Multi-nuclear magnetic resonance phosphorus spectrum data processing method, device and processing equipment

Abstract

The invention provides a multi-nuclear magnetic resonance phosphorus spectrum data processing method, device and processing equipment, designs a novel multi-nuclear magnetic resonance phosphorus spectrum data processing scheme, and builds a corresponding specific model processing architecture on the basis of deep learning, so that the resolution of multi-nuclear magnetic resonance phosphorus spectrum data can be effectively improved, and the accuracy of the multi-nuclear magnetic resonance phosphorus spectrum data processing is improved. And high-quality data support is provided for related applications developed on the basis of the multi-nuclear magnetic resonance phosphorus spectrum data, so that the corresponding medical service quality can be further improved.

Description

Technical Field

[0001] The present application relates to the field of magnetic resonance, and particularly to a method, apparatus, and processing device for processing multi-nuclear magnetic resonance phosphorus spectrum data. Background Art

[0002] As a highly energy-consuming organ, the brain relies on glucose as the main carbon substrate and supports oxygen metabolism and the production of adenosine triphosphate (ATP) molecules through oxidative phosphorylation in mitochondria. ATP is the main source of biochemical energy necessary for neurophysiology and brain function. Approximately one-quarter of the total brain ATP energy is used to maintain cell integrity and brain tissue vitality, and the remaining ATP energy supports electrophysiological activities in the resting or working brain state. Abnormal brain energy metabolism is related to many brain diseases and neurodegenerative diseases, including schizophrenia, Alzheimer's disease, Huntington's disease, Parkinson's disease, mitochondrial dysfunction, and the aging process. Therefore, the development of quantitative neuroimaging techniques helps to non-invasively evaluate regional brain energy metabolism in vivo under normal and disease conditions.

[0003] Multi-nuclear magnetic resonance phosphorus spectrum (which can be referred to as 31 P-MRS or 31 P magnetic resonance phosphorus spectrum) can detect other phosphate metabolites in addition to ATP, phosphocreatine (PCr), and inorganic phosphate (Pi), including nicotinamide adenine dinucleotide (NADH), glycerophosphoethanolamine (GPE), glycerophosphocholine (GPC), phosphoethanolamine (PE), and phosphocholine (PC), which are actively involved in oxidative phosphorylation, NAD redox reactions, and membrane phospholipid metabolism through phospholipid biosynthetic enzymes.

[0004] However, 31 The clinical application of 31 P multi-nuclear magnetic resonance technology (which can also be referred to as 1 P multi-nuclear magnetic resonance technology) faces many challenges. The main limitation is the low concentration of detectable metabolites. The concentration of phosphate metabolites detected by in vivo multi-nuclear magnetic resonance phosphorus spectrum in the human brain is only a few millimoles, which is thousands of times lower than the tissue water proton concentration detected by

[0005] Therefore, in practical applications, a large number of signal averages are often required to obtain a reasonable signal-to-noise ratio and phosphorus spectrum quality, which obviously limits the resolution of in vivo 31 P multi-nuclear magnetic resonance technology, thereby affecting the reliability and applicability of the examination. Summary of the Invention

[0006] The present application provides a method, apparatus, and processing device for processing multi-nuclear magnetic resonance phosphorus spectrum data, designs a novel multi-nuclear magnetic resonance phosphorus spectrum data processing scheme, and constructs a corresponding specific model processing architecture based on deep learning, so as to effectively improve the resolution of multi-nuclear magnetic resonance phosphorus spectrum data, provide high-quality data support for related applications based on multi-nuclear magnetic resonance phosphorus spectrum data, and further ensure that the corresponding medical service quality is further improved.

[0007] In a first aspect, the present application provides a method for processing multi-nuclear magnetic resonance phosphorus spectrum data, the method comprising: Obtaining first nuclear magnetic resonance phosphorus spectrum data currently to be resolution-enhanced; Inputting the first multi-nuclear magnetic resonance phosphorus spectrum data into a multi-nuclear magnetic resonance phosphorus spectrum data processing model for processing, wherein the multi-nuclear magnetic resonance phosphorus spectrum data processing model is used to improve the resolution of the multi-nuclear magnetic resonance phosphorus spectrum data input into the model, the multi-nuclear magnetic resonance phosphorus spectrum data processing model is pre-trained by labeled sample multi-nuclear magnetic resonance phosphorus spectrum data, the multi-nuclear magnetic resonance phosphorus spectrum data processing model includes an input side and an output side, the input side includes 5 first dense convolutional modules and 5 max pooling layers, the first dense convolutional modules are used for feature extraction, the max pooling layers are used for downsampling, the 1st first dense convolutional module corresponds to the model input, the first dense convolutional modules and the max pooling layers are alternately connected, the 5th max pooling layer corresponds to the output of the input side, the output side includes 5 self-attention modules, 5 second dense convolutional modules and 4 upsampling layers, the self-attention modules are used for encoding, the second dense convolutional modules are used for decoding, the upsampling layers are used for upsampling, the second dense convolutional modules and the upsampling layers are alternately connected, the 5th second dense convolutional module corresponds to the model output, and each first dense convolutional module is connected to a second dense convolutional module with an inverted order relationship with its own module order through a self-attention module; Obtaining second multi-nuclear magnetic resonance phosphorus spectrum data corresponding to the first multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model.

[0008] In a second aspect, the present application provides a multi-nuclear magnetic resonance phosphorus spectrum data processing apparatus, the apparatus comprising: A first obtaining unit, configured to obtain first multi-nuclear magnetic resonance phosphorus spectrum data currently to be resolution-enhanced; A processing unit for inputting first multi-nuclear magnetic resonance phosphorus spectrum data into a multi-nuclear magnetic resonance phosphorus spectrum data processing model for processing. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is used to improve the resolution of the multi-nuclear magnetic resonance phosphorus spectrum data input into the model. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is pre-trained by labeled sample multi-nuclear magnetic resonance phosphorus spectrum data. The multi-nuclear magnetic resonance phosphorus spectrum data processing model includes an input side and an output side. The input side includes 5 first dense convolutional modules and 5 max pooling layers. The first dense convolutional modules are used for feature extraction, and the max pooling layers are used for downsampling. The first first dense convolutional module corresponds to the model input. The first dense convolutional modules and the max pooling layers are alternately connected. The 5th max pooling layer corresponds to the output of the input side. The output side includes 5 self-attention modules, 5 second dense convolutional modules and 4 upsampling layers. The self-attention modules are used for encoding, the second dense convolutional modules are used for decoding, and the upsampling layers are used for upsampling. The second dense convolutional modules and the upsampling layers are alternately connected. The 5th second dense convolutional module corresponds to the model output. Each first dense convolutional module is connected to a second dense convolutional module with a reverse order relationship through a self-attention module; A second acquisition unit for acquiring second multi-nuclear magnetic resonance phosphorus spectrum data corresponding to the first multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model.

