Time-frequency domain signal filtering method for magnetic particle imaging based on multi-scale feature fusion

Abstract

The present invention belongs to the field of magnetic particle imaging, and specifically relates to a magnetic particle imaging time-frequency domain signal filtering method, system, and device based on multi-scale feature fusion, aiming to solve the problem that the existing technology cannot effectively eliminate noise in MPI, resulting in poor magnetic particle imaging quality. The method of the present invention comprises: collecting the original frequency domain signal in the magnetic particle imaging system as the input signal; filtering the input signal through a pre-constructed multi-scale feature filtering neural network model to obtain a filtered time domain signal; the multi-scale feature filtering neural network model includes four filtering submodules; the filtering submodules are constructed based on a sequentially connected convolution layer, a feature filtering structure, and two fully connected layers. The present invention can extract particle and noise-related features in the signal, filter out noise signals, improve the signal-to-noise ratio, and enhance the quality of magnetic particle imaging.

Description

Technical Field

[0001] The present invention belongs to the field of magnetic particle imaging, and in particular relates to a magnetic particle imaging time-frequency domain signal filtering method, system and device based on multi-scale feature fusion. Background Art

[0002] Magnetic particle imaging (MPI) is a non-invasive medical imaging technique that precisely locates targets through the nonlinear response of superparamagnetic nanoparticles in a magnetic field. It boasts high temporal and spatial resolution and high sensitivity. This makes MPI promising for applications in cell tracking, perfusion imaging, and tumor detection.

[0003] In practice, the thermal effects of the resistive elements in the MPI receiving coil alter the impedance of the electromagnetic transmitting coil, causing magnetic field drift. Furthermore, the nonlinear response of the coil's nonlinear electronic components to the input signal generates high-order harmonic interference, rendering simple filtering methods insufficient for achieving satisfactory results. Therefore, a method that effectively removes complex noise from MPI is crucial for improving MPI imaging performance. Based on this, the present invention proposes a time-frequency domain signal filtering method for magnetic particle imaging based on multi-scale feature fusion. Summary of the Invention

[0004] In order to solve the above-mentioned problems in the prior art, that is, to solve the problem that the prior art cannot effectively eliminate noise in MPI, resulting in poor quality of magnetic particle imaging, the present invention proposes a magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion, which is used to filter the collected MPI frequency domain signal to obtain a filtered time domain signal, and then perform magnetic particle imaging based on the filtered time domain signal. The method includes:

[0005] S100, collecting the original frequency domain signal in the magnetic particle imaging system as an input signal;

[0006] S200, filtering the input signal using a pre-built multi-scale feature filtering neural network model to obtain a filtered time domain signal;

[0007] The multi-scale feature filtering neural network model includes four filtering submodules; the four filtering submodules are connected in sequence, wherein the frequency domain data passes through the first two filtering submodules, is inversely Fourier transformed into a time domain signal, and then is processed by the last two filtering submodules;

[0008] The filtering submodule is constructed based on sequentially connected convolutional layers, feature filtering structures, and two fully connected layers; the feature filtering structure includes a multi-channel signal feature encoding module and an attention-based filtering processing module;

[0009] The multi-channel signal feature encoding module includes two multi-scale convolution units; the first multi-scale convolution unit is constructed based on four parallel convolution layers, and the features processed by the four parallel convolution layers are spliced ​​and input into the second multi-scale convolution unit; the second multi-scale convolution unit is constructed based on two parallel convolution layers;

[0010] The attention-based filtering processing module is constructed based on a global average pooling layer, two fully connected layers, and a Sigmoid activation function connected in sequence; the input of the attention-based filtering processing module is the concatenated features of the outputs of the two parallel convolutional layers in the second multi-scale convolutional unit;

[0011] The output of the filtering submodule is the product of the input of the attention-based filtering processing module and the features processed by the Sigmoid activation function.

[0012] In some preferred embodiments, the convolution kernel sizes of the four parallel convolution layers in the first multi-scale convolution unit are 1, 3, 5, and 7, respectively; the convolution kernel sizes of the two parallel convolution layers in the second multi-scale convolution unit are 1 and 3, respectively.

