A fast system matrix calibration algorithm for magnetic particle imaging

Abstract

The application belongs to the field of magnetic particle imaging, and particularly relates to a fast system matrix calibration algorithm applied to magnetic particle imaging, and aims to solve the problem that the prior art cannot simultaneously consider fast calibration of a system matrix and high-resolution imaging. The application comprises the following steps: acquiring a public data set, screening a high signal-to-noise ratio system matrix and converting the high signal-to-noise ratio system matrix into a three-dimensional cubic format, and dividing the three-dimensional cubic format into a training set and a test set according to scanning parameters; collecting a frequency domain response of a magnetic nanoparticle point phantom under a scanning / excitation magnetic field through actual measurement, and generating a measured system matrix set through moving sampling and frequency point screening; constructing a magnetic particle concentration distribution matrix, generating a simulated system matrix set through numerical simulation and frequency point screening; constructing a deep learning model, including an image encoding unit, a feature processing unit and an image reconstruction unit; inputting a matrix to be calibrated into the trained model, and outputting a calibrated high-resolution system matrix. The method significantly improves calibration efficiency and maintains imaging resolution.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of magnetic particle imaging, and particularly relates to a fast system matrix calibration algorithm applied to magnetic particle imaging. BACKGROUND

[0002] Magnetic particle imaging (MPI) is a tomographic imaging technique that can detect the magnetism of iron oxide nanoparticles injected into the blood to generate three-dimensional images. In recent years, due to its characteristics of improving imaging resolution while not using radioactive substances, the magnetic particle imaging technology has been widely studied in the fields of multi-modal in vivo imaging, blood vessel imaging, cell tracking, cell therapy monitoring, targeted drug delivery, tumor imaging, nanoparticle development, etc., and has made breakthrough progress in clinical transformation.

[0003] There are two traditional MPI reconstruction methods: X-space-based method and system matrix (SM)-based method. Compared with the X-space-based method, the SM-based method achieves higher image quality, but the system matrix measurement takes a long time. This method requires a complex-valued SM, which is obtained through a time-consuming calibration process involving measuring the voxel-level response of the MPI scanner within the field of view (FOV), which is very time-consuming. At the same time, the long operation of the MPI scanner can cause overheating, resulting in thermal noise and negatively affecting the imaging performance.

[0004] Sparse sampling can accelerate calibration, for example, it takes about 32 hours to obtain a 37x37x37 size system matrix, while a 9x9x9 system matrix only takes about 37 minutes. However, this method often leads to a decrease in system matrix and image resolution. Therefore, there is an urgent need for an effective solution to balance the trade-off between faster calibration acceleration and maintaining high image resolution.

[0005] Based on this, the application provides a fast system matrix calibration algorithm applied to magnetic particle imaging. SUMMARY

[0006] In order to solve the above problems in the prior art, i.e., the problem that the prior art cannot simultaneously consider fast calibration of the system matrix and high-resolution imaging, the application provides a fast system matrix calibration algorithm applied to magnetic particle imaging, which comprises:

[0007] obtaining a public system matrix dataset, screening effective system matrix data with a signal-to-noise ratio higher than a set threshold, and uniformly converting and merging the effective system matrix data into a first set in a three-dimensional cubic format; dividing the first set into a training dataset and a test dataset based on different scanning parameters;

[0008] In the magnetic particle imaging device, a magnetic nanoparticle point phantom is placed in the center of the imaging region, and a scanning magnetic field and an excitation magnetic field are applied; by gradually moving the magnetic nanoparticle point phantom and screening effective frequency points, a set of measured system matrices is obtained;

[0009] A magnetic particle concentration distribution matrix is constructed and an initial value is set, a simulation system matrix is generated by numerical simulation calculation, and a set of simulation system matrices is obtained after screening effective frequency points;

[0010] A deep learning model is constructed, and the deep learning model is trained using the training data set, the deep learning model comprising:

[0011] An image encoding unit for feature extraction of an input system matrix, and separating and outputting real part features and imaginary part features of the input system matrix;

[0012] A feature processing unit for separating and enhancing high-frequency and low-frequency features of the real part features and the imaginary part features, respectively, to generate real part enhanced features and imaginary part enhanced features;

[0013] An image reconstruction unit for upsampling and fusing the real part enhanced features and the imaginary part enhanced features to generate a high-resolution system matrix;

[0014] The trained deep learning model is input into the trained deep learning model, and a calibrated high-resolution system matrix is output.