[0009] In a third aspect, the present application provides a processing device, including a processor and a memory. A computer program is stored in the memory. When the processor calls the computer program in the memory, it executes the method provided in the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0010] In a fourth aspect, the present application provides a computer-readable storage medium. The computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the method provided in the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0011] From the above content, the following beneficial effects of the present application can be obtained: Aiming at the goal of improving the resolution of multi-nuclear magnetic resonance phosphorus spectrum data, the present application designs a novel multi-nuclear magnetic resonance phosphorus spectrum data processing scheme and builds a corresponding specific model processing architecture based on deep learning. In this way, the resolution of multi-nuclear magnetic resonance phosphorus spectrum data can be effectively improved, providing high-quality data support for related applications based on multi-nuclear magnetic resonance phosphorus spectrum data, and further ensuring that the corresponding medical service quality can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those skilled in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0013] Figure 1 It is a schematic flowchart of a method for processing multi-nuclear magnetic resonance phosphorus spectrum data of the present application; Figure 2 It is a schematic scenario diagram of the overall model processing architecture of the present application; Figure 3 It is a schematic structural diagram of a dense convolution module of the present application; Figure 4 It is a schematic structural diagram of a self-attention module of the present application; Figure 5 It is a schematic structural diagram of a multi-nuclear magnetic resonance phosphorus spectrum data processing device of the present application; Figure 6 It is a schematic structural diagram of a processing device of the present application. Detailed implementation manners

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the protection scope of the present application.

[0015] The terms "first", "second", etc. in the specification and claims of the present application and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order different from that shown or described here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or modules does not necessarily need to be limited to those clearly listed steps or modules, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. In the present application, the naming or numbering of steps does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The already named or numbered process steps can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0016] The division of modules in this application is a logical division. In actual implementation, there may be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Additionally, the coupling, direct coupling, or communication connection shown or discussed among each other can be through some interfaces. The indirect coupling or communication connection between modules can be in an electrical or other similar form, which is not limited in this application. Moreover, the modules or sub-modules described as separate components can be physically separated or not, can be physical modules or not, or can be distributed into multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application.

[0017] Before introducing the multi-nuclear magnetic resonance phosphorus spectrum data processing method provided by this application, the background content involved in this application is first introduced.

[0018] The multi-nuclear magnetic resonance phosphorus spectrum data processing method, device, and computer-readable storage medium provided by this application can be applied to processing devices. A novel multi-nuclear magnetic resonance phosphorus spectrum data processing solution is designed, and a corresponding specific model processing architecture is built based on deep learning. In this way, the resolution of multi-nuclear magnetic resonance phosphorus spectrum data can be effectively improved, providing high-quality data support for related applications based on multi-nuclear magnetic resonance phosphorus spectrum data, and further ensuring the improvement of the corresponding medical service quality.

[0019] The execution subject of the multi-nuclear magnetic resonance phosphorus spectrum data processing method mentioned in this application can be a multi-nuclear magnetic resonance phosphorus spectrum data processing device, or different types of processing devices such as a server, physical host, or user equipment (UE) that integrates this multi-nuclear magnetic resonance phosphorus spectrum data processing device. Among them, the multi-nuclear magnetic resonance phosphorus spectrum data processing device can be implemented in a hardware or software manner. The UE can specifically be a terminal device such as a smart phone, tablet computer, notebook computer, desktop computer, or personal digital assistant (PDA). The processing devices can be set up in the form of a device cluster.

[0020] It can be understood that in specific applications, the focus of the solution of this application is to perform corresponding magnetic resonance phosphorus spectrum data processing based on a pre-configured magnetic resonance phosphorus spectrum data processing model. In this case, the processing device that executes the multi-nuclear magnetic resonance phosphorus spectrum data processing method of this application or, in other words, the processing device that carries the corresponding application service of the multi-nuclear magnetic resonance phosphorus spectrum data processing method of this application usually only needs to meet the required data processing capabilities. The specific device types and specific device deployment forms involved are relatively flexible and can be flexibly configured according to actual needs, which are not specifically limited in this application.

[0021] If further involved in the training of the magnetic resonance phosphorus spectrum data processing model, further adaptive adjustments can be made according to the actual situation.

[0022] As an example, the processing device can be divided into two parts, that is, the processing device can include a first processing device that executes the training task of the magnetic resonance phosphorus spectrum data processing model and a second processing device that executes the application task of the magnetic resonance phosphorus spectrum data processing model.

[0023] Furthermore, in some cases, the solution of the present application can also involve further data support work for auxiliary disease diagnosis such as further lesion identification / localization, etc., then further adaptive adjustments can be made to the processing device corresponding to further data analysis and processing. Corresponding to the above example, in some cases, the processing device can also include a third processing device that executes the data analysis and processing based on the processing results of multiple nuclear magnetic resonance phosphorus spectrum data.

[0024] Furthermore, in some cases, the solution of the present application can also involve further result display, then the processing device itself can be configured with a display screen (including a touch screen), or an external display device, or an external other device with a display screen to meet the requirements of the result display function.

[0025] Next, the multi-nuclear magnetic resonance phosphorus spectrum data processing method provided by the present application will be introduced.

[0026] First, refer to Figure 1 , Figure 1 FIG. shows a schematic flowchart of a multi-nuclear magnetic resonance phosphorus spectrum data processing method of the present application. The multi-nuclear magnetic resonance phosphorus spectrum data processing method provided by the present application can specifically include the following steps S101 to step S103: Step S101, obtain the first multi-nuclear magnetic resonance phosphorus spectrum data whose resolution needs to be improved currently; It can be understood that corresponding to the actual processing requirements for multi-nuclear magnetic resonance phosphorus spectrum data, more specifically, corresponding to the actual processing requirements for improving the resolution of multi-nuclear magnetic resonance phosphorus spectrum data, the present application can obtain the multi-nuclear magnetic resonance phosphorus spectrum data whose resolution needs to be improved currently or to be improved. Herein, for the convenience of description, the multi-nuclear magnetic resonance phosphorus spectrum data obtained here is denoted as the first multi-nuclear magnetic resonance phosphorus spectrum data, and the multi-nuclear magnetic resonance phosphorus spectrum data obtained by subsequent model processing is denoted as the second multi-nuclear magnetic resonance phosphorus spectrum data.

