection gradients, first stimulating a specific layer of interest. Spatial positioning is then performed using gradients in two other directions (frequency encoding gradient and phase encoding gradient).
[0053] However, this type of scanning process takes a long time, and the patient has to remain still for a long time, which may cause some discomfort. In addition, due to the noise generated during the MRI operation, the hearing of some patients may also be damaged.
[0054] To this end, the present application provides a corresponding MRI image sequence processing solution, which reduces the MRI scanning time while completing inter-layer resolution reconstruction and improving user experience.
[0055] See also Figure 1 As shown, the embodiment of the present application discloses a method for processing a nuclear magnetic resonance image sequence, comprising:
[0056] Step S11: Construct an initial inter-layer resolution reconstruction neural network.
[0057] In this embodiment, constructing an initial inter-layer resolution reconstruction neural network includes constructing an initial inter-layer resolution reconstruction neural network using a backbone generation network and a bypass generation network. The MRI image inter-layer resolution reconstruction neural network, i.e., the initial inter-layer resolution reconstruction neural network, is constructed. The initial inter-layer resolution reconstruction neural network is composed of a backbone generation network and one or more bypass generation networks.
[0058] The backbone generative network can input 4D or 5D image data and can be composed of single or multi-channel inputs, multiple 2D convolutional layers, nonlinear activation function layers, residual connections, and 3D convolutions. It can also include spatial / channel attention modules. Residual connections utilize 1x3x3 and 3x1x1 convolutions instead of conventional 3x3x3 convolutions, reducing computational complexity. Multiple residual blocks are used to construct a densely connected network. In other words, a densely connected network (DenseNet) can be simply understood as a more complex form of residual connections, effectively avoiding the vanishing gradient problem.
[0059] The backbone generative network includes self-attention, masked self-attention, and position encoding. The bypass generative network consists of multiple convolutional layers, nonlinear activation layers, and residual networks, and may also include spatial / channel attention modules.
[0060] It is understandable that it is also necessary to construct a deep learning network loss, that is, a target loss function. The deep learning network loss is composed of one or more pixel feature losses, image feature losses, adversarial generation losses, and k-space feature losses. Pixel feature losses include L1loss (L1 Norm Loss), L2loss, and SmoothL1loss. Image feature losses include SSIM (Structural Similarity Index) loss and PSNR (Peak Signal-to-Noise Ratio) loss. The loss of pre-trained network feature maps includes but is not limited to those based on CNN (Convolutional Neural Network), Transformer, and Diffusion. The adversarial generation loss is composed of a trainable discriminator network and GAN (Generative Adversarial Networks) loss. The discriminant network can be composed of different VGG (Visual Geometry Group) networks, FCN networks (Fully Convolutional Networks), U-net networks based on spectral normalization and batch normalization, and VIT (Vision Inference Network) based on Transformer. Transformer) segmentation network, GAN loss can be basic GAN loss, WGAN (Wasserstein GAN) loss or RGAN (Relative GAN) loss.
[0061] Step S12: using the training data to train the initial inter-layer resolution reconstruction neural network to obtain the target inter-layer resolution reconstruction neural network.
[0062] It should be noted that the resolution can be divided into inter-layer resolution and intra-layer resolution, and the target inter-layer resolution reconstruction neural network obtained after training in this embodiment only reconstructs the inter-layer resolution.