[0013] In some preferred embodiments, the training method of the neural network model of the multi-scale feature filtering is:

[0014] A100 generates a simulation signal of a magnetic particle imaging time domain signal and performs Fourier transform to obtain a frequency domain signal; the frequency domain signal is a one-dimensional array containing only the amplitude of the harmonic component and is a complex number;

[0015] A200 adds noise to the frequency domain signal in A100 to obtain a noisy frequency domain signal; a training dataset is constructed based on the noisy frequency domain signal and its corresponding true value label of the time domain signal;

[0016] A300, inputting the noisy frequency domain signal in the training data set into the neural network model of the multi-scale feature filtering to obtain a filtered time domain signal;

[0017] A400 calculates a loss value using a pre-built loss function based on the filtered time domain signal obtained by A300 and the true value label of the time domain signal, and updates network parameters of the neural network model of the multi-scale feature filtering;

[0018] A500, looping steps A300-A400 until a trained multi-scale feature filtering neural network model is obtained.

[0019] In some preferred embodiments, the noise added to the frequency domain signal includes Gaussian noise, harmonic interference, and mixed noise thereof.

[0020] In some preferred embodiments, the pre-constructed loss function is:

[0021] L=L MSE +α3L2

[0022] L MSE =α1L freq +α2L Time

[0023]

[0024]

[0025]

[0026] Among them, L is the total loss function, that is, the pre-built loss function, L freq is the frequency domain loss, L Time is the time domain loss, L MSE is the sum of error losses, L2 is the network regularization term, α1, α2, α3 are hyperparameters, is the frequency domain prediction result, that is, the feature after inverse Fourier transform, u(n) is the true value label of the time domain signal, is the time domain prediction result, N represents the length of the sampling signal, P represents the number of neural network parameters, θ i represents the i-th neural network parameter.

[0027] In a second aspect of the present invention, a magnetic particle imaging time-frequency domain signal filtering system based on multi-scale feature fusion is proposed, which is used to filter the collected MPI frequency domain signal to obtain a filtered time domain signal, and then perform magnetic particle imaging based on the filtered time domain signal. The system includes: a signal acquisition module and a filtering processing module;

[0028] The signal acquisition module is configured to acquire the original frequency domain signal in the magnetic particle imaging system as an input signal;

[0029] The filtering processing module is configured to filter the input signal through a pre-built multi-scale feature filtering neural network model to obtain a filtered time domain signal;

[0030] The multi-scale feature filtering neural network model includes four filtering submodules; the four filtering submodules are connected in sequence, wherein the frequency domain data passes through the first two filtering submodules, is inversely Fourier transformed into a time domain signal, and then is processed by the last two filtering submodules;

[0031] The filtering submodule is constructed based on sequentially connected convolutional layers, feature filtering structures, and two fully connected layers; the feature filtering structure includes a multi-channel signal feature encoding module and an attention-based filtering processing module;

[0032] The multi-channel signal feature encoding module includes two multi-scale convolution units; the first multi-scale convolution unit is constructed based on four parallel convolution layers, and the features processed by the four parallel convolution layers are spliced ​​and input into the second multi-scale convolution unit; the second multi-scale convolution unit is constructed based on two parallel convolution layers;

[0033] The attention-based filtering processing module is constructed based on a global average pooling layer, two fully connected layers, and a Sigmoid activation function connected in sequence; the input of the attention-based filtering processing module is the concatenated features of the outputs of the two parallel convolutional layers in the second multi-scale convolutional unit;

[0034] The output of the filtering submodule is the product of the input of the attention-based filtering processing module and the features processed by the Sigmoid activation function.

[0035] In a third aspect of the present invention, an electronic device is proposed, comprising: at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion.

[0036] In a fourth aspect of the present invention, a computer-readable storage medium is proposed, characterized in that the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by a computer to implement the above-mentioned magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion.

[0037] Beneficial effects of the present invention:

[0038] The present invention can extract the particle and noise related features in the signal, filter out the noise signal, improve the signal-to-noise ratio, and enhance the quality of magnetic particle imaging.

[0039] Based on the raw time-frequency domain signals acquired from a magnetic particle imaging system, this method uses a deep learning (DL)-based collaborative filtering method to construct a neural network that fuses multi-scale features. This filtering process ultimately yields high-resolution MPI signals. This method extracts particle- and noise-related features from the signal, filters out noise signals such as dynamic background noise, and improves the signal-to-noise ratio. It also demonstrates generalizability across MPI signals of varying scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Other features, objects and advantages of the present application will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings.

