Method for optimizing the sampling trajectory in molecular exchange magnetic resonance measurements, molecular exchange magnetic resonance measurement method and apparatus
An integrated network model optimizes molecular exchange magnetic resonance measurement sampling trajectories and parameter estimation, addressing complexity and freedom issues, resulting in improved efficiency and accuracy for real-time monitoring and analysis.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2024-07-01
- Publication Date
- 2026-07-02
AI Technical Summary
Existing molecular exchange magnetic resonance measurement methods face challenges in achieving efficient and accurate parameter estimation due to high complexity and freedom in optimization parameters, limiting their application in real-time monitoring and analysis of complex systems.
A method involving an end-to-end integrated network model comprising a sampling trajectory optimization network, a noise signal generation network, and a parameter estimation network is used to optimize the sampling trajectory, training the network to improve parameter estimation accuracy by setting relevant parameters as trainable and optimizing them directly.
The method provides optimized sampling trajectories and parameter estimation networks, enhancing efficiency and accuracy in molecular exchange magnetic resonance measurements, enabling real-time monitoring and quantitative analysis in various systems.
Smart Images
Figure 2026521969000001_ABST
Abstract
Description
[Technical Field]
[0001] The present invention belongs to the field of quantitative magnetic resonance measurement, and more particularly to a method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement, a molecular exchange magnetic resonance measurement method, and an apparatus. [Background technology]
[0002] Multidimensional nuclear magnetic resonance (NMR) spectroscopy enables comprehensive and detailed analysis of complex molecular systems and is widely used in biological research as a non-invasive technique [Bai R, Benjamini D, Cheng J, et al. Fast, accurate 2D-MR relaxation exchange spectroscopy (REXSY): Beyond compressed sensing[J]. The Journal of chemical physics, 2016, 145(15): 154202]. However, the long acquisition time of multidimensional spectroscopy limits its application to the analysis of fast reaction kinetics under specific physical conditions and to the study of human / animal bodies. For example, diffusion exchange spectroscopy (DEXSY) [Callaghan PT, Furo I. Diffusion-diffusion correlation and exchange as a signature for local order and dynamics[J]. The Journal of Chemical Physics, 2004, 120(8): 4032-4038] measures molecular exchange processes based on differences in the apparent diffusion rates of molecules in different microenvironments or compartments. Its sequence diagram is shown in Figure 2(a) and consists of two independent diffusion encoding modules separated by a time mixing module, forming a three-dimensional acquisition sequence including the first diffusion module - mixing module - second diffusion module. The acquisition variables corresponding to each module are diffusion-weighted b1, exchange / mixing time tm, and diffusion-weighted b2. The DEXSY method is very time-consuming because each signal point needs to be acquired individually at specific first and second diffusion encoding and mixing times.
[0003] Real-time measurement of molecular exchange is extremely important in chemical dynamics and in vivo biological research, and many studies are currently proposing methods to accelerate molecular exchange measurements. Ultrafast nuclear magnetic resonance (UF NMR) spectroscopy utilizes the basic principles of magnetic resonance imaging (MRI) to spatially encode incremental evolution time in each layer of the sample, enabling the acquisition of multidimensional spectra in a single scan [Mankinen O, Zhivonitko VV, Selent A, et al. Ultrafast diffusion exchange nuclear magnetic resonance[J].Nature Communications,2020,11(1):3251]. However, UF NMR techniques have limitations in the sensitivity of results obtained in a single scan, making it impossible to simultaneously achieve short scan times and high resolution [Giraudeau P, Akoka S. Sources of sensitivity losses in ultrafast 2D NMR[J]. Journal of Magnetic Resonance,2008,192(1):151-158]. Therefore, it is difficult to apply this technique to the exploration of complex systems. Another series of studies has proposed reducing scan time by b1-b2 spatial downsampling; for example, a 10% sampling ratio corresponds to a 10-fold speedup. S. Ramadan proposed diffusion-exchange weighted imaging (DEWI) [Ramadan S. Diffusion-exchange weighted imaging[J]. Magnetic Resonance Insights,2009,3:MRI.S3504], but in this technique, the b-values of the first and second diffusion encoding modules must be equal, for example, b1=b2. Nilsson et al. proposed filter exchange spectroscopy. TIFF2026521969000002.tif37144 They proposed the following technique. This technique reduces scan time by setting the b-value of the first diffusion encoding module to a fixed value and reducing the number of scan parameters to be set. Cai et al. proposed the antidiagonal strategy (ANTI) [Williamson NH, Ravin R, Cai TX, et al. Real-time measurement of diffusion exchange rate in biological tissue [J]. Journal of Magnetic Resonance, 2020, 317:106782]. In this sampling strategy, the sum of the b-values of the first and second diffusion encoding modules must be a fixed value. Ordinola et al. proposed a uniform undersampling strategy TIFF2026521969000003.tif31144 We proposed the following sampling strategy: This sampling strategy proposes uniform sampling in the upper triangular region of the b1-b2 space. However, the optimality of these sampling patterns obtained based on heuristic methods cannot be proven.