[0063] In this embodiment, the method of training the initial inter-layer resolution reconstruction neural network using training data to obtain a target inter-layer resolution reconstruction neural network includes: acquiring an initial nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the initial nuclear magnetic resonance image sequence is a preset high-layer resolution category; acquiring first target training data based on the initial nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the first target training data is a preset low-layer resolution category; sharpening the initial nuclear magnetic resonance image sequence to obtain second target training data; and training the initial inter-layer resolution reconstruction neural network using the first target training data and the second target training data to obtain a target inter-layer resolution reconstruction neural network. Acquiring an initial nuclear magnetic resonance image sequence with high-layer resolution, processing the initial nuclear magnetic resonance image sequence with high-layer resolution to obtain first target training data with low-layer resolution, then sharpening the initial nuclear magnetic resonance image sequence with high-layer resolution to obtain second target training data, and then training the initial inter-layer resolution reconstruction neural network using the first target training data and the second target training data to obtain a target inter-layer resolution reconstruction neural network. Among them, image sequences of inter-layer resolution categories can be obtained based on the specific MRI scanning scenario. For example, sequences with a layer thickness of less than 1mm can be considered as high-layer resolution sequences, while layers with a thickness of more than 1mm can be considered as low-layer resolution sequences. The initial MRI image sequence is sharpened to obtain the second target training data. The sharpening process includes differentiation, high-pass filtering, and differentiation methods including gradient method, Sobel operator (Sobel operator) method, and Laplace operator method.
[0064] In a specific embodiment, the acquisition of the initial nuclear magnetic resonance image sequence includes: performing 2D nuclear magnetic resonance scans on each object in different planes to obtain a first initial nuclear magnetic resonance image sequence. That is, three ordinary 2D MRI scans are performed on each object in different planes (i.e., sagittal, coronal and axial planes) to obtain three high-resolution MRI image sequences of different planes. Correspondingly, the acquisition of the first target training data based on the initial nuclear magnetic resonance image sequence includes: normalizing the first initial nuclear magnetic resonance image sequence to obtain a first normalized image sequence; processing the first normalized image sequence using a downsampling algorithm to obtain the first target training data; wherein the downsampling algorithm is any one or more of a pixel degradation algorithm, a blurring algorithm, an image resolution reduction algorithm, a noise addition algorithm, a ringing effect addition algorithm and an image compression algorithm. The first initial nuclear magnetic resonance image sequence is normalized, that is, normalized by analyzing the statistical histogram of the pixel values of the image. The specific formula is as follows:
[0065] Norm_val=(Real_val-Min) / (Max-Min);
[0066] Where Min is the minimum MRI value (set to 0 here, any value less than 0 is set to 0), Max is the maximum MRI value, and Real_val represents the true value of the pixel;
[0067] The first normalized image sequence is processed using a downsampling algorithm to obtain first target training data; wherein the downsampling algorithm is any one or more of a pixel degradation algorithm, a blur processing algorithm, an image resolution reduction algorithm, a noise addition algorithm, a ringing effect addition algorithm, and an image compression algorithm; the blur processing algorithm includes Gaussian blur (isotropic, anisotropic), motion blur, disk blur, image shadow, tilt-shift blur, path blur, scene blur, rotation blur, and real blur simulation based on deep learning; the image resolution reduction algorithm includes unilinear interpolation, bilinear interpolation, and bicubic linear interpolation; the noise addition algorithm includes Gaussian noise, Rice noise, salt and pepper noise, Poisson noise, and a noise generation network obtained from a real two-dimensional data image based on deep learning technology.