[0041] Figure 1 It is a flowchart of a method for filtering time-frequency domain signals in magnetic particle imaging based on multi-scale feature fusion according to an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the framework of a magnetic particle imaging time-frequency domain signal filtering system based on multi-scale feature fusion according to an embodiment of the present invention;

[0043] Figure 3 This is a simplified flowchart of a method for filtering time-frequency domain signals in magnetic particle imaging based on multi-scale feature fusion according to an embodiment of the present invention;

[0044] Figure 4 (a) is a schematic diagram of a time domain signal according to an embodiment of the present invention;

[0045] Figure 4 (b) is a schematic diagram of a frequency domain signal according to an embodiment of the present invention;

[0046] Figure 5 (a) is a schematic diagram of a noisy time domain signal according to an embodiment of the present invention;

[0047] Figure 5 (b) is a schematic diagram of a frequency domain signal with noise according to an embodiment of the present invention;

[0048] Figure 6 1 is a schematic structural diagram of a neural network model for multi-scale feature filtering according to an embodiment of the present invention;

[0049] Figure 7 (a) is a schematic diagram of a time domain signal of magnetic particle imaging after filtering according to an embodiment of the present invention;

[0050] Figure 7 (b) is a schematic diagram of a filtered magnetic particle imaging frequency domain signal according to an embodiment of the present invention;

[0051] Figure 8 It is a structural diagram of a computer system suitable for implementing an electronic device of an embodiment of the present application according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0053] The present application will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to explain the relevant invention and are not intended to limit the invention. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application may be combined with each other.

[0054] The first embodiment of the present invention is a magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion, which is used to filter the collected MPI frequency domain signal to obtain a filtered time domain signal, and then perform magnetic particle imaging based on the filtered time domain signal, such as Figure 1 As shown, the method includes:

[0055] S100, collecting the original frequency domain signal in the magnetic particle imaging system as an input signal;

[0056] S200, filtering the input signal using a pre-built multi-scale feature filtering neural network model to obtain a filtered time domain signal;

[0057] The multi-scale feature filtering neural network model includes four filtering submodules; the four filtering submodules are connected in sequence, wherein the frequency domain data passes through the first two filtering submodules, is inversely Fourier transformed into a time domain signal, and then is processed by the last two filtering submodules;

[0058] The filtering submodule is constructed based on sequentially connected convolutional layers, feature filtering structures, and two fully connected layers; the feature filtering structure includes a multi-channel signal feature encoding module and an attention-based filtering processing module;

[0059] The multi-channel signal feature encoding module includes two multi-scale convolution units; the first multi-scale convolution unit is constructed based on four parallel convolution layers, and the features processed by the four parallel convolution layers are spliced ​​and input into the second multi-scale convolution unit; the second multi-scale convolution unit is constructed based on two parallel convolution layers;

[0060] The attention-based filtering processing module is constructed based on a global average pooling layer, two fully connected layers, and a Sigmoid activation function connected in sequence; the input of the attention-based filtering processing module is the concatenated features of the outputs of the two parallel convolutional layers in the second multi-scale convolutional unit;

[0061] The output of the filtering submodule is the product of the input of the attention-based filtering processing module and the features processed by the Sigmoid activation function.

[0062] In order to more clearly illustrate the magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion of the present invention, each step in an embodiment of the method of the present invention is described in detail below with reference to the accompanying drawings.

[0063] In the following embodiments, the structure and training process of the neural network model of multi-scale feature filtering are first described in detail, and then the process of filtering the frequency domain signal by the magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion is described in detail.

[0064] 1. The structure and training process of the neural network model of multi-scale feature filtering, such as Figure 3 shown

[0065] A100 generates a simulation signal of a magnetic particle imaging time domain signal and performs Fourier transform to obtain a frequency domain signal; the frequency domain signal is a one-dimensional array containing only the amplitude of the harmonic component and is a complex number;

[0066] In this embodiment, firstly, the time domain signal of magnetic particle imaging in actual situation is simulated to generate a simulated time domain signal, that is, the simulated signal of the time domain signal ( Figure 3 The frequency domain signal is obtained by Fourier transforming the simulation signal. The frequency domain signal is a one-dimensional array containing only the amplitude of the harmonic component and is a complex number. The time domain signal without adding noise is as follows Figure 4 As shown in (a) in the figure, the frequency domain signal without adding noise is as follows Figure 4 As shown in (b) in .

