Accelerated multi-contrast imaging method, system and terminal based on diffusion model

Abstract

The application discloses an accelerated multi-contrast imaging method and system based on a diffusion model and a terminal, and the method comprises the following steps: acquiring an undersampling measurement signal in magnetic resonance, inputting the undersampling measurement signal into a trained convolutional neural network, and outputting joint distribution data from the trained convolutional neural network; determining an accelerated multi-contrast image generation problem based on the undersampling measurement signal, and constructing a joint distribution diffusion model by using a Bayesian method and a Langevin dynamics Markov Monte Carlo sampling method; and iteratively processing the joint distribution diffusion model based on the joint distribution data to obtain a target multi-contrast weighting image and output the target multi-contrast weighting image. The method can obtain a multi-contrast weighting image with good generalization and acceleration performance, and solves the problem that the current accelerated multi-contrast imaging method cannot simultaneously meet the generalization and acceleration performance.

Description

Technical Field

[0001] This invention relates to the field of magnetic resonance imaging, and more particularly to an accelerated multi-contrast imaging method, system, and terminal based on a diffusion model. Background Technology

[0002] Currently, in multi-parameter imaging of magnetic resonance imaging (MRI), it is typically necessary to adjust different TR (Repetition Time), TE (echo time), or TI (Longitudinal Relaxation Time) values ​​to run multiple encoding and acquisition sequences separately, thereby generating various contrast-weighted images. Then, exponential fitting is performed on the weighted images corresponding to different parameters to obtain the corresponding parametric map. However, in MRI parametric imaging, multiple weighted images need to be acquired independently for weighting, increasing the overall parametric imaging time and the possibility of motion artifacts. Therefore, various acceleration methods have emerged.

[0003] Current methods for accelerating multi-contrast imaging are mainly divided into two categories: the first is based on traditional compressed sensing methods, and the other is based on deep learning methods. However, the first method requires manual tuning to balance the relationship between constraint terms and data fidelity terms, which easily leads to suboptimal solutions and requires many reconstruction steps to achieve the desired effect. Although the second method can solve the problem of manual tuning, it uses a designed black-box neural network block to capture the mapping relationship between outputs, which lacks interpretability and requires retraining for different sampling templates, resulting in poor generalization.

[0004] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0005] The main objective of this invention is to provide a diffusion-based accelerated multi-contrast imaging method, system, terminal, and storage medium, which aims to solve the problem that existing diffusion-based accelerated multi-contrast imaging methods cannot simultaneously satisfy generalization and acceleration performance.

[0006] To achieve the above objectives, a first aspect of the present invention provides an accelerated multi-contrast imaging method based on a diffusion model, wherein the accelerated multi-contrast imaging method based on a diffusion model includes:

[0007] The undersampled measurement signal in the magnetic resonance imaging is acquired, and the undersampled measurement signal is input into the trained convolutional neural network. The trained convolutional neural network outputs joint distribution data. The trained convolutional neural network is trained according to a preset training process.

[0008] Based on the undersampled measurement signal, the problem of accelerating multi-contrast image generation is determined, and a joint distribution diffusion model is constructed for the problem of accelerating multi-contrast image generation using Bayesian method and Langevin dynamics Markov Monte Carlo sampling method.

[0009] Based on the joint distribution data, the joint distribution diffusion model is iteratively processed to obtain the target multi-contrast weighted map, which is then output.

[0010] Optionally, the steps of the preset training process include:

[0011] Acquire a preset number of full sample data, perturb the full sample data according to a preset number of times, and acquire the perturbation data generated after each perturbation of the full sample data during the perturbation process;

[0012] The fully sampled data and the perturbation data are input into the convolutional neural network for training to obtain the trained convolutional neural network.

[0013] Optionally, the step of inputting the fully sampled data and the perturbation data into the convolutional neural network for training to obtain the trained convolutional neural network includes:

[0014] Obtain the perturbation data corresponding to each full sample data;

[0015] Each fully sampled data point and its corresponding perturbation data are input into the convolutional neural network, and the parameters of the convolutional neural network are optimized according to the optimization parameter loss function.

[0016] When the training iterations of the convolutional neural network reach the preset number, the training is completed, and the trained convolutional neural network is output.

[0017] Optionally, the step of iteratively processing the joint distribution diffusion model based on the joint distribution data to obtain the target multi-contrast weighted map and outputting it includes:

[0018] The joint distribution diffusion model is iteratively calculated using a preset algorithm and the joint distribution data.

[0019] When the number of iterations reaches the preset number, the iteration ends, and the target multi-contrast weighted image obtained from the last iteration is output.

