Intelligent dynamic contrast-enhanced magnetic resonance parametric imaging method and apparatus

Abstract

The present application relates to the technical field of medical magnetic resonance imaging, and discloses an intelligent dynamic contrast-enhanced magnetic resonance parametric imaging method and apparatus. In the method, a pixel-level vision transformer is integrated at the connection position of an encoder and a decoder of a generator to enhance the learning capability for non-local patterns and time series information in medical images, thereby improving the parameter estimation precision in noisy and variable environments. Gradient penalty is introduced into a loss function to optimize the performance of a discriminator, so that the discriminator can more effectively distinguish between real and generated images. A CycleGAN-like model can fully extract and use spatial features while performing time series analysis, thereby providing a more comprehensive and detailed PK parameter map, and providing powerful technical support for improving the accuracy of diagnosis of diseases such as cervical cancer.

Description

Intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method and device TECHNICAL FIELD

[0001] The present application relates to the technical field of medical magnetic resonance imaging, in particular to an intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method and device. BACKGROUND

[0002] Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is an advanced medical imaging technique that plays a crucial role in diagnosing and treatment planning for diseases such as cervical cancer. DCE-MRI captures the transport of contrast agent within the region of interest by repeatedly acquiring T1-weighted images after contrast agent injection and based on pharmacokinetic (PK) models such as the Tofts model and its extended versions (ETK). This non-invasive microvascular parameter quantification and analysis, such as the forward volume transfer constant (K trans ) and extracellular space volume (v e ), enables a deep understanding of the physiological characteristics of tissues.

[0003] In traditional DCE-MRI data analysis, solving these PK model parameters usually relies on methods such as nonlinear least squares (NLLS) and Bayesian estimation. Although these methods are effective in certain scenarios, they are sensitive to noise, computationally expensive, and require accurate estimation of the arterial input function (AIF), which often poses a challenge in practical applications. Therefore, these traditional methods may lead to inaccurate and inconsistent parameter estimation when dealing with large volumes or high-noise data.

[0004] In recent years, deep learning methods have shown great potential in handling DCE-MRI data. Various advanced deep learning architectures, such as convolutional neural networks (CNNs) and long short-term memory networks (LSTMs), have been used to automate and accurately estimate the PK parameters of DCE-MRI. CNNs are good at extracting spatial features, but cannot fully capture the dynamic information of time series data. LSTMs are suitable for processing time series data and learning long-term dependencies, but often ignore key spatial feature extraction in the process of mapping contrast agent concentration time curves and arterial input functions to PK parameters, thereby limiting the overall accuracy of parameter estimation. Therefore, deep learning methods still face challenges in dealing with complex spatiotemporal dependencies and high-dimensional spatial features.

[0005] SUMMARY

[0006] The present application provides an intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method to solve the problem that deep learning methods still face challenges in dealing with complex spatiotemporal dependencies and high-dimensional spatial features in the prior art.

[0007] Correspondingly, the application further provides an intelligent dynamic contrast-enhanced magnetic resonance parameter imaging device, an electronic device, and a computer readable storage medium, which are used to ensure the implementation and application of the method.

[0008] To solve the above technical problems, the application discloses an intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method, which comprises the following steps:

[0009] A CycleGAN model is constructed, the CycleGAN model comprises a generator and a discriminator, the generator comprises an encoder and a decoder, and a pixel-level visual transformer is arranged at the connection position of the encoder and the decoder;

[0010] The dynamic contrast-enhanced magnetic resonance image sequence is input into the encoder to obtain an encoded image;

[0011] The pixel-level visual transformer is used to construct the encoded image into an image sequence, and a feature sequence is extracted based on the image sequence; the image sequence comprises pixel information and position information;

[0012] The feature sequence is input into the decoder to obtain a multi-parameter image;

[0013] The CycleGAN model uses a loss function to optimize the performance of the discriminator; the loss function is constructed based on a least square GAN loss and a gradient penalty enhancement loss.

[0014] The application further discloses an intelligent dynamic contrast-enhanced magnetic resonance parameter imaging device, which comprises:

[0015] A model construction module is configured to construct a CycleGAN model; the CycleGAN model comprises a generator and a discriminator, the generator comprises an encoder and a decoder, and a pixel-level visual transformer is arranged at the connection position of the encoder and the decoder;

[0016] An encoding module is configured to input a dynamic contrast-enhanced magnetic resonance image sequence into the encoder to obtain an encoded image;

[0017] A time sequence learning module is configured to use a pixel-level visual transformer to construct the encoded image into an image sequence, and extract a feature sequence based on the image sequence; the image sequence comprises pixel information and position information;

[0018] A decoding module is configured to input the feature sequence into the decoder to obtain a multi-parameter image;

[0019] The CycleGAN model uses a loss function to optimize the performance of the discriminator; the loss function is constructed based on a least square GAN loss and a gradient penalty enhancement loss.

[0020] The application further discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method in one or more of the embodiments of the application when executing the program.