[0067] A200 adds noise to the frequency domain signal in A100 to obtain a noisy frequency domain signal; a training dataset is constructed based on the noisy frequency domain signal and its corresponding true value label of the time domain signal;

[0068] In this embodiment, different levels of noise are added to the frequency domain signal to obtain 27,300 sets of noisy frequency domain signals and their corresponding clean time domain signals as the data set for neural network training, which are randomly divided into training set and validation set. Figure 5 As shown in (a) in the figure, the frequency domain signal with noise is as follows Figure 5 As shown in (b) in .

[0069] In the present invention, the noise added to the frequency domain signal preferably includes Gaussian noise, harmonic interference and mixed noise thereof. In other embodiments, the noise can be selected and added according to actual conditions.

[0070] A300, inputting the noisy frequency domain signal in the training data set into the neural network model of the multi-scale feature filtering to obtain a filtered time domain signal;

[0071] In this embodiment, a neural network model of multi-scale feature filtering is first built, such as Figure 6 As shown, the neural network model of the multi-scale feature filtering includes four filtering submodules; the four filtering submodules are connected in sequence, wherein the frequency domain data passes through the first two filtering submodules and is converted into a time domain signal by inverse Fourier transform and then processed by the last two filtering submodules, i.e. Figure 6 IDFT in (a);

[0072] The filtering submodule is constructed based on sequentially connected convolutional layers (kernel size is 1), a feature filtering structure, and two fully connected layers; the feature filtering structure includes a multi-channel signal feature encoding module and an attention-based filtering processing module; the feature filtering structure is used for filtering processing.

[0073] The multi-channel signal feature encoding module includes two multi-scale convolution units; the first multi-scale convolution unit is constructed based on four parallel convolution layers, which are divided into four parallel convolution layers (the convolution kernel sizes are 1, 3, 5, and 7, respectively). Figure 6 The processed features are spliced ​​and input into the second multi-scale convolution unit; the second multi-scale convolution unit is constructed based on two parallel convolution layers (the convolution kernel sizes are 1 and 3 respectively);

[0074] The attention-based filtering processing module is constructed based on a global average pooling layer, two fully connected layers, and a Sigmoid activation function connected in sequence; the input of the attention-based filtering processing module is the features of the spliced ​​outputs of the two parallel convolutional layers in the second multi-scale convolutional unit; that is, the features of the spliced ​​outputs of the two parallel convolutional layers in the second multi-scale convolutional unit are input into the global average pooling layer to obtain global features, and then the weight of each feature is obtained through two fully connected layers, and the Sigmoid activation function is used to map the weight to [-1,1] to filter out noise-related features.

[0075] The output of the filtering submodule is the product of the input of the attention-based filtering processing module and the features processed by the Sigmoid activation function (i.e., element-by-element multiplication), and the result is: Figure 6 The improved features shown in (b).

[0076] After the output of the filter submodule, the last two fully connected layers map the features to the signal space.

[0077] Based on the constructed multi-scale feature filtering neural network model, the noisy frequency domain signal is filtered to obtain the filtered time domain signal, that is, the time domain prediction result. The filtered time domain signal is as follows Figure 7 As shown in (a) in the figure, the time domain signal after filtering is as follows Figure 7 As shown in (b) in .

[0078] A400 calculates a loss value using a pre-built loss function based on the filtered time domain signal obtained by A300 and the true value label of the time domain signal, and updates network parameters of the neural network model of the multi-scale feature filtering;

[0079] In this example, a noisy frequency-domain signal is used as input, and a clean time-domain signal is used as the label (i.e., the ground-truth label of the time-domain signal). A neural network model is trained using mean absolute error as the loss function. The sum of the frequency-domain signal output loss and the time-domain signal output loss is calculated, and an L2 regularization term is used to reduce network overfitting.

[0080] Among them, the constructed loss function is:

[0081]

[0082]

[0083] L MSE =α1L freq +α2L Time (3)

[0084]

[0085] L=L MSE +α3L2 (5)

[0086] Among them, L is the total loss function, that is, the pre-built loss function, L freq is the frequency domain loss, L Time is the time domain loss, L MSE is the sum of error losses, L2 is the network regularization term, α1, α2, α3 are hyperparameters, is the frequency domain prediction result, that is, the feature after inverse Fourier transform, u(n) is the true value label of the time domain signal, is the time domain prediction result, N represents the length of the sampling signal, P represents the number of neural network parameters, θ i represents the i-th neural network parameter.