[0020] Optionally, the step of determining the accelerated multi-contrast image generation problem based on the undersampled measurement signal, and constructing a joint distribution diffusion model of the accelerated multi-contrast image generation problem using Bayesian methods and Langevin dynamics Markov Monte Carlo sampling methods, includes:

[0021] Determine the problem of accelerating multi-contrast image generation based on undersampled measurement signals;

[0022] The problem of accelerating multi-contrast image generation is transformed into a maximum likelihood estimation problem using the Bayesian method.

[0023] The joint distribution diffusion model is constructed based on Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem.

[0024] Optionally, the step of transforming the accelerated multi-contrast image generation problem into a maximum likelihood estimation problem according to the Bayesian method includes:

[0025] The accelerated multi-contrast image is transformed into a preliminary maximum likelihood estimation problem based on the Bayesian framework;

[0026] The initial maximum likelihood estimation problem is expanded using the Bayesian method, and the maximum likelihood estimation problem is obtained through logarithmic processing.

[0027] Optionally, the step of constructing the joint distribution diffusion model based on the Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem includes:

[0028] The Langevin dynamics Markov Monte Carlo sampling method is applied to the posterior probability in the maximum likelihood estimation problem to obtain a joint distribution diffusion model.

[0029] A second aspect of the present invention provides an accelerated multi-contrast imaging system based on a diffusion model, wherein the accelerated multi-contrast imaging system based on a diffusion model includes:

[0030] The data acquisition module is used to acquire undersampled measurement signals in magnetic resonance imaging, input the undersampled measurement signals into a trained convolutional neural network, and output joint distribution data from the trained convolutional neural network. The trained convolutional neural network is trained according to a preset training process.

[0031] The model building module is used to determine the problem of accelerating multi-contrast image generation based on the undersampled measurement signal, and to construct a joint distribution diffusion model of the problem of accelerating multi-contrast image generation using Bayesian method and Langevin dynamics Markov Monte Carlo sampling method.

[0032] The output module is used to iteratively process the joint distribution diffusion model based on the joint distribution data to obtain the target multi-contrast weighted map and output it.

[0033] A third aspect of the present invention provides a terminal, the terminal including a memory, a processor, and a diffusion-based accelerated multi-contrast imaging program stored in the memory and executable on the processor, wherein the diffusion-based accelerated multi-contrast imaging program, when executed by the processor, implements any of the steps of the diffusion-based accelerated multi-contrast imaging method.

[0034] A fourth aspect of the present invention provides a computer-readable storage medium storing a diffusion-based accelerated multi-contrast imaging program, wherein the diffusion-based accelerated multi-contrast imaging program, when executed by a processor, implements any of the steps of the diffusion-based accelerated multi-contrast imaging method.

[0035] As can be seen from the above, in the present invention, an undersampled measurement signal in magnetic resonance is acquired, and the undersampled measurement signal is input into a trained convolutional neural network. The trained convolutional neural network outputs joint distribution data. The trained convolutional neural network is trained according to a preset training process. Based on the undersampled measurement signal, an accelerated multi-contrast image generation problem is determined, and a joint distribution diffusion model is constructed for the accelerated multi-contrast image generation problem using a Bayesian method and a Langevin dynamics Markov Monte Carlo sampling method. Based on the joint distribution data, the joint distribution diffusion model is iteratively processed to obtain a target multi-contrast weighted map, which is then output.

[0036] Compared with existing technologies, this invention addresses the problem that current accelerated multi-contrast imaging methods based on diffusion models cannot simultaneously satisfy generalization and acceleration performance. Within a Bayesian framework, it proposes an accelerated multi-contrast imaging method based on a diffusion model. Compared to traditional black-box-designed neural network deep learning methods, the proposed method considers the problem from a Bayesian perspective, introducing higher-dimensional joint distribution prior information from the full sampling, resulting in greater flexibility and generalization. Furthermore, the convolutional neural network in the proposed method does not require retraining for different sampling templates, further enhancing its generalization ability. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic flowchart of an accelerated multi-contrast imaging method based on a diffusion model provided in an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of the forward perturbation process and the backward sampling generation process provided in the embodiments of the present invention;

[0040] Figure 3 This is a schematic diagram of the preset algorithm provided in the embodiment of the present invention;

[0041] Figure 4 This is a schematic diagram of the constituent modules of an accelerated multi-contrast imaging system based on a diffusion model provided in an embodiment of the present invention;

[0042] Figure 5 This is a block diagram illustrating the internal structure of a terminal provided in an embodiment of the present invention. Detailed Implementation

[0043] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0044] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0045] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0046] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0047] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to classification." Similarly, the phrases "if determined" or "if classified to [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once classified to [the described condition or event]," or "in response to classification to [the described condition or event]."