[0068] In another specific embodiment, the acquisition of the initial nuclear magnetic resonance image sequence includes: performing multi-layer 2D nuclear magnetic resonance scans or 3D nuclear magnetic resonance scans on each object to obtain a second initial nuclear magnetic resonance image sequence. The acquisition of multi-layer 2D MRI scans or 3D MRI scans of the object to obtain a second initial nuclear magnetic resonance image sequence. Correspondingly, the acquisition of the first target training data based on the initial nuclear magnetic resonance image sequence includes: normalizing the second initial nuclear magnetic resonance image sequence to obtain a second normalized image sequence; processing the second normalized image sequence using a downsampling algorithm to obtain the first target training data; wherein the downsampling algorithm is any one or more of a voxel degradation algorithm, a voxel blurring algorithm, a voxel noise addition algorithm, a mosaic algorithm, a sequence resampling algorithm, and a voxel resolution reduction algorithm. The second initial nuclear magnetic resonance image sequence is normalized, that is, normalized by analyzing the statistical histogram of the pixel values of the image. The specific formula is as follows:
[0069] Norm_val=(Real_val-Min) / (Max-Min);
[0070] Where Min is the minimum MRI value (set to 0 here, any value less than 0 is set to 0), Max is the maximum MRI value, and Real_val represents the true value of the pixel;
[0071] The second normalized image sequence is processed using a downsampling algorithm to obtain first target training data; wherein the downsampling algorithm is any one or more of a voxel degradation algorithm, a voxel blur processing algorithm, a voxel noise addition algorithm, a mosaic algorithm, a sequence resampling algorithm, and a voxel resolution reduction algorithm; the voxel blur processing algorithm includes Gaussian blur (isotropic, anisotropic), motion blur, disk blur, image shadow, tilt-shift blur, path blur, scene blur, rotation blur, and real blur simulation based on deep learning; the voxel resolution reduction algorithm includes unilinear interpolation, bilinear interpolation, and bicubic linear interpolation; the voxel noise addition processing includes Gaussian noise, Rice noise, salt and pepper noise, Poisson noise, and a noise generation network obtained from real three-dimensional data images based on deep learning technology.
[0072] In this embodiment, the initial inter-layer resolution reconstruction neural network is trained using training data to obtain a target inter-layer resolution reconstruction neural network, including: determining the initial inter-layer resolution reconstruction neural network as the current inter-layer resolution reconstruction neural network; inputting the first target training data into the current inter-layer resolution reconstruction neural network, so that the current inter-layer resolution reconstruction neural network uses the backbone generation network to extract features from the training data to obtain a feature image, and uses the bypass generation network to process the feature image to obtain a reconstructed image sequence; based on the reconstructed image sequence, the parameters of the current inter-layer resolution reconstruction neural network are updated to obtain the next inter-layer resolution reconstruction neural network. The backbone generation network extracts features from the input image sequence, and the input of the bypass generation network is the feature image output by the backbone generation network. The feature image includes a wavelet transform image, a K-space projection image, etc., and may also include a mask feature image, which is a mask feature image of key tissues and lesions.
[0073] In this embodiment, the updating of the parameters of the current inter-layer resolution reconstruction neural network based on the reconstructed image sequence to obtain the next inter-layer resolution reconstruction neural network includes: using a target loss function to calculate the loss value between the reconstructed image sequence and the second target training data; wherein the target loss function is any one or more of a pixel feature loss function, an image feature loss function, an adversarial generation loss function, and a K-level feature loss function; based on the loss value, updating the parameters of the current inter-layer resolution reconstruction neural network to obtain the next inter-layer resolution reconstruction neural network. The first target training data of low inter-layer resolution is used as the input of the network, and the reconstructed image sequence of high inter-layer resolution is used as the output of the network. The loss value between the reconstructed image sequence and the second target training data is calculated, and the parameters of the current inter-layer resolution reconstruction neural network are updated based on the loss value, for example, by iteratively updating the network weights through stochastic gradient descent or Adam (Adaptive Moment Estimation) optimizer to obtain the next inter-layer resolution reconstruction neural network.
[0074] Step S13: outputting the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category.
[0075] In practical applications, if a current MRI image sequence with low inter-layer resolution is obtained, the current MRI image sequence is input into the target inter-layer resolution reconstruction neural network. The scanning time required to obtain the current MRI image sequence with low inter-layer resolution is shorter, so the patient does not need to remain still for a long time and avoids excessive exposure to noise generated by MRI operation.
[0076] Step S14: reconstructing the inter-layer resolution of the current MRI image sequence through the target inter-layer resolution reconstruction neural network, and outputting a processed MRI image sequence; wherein the inter-layer resolution category of the processed MRI image sequence is a preset high-layer inter-resolution category.
[0077] The target inter-layer resolution reconstruction neural network receives the current nuclear magnetic resonance image sequence, reconstructs the inter-layer resolution of the current nuclear magnetic resonance image sequence, and can automatically and quickly output the processed nuclear magnetic resonance image sequence with high inter-layer resolution.