[0027] As for the multi-nuclear magnetic resonance phosphorus spectrum data itself, it is understandable that this is an existing concept. The purpose of the solution of this application is to process the ready-made or preliminarily data-collected multi-nuclear magnetic resonance phosphorus spectrum data through the multi-nuclear magnetic resonance phosphorus spectrum data processing model built by the multi-nuclear magnetic resonance phosphorus spectrum data processing logic designed in this application, so as to further improve its resolution. At the same time, it can also avoid the situation of originally needing to collect multi-nuclear magnetic resonance phosphorus spectrum data multiple times, reduce the overall data collection duration, enhance the reliability and applicability of the data, and achieve better auxiliary data support.

[0028] As an example, for the low-resolution multi-nuclear magnetic resonance phosphorus spectrum data, it can specifically involve the multi-nuclear magnetic resonance phosphorus spectrum data collected by a 3T multi-nuclear magnetic resonance system and a dual-tuned (1H / 31P) "birdcage" coil. The scanning scheme can include a low-time-resolution 2D-CSI-31P-MRS sequence.

[0029] Correspondingly, the scanning parameters of the low-time-resolution MRS are: TR / TE = 3500 / 0.24 ms, acquisition bandwidth = 3000 Hz, number of acquisition points = 2048, number of excitations (NSA) = 6, using an automatic first-order pencil shim, voxel = 20×20×40 mm3, a total of 256 voxels are collected, and the scanning time is 9 minutes and 24 seconds.

[0030] Among them, for the acquisition and processing of the first multi-nuclear magnetic resonance phosphorus spectrum data, it can be either manually input or automatically acquired. The device can extract it from the local storage space or the storage space of other devices, or the device can receive the data sent by other devices, which is all possible, corresponding to the flexible application requirements in actual situations.

[0031] In addition, in actual applications, the solution of this application is usually carried out in the form of a work task, that is, a multi-nuclear magnetic resonance phosphorus spectrum data processing task. The initiation of the task can be either manually input or automatically initiated. The device can initiate it independently according to the preset autonomous initiation strategy / rules, or the device can receive the task sent by other devices, which is all possible. Similar to the above, it meets the diverse application requirements of the solution.

[0032] It should be noted that the concept of "multi-nuclear magnetic resonance phosphorus spectrum" involved in this application can also be referred to as " 31 P multi-nuclear magnetic resonance".

[0033] Step S102, input the first multi-nuclear magnetic resonance phosphorus spectrum data into the multi-nuclear magnetic resonance phosphorus spectrum data processing model for processing. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is used to improve the resolution of the input multi-nuclear magnetic resonance phosphorus spectrum data. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is pre-trained by the labeled sample multi-nuclear magnetic resonance phosphorus spectrum data. The multi-nuclear magnetic resonance phosphorus spectrum data processing model includes an input side and an output side. The input side includes 5 first dense convolutional modules and 5 max pooling layers. The first dense convolutional modules are used for feature extraction, and the max pooling layers are used for downsampling. The 1st first dense convolutional module corresponds to the model input. The first dense convolutional modules and the max pooling layers are alternately connected. The 5th max pooling layer corresponds to the output of the input side. The output side includes 5 self-attention modules, 5 second dense convolutional modules, and 4 upsampling layers. The self-attention modules are used for encoding, the second dense convolutional modules are used for decoding, and the upsampling layers are used for upsampling. The second dense convolutional modules and the upsampling layers are alternately connected. The 5th second dense convolutional module corresponds to the model output. Each first dense convolutional module is connected to the second dense convolutional module with the reverse order relationship of its own module through a self-attention module; It can be seen that this application designs a novel multi-nuclear magnetic resonance phosphorus spectrum data processing logic / scheme, which can be specifically implemented by a corresponding deep learning model or neural network model. For this, this application can pre-configure the corresponding multi-nuclear magnetic resonance phosphorus spectrum data processing model, and this model is trained by the sample multi-nuclear magnetic resonance phosphorus spectrum data in the prior model training work / session.

[0034] Specifically, referring to Figure 2 a schematic diagram of a scenario of the overall model processing architecture of this application shown, the processing architecture of the multi-nuclear magnetic resonance phosphorus spectrum data processing model of this application is composed of 10 dense convolutional modules, 5 max pooling layers, 5 self-attention modules, and 4 upsampling layers, where: 1) The entire multi-nuclear magnetic resonance phosphorus spectrum data processing model can be divided into an input side and an output side. Among the 10 dense convolutional modules, 5 are on the input side (that is, Figure 2 the 5 dense convolutional modules on the left), for the convenience of description, denoted as 5 first dense convolutional modules, and the other 5 second dense convolutional modules are on the output side (that is, Figure 2 the 5 dense convolutional modules on the right).

[0035] 2) On the input side, the first dense convolutional modules and the max pooling layers are alternately connected. In each group of first dense convolutional modules and max pooling layers, in the direction from the model input to the output (corresponding to Figure 2In the left-to-right direction (in the figure), the first dense convolution module is in the front position, and the first first dense convolution module is at the model input position. The max pooling layer is in the rear position. That is, the output of the first first dense convolution module is connected to the input of the first max pooling layer, and the output of the first max pooling layer is connected to the input of the second first dense convolution module... The output of the fifth first dense convolution module is connected to the input of the fifth max pooling layer. In this way, a cascaded input-side structure is formed.

[0036] In the specific working process, the first dense convolution module, as a feature extractor, extracts features from the phosphorus spectrum data of the input module to obtain the corresponding feature map, while the max pooling layer is responsible for further downsampling the feature map output by the first dense convolution module.