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0050] In multi-parameter MRI imaging, different TR, TE, or TI values ​​are typically adjusted to run multiple encoding and acquisition sequences to generate various contrast-weighted images. After measuring the weighted images corresponding to different parameters, exponential fitting is performed to obtain the corresponding parametric map. However, in MRI parametric imaging, multiple weighted images need to be acquired independently, increasing the overall imaging time and the likelihood of motion artifacts. Therefore, various acceleration methods have emerged. Current accelerated multi-contrast imaging methods are mainly divided into two categories: the first is based on traditional compressed sensing methods, and the other utilizes deep learning methods. The former requires manual tuning to balance the relationship between constraint terms and data fidelity terms, easily reaching suboptimal solutions, and requires many reconstruction steps to achieve the desired effect. The latter, although it can solve the problem of manual tuning, uses designed black-box neural network blocks to capture the mapping relationship between outputs, lacking interpretability, and requires retraining for different sampling templates, resulting in poor generalization.

[0051] To address at least one of the aforementioned problems, this invention provides an accelerated multi-contrast imaging method, system, terminal, and storage medium based on a diffusion model. Specifically, it involves acquiring undersampled measurement signals from magnetic resonance imaging (MRI), inputting these signals into a trained convolutional neural network (CNN), and the trained CNN outputting joint distribution data. The trained CNN is trained according to a pre-defined training process. Based on the undersampled measurement signals, an accelerated multi-contrast image generation problem is determined, and a joint distribution diffusion model is constructed using a Bayesian method and a Langevin dynamics Markov Monte Carlo sampling method. Based on the joint distribution data, the joint distribution diffusion model is iteratively processed to obtain a target multi-contrast weighted map, which is then output.

[0052] This invention proposes an accelerated multi-contrast imaging method based on a diffusion model within a Bayesian framework. Compared to traditional neural network deep learning methods based on black-box design, the proposed method considers the problem from a Bayesian perspective, introducing higher-dimensional joint distribution prior information from the full sampling, resulting in greater flexibility and generalization. Furthermore, the convolutional neural network in the proposed method does not require retraining for different sampling templates, further enhancing its generalization ability.

[0053] Exemplary methods

[0054] like Figure 1 As shown, this embodiment of the invention provides an accelerated multi-contrast imaging method based on a diffusion model. Specifically, the accelerated multi-contrast imaging method based on a diffusion model includes the following steps:

[0055] Step S100: Obtain the undersampled measurement signal in the magnetic resonance imaging, input the undersampled measurement signal into the trained convolutional neural network, and output the joint distribution data of the trained convolutional neural network. The trained convolutional neural network is trained according to a preset training process.

[0056] Specifically, the undersampled measurement signal in the aforementioned magnetic resonance imaging (MRI) is represented in the form of an image. Undersampling is a technique employed when the bandwidth of the MRI equipment is insufficient, effectively increasing the bandwidth of the testing equipment to enable the sampling of higher frequency signals. The aforementioned Convolutional Neural Networks (CNNs) are a type of feedforward neural network that includes convolutional computations and has a deep structure.

[0057] It is understandable that when acquiring the undersampled measurement signal during the magnetic resonance process, it is necessary to recover the target multi-contrast weighted image from the undersampled measurement signal. Therefore, a trained convolutional neural network is first used to process the undersampled measurement signal to obtain the joint distribution data required for recovering the target multi-contrast weighted image.

[0058] It should be noted that for some independent and identically distributed samples from an unknown distribution p(x), the gradient of the logarithm of the probability density function of p(x) is approximated as... It can be further expressed as Where x represents the target multi-contrast weighted image, and i represents a set of indices, the corresponding Langevin dynamics Markov Monte Carlo sampling is as shown in formula (1):

[0059]

[0060] in This represents the standard Wiener process during iterative sampling, g i Indicates the step size. This represents a standard Gaussian distribution. It can be solved directly in formula (1). Right now It is difficult, therefore a convolutional neural network needs to be trained to approximate it. In this application, the target multi-contrast weighted map is obtained by means of formula (1). Therefore, the undersampled measurement signal is processed by the trained convolutional neural network to obtain the joint distribution data required in the process of recovering the target multi-contrast weighted map.

[0061] Furthermore, the steps of the preset training process include:

[0062] Acquire a preset number of full sample data, perturb the full sample data according to a preset number of times, and acquire the perturbation data generated after each perturbation of the full sample data during the perturbation process;

[0063] The fully sampled data and the perturbation data are input into the convolutional neural network for training to obtain the trained convolutional neural network.