[0015] Further, the image encoding unit comprises:

[0016] An initial feature extraction module for preliminary feature extraction of an input system matrix;

[0017] A feature enhancement module comprising at least one dense block comprising a residual connection for refining and enhancing the preliminary extracted features;

[0018] A first frequency separation enhancement module for receiving refined and enhanced features and separating them into real part features and imaginary part features for output.

[0019] Further, the feature processing unit comprises a real part branch and an imaginary part branch working in parallel, the real part branch being used for processing the real part features, and the imaginary part branch being used for processing the imaginary part features;

[0020] The real part branch and the imaginary part branch both comprise a second frequency separation enhancement module and a feature addition layer;

[0021] The second frequency separation enhancement module is used for separating and enhancing high-frequency features and low-frequency features;

[0022] The feature addition layer is configured to add the enhanced high-frequency features and the enhanced low-frequency features element by element to obtain real part enhanced features and imaginary part enhanced features, and input the real part enhanced features and the imaginary part enhanced features into the image reconstruction unit.

[0023] Further, the first frequency separation enhancement module and the second frequency separation enhancement module each include:

[0024] The group normalization layer is configured to receive the input features and perform channel group normalization processing on the input features.

[0025] The frequency domain weighting layer is configured to perform a Softmax operation on the features after the group normalization processing to generate a frequency domain weighting representation.

[0026] The weight separation layer is configured to multiply the frequency domain weighting representation and the features after the group normalization processing element by element, and then perform normalization on the multiplied features to obtain a normalized weighting representation.

[0027] The threshold gating layer is configured to gate the normalized weighting representation with a set threshold to generate high-frequency weights and low-frequency weights.

[0028] The feature mapping layer is configured to map the input features to high-frequency components according to the high-frequency weights, and map the input features to low-frequency components according to the low-frequency weights.

[0029] The high-frequency processing layer is configured to perform group convolution and point-by-point convolution on the high-frequency components synchronously, and output fused enhanced high-frequency features.

[0030] The low-frequency processing layer is configured to perform point-by-point convolution on the low-frequency components, and concatenate the output of the point-by-point convolution and the low-frequency components to generate enhanced low-frequency features.

[0031] Further, the image reconstruction unit includes:

[0032] The up-sampling module is configured to perform spatial dimension lifting operations on the real part enhanced features and the imaginary part enhanced features, respectively.

[0033] The feature fusion module is configured to fuse the real part features and the imaginary part features after the up-sampling processing to generate the high-resolution system matrix.

[0034] Further, the up-sampling module specifically includes:

[0035] The feature extraction submodule is configured to perform feature extraction on the up-sampled input features.

[0036] The spatial lifting submodule includes a cascaded transpose convolution layer and a standard convolution layer, and is configured to gradually multiply the spatial dimensions of the features.

[0037] Further, the system matrix with a frequency point signal-to-noise ratio lower than a set threshold is determined as invalid data and is removed.

[0038] Further, the calibrated high-resolution system matrix is used for magnetic particle imaging reconstruction to improve the resolution of the reconstructed image.

[0039] Further, the frequency domain weighting layer is a Softmax activation function layer.

[0040] Further, the normalized weighted representation is realized by a Sigmoid activation function layer.

[0041] The beneficial effects of the present application are:

[0042] First, a multi-source data collaborative training mechanism is used to integrate public data set screening, measured data acquisition and simulation data generation, effectively enhancing the generalization ability of the model, so that the calibration algorithm adapts to the complex and variable actual imaging environment.

[0043] Second, the real part and the imaginary part features of the system matrix are independently processed by a complex domain feature separation architecture, which completely retains the physical response information and avoids the loss of high-frequency details and low-frequency structure information caused by the real number operation of the complex matrix in the traditional method.

[0044] Further, a frequency domain guided double-branch feature enhancement mechanism is adopted to specifically strengthen the effective frequency band features and suppress the noise interference introduced by equipment overheating, significantly improving the signal-to-noise ratio of the output matrix; in the reconstruction stage, the cascaded upsampling operation combined with physical constraints is used to realize the spatial dimension promotion by the cooperation of transpose convolution and standard convolution, breaking through the resolution limit of sparse sampling and realizing high-quality reconstruction with edge sharpness preservation.

