Nuclear magnetic resonance pure chemical shift data optimization method and device based on deep learning

Abstract

The application discloses a kind of nuclear magnetic resonance pure chemical shift data optimization method and device based on deep learning, comprising: constructing nuclear magnetic resonance pure chemical shift data optimization model, first training data and second training data, using first training data to pre-train nuclear magnetic resonance pure chemical shift data optimization model, obtain the pre-trained nuclear magnetic resonance pure chemical shift data optimization model, using second training data to fine-tune the pre-trained nuclear magnetic resonance pure chemical shift data optimization model, obtain the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model;Extract FID signal in the nuclear magnetic resonance pure chemical shift data before optimization and carry out Fourier transform, obtain corresponding Fourier transform result and input fine-tuned nuclear magnetic resonance pure chemical shift data optimization model, obtain the frequency domain signal of optimized nuclear magnetic resonance pure chemical shift data.The present application solves the problems of insufficient generalization and low precision of existing deep learning model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of nuclear magnetic resonance spectroscopy detection, and in particular to a nuclear magnetic resonance pure chemical shift data optimization method and device based on deep learning. BACKGROUND

[0002] Nuclear magnetic resonance technology (NMR) is a method of analyzing molecular structure and dynamics, and due to its non-invasive nature, it is also widely used in the fields of biology, medicine and chemical analysis. For conventional nuclear magnetic resonance spectroscopy, researchers only need to perform a simple Fourier transform operation on the FID signal collected from the magnetic resonance spectrometer to obtain the required spectrum. However, due to the limitations of conventional magnetic resonance spectroscopy methods, such as high peak overlap rate of one-dimensional spectrum and low sampling efficiency of multi-dimensional spectrum, people have proposed pure chemical shift technology to improve the resolution of the spectrum. The pure chemical shift magnetic resonance spectroscopy method selectively inverts the spin nuclei at a specific frequency, thereby realizing the re-aggregation of scalar coupling evolution, simplifying the multiple peaks in the conventional spectrum due to scalar coupling splitting into single peaks, and improving the resolution of the spectrum. However, after the pure chemical shift magnetic resonance spectroscopy method is processed, the modulation of the signal caused by residual scalar coupling will cause some periodic pseudo-peaks in the spectrum, which are called "block sideband pseudo-peaks". In addition, due to the long experimental time of high-dimensional magnetic resonance spectroscopy, researchers usually undersample the signal to accelerate the experiment, and then use the sparsity and low rank of the magnetic resonance spectroscopy signal as prior attributes to restore high-quality magnetic resonance spectroscopy through corresponding signal reconstruction methods. Since undersampling will bring undersampling artifacts and reduce the signal-to-noise ratio of the magnetic resonance spectrum, removing the block sideband pseudo-peaks in the process of reconstructing the spectrum becomes a more complex task.

[0003] The current mainstream methods for removing such pseudo-peaks include traditional algorithms and deep learning-based methods. Traditional algorithms rely on complex mathematical models and optimization algorithms, and generally have problems such as sensitivity to noise, large computational load, long time consumption, and excessive dependence on hand-designed mathematical models. Deep learning-based methods use a large number of simulated pure chemical shift spectrum data to train neural networks, but the actual experimental conditions are complex and varied, and the simulation data set can only represent a small class of experimental data distribution. When the trained network is used to process different types of spectra under different experimental conditions, it often has the problem of insufficient generalization ability, such as incomplete pseudo-peak removal or small peaks being missed when the experimental signal length or undersampling template is different from the training data set. SUMMARY

[0004] The present application aims to solve the problems of insufficient generalization and low precision of existing deep learning algorithms mentioned above, and proposes a nuclear magnetic resonance pure chemical shift data optimization method and device based on deep learning.

[0005] In a first aspect, the present application provides a deep learning-based nuclear magnetic resonance pure chemical shift data optimization method, comprising the following steps:

[0006] Obtaining nuclear magnetic resonance pure chemical shift data before optimization;

[0007] Constructing a nuclear magnetic resonance pure chemical shift data optimization model, first training data and second training data, the nuclear magnetic resonance pure chemical shift data optimization model comprising a first convolutional module, a plurality of residual modules, a first convolutional layer and a J coupling modulation simulation module connected in sequence, the residual module comprising two second convolutional modules connected in sequence, the output feature of the first second convolutional module being added to the output feature of the second second convolutional module to obtain the output feature of the residual module;

[0008] Pre-training the nuclear magnetic resonance pure chemical shift data optimization model using the first training data to obtain a pre-trained nuclear magnetic resonance pure chemical shift data optimization model, and fine-tuning the pre-trained nuclear magnetic resonance pure chemical shift data optimization model using the second training data to obtain a fine-tuned nuclear magnetic resonance pure chemical shift data optimization model;

[0009] Extracting the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization and performing Fourier transform to obtain the corresponding Fourier transform result, inputting the Fourier transform result of the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization into the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model, inputting the Fourier transform result into the first convolutional module to obtain first features, inputting the features into the plurality of residual modules in sequence to obtain second features, inputting the second features into the first convolutional layer to obtain frequency domain features, and inputting the frequency domain features into the J coupling modulation simulation module to obtain the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data.

[0010] As a preferred, the frequency domain features are input into the J coupling modulation simulation module, first converted to the time domain through inverse Fourier transform, then multiplied point by point with the equivalent envelope signal f(t) of J coupling modulation to generate time domain features containing simulated J coupling modulation effects, and then the time domain features are converted to the frequency domain through Fourier transform to output the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data.