[0064] Specifically, the full sampled data and the perturbation data are denoted as... Where i represents a sequence number, N represents the number of steps in the perturbation process, and d represents the number of weighted images contained in each full sample data and perturbation data. For example, if a multi-contrast weighted image contains 4 weighted images, then d = 4; x i This represents a set of multi-contrast weighted images in each full-sampled data. The process of performing corresponding denoising and score matching on the above full-sampled data involves perturbing the data using a Gaussian perturbation kernel with different variances, and then using a network to approximate the logarithmic gradient of the perturbed data. The approximation form is shown in the following formula (2):

[0065]

[0066] in, This represents the data distribution after perturbation with the addition of a perturbation kernel. In other words, the perturbation process ensures that the resulting perturbed data conforms to this distribution; specifically, the perturbation process passes through the perturbation kernel. Gradually disturb the clean, fully sampled data until pure noise is obtained after a preset number of perturbation processes. Expressing expectations, s represents the data after perturbation. θ Let ||·|| represent the corresponding convolutional neural network, ||·|| represent the modulus calculation process, and σ represents considering a series of noise scales. min =σ1<σ2<…<σ L =σ maxThe noise scale is divided into 1, 2, 3, ..., L from smallest to largest. The perturbation process can be regarded as a continuous diffusion process. After a preset number of perturbation processes, a set of fully sampled data is obtained. It can be approximated as standard Gaussian noise. This process is also known as the forward process, where σ represents the noise variance and can be used to represent the magnitude of the noise scale.

[0067] During the perturbation process, each perturbation will result in a set of perturbation data corresponding to the full sampled data. Each set of perturbation data has a certain correlation with the perturbation data obtained from the previous perturbation. Therefore, the full sampled data and the perturbation data are input into the convolutional neural network to train the convolutional neural network, learn this correlation, and thus obtain the trained convolutional neural network.

[0068] Furthermore, in this application, the specific perturbation processing, such as Figure 2 As shown in the forward process, the perturbation process of the clean full sampled data is to add noise to it accordingly until a preset number of noise addition processes are passed, that is, after a preset number of perturbation processes, a pure noise map is obtained. In this process, the noise addition process, i.e. the perturbation process, follows the Brownian motion rule.

[0069] Furthermore, the step of inputting the fully sampled data and the perturbation data into the convolutional neural network for training to obtain the trained convolutional neural network includes:

[0070] Obtain the perturbation data corresponding to each full sample data;

[0071] Each fully sampled data point and its corresponding perturbation data are input into the convolutional neural network, and the parameters of the convolutional neural network are optimized according to the optimization parameter loss function.

[0072] When the training iterations of the convolutional neural network reach the preset number, the training is completed, and the trained convolutional neural network is output.

[0073] Specifically, each full sample data contains multiple contrast images. After multiple perturbation processes, multiple perturbation data are obtained. Each full sample data and its corresponding perturbation data are input into the convolutional neural network. The convolutional neural network learns the joint distribution of the multi-contrast images in each perturbation process. Specifically, the convolutional neural network optimizes its parameters according to the optimization parameter loss function, which is a transformation of formula (2) and can be applied to the convolutional neural network. Specifically, it is expressed as formula (3):

[0074]

[0075] Where, θ * X represents the optimized parameters, θ represents the unoptimized parameters, and X represents the unoptimized parameters. q This represents the full sample data. This represents the data after each perturbation, where i represents the ordinal number, and {x} 1 x 2 , ..., x m}~P data (X q ) represents the probability distribution of the full sampled data, where x 1 x 2 , ..., x m This indicates that a fully sampled dataset contains m weighted images. Given sufficient fully sampled data and its corresponding perturbation data, after training with a preset number of training samples, the trained convolutional neural network obtains the target multi-contrast weighted image s from the input undersampled data. θ* (X, σ) can be matched to its Where X is the target multi-contrast weighted map. That is, the network can capture the true distribution information of the data, and has stronger interpretability compared to black-box designed neural networks that learn the joint distribution information between data.

[0076] Step S200: Based on the undersampled measurement signal, determine the problem of accelerating multi-contrast image generation, and construct a joint distribution diffusion model for the problem of accelerating multi-contrast image generation using the Bayesian method and the Langevin dynamics Markov Monte Carlo sampling method.

[0077] For the input undersampled measurement signal, the sampled measurement signal is represented in the form of a noise graph, denoted as Y, where Y = [y1, y2, ..., y]. m The image indicates that the acquired undersampled measurement signal contains multiple weighted images, and the corresponding final target multi-contrast weighted image is represented as X = [x1, x2, ..., x]. m The image represents the final stitched target multi-contrast weighted image, where X and Y have the same dimension. When there are fewer than m weighted images collected in Y, and a Y containing m contrast weighted images is needed, the missing weighted images in X are filled with 0, i.e., filled with a completely noisy image. In the process of obtaining X from Y, it is necessary to determine the problem of accelerating multi-contrast image generation based on the undersampled measurement signal. This problem is then used to construct a joint distribution diffusion model using Bayesian methods and Langevin dynamics Markov Monte Carlo sampling. The final Y is obtained based on this joint distribution diffusion model.