[0078] The beneficial effects of the present application are as follows: the present application constructs an initial inter-layer resolution reconstruction neural network; uses training data to train the initial inter-layer resolution reconstruction neural network to obtain a target inter-layer resolution reconstruction neural network; outputs the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category; reconstructs the inter-layer resolution of the current nuclear magnetic resonance image sequence through the target inter-layer resolution reconstruction neural network, and outputs a processed nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high inter-layer resolution category. It can be seen that the present application uses training data to train the initial inter-layer resolution reconstruction neural network to obtain the target inter-layer resolution reconstruction neural network. In this way, the current MRI image sequence of the preset low inter-layer resolution category is input into the target inter-layer resolution reconstruction neural network, and the target inter-layer resolution reconstruction neural network can automatically output the processed MRI image sequence of the preset high inter-layer resolution category. That is to say, the present application can use the target inter-layer resolution reconstruction neural network to process the current MRI image sequence of the preset low inter-layer resolution category with a shorter MRI scanning time to obtain the processed MRI image sequence of the preset high inter-layer resolution category, thereby reducing the MRI scanning time and the discomfort brought to the patient during the scan, and improving the user experience.
[0079] Below is Figure 2 Taking a specific nuclear magnetic resonance sequence comparison diagram shown as an example, the present application is described accordingly:
[0080] 1) Training data acquisition: A mixture of real 7T data and high-quality 3T data (multi-layer 2D scans) is used as high-resolution data between layers to obtain the second target training data, and simulated low-resolution data between layers as the first target training data;
[0081] 2) Constructing an initial inter-layer resolution reconstruction neural network, wherein the initial inter-layer resolution reconstruction neural network is a 6-layer 3D_P3D_Block network;
[0082] 3) Constructing the target loss function: The deep learning network losses are L1 loss, focal frequency loss (FFL), VGGNet-19 layer feature map loss, basic GAN loss, and discriminant loss based on spectral normalization U-net;
[0083] 4) Use the training set data to train the initial inter-layer resolution reconstruction neural network: Take the low-resolution matrix and high-resolution matrix in the training set data as input, output comparison samples, and iteratively update the network weights through the Adam optimizer with a learning rate of 0.001 for 200 cycles. In this way, the target inter-layer resolution reconstruction neural network is obtained;
[0084] 5) Output the current MRI image sequence to the target inter-layer resolution reconstruction neural network, wherein the current MRI image sequence is, for example, Figure 2 (a), (b), and (c) in the text;
[0085] 6) Target inter-layer resolution reconstruction The neural network performs inter-layer resolution reconstruction and obtains, for example, Figure 2 The processed MRI image sequences shown in (d), (e), and (f).
[0086] See also Figure 3 As shown, the embodiment of the present application discloses a nuclear magnetic resonance image sequence processing device, comprising:
[0087] A network construction module 11 is used to construct an initial inter-layer resolution reconstruction neural network;
[0088] A network training module 12 is used to train the initial inter-layer resolution reconstruction neural network using training data to obtain a target inter-layer resolution reconstruction neural network;
[0089] A sequence input module 13 is configured to output the current MRI image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current MRI image sequence is a preset low inter-layer resolution category;
[0090] The sequence reconstruction module 14 is used to reconstruct the inter-layer resolution of the current nuclear magnetic resonance image sequence through the target inter-layer resolution reconstruction neural network, and output a processed nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high-layer inter-resolution category.
