Magnetic particle imaging equipment system matrix calibration method, device and equipment

Abstract

The invention provides a magnetic particle imaging equipment system matrix calibration method, device and equipment. The method comprises the steps of performing sampling processing on a preset two-dimensional slice layer based on a cross down-sampling method to obtain original LR SM data; the preset two-dimensional slice layer is each two-dimensional slice layer in the three-dimensional FOV of the MPI equipment; performing frequency screening on the original LR SM data to obtain high signal-to-noise ratio LR SM data; converting each row of the high-signal-to-noise-ratio LR SM data into a three-dimensional structure to obtain a plurality of pieces of three-dimensional LR SM data; taking the L layer and the L + 1 layer of each three-dimensional LR SM data as first input data, and inputting the first input data into a pre-training system matrix calibration network to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data to obtain calibrated HR SM data; the adjacent layers of the preset two-dimensional slice layer are subjected to dislocation sampling, so that more prior information of each slice layer can be obtained, and the system matrix calibration precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical imaging, and in particular to a method, device and equipment for calibrating a matrix of a magnetic particle imaging device system. Background Art

[0002] As a new type of molecular imaging technology, MPI (Magnetic Particle Imaging) has shown broad application prospects in the biomedical field with its unique advantages of high sensitivity, high temporal resolution and no spatial depth restriction. MPI uses the nonlinear magnetization response of magnetic particles. Under the joint action of uniform driving field and static gradient field, it can realize the generation of magnetic particle signals in the encoding space and spatial position encoding. By reconstructing the received magnetic particle signals in the entire imaging field of view (Field-of-View, FOV), the visualized concentration distribution of magnetic particles in the field of view can be obtained. The MPI reconstruction method based on the system matrix (System Matrix, SM) can achieve higher reconstructed image quality. In MPI, the system matrix based on measurement is obtained by traversing all grid positions of the sample in the entire FOV through a mechanical translation stage and recording the magnetic particle response signals at the corresponding positions. This is very time-consuming for the more advanced 3D imaging equipment suitable for human size. Therefore, how to speed up the system matrix calibration process of MPI equipment has become a research hotspot in the MPI field.

[0003] The system matrix calibration method in the prior art mainly includes: using a compressed sensing-based method to recover the fully sampled system matrix from the under-sampled system matrix. However, the quality of the fully sampled system matrix recovered by this method will decrease as the measurement position decreases, and when the measurement position increases, the time cost of calibration will be greatly increased; some studies have also applied the super-resolution method based on deep learning to the MPI system matrix calibration, that is, using equally spaced sampling to obtain LR SM (Low Resolution System Matrix) as the prior information of HR SM (High Resolution System Matrix), and then sending the obtained pairs of LR SM and HR SM to the CNN-based super-resolution network for training, but this method has less prior information available, resulting in low recovery accuracy.

[0004] Therefore, the existing system matrix calibration method of magnetic particle imaging equipment has the problems of long calibration time and low calibration accuracy. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a method, device and equipment for calibrating a matrix of a magnetic particle imaging device system.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a method for calibrating a system matrix of a magnetic particle imaging device, comprising:

[0008] Sampling is performed based on a cross-downsampling method on a preset two-dimensional slice layer to obtain original LR SM data; the preset two-dimensional slice layer is each two-dimensional slice layer within the three-dimensional FOV of the MPI device;

[0009] Perform frequency screening on the original LR SM data to obtain high signal-to-noise ratio LR SM data;

[0010] Convert each row of the high signal-to-noise ratio LR SM data into a three-dimensional structure to obtain a plurality of three-dimensional LR SM data;

[0011] The L layer and L+1 layer of each three-dimensional LR SM data are taken as the first input data and input into the pre-trained system matrix calibration network to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data to obtain calibrated HR SM data; wherein, L∈{1,2,...,Z1}, Z1 represents the total number of preset two-dimensional slice layers; the cross downsampling method is to perform staggered sampling on adjacent layers of the preset two-dimensional slice layer; the pre-trained system matrix calibration network is trained jointly using HR SM sample data and three-dimensional LR SM training data; the three-dimensional LR SM training data is sampled from the HR SM sample data based on the cross downsampling method.