[0078] Furthermore, the step of determining the accelerated multi-contrast image generation problem based on the undersampled measurement signal, and constructing a joint distribution diffusion model for the accelerated multi-contrast image generation problem using Bayesian methods and Langevin dynamics Markov Monte Carlo sampling methods, includes:

[0079] Determine the problem of accelerating multi-contrast image generation based on undersampled measurement signals;

[0080] The problem of accelerating multi-contrast image generation is transformed into a maximum likelihood estimation problem using the Bayesian method.

[0081] The joint distribution diffusion model is constructed based on Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem.

[0082] In the problem of accelerating multi-contrast image generation, the process of accelerating multi-contrast image generation can be regarded as solving the inverse problem in a forward process. The specific problem of accelerating multi-contrast image generation can be expressed as formula (4):

[0083]

[0084] Subject to Y=EX, (4);

[0085] Where E represents the encoding matrix and Ψ(·) represents the sparse transformation. X can be solved using traditional CS-based accelerated reconstruction methods. In this application, the problem of accelerating multi-contrast image generation is transformed into a maximum likelihood estimation problem using a Bayesian method; the joint distribution diffusion model is constructed based on the Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem.

[0086] Furthermore, the step of transforming the accelerated multi-contrast image generation problem into a maximum likelihood estimation problem using the Bayesian method includes:

[0087] The accelerated multi-contrast image is transformed into a preliminary maximum likelihood estimation problem based on the Bayesian framework;

[0088] The initial maximum likelihood estimation problem is expanded using the Bayesian method, and the maximum likelihood estimation problem is obtained through logarithmic processing.

[0089] Specifically, within the Bayesian framework, formula (4) corresponding to the problem of accelerating multi-contrast image generation will be transformed into the following paradigm (5), namely, the preliminary maximum likelihood estimation problem:

[0090]

[0091] Solving for X (where X represents the final stitched target multi-contrast weighted image, which is not solved here but only used to construct the paradigm) becomes the maximum likelihood estimation process for X given a noisy undersampled measurement signal Y. In this way, the present invention can recover a clean weighted image more accurately using maximum likelihood estimation. For (5), Bayesian expansion is used, and then the logarithm of both sides is taken simultaneously to obtain the maximum likelihood estimation problem, expressed as the following formula (6):

[0092]

[0093] As can be seen from (6), the present application embodiment transforms the solution process of X in (4) into a maximum a posteriori estimation of X given Y, where p(Y|X) represents the conditional probability and follows a Gaussian distribution.

[0094] Furthermore, the step of constructing the joint distribution diffusion model based on the Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem includes:

[0095] The Langevin dynamics Markov Monte Carlo sampling method is applied to the posterior probability in the maximum likelihood estimation problem to obtain a joint distribution diffusion model.

[0096] Specifically, assuming there is already Then X can be solved from the posterior p(X|Y) by performing the Langevin dynamics Markov Monte Carlo sampling method, that is, multiple weighted graphs can be recovered. That is, the joint distribution diffusion model can be obtained by performing the Langevin dynamics Markov Monte Carlo sampling method according to the posterior probability in formula (6), which is specifically expressed as the following formula (7):

[0097]

[0098] Since the variable to be inferred in formula (6) is X, and logp(Y) can be eliminated, in formula (7), k represents the number of inference steps, γ is a scalar controlling the step size, and ζ k This represents standard Gaussian noise, whose dimension is consistent with the Y dimension of the input.

[0099] Based on the obtained joint distribution diffusion model, in Given the information, the corresponding X can be obtained.

[0100] Step S300: Based on the joint distribution data, iteratively process the joint distribution diffusion model to obtain the target multi-contrast weighted map and output it.

[0101] The obtained joint distribution diffusion model is iteratively processed to obtain the corresponding target multi-contrast weighted map. Specifically, as follows... Figure 2 As shown in the backsampling generation process, for the input full noise map, the final Y can be generated through backsampling. In this process, Figure 2 The formula used in the direction sampling generation process is a variation of formula (7).

[0102] Furthermore, the step of iteratively processing the joint distribution diffusion model based on the joint distribution data to obtain the target multi-contrast weighted map and outputting it includes:

[0103] The joint distribution diffusion model is iteratively calculated using a preset algorithm and the joint distribution data.