[0011] As a preferred, in the pre-training process of the nuclear magnetic resonance pure chemical shift data optimization model, the equivalent envelope signal f(t) in the J coupling modulation simulation module is fixed, and the parameters of the first convolutional module, the plurality of residual modules and the first convolutional layer are adjusted; at this time, the equivalent envelope signal f(t) is set to an initial value f input (t), and the expression is as follows:

[0012]

[0013] wherein J input represents the equivalent factor of J coupling modulation, τ represents the duration of each data block in the pseudo two-dimensional sampling mode, and is described by the following formula:

[0014]

[0015] wherein SW1 represents the indirect dimension spectral width of the pseudo two-dimensional sampling, t2 represents the evolution time within each data block, and is calculated from the coordinate time t of the FID signal of the pre-optimized magnetic resonance pure chemical shift spectral data, as shown in the following formula:

[0016] t2 = (t mod τ);

[0017] wherein "mod" represents the modulo operation, the sampling interval of the time coordinate t is 1 / SW, and SW represents the direct dimension spectral width of the pseudo two-dimensional sampling;

[0018] In the fine-tuning process of the pre-trained magnetic resonance pure chemical shift data optimization model, the parameters of the equivalent envelope signal f(t) in the J coupling modulation simulation module, the first convolution module, the plurality of residual modules and the first convolution layer are adjusted, and the equivalent envelope signal f(t) in the J coupling modulation simulation module is adjusted from the initial value f input (t).

[0019] As preferred, the construction process of the first training data is as follows:

[0020] The Fourier transform result and the label of the FID signal of the pre-optimized magnetic resonance pure chemical shift data in the first training data are generated by using a mathematical model based on Fourier transform and a specific attenuation function;

[0021] The mathematical model is represented as:

[0022]

[0023] wherein s ideal (t) is the constructed ideal FID signal of the magnetic resonance pure chemical shift data, s origin (t) is the constructed pre-optimized FID signal of the magnetic resonance pure chemical shift data, N is the total number of spectral peak components in a single sample, and for each sample, N is set to a random integer in the interval of 7 to 10; A i , F i , T 2i , and J i respectively represent the amplitude, frequency, transverse relaxation time and scalar coupling constant of the i-th spectral peak, and for each spectral peak component, A i is set to a random number in the interval of 1 to 30, F i is set to a random number in the interval of 80 to 2048, and T2i Set to a random number between 0.08 and 0.6, J i Set to a random number between 2 and 12; t1 represents the indirect dimension evolution time, j represents the imaginary unit, n(t) represents Gaussian white noise, and NUS(t) represents the sampling template of the Poisson exponential distribution segment NUS. The sampling template is shown in the following formula:

[0024]

[0025] Where t′ p Indicates the coordinate time corresponding to the p-th sampling point, where p is the sampling point subscript, generated according to Poisson distribution, and the total number of sampling points in each sampling segment is set to 128;

[0026] The FID signal s of the ideal pure chemical shift data of NMR will be constructed origin The Fourier transform result of (t) is used as the label;

[0027] The construction process of the second training data is as follows:

[0028] Acquire the pre-optimization pure chemical shift data collected experimentally, extract the FID signal from the pre-optimization pure chemical shift data, and perform Fourier transform to obtain a corresponding Fourier transform result, which is the Fourier transform result of the FID signal of the pre-optimization pure chemical shift data of the NMR in the second training data.

[0029] Preferably, the loss function used in the pre-training process of the NMR pure chemical shift data optimization model is a mean square error loss function between the frequency domain signal of the optimized NMR pure chemical shift data and the label obtained by inputting the Fourier transform result of the FID signal of the NMR pure chemical shift data before optimization in the second training data into the NMR pure chemical shift data optimization model;

[0030] The loss function used in the pre-training process of the NMR pure chemical shift data optimization model is:

[0031]

[0032] in,‖·‖ F represents the F norm, ‖·‖1 represents the L1 norm, λ represents the regularization parameter, s′ origin (t) represents the FID signal of the pure chemical shift data of nuclear magnetic resonance before optimization in the second training data, s′ predict(t) represents the Fourier transform result of the FID signal of the optimized nuclear magnetic resonance pure chemical shift data in the second training data, and FT(·) represents the Fourier transform function.

[0033] Preferably, the first convolutional layer has a kernel size of 1x1; the first convolutional module comprises a second convolutional layer and a first LeakyReLU activation function layer connected in sequence, and the second convolutional layer has a kernel size of 15x1;

[0034] The number of residual modules is 7, and the calculation process of the residual module is as follows:

[0035] Y k =CONV k1 (X k )+CONV k2 (CONV k1 (X k ));

[0036] Wherein, X k represents the input feature of the kth residual module, CONV k1 (·) represents the output feature of the first second convolutional module in the kth residual module, CONV k2 (·) represents the output feature of the second second convolutional module in the kth residual module, and Y k represents the output feature of the kth residual module.

[0037] Both the two second convolutional modules comprise a third convolutional layer and a second LeakyReLU activation function layer connected in sequence, and the third convolutional layer has a kernel size of 29x1.

[0038] In a second aspect, the present application provides a nuclear magnetic resonance pure chemical shift data optimization device based on deep learning, comprising:

[0039] The data acquisition module is configured to acquire the nuclear magnetic resonance pure chemical shift data before optimization.

[0040] The model construction module is configured to construct the nuclear magnetic resonance pure chemical shift data optimization model, the first training data and the second training data, wherein the nuclear magnetic resonance pure chemical shift data optimization model comprises a first convolutional module, a plurality of residual modules, a first convolutional layer and a J coupling modulation simulation module connected in sequence, the residual module comprises two second convolutional modules connected in sequence, and the output feature of the first second convolutional module is added to the output feature of the second second convolutional module to obtain the output feature of the residual module.

[0041] The training fine-tuning module is configured to pre-train the nuclear magnetic resonance pure chemical shift data optimization model by using first training data to obtain a pre-trained nuclear magnetic resonance pure chemical shift data optimization model, and fine-tune the pre-trained nuclear magnetic resonance pure chemical shift data optimization model by using second training data to obtain a fine-tuned nuclear magnetic resonance pure chemical shift data optimization model.