[0012] Optionally, sampling processing is performed on a preset two-dimensional slice layer based on a cross downsampling method to obtain original LR SM data, including:

[0013] Based on the cross-down sampling method, corresponding sampling points are obtained in each preset two-dimensional slice layer;

[0014] The preset two-dimensional slice layer is sampled according to the sampling points to obtain sub-LR SM data; the sub-LR SM data is the LR SM sampling data of each two-dimensional slice layer;

[0015] The sub-LR SM data are processed by matrix combination to obtain the original LR SM data.

[0016] Alternatively, the cross-downsampling method is expressed as:

[0017] When the number of preset 2D slice layers L is an odd number:

[0018]

[0019] When the number of preset 2D slice layers L is an even number:

[0020]

[0021] Among them, p L (x, y) represents a sampling position point, x represents a horizontal coordinate position index of a preset two-dimensional slice layer, and y represents a vertical coordinate position index of a preset two-dimensional slice layer.

[0022] Optionally, frequency screening is performed on the original LR SM data to obtain high signal-to-noise ratio LR SM data, including:

[0023] Perform Fourier transform on the original LR SM data to obtain frequency domain LR SM data;

[0024] The data in the frequency domain LR SM data that is greater than a preset frequency threshold constitutes high signal-to-noise ratio LR SM data.

[0025] Optionally, the pre-trained system matrix calibration network includes: a data fusion module, a feature encoding module and an upsampling reconstruction module; the upsampling reconstruction module includes: a first upsampling reconstruction module and a second upsampling reconstruction module;

[0026] The data fusion module, the feature encoding module and the second upsampling reconstruction module are connected in series in sequence to form a first structure;

[0027] The first upsampling reconstruction module is connected in parallel with the first structure.

[0028] Optionally, the training process of the pre-trained system matrix calibration network includes:

[0029] Obtain HR SM sample data;

[0030] Perform frequency screening on HR SM sample data to obtain HR SM sample data with high signal-to-noise ratio;

[0031] Perform three-dimensional conversion processing on each row of high signal-to-noise ratio HR SM sample data to obtain multiple three-dimensional HR SM initial data;

[0032] Based on the cross-down sampling method, multiple three-dimensional HR SM initial data are sampled and processed to obtain multiple three-dimensional LRSM training data;

[0033] The L layer of each three-dimensional LRSM training data, the L+1 layer of the three-dimensional LRSM training data, and the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of the three-dimensional LR SM training data are input into the initial system matrix calibration network, and the initial system matrix calibration network is trained to obtain the pre-trained system matrix calibration network; wherein, L∈{1,2,...,Z2}, Z2 represents the total number of layers of the three-dimensional LRSM training data; wherein, the L layer of the three-dimensional LRSM sample data and the L+1 layer of the three-dimensional LRSM sample data are the second input data, and the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of the three-dimensional LR SM training data are the label data; the initial system matrix calibration network has the same structure as the pre-trained system matrix calibration network.

[0034] Optionally, the L layer of each three-dimensional LRSM training data, the L+1 layer of the three-dimensional LRSM training data, and the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of the three-dimensional LRSM training data are input into the initial system matrix calibration network, and the initial system matrix calibration network is trained to obtain a pre-trained system matrix calibration network, including:

[0035] Input the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of each three-dimensional LRSM training data, the L+1th layer of the three-dimensional LRSM training data, and the Lth layer of the three-dimensional LR SM training data into the initial system matrix calibration network;

[0036] The initial system matrix calibration network is continuously trained in the direction of decreasing loss function;

[0037] When the loss function meets the preset loss threshold or the number of iterations meets the preset iteration threshold, the corresponding initial system matrix calibration network is used as the pre-trained system matrix calibration network.

[0038] Optionally, the feature encoding module includes: a convolution module and a Transformer module connected in sequence;

[0039] The Transformer module includes: N Transformer encoders connected in series.