[0004] In recent years, the advent of deep learning has provided new data-driven optimization techniques for optimizing sampling patterns. Under specific experimental conditions, studies have been published that have successfully optimized frequency-domain spatial sampling patterns for MRI using this technique [Xue S, Cheng Z, Han G, et al. 2D probabilistic undersampling pattern optimization for MR image reconstruction [J]. Medical Image Analysis, 2022, 77:102346]. The quality of magnetic resonance (MR) images reconstructed using this technique is superior to images obtained using other sampling patterns.
[0005] Therefore, methods for further improving the efficiency and accuracy of parameter estimation are currently a hot spot in the research of this field.
Summary of the Invention
Problems to be Solved by the Invention
[0006] An object of the present invention is to provide a method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement. This method can improve the efficiency and accuracy of physiological / physical parameter estimation and is applicable to various molecular exchange magnetic resonance measurement methods.
[0007] The present invention provides the following technical solutions.
[0008] A method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement, the method comprising: Determining the number and corresponding range of physiological / physical parameters of the molecular exchange system, using them as simulation parameters, and performing uniformly random sampling independent of each other within the specified range to generate a simulated physiological / physical parameter data set in step (1); Determining a molecular exchange magnetic resonance measurement method and setting collection parameters in step (2); Constructing an end-to-end integrated network model including a sampling trajectory optimization network, a noise signal generation network, and a parameter estimation network, wherein the sampling trajectory optimization network simulates the signal generation process of the molecular exchange magnetic resonance measurement method and generates a magnetic resonance signal based on the corresponding physiological / physical parameters, sampling trajectory, and collection parameters, the noise signal generation network simulates the noise distribution in the actual magnetic resonance measurement scenario and generates a magnetic resonance signal containing noise, and the parameter estimation network realizes the fitting function of the magnetic resonance signal and outputs physiological / physical parameters based on the magnetic resonance signal containing noise in step (3); Step (4) includes training the end-to-end integrated network model from step (3) with specified acquisition parameters, using the physiological / physical parameters simulated in step (1) as input, and obtaining the optimized sampling trajectory and the corresponding parameter estimation network once the network has converged and training is complete.
[0009] The method for optimizing the sampling trajectory of molecular exchange magnetic resonance (MSR) measurements provided by the present invention mainly consists of two parts: a sampling trajectory optimization network that simulates the signal generation process of the MSRS measurement method, and a parameter estimation network that realizes the magnetic resonance signal fitting process. These two networks are connected by a noise signal generation network, integrating the optimization of the sampling trajectory and parameter estimation into an end-to-end integrated network model. This allows the network to be trained with the direct goal of improving the accuracy of parameter estimation, resulting in an optimized sampling trajectory and parameter estimation network.
[0010] In step (1), to ensure data diversity and methodological generalization, all parameters are independently obtained by uniform random sampling within their respective ranges.
[0011] In this invention, physiological / physical parameters refer to relevant parameters that describe the molecular exchange system.
[0012] The molecular exchange system is a two-compartment exchange system or a multi-compartment exchange system, the magnetic resonance measurement method is a diffusion exchange spectroscopy DEXSY sequence or a relaxation exchange spectroscopy REXSY sequence, and in the two-compartment exchange system, the physiological / physical parameter is the diffusion rate D of the fast component. f , the diffusion rate D of the slow component s , exchange rate K from high-speed component to low-speed component fs , the proportion coefficient f of the fast component in the equilibrium statef eq , and the signal S0 in the case of no spreading encoding gradient.
[0013] Since relaxation exchange spectroscopy (REXSY) sequences are very similar to diffusion exchange spectroscopy (DEXSY) sequences, and the integrated network models constructed are also similar, this invention primarily uses diffusion exchange spectroscopy (DEXSY) sequences as examples for illustrative purposes.
[0014] In step (2), the acquisition parameters include the value and number of exchange times, the number and distribution constraints of sampling trajectory points in a single exchange time, and the signal-to-noise ratio acquired in a single magnetic resonance.