[0042] The optimization module is configured to extract FID signals in the pre-optimized nuclear magnetic resonance pure chemical shift data and perform Fourier transform to obtain corresponding Fourier transform results, input the Fourier transform results of the FID signals in the pre-optimized nuclear magnetic resonance pure chemical shift data into the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model, input the Fourier transform results into the first convolution module to obtain first features, input the features through a plurality of residual modules in sequence to obtain second features, input the second features into the first convolution layer to obtain frequency domain features, and input the frequency domain features into the J coupling modulation simulation module to obtain frequency domain signals of the optimized nuclear magnetic resonance pure chemical shift data.

[0043] In a third aspect, the present application provides an electronic device, comprising one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the implementation manners of the first aspect.

[0044] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, when the computer program is executed by a processor, the method described in any of the implementation manners of the first aspect is implemented.

[0045] In a fifth aspect, the present application provides a computer program product comprising a computer program, when the computer program is executed by a processor, the method described in any of the implementation manners of the first aspect is implemented.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] (1) The nuclear magnetic resonance pure chemical shift data optimization method based on deep learning provided by the present application is pre-trained in the first stage, the parameters of the nuclear magnetic resonance pure chemical shift data optimization model reach a good initial value, and the nuclear magnetic resonance pure chemical shift data optimization model has the ability to preliminarily optimize pure chemical shift experimental data, including removing undersampling artifacts and residual J coupling modulation generated pseudo-peaks to a certain extent, which makes the single-sample customized optimization in the second stage be completed without too many iteration times, improves the algorithm efficiency, balances the algorithm efficiency and the accuracy for single samples, and is especially suitable for efficient processing and optimization of data under complex experimental conditions.

[0048] (2) The deep learning-based nuclear magnetic resonance pure chemical shift data optimization method proposed in the present application performs unsupervised training on the experimental data to be optimized in the fine-tuning process of the second stage, that is, further customized training on individual data based on the results of the first stage network, and the training process is guided by a physical model without the need for an ideal spectrum as a reference. This scheme has stronger generalization and can obtain accurate reconstruction results when used for experimental data collected under different experimental conditions.

[0049] (3) The deep learning-based nuclear magnetic resonance pure chemical shift data optimization method proposed in the present application can complete the undersampling experimental signal points for fast sampling data and accurately remove the periodic pseudo-peak caused by residual J coupling modulation, has fast processing speed, high peak reconstruction accuracy, and strong robustness. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0051] Figure 1 The flowchart of the deep learning-based nuclear magnetic resonance pure chemical shift data optimization method of the embodiments of the present application;

[0052] Figure 2 The structure diagram of the nuclear magnetic resonance pure chemical shift data optimization model of the deep learning-based nuclear magnetic resonance pure chemical shift data optimization method of the embodiments of the present application;

[0053] Figure 3 The schematic diagram of the training process of the nuclear magnetic resonance pure chemical shift data optimization model of the deep learning-based nuclear magnetic resonance pure chemical shift data optimization method of the embodiments of the present application;

[0054] Figure 4 The optimized nuclear magnetic resonance pure chemical shift spectrum obtained by processing the full-sampling nuclear magnetic resonance pure chemical shift data by the pre-trained nuclear magnetic resonance pure chemical shift data optimization model obtained through the first stage pre-training process in the deep learning-based nuclear magnetic resonance pure chemical shift data optimization method of the embodiments of the present application;

[0055] Figure 5 The optimized nuclear magnetic resonance pure chemical shift spectrum obtained by processing the full-sampling nuclear magnetic resonance pure chemical shift data by the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model obtained through the second stage fine-tuning process in the deep learning-based nuclear magnetic resonance pure chemical shift data optimization method of the embodiments of the present application;

[0056] Figure 6 The pre-optimized nuclear magnetic resonance pure chemical shift data optimization model obtained through the first stage pre-training process in the deep learning-based nuclear magnetic resonance pure chemical shift data optimization method of the embodiments of the present application is used to process the undersampled nuclear magnetic resonance pure chemical shift data to obtain the optimized nuclear magnetic resonance pure chemical shift spectrum.

[0057] Figure 7 The fine-tuned nuclear magnetic resonance pure chemical shift data optimization model obtained through the second stage fine-tuning process in the deep learning-based nuclear magnetic resonance pure chemical shift data optimization method of the embodiments of the present application is used to process the undersampled nuclear magnetic resonance pure chemical shift data to obtain the optimized nuclear magnetic resonance pure chemical shift spectrum.

[0058] Figure 8 The schematic diagram of the deep learning-based nuclear magnetic resonance pure chemical shift data optimization device of the embodiments of the present application is shown.

[0059] Figure 9 The hardware structure schematic diagram of the electronic device provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0061] Figure 1 A deep learning-based nuclear magnetic resonance pure chemical shift data optimization method provided by the embodiments of the present application is shown, which includes the following steps:

[0062] S1, obtaining nuclear magnetic resonance pure chemical shift data before optimization.

[0063] For the convenience of description, the related professional terms appearing in the specific embodiments are explained first:

[0064] FID (Free Induction Decay): Free Induction Decay;

[0065] FT (Fourier Transform): Fourier Transform;

[0066] IFT (Inverse Fourier Transform): Inverse Fourier Transform;

[0067] ReLU (Rectified Linear Unit): a nonlinear activation function;

[0068] NUS (Non-Uniform Sampling): Poisson exponential distribution segment undersampling;

[0069] Adam parameter optimization algorithm: Adaptive Moment Estimation.

[0070] Specifically, the nuclear magnetic resonance pure chemical shift data to be optimized is obtained, which is used as the nuclear magnetic resonance pure chemical shift data before optimization. The nuclear magnetic resonance pure chemical shift data can be obtained by experiment.