[0040] In a second aspect, the present invention provides a magnetic particle imaging device system matrix calibration device, the magnetic particle imaging device system matrix calibration device comprising: a sampling unit, a screening unit, a conversion unit and an input-output unit;

[0041] The sampling unit is used to: perform sampling processing based on a cross-down sampling method on a preset two-dimensional slice layer to obtain original LR SM data; the preset two-dimensional slice layer is each two-dimensional slice layer within the three-dimensional FOV of the MPI device;

[0042] The screening unit is used to: perform frequency screening on the original LR SM data to obtain LR SM data with a high signal-to-noise ratio;

[0043] The conversion unit is used to: convert each row of the high signal-to-noise ratio LR SM data into a three-dimensional structure to obtain a plurality of three-dimensional LRSM data;

[0044] The input-output unit is used to: take the L layer and L+1 layer of each three-dimensional LR SM data as the first input data, input them into the pre-trained system matrix calibration network, so as to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data, and obtain calibrated HR SM data; wherein, L∈{1,2,...,Z1}, Z1 represents the total number of preset two-dimensional slice layers; the cross-downsampling method is to perform staggered sampling on adjacent layers of the preset two-dimensional slice layer; the pre-trained system matrix calibration network adopts HR SM sample data and three-dimensional LR SM training data for joint training; the three-dimensional LR SM training data is sampled from the HR SM sample data based on the cross-downsampling method.

[0045] In a third aspect, the present invention provides a magnetic particle imaging device system matrix calibration device, comprising: a processor, a storage medium and a bus, the storage medium storing machine-readable instructions executable by the processor, and when the magnetic particle imaging device system matrix calibration is ready to run, the processor and the storage medium communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the magnetic particle imaging device system matrix calibration method as described in the first aspect above.

[0046] The invention provides a method, device and equipment for calibrating a system matrix of a magnetic particle imaging device. Among them, the system matrix calibration method of the magnetic particle imaging device includes: performing sampling processing based on the cross-downsampling method on the preset two-dimensional slice layer to obtain original LR SM data; the preset two-dimensional slice layer is each two-dimensional slice layer within the three-dimensional FOV of the MPI device; performing frequency screening on the original LRSM data to obtain high signal-to-noise ratio LR SM data; converting each row of the high signal-to-noise ratio LR SM data into a three-dimensional structure to obtain multiple three-dimensional LR SM data; using the L layer and L+1 layer of each three-dimensional LR SM data as the first input data, inputting the pre-trained system matrix calibration network to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data to obtain calibrated HR SM data; wherein L∈{1,2,...,Z1}, Z1 represents the total number of preset two-dimensional slice layers; the cross-downsampling method is to perform staggered sampling on adjacent layers of the preset two-dimensional slice layer; the pre-trained system matrix calibration network adopts HR SM sample data and three-dimensional LR SM training data for joint training; the three-dimensional LR SM training data is sampled from the HR SM sample data based on the cross-downsampling method. In the present invention, adjacent layers of a preset two-dimensional slice layer are staggered for sampling, and system matrix calibration processing is performed on the original LR SM data obtained by the staggered sampling based on a pre-trained system matrix calibration network, so as to finally obtain calibrated HR SM data. Since this method does not need to measure the magnetic particle response signal at all measurement positions, the system matrix calibration time of the MPI device is improved. In addition, since the present invention adopts staggered sampling of adjacent layers in the preset two-dimensional slice layer, the problem of fixed data information and little prior information caused by the existing equally spaced interval sampling is avoided, more prior information can be obtained at the same sampling rate, and the system matrix calibration accuracy is improved.

[0047] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of a flow chart of a method for calibrating a matrix of a magnetic particle imaging device system provided by an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of 2x cross-down sampling provided in an embodiment of the present invention;

[0050] Figure 3 A schematic diagram of the calibration process of a pre-trained system matrix calibration network provided in an embodiment of the present invention;

[0051] Figure 4 A schematic diagram of a calibrated system matrix image generated based on the method of the present invention provided in an embodiment of the present invention;

[0052] Figure 5 A schematic diagram of the structure of a matrix calibration device for a magnetic particle imaging device system provided by an embodiment of the present invention;

[0053] Figure 6 A schematic structural diagram of a matrix calibration device for a magnetic particle imaging device system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0055] In order to improve the system matrix calibration time of an MPI device and improve the system matrix calibration accuracy, an embodiment of the present invention provides a system matrix calibration method for a magnetic particle imaging device. Figure 1 The present invention provides a flow chart of a method for calibrating a matrix of a magnetic particle imaging device system. Figure 1 As shown, including:

[0056] S101. Perform sampling processing on a preset two-dimensional slice layer based on a cross downsampling method to obtain original LR SM data, where the preset two-dimensional slice layer is each two-dimensional slice layer within a three-dimensional FOV of an MPI device.