[0037] 3) On the output side, similar to the input side, the second dense convolution module and the upsampling layer are alternately connected. In each group of the second dense convolution module and the upsampling layer, in the direction from the input to the output of the model (corresponding to Figure 2 the left-to-right direction in the figure), the second dense convolution module is in the front position, the upsampling layer is in the rear position, and the fifth second dense convolution module is at the model output position. That is, the output of the first second dense convolution module is connected to the input of the first upsampling layer, and the output of the first upsampling layer is connected to the input of the second second dense convolution module... The output of the fourth upsampling layer is connected to the input of the fifth second dense convolution module; and the self-attention module connects the first dense convolution module and the second dense convolution module before and after (left and right), and for the two connected convolution modules, there is a characteristic of reversed order in their modules. That is, there is one self-attention module connected between the first first dense convolution module and the fifth second dense convolution module (corresponding to Figure 2 the top), there is one self-attention module connected between the second first dense convolution module and the fourth second dense convolution module... There is one self-attention module connected between the fifth first dense convolution module and the first second dense convolution module (corresponding to Figure 2 the bottom) to implement the corresponding self-attention mechanism. In this way, the output side and the previous input side form a nested complex stacked model structure.

[0038] In the specific working process, the self-attention module, as an encoder, encodes the feature map of the input module, and the second dense convolution module, as a decoder, decodes the feature map of the input module. The input of the second dense convolution module (decoder) includes the output of the self-attention module (encoder) of the current layer and the output of the previous layer decoder after upsampling.

[0039] Under the above processing mechanism, it can be seen that the present application utilizes a specially designed and densely connected neural network structure to continuously and deeply extract the characteristics of phosphorus spectrum data, and screens and enhances the phosphorus spectrum data characteristics through a self-attention module. In this way, while paying attention to different bands, the error caused by noise is reduced, and the improvement effect of phosphorus spectrum resolution is stably and effectively achieved. At the same time, compared with the existing problem that the gyromagnetic ratio is about 2 times lower than that of 1H, which further reduces the inherent detection sensitivity and signal-to-noise ratio, the signal-to-noise ratio can be further improved, and the overall sampling time can be reduced. Generally speaking, the acquisition effect of multi-nuclear magnetic resonance phosphorus spectrum data with high efficiency and high precision is achieved.

[0040] Step S103: Obtain the second multi-nuclear magnetic resonance phosphorus spectrum data corresponding to the first multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model.

[0041] After processing the current first multi-nuclear magnetic resonance phosphorus spectrum data through the multi-nuclear magnetic resonance phosphorus spectrum data processing model to obtain the corresponding second multi-nuclear magnetic resonance phosphorus spectrum data with significantly improved resolution, the model will output it.

[0042] Correspondingly, the second multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model can be extracted.

[0043] At this time, it can be understood that for the processing result of the second multi-nuclear magnetic resonance phosphorus spectrum data of the magnetic resonance phosphorus spectrum data, local storage, off-site storage, result forwarding, outputting a completion processing prompt, result display, or further data processing and analysis (such as the data support work for lesion recognition / location and other auxiliary disease diagnoses mentioned above) can be carried out next. Obviously, the specific data application content involved is relatively flexible and can be adjusted according to the pre-set and real-time configured data application strategies / rules. The present application does not make specific limitations.

[0044] Next, continue to elaborate in detail on each step of the above Figure 1 illustrated embodiment and its possible implementation manners in actual applications.

[0045] In terms of specific network structure parameters, the present application provides a further and practical implementation solution.

[0046] Specifically, as an exemplary embodiment, corresponding to the input side, the pooling size of the max-pooling layer can be specifically 2, the pooling stride can be specifically 2, and the sampling multiple of the up-sampling layer can be specifically 2; Corresponding to the input side, the output feature map sizes of the 5 first dense convolution modules can be specifically (corresponding to Figure 2 the first first dense convolution module from left to right, and so on later), , , and ; For the corresponding output side, the sizes of the output feature maps of the 5 self-attention modules can specifically be (corresponding to Figure 2 the 1st self-attention module from left to right in , and so on), , and ; For the corresponding output side, the sizes of the output feature maps of the 5 second dense convolution modules can specifically be (corresponding to Figure 2 the 1st second dense convolution module from left to right in , and so on), , and .

[0047] It can be seen that although each network structure / module can have the same name and correspond to the same processing mode or function, their specific parameters are different. In the embodiments herein, for each network structure / module involved in the multi-nuclear magnetic resonance phosphorus spectrum data processing model established in this application, specific parameter designs are given from a deep level, which has better practical significance.

[0048] Meanwhile, in terms of the specific module structure, as an exemplary embodiment, for the dense convolution modules in the model, namely the first dense convolution module and the second dense convolution module, they can both be composed of a convolutional layer, a normalization layer, and an activation layer.

[0049] Furthermore, referring to Figure 3 a schematic structural diagram of the dense convolution module of this application shown, as an exemplary embodiment, for the specific structure or processing logic of the first dense convolution module and the second dense convolution module of this application, specifically, there can be: Assume the input feature map is , , is the length of the feature map, is the number of channels of the feature map, and the output feature map is , , which is specifically obtained by the following formula: , , , where is the feature map after the first layer of convolution, , is the feature map after the second layer of convolution, , is the number of convolutional kernels, is the ReLU activation function, is the batch normalization function, is the one-dimensional convolution function, is the concatenation function.

[0050] Taking the first dense convolution module in the model, i.e., the first first dense convolution module, as an example, the specific length of the input phosphorus spectrum can be . The number of convolutional kernels in the convolutional layer is 64, the convolutional stride is 1, and the size of the convolutional kernel is 13. Therefore, the size of the final output feature map of this dense convolution module is , which also corresponds to the parameter design example involved in the first first dense convolution module mentioned above.

[0051] It can be seen that the embodiment here is the specific working logic of the first dense convolution module and the second dense convolution module of the present application. Starting from the quantization formula level, a set of specific implementation supporting schemes is given, which can preferably implement the module functions of the feature extractor / decoder designed in the model architecture of the present application, with specific and better practicability.

[0052] Next, continue to focus on the self-attention module involved in the model architecture of the present application. It can be understood that its purpose is to implement the self-attention mechanism to guide the model to focus on more delicate phosphorus spectrum data features, thereby promoting a better resolution improvement effect, and further strengthening this feature through a multi-layer nested self-attention mechanism.

[0053] In this regard, in terms of the specific module structure, as an exemplary embodiment, for the self-attention module, it can specifically be composed of an embedding layer, a multi-self-attention layer, a concatenation layer, a normalization layer, and a reduction layer. The multi-self-attention layer includes (for example, specifically 12) self-attention layers.