[0071] S2, a nuclear magnetic resonance pure chemical shift data optimization model, first training data and second training data are constructed, the nuclear magnetic resonance pure chemical shift data optimization model comprising a first convolutional module, a plurality of residual modules, a first convolutional layer and a J coupling modulation simulation module connected in sequence, the residual module comprising two second convolutional modules connected in sequence, the output feature of the first second convolutional module and the output feature of the second second convolutional module being added to obtain the output feature of the residual module.

[0072] In specific embodiments, the size of the convolution kernel of the first convolutional layer is 1x1; the first convolutional module comprises a second convolutional layer and a first LeakyReLU activation function layer connected in sequence, and the size of the convolution kernel of the second convolutional layer is 15x1;

[0073] The number of residual modules is 7, and the calculation process of the residual module is as follows:

[0074] Y k =CONV k1 (X k )+CONV k2 (CONV k1 (X k ));

[0075] Wherein, X k represents the input feature of the kth residual module, CONV k1 (·) represents the output feature after the first second convolutional module in the kth residual module, CONV k2 (·) represents the output feature after the second second convolutional module in the kth residual module, and Y k represents the output feature of the kth residual module.

[0076] Both the second convolutional modules comprise a third convolutional layer and a second LeakyReLU activation function layer connected in sequence, and the size of the convolution kernel of the third convolutional layer is 29x1.

[0077] Specifically, as shown in Figure 2 The first step is to construct a deep learning-based nuclear magnetic resonance pure chemical shift data optimization model. The model includes a first convolutional block 1 composed of a second convolutional layer and a LeakyReLU activation function layer, followed by N residual blocks, each containing two second convolutional blocks 2, each of which contains a third convolutional layer and a LeakyReLU activation function layer. After the N residual blocks, there is a first convolutional layer. The output features of the first convolutional block are input into the first residual block, the output features of the previous residual block are input into the next residual block, and the output features of the last residual block are input into the first convolutional layer. After the first convolutional layer, there is a J coupling modulation simulation module to simulate the effect of J coupling modulation. Specifically, the J coupling modulation simulation module first converts the frequency domain features input into the J coupling modulation simulation module into time domain through inverse Fourier transform, then performs point-by-point multiplication operation with the equivalent envelope signal f(t) of J coupling modulation to generate time domain features containing simulated J coupling modulation effect, and finally converts the time domain features into frequency domain through Fourier transform to output the frequency domain signal, which corresponds to the Fourier transform result of the optimized FID signal of nuclear magnetic resonance pure chemical shift data. The convolutional layers in the nuclear magnetic resonance pure chemical shift data optimization model are used to extract the features of the input data, the LeakyReLU activation function layer is used to introduce non-linear characteristics, and the residual blocks are used to form effective feature fusion to solve the gradient vanishing problem in deep networks, thereby improving the training effect and performance of the network.

[0078] S3, using the first training data to pre-train the nuclear magnetic resonance pure chemical shift data optimization model, obtaining a pre-trained nuclear magnetic resonance pure chemical shift data optimization model, using the second training data to fine-tune the pre-trained nuclear magnetic resonance pure chemical shift data optimization model, obtaining a fine-tuned nuclear magnetic resonance pure chemical shift data optimization model.

[0079] In specific embodiments, during the pre-training process of the nuclear magnetic resonance pure chemical shift data optimization model, the equivalent envelope signal f(t) in the J coupling modulation simulation module is fixed, and the parameters of the first convolutional block, the residual blocks, and the first convolutional layer are adjusted. At this time, the equivalent envelope signal f(t) is set to the initial value f input (t), and the expression is as follows:

[0080]

[0081] where Jinput An equivalent factor of J-coupling modulation, τ represents the duration of each data block in the pseudo two-dimensional sampling mode, and is described by the following formula:

[0082]

[0083] wherein, SW1 represents the indirect dimension spectral width of the pseudo two-dimensional sampling, t2 represents the evolution time within each data block, and is calculated from the coordinate time t of the FID signal of the pre-optimized magnetic resonance pure chemical shift spectral data, as shown in the following formula:

[0084] t2=(t modτ);

[0085] wherein, "mod" represents a modulo operation, the sampling interval of the time coordinate t is 1 / SW, and SW represents the direct dimension spectral width of the pseudo two-dimensional sampling;

[0086] In the fine-tuning process of the pre-trained magnetic resonance pure chemical shift data optimization model, the parameters of the equivalent envelope signal f(t) in the J-coupling modulation simulation module, the first convolution module, the plurality of residual modules and the first convolution layer are adjusted, and the equivalent envelope signal f(t) in the J-coupling modulation simulation module is adjusted from the initial value f input (t).

[0087] In specific embodiments, the construction process of the first training data is as follows:

[0088] The Fourier transform result and the label of the FID signal of the pre-optimized magnetic resonance pure chemical shift data in the first training data are generated by using a mathematical model based on Fourier transform and a specific attenuation function;

[0089] The mathematical model is represented as:

[0090]

[0091] wherein, s ideal (t) is the constructed ideal FID signal of the magnetic resonance pure chemical shift data, s origin (t) is the constructed pre-optimized FID signal of the magnetic resonance pure chemical shift data, N is the total number of spectral peak components in a single sample, and for each sample, N is set to a random integer in the interval of 7 to 10; A i , F i , T 2i , and J i represent the amplitude, frequency, transverse relaxation time and scalar coupling constant of the i-th spectral peak, respectively, and for each spectral peak component, A i is set to a random number in the interval of 1 to 30, F i is set to a random number in the interval of 80 to 2048, and T 2ia random number set to the interval of 0.08 to 0.6, J i a random number set to the interval of 2 to 12; t1 represents an indirect evolution time, j represents an imaginary unit, n(t) represents a Gaussian white noise, NUS(t) represents a sampling template of a segment NUS of a Poisson exponential distribution, and the sampling template is shown in the following formula:

[0092]

[0093] wherein, t' p represents a coordinate time corresponding to a pth sampling point, p is a sampling point subscript, and the total number of sampling points in each sampling segment is set to 128, which is generated according to a Poisson distribution;

[0094] the Fourier transform result of the FID signal s origin (t) of the ideal nuclear magnetic resonance pure chemical shift data constructed as a label;

[0095] The construction process of the second training data is as follows:

[0096] The nuclear magnetic resonance pure chemical shift data collected in the experiment before optimization is obtained, the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization is extracted and Fourier transformed, and the corresponding Fourier transform result is obtained, that is, the Fourier transform result of the FID signal of the nuclear magnetic resonance pure chemical shift data before optimization in the second training data.