[0057] Optionally, sampling processing is performed on a preset two-dimensional slice layer based on a cross downsampling method to obtain original LR SM data, including:

[0058] Based on the cross-down sampling method, corresponding sampling points are obtained in each preset two-dimensional slice layer;

[0059] The preset two-dimensional slice layer is sampled according to the sampling points to obtain sub-LR SM data; the sub-LR SM data is the LR SM sampling data of each two-dimensional slice layer;

[0060] The sub-LR SM data are processed by matrix combination to obtain the original LR SM data.

[0061] Alternatively, the cross-downsampling method is expressed as:

[0062] When the number of preset 2D slice layers L is an odd number:

[0063]

[0064] When the number of preset 2D slice layers L is an even number:

[0065]

[0066] Among them, p L(x, y) represents a sampling position point, x represents a horizontal coordinate position index of a preset two-dimensional slice layer, and y represents a vertical coordinate position index of a preset two-dimensional slice layer.

[0067] In addition, in the embodiment of the present invention, it can also be set that when p L When (x,y) takes the value 1, the position is selected. L When the value of (x,y) is 0, the position is not selected.

[0068] S102, performing frequency screening on the original LR SM data to obtain LR SM data with a high signal-to-noise ratio.

[0069] Optionally, S102 may specifically include:

[0070] Perform Fourier transform on the original LR SM data to obtain frequency domain LR SM data;

[0071] The data in the frequency domain LR SM data that is greater than a preset frequency threshold constitutes high signal-to-noise ratio LR SM data.

[0072] S103, converting each row of the high signal-to-noise ratio LR SM data into a three-dimensional structure to obtain a plurality of three-dimensional LR SM data.

[0073] S104. Use the L layer and L+1 layer of each three-dimensional LR SM data as the first input data and input them into the pre-trained system matrix calibration network to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data to obtain calibrated HR SM data.

[0074] Among them, L∈{1,2,...,Z1}, Z1 represents the total number of preset two-dimensional slicing layers; the cross downsampling method is to staggered sampling of adjacent layers of the preset two-dimensional slicing layer; the pre-trained system matrix calibration network is trained jointly using HR SM sample data and three-dimensional LR SM training data; the three-dimensional LR SM training data is sampled from the HR SM sample data based on the cross downsampling method.

[0075] It should be noted that when restoring to the Z1 layer, since the Z1+1 layer does not exist, the Z1 layer and the Z1-1 layer can be used as the first input data.

[0076] An embodiment of the present invention provides a method for calibrating a system matrix of a magnetic particle imaging device, comprising: performing staggered sampling on adjacent layers of a preset two-dimensional slice layer, and performing system matrix calibration processing on original LR SM data obtained by the staggered sampling based on a pre-trained system matrix calibration network, and finally obtaining calibrated HR SM data. Since the method does not require traversal measurement at multiple measurement positions, the system matrix calibration time of the MPI device is improved. In addition, since the present invention adopts staggered sampling of adjacent layers in the preset two-dimensional slice layer, the problem of fixed data information and little prior information caused by existing equally spaced interval sampling is avoided, more prior information can be obtained at the same sampling rate, thereby improving the accuracy of system matrix calibration.

[0077] Optionally, the pre-trained system matrix calibration network includes: a data fusion module, a feature encoding module and an upsampling reconstruction module; the upsampling reconstruction module includes: a first upsampling reconstruction module and a second upsampling reconstruction module;

[0078] The data fusion module, the feature encoding module and the second upsampling reconstruction module are connected in series in sequence to form a first structure;

[0079] The first upsampling reconstruction module is connected in parallel with the first structure.