[0087] A500, looping steps A300-A400 until a trained multi-scale feature filtering neural network model is obtained.

[0088] In this embodiment, the multi-scale feature filtering neural network model is trained cyclically until a trained multi-scale feature filtering neural network model is obtained. In the present invention, the number of network iterative training times is preferably set to 1000 times.

[0089] 2. Magnetic Particle Imaging Time-Frequency Domain Signal Filtering Method Based on Multi-Scale Feature Fusion

[0090] S100, collecting the original frequency domain signal in the magnetic particle imaging system as an input signal;

[0091] In this embodiment, a real MPI frequency domain signal (or referred to as an original frequency domain signal) is randomly collected as input.

[0092] S200 , filtering the input signal using a pre-built multi-scale feature filtering neural network model to obtain a filtered time domain signal.

[0093] In this embodiment, the original frequency domain signal is input into the above-trained multi-scale feature filtering neural network model to obtain a filtered time domain signal.

[0094] The second embodiment of the present invention is a magnetic particle imaging time-frequency domain signal filtering system based on multi-scale feature fusion, which is used to filter the collected MPI frequency domain signal to obtain a filtered time domain signal, and then perform magnetic particle imaging based on the filtered time domain signal, such as Figure 2 As shown, the system includes: a signal acquisition module 100, a filtering processing module 200;

[0095] The signal acquisition module 100 is configured to acquire the original frequency domain signal in the magnetic particle imaging system as an input signal;

[0096] The filtering processing module 200 is configured to filter the input signal through a pre-built multi-scale feature filtering neural network model to obtain a filtered time domain signal;

[0097] The multi-scale feature filtering neural network model includes four filtering submodules; the four filtering submodules are connected in sequence, wherein the frequency domain data passes through the first two filtering submodules, is inversely Fourier transformed into a time domain signal, and then is processed by the last two filtering submodules;

[0098] The filtering submodule is constructed based on sequentially connected convolutional layers, feature filtering structures, and two fully connected layers; the feature filtering structure includes a multi-channel signal feature encoding module and an attention-based filtering processing module;

[0099] The multi-channel signal feature encoding module includes two multi-scale convolution units; the first multi-scale convolution unit is constructed based on four parallel convolution layers, and the features processed by the four parallel convolution layers are spliced ​​and input into the second multi-scale convolution unit; the second multi-scale convolution unit is constructed based on two parallel convolution layers;

[0100] The attention-based filtering processing module is constructed based on a global average pooling layer, two fully connected layers, and a Sigmoid activation function connected in sequence; the input of the attention-based filtering processing module is the concatenated features of the outputs of the two parallel convolutional layers in the second multi-scale convolutional unit;

[0101] The output of the filtering submodule is the product of the input of the attention-based filtering processing module and the features processed by the Sigmoid activation function.

[0102] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0103] It should be noted that the magnetic particle imaging time-frequency domain signal filtering system based on multi-scale feature fusion provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiment can be combined into a single module or further divided 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 invention are merely for the purpose of distinguishing the modules or steps and are not to be considered as improper limitations of the present invention.

[0104] A third embodiment of the present invention provides an electronic device comprising at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to implement the method for filtering time-frequency domain signals in magnetic particle imaging based on multi-scale feature fusion as claimed in the claim above.

[0105] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, which are used to be executed by a computer to implement the magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion as claimed in the claim.

[0106] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes and related instructions of the electronic device and computer-readable storage medium described above can refer to the corresponding processes in the aforementioned method examples and will not be repeated here.

[0107] Reference below Figure 8 , which shows a structural diagram of a computer system of a server suitable for implementing the system, method, and device embodiments of the present application. Figure 8 The server shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0108] like Figure 8 As shown, the computer system includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 802 or the program loaded from the storage part 808 into the random access memory (RAM) 803. Various programs and data required for system operation are also stored in the RAM 803. The CPU 801, ROM 802 and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0109] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, and the like; an output section 807 including devices such as a cathode ray tube, a liquid crystal display, and a speaker; a storage section 808 including devices such as a hard disk; and a communication section 809 including a network interface card such as a local area network card or a modem. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. Removable media 811, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 810 as needed, so that computer programs read from the removable media can be installed in the storage section 808 as needed.