[0097] In specific embodiments, the loss function used in the pre-training process of the nuclear magnetic resonance pure chemical shift data optimization model is a mean square error loss function between the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data obtained after the Fourier transform result of the FID signal of the nuclear magnetic resonance pure chemical shift data before optimization in the second training data is input into the nuclear magnetic resonance pure chemical shift data optimization model and the label;

[0098] The loss function loss used in the pre-training process of the nuclear magnetic resonance pure chemical shift data optimization model is:

[0099]

[0100] wherein, ‖·‖ F represents an F norm, ‖·‖1 represents an L1 norm, λ represents a regularization parameter, s' origin (t) represents the FID signal of the nuclear magnetic resonance pure chemical shift data before optimization in the second training data, s' predict (t) represents the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data obtained after the Fourier transform result of the FID signal of the nuclear magnetic resonance pure chemical shift data before optimization in the second training data is input into the pre-trained nuclear magnetic resonance pure chemical shift data optimization model, and FT(·) represents a Fourier transform function.

[0101] Specifically, referring to Figure 3 , a first training data is constructed according to a mathematical model of a free induction decay (FID) signal, a first-stage pre-training of a model for optimizing pure chemical shift data of nuclear magnetic resonance is performed in a data-driven manner, a pre-trained model for optimizing pure chemical shift data of nuclear magnetic resonance is obtained, and when an experimental signal to be optimized (pure chemical shift data of nuclear magnetic resonance collected by experiment before optimization) is input into the pre-trained model for optimizing pure chemical shift data of nuclear magnetic resonance, an output obtained is a frequency domain signal of optimized pure chemical shift data of nuclear magnetic resonance, and the frequency domain signal of optimized pure chemical shift data of nuclear magnetic resonance is a partially optimized pseudo-peak-containing signal.

[0102] In the construction of the first training data, in order to simulate the dynamic evolution process of the FID signal, the embodiments of the present application follow the physical characteristics of the nuclear magnetic resonance signal, and use a mathematical model based on Fourier transform and a specific decay function to generate, which contains multiple frequency components and different decay characteristics, ensuring the consistency of the characteristics of the simulation data and the experimental data and the reliability of the training set. In order to be closer to the complexity of the real data, the embodiments of the present application introduce random noise in the model, and adjust the signal-to-noise ratio according to the actual acquisition conditions. In addition, considering the non-uniform undersampling technology commonly used in actual experiments, the embodiments of the present application also introduce the corresponding simulation process in the above mathematical model. The Fourier transform result of s origin (t) is used as the input signal of the model, and the Fourier transform result of s ideal (t) is used as the label. The training parameters include learning rate, batch size, training round number, etc., optimization algorithm, etc.

[0103] In one of the embodiments, in the pre-training process of the first stage, the dimension of the first training data is S1: 5000x4096x2, where 5000 is the number of samples of the training data, 4096 is the length of the input signal, and 2 is the distinguishing channel of the real part and the imaginary part. The dimension of the output data is S2: 1x4096x2, where 4096 is the length of the input signal, and 2 is the distinguishing channel of the real part and the imaginary part. In other embodiments, parameters can be selected according to actual needs.

[0104] In the embodiment of the present application, the kernel size of the second convolutional layer is set to 15, the number N of residual modules is set to 7, and the kernel size of the third convolutional layer is set to 29. In the embodiment of the present application, the Adam optimizer is selected, the learning rate of the Adam optimizer is set to 0.0001, and the number of training rounds is set to 100. In the pre-training process of the first stage, the number of training data samples as labels is set to 5000, the loss value is calculated by the mean square error (MSE) between the output of the model and the label, the parameter gradient is obtained by the back propagation algorithm, and the network parameters are updated by the Adam optimization algorithm according to the calculated parameter gradient, until the loss value converges to end the pre-training.

[0105] For the obtained pre-trained nuclear magnetic resonance pure chemical shift data optimization model, single-sample customized training is performed on the nuclear magnetic resonance pure chemical shift data collected by the experiment to be optimized, the network model parameters are fine-tuned, the input is the FID signal of the single experiment collected nuclear magnetic resonance pure chemical shift data to be optimized, the output of the pre-trained nuclear magnetic resonance pure chemical shift data optimization model after each iteration is substituted into the loss function for calculation, and the neural network parameters are updated by the back propagation algorithm to minimize the loss function. The optimization process is based on the known physical model, which guides the pre-trained nuclear magnetic resonance pure chemical shift data optimization model to further learn the characteristics of the experimental data, and improves the adaptability and accuracy of the model under specific experimental data.

[0106] For the nuclear magnetic resonance pure chemical shift data obtained under different experimental conditions, the pre-training process does not need to be repeated. Only a small amount of iterative optimization of specific nuclear magnetic resonance pure chemical shift data is required in the fine-tuning process, and a pure pure chemical shift spectrum output can be achieved. In other words, after completing the pre-training process and obtaining the pre-trained nuclear magnetic resonance pure chemical shift data optimization model, only the fine-tuning process iteration with a time scale of seconds is required for various experimental signals to be optimized, and the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data can be obtained. In the embodiment of the present application, the network parameters are optimized by a few-step iterative training in the fine-tuning process, the learning rate of the Adam optimizer is set to 0.001, and the number of training rounds is set to 50.