[0015] In step (3), the sampling trajectory optimization network simulates the signal generation process of the DEXSY sequence in the molecular exchange magnetic resonance measurement method, and generates a magnetic resonance signal based on the corresponding physiological / physical parameters, sampling trajectory, and acquisition parameters, the expression of which is as follows:
number
number
[0016] Operator O E describes the signal change process in the mixing module of the DEXSY sequence, and its expression is as follows:
Number
[0017] To ensure that both b1 and b2 are fully optimized within the constraints of the sampling trajectory distribution, the present invention defines as follows in the sampling trajectory optimization network. In step (3), in the sampling trajectory optimization network, b1 and b2 satisfy the following conditions:
Number
[0018] In step (3), the noise signal generation network simulates an actual magnetic resonance scan scenario, adding independently and identically distributed Gaussian noise to the real and imaginary parts of the signal to obtain a magnetic resonance signal containing Rice noise, which is expressed by the following equation:
number
[0019] In step (3), the parameter estimation network is constructed by fully connected layers, comprising one input layer, five hidden layers, and one output layer, wherein the number of neurons in the input layer is equal to the number of magnetic resonance signals, and the number of neurons in the output layer is equal to the number of physiological / physical parameters, and outputs the physiological / physical parameters based on the input magnetic resonance signal containing Rice noise.
number
[0020] Preferably, in order to accelerate the convergence of the network, the values of the number of neurons in the output layer are not used directly as physiological / physical parameters, but rather as scaling values for these parameters. In step (3), the outputted physiological / physical parameters to be estimated are obtained by scaling the maximum-minimum values as shown below.
number
[0021] In step (4), the optimization goal is to improve the accuracy of parameter estimation, and the loss function to be trained is based on the conventional mean squared error and consists of losses for all physiological / physical parameters.
number
number
[0022] Preferably, in step (4), after the training of the integrated network model is completed, the obtained optimized sampling trajectories are discretized into the b1-b2 sampling space, b1 and b2 in the sampling trajectory optimization network are set according to the discretized sampling trajectories and fixed as untrainable parameters, then the parameter estimation network is trained, and when the network converges and training is complete, an optimal parameter estimation network corresponding to the discretized sampling trajectories is obtained.
[0023] Preferably, the method further includes the steps of: (2) adjusting the signal-to-noise ratio acquired by a single magnetic resonance in the acquisition parameters; (3) setting the standard deviation of Gaussian noise in the noise signal network generated in step (3); and (4) training an integrated network model to obtain optimized sampling trajectories and corresponding parameter estimation networks under different signal-to-noise ratios.
[0024] Preferably, the method further includes the steps of: adjusting the number of sampling trajectory points in a single exchange time for the acquisition parameters in step (2); setting the number of sampling points in the b1-b2 space of the sampling trajectory optimization network in step (3); and training the integrated network model in step (4) to obtain optimized sampling trajectories and corresponding parameter estimation networks for different sampling ratios.
[0025] The present invention provides a method for optimizing sampling trajectories in molecular exchange magnetic resonance measurements. Based on a data-driven method, it trains an integrated network model under specific experimental settings to obtain optimized sampling trajectories and their corresponding parameter estimation networks, and then analyzes the distribution characteristics of optimal sampling trajectories at different sampling trajectory point counts and signal-to-noise ratios. This technique has a certain degree of versatility and can be applied to optimizing sampling trajectories in other molecular exchange magnetic resonance measurements.
[0026] The present invention further provides a method for measuring molecular exchange magnetic resonance, and the method is For the molecular exchange system to be measured, the sampling trajectory optimized by the above-described method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement is set as the sampling trajectory for molecular exchange magnetic resonance measurement, and the magnetic resonance signal is collected. The method includes the step of outputting the physiological / physical parameters of the object to be measured based on a corresponding parameter estimation network in a method for optimizing the sampling trajectory of the molecular exchange magnetic resonance measurement described above, using the collected signals as input.
[0027] The present invention further provides a molecular exchange magnetic resonance measurement apparatus, the apparatus comprising a memory and one or more processors, wherein executable code is stored in the memory, and the molecular exchange magnetic resonance measurement method described above is realized when the one or more processors execute the executable code.
[0028] The above-described device requires the processor to be an Intel(R) Xeon(R) CPU @ 3.20 GHz, with 256 GB of memory and one Nvidia Tesla P4 (8GB) GPU.