[0104] When the number of iterations reaches the preset number, the iteration ends, and the target multi-contrast weighted image obtained from the last iteration is output.

[0105] Specifically, the preset algorithm is the multi-contrast joint sampling algorithm, as follows: Figure 3 As shown, firstly, noise σ at N scales is given. i A constant γ, iteration number N, where the number of noise scales is the same as the iteration number, and the total number of calibration steps T; initialization and contrast are performed using Gaussian-distributed noise of the same number. Then, a standard Gaussian noise z following a standard Gaussian distribution is given. The iteration loop is set to iterate from N-1 to 0.

[0106] First, let's define the noise X at the current scale. i+1 The input is fed into the trained network, and then the intermediate value g is obtained according to formula (8). i+1 Formula (8) is specifically expressed as follows:

[0107]

[0108] Among them, g i+1 Indicates the step size. This represents the joint distribution data obtained from the trained convolutional neural network, where i represents the index number, and Y... i+1 It is the same in each loop, only the sequence number changes.

[0109] g i+1 and Multiply, then combine with X i+1 Obtain the iterated X′ i Specifically, this is expressed as formula (9):

[0110]

[0111] X′i By combining the random noise z, the value X corresponding to the intermediate step is obtained. i Specifically, as shown in formula (10):

[0112]

[0113] In each iteration, performing the calibration process provides enhanced performance. Given T iterations, a noise z following a standard Gaussian distribution is first sampled from the Gaussian distribution, and then the calibration step size η is calculated. i The calculation process is shown in formula (11):

[0114]

[0115] Multiply the calculated calibration step size by in It is an intermediate variable in the calibration process; when t=1, equals X i A is the forward encoding matrix of MRI, and y is the measurement signal of multiple contrast images or a single contrast image, i.e., the measurement signal corresponding to multiple contrast images or a single contrast image in the undersampled measurement signal. The calibration signal is obtained through the above.

[0116] This loop continues for T iterations. When t = T, the loop exits. At that time, the outer loop process continues to execute.

[0117] When the number of iterations reaches the preset number, that is, during the iteration cycle from N-1 to 0, the iteration ends when it reaches 0, and the output is a clean multi-contrast image, that is, the target multi-contrast weighted image.

[0118] As can be seen from the above, compared with the existing technology, the current accelerated multi-contrast imaging method based on the diffusion model cannot simultaneously satisfy the problems of generalization and acceleration performance. In this invention, under the Bayesian framework, an accelerated multi-contrast imaging method based on the diffusion model is proposed. Compared with the traditional neural network deep learning method based on black box design, the proposed method considers the problem from a Bayesian perspective, introduces higher-dimensional joint distribution prior information in the full sampling, and has stronger flexibility and generalization. Moreover, the convolutional neural network in the proposed method does not need to be retrained for different sampling templates, which further enhances its generalization.

[0119] Exemplary device

[0120] like Figure 4 As shown, corresponding to the above-described accelerated multi-contrast imaging method based on a diffusion model, this embodiment of the invention also provides an accelerated multi-contrast imaging system based on a diffusion model, the above-described accelerated multi-contrast imaging system based on a diffusion model includes:

[0121] The data acquisition module 41 is used to acquire the undersampled measurement signal in the magnetic resonance, input the undersampled measurement signal into the trained convolutional neural network, and output joint distribution data from the trained convolutional neural network. The trained convolutional neural network is trained according to a preset training process.

[0122] Model building module 42 is used to determine the problem of accelerating multi-contrast image generation based on the undersampled measurement signal, and to build a joint distribution diffusion model of the problem of accelerating multi-contrast image generation using Bayesian method and Langevin dynamics Markov Monte Carlo sampling method.

[0123] Output module 43 is used to iteratively process the joint distribution diffusion model based on the joint distribution data to obtain a target multi-contrast weighted map and output it.

[0124] It should be noted that the specific structure and implementation of the above-mentioned accelerated multi-contrast imaging system based on the diffusion model and its various modules or units can be referred to the corresponding descriptions in the above method embodiments, and will not be repeated here.

[0125] It should be noted that the division of the modules in the above-mentioned diffusion-based accelerated multi-contrast imaging system is not unique and is not intended as a specific limitation.

[0126] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 5 As shown. The terminal includes a processor 10, a memory 20, a network interface, and a display screen 30 connected via a system bus. In one embodiment, when the processor 10 executes the diffusion-based accelerated multi-contrast imaging program 40 in the memory 20, the following steps are implemented:

[0127] The undersampled measurement signal in the magnetic resonance imaging is acquired, and the undersampled measurement signal is input into the trained convolutional neural network. The trained convolutional neural network outputs joint distribution data. The trained convolutional neural network is trained according to a preset training process.