[0045] Finally, an end-to-end deep learning calibration framework is constructed, which optimizes the traditional multi-step calibration process into an integrated model and eliminates the error accumulation of intermediate links. This scheme greatly shortens the system matrix acquisition time while ensuring high-resolution imaging, and avoids the problem of thermal noise caused by long-term operation of the equipment from the source. BRIEF DESCRIPTION OF DRAWINGS

[0046] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0047] Figure 1 is a process schematic diagram of a fast system matrix calibration algorithm applied to magnetic particle imaging of the present application;

[0048] Figure 2 is a structure schematic diagram of a deep learning model of a fast system matrix calibration algorithm applied to magnetic particle imaging of the present application;

[0049] Figure 3 is a structure schematic diagram of a feature processing unit in a fast system matrix calibration algorithm applied to magnetic particle imaging of the present application;

[0050] Figure 4 is a reconstruction result image based on a calibrated system matrix in a fast system matrix calibration algorithm applied to magnetic particle imaging of the application;

[0051] Figure 5 is a structural schematic diagram of a computer system of a server for implementing the method, system and device embodiments of the application. DETAILED DESCRIPTION

[0052] The application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the related application, and not to limit the application. In addition, it should be noted that only the parts related to the application are shown in the drawings for ease of description.

[0053] It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict. The application will be described in detail below with reference to the accompanying drawings and embodiments.

[0054] The first embodiment of the application provides a fast system matrix calibration algorithm applied to magnetic particle imaging, which comprises the following steps:

[0055] Step S10, obtain a public system matrix data set, filter effective system matrix data with a signal-to-noise ratio higher than a set threshold, and uniformly convert to a three-dimensional cubic format and merge into a first set; based on different scanning parameters, the first set is divided into a training data set and a test data set;

[0056] Step S20, in a magnetic particle imaging device, place a magnetic nanoparticle point phantom in the center of the imaging area, apply a scanning magnetic field and an excitation magnetic field; by gradually moving the magnetic nanoparticle point phantom and filtering effective frequency points, a measured system matrix set is obtained;

[0057] Step S30, construct a magnetic particle concentration distribution matrix and set an initial value, generate a simulation system matrix through numerical simulation calculation, and obtain a simulation system matrix set after filtering effective frequency points;

[0058] Step S40, construct a deep learning model, train the deep learning model using the training data set, and the deep learning model comprises:

[0059] An image encoding unit is configured to extract features from an input system matrix, and separate and output real part features and imaginary part features of the input system matrix;

[0060] A feature processing unit is configured to separately separate and enhance the real part features and the imaginary part features to generate real part enhanced features and imaginary part enhanced features;

[0061] an image reconstruction unit configured to up-sample and fuse the real part enhanced feature and the imaginary part enhanced feature to generate a high-resolution system matrix;

[0062] Step S50, inputting the to-be-calibrated system matrix into the trained deep learning model to output a calibrated high-resolution system matrix.

[0063] In order to more clearly describe the fast system matrix calibration algorithm applied to magnetic particle imaging, the following will be described in combination with Figure 1 The steps in the embodiments of the present application will be described in detail, including steps S10-S50, and the detailed description of each step is as follows:

[0064] Step S10, obtaining a public system matrix dataset, screening effective system matrix data with a signal-to-noise ratio higher than a set threshold, and uniformly converting into a three-dimensional cubic format and merging into a first set; based on different scanning parameters, the first set is divided into a training dataset and a test dataset;

[0065] Among them, the system matrix with a frequency point signal-to-noise ratio lower than the set threshold is determined as invalid data and is removed.

[0066] In the implementation process of step S10, first, the public magnetic particle imaging system matrix dataset is collected. The signal-to-noise ratio of each system matrix frequency point in the data set is analyzed, and invalid frequency points with a signal-to-noise ratio lower than 3 are strictly removed, and only effective data with a signal-to-noise ratio higher than 3 is retained. Subsequently, all screened system matrices are uniformly converted into a three-dimensional cubic format.

[0067] After completing the format conversion, a plurality of processed system matrices are merged into a first data set S O Based on the parameter configuration of different scanning devices, the data set is divided: all system matrices derived from a complete scanning sequence are classified into the training data set, and all system matrices of another independent scanning sequence are classified into the test data set, ensuring that the training data and the test data are completely separated.