[0091] The beneficial effects of the present application are as follows: the present application constructs an initial inter-layer resolution reconstruction neural network; uses training data to train the initial inter-layer resolution reconstruction neural network to obtain a target inter-layer resolution reconstruction neural network; outputs the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category; reconstructs the inter-layer resolution of the current nuclear magnetic resonance image sequence through the target inter-layer resolution reconstruction neural network, and outputs a processed nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high inter-layer resolution category. It can be seen that the present application uses training data to train the initial inter-layer resolution reconstruction neural network to obtain the target inter-layer resolution reconstruction neural network. In this way, the current MRI image sequence of the preset low inter-layer resolution category is input into the target inter-layer resolution reconstruction neural network, and the target inter-layer resolution reconstruction neural network can automatically output the processed MRI image sequence of the preset high inter-layer resolution category. That is to say, the present application can use the target inter-layer resolution reconstruction neural network to process the current MRI image sequence of the preset low inter-layer resolution category with a shorter MRI scanning time to obtain the processed MRI image sequence of the preset high inter-layer resolution category, thereby reducing the MRI scanning time and the discomfort brought to the patient during the scan, and improving the user experience.
[0092] Furthermore, an embodiment of the present application also provides an electronic device. Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram should not be considered as any limitation to the scope of application of the present application.
[0093] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Specifically, the device may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the method for processing a nuclear magnetic resonance image sequence performed by an electronic device as disclosed in any of the aforementioned embodiments.
[0094] In this embodiment, the power supply 23 is used to provide operating voltage for various hardware devices on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and external devices. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.
[0095] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.
[0096] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon include an operating system 221, a computer program 222 and data 223, etc. The storage method can be temporary storage or permanent storage.
[0097] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device, enabling the processor 21 to calculate and process the massive amount of data 223 in the memory 22. It can be Windows, Unix, Linux, etc. In addition to including computer programs capable of performing the magnetic resonance image sequence processing method disclosed in any of the aforementioned embodiments by the electronic device, the computer program 222 may further include computer programs capable of performing other specific tasks. Data 223 may include data received by the electronic device from external devices as well as data collected by its own input / output interface 25.
[0098] Furthermore, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when executed by a processor, the computer program implements the aforementioned method for processing nuclear magnetic resonance image sequences. The specific steps of this method can be referred to the corresponding contents disclosed in the aforementioned embodiments and will not be repeated here.
[0099] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.
[0100] Professionals may further appreciate that the units and algorithmic steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application. The steps of the method or algorithm described in conjunction with the embodiments disclosed herein can be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in random access memory (RAM), memory, read-only memory (ROM), electrically programmable EPROM (Erasable Programmable Read Only Memory), electrically erasable programmable EEPROM (Electrically Erasable Programmable read only memory), registers, hard disk, removable disk, CD-ROM (CoMP24000881act Disc Read-Only Memory), or any other form of storage medium known in the technical field.
[0101] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.
[0102] The above is a detailed introduction to the method, device, equipment and medium for processing nuclear magnetic resonance image sequences provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only intended to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.
Claims
1. A method for processing a nuclear magnetic resonance image sequence, characterized in that: include: Constructing an initial inter-layer resolution reconstruction neural network; Training the initial inter-layer resolution reconstruction neural network using the training data to obtain a target inter-layer resolution reconstruction neural network; Outputting the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category; The inter-layer resolution of the current nuclear magnetic resonance image sequence is reconstructed through the target inter-layer resolution reconstruction neural network, and a processed nuclear magnetic resonance image sequence is output; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high-layer inter-resolution category.
2. The method for processing nuclear magnetic resonance image sequences according to claim 1, wherein: The initial inter-layer resolution reconstruction neural network is trained using the training data to obtain a target inter-layer resolution reconstruction neural network, including: Acquiring an initial nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the initial nuclear magnetic resonance image sequence is a preset high-layer inter-layer resolution category; Acquire first target training data based on the initial nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the first target training data is a preset low inter-layer resolution category; performing a sharpening process on the initial nuclear magnetic resonance image sequence to obtain second target training data; The initial inter-layer resolution reconstruction neural network is trained using the first target training data and the second target training data to obtain a target inter-layer resolution reconstruction neural network.