[0107] Further, the equivalent envelope signal f(t) of the J coupling modulation in the J coupling modulation simulation module is set as a non-trainable parameter of the model in the pre-training process of the first stage, and the value is always equal to the initial value f input (t). In the fine-tuning process of the second stage, it is set as a trainable parameter of the model, and the value has a slight optimization change. The parameters of the remaining structures in the pre-training process and the fine-tuning process need to be trainable.

[0108] S4, extract the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization and perform Fourier transform to obtain the corresponding Fourier transform result, input the Fourier transform result of the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization into the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model, input the Fourier transform result into a first convolution module to obtain a first feature, input the features into a plurality of residual modules in sequence to obtain a second feature, input the second feature into a first convolution layer to obtain a frequency domain feature, and input the frequency domain feature into a J coupling modulation simulation module to obtain the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data.

[0109] In specific embodiments, the frequency domain feature is input into the J coupling modulation simulation module, first converted into the time domain through inverse Fourier transform, then subjected to point-by-point multiplication operation with the J coupling modulation equivalent envelope signal f(t) to generate a time domain feature containing the simulated J coupling modulation effect, and then converted into the frequency domain through Fourier transform to output the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data.

[0110] Specifically, after the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model is deployed, the Fourier transform result of the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization obtained through experiment can be input into the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model to predict the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data.

[0111] The above steps S1-S4 do not necessarily represent the order between the steps, but the step symbols represent that the order between the steps can be adjusted.

[0112] The application will be described below with specific examples.

[0113] The nuclear magnetic resonance pure chemical shift data before optimization selected the full-sampling nuclear magnetic resonance pure chemical shift spectrum and the 30% NUS nuclear magnetic resonance pure chemical shift spectrum to demonstrate that the application has the functions of removing the periodic pseudo-peak caused by J coupling modulation and completing the undersampling signal. In this example, the nuclear magnetic resonance pure chemical shift spectrum of a propylene glycol sample was used for optimization, and the results are shown in FIGS. 1-4. Figures 4 to 7 Figure 4 Figure 5 are the results corresponding to the input full-sampling nuclear magnetic resonance pure chemical shift spectrum, Figure 4 (a) and Figure 5 (a) are the nuclear magnetic resonance pure chemical shift spectra corresponding to the FID signal of the full-sampling propylene glycol sample collected through experiment, Figure 4 (b) and Figure 5 (b) are the output results of the pre-trained nuclear magnetic resonance pure chemical shift data optimization model obtained through the first stage pre-training of the embodiment of the application. Figure 6 and​​Figure 7 The result corresponding to the pure chemical shift spectrum of nuclear magnetic resonance of the input 30% NUS, Figure 6 (a) and Figure 7 (a) is the pure chemical shift spectrum of nuclear magnetic resonance corresponding to the FID signal of the propylene glycol sample after 30% NUS, Figure 6 (b) and Figure 7 (b) is the output result of the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model obtained in the second stage of the embodiment of the present application. As can be seen from the figure, Figure 4 (b) and Figure 6 The reconstructed signal of (b) contains obvious false peaks, and the spectrum is not clean and smooth, while Figure 5 (b) and Figure 7 The reconstructed signal of (b) is a good fitting of the ideal pure chemical shift spectrum, and the component spectrum peak is clear and complete, and does not contain false peaks.

[0114] Further referring to Figure 8 As an implementation of the method shown in the above figures, the present application provides an embodiment of a nuclear magnetic resonance pure chemical shift data optimization device based on deep learning. The device embodiment corresponds to the method embodiment shown in Figure 1 , and the device can be applied in various electronic devices.

[0115] The present application provides a nuclear magnetic resonance pure chemical shift data optimization device based on deep learning, comprising:

[0116] A data acquisition module 1 configured to acquire nuclear magnetic resonance pure chemical shift data before optimization;

[0117] A model construction module 2 configured to construct a nuclear magnetic resonance pure chemical shift data optimization model, first training data and second training data. The nuclear magnetic resonance pure chemical shift data optimization model comprises a first convolutional module, a plurality of residual modules, a first convolutional layer and a J coupling modulation simulation module connected in sequence. The residual module comprises two second convolutional modules connected in sequence. The output features of the first second convolutional module and the output features of the second second convolutional module are added to obtain the output features of the residual module.

[0118] A training fine-tuning module 3 configured to pre-train the nuclear magnetic resonance pure chemical shift data optimization model using the first training data to obtain a pre-trained nuclear magnetic resonance pure chemical shift data optimization model, and fine-tune the pre-trained nuclear magnetic resonance pure chemical shift data optimization model using the second training data to obtain a fine-tuned nuclear magnetic resonance pure chemical shift data optimization model.

[0119] The optimization module 4 is configured to extract the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization and perform Fourier transform to obtain a corresponding Fourier transform result, input the Fourier transform result of the FID signal in the nuclear magnetic resonance pure chemical shift data before optimization into the fine-tuned nuclear magnetic resonance pure chemical shift data optimization model, input the Fourier transform result into the first convolution module to obtain first features, input the features into a plurality of residual modules in sequence to obtain second features, input the second features into the first convolution layer to obtain frequency domain features, input the frequency domain features into the J coupling modulation simulation module to obtain the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data.

[0120] Figure 9 A hardware structure schematic diagram of an electronic device provided by the embodiment of the present application is shown in the figure. Figure 9 As shown in the figure, the electronic device of the embodiment includes a processor 901 and a memory 902; the memory 902 is used for storing computer execution instructions; the processor 901 is used for executing the computer execution instructions stored by the memory to realize each step executed by the electronic device in the above-mentioned embodiment. For details, please refer to the related description in the foregoing method embodiment.