[0068] Step S20, in the magnetic particle imaging device, a magnetic nanoparticle point phantom is placed at the center of the imaging area, and a scanning magnetic field and an excitation magnetic field are applied; by gradually moving the magnetic nanoparticle point phantom and screening effective frequency points, a measured system matrix set is obtained;

[0069] Step S20, in the implementation process, first based on the actual Filed Free Line (FFL) based magnetic particle imaging device, the magnetic nanoparticle point phantom with concentration c is accurately positioned at the center position of the imaging area of the magnetic particle imaging device. After starting the imaging device, the scanning magnetic field and the excitation magnetic field with preset parameters are applied synchronously. Then, according to the preset step and the three-dimensional space path, the point phantom is gradually moved to different spatial coordinate points;

[0070] The original frequency domain response signal is collected at each coordinate point, and the signal-to-noise ratio of the full frequency band signal is analyzed: the signal-to-noise ratio of each frequency point is calculated, all invalid frequency points with a signal-to-noise ratio lower than 3 are removed, and only valid frequency point data is retained. The moving, collecting and screening process is repeated until the entire preset field of view range is covered.

[0071] Finally, all coordinate point effective frequency point data is integrated to form a measured system matrix set S R .

[0072] Step S30, construct a magnetic particle concentration distribution matrix and set an initial value, generate a simulation system matrix through numerical simulation calculation, and obtain a simulation system matrix set after screening effective frequency points;

[0073] In the implementation process of step S30, first, a magnetic particle concentration distribution matrix C is constructed and an initial value is set. A numerical simulation method is used to simulate the magnetic particle imaging physical process: based on the Filed Free Point (FFP) scanning mechanism, corresponding scanning magnetic field and excitation magnetic field parameters are applied in the imaging area, and the frequency domain signal of each spatial position is calculated by solving the magnetic nanoparticle response equation.

[0074] The full frequency band signal generated by simulation is screened for effectiveness: the signal-to-noise ratio value is calculated point by point, and invalid frequency point data with a signal-to-noise ratio lower than 3 is strictly removed, and only valid frequency points are retained. The above simulation calculation and frequency point screening operation is repeated to cover all spatial coordinate points in the preset field of view range.

[0075] Finally, all effective frequency point data is integrated to form a simulation system matrix set. The set is independent of the measured system matrix set and is used to verify the generalization performance of the algorithm.

[0076] After obtaining the above training data set and test data set, a suitable loss function, learning rate, etc. are selected to train the deep learning model proposed in the application. Those skilled in the art can select a suitable loss function according to the actual situation, and the application does not make specific limitations. The construction process of the deep learning model is as follows:

[0077] Referring to Figure 2 and Figure 3, step S40, constructing a deep learning model, training the deep learning model using the training data set, the deep learning model comprising:

[0078] An image encoding unit is configured to perform feature extraction on an input system matrix and separate real part features and imaginary part features of the input system matrix;

[0079] A feature processing unit is configured to separate and enhance high-frequency features and low-frequency features of the real part features and the imaginary part features respectively, to generate real part enhanced features and imaginary part enhanced features;

[0080] An image reconstruction unit is configured to perform up-sampling processing on the real part enhanced features and the imaginary part enhanced features and fuse them to generate a high-resolution system matrix;

[0081] In this embodiment, the image encoding unit is an initial stage of generating a high-resolution image and is consistent with the Fourier transform process. This module aims to effectively extract shallow features, which include:

[0082] An initial feature extraction module is configured to perform preliminary feature extraction on an input system matrix;

[0083] A feature enhancement module includes at least one dense block with a residual connection, configured to refine and enhance the preliminary extracted features;

[0084] A first frequency separation enhancement module is configured to receive the refined and enhanced features and separate them into real part features and imaginary part features for output.

[0085] Specifically, in the specific implementation process, the image encoding unit performs operations according to the following flow:

[0086] Initial feature extraction: the input system matrix is first subjected to preliminary feature extraction by a 3x3x3 convolution block. The convolution operation captures the basic spatial features of the system matrix to form an initial feature map.

[0087] Feature refinement and enhancement: the initial feature map is input into two cascaded residual-residual dense blocks (RRDB). Each RRDB extracts multi-level features through a dense connection structure and combines residual connection to preserve original information, achieving progressive refinement and enhancement of features, and outputting optimized deep features.