[0080] It should be noted that, in the embodiment of the present invention, the data processing process of the feature encoding module is specifically as follows:

[0081] (1) Use a convolution layer to perform convolution operations on the base layer image (the L layer of the 3D LR SM data) and the reference layer image (the L+1 layer of the 3D LR SM data) to obtain the base layer image feature F1 and the reference layer image feature F2;

[0082] (2) Perform dot product operations on the baseline image feature F1 and the reference image feature F2 with the baseline image feature F1 to obtain affinity mappings A between F1, F2 and F1 respectively. 1,1 and A 1,2 , the calculation formula is:

[0083] A 11 =F1·F1;

[0084] A 12 =F1·F2;

[0085] (3) Calculate the self-affinity feature M1 of the reference layer image:

[0086]

[0087] (4) Calculate the unique feature M2 of the reference layer image independent of the base layer image:

[0088]

[0089] in, Represents the multiplication of corresponding elements.

[0090] (5) Concatenate M1 and M2 to obtain the complementary fusion feature F.

[0091] Corresponding to the above process, schematically, Figure 3 A schematic diagram of the calibration process of the pre-trained system matrix calibration network provided in an embodiment of the present invention. Figure 3 The reference layer system matrix image in corresponds to the L layer of the three-dimensional LR SM data, and the adjacent layer system matrix image corresponds to the L+1 layer of the three-dimensional LR SM data.

[0092] In the embodiment of the present invention, the upsampling reconstruction module is composed of two parts: a pixelshuffle upsampling module (a second upsampling reconstruction module) and an interpolation upsampling module (a first upsampling reconstruction module); wherein the pixelshuffle upsampling module is composed of a pixelshuffle layer and two 3x3 convolution blocks, and the interpolation upsampling module is composed of a linear interpolation layer and two 3x3 convolution blocks. The calculation process of the upsampling reconstruction module is as follows:

[0093] (1) The global feature representation output by the feature encoding module is sent to the pixel shuffle upsampling module for operation to obtain a high-resolution feature map I1;

[0094] (2) Send the input base layer image to the interpolation upsampling module to obtain the high-resolution feature map I2;

[0095] (3) Add the high-resolution feature map I1 and the high-resolution feature map I2 to output the calibrated HR SM data.

[0096] Optionally, the training process of the pre-trained system matrix calibration network includes:

[0097] Obtain HR SM sample data;

[0098] Perform frequency screening on HR SM sample data to obtain HR SM sample data with high signal-to-noise ratio;

[0099] Perform three-dimensional conversion processing on each row of high signal-to-noise ratio HR SM sample data to obtain multiple three-dimensional HR SM initial data;

[0100] Based on the cross-down sampling method, multiple three-dimensional HR SM initial data are sampled and processed to obtain multiple three-dimensional LRSM training data;

[0101] The L layer of each three-dimensional LRSM training data, the L+1 layer of the three-dimensional LRSM training data, and the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of the three-dimensional LR SM training data are input into the initial system matrix calibration network, and the initial system matrix calibration network is trained to obtain the pre-trained system matrix calibration network; wherein, L∈{1,2,...,Z2}, Z2 represents the total number of layers of the three-dimensional LRSM training data; wherein, the L layer of the three-dimensional LRSM sample data and the L+1 layer of the three-dimensional LRSM sample data are the second input data, and the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of the three-dimensional LR SM training data are the label data; the initial system matrix calibration network has the same structure as the pre-trained system matrix calibration network.

[0102] In the embodiment of the present invention, the HR SM sample data obtained by full sampling measurement is firstly subjected to frequency screening to obtain high signal-to-noise ratio HR SM sample data; each row of the screened high signal-to-noise ratio HR SM sample data is converted into a three-dimensional structure, and sliced ​​along the z-axis direction, and the cross-down sampling method (see Figure 2 ) calculates the sampling position points of each slice layer, extracts the data of the corresponding points, and obtains the three-dimensional LR SM training data. Specifically, Figure 2 A schematic diagram of 2-fold cross-down sampling provided in an embodiment of the present invention is shown in FIG. Figure 2 As shown, a black grid indicates that the position point is selected, and a blank grid indicates that the position point is not selected. In addition, it should be noted that the 2-fold cross-down sampling method provided in the embodiment of the present invention is only exemplary, and can be adjusted according to device performance and actual needs. For example, it can also be set to 2 T times, where T is a positive integer.