[0021] The application further discloses a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method in one or more of the embodiments of the application.

[0022] In the application, a pixel-level visual transformer is integrated at the connection position of the encoder and the decoder of the generator to enhance the learning ability of non-local patterns and time sequence information in medical images, and the parameter estimation accuracy in a variable noise environment can be improved. Gradient penalty is introduced into the loss function to optimize the performance of the discriminator, so that it can more effectively distinguish between real and generated images. The CycleGAN model can fully extract and utilize spatial features while performing time series analysis, provide more comprehensive and detailed PK parameter maps, and provide strong technical support for improving the accuracy of cervical cancer and other disease diagnosis.

[0023] Additional aspects and advantages of the application will be described in the following description part, which will become apparent from the following description or be understood by practicing the application. BRIEF DESCRIPTION OF DRAWINGS

[0024] The above and / or additional aspects and advantages of the application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

[0025] Fig. 1 is a flowchart of an intelligent dynamic contrast enhancement magnetic resonance parameter imaging method provided by an embodiment of the application;

[0026] Fig. 2 is a structure diagram of a Unet generator integrated with a pixel-level visual transformer provided by an embodiment of the application;

[0027] Fig. 3 is a structure diagram of a pixel-level visual transformer module provided by an embodiment of the application;

[0028] Fig. 4 is a structure diagram of an overall network provided by an embodiment of the application;

[0029] Fig. 5 is an experimental result diagram of cervical cancer DCE MRI parameter imaging provided by an embodiment of the application;

[0030] Fig. 6 is a multi-parameter map profile analysis result diagram provided by an embodiment of the application;

[0031] Fig. 7 is a structure diagram of an intelligent dynamic contrast enhancement magnetic resonance parameter imaging device provided by an embodiment of the application;

[0032] FIG. 8 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0033] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which examples of embodiments are shown, and in which like or similar designations denote like or similar elements or elements having the same or similar functions throughout the present description. The embodiments described below by reference to the drawings are exemplary only, and are used only for explanation of the present application, and cannot be interpreted as limiting the present application.

[0034] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "a", "an", and "the" as used herein include plural referents. It should be further understood that the word "comprise" as used in the specification of the present application means that a feature, integer, step, operation, element, and / or component is present, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or combinations thereof. It should be understood that when we say an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be an intermediate element. In addition, "connected" or "coupled" as used herein can include wireless connection or wireless coupling. The phrase "and / or" as used herein includes all or any one of the associated listed items, and all combinations thereof.

[0035] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as that understood by a person of ordinary skill in the art to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have meanings consistent with those in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as such.

[0036] The scheme provided by the embodiments of the present application can be executed by any electronic device, such as a terminal device or a server. The server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be a smartphone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server can be connected directly or indirectly through wired or wireless communication, which is not limited in the present application. For the technical problems existing in the prior art, the intelligent dynamic contrast enhancement magnetic resonance parameter imaging method and device provided by the present application aims to solve at least one of the technical problems in the prior art.

[0037] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0038] The embodiments of the present application provide a possible implementation manner, as shown in FIG. 1, a flowchart of an intelligent dynamic contrast enhancement magnetic resonance parameter imaging method is provided, which can be executed by any electronic device, and can be executed on a server side or a terminal device.

[0039] As shown in FIG. 1, the method can include the following steps:

[0040] Step 101, constructing a CycleGAN-like model; the CycleGAN-like model includes a generator and a discriminator, the generator includes an encoder and a decoder, and a pixel-level visual transformer is arranged at a connection position of the encoder and the decoder.

[0041] The embodiments of the present application propose a CycleGAN-like model, which is composed of a generator and a discriminator to form a generative adversarial network, so as to realize the transformation of a dynamic contrast enhancement magnetic resonance image sequence to a multi-parameter graph. The generator is used to learn the mapping from the dynamic contrast enhancement magnetic resonance image sequence to the multi-parameter graph, and then generate the multi-parameter graph, and the discriminator is used to determine whether the multi-parameter graph generated by the generator is a real image.

[0042] A pixel-level visual transformer is introduced in the generator to improve the ability of the model in non-local pattern recognition and time sequence information extraction. The structure of the encoder and the decoder can be a UNet structure, or a UNet structure with a convolution block attention module (CBAM), or a UNet structure with other attention mechanisms, so as to enhance the feature extraction ability of the model. The structure of the encoder and the decoder can also be other network structures, such as a deep residual network (ResNet) and a densely connected network (DenseNet), which are not specifically limited in the embodiments of the present application.

[0043] Step 102, inputting the dynamic contrast enhancement magnetic resonance image sequence into the encoder to obtain an encoded image.

[0044] In the embodiments of the present application, the features in the dynamic contrast enhancement magnetic resonance imaging (DCE-MRI) image are extracted by the encoder to obtain an encoded image.

[0045] Step 103, constructing the encoded image into an image sequence by using a pixel-level visual transformer, and extracting a feature sequence based on the image sequence; wherein the image sequence includes pixel information and position information.