[0054] Next, referring to Figure 4 a schematic structural diagram of the self-attention module of the present application shown, as an exemplary embodiment, for the specific structure or processing logic of the self-attention module of the present application, specifically, there can be: Assume that the input feature map is , , is the length of the feature map, is the number of convolutional kernels, and the output feature map is , , which is specifically obtained by the following processing content: 1) , wherein, is a deformation function, transforms the shape of into , is a one-dimensional convolution function, , , , is the convolution layer stride, is the number of convolution kernels, indicates that the embedding layer divides the output feature map of the first dense convolution module into equal parts for re-encoding; It can be understood that in this step, the embedding layer performs re-encoding on through one-dimensional convolution to obtain .

[0055] 2) , wherein, , , is the positional encoding; It can be understood that in this step, after the embedding layer performs re-encoding, the positional encoding is further added for subsequent processing.

[0056] 3) , wherein, is the layer normalization function, is the splicing function, is the output set of self-attention layers, , , , , , is the normalization exponential function, , and are respectively the weight matrices of the query matrix, key matrix and value matrix in the th self-attention layer, , and , is the hidden layer dimension; It can be understood that in this step, for each self-attention layer, the query matrix is calculated through and the key matrix The similarity matrix is normalized, The function then probabilizes the similarity matrix to obtain a probability distribution, and finally uses the probability distribution as a weight matrix to perform a weighted sum on the value matrix to obtain the self-attention matrix .

[0057] Then, for each self-attention matrix, i.e., , after passing through the concatenation layer and the normalization layer, we obtain .

[0058] 4) Repeat the operation in 3) to obtain the feature vectors , ; It can be understood that the repeated operation here is to use the obtained in the previous round as the new input (corresponding to the input at the first time) for cyclic processing. The number of cycles R can be fixed or dynamically adjusted, for example, it can be specifically 8.

[0059] 5) , where, is the deformation function, which transforms the shape of the feature vector from to , is the upsampling function, and the upsampling rate is . Here, the number of convolutional kernels of the one-dimensional convolutional function is .

[0060] It can be seen that the one-dimensional convolutional function is different from the previous one in terms of specific settings.

[0061] Taking the first self-attention module in the model (located at the Figure 2 bottom) as an example, the size of the feature map of the input module is . The size of the convolutional kernel of the convolutional layer in the embedding layer is 13, the convolutional stride is 4, and the number of convolutional kernels is 2048, obtaining the encoded feature map . The self-attention module uses 12 self-attention layers, the hidden layer dimension is 256, the number of repeated operations is 8, the size of the convolutional kernel of the convolutional layer in the reduction layer is 13, the convolutional stride is 1, and the number of convolutional kernels is 512. The upsampling rate of the upsampling layer is 4. Therefore, the size of the final output feature map of the self-attention module is .

[0062] Similar to the previous embodiments, it can be seen that the embodiment here is the specific working logic of the self-attention module of the present application. Starting from the quantization formula level, a set of specific implementation supporting solutions is given, which can preferably implement the self-attention module function involved in the self-attention module mechanism designed in the model architecture of the present application, with specific and preferably practicality.

[0063] The above content is an introduction to the specific model structure / working logic involved in the multi-nuclear magnetic resonance phosphorus spectrum data processing model of the present application. It is easy to understand that in specific applications, as mentioned above, it may involve prior model training processing. In this regard, the method of the present application may also involve corresponding model training links.

[0064] Correspondingly, as an exemplary embodiment, the method of the present application may further include: Obtain labeled sample multi-nuclear magnetic resonance phosphorus spectrum data; Based on the sample multi-nuclear magnetic resonance phosphorus spectrum data, train the multi-nuclear magnetic resonance phosphorus spectrum data processing model.

[0065] The labeling process involved therein can be understood as configuring the theoretical or standard model processing results for the sample multi-nuclear magnetic resonance phosphorus spectrum data, that is, high-resolution multi-nuclear magnetic resonance phosphorus spectrum data, to assist in training the model's pertinence for multi-nuclear magnetic resonance phosphorus spectrum data processing during the model training process.

[0066] In specific operations, the labeling process can be completed either manually or by handing it over to corresponding automated labeling tools. When using automated labeling tools in specific applications, the corresponding automated labeling logic also needs to be pre-configured.

[0067] Among them, the labeling can specifically also use ready-made high-resolution data.

[0068] As an example, for the sample multi-nuclear magnetic resonance phosphorus spectrum data and its labeling, multi-nuclear magnetic resonance phosphorus spectrum data can be specifically collected based on a clinical 3T multi-nuclear magnetic resonance system and a dual-tuned (1H / 31P) "birdcage" coil. The scanning protocol includes a high-resolution 3D T1-weighted structural image and a low-time-resolution 2D-CSI-31P-MRS sequence and a high-time-resolution 2D-CSI-31P-MRS sequence.

[0069] The scanning parameters of the 3D T1-weighted structural image are as follows: TR / TE = 4.3 / 1.97 ms, voxel = 2 × 2.15 × 2 mm3, acquisition time = 3 minutes and 39 seconds. The structural image is used to place the MRS acquisition voxel frame.

[0070] The scanning parameters for low temporal resolution MRS are: TR / TE = 3500 / 0.24 ms, acquisition bandwidth = 3000 Hz, number of acquisition points = 2048, number of excitations (NSA) = 6, using automatic first-order pencil shimming, voxel = 20×20×40 mm3, a total of 256 voxels are acquired, and the scanning time is 9 minutes and 24 seconds.

[0071] High temporal resolution MRS phosphorus spectra are obtained by increasing the number of excitations. The number of excitations is increased from 6 to 30, and the other parameters remain unchanged. The scanning time is 40 minutes. The two MRS acquisitions maintain the same localization. The thalamus region is selected to be at the center of the CSI grid, and the grid is parallel to the anterior-posterior commissure line of the corpus callosum in the sagittal plane.

[0072] Among them, each file is acquired 100 times repeatedly to obtain a data set , where is the average of 10 random repeated acquisitions, is the average of 100 repeated acquisitions. Among them, the input data and the output have a length of .

[0073] In this way, the completed labeled sample multi-nuclear magnetic resonance phosphorus spectrum data can be used for specific model training.

[0074] During the model training process, it generally includes the following: In each model training session, a training sample is input into the model, enabling the model to perform corresponding multi-nuclear magnetic resonance phosphorus spectrum data processing to achieve forward propagation. Then, based on the multi-nuclear magnetic resonance phosphorus spectrum data results output by the model, the loss function is calculated in combination with the labels, and the model parameters are optimized according to the calculation results of the loss function to achieve backward propagation. After a large number of trainings like this, if the preset model training requirements such as the number of training times, training duration, or prediction accuracy are met, the model training can be completed, and a multi-nuclear magnetic resonance phosphorus spectrum data processing model that can be put into practical application can be obtained.