[0121] Optionally, the memory 902 can be independent or integrated with the processor 901.

[0122] When the memory 902 is independently arranged, the electronic device further includes a bus 903 for connecting the memory 902 and the processor 901.

[0123] The embodiment of the present application further provides a computer storage medium, and the computer storage medium stores computer execution instructions; when the processor 901 executes the computer execution instructions, the method as above is realized.

[0124] The embodiment of the present application further provides a computer program product, and the computer program product includes a computer program; when the computer program is executed by the processor 901, the method as above is realized.

[0125] In the embodiments provided by the present application, it should be understood that the disclosed device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the modules is only a logical function division. There can be another division manner in actual implementation, for example, a plurality of modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or in other forms.

[0126] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to implement the embodiments of the present application.

[0127] In addition, each functional module in various embodiments of the present application can be integrated in one processing unit, or each module can be physically present alone, or two or more modules can be integrated in one unit. The unit formed by the above modules can be realized in the form of hardware or in the form of hardware plus software functional unit.

[0128] The integrated modules realized in the form of software functional modules can be stored in a computer readable storage medium. The software functional modules stored in a storage medium include a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor 901 to execute part of the steps of the method of various embodiments of the present application.

[0129] It should be understood that the processor 901 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor 901 can be any conventional processor 901, etc. The steps of the method disclosed in the application can be directly embodied as hardware processor 901 execution, or executed by a combination of hardware and software modules in the processor 901.

[0130] The memory 902 can include a high-speed RAM memory, and can also include a non-volatile storage NVM, for example, at least one disk memory, and can also be a U disk, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0131] The bus 903 can be an Industry Standard Architecture (ISA), Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus 903 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus 903 in the drawings of the present application does not limit to only one bus 903 or one type of bus 903.

[0132] The storage medium described above can be realized by any type of volatile or nonvolatile storage devices or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk, or an optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0133] An exemplary storage medium is coupled to the processor 901, so that the processor 901 can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor 901. The processor 901 and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor 901 and the storage medium can also exist as discrete components in an electronic device or a host device.

[0134] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by relevant hardware instructed by programs. The foregoing programs can be stored in a computer readable storage medium. When the programs are executed, the steps of the above-mentioned method embodiments are executed; and the foregoing storage medium includes various media that can store program codes, such as ROM, RAM, magnetic disks, or optical disks.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some or all of the technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for optimizing pure chemical shift data of nuclear magnetic resonance based on deep learning, characterized in that: The following steps are involved: Obtain pure chemical shift data of NMR before optimization; Constructing a nuclear magnetic resonance pure chemical shift data optimization model, first training data, and second training data, wherein the nuclear magnetic resonance pure chemical shift data optimization model includes a first convolution module, a plurality of residual modules, a first convolution layer, and a J-coupling modulation simulation module connected in sequence, wherein the residual module includes two second convolution modules connected in sequence, and the output feature of the first second convolution module and the output feature of the second second convolution module are added together to form the output feature of the residual module; Pre-training the nuclear magnetic resonance pure chemical shift data optimization model using the first training data to obtain a pre-trained nuclear magnetic resonance pure chemical shift data optimization model, and fine-tuning the pre-trained nuclear magnetic resonance pure chemical shift data optimization model using the second training data to obtain a fine-tuned nuclear magnetic resonance pure chemical shift data optimization model; Extract the FID signal from the pure chemical shift data of nuclear magnetic resonance before optimization and perform Fourier transform to obtain a corresponding Fourier transform result. Input the Fourier transform result of the FID signal in the pure chemical shift data of nuclear magnetic resonance before optimization into the fine-tuned pure chemical shift data optimization model of nuclear magnetic resonance. Input the Fourier transform result into a first convolution module to obtain a first feature. The first feature is sequentially passed through several residual modules to obtain a second feature. The second feature is input into a first convolution layer to obtain a frequency domain feature. The frequency domain feature is input into a J-coupled modulation simulation module to obtain a frequency domain signal of the optimized pure chemical shift data of nuclear magnetic resonance.

2. The method for optimizing pure chemical shift data of nuclear magnetic resonance based on deep learning according to claim 1, characterized in that: The frequency domain features are input into the J-coupling modulation simulation module, first converted to the time domain through inverse Fourier transform, and then point-by-point multiplication operation is performed with the equivalent envelope signal f(t) of the J-coupling modulation to generate a time domain feature containing the simulated J-coupling modulation effect. The time domain features are then converted to the frequency domain through Fourier transform and output as a frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data.

3. The method for optimizing pure chemical shift data of nuclear magnetic resonance based on deep learning according to claim 2, characterized in that: During the pre-training process of the NMR pure chemical shift data optimization model, the equivalent envelope signal f(t) in the J-coupling modulation simulation module is fixed, and the parameters of the first convolution module, the plurality of residual modules, and the first convolution layer are adjusted; at this time, the equivalent envelope signal f(t) is set to the initial value f input (t), the expression is as follows: Among them, J input represents the equivalent factor of J-coupling modulation, and τ represents the duration of each data block in the pseudo-two-dimensional sampling mode, which is described by the following formula: Where SW1 represents the indirect dimensional spectrum width of pseudo-two-dimensional sampling, and t2 represents the evolution time within each data block, which is calculated from the coordinate time t of the FID signal of the pure chemical shift spectroscopy data before optimization, as shown in the following formula: t2=(t mod τ); Where "mod" represents the modulo operation, the sampling interval of the time coordinate t is 1 / SW, and SW represents the direct dimension spectrum width of the pseudo-two-dimensional sampling; During the fine-tuning process of the pre-trained NMR pure chemical shift data optimization model, the parameters of the equivalent envelope signal f(t), the first convolution module, the plurality of residual modules, and the first convolution layer in the J-coupling modulation simulation module are adjusted, and the equivalent envelope signal f(t) in the J-coupling modulation simulation module is adjusted from the initial value f input (t) Start adjustment.