[0046] The powerful timing processing capability of the pixel-level visual transformer can improve the parameter estimation accuracy of the model in a variable noise environment. The deep timing analysis function of the pixel-level visual transformer can more comprehensively capture the time series dynamics of the image sequence to achieve more accurate PK parameter estimation, and is suitable for processing dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) sequences.

[0047] In step 104, the feature sequence is input into the decoder to obtain a multi-parameter map.

[0048] The feature sequence not only includes the spatial features extracted in the encoder, but also includes the analysis of the time series in the pixel-level visual transformer. The decoder processes the feature sequence to further output a more comprehensive and detailed PK parameter map.

[0049] The CycleGAN-like model uses a loss function to optimize the performance of the discriminator; the loss function is constructed based on the least squares GAN loss and the gradient penalty enhancement loss.

[0050] The CycleGAN-like model in the embodiments of the present application introduces gradient penalty into the loss function to optimize the performance of the discriminator, so that the discriminator can more effectively distinguish between real and generated images.

[0051] In the embodiments of the present application, a pixel-level visual transformer is integrated at the connection position of the encoder and the decoder of the generator to enhance the learning ability of non-local patterns and time series information in medical images, which can improve the parameter estimation accuracy in a variable noise environment. Gradient penalty is introduced into the loss function to optimize the performance of the discriminator, so that it can more effectively distinguish between real and generated images. The CycleGAN-like model can simultaneously extract and utilize spatial features while analyzing time series, providing a more comprehensive and detailed PK parameter map, and providing strong technical support for improving the accuracy of cervical cancer and other disease diagnosis.

[0052] In an optional embodiment, the encoder and the decoder are UNet encoding-decoding structures, and the encoder includes a plurality of down-sampling encoding blocks.

[0053] The dynamic contrast-enhanced magnetic resonance image sequence is input into the encoder to obtain an encoded image, including:

[0054] The dynamic contrast-enhanced magnetic resonance image sequence is preprocessed and converted into a three-dimensional tensor.

[0055] The three-dimensional tensor is dimensionally reduced layer by layer using a plurality of down-sampling encoding blocks to obtain an encoded image.

[0056] As shown in FIG. 2, the down-sampling encoding block includes a basic convolution block and a down-sampling convolution block with a step of 2, first converts the input dynamic contrast-enhanced magnetic resonance image sequence into a three-dimensional tensor through a preprocessing layer, and then reduces the dimension layer by layer through the alternating basic convolution block and the down-sampling convolution block with a step of 2 in the encoder, until the encoded image reaches the size and dimension suitable for ViT (Vision Transformer) processing. In the embodiment of the application, the width w and the height h are halved each time the down-sampling is implemented, so as to gradually reduce the image size and increase the depth f of the feature. For each down-sampling convolution block, the design of doubling the feature dimension is to better capture complex image features and form a feature representation with sufficient information for subsequent pixel-level ViT processing.

[0057] Exemplarily, the encoder includes 4 down-sampling encoding blocks, and the size of the dynamic contrast-enhanced magnetic resonance image sequence is (256, 256, 40). The three-dimensional tensor (w0=h0=256, f0=48) after preprocessing is input into the basic convolution block of the first down-sampling encoding block to obtain a feature with a size of (256, 256, 48); after the feature is subjected to down-sampling processing by the down-sampling convolution block D4 of the first down-sampling encoding block, the basic convolution block of the second down-sampling encoding block is input to obtain a feature with a size of (128, 128, 96); after the feature is subjected to down-sampling processing by the down-sampling convolution block D3 of the second down-sampling encoding block, the basic convolution block of the third down-sampling encoding block is input to obtain a feature with a size of (64, 64, 192); after the feature is subjected to down-sampling processing by the down-sampling convolution block D2 of the third down-sampling encoding block, the basic convolution block of the fourth down-sampling encoding block is input to obtain a feature with a size of (32, 32, 384); after the feature is subjected to down-sampling processing by the down-sampling convolution block D1 of the fourth down-sampling encoding block, the size (w=h=16, f0=384) suitable for VIT processing is obtained.

[0058] In an optional embodiment, the pixel-level vision transformer includes a Transformer encoder;

[0059] The pixel-level vision transformer is used to construct the encoded image into an image sequence, and a feature sequence is extracted based on the image sequence, including:

[0060] The token sequence is constructed in units of each pixel point in the encoded image;

[0061] The position information of each pixel point in the token sequence is generated by means of two-dimensional Fourier position embedding;

[0062] Based on the pixel information and position information of each pixel point in the token sequence, an image sequence is generated.

[0063] A Transformer encoder is used to extract features of the image sequence to obtain a feature sequence.