[0128] Based on the undersampled measurement signal, the problem of accelerating multi-contrast image generation is determined, and a joint distribution diffusion model is constructed for the problem of accelerating multi-contrast image generation using Bayesian method and Langevin dynamics Markov Monte Carlo sampling method.

[0129] Based on the joint distribution data, the joint distribution diffusion model is iteratively processed to obtain the target multi-contrast weighted map, which is then output.

[0130] The steps of the preset training process include:

[0131] Acquire a preset number of full sample data, perturb the full sample data according to a preset number of times, and acquire the perturbation data generated after each perturbation of the full sample data during the perturbation process;

[0132] The fully sampled data and the perturbation data are input into the convolutional neural network for training to obtain the trained convolutional neural network.

[0133] The step of inputting the fully sampled data and the perturbation data into the convolutional neural network for training to obtain the trained convolutional neural network includes:

[0134] Obtain the perturbation data corresponding to each full sample data;

[0135] Each fully sampled data point and its corresponding perturbation data are input into the convolutional neural network, and the parameters of the convolutional neural network are optimized according to the optimization parameter loss function.

[0136] When the training iterations of the convolutional neural network reach the preset number, the training is completed, and the trained convolutional neural network is output.

[0137] The step of iteratively processing the joint distribution diffusion model based on the joint distribution data to obtain the target multi-contrast weighted map and outputting it includes:

[0138] The joint distribution diffusion model is iteratively calculated using a preset algorithm and the joint distribution data.

[0139] When the number of iterations reaches the preset number, the iteration ends, and the target multi-contrast weighted image obtained from the last iteration is output.

[0140] The steps of determining the accelerated multi-contrast image generation problem based on the undersampled measurement signal, and constructing a joint distribution diffusion model of the accelerated multi-contrast image generation problem using Bayesian methods and Langevin dynamics Markov Monte Carlo sampling method, include:

[0141] Determine the problem of accelerating multi-contrast image generation based on undersampled measurement signals;

[0142] The problem of accelerating multi-contrast image generation is transformed into a maximum likelihood estimation problem using the Bayesian method.

[0143] The joint distribution diffusion model is constructed based on Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem.

[0144] The step of transforming the accelerated multi-contrast image generation problem into a maximum likelihood estimation problem using the Bayesian method includes:

[0145] The accelerated multi-contrast image is transformed into a preliminary maximum likelihood estimation problem based on the Bayesian framework;

[0146] The initial maximum likelihood estimation problem is expanded using the Bayesian method, and the maximum likelihood estimation problem is obtained through logarithmic processing.

[0147] The steps for constructing the joint distribution diffusion model based on the Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem include:

[0148] The Langevin dynamics Markov Monte Carlo sampling method is applied to the posterior probability in the maximum likelihood estimation problem to obtain a joint distribution diffusion model.

[0149] The terminal's processor provides computing and control capabilities. The terminal's memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and a diffusion-based accelerated multi-contrast imaging program. The internal memory provides the environment for the operation of the operating system and the diffusion-based accelerated multi-contrast imaging program stored in the non-volatile storage medium. The terminal's network interface is used for communication with external terminals via a network connection. When the diffusion-based accelerated multi-contrast imaging program is executed by the processor, it implements the steps of any of the aforementioned diffusion-based accelerated multi-contrast imaging methods. The terminal's display screen can be a liquid crystal display (LCD) or an e-ink display.

[0150] Those skilled in the art will understand that Figure 5 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0151] In one embodiment, a terminal is provided, the terminal including a memory, a processor, and a diffusion-based accelerated multi-contrast imaging program stored in the memory and executable on the processor. When the diffusion-based accelerated multi-contrast imaging program is executed by the processor, it implements the steps of any of the diffusion-based accelerated multi-contrast imaging methods provided in the embodiments of the present invention.

[0152] This invention also provides a computer-readable storage medium storing a diffusion-based accelerated multi-contrast imaging program. When executed by a processor, the diffusion-based accelerated multi-contrast imaging program implements the steps of any of the diffusion-based accelerated multi-contrast imaging methods provided in this invention.