[0103] Optionally, the L layer of each three-dimensional LRSM training data, the L+1 layer of the three-dimensional LRSM training data, and the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of the three-dimensional LRSM training data are input into the initial system matrix calibration network, and the initial system matrix calibration network is trained to obtain a pre-trained system matrix calibration network, including:

[0104] Input the high signal-to-noise ratio HR SM sample data corresponding to the Lth layer of each three-dimensional LRSM training data, the L+1th layer of the three-dimensional LRSM training data, and the Lth layer of the three-dimensional LR SM training data into the initial system matrix calibration network;

[0105] The initial system matrix calibration network is continuously trained in the direction of decreasing loss function;

[0106] When the loss function meets the preset loss threshold or the number of iterations meets the preset iteration threshold, the corresponding initial system matrix calibration network is used as the pre-trained system matrix calibration network.

[0107] Optionally, the feature encoding module includes: a convolution module and a Transformer module connected in sequence;

[0108] The Transformer module includes: N Transformer encoders connected in series.

[0109] It should be noted that, in an embodiment of the present invention, the Transformer module is used to extract the global feature representation of the input data. The Transformer module can be exemplarily composed of 6 Transformer encoders in series, each of which is composed of a multi-head self-attention (MHSA) module and a feedforward network (FFN) module; the MHSA module uses multiple self-attention layers to calculate the relevant information between each two embedded block sequences; after each MHSA module, an FFN module is used to map the data from low dimension to high dimension and then to low dimension improved feature representation, and the FFN module is composed of two fully connected layers and a GeLU activation function.

[0110] In order to verify the effectiveness of the matrix calibration method for a magnetic particle imaging device system provided by the embodiment of the present invention, the embodiment of the present invention also carried out simulation verification. Figure 4 The schematic diagram of the calibrated system matrix image generated by the method of the present invention is provided in an embodiment of the present invention. Figure 4 As shown, the first row of data is a low-resolution system matrix image obtained by sampling based on the cross-downsampling method, the second row of data is the corresponding original system matrix image (for comparison with the generated calibrated system matrix image schematic), and the third row of data is the calibrated system matrix image. Figure 4 It can be seen that the calibrated system matrix image obtained based on the method of the present invention is consistent with the original system matrix image in appearance, which proves the effectiveness of the method of the embodiment of the present invention.

[0111] In summary, the embodiment of the present invention provides a method for calibrating the system matrix of a magnetic particle imaging device, cross-sampling adjacent layers in a 3DFOV to obtain information of two adjacent layers as prior information, thereby solving the problem of low quality of system matrix recovery caused by insufficient input information. In addition, a pre-trained system matrix calibration network is provided, which includes: an information fusion module, a feature encoding module and an upsampling reconstruction module connected in series by CNN-Transformer; the information fusion module is used to effectively fuse the complementary features of multiple LR SMs; the feature encoding module connected in series by CNN-Transformer is used to extract local and global feature representations of the fused features; the upsampling reconstruction module is used to reconstruct the acquired feature representation into an HR SM, and the combination of the above modules is conducive to fusing the complementary features of multiple LR SMs and improving the recovery quality of the system matrix.

[0112] The method provided in the embodiment of the present invention can be applied to an electronic device. Specifically, the electronic device can be: a desktop computer, a portable computer, an intelligent mobile terminal, a server, etc., which is not limited in the embodiment of the present invention.

[0113] Based on the same inventive concept, an embodiment of the present invention further provides a matrix calibration device for a magnetic particle imaging device system. Figure 5 The present invention provides a schematic diagram of a magnetic particle imaging device system matrix calibration device. Figure 5 As shown, it includes: a sampling unit 501, a screening unit 502, a conversion unit 503 and an input-output unit 504;

[0114] The sampling unit 501 is used to: perform sampling processing based on a cross downsampling method on a preset two-dimensional slice layer to obtain original LR SM data; the preset two-dimensional slice layer is each two-dimensional slice layer within the three-dimensional FOV of the MPI device;

[0115] The screening unit 502 is used to: perform frequency screening on the original LR SM data to obtain high signal-to-noise ratio LR SM data;

[0116] The conversion unit 503 is used to: convert each row of the high signal-to-noise ratio LR SM data into a three-dimensional structure to obtain a plurality of three-dimensional LR SM data;

[0117] The input-output unit 504 is used to: take the L layer and L+1 layer of each three-dimensional LR SM data as the first input data, input them into the pre-trained system matrix calibration network, so as to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data, and obtain calibrated HR SM data; wherein, L∈{1,2,...,Z1}, Z1 represents the total number of preset two-dimensional slice layers; and the cross downsampling method is to perform staggered sampling on adjacent layers of the preset two-dimensional slice layer.