[0064] In order to accurately extract the time sequence and spatial features in the dynamic contrast-enhanced magnetic resonance image sequence, an embodiment of the present application introduces a pixel-level vision transformer (ViT) at the end of the encoder, as shown in FIG. 3. This module takes each pixel point as a unit to construct a token sequence representing the entire image sequence, each token carrying pixel-level information. The token sequence also enhances the model's ability to recognize image spatial positions through two-dimensional Fourier position embedding, thereby generating an image sequence with spatial features and position features. Through this design, the model can capture finer local details and understand the time sequence changes of the image at a global level, thereby improving the generation quality of the multi-parameter map.

[0065] In an optional embodiment, a Transformer encoder is used to extract features of the image sequence to obtain a feature sequence, including:

[0066] The image sequence is processed multiple times using residual connection and continuous layer normalization operation to obtain a feature sequence.

[0067] Optionally, as shown in FIG. 3, the ViT includes a position embedding layer, a linear layer, and a Transformer encoder (i.e., a transformer encoding module). The Transformer encoder is provided with a linear layer before and after it. The ViT first changes the shape of the image, i.e., flattens the image along the spatial dimension to form a token sequence with a length of 256, and each token in the sequence is a vector with a length of f. Then, the position embedding layer is used for position embedding to obtain an image sequence with spatial features and time features with a dimension of f+f p , and then the first linear layer is used to linearly map the dimension of the result to f vThe input is input into a Transformer encoder. The Transformer encoder includes a normalization layer, a multi-head self-attention mechanism, and a feedforward neural network. The image sequence input through the first linear layer is input into the first normalization layer for layer normalization processing and then input into the multi-head self-attention mechanism to capture features in the image sequence; the features output by the multi-head self-attention mechanism are input into the second normalization layer for layer normalization processing, and then input into the feedforward neural network for nonlinear mapping to obtain a feature sequence. The feature sequence output by the Transformer encoder is linearly projected back to the dimension f through the second linear layer, and then output after changing the shape. In the embodiment of the application, the input of the first normalization layer and the input of the second normalization layer are connected through a residual connection. The input of the second normalization layer and the input of the second linear layer are connected through a residual connection. In the embodiment of the application, the Transformer encoder can process the image sequence multiple times to enhance the ability of the model to capture sequence data features. Specifically, the image sequence can be processed multiple times through residual connection and continuous layer normalization operations.

[0068] In the embodiment of the application, the residual connection can be optimized through ReZero regularization and its learnable parameter a to enhance the stability of model training.

[0069] Alternatively, other regularization techniques (such as Layer Normalization or Batch Normalization regularization techniques) can also be used in the embodiment of the application to test the influence of different regularization methods on the training effect of the model.

[0070] In an optional embodiment, the decoder includes a plurality of up-sampling encoding blocks, and the up-sampling encoding blocks are connected to the corresponding down-sampling encoding blocks in the decoder through a skip connection.

[0071] The feature sequence is input into the decoder to obtain a multi-parameter map, including:

[0072] Based on the feature sequence and the features of each up-sampling encoding block corresponding to the down-sampling encoding block, the plurality of up-sampling encoding blocks are used for processing to obtain a multi-parameter map.

[0073] As shown in FIG. 2, the up-sampling encoding block includes an up-sampling convolution block with a step of 2 and a basic convolution block. The feature is input into the basic convolution block after the up-sampling processing of the up-sampling convolution block. The input of the basic convolution block in the up-sampling encoding block is connected to the output of the basic convolution block in the down-sampling encoding block through a skip connection to fuse the features in the corresponding down-sampling encoding block. In the embodiment of the application, the decoder is allowed to access the intermediate features in the encoder stage, which can improve the detail preservation capability in the image reconstruction process, so that the final output can accurately reconstruct the image through the post-processing layer.

[0074] Exemplarily, in combination with the example of the encoder part, the decoder includes 4 up-sampling encoding blocks, the size of the features in the feature sequence input after VIT processing is (16, 16, 384), the feature sequence is input into the up-sampling convolution block U1 of the first up-sampling encoding block for up-sampling processing, and a feature with a size of (32, 32, 384) is obtained, the feature is combined with the basic convolution block B4 of the fourth down-sampling encoding block through a jump connection , and the combined feature is input into the up-sampling convolution block U4 of the fourth up-sampling encoding block for processing; the up-sampling convolution block U4 ; The output feature is input into the up-sampling convolution block U2 of the second up-sampling encoding block for up-sampling processing, and a feature with a size of (64, 64, 192) is obtained, the feature is combined with the basic convolution block B3 of the third down-sampling encoding block through a jump connection , and the combined feature is input into the up-sampling convolution block U3 of the third up-sampling encoding block for processing; the up-sampling convolution block U3 ; The output feature is input into the up-sampling convolution block U3 of the third up-sampling encoding block for up-sampling processing, and a feature with a size of (128, 128, 96) is obtained, the feature is combined with the basic convolution block B2 of the second down-sampling encoding block through a jump connection , and the combined feature is input into the up-sampling convolution block U2 of the second up-sampling encoding block for processing; the up-sampling convolution block U2 ; The output feature is input into the up-sampling convolution block U4 of the fourth up-sampling encoding block for up-sampling processing, and a feature with a size of (256, 256, 48) is obtained, the feature is combined with the basic convolution block B4 of the fourth down-sampling encoding block through a jump connection , and the combined feature is input into the up-sampling convolution block U4 of the fourth up-sampling encoding block for processing; the up-sampling convolution block U4 ;