[0088] Complex feature separation: the optimized deep features are input into a first frequency separation enhancement module (FSEM), which maps the input features to a complex domain space according to a learnable frequency domain filtering operation based on the Fourier transform principle;

[0089] The complex representation is optimized by a deep learning network, and the real part features X R and the imaginary part features X I are separated and output.

[0090] The feature processing unit comprises a real branch and an imaginary branch working in parallel, the real branch being configured to process the real part features, and the imaginary branch being configured to process the imaginary part features;

[0091] The real branch and the imaginary branch each comprise a second frequency separation and enhancement module and a feature addition layer;

[0092] The second frequency separation and enhancement module is configured to perform separation and enhancement operations on high-frequency features and low-frequency features;

[0093] The feature addition layer is configured to perform element-wise addition of the enhanced high-frequency features and the enhanced low-frequency features to obtain real part enhanced features and imaginary part enhanced features, and input the real part enhanced features and the imaginary part enhanced features into an image reconstruction unit.

[0094] In the implementation process, the feature processing unit performs operations according to the following flow:

[0095] Double-branch parallel processing: the real branch receives the real part features output by the image encoding unit X R , and the imaginary branch receives the imaginary part features X I ; the two branches independently run using the same network structure and do not share weight parameters with each other.

[0096] Each branch sub-module performs band separation and enhancement operations;

[0097] Branch output generation: the real branch outputs the enhanced real part features X RR , and the imaginary branch outputs the enhanced imaginary part features X IR .

[0098] The first frequency separation and enhancement module and the second frequency separation and enhancement module in the embodiment each comprise:

[0099] A group normalization layer configured to receive input features and perform channel group normalization processing;

[0100] A frequency domain weighting layer configured to perform Softmax operation on the features after the group normalization processing to generate a frequency domain weighted representation;

[0101] A weight separation layer configured to multiply the frequency domain weighted representation and the features after the group normalization processing element by element, and then perform normalization weighting representation on the multiplication result;

[0102] A threshold gating layer configured to generate high-frequency weights and low-frequency weights by gating the normalized weighted representation using a set threshold value;

[0103] a feature mapping layer, configured to map the input features into high-frequency components according to high-frequency weights and into low-frequency components according to low-frequency weights;

[0104] a high-frequency processing layer, configured to synchronously perform grouped convolution and point-wise convolution on the high-frequency components and output fused enhanced high-frequency features;

[0105] a low-frequency processing layer, configured to perform point-wise convolution on the low-frequency components and splice an output of the point-wise convolution with the low-frequency components to generate enhanced low-frequency features.

[0106] The frequency domain weighting layer is a Softmax activation function layer. The normalized weighting is realized by a Sigmoid activation function layer.

[0107] The first and second frequency separation enhancement modules in the embodiment are configured to accurately separate and enhance different frequency components in a feature extraction process. The modules first perform grouped normalization on input features, obtain frequency domain weighting by Softmax operation, and then perform segmentation by Sigmoid function normalization with a 0.5 threshold to generate high-frequency weights and low-frequency weights. The input features are mapped into high-frequency features and low-frequency features according to the weights. The high-frequency branch adopts parallel processing of grouped convolution (GWC) and point-wise convolution (PWC) to retain local details such as edges and textures and fuse cross-channel information. The low-frequency branch adopts point-wise convolution to extract global structure information and splice with the original low-frequency features to suppress noise and retain overall shape. Through the nonlinear and learnable frequency separation and enhancement mechanism, the FSEM can simultaneously retain details and global structure in fast system matrix calibration and other tasks, and significantly improve feature expression capability and reconstruction quality.

[0108] The image reconstruction unit comprises:

[0109] an upsampling module configured to perform spatial dimension lifting operations on the real part enhanced features and the imaginary part enhanced features respectively;

[0110] a feature fusion module configured to fuse the real part features and the imaginary part features after the upsampling processing to generate the high-resolution system matrix.

[0111] The upsampling module specifically comprises:

[0112] a feature extraction submodule configured to perform feature extraction on the upsampling input features;

[0113] a spatial lifting submodule comprising cascaded transpose convolution layers and standard convolution layers and configured to gradually multiply the spatial dimensions of the features.