[0029] Compared to conventional technologies, the present invention has the following technical advantages.
[0030] In the method for optimizing the sampling trajectory of molecular exchange magnetic resonance (MSR) measurements proposed in this invention, the sampling trajectory optimization network that simulates the signal generation process of the MSRS measurement method overcomes the problem in conventional optimization methods where the parameters to be optimized have a high degree of freedom and the model is complex by setting the relevant parameters of the sampling trajectory as trainable parameters. The parameter estimation network for realizing the magnetic resonance signal fitting process proposed in this invention utilizes the advantages of deep learning when solving the inverse problem, improving the efficiency and accuracy of parameter estimation compared to conventional parameter estimation methods. With the integrated network model proposed in this invention, the network can be trained with parameter estimation accuracy as the direct target in a specific MSRS measurement method, and an optimized sampling trajectory and parameter estimation network can be obtained.
[0031] Compared to conventional sampling trajectory and parameter estimation methods, the sampling trajectory and parameter estimation method obtained based on the present invention can provide the most reliable parameters at the same sampling ratio. Therefore, by applying the present invention to molecular exchange magnetic resonance measurement methods, an efficient and accurate method can be obtained for real-time monitoring and quantitative measurement of molecular exchange in different microenvironments or between compartments. Furthermore, this method has a certain degree of versatility and can be applied to various molecular exchange magnetic resonance measurement methods. [Brief explanation of the drawing]
[0032] [Figure 1] Figure 1 shows a flowchart illustrating the application of the method for optimizing the sampling trajectory in molecular exchange magnetic resonance measurements in the example to molecular exchange magnetic resonance measurements. [Figure 2]Figure 2 shows the structure of the integrated network model provided by the present invention: (a) an optimized DEXSY scan sequence of an embodiment, and (b) the network structure proposed by the present invention: an input layer, a DEXSY sequence sampling trajectory optimization network, a noise signal generation network, a parameter estimation network, and an output layer. [Figure 3] Figure 3 shows a two-compartment exchange system based on a cell model. [Figure 4] Figure 4 shows comparative b1-b2 spatial sampling patterns and b1-b2 spatial sampling patterns obtained by optimizing according to the present invention under different experimental conditions. (a) Comparative sampling patterns: FEXSY, DEWI, ANTI, and UNIFORM. (b) Optimization results obtained by repeating 100 experiments with the number of b1-b2 spatial sampling points fixed at 12 and the SNR set to 10, 40, and 400. (c) Optimization results obtained by repeating 100 experiments with the SNR fixed at 40 and the number of b1-b2 spatial sampling points set to 3, 6, 12, and 18. [Figure 5] Figure 5 shows a set of sampling patterns obtained by optimizing with an SNR of 40 and 12 spatial sampling points in the b1-b2 space. [Figure 6] Figure 6 shows the results of parameter error estimation based on different sampling patterns under different SNRs and parameter estimation methods. [Figure 7] Figure 7 shows the reproducibility of parameter estimation under different sampling patterns: (a) reproducibility of parameter estimation based on simulation data, and (b) reproducibility of parameter estimation based on real data. [Modes for carrying out the invention]
[0033] To further clarify the object, technical solutions, and advantages of the present invention, the invention will be described in more detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are used solely for illustrative purposes and are not intended to limit the scope of protection of the invention.
[0034] The model proposed in this invention is implemented based on the PyTorch framework.
[0035] Figure 1 shows a flowchart illustrating the application of the method for optimizing the sampling trajectory in molecular exchange magnetic resonance measurements in the examples to molecular exchange magnetic resonance measurements, and mainly includes the following steps.
[0036] (1) To optimize the sampling trajectory for molecular exchange magnetic resonance measurements, the physiological / physical parameters and corresponding ranges of the molecular exchange system to be measured are determined and used as simulation parameters. Within the specified range, mutually independent, uniform random sampling is performed to generate a certain number of simulated physiological / physical parameters, which are then randomly divided into three parts: a training set, a validation set, and a test set.
[0037] In this example, a two-compartment exchange system is used as the experimental model, as shown in Figure 3. This system includes two components: fast (f) and slow (s). The physiological parameters / molecular exchange-related parameters to be measured in this system and their reasonable ranges are as follows: The diffusion rate of the fast component is D f ∈[1.5,2.0]×10 -3 mm 2 The diffusion rate of the slow component is D s ∈[0.03,0.06]×10 -3 mm 2 The speed of exchange from the high-speed component to the low-speed component is k. fs ∈[0,10] s -1 The ratio coefficient f of the fast component in the equilibrium state is such that f eq The signal S0 is ∈[0.05,0.5] and the signal S0 ∈[500,1500] when there is no spreading encoding gradient.