3. The method for processing nuclear magnetic resonance image sequences according to claim 2, characterized in that: The acquiring of an initial nuclear magnetic resonance image sequence comprises: Performing 2D MRI scans on different planes on each subject to obtain a first initial MRI image sequence; Accordingly, the acquiring first target training data based on the initial nuclear magnetic resonance image sequence includes: performing normalization processing on the first initial nuclear magnetic resonance image sequence to obtain a first normalized image sequence; The first normalized image sequence is processed using a downsampling algorithm to obtain first target training data; wherein the downsampling algorithm is any one or more of a pixel degradation algorithm, a blur processing algorithm, an image resolution reduction algorithm, a noise addition algorithm, a ringing effect addition algorithm, and an image compression algorithm.
4. The method for processing nuclear magnetic resonance image sequences according to claim 2, wherein: The acquiring of an initial nuclear magnetic resonance image sequence comprises: performing a multi-slice 2D MRI scan or a 3D MRI scan on each subject to obtain a second initial MRI image sequence; Accordingly, the acquiring first target training data based on the initial nuclear magnetic resonance image sequence includes: performing normalization processing on the second initial nuclear magnetic resonance image sequence to obtain a second normalized image sequence; The second normalized image sequence is processed using a downsampling algorithm to obtain first target training data; wherein the downsampling algorithm is any one or more of a voxel degradation algorithm, a voxel blurring algorithm, a voxel noise addition algorithm, a mosaic algorithm, a sequence resampling algorithm, and a voxel resolution reduction algorithm.
5. The method for processing nuclear magnetic resonance image sequences according to any one of claims 2 to 4, characterized in that: The initial inter-layer resolution reconstruction neural network is constructed, comprising: The backbone generation network and the bypass generation network are used to construct the initial inter-layer resolution reconstruction neural network.
6. The method for processing nuclear magnetic resonance image sequences according to claim 5, characterized in that: The initial inter-layer resolution reconstruction neural network is trained using the training data to obtain a target inter-layer resolution reconstruction neural network, including: Determining the initial inter-layer resolution reconstruction neural network as the current inter-layer resolution reconstruction neural network; Inputting the first target training data into the current inter-layer resolution reconstruction neural network, so that the current inter-layer resolution reconstruction neural network uses the backbone generation network to extract features from the training data to obtain feature images, and uses the bypass generation network to process the feature images to obtain a reconstructed image sequence; The parameters of the current inter-layer resolution reconstruction neural network are updated based on the reconstructed image sequence to obtain the next inter-layer resolution reconstruction neural network.
7. The method for processing nuclear magnetic resonance image sequences according to claim 6, characterized in that: The updating of the parameters of the current inter-layer resolution reconstruction neural network based on the reconstructed image sequence to obtain a next inter-layer resolution reconstruction neural network includes: Calculating a loss value between the reconstructed image sequence and the second target training data using a target loss function; wherein the target loss function is any one or more of a pixel feature loss function, an image feature loss function, an adversarial generation loss function, and a K-space feature loss function; The parameters of the current inter-layer resolution reconstruction neural network are updated based on the loss value to obtain the next inter-layer resolution reconstruction neural network.
8. A nuclear magnetic resonance image sequence processing device, characterized in that: include: Network building module, used to build the initial inter-layer resolution reconstruction neural network; A network training module is used to train the initial inter-layer resolution reconstruction neural network using training data to obtain a target inter-layer resolution reconstruction neural network; A sequence input module is used to output the current nuclear magnetic resonance image sequence to the target inter-layer resolution reconstruction neural network; wherein the inter-layer resolution category of the current nuclear magnetic resonance image sequence is a preset low inter-layer resolution category; A sequence reconstruction module is used to reconstruct the inter-layer resolution of the current nuclear magnetic resonance image sequence through the target inter-layer resolution reconstruction neural network, and output a processed nuclear magnetic resonance image sequence; wherein the inter-layer resolution category of the processed nuclear magnetic resonance image sequence is a preset high-layer inter-resolution category.
9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the method for processing a nuclear magnetic resonance image sequence according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that Used to store a computer program; wherein, when the computer program is executed by a processor, the steps of the method for processing a nuclear magnetic resonance image sequence according to any one of claims 1 to 7 are implemented.