[0114] Specifically, the feature fusion module averages the real part enhanced feature and the imaginary part enhanced feature after upsampling according to the number of channels, and each is divided into two sub-features, each of which contains one half of the original feature channel number;

[0115] The two sub-features of the real part enhanced feature after upsampling are a first real part enhanced sub-feature and a second real part enhanced sub-feature, and the two sub-features of the imaginary part enhanced feature after upsampling are a first imaginary part enhanced sub-feature and a second imaginary part enhanced sub-feature. The first real part enhanced sub-feature and the first imaginary part enhanced sub-feature are added to obtain a first fusion feature, and the second real part enhanced sub-feature and the second imaginary part enhanced sub-feature are added to obtain a second fusion feature. The first fusion feature and the second fusion feature are merged to generate a final high-resolution system matrix.

[0116] In the specific implementation process, the image reconstruction unit performs the following operation process:

[0117] First, the real part enhanced feature is input into the feature extraction sub-module to perform deep feature extraction through the residual-residual dense block; at the same time, the imaginary part enhanced feature is input into the same structure of the independent feature extraction sub-module for processing.

[0118] Subsequently, the spatial lifting operation is performed: a three-dimensional transpose convolution is performed on the extracted real part feature to expand its spatial dimension to twice, and a 3x3x3 standard convolution layer is immediately connected for feature refinement; the cascade operation of the transpose convolution and the standard convolution is repeated U times (where 2 U equals the target upsampling rate) until the real part feature reaches the target spatial dimension. The imaginary part feature is processed synchronously through the same independent cascade operation process, and the weight parameters of the real part and the imaginary part paths are not shared.

[0119] After upsampling, the processed real part feature and the imaginary part feature are input into the feature fusion module: the two are combined in the form of real part plus imaginary part according to the complex relationship, and finally the complex feature is fused through a single layer 3x3x3 convolution operation to generate a final high-resolution system matrix.

[0120] This process strictly follows three technical constraints: a 3x3x3 standard convolution must be connected after each transpose convolution; the number of cascade operations U is uniquely determined by the target upsampling rate; the real part and the imaginary part processing path maintain parameter independence throughout.

[0121] Step S50, input the to-be-calibrated system matrix into the trained deep learning model, and output a calibrated high-resolution system matrix.

[0122] The calibrated high-resolution system matrix is used for magnetic particle imaging reconstruction to improve the resolution of the reconstructed image.

[0123] We use the algorithm of the present application in the FFP MPI system to improve the quality of image reconstruction. The measured low-resolution SM is used to generate a high-resolution SM, and the measured SM and the generated high-resolution SM are used to reconstruct images, respectively. The reconstruction results of the phantom are shown in FIG. 8. Figure 4 The image reconstructed from the original SM looks very harsh, while the algorithm of the present application can obtain smoother results in the case of upsampling ratios of 2 and 4. In addition, for the resolution phantom, the image reconstructed using the original low-resolution SM is blurred at the boundary, while the image reconstructed using the high-resolution SM shows better quality in resolution. It is worth noting that the initial SM reconstruction fails to distinguish the gap between the two rod-shaped phantoms, resulting in them appearing to be connected. In contrast, the algorithm of the present application effectively distinguishes them at intervals of 3 mm and 4 mm, with an upsampling ratio of 2 and 4, and the boundary is obvious.

[0124] Although the steps in the above embodiments are described in the above order, it should be understood by those skilled in the art that, in order to achieve the effect of the embodiments, the steps do not have to be executed in such an order, and they can be executed simultaneously (in parallel) or in a reversed order, and these simple changes are within the protection scope of the present application.

[0125] The second embodiment of the present application provides a fast system matrix calibration system applied to magnetic particle imaging, which is used to implement a fast system matrix calibration algorithm applied to magnetic particle imaging, and the system comprises:

[0126] A data set construction module is configured to obtain a public system matrix data set, filter effective system matrix data with a signal-to-noise ratio higher than a set threshold, and uniformly convert and merge into a first set in a three-dimensional cubic format; based on different scanning parameters, the first set is divided into a training data set and a test data set;

[0127] A measured system matrix set construction module is configured to place a magnetic nanoparticle point phantom in the center of an imaging area in a magnetic particle imaging device, apply a scanning magnetic field and an excitation magnetic field; by gradually moving the magnetic nanoparticle point phantom and filtering effective frequency points, a measured system matrix set is obtained;

[0128] A simulated system matrix set construction module is configured to construct a magnetic particle concentration distribution matrix and set an initial value, generate a simulated system matrix through numerical simulation calculation, and obtain a simulated system matrix set after filtering effective frequency points;