[0118] Figure 6 A schematic diagram of the structure of a magnetic particle imaging device system matrix calibration device provided by an embodiment of the present invention includes: a processor 710, a storage medium 720 and a bus 730, wherein the storage medium 720 stores machine-readable instructions executable by the processor 710, and when the magnetic particle imaging device system matrix calibration device is running, the processor 710 communicates with the storage medium 720 via the bus 730, and the processor 710 executes the machine-readable instructions to execute the steps of the above method embodiment. The specific implementation method and technical effect are similar and will not be repeated here.

[0119] The storage medium may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk storage. Optionally, the storage medium may also be at least one storage device located away from the aforementioned processor.

[0120] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0121] It should be noted that the terms "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention.

[0122] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0123] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other changes of the above disclosed embodiments by viewing the drawings and the disclosed content. In the description of the present invention, the term "comprising" does not exclude other components or steps, "one" or "an" does not exclude multiple situations, and the meaning of "multiple" is two or more, unless otherwise clearly and specifically limited. In addition, certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0124] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.

Claims

1. A method for calibrating a magnetic particle imaging device system matrix. It is characterized in that include: The sampling process is performed based on the cross-down sampling method at the preset two-dimensional slice layer to obtain the original LR SM data; The preset two-dimensional slice layer is each two-dimensional slice layer within the three-dimensional FOV of the MPI device; Performing frequency screening on the original LR SM data to obtain high signal-to-noise ratio LR SM data; Convert each row of the high signal-to-noise ratio LR SM data into a three-dimensional structure to obtain a plurality of three-dimensional LR SM data; The L layer and L+1 layer of each of the three-dimensional LR SM data are used as the first input data and input into the pre-trained system matrix calibration network to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data to obtain calibrated HR SM data; wherein L∈{1,2,...,Z1}, Z1 represents the total number of preset two-dimensional slice layers; the cross downsampling method is to perform staggered sampling on adjacent layers of the preset two-dimensional slice layer; The pre-trained system matrix calibration network is trained jointly using HR SM sample data and three-dimensional LR SM training data; the three-dimensional LR SM training data is sampled from the HR SM sample data based on the cross-downsampling method.

2. The method for calibrating a magnetic particle imaging device system matrix according to claim 1, It is characterized in that The sampling process is performed based on a cross down sampling method on a preset two-dimensional slice layer to obtain original LR SM data, including: Based on the cross-down sampling method, obtain corresponding sampling points in each of the preset two-dimensional slice layers; Sampling the preset two-dimensional slice layer according to the sampling point to obtain sub-LR SM data; the sub-LR SM data is the LR SM sampling data of each of the two-dimensional slice layers; The sub-LR SM data are subjected to matrix combination processing to obtain the original LR SM data.

3. The method for calibrating a magnetic particle imaging device system matrix according to claim 1 or 2, It is characterized in that The cross-down sampling method is expressed as: When the number of preset 2D slice layers L is an odd number: When the number of preset 2D slice layers L is an even number: Among them, p L (x, y) represents a sampling position point, x represents a horizontal coordinate position index of a preset two-dimensional slice layer, and y represents a vertical coordinate position index of a preset two-dimensional slice layer.

4. The method for calibrating a magnetic particle imaging device system matrix according to claim 1, It is characterized in that The frequency screening of the original LR SM data to obtain high signal-to-noise ratio LR SM data includes: Performing Fourier transform on the original LR SM data to obtain frequency domain LR SM data; The data in the frequency domain LR SM data that is greater than a preset frequency threshold constitutes the high signal-to-noise ratio LR SM data.

5. The method for calibrating a magnetic particle imaging device system matrix according to claim 1, It is characterized in that The pre-trained system matrix calibration network includes: a data fusion module, a feature encoding module and an upsampling reconstruction module; the upsampling reconstruction module includes: a first upsampling reconstruction module and a second upsampling reconstruction module; The data fusion module, the feature encoding module and the second upsampling reconstruction module are connected in series in sequence to form a first structure; The first upsampling reconstruction module is connected in parallel with the first structure.