4. The method for optimizing pure chemical shift data of nuclear magnetic resonance based on deep learning according to claim 3, characterized in that: The construction process of the first training data is as follows: Generate a Fourier transform result and a label of an FID signal of the pre-optimization nuclear magnetic resonance pure chemical shift data in the first training data using a mathematical model based on Fourier transform and a specific attenuation function; The mathematical model is expressed as: Among them, s ideal (t) is the FID signal of the ideal pure chemical shift data of NMR, s origin (t) is the FID signal of the pure chemical shift data of NMR before optimization, N is the total number of peak components in a single sample, and for each sample, N is set to a random integer between 7 and 10; A i 、F i 、T 2i , and J i Represent the amplitude, frequency, transverse relaxation time and scalar coupling constant of the i-th spectral peak respectively. For each spectral peak component, A i Set to a random number between 1 and 30, F i Set to a random number between 80 and 2048, T 2i Set to a random number between 0.08 and 0.6, J i Set to a random number between 2 and 12; t1 represents the indirect dimension evolution time, j represents the imaginary unit, n(t) represents Gaussian white noise, and NUS(t) represents the sampling template of the Poisson exponential distribution segment NUS. The sampling template is shown in the following formula: Where t′ p Indicates the coordinate time corresponding to the p-th sampling point, where p is the sampling point subscript, generated according to Poisson distribution, and the total number of sampling points in each sampling segment is set to 128; The FID signal s of the ideal pure chemical shift data of NMR will be constructed origin The Fourier transform result of (t) is used as the label; The construction process of the second training data is as follows: Acquire the pre-optimization pure nuclear magnetic resonance chemical shift data collected in the experiment, extract the FID signal in the pre-optimization pure nuclear magnetic resonance chemical shift data and perform Fourier transform to obtain a corresponding Fourier transform result, which is the Fourier transform result of the FID signal of the pre-optimization pure nuclear magnetic resonance chemical shift data in the second training data.

5. The method for optimizing pure chemical shift data of nuclear magnetic resonance based on deep learning according to claim 4, characterized in that: The loss function used in the pre-training process of the nuclear magnetic resonance pure chemical shift data optimization model is a mean square error loss function between the frequency domain signal of the optimized nuclear magnetic resonance pure chemical shift data and the label obtained by inputting the Fourier transform result of the FID signal of the nuclear magnetic resonance pure chemical shift data before optimization in the second training data into the nuclear magnetic resonance pure chemical shift data optimization model; The loss function used in the pre-training process of the NMR pure chemical shift data optimization model is: in,‖·‖ F represents the F norm, ‖·‖1 represents the L1 norm, λ represents the regularization parameter, s′ origin (t) represents the FID signal of the pure chemical shift data of nuclear magnetic resonance before optimization in the second training data, s′ predi (t) represents the frequency domain signal of the optimized pure nuclear magnetic resonance chemical shift data obtained by inputting the Fourier transform result of the FID signal of the pre-optimization pure nuclear magnetic resonance chemical shift data in the second training data into the pre-trained nuclear magnetic resonance pure chemical shift data optimization model, and FT(·) represents the Fourier transform function.

6. The method for optimizing pure chemical shift data of nuclear magnetic resonance based on deep learning according to claim 1, characterized in that: The convolution kernel size of the first convolution layer is 1×1; the first convolution module includes a second convolution layer and a first LeakyReLU activation function layer connected in sequence, and the convolution kernel size of the second convolution layer is 15×1; The number of the residual modules is 7, and the calculation process of the residual modules is as follows: Y k =CONV k1 (X k )+CONV k2 (CONV k1 (X k )); Among them, X k Represents the input features of the kth residual module, CONV k1 (·) represents the output feature of the first second convolution module in the kth residual module, CONV k2 (·) represents the output feature of the second convolution module in the kth residual module, Y k represents the output feature of the kth residual module; The two second convolution modules each include a third convolution layer and a second LeakyReLU activation function layer connected in sequence, and the convolution kernel size of the third convolution layer is 29×1.

7. A deep learning-based NMR pure chemical shift data optimization device, characterized in that: include: a data acquisition module configured to acquire pure chemical shift data of nuclear magnetic resonance before optimization; a model construction module configured to construct a nuclear magnetic resonance pure chemical shift data optimization model, first training data, and second training data, wherein the nuclear magnetic resonance pure chemical shift data optimization model includes a first convolution module, a plurality of residual modules, a first convolution layer, and a J-coupling modulation simulation module connected in sequence, wherein the residual module includes two second convolution modules connected in sequence, and the output feature of the first second convolution module and the output feature of the second second convolution module are added together to form the output feature of the residual module; a training and fine-tuning module configured to pre-train the nuclear magnetic resonance pure chemical shift data optimization model using the first training data to obtain a pre-trained nuclear magnetic resonance pure chemical shift data optimization model, and fine-tune the pre-trained nuclear magnetic resonance pure chemical shift data optimization model using the second training data to obtain a fine-tuned nuclear magnetic resonance pure chemical shift data optimization model; The optimization module is configured to extract the FID signal in the pure chemical shift data of nuclear magnetic resonance before optimization and perform Fourier transform to obtain a corresponding Fourier transform result, input the Fourier transform result of the FID signal in the pure chemical shift data of nuclear magnetic resonance before optimization into the pure chemical shift data optimization model after fine-tuning, input the Fourier transform result into a first convolution module to obtain a first feature, the first feature is sequentially passed through a plurality of residual modules to obtain a second feature, the second feature is input into a first convolution layer to obtain a frequency domain feature, the frequency domain feature is input into a J-coupled modulation simulation module to obtain a frequency domain signal of the optimized pure chemical shift data of nuclear magnetic resonance.

8. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images