[0038] A simulation dataset is generated by performing mutually independent, uniformly random sampling based on a specified range for each physiological parameter. From this dataset, 1,000,000 sets are selected as the training set, 10,000 sets as the validation set, and 10,000 sets as the test set.
[0039] (2) In order to optimize the sampling trajectory of molecular exchange magnetic resonance measurements, the relevant experimental settings in the molecular exchange magnetic resonance measurement experiment are determined (or the molecular exchange magnetic resonance measurement method is determined and the acquisition parameters are set), including settings such as the value and number of exchange times, constraints on the number and distribution of sampling trajectory points in a single exchange time, and the signal-to-noise ratio collected by a single magnetic resonance.
[0040] In this embodiment, the DEXSY sequence is used as the molecular exchange magnetic resonance measurement method. The acquisition parameters include the following, and the exchange time t m The time interval is fixed at [10,100,200,350]ms, the number of sampling trajectory points in a single exchange time is 12, the sampling space is the b1-b2 space, and b min ≤ b1, b2 ≤ b max Satisfying the condition, where b min = 100 s / mm 2 and b max = 1200 s / mm 2 Therefore, the sampling interval for the discrete b1-b2 space is 100 s / mm. 2 The signal-to-noise ratio collected by a single magnetic resonance is 40.
[0041] (3) Construct a sampling trajectory optimization network to simulate the signal generation process of molecular exchange magnetic resonance measurement, and generate a magnetic resonance signal based on the corresponding physiological / physical parameters, sampling trajectory, and experimental conditions (collection parameters).
[0042] In this example, the DEXSY sequence is used as the molecular exchange magnetic resonance measurement method. The equation for generating the magnetic resonance signal is as follows:
number
number
number
[0043] Operator O E This describes the signal change process in the mixing module, and its expression is as follows:
number
[0044] (4) Construct a noise signal generation network to simulate the noise distribution in an actual magnetic resonance experiment scenario, and take the magnetic resonance signal output by the sampling trajectory optimization network in step (3) as input, add independently identically distributed Gaussian noise to the real and imaginary parts of the signal, respectively, and output a signal containing Rice noise.
number
[0045] (5) Construct a parameter estimation network and implement a magnetic resonance signal fitting function. The signal containing the Rice noise generated in step (4) is used as input to the network, and the physiological / physical parameter values to be estimated are output.
number
[0046] Specifically, in this embodiment, a parameter estimation network is constructed using only fully connected layers, and the network includes one input layer, five hidden layers, and one output layer. The number of neurons in the input layer is equal to the number of acquired signals, and in this embodiment, it is equal to the product of the number of exchange times and the number of sampling trajectory points in a single exchange time. According to the experimental setup in step (2), the number of exchange times is 4 and the number of sampling trajectory points in a single exchange time is 12, so the number of neurons in the input layer is set to 48. The number of neurons in the five hidden layers is set to 128, 256, 512, 256, and 128, respectively, and each layer is followed by an Exponential Linear Unit (ELU) activation function. The number of neurons in the output layer is equal to the number of molecular exchange-related parameters / physiological parameters to be estimated, and in this embodiment, according to the experimental model described in step (1), the estimated physiological / physical parameters TIFF2026521969000036.tif7144 , TIFF2026521969000037.tif7144 , TIFF2026521969000038.tif7144 , TIFF2026521969000039.tif13144 and TIFF2026521969000040.tif7144 It is set to 5, corresponding to each of them.
[0047] (6) Integrate the three network segments obtained in steps (3) to (5) into an end-to-end integrated network. Train the network with the specified acquisition parameters, based on the simulation dataset obtained in step (1), until the network converges and an optimized sampling trajectory is obtained.
[0048] A computer program is created to implement the three networks in steps (3), (4), and (5), and an end-to-end integrated network model is constructed as shown in Figure 2(b). The network is trained with parameter estimation accuracy as the direct goal for a specific molecular exchange magnetic resonance measurement method, and an optimized b1-b2 spatial sampling trajectory and parameter estimation network are obtained. The loss function to be trained consists of the loss of all physiological parameters based on the mean squared error of the parameter estimates.
number
number
[0049] (7) In accordance with the distribution constraints of the sampling trajectories in step (2), the sampling trajectories optimized in step (6) are discretized into the b1-b2 sampling space so as to satisfy the sampling interval constraints. Depending on the discretized sampling trajectories, b1 and b2 in the sampling trajectory optimization network are set and fixed as parameters that cannot be trained. Based on the simulation dataset from step (1), the integrated network model is trained until the network converges to obtain the optimal parameter estimation network corresponding to the discretized sampling trajectories.