[0075] Among them, it can be understood that the specific model training architecture and the specific loss function adopted during the training process can either adopt the existing scheme, or be further optimized and improved on the basis of the existing scheme, or a novel self-developed scheme can be adopted. These are all acceptable and can be configured according to the actual situation.

[0076] As an example, as a model training architecture, for the sample multi-nuclear magnetic resonance phosphorus spectrum data, it can be specifically divided into a training set and a test set , and The ratio is 7:3. Further, is evenly divided into K subsets (K≥5) to ensure that the data distribution of each subset is similar. During each training, one of the subsets is used as the validation set, and the other K - 1 subsets are used as the training set.

[0077] K-fold cross-validation can be used for model training. Specifically, first, initialize the hyperparameters of the model, including the learning rate, batch size, number of iterations, etc. Then, perform K rounds of training and validation on the divided K subsets. In each round, without repetition, select 1 subset as the validation set to evaluate the model performance, and the other K - 1 subsets are used as the training set to train the model. During model training, the optimizer is used to perform gradient descent to approximate the optimal value of the loss function, thereby adjusting the network weight matrix of the entire model. The termination condition of the training process is to reach the upper limit of the number of iterations or the loss function value no longer decreases within n iteration cycles. After training, select performance evaluation metrics such as accuracy, precision, sensitivity, specificity, AUC, F1-score, etc. to measure the model performance, and obtain the performance evaluation results of K models on the validation set. The overall performance and performance stability of the model are reflected by calculating the average value and standard deviation of the K performance evaluation results.

[0078] For the hyperparameters among them, model hyperparameter optimization can be carried out through grid search. Specifically, first define the value range of the hyperparameters to be optimized (such as learning rate, batch size, number of iterations, etc.). The value range can be a continuous interval or a discrete set. Then, combine the value ranges of each hyperparameter to generate a hyperparameter grid. Perform K-fold cross-validation on each hyperparameter grid and record the model performance evaluation results. Compare the performance evaluation results corresponding to all hyperparameter grids by selecting single or multiple evaluation metrics, and finally find the best hyperparameter combination and .

[0079] As an example, the model of the present application can be specifically implemented on the PyTorch platform, trained using an NVIDIA GeForce GTX 3090 Ti (GPU 24GB) graphics card, and adopt a five-fold cross-validation strategy. The model uses the Adam optimizer, and the loss function uses the cross-entropy loss function. The quantitative evaluation indicators include: accuracy, precision, sensitivity, specificity, AUC, and F1. Grid search is used to compare the five-fold cross-validation results of different hyperparameter combinations. The range of the learning rate is 0.00001 - 0.0001, with an interval of 0.00001; the range of the batch size is 2 - 16, with an interval of 2; the range of the total number of iterations is 100 - 400, with an interval of 50. Finally, the model hyperparameter combination is determined as: learning rate 0.00005, batch size 8, and number of iterations 300. Under the setting of the final model hyperparameter combination, the training set is trained until the value of the loss function no longer decreases within 10 epochs or the total number of iterations is completed. The model with the smallest loss function on the training set is selected for testing to make the output result after training more accurate.

[0080] Finally, for the above solution content, generally speaking, aiming at the goal of improving the resolution of multi-nuclear magnetic resonance phosphorus spectrum data, the present application designs a novel multi-nuclear magnetic resonance phosphorus spectrum data processing solution and builds a corresponding specific model processing architecture based on deep learning. In this way, the resolution of multi-nuclear magnetic resonance phosphorus spectrum data can be effectively improved, providing high-quality data support for related applications based on multi-nuclear magnetic resonance phosphorus spectrum data, and further ensuring the improvement of the corresponding medical service quality.

[0081] The above is the introduction of the multi-nuclear magnetic resonance phosphorus spectrum data processing method provided by the present application. To facilitate the better implementation of the multi-nuclear magnetic resonance phosphorus spectrum data processing method provided by the present application, the present application also provides a multi-nuclear magnetic resonance phosphorus spectrum data processing device from the perspective of functional modules.

[0082] Refer to Figure 5 , Figure 5 which is a schematic structural diagram of a multi-nuclear magnetic resonance phosphorus spectrum data processing device of the present application. In the present application, the multi-nuclear magnetic resonance phosphorus spectrum data processing device 500 may specifically include the following structure: The first acquisition unit 501 is used to acquire the first multi-nuclear magnetic resonance phosphorus spectrum data whose resolution needs to be improved currently. A processing unit 502 is configured to input the first multi-nuclear magnetic resonance phosphorus spectrum data into a multi-nuclear magnetic resonance phosphorus spectrum data processing model for processing. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is used to improve the resolution of the multi-nuclear magnetic resonance phosphorus spectrum data input to the model. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is pre-trained by using labeled sample multi-nuclear magnetic resonance phosphorus spectrum data. The multi-nuclear magnetic resonance phosphorus spectrum data processing model includes an input side and an output side. The input side includes 5 first dense convolutional modules and 5 max pooling layers. The first dense convolutional modules are used for feature extraction, and the max pooling layers are used for downsampling. The first first dense convolutional module corresponds to the model input. The first dense convolutional modules and the max pooling layers are alternately connected. The 5th max pooling layer corresponds to the output of the input side. The output side includes 5 self-attention modules, 5 second dense convolutional modules, and 4 upsampling layers. The self-attention modules are used for encoding, the second dense convolutional modules are used for decoding, and the upsampling layers are used for upsampling. The second dense convolutional modules and the upsampling layers are alternately connected. The 5th second dense convolutional module corresponds to the model output. Each first dense convolutional module is connected to a second dense convolutional module with an inverted order relationship through a self-attention module; A second acquisition unit 503 is configured to acquire second multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model and corresponding to the first multi-nuclear magnetic resonance phosphorus spectrum data.

[0083] In an exemplary embodiment, the pooling size of the max pooling layer is 2, the pooling stride is 2, and the sampling multiple of the upsampling layer is 2; The output feature map sizes of the 5 first dense convolutional modules are, in sequence, 、 、 、 and ; The output feature map sizes of the 5 self-attention modules are, in sequence, 、 、 、 and ; The output feature map sizes of the 5 second dense convolutional modules are, in sequence, 、 、 、 and .

[0084] In another exemplary embodiment, for both the first dense convolutional module and the second dense convolutional module, it consists of a convolutional layer, a normalization layer, and an activation layer.