[0075] In an optional embodiment, as shown in FIG. 4, the generator includes a first generator A and a second generator B, and the discriminator includes a first discriminator D and a second discriminator D B ;

[0076] The first generator A is used to convert the dynamic contrast-enhanced magnetic resonance image sequence into a multi-parameter map, and the second generator B is used to convert the multi-parameter map into a dynamic contrast-enhanced magnetic resonance image sequence;

[0077] The first discriminator D A second discriminator D for judging the authenticity of the converted dynamic contrast-enhanced magnetic resonance image sequence B A second discriminator D for judging the authenticity of the converted dynamic contrast-enhanced magnetic resonance image sequence

[0078] As shown in FIG. 4, the CycleGAN-like model interweaves two generator-discriminator pairs for unpaired image-to-image translation. Denote two image domains as A and B, the CycleGAN-like model uses a generator G A→B to convert images from A to B, a generator G B→A to convert images from B to A. A discriminator D A for distinguishing images in A from images converted from B A f , a discriminator D B for distinguishing images in B from images converted from A B f . The CycleGAN-like model uses identity loss and cycle consistency loss to ensure the fidelity of image content during translation. Wherein A represents the dynamic contrast-enhanced magnetic resonance image sequence domain, and B represents the multi-parameter image domain. A generator G A→B is a first generator that generates multi-parameter images from the dynamic contrast-enhanced magnetic resonance image sequence domain, and a generator G B→A is a second generator that generates dynamic contrast-enhanced magnetic resonance image sequences from the multi-parameter image domain.

[0079] Taking the second discriminator as an example, the loss function of the second discriminator is:

[0080] wherein,

[0081] wherein, is the loss of the second discriminator; denotes the loss of the multi-parameter image obtained by converting from the dynamic contrast-enhanced magnetic resonance image sequence corresponding domain A to the multi-parameter image corresponding domain B, denotes the loss of the real multi-parameter image in the multi-parameter image corresponding domain B; λ GP is the weight of the gradient penalty term, and γ is the target gradient size, denotes the gradient of x, denotes the expected value; denotes the first discriminator; D B denotes the second discriminator; denotes the conversion from the dynamic contrast-enhanced magnetic resonance image sequence corresponding domain A to the multi-parameter image corresponding domain B.

[0082] The loss function of the first discriminator refers to the loss function of the second discriminator, and the calculation methods of the two are the same.

[0083] The discriminator loss function combines the LSGAN loss to reduce oscillation in the training process, optimize the image quality after reconstruction, and perform better than traditional GAN loss functions in preserving image details. The introduction of a gradient penalty term in the loss function helps balance the relationship between the generator and the discriminator during training, making the model training more stable and reducing potential training oscillation. This loss function design aims to improve the performance of the discriminator in distinguishing between real and generated images, thereby encouraging the generator to produce higher quality images and ensuring high consistency between the images and the real situation. This enables the model to generate multi-parameter maps that are more realistic and accurate in both visual and statistical characteristics.

[0084] In the embodiments of the present application, by using the above loss function and combining pixel-level ViT and encoding-decoding architecture, efficient and accurate mapping from dynamic contrast-enhanced magnetic resonance image sequences to multi-parameter maps is successfully achieved, significantly improving the parameter estimation capability of dynamic contrast-enhanced MRI.

[0085] Optionally, other types of loss functions such as Wasserstein GAN loss or Conditional GAN loss can also be used in the embodiments of the present application to further improve image quality and model stability.

[0086] In an optional embodiment, after constructing the CycleGAN-like model, the method further comprises:

[0087] The paired image blocks are extracted from the pre-obtained dynamic contrast-enhanced magnetic resonance image sequence and corresponding multi-parameter map data set, and the Adam optimizer is used to optimize the loss function to obtain the optimized CycleGAN-like model.

[0088] Optionally, the discriminator can be optimized by introducing other forms of regularization or deeper network architecture to improve its ability to recognize details, which is not limited in the embodiments of the present application.

[0089] The method in the embodiments of the present application is verified based on a cervical cancer DCE MRI sequence data set, and the experimental results are shown in FIG. 5:

[0090] In FIG. 5, the first row is the experimental results of the method in the embodiments of the present application (i.e., the experimental results of the present application in FIG. 5), which respectively shows the generated K trans parameter maps and their ROI enlarged images, the generated v e parameter maps and their ROI enlarged images; the second row is the corresponding real reference images.