6. The method for calibrating a magnetic particle imaging device system matrix according to claim 1, It is characterized in that The training process of the pre-trained system matrix calibration network includes: Obtain HR SM sample data; Performing frequency screening on the HR SM sample data to obtain HR SM sample data with a high signal-to-noise ratio; Performing three-dimensional conversion processing on each row of the high signal-to-noise ratio HR SM sample data to obtain a plurality of three-dimensional HR SM initial data; Sampling the plurality of three-dimensional HR SM initial data based on a cross downsampling method to obtain a plurality of three-dimensional LRSM training data; The L layers of each of the three-dimensional LRSM training data, the L+1 layers of the three-dimensional LRSM training data, and the high signal-to-noise ratio HRSM sample data corresponding to the Lth layer of the three-dimensional LRSM training data are input into the initial system matrix calibration network, and the initial system matrix calibration network is trained to obtain the pre-trained system matrix calibration network; wherein, L∈{1,2,...,Z2}, Z2 represents the total number of layers of three-dimensional LRSM training data; wherein, the L layers of the three-dimensional LRSM sample data and the L+1 layers of the three-dimensional LRSM sample data are the second input data, and the high signal-to-noise ratio HRSM sample data corresponding to the Lth layer of the three-dimensional LRSM training data are label data; the initial system matrix calibration network has the same structure as the pre-trained system matrix calibration network.

7. The method for calibrating a magnetic particle imaging device system matrix according to claim 1, It is characterized in that The method comprises inputting the high signal-to-noise ratio HR SM sample data corresponding to the L layer of each three-dimensional LRSM training data, the L+1 layer of the three-dimensional LRSM training data, and the Lth layer of the three-dimensional LR SM training data into the initial system matrix calibration network, and training the initial system matrix calibration network to obtain the pre-trained system matrix calibration network, including: Inputting the high signal-to-noise ratio HR SM sample data corresponding to each L layer of the three-dimensional LRSM training data, the L+1 layer of the three-dimensional LRSM training data, and the Lth layer of the three-dimensional LRSM training data into the initial system matrix calibration network; Continuously training the initial system matrix calibration network in a direction in which the loss function decreases; When the loss function meets a preset loss threshold or the number of iterations meets a preset iteration threshold, the corresponding initial system matrix calibration network is used as the pre-trained system matrix calibration network.

8. The method for calibrating a magnetic particle imaging device system matrix according to claim 5, It is characterized in that The feature encoding module includes: a convolution module and a Transformer module connected in sequence; The Transformer module includes: N Transformer encoders connected in series.

9. A magnetic particle imaging equipment system matrix calibration device, It is characterized in that The magnetic particle imaging equipment system matrix calibration device comprises: a sampling unit, a screening unit, a conversion unit and an input and output unit; The sampling unit is used to: perform sampling processing based on a cross-down sampling method on a preset two-dimensional slice layer to obtain original LR SM data; the preset two-dimensional slice layer is each two-dimensional slice layer within the three-dimensional FOV of the MPI device; The screening unit is used to: perform frequency screening on the original LR SM data to obtain high signal-to-noise ratio LR SM data; The conversion unit is used to: convert each row of the high signal-to-noise ratio LR SM data into a three-dimensional structure to obtain a plurality of three-dimensional LR SM data; The input-output unit is used to: take the L layer and L+1 layer of each three-dimensional LR SM data as the first input data, input them into the pre-trained system matrix calibration network, so as to perform system matrix calibration processing on the L layer of the three-dimensional LR SM data to obtain calibrated HR SM data; wherein, L∈{1,2,...,Z1}, Z1 represents the total number of preset two-dimensional slice layers; the cross-downsampling method is to perform staggered sampling on adjacent layers of the preset two-dimensional slice layers; the pre-trained system matrix calibration network adopts HR SM sample data and three-dimensional LR SM training data for joint training; the three-dimensional LR SM training data is sampled from the HR SM sample data based on the cross-downsampling method.

10. A magnetic particle imaging device system matrix calibration device, It is characterized in that include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the magnetic particle imaging device system matrix calibration is ready to run, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the magnetic particle imaging device system matrix calibration method as described in any one of claims 1-8.

Images