[0085] In another exemplary embodiment, for the first dense convolutional module and the second dense convolutional module, there is: Assume that the input feature map is , , is the length of the feature map, is the number of channels of the feature map, and the output feature map is , , specifically obtained by the following formula: , , , where is the feature map after the first layer of convolution, , is the feature map after the second layer of convolution, , is the number of convolution kernels, is the ReLU activation function, is the batch normalization function, is the one-dimensional convolution function, is the concatenation function.

[0086] In another exemplary embodiment, for the self-attention module, it consists of an embedding layer, multiple self-attention layers, a concatenation layer, a normalization layer, and a reduction layer. The multiple self-attention layers include self-attention layers.

[0087] In another exemplary embodiment, for the self-attention module, there is: Assume that the input feature map is , , is the length of the feature map, is the number of convolution kernels, and the output feature map is , , specifically obtained by the following processing content: 1) , where is the deformation function, transforms from the shape of to , is the one-dimensional convolution function, , , , is the convolution layer stride, is the number of convolution kernels, indicates that the embedding layer divides the output feature map of the first dense convolution module into equal parts for re-encoding; 2) , wherein , , are position encodings; 3) , wherein is a layer normalization function, is a concatenation function, is the output set of self-attention layers, , , , , , is a normalization exponential function, , and are respectively the weight matrices of the query matrix, the key matrix and the value matrix in the th self-attention layer, , and , is the hidden layer dimension; 4) Repeat the operation in 3) to obtain the feature vectors , ; 5) , wherein is a deformation function, transforms the shape of the feature vector from to , is an upsampling function, and the upsampling rate is . Here, the number of convolution kernels of the one-dimensional convolution function is .

[0088] In another exemplary embodiment, the apparatus further includes a model training unit 504, configured to: Obtain labeled sample multi-nuclear magnetic resonance phosphorus spectrum data; Based on the sample multi-nuclear magnetic resonance phosphorus spectrum data, train a multi-nuclear magnetic resonance phosphorus spectrum data processing model.

[0089] This application also provides a processing device from the perspective of the hardware structure. Refer to Figure 6 , Figure 6A schematic structural diagram of the processing device of the present application is shown. Specifically, the processing device of the present application may include a processor 601, a memory 602, and an input / output device 603. When the processor 601 executes the computer program stored in the memory 602, it realizes the steps of the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment; or, when the processor 601 executes the computer program stored in the memory 602, it realizes the functions of each unit in the corresponding embodiment. The memory 602 is used to store the computer program required for the processor 601 to execute the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment. Figure 1 When the processor 601 executes the computer program stored in the memory 602, it realizes the steps of the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment; or, when the processor 601 executes the computer program stored in the memory 602, it realizes the functions of each unit in the corresponding embodiment. Figure 5 The memory 602 is used to store the computer program required for the processor 601 to execute the above Figure 1 corresponding embodiment of the multi-nuclear magnetic resonance phosphorus spectrum data processing method.

[0090] Exemplarily, the computer program may be divided into one or more modules / units. One or more modules / units are stored in the memory 602 and executed by the processor 601 to complete the present application. One or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.

[0091] The processing device may include, but is not limited to, the processor 601, the memory 602, and the input / output device 603. Those skilled in the art can understand that the schematic diagram is only an example of the processing device and does not constitute a limitation on the processing device. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the processing device may also include a network access device, a bus, etc. The processor 601, the memory 602, the input / output device 603, etc. are connected through a bus.

[0092] The processor 601 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc. The processor is the control center of the processing device and connects various parts of the entire device through various interfaces and lines.

[0093] The memory 602 can be used to store computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory 602 and calling the data stored in the memory 602, the processor 601 realizes various functions of the computer device. The memory 602 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc.; the data storage area can store data created according to the use of the processing device, etc. In addition, the memory can include high-speed random access memory and can also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.

[0094] When the processor 601 is used to execute the computer program stored in the memory 602, the following functions can be specifically realized: Obtain first multi-nuclear magnetic resonance phosphorus spectrum data whose resolution is to be improved currently; Input the first multi-nuclear magnetic resonance phosphorus spectrum data into a multi-nuclear magnetic resonance phosphorus spectrum data processing model for processing. Among them, the multi-nuclear magnetic resonance phosphorus spectrum data processing model is used to improve the resolution of the multi-nuclear magnetic resonance phosphorus spectrum data input into the model. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is pre-trained by the labeled sample multi-nuclear magnetic resonance phosphorus spectrum data. The multi-nuclear magnetic resonance phosphorus spectrum data processing model includes an input side and an output side. The input side includes 5 first dense convolution modules and 5 max pooling layers. The first dense convolution modules are used for feature extraction, and the max pooling layers are used for downsampling. The first first dense convolution module corresponds to the model input. The first dense convolution modules and the max pooling layers are alternately connected. The 5th max pooling layer corresponds to the output of the input side. The output side includes 5 self-attention modules, 5 second dense convolution modules, and 4 upsampling layers. The self-attention modules are used for encoding, the second dense convolution modules are used for decoding, and the upsampling layers are used for upsampling. The second dense convolution modules and the upsampling layers are alternately connected. The 5th second dense convolution module corresponds to the model output. Each first dense convolution module is connected to a second dense convolution module with an inverted order relationship of its own module order through a self-attention module; Obtain second multi-nuclear magnetic resonance phosphorus spectrum data corresponding to the first multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model.

[0095] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described multi-nuclear magnetic resonance phosphorus spectrum data processing device, processing equipment, and their corresponding units can be referred to as Figure 1For the description of the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment, it will not be elaborated here specifically.

[0096] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling relevant hardware through instructions. The instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0097] For this reason, the present application provides a computer-readable storage medium, in which multiple instructions are stored, and the instructions can be loaded by a processor to execute the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment of the present application as Figure 1 the steps of the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment. For specific operations, reference can be made to Figure 1 the description of the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment, which will not be elaborated here.

[0098] Among them, the computer-readable storage medium may include: Read Only Memory (ROM), Random Access Memory (RAM), a magnetic disk or an optical disc, etc.

[0099] Since the instructions stored in the computer-readable storage medium can execute the steps of the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment of the present application as Figure 1 the steps of the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment, therefore, the beneficial effects that can be achieved by the multi-nuclear magnetic resonance phosphorus spectrum data processing method in the corresponding embodiment of the present application can be realized. For details, refer to the previous description, which will not be elaborated here. Figure 1 For details, refer to the previous description, which will not be elaborated here.