[0091] For K trans real images and v eReal image profile analysis curve analysis, results as shown in Figure 6, the left shows the generated K trans Real image and v e Real image, where the white line marks the specific profile analyzed. The deep learning methods used for comparison with the embodiments of the present application include Pix2Pix model, CycleGAN model, UNet model and AttUnet model. The graph on the right shows the parameter value changes along the white profile line position (from left to right), where the blue curve Ours represents the analysis results of the parameter map generated by the method in the embodiments of the present application, and the red curve Ground Truth represents the real results. The area framed by the red dotted line in the figure highlights the specific distribution of profile values in the cervical and uterine regions (ROI), verifying the accuracy of the method in the embodiments of the present application in the key area.

[0092] It should be noted that although the embodiments of the present application mainly aim at DCE-MRI parameter mapping, they are also applicable to other medical imaging techniques and disease types:

[0093] 1) Parameter image generation for other diseases: The method of the embodiments of the present application can be applied to parameter image generation for other diseases such as heart disease, brain tumor, etc., providing support for accurate diagnosis and treatment planning under different pathological conditions.

[0094] 2) Parameter generation for other modal medical images: For example, for positron emission tomography (PET) or computed tomography (CT) images, the embodiments of the present application can also be used to generate parameter images reflecting the functional state or pathological changes of tissues.

[0095] 3) Cross-modality medical image conversion: The model of the embodiments of the present application can be used for image conversion between different medical imaging modalities, such as converting CT images to dynamic contrast-enhanced magnetic resonance images, or vice versa, thereby providing more rich clinical information.

[0096] The method in the embodiments of the present application can also be applied outside medical imaging and can be applied to image-to-image conversion tasks in other fields, such as satellite image processing, artistic style conversion, etc.

[0097] The method in the embodiments of the present application is also applicable to multi-modality imaging fusion: The method in the embodiments of the present application is also applicable to conversion between different imaging modalities, such as PET and CT, MRI and CT, etc. image fusion to obtain more rich diagnostic information.

[0098] The method in the embodiments of the present application is applicable to image processing in non-medical fields: the improved design of the CycleGAN-like model can be applied to image processing tasks in non-medical fields, such as video frame interpolation, high-resolution imaging, and real-time image recognition systems in autonomous vehicles.

[0099] Based on the same principle as the method provided in the embodiments of the present application, the embodiments of the present application also provide an intelligent dynamic contrast-enhanced magnetic resonance parameter imaging device, as shown in FIG. 7, which comprises:

[0100] The model construction module 701 is configured to construct a CycleGAN-like model; the CycleGAN-like model comprises a generator and a discriminator, the generator comprises an encoder and a decoder, and a pixel-level visual transformer is arranged at the connection position of the encoder and the decoder;

[0101] The encoding module 702 is configured to input the dynamic contrast-enhanced magnetic resonance image sequence into the encoder to obtain an encoded image;

[0102] The time sequence learning module 703 is configured to construct the encoded image into an image sequence by using the pixel-level visual transformer, and extract a feature sequence based on the image sequence; the image sequence comprises pixel information and position information;

[0103] The decoding module 704 is configured to input the feature sequence into the decoder to obtain a multi-parameter map;

[0104] The CycleGAN-like model uses a loss function to optimize the performance of the discriminator; the loss function is constructed based on a least squares GAN loss and a gradient penalty enhancement loss.

[0105] In the embodiments of the present application, the pixel-level visual transformer is integrated at the connection position of the encoder and the decoder of the generator to enhance the learning ability of non-local patterns and time sequence information in medical images, which can improve the parameter estimation accuracy in a variable noise environment. The gradient penalty is introduced into the loss function to optimize the performance of the discriminator, so that it can more effectively distinguish between real and generated images. The CycleGAN-like model can fully extract and utilize spatial features while performing time sequence analysis, providing more comprehensive and fine PK parameter maps, and providing strong technical support for improving the accuracy of cervical cancer and other disease diagnosis.

[0106] The intelligent dynamic contrast-enhanced magnetic resonance parameter imaging device provided in the embodiments of the present application can realize the various processes realized in the method embodiments of FIGS. 1 to 6, and thus the details are not repeated here.

[0107] The intelligent dynamic contrast-enhanced magnetic resonance parameter imaging device provided in the embodiments of the present application can execute the intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method provided in the embodiments of the present application, and the implementation principles are similar. The actions performed by each module and unit in the intelligent dynamic contrast-enhanced magnetic resonance parameter imaging device in the embodiments of the present application are corresponding to the steps in the intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method in the embodiments of the present application. For the detailed function description of each module of the intelligent dynamic contrast-enhanced magnetic resonance parameter imaging device, refer to the description of the corresponding intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method shown in the foregoing, which will not be repeated here.