[0129] A model construction and training module is configured to construct a deep learning model, and train the deep learning model using the training data set, wherein the deep learning model comprises:

[0130] An image encoding unit is configured to perform feature extraction on an input system matrix, and separate real part features and imaginary part features of the input system matrix;

[0131] A feature processing unit is configured to perform separation and enhancement processing of high-frequency features and low-frequency features on the real part features and the imaginary part features respectively, to generate real part enhanced features and imaginary part enhanced features;

[0132] An image reconstruction unit is configured to perform up-sampling processing on the real part enhanced features and the imaginary part enhanced features and fusion, to generate a high-resolution system matrix;

[0133] The high-resolution system matrix generation module is configured to input a to-be-calibrated system matrix into the trained deep learning model, and output a calibrated high-resolution system matrix.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related description of the system described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0135] It should be noted that the above embodiment provides a fast system matrix calibration algorithm applied to magnetic particle imaging, which is only exemplified by the division of the above functional modules. In actual application, the above functions can be completed by different functional modules according to needs, that is, the modules or steps in the embodiment of the present application are further decomposed or combined, for example, the modules of the above embodiment can be combined into one module, or can be further split into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present application are only for distinguishing the modules and steps, and should not be considered as an improper limitation of the present application.

[0136] The third embodiment of the electronic device of the present application comprises:

[0137] at least one processor; and

[0138] a memory in communication connection with the at least one processor; wherein

[0139] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to realize the above-mentioned fast system matrix calibration algorithm applied to magnetic particle imaging.

[0140] The fourth embodiment of the computer readable storage medium of the present application stores computer instructions, and the computer instructions are used to be executed by the computer to realize the above-mentioned fast system matrix calibration algorithm applied to magnetic particle imaging.

[0141] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the storage device and the processing device described above and the related descriptions can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0142] Those skilled in the art will realize that the modules, method steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or combinations of both. The software modules, method steps corresponding to the functions can be stored in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0143] Reference is made below to Figure 5 which shows a structural schematic diagram of a computer system of a server for implementing the method, system, and device embodiments of the present application. Figure 5 The server shown is merely an example and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0144] As Figure 5 shown, the computer system includes a central processing unit (CPU) 501 which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 502 or programs loaded from a storage portion 508 into a random access memory (RAM) 503. Various programs and data required for system operation are also stored in the RAM 503. The CPU 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0145] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as necessary. A removable recording medium 511 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 510 as necessary, so that a computer program read out therefrom is installed in the storage section 508 as necessary.

[0146] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 509, and / or installed from the removable medium 511. When the computer program is executed by the central processing unit (CPU) 501, the above-described functions defined in the methods of the present application are performed. It should be noted that the computer readable medium of the present application can be a computer readable signal medium or a computer readable storage medium or any combination of the two. The computer readable storage medium may, for example, be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device. In the present application, the computer readable signal medium can include a data signal carried in a baseband or as part of a carrier wave, in which the computer readable program code is carried. Such a propagated data signal can take any of a variety of forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device. Program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire line, optical fiber cable, RF, etc., or any suitable combination of the above.

[0147] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0148] The computer program instructions can also be loaded onto a computer or other programmable information processing apparatus to cause a series of operations to be performed on the computer or other programmable information processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable information processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0149] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. For example, singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0150] The description and drawings are illustrative and are not intended to be limiting. Numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is available. It is intended that the scope of the application encompass the various modifications and alterations of the application.

[0151] The technical scheme of the present application has been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical schemes after the changes or replacements will all fall within the protection scope of the present application.