[0100] The multi-nuclear magnetic resonance phosphorus spectrum data processing method, device, processing equipment, and computer-readable storage medium provided by the present application have been introduced in detail above. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for processing multinuclear magnetic resonance phosphorus spectrum data, characterized in that, The method includes: Obtaining first multi-nuclear magnetic resonance phosphorus spectrum data of the current to-be-improved resolution; Inputting the first multi-nuclear magnetic resonance phosphorus spectrum data into a multi-nuclear magnetic resonance phosphorus spectrum data processing model for processing, where the multi-nuclear magnetic resonance phosphorus spectrum data processing model is used to improve the resolution of the multi-nuclear magnetic resonance phosphorus spectrum data input into the model. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is pre-trained by labeled sample multi-nuclear magnetic resonance phosphorus spectrum data. The multi-nuclear magnetic resonance phosphorus spectrum data processing model includes an input side and an output side. The input side includes 5 first dense convolutional modules and 5 max pooling layers. The first dense convolutional modules are used for feature extraction, and the max pooling layers are used for downsampling. The 1st first dense convolutional module corresponds to the model input. The first dense convolutional modules and the max pooling layers are alternately connected. The 5th max pooling layer corresponds to the output of the input side. The output side includes 5 self-attention modules, 5 second dense convolutional modules, and 4 upsampling layers. The self-attention modules are used for encoding, the second dense convolutional modules are used for decoding, and the upsampling layers are used for upsampling. The second dense convolutional modules and the upsampling layers are alternately connected. The 5th second dense convolutional module corresponds to the model output. Each first dense convolutional module is connected to the second dense convolutional module with an inverted order relationship with its own module order through the self-attention module; Obtaining second multi-nuclear magnetic resonance phosphorus spectrum data corresponding to the first multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model.

2. The method according to claim 1, wherein The pooling size of the max pooling layer is 2, and the pooling stride is 2. The upsampling multiple of the upsampling layer is 2; The output feature map sizes of the five first dense convolution modules are, in sequence, , , , and ; The output feature map sizes of the 5 self-attention modules are, in sequence, , , , and ; The output feature map sizes of the five second dense convolution modules are, in sequence, , , , and .

3. The method according to claim 1, wherein For both the first dense convolutional module and the second dense convolutional module, they are composed of a convolutional layer, a normalization layer, and an activation layer.

4. The method according to claim 1, wherein For the first dense convolutional module and the second dense convolutional module, there is: Assume the input feature map is , , , where is the length of the feature map, is the number of channels of the feature map, and the output feature map is , , which is specifically obtained by the following formula: ， ， ， Among them, is the feature map after the first layer of convolution, , is the feature map after the second layer of convolution, , is the number of convolutional kernels, is the ReLU activation function, is the batch normalization function, is the one-dimensional convolution function, is the concatenation function.

5. The method according to claim 1, wherein For the self-attention module, it consists of an embedding layer, a multi-head self-attention layer, a splicing layer, a normalization layer, and a reduction layer. The multi-head self-attention layer includes self-attention layers.

6. The method according to claim 5, wherein For the self-attention module, there is: Assume the input feature map is , , is the length of the feature map, is the number of convolutional kernels, and the output feature map is , , which is specifically obtained by the following processing content: 1） ， Among them, is a deformation function, transforms the shape of from to , is a one-dimensional convolution function, , , , is the convolution layer stride, is the number of convolution kernels, indicates that the embedding layer divides the output feature map of the first dense convolution module into equal parts for re-encoding; 2） ， Among them, , , are position encodings; 3） ， Among them, is the layer normalization function, is the concatenation function, is the output set of the said self-attention layers, is the multi-head self-attention weight matrix, , , , , , is the normalization exponential function, , and are respectively the weight matrices of the query matrix, key matrix and value matrix in the th self-attention layer, , and , is the hidden layer dimension; 4) Repeat the operation in 3) to obtain the feature vector , ; 5） ， Among them, is a deformation function, transforms the shape of the feature vector from to , is an upsampling function, and the upsampling rate is . Here, the number of convolution kernels of the one-dimensional convolution function is .

7. The method according to claim 1, wherein The method further includes: Obtaining the labeled sample multi-nuclear magnetic resonance phosphorus spectrum data; Training the magnetic resonance phosphorus spectrum data processing model based on the sample multi-nuclear magnetic resonance phosphorus spectrum data.

8. A multi-nuclear magnetic resonance phosphorus spectrum data processing device, characterized in that, The device includes: A first acquisition unit for acquiring first multi-nuclear magnetic resonance phosphorus spectrum data of the current to-be-improved resolution; A processing unit for inputting the first multi-nuclear magnetic resonance phosphorus spectrum data into a multi-nuclear magnetic resonance phosphorus spectrum data processing model for processing, wherein the multi-nuclear magnetic resonance phosphorus spectrum data processing model is used to improve the resolution of the multi-nuclear magnetic resonance phosphorus spectrum data input into the model. The multi-nuclear magnetic resonance phosphorus spectrum data processing model is pre-trained by labeled sample multi-nuclear magnetic resonance phosphorus spectrum data. The magnetic resonance phosphorus spectrum data processing model includes an input side and an output side. The input side includes 5 first dense convolutional modules and 5 max pooling layers. The first dense convolutional module is used for feature extraction, and the max pooling layer is used for downsampling. The first of the first dense convolutional modules corresponds to the model input. The first dense convolutional module and the max pooling layer are alternately connected. The fifth max pooling layer corresponds to the output of the input side. The output side includes 5 self-attention modules, 5 second dense convolutional modules, and 4 upsampling layers. The self-attention module is used for encoding, the second dense convolutional module is used for decoding, and the upsampling layer is used for upsampling. The second dense convolutional module and the upsampling layer are alternately connected. The fifth second dense convolutional module corresponds to the model output. Each of the first dense convolutional modules is connected to the second dense convolutional module with an inverted order relationship through the self-attention module; A second acquisition unit for acquiring second multi-nuclear magnetic resonance phosphorus spectrum data output by the multi-nuclear magnetic resonance phosphorus spectrum data processing model and corresponding to the first multi-nuclear magnetic resonance phosphorus spectrum data.

9. A processing device, characterized in that, It includes a processor and a memory. A computer program is stored in the memory. When the processor calls the computer program in the memory, it executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by the processor to execute the method according to any one of claims 1 to 7.

Images