[0108] Based on the same principle as the method shown in the embodiments of the present application, the embodiments of the present application also provide an electronic device, which can include but is not limited to a processor and a memory. The memory is used to store a computer program. The processor is used to execute the intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method shown in any optional embodiment of the present application by calling the computer program. Compared with the prior art, the intelligent dynamic contrast-enhanced magnetic resonance parameter imaging method provided in the present application integrates a pixel-level visual transformer at the connection position of the encoder and the decoder of the generator to enhance the learning ability of non-local patterns and time series information in medical images, which can improve the parameter estimation accuracy in a variable noise environment. The gradient penalty is introduced into the loss function to optimize the performance of the discriminator, so that it can more effectively distinguish between real and generated images. The CycleGAN model can fully extract and utilize spatial features while analyzing time series, providing more comprehensive and detailed PK parameter maps, and providing strong technical support for improving the accuracy of cervical cancer and other disease diagnosis.

[0109] In an optional embodiment, an electronic device is also provided, as shown in FIG. 8. The electronic device 800 shown in FIG. 8 can be a server, which includes a processor 801 and a memory 803. The processor 801 and the memory 803 are connected, such as through a bus 802. Optionally, the electronic device 800 can also include a transceiver 804. It should be noted that the transceiver 804 is not limited to one in actual application, and the structure of the electronic device 800 does not constitute a limitation on the embodiments of the present application.

[0110] The processor 801 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It can implement or execute the various exemplary logical blocks, modules and circuits described in connection with the disclosure. The processor 801 can also be a combination of computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0111] The bus 802 can include a path for transmitting information between the above-mentioned components. The bus 802 can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 802 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is shown in FIG. 8, but it does not mean that there is only one bus or only one type of bus.

[0112] The memory 803 can be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.

[0113] The memory 803 is configured to store application program codes for implementing the solutions of the present application, and the processor 801 is configured to execute the application program codes stored in the memory 803.

[0114] The electronic device includes, but is not limited to, a mobile terminal such as a mobile phone, a notebook computer, a digital broadcast receiver, a PDA (Personal Digital Assistant), a PAD (Tablet Personal Computer), a PMP (Portable Multimedia Player), a car terminal (for example, a car navigation terminal), and the like, and a fixed terminal such as a digital TV, a desktop computer, and the like. The electronic device shown in FIG. 8 is merely an example, and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0115] The server provided in the present application can be a stand-alone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, and basic cloud computing services such as big data and artificial intelligence platforms. The terminal can be a smartphone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, and the like, but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication, which is not limited in the present application.

[0116] The embodiments of the present application provide a computer readable storage medium, which stores a computer program. When the computer program is executed on a computer, the computer can execute the corresponding content in the foregoing method embodiments.

[0117] It should be understood that, although each step in the flowchart of the accompanying drawings is shown in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and they can be executed in other sequences. Moreover, at least part of the steps in the flowchart of the accompanying drawings can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or sub-steps or stages of other steps.

[0118] It should be noted that the computer readable storage medium in the above application can also be a computer readable signal medium or a combination of a computer readable storage medium and a computer readable storage medium. The computer readable storage medium may, for example, be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this application, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In this application, the computer readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer readable program code. Such a propagated data signal can take many forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer readable signal medium can also be any computer readable medium other than the computer readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained in the computer readable medium can be transmitted by any suitable medium, including but not limited to a wire, a cable, an RF (radio frequency) or the like, or any suitable combination of the above.

[0119] The above computer readable medium can be contained in the above electronic device; or can exist separately and not be assembled into the electronic device.

[0120] The above computer readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods shown in the above embodiments.

[0121] According to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the intelligent dynamic contrast enhancement magnetic resonance parameter imaging method and device provided in the various optional implementation manners.

[0122] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0123] The computer program instructions can also be loaded onto a computer or other programmable information processing apparatus to cause a series of operations to be performed on the computer or other programmable information processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable information processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0124] The modules involved in the embodiments of the present application can be implemented in software or hardware. In some cases, the name of a module does not constitute a limitation on the module itself. For example, the model construction module can also be described as a "model construction module for constructing a CycleGAN model".

[0125] The above description is merely that of the preferred embodiments of the present application and of the principles thereof. It is to be understood that the disclosed scope of the present application is not limited to the specific combinations of technical features described above, and also encompasses other technical solutions formed by any combination of the technical features described above or equivalent features, without departing from the disclosed concept. For example, the above technical features can be replaced with other technical features disclosed in the present application (but not limited to) having similar functions to form technical solutions.

Claims

1. A method of intelligent dynamic contrast-enhanced magnetic resonance parametric imaging, characterized in that, The method comprises: constructing a CycleGAN-like model; the CycleGAN-like model comprises a generator and a discriminator, the generator comprises an encoder and a decoder, and a pixel-level visual transformer is arranged at the connection position of the encoder and the decoder; inputting a dynamic contrast-enhanced magnetic resonance image sequence into the encoder to obtain an encoded image; using the pixel-level visual transformer to construct the encoded image into an image sequence, and extracting a feature sequence based on the image sequence; the image sequence comprises pixel information and position information; inputting the feature sequence into the decoder to obtain a multi-parameter map; wherein the CycleGAN-like model uses a loss function to optimize the performance of the discriminator; the loss function is constructed based on a least squares GAN loss and a gradient penalty enhancement loss.