[0050] To demonstrate the superior performance of sampling trajectories optimized based on the data-driven method proposed in this invention, four b1-b2 spatial sampling trajectories—FEXSY, DEWI, ANTI, and UNIFORM—were compared (see Figure 4(a)). To ensure fairness in the comparison, b1 and b2 in the corresponding sampling trajectory optimization network were set according to the different sampling trajectories, fixed as non-trainable parameters, and trained with the same experimental parameters to obtain the corresponding parameter estimation network. The performance of the sampling trajectories was evaluated in terms of accuracy and reproducibility on the simulation dataset, and further, the reproducibility of the sampling trajectories was evaluated using real data.
[0051] As shown in Figure 6, the test results of parameter estimation accuracy in simulation datasets using different signal-to-noise ratios and different parameter estimation methods are presented. These results demonstrate that the sampling trajectory proposed in this invention can significantly improve parameter estimation accuracy, whether based on a parameter estimation network or on conventional nonlinear least squares model fitting.
[0052] Figure 7 shows the results of reproducibility experiments using simulation data and real data. As can be seen from Figure 7, the present invention demonstrates optimal parameter reproducibility in both datasets, demonstrating its stability and proving that the sampling trajectory optimized based on the present invention has a certain degree of universality.
[0053] This invention qualitatively analyzes the relationship between the optimal sampling trajectory and the signal-to-noise ratio (SNR) and the number of sampling trajectory points in a single exchange time, by setting the SNR and the number of sampling trajectory points to constant values and varying another parameter to analyze the optimal sampling trajectory and its relationship to them. The specific procedure is as follows.
[0054] (8) The number of sampling trajectory points in the b1-b2 sampling space of the sampling trajectory optimization network from step (3) is fixed to 12, and the SNR of the noise signal generation network from step (4) is set to 10, 40, and 400, respectively. Step (6) is repeated to train the network 100 times and obtain optimized b1-b2 space sampling trajectories at different signal-to-noise ratios.
[0055] (9) The SNR of the noise signal generation network in step (4) is fixed at 40, and the number of sampling trajectory points in the b1-b2 sampling space of the sampling trajectory optimization network in step (3) is set to 3, 6, 12, and 18, respectively. Step (6) is repeated to train the network 100 times and obtain optimized b1-b2 space sampling trajectories for different sampling ratios.
[0056] (9) Qualitatively analyze the optimal sampling trajectory patterns and characteristics corresponding to different sampling ratios and signal-to-noise ratios.
[0057] For specific SNR and sampling ratios, repeat the training of the network 100 times based on step (6). Since training a deep neural network is a random process, different network weights may be obtained for the same data, and to obtain statistically significant results, repeat the training 100 times with the same experimental parameter settings.
[0058] The distribution of optimal sampling trajectories at different SNRs obtained in step (8) is quite regular. When the SNR is low, the sampling trajectory tends to improve the signal-to-noise ratio by repeatedly sampling in important regions. When the SNR is relatively high, the sampling trajectory tends to perform uniform sampling within the b1-b2 space in order to obtain more information.
[0059] The distribution of optimal sampling trajectories at different sampling ratios obtained in step (9) indicates that each sampling point in the b1-b2 space has a different importance.
[0060] The results are shown in Figures 4(b) and (c), respectively. As shown in Figure 4(b), when the number of sampling trajectory points is fixed, the optimal sampling trajectory is affected by the SNR. When the SNR is relatively low, the sampling trajectory tends to repeat sampling in certain important regions, reducing the impact of noise on the signal. When the SNR is relatively high, the optimal sampling trajectory tends to be evenly distributed to increase signal diversity. However, since the high b-value region is more affected by noise than the low b-value region, the sampling density remains relatively high in the high b-value region. As can be seen from Figure 4(c), when the SNR is fixed, the optimal sampling trajectory is affected by the sampling ratio. When the sampling ratio is relatively low, the sampling trajectory shows a distribution mainly similar to ANTI, and as the sampling ratio gradually increases, the sampling trajectory gradually expands along the diagonal of the b1-b2 sampling space and the optimization result at low sampling ratios. As can be seen from the above two results, it is necessary to design the optimal sampling trajectory for the specific experimental environment.