Claims

1. A fast system matrix calibration algorithm for magnetic particle imaging, characterized in that, The algorithm includes: Obtain a publicly available system matrix dataset, filter out valid system matrix data with a signal-to-noise ratio higher than a set threshold, and uniformly convert them into a three-dimensional cube format and merge them into a first set; based on different scanning parameters, divide the first set into a training dataset and a test dataset; In a magnetic particle imaging device, a magnetic nanoparticle dot phantom is placed at the center of the imaging area, and a scanning magnetic field and an excitation magnetic field are applied; by gradually moving the magnetic nanoparticle dot phantom and screening effective frequency points, a set of measured system matrices is obtained. A magnetic particle concentration distribution matrix is ​​constructed and initial values ​​are set. A simulation system matrix is ​​generated through numerical simulation calculation. After filtering effective frequency points, a set of simulation system matrices is obtained. The set of measured system matrices and the set of simulation system matrices are used to verify the generalization performance of the trained deep learning model. Construct a deep learning model, train the deep learning model using the training dataset, and the deep learning model includes: An image coding unit is used to extract features from an input system matrix and separate and output the real and imaginary features of the input system matrix. The feature processing unit is used to separate and enhance the high-frequency and low-frequency features of the real part features and the imaginary part features respectively, and generate real part enhanced features and imaginary part enhanced features; An image reconstruction unit is used to upsample and fuse the real part enhancement features and the imaginary part enhancement features to generate a high-resolution system matrix. Input the system matrix to be calibrated into the trained deep learning model, and output the calibrated high-resolution system matrix.

2. The fast system matrix calibration algorithm for magnetic particle imaging according to claim 1, characterized in that, The image encoding unit includes: The initial feature extraction module is used to perform preliminary feature extraction on the input system matrix; The feature enhancement module includes at least one dense block containing residual connections for refining and enhancing the initially extracted features; The first frequency separation and enhancement module is used to receive the refined and enhanced features and separate them into real features and imaginary features for output.

3. The fast system matrix calibration algorithm for magnetic particle imaging according to claim 1, characterized in that, The feature processing unit includes a real branch and an imaginary branch that operate in parallel. The real branch is used to process the real features, and the imaginary branch is used to process the imaginary features. Both the real and imaginary branches include a second frequency separation enhancement module and a feature addition layer; The second frequency separation and enhancement module is used to separate and enhance high-frequency and low-frequency features; The feature addition layer is used to add the enhanced high-frequency features and low-frequency features element by element to obtain real part enhancement features and imaginary part enhancement features, which are then input into the image reconstruction unit.

4. A fast system matrix calibration algorithm for magnetic particle imaging according to claim 2 or 3, characterized in that, Both the first frequency separation enhancement module and the second frequency separation enhancement module include: The grouping normalization layer is used to receive input features and perform channel grouping normalization processing; The frequency domain weighting layer is used to perform a Softmax operation on the grouped and normalized features to generate a frequency domain weighted representation. The weight separation layer is used to multiply the frequency domain weighted representation with the grouped and normalized features element by element, and then perform normalized weighted representation after multiplication. The threshold gating layer is used to generate high-frequency weights and low-frequency weights by setting a threshold gating to the normalized weighted representation; The feature mapping layer is used to map input features to high-frequency components based on high-frequency weights and to map input features to low-frequency components based on low-frequency weights. The high-frequency processing layer is used to simultaneously perform grouped convolution and pointwise convolution on the high-frequency components, and output the fused enhanced high-frequency features; The low-frequency processing layer is used to perform pointwise convolution on the low-frequency components and concatenate the output of the pointwise convolution with the low-frequency components to generate enhanced low-frequency features.

5. The fast system matrix calibration algorithm for magnetic particle imaging according to claim 1, characterized in that, The image reconstruction unit includes: The upsampling module is used to perform spatial dimension upsampling operations on the real part enhancement features and the imaginary part enhancement features, respectively; The feature fusion module is used to fuse the real and imaginary features after upsampling to generate the high-resolution system matrix.

6. The fast system matrix calibration algorithm for magnetic particle imaging according to claim 5, characterized in that, The upsampling module specifically includes: The feature extraction submodule is used to extract features from the upsampled input features. The spatial enhancement submodule contains cascaded transposed convolutional layers and standard convolutional layers, used to progressively multiply the spatial dimension of features.

7. The fast system matrix calibration algorithm for magnetic particle imaging according to claim 1, characterized in that, System matrices with a frequency signal-to-noise ratio below a set threshold are identified as invalid data and removed.

8. The fast system matrix calibration algorithm for magnetic particle imaging according to claim 1, characterized in that, The calibrated high-resolution system matrix is ​​used for magnetic particle imaging reconstruction to improve the resolution of the reconstructed image.

9. A fast system matrix calibration algorithm for magnetic particle imaging according to claim 4, characterized in that, The frequency domain weighting layer is a Softmax activation function layer.

10. A fast system matrix calibration algorithm for magnetic particle imaging according to claim 4, characterized in that, The normalized weighted representation is implemented through a Sigmoid activation function layer.

Images