2. The intelligent dynamic contrast enhanced magnetic resonance parametric imaging method of claim 1, wherein, The encoder and the decoder are UNet encoding-decoding structures, and the encoder comprises a plurality of down-sampling encoding blocks; the step of inputting the dynamic contrast-enhanced magnetic resonance image sequence into the encoder to obtain an encoded image comprises: preprocessing the dynamic contrast-enhanced magnetic resonance image sequence to convert it into a three-dimensional tensor; using a plurality of down-sampling encoding blocks to perform layer-by-layer dimension reduction on the three-dimensional tensor to obtain the encoded image.

3. The intelligent dynamic contrast enhanced magnetic resonance parametric imaging method of claim 1, wherein, The pixel-level visual transformer comprises a Transformer encoder; the step of using the pixel-level visual transformer to construct the encoded image into an image sequence and extracting a feature sequence based on the image sequence comprises: constructing a token sequence in units of each pixel point in the encoded image; generating position information of each pixel point in the token sequence by means of two-dimensional Fourier position embedding; generating the image sequence based on the pixel information and the position information of each pixel point in the token sequence; using the Transformer encoder to perform feature extraction on the image sequence to obtain the feature sequence. using the Transformer encoder to perform feature extraction on the image sequence to obtain the feature sequence comprises:

4. The intelligent dynamic contrast enhanced magnetic resonance parametric imaging method of claim 1, wherein, using residual connection and continuous layer normalization operations to process the image sequence multiple times to obtain the feature sequence. The decoder comprises a plurality of up-sampling encoding blocks, and the up-sampling encoding blocks are connected to corresponding down-sampling encoding blocks in the decoder by skip connection; 5. The intelligent dynamic contrast enhanced magnetic resonance parametric imaging method of claim 2, wherein, the step of inputting the feature sequence into the decoder to obtain a multi-parameter map comprises: using a plurality of up-sampling encoding blocks to process the feature sequence and the features of the corresponding down-sampling encoding blocks of each up-sampling encoding block to obtain the multi-parameter map. The generator comprises a first generator and a second generator, and the discriminator comprises a first discriminator and a second discriminator; 6. The intelligent dynamic contrast enhanced magnetic resonance parametric imaging method of claim 1, wherein, the first generator is used to convert a dynamic contrast-enhanced magnetic resonance image sequence into a multi-parameter map, and the second generator is used to convert a multi-parameter map into a dynamic contrast-enhanced magnetic resonance image sequence; the first discriminator is used to judge the authenticity of the converted dynamic contrast-enhanced magnetic resonance image sequence, and the second discriminator is used to judge the authenticity of the converted multi-parameter map; ​ The loss function of the second discriminator is: wherein wherein for the second discriminator; a loss representing a multi-parametric map obtained from converting a dynamic contrast-enhanced magnetic resonance image sequence corresponding to domain A to domain B corresponding to the multi-parametric map, denotes the loss of the real multi-parameter graph in the domain B corresponding to the multi-parameter graph; λ GP is the weight of the gradient penalty term, and γ is the target gradient size, denotes the gradient of x, represents an expected value; denotes a first discriminator; D B denotes a second discriminator; indicates a conversion from a domain A corresponding to a dynamic contrast-enhanced magnetic resonance image sequence to a domain B corresponding to a multi-parametric map.

7. The intelligent dynamic contrast enhanced magnetic resonance parametric imaging method of claim 1, wherein, After the CycleGAN-like model is constructed, the method further includes: Pairing image blocks are extracted from a pre-obtained dynamic contrast-enhanced magnetic resonance image sequence and corresponding multi-parametric map data set, and input into the CycleGAN-like model, and an Adam optimizer is used to optimize the loss function, thereby obtaining an optimized CycleGAN-like model.

8. An intelligent dynamic contrast enhanced magnetic resonance parametric imaging apparatus, characterized by, The device includes: a model construction module configured to construct a CycleGAN-like model, the CycleGAN-like model including a generator and a discriminator, the generator including an encoder and a decoder, and a pixel-level visual transformer being arranged at a connection position of the encoder and the decoder; an encoding module configured to input a dynamic contrast-enhanced magnetic resonance image sequence into the encoder to obtain an encoded image; a time sequence learning module configured to construct an image sequence from the encoded image by using the pixel-level visual transformer, and extract a feature sequence based on the image sequence, the image sequence including pixel information and position information; a decoding module configured to input the feature sequence into the decoder to obtain a multi-parametric map; wherein the CycleGAN-like model optimizes performance of the discriminator by using a loss function, and the loss function is constructed based on a least squares GAN loss and a gradient penalty enhancement loss.

9. An electronic device, comprising: The computer program is stored on the computer readable storage medium and executable on the processor, and the processor executes the program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored on the computer readable storage medium and executable on the processor, and the processor executes the program to implement the method of any one of claims 1 to 7.

Images