[0061] The apparatus for optimizing the sampling trajectory of molecular exchange magnetic resonance measurements provided in this embodiment includes one or more processors, and executable code is stored in memory. When the processors execute the executable code, the method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurements of the embodiment is realized. Taking a software implementation as an example, the logic device is formed when the processor of any device having data processing capabilities on which it is located reads corresponding computer program instructions from non-volatile memory into memory and executes them. From a hardware perspective, in addition to the processor, memory, network interface, and non-volatile memory, the embodiment may include other hardware depending on the actual functionality of the device having data processing capabilities, but this will not be described in further detail.
[0062] The foregoing describes only preferred specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any modifications or substitutions that a person skilled in the art can easily conceive of within the scope of the art disclosed herein shall be included in the scope of protection of the present application. Accordingly, the scope of protection of the present application should be subject to the scope of protection of the claims.
Claims
1. A method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement, wherein the method is: Step (1) determines the number and corresponding ranges of physiological / physical parameters of the molecular exchange system, uses them as simulation parameters, and generates a simulated physiological / physical parameter dataset by performing mutually independent, uniform random sampling within the specified range. Step (2) is to determine the molecular exchange magnetic resonance measurement method and set the acquisition parameters, Step (3) is to construct an end-to-end integrated network model including a sampling trajectory optimization network, a noise signal generation network, and a parameter estimation network, wherein the sampling trajectory optimization network simulates the signal generation process of a molecular exchange magnetic resonance measurement method and generates a magnetic resonance signal based on corresponding physiological / physical parameters, sampling trajectory, and acquisition parameters; the noise signal generation network simulates the noise distribution in an actual magnetic resonance measurement scenario and generates a magnetic resonance signal including noise; and the parameter estimation network implements a magnetic resonance signal fitting function and outputs physiological / physical parameters based on the magnetic resonance signal including noise. A method for optimizing a sampling trajectory for molecular exchange magnetic resonance measurements, comprising step (4) training the end-to-end integrated network model of step (3) with the physiological / physical parameters simulated in step (1) as input under specified acquisition parameters, and obtaining an optimized sampling trajectory and a corresponding parameter estimation network once the network has converged and training is complete.
2. The molecular exchange system is a two-compartment exchange system or a multi-compartment exchange system, the magnetic resonance measurement method is a diffusion exchange spectroscopy DEXSY sequence or a relaxation exchange spectroscopy REXSY sequence, and in the two-compartment exchange system, the physiological / physical parameter is the diffusion velocity D of the fast component. ｆ , the diffusion velocity D of the slow component ｓ , exchange rate K from high-speed component to low-speed component ｆｓ , the proportion coefficient f of the fast component in the equilibrium state ｆ ｅｑ , and the signal S in the case of no spreading encoding gradient ０ A method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement according to claim 1, characterized in that it is the method described in claim 1.
3. Step (2) is characterized in that the sampling trajectory for molecular exchange magnetic resonance measurement according to claim 1 includes the value and number of exchange times, the number and distribution constraints of sampling trajectory points in a single exchange time, and the signal-to-noise ratio collected in a single magnetic resonance.
4. In step (3), the sampling trajectory optimization network simulates the signal generation process of the DEXSY sequence in molecular exchange magnetic resonance (MSR) measurement, and generates a magnetic resonance signal based on the corresponding physiological / physical parameters, sampling trajectory, and acquisition parameters, the expression of which is as follows: Here, I = (1, 1), 、 And, Operator O Ｄ１ and O Ｄ２ This represents the signal change process in the first and second spreading encoding modules of the DEXSY sequence, and its expression is as follows: Here, is a diffusion matrix, b １ and b ２ respectively represent the b-values in the first and second diffusion encoding modules, and the sampling points b １ and b ２ The number of is set according to the number of sampling trajectory points at a single exchange time in step (2), and the sampling points b １ and b ２ The values of are related parameters of the sampling trajectory to be optimized, and thus are set as trainable parameters and satisfy the following conditions: Here, the range of values for the variable v is [-∞, +∞], σ represents the sigmoid activation function, and b ｍａｘ and b ｍｉｎ The terms b represent the distribution constraints of the sampling trajectory, respectively. １ and b ２ These are the maximum and minimum values within the range, Operator O Ｅ This describes the signal change process in the DEXSY sequence exchange module, and its expression is as follows: Here, t ｍ This is the exchange time, This is the commutation matrix, and This is a vertical relaxation matrix, R １，ｓ（ｆ） = 1 / T １，ｓ（ｆ） and T １，ｓ（ｆ） The method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement according to claim 3, characterized in that is the longitudinal relaxation time.