[0153] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0154] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the above device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0155] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0156] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0157] In the embodiments provided by this invention, it should be understood that the disclosed system / terminal device and method can be implemented in other ways. For example, the system / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units described above is merely a logical functional division, and in actual implementation, it can be divided in other ways. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0158] If the integrated modules / units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0159] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not mean that the essence of the corresponding technical solutions deviates from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for accelerating multi-contrast imaging based on a diffusion model, characterized in that, The accelerated multi-contrast imaging method based on the diffusion model includes: The undersampled measurement signal in the magnetic resonance imaging is acquired, and the undersampled measurement signal is input into the trained convolutional neural network. The trained convolutional neural network outputs joint distribution data. The trained convolutional neural network is trained according to a preset training process. Based on the undersampled measurement signal, the problem of accelerating multi-contrast image generation is determined, and a joint distribution diffusion model is constructed for the problem of accelerating multi-contrast image generation using Bayesian method and Langevin dynamics Markov Monte Carlo sampling method. Based on the joint distribution data, the joint distribution diffusion model is iteratively processed to obtain the target multi-contrast weighted map, which is then output.

2. The accelerated multi-contrast imaging method based on a diffusion model according to claim 1, characterized in that, The steps of the preset training process include: Acquire a preset number of full sample data, perturb the full sample data according to a preset number of times, and acquire the perturbation data generated after each perturbation of the full sample data during the perturbation process; The fully sampled data and the perturbation data are input into the convolutional neural network for training to obtain the trained convolutional neural network.

3. The accelerated multi-contrast imaging method based on a diffusion model according to claim 2, characterized in that, The step of inputting the fully sampled data and the perturbation data into the convolutional neural network for training to obtain the trained convolutional neural network includes: Obtain the perturbation data corresponding to each full sample data; Each fully sampled data point and its corresponding perturbation data are input into the convolutional neural network, and the parameters of the convolutional neural network are optimized according to the optimization parameter loss function. When the training iterations of the convolutional neural network reach the preset number, the training is completed, and the trained convolutional neural network is output.

4. The accelerated multi-contrast imaging method based on a diffusion model according to claim 2, characterized in that, The steps of iteratively processing the joint distribution diffusion model based on the joint distribution data to obtain the target multi-contrast weighted map and outputting it include: The joint distribution diffusion model is iteratively calculated using a preset algorithm and the joint distribution data. When the number of iterations reaches the preset number, the iteration ends, and the target multi-contrast weighted image obtained from the last iteration is output.

5. The accelerated multi-contrast imaging method based on a diffusion model according to claim 1, characterized in that, The steps of determining the accelerated multi-contrast image generation problem based on the undersampled measurement signal, and constructing a joint distribution diffusion model for the accelerated multi-contrast image generation problem using Bayesian methods and Langevin dynamics Markov Monte Carlo sampling methods, include: Determine the problem of accelerating multi-contrast image generation based on undersampled measurement signals; The problem of accelerating multi-contrast image generation is transformed into a maximum likelihood estimation problem using the Bayesian method. The joint distribution diffusion model is constructed based on Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem.

6. The accelerated multi-contrast imaging method based on a diffusion model according to claim 4, characterized in that, The step of transforming the accelerated multi-contrast image generation problem into a maximum likelihood estimation problem using a Bayesian method includes: The accelerated multi-contrast image is transformed into a preliminary maximum likelihood estimation problem based on the Bayesian framework; The initial maximum likelihood estimation problem is expanded using the Bayesian method, and the maximum likelihood estimation problem is obtained through logarithmic processing.

7. The accelerated multi-contrast imaging method based on a diffusion model according to claim 4, characterized in that, The steps for constructing the joint distribution diffusion model based on the Langevin dynamics Markov Monte Carlo sampling method and the maximum likelihood estimation problem include: The Langevin dynamics Markov Monte Carlo sampling method is applied to the posterior probability in the maximum likelihood estimation problem to obtain a joint distribution diffusion model.

8. An accelerated multi-contrast imaging system based on a diffusion model, characterized in that, The diffusion-based accelerated multi-contrast imaging system includes: The data acquisition module is used to acquire undersampled measurement signals in magnetic resonance imaging, input the undersampled measurement signals into a trained convolutional neural network, and output joint distribution data from the trained convolutional neural network. The trained convolutional neural network is trained according to a preset training process. The model building module is used to determine the problem of accelerating multi-contrast image generation based on the undersampled measurement signal, and to construct a joint distribution diffusion model of the problem of accelerating multi-contrast image generation using Bayesian method and Langevin dynamics Markov Monte Carlo sampling method. The output module is used to iteratively process the joint distribution diffusion model based on the joint distribution data to obtain the target multi-contrast weighted map and output it.

9. A terminal, characterized in that, The terminal includes a memory, a processor, and a diffusion-based accelerated multi-contrast imaging program stored in the memory and executable on the processor. When executed by the processor, the diffusion-based accelerated multi-contrast imaging program implements the steps of the diffusion-based accelerated multi-contrast imaging method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a diffusion-based accelerated multi-contrast imaging program, which, when executed by a processor, implements the steps of the diffusion-based accelerated multi-contrast imaging method as described in any one of claims 1-7.

Images