5. In step (3), the noise signal generation network simulates an actual magnetic resonance scan scenario, adding independently and identically distributed Gaussian noise to the real and imaginary parts of the magnetic resonance signal to obtain a magnetic resonance signal containing Rice noise, which is expressed by the following equation: Here, n ｒｅ and n ｉｍ These represent independent and identically distributed Gaussian noise added to the real and imaginary parts, respectively, and the standard deviation of this Gaussian noise is σ = S ０ / SNR, where SNR is the signal-to-noise ratio, S ＤＥＸＳＹ This is the noise-free magnetic resonance signal output by the sampling trajectory optimization network in step (3), The method for optimizing the sampling trajectory of a molecular exchange magnetic resonance measurement according to claim 4, characterized in that it represents a magnetic resonance signal including Rice noise.
6. In step (3), the parameter estimation network is constructed of fully connected layers including one input layer, five hidden layers, and one output layer, the number of neurons in the input layer being equal to the number of magnetic resonance signals, and the number of neurons in the output layer being equal to the number of physiological / physical parameters, and outputs physiological / physical parameters based on the input magnetic resonance signal including Rice noise. Here, represents the estimated physiological parameter, f ＤＮＮ represents the parameter estimation network, and ( is f ＤＮＮ A method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement according to claim 5, characterized in that it represents the network parameters to be optimized.
7. In step (3), the output physiological and physical parameters are obtained by scaling the maximum-minimum values and are expressed as follows: Here, P ｍａｘ and P ｍｉｎ These represent the upper and lower limits of the physiological / physical parameter range, respectively, and Output ｐ The parameter estimation network in a method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement according to claim 1, characterized in that it represents the value of the number of neurons for the corresponding parameter P in the output layer.
8. In step (4), the loss function trained with the optimization goal of improving the accuracy of parameter estimation is based on the conventional mean squared error and consists of losses for all physiological / physical parameters. Here, the loss of a single physiological parameter is defined as follows: Here, P represents the actual value of the physiological parameter. The method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement according to claim 1, characterized in that represents the physiological parameters estimated by the network, and N represents the number of physiological parameters P.
9. In step (4), after the training of the integrated network model is complete, the obtained optimized sampling trajectory is b １ -b ２ Discretize into a sampling space, and in the sampling trajectory optimization network, b １ and b ２ A method for optimizing the sampling trajectory of molecular exchange magnetic resonance measurement according to claim 1, characterized in that a parameter is set according to the discretized sampling trajectory, fixed as an untrainable parameter, then a parameter estimation network is trained, and when the network converges and training is complete, a parameter estimation network corresponding to the discretized sampling trajectory is obtained.
10. The method for optimizing a sampling trajectory for a molecular exchange magnetic resonance measurement according to claim 1, further comprising the steps of: (2) adjusting the signal-to-noise ratio acquired by a single magnetic resonance in the acquisition parameters; (3) setting the standard deviation of Gaussian noise in the noise signal network generated in step (3); and (4) training an integrated network model to obtain an optimized sampling trajectory and a corresponding parameter estimation network under different signal-to-noise ratios.
11. The method involves, in step (2), adjusting the number of sampling trajectory points in a single exchange time for the acquisition parameters, and in step (3), the b of the sampling trajectory optimization network. １ -b ２ A method for optimizing a sampling trajectory for molecular exchange magnetic resonance measurements according to claim 1, further comprising the steps of setting the number of sampling points in space, training an integrated network model in step (4) to obtain optimized sampling trajectories and corresponding parameter estimation networks for different sampling ratios.
12. A method for measuring molecular exchange magnetic resonance, wherein the method is The steps include setting the optimized sampling trajectory described in any one of claims 1 to 11 as the sampling trajectory in molecular exchange magnetic resonance measurement for the molecular exchange system to be measured, and collecting the magnetic resonance signal, A molecular exchange magnetic resonance measurement method characterized by comprising the step of using the collected signal as input and outputting a physiological / physical parameter to be measured based on the corresponding parameter estimation network described in any one of claims 1 to 11.
13. A molecular exchange magnetic resonance (MEL) measurement apparatus, wherein the apparatus includes a memory and one or more processors, the memory stores executable code, and when the one or more processors execute the executable code, the molecular exchange magnetic resonance measurement method described in claim 12 is realized.