[0110] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the CPU 801, the above-mentioned functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, RAM, ROM, erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including, but not limited to, wirelessly, by wire, by optical cable, or any suitable combination thereof.

[0111] The computer program code for performing the operations of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as C or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network or a wide area network, or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet).

[0112] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0113] The terms "first", "second", etc. are used to distinguish similar objects, rather than to describe or indicate a particular order or sequence.

[0114] The term "comprise" or any other similar term is intended to cover non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0115] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion, which is used to filter the collected MPI frequency domain signal to obtain a filtered time domain signal, and then perform magnetic particle imaging based on the filtered time domain signal, characterized in that: The method includes: S100, collecting the original frequency domain signal in the magnetic particle imaging system as an input signal; S200, filtering the input signal using a pre-built multi-scale feature filtering neural network model to obtain a filtered time domain signal; The multi-scale feature filtering neural network model includes four filtering submodules; the four filtering submodules are connected in sequence, wherein the frequency domain data passes through the first two filtering submodules, is inversely Fourier transformed into a time domain signal, and then is processed by the last two filtering submodules; The filtering submodule is constructed based on sequentially connected convolutional layers, feature filtering structures, and two fully connected layers; the feature filtering structure includes a multi-channel signal feature encoding module and an attention-based filtering processing module; The multi-channel signal feature encoding module includes two multi-scale convolution units; the first multi-scale convolution unit is constructed based on four parallel convolution layers, and the features processed by the four parallel convolution layers are spliced ​​and input into the second multi-scale convolution unit; the second multi-scale convolution unit is constructed based on two parallel convolution layers; The attention-based filtering processing module is constructed based on a global average pooling layer, two fully connected layers, and a Sigmoid activation function connected in sequence; the input of the attention-based filtering processing module is the concatenated features of the outputs of the two parallel convolutional layers in the second multi-scale convolutional unit; The output of the filtering submodule is the product of the input of the attention-based filtering processing module and the features processed by the Sigmoid activation function.

2. The magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion according to claim 1 is characterized in that: The convolution kernel sizes of the four parallel convolution layers in the first multi-scale convolution unit are 1, 3, 5, and 7 respectively; the convolution kernel sizes of the two parallel convolution layers in the second multi-scale convolution unit are 1 and 3 respectively.

3. The magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion according to claim 1 is characterized in that: The training method of the neural network model of the multi-scale feature filtering is: A100 generates a simulation signal of a magnetic particle imaging time domain signal and performs Fourier transform to obtain a frequency domain signal; the frequency domain signal is a one-dimensional array containing only the amplitude of the harmonic component and is a complex number; A200 adds noise to the frequency domain signal in A100 to obtain a noisy frequency domain signal; Construct a training dataset based on the noisy frequency domain signal and its corresponding time domain signal true value label; A300, inputting the noisy frequency domain signal in the training data set into the neural network model of the multi-scale feature filtering to obtain a filtered time domain signal; A400 calculates a loss value using a pre-built loss function based on the filtered time domain signal obtained by A300 and the true value label of the time domain signal, and updates network parameters of the neural network model of the multi-scale feature filtering; A500, looping steps A300-A400 until a trained multi-scale feature filtering neural network model is obtained.

4. The magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion according to claim 3 is characterized in that: The noise added to the frequency domain signal includes Gaussian noise, harmonic interference and their mixed noise.

5. The magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion according to claim 3 is characterized in that: The pre-built loss function is: L=L MSE +α3l2 L MSE =α1L freq +α2L Time Among them, L is the total loss function, that is, the pre-built loss function, L freq is the frequency domain loss, L Time is the time domain loss, L MSE is the sum of error losses, L2 is the network regularization term, α1, α2, α3 are hyperparameters, is the frequency domain prediction result, that is, the feature after inverse Fourier transform, u(n) is the true value label of the time domain signal, is the time domain prediction result, N represents the length of the sampling signal, P represents the number of neural network parameters, θ i represents the i-th neural network parameter.

6. A magnetic particle imaging time-frequency domain signal filtering system based on multi-scale feature fusion, which is used to filter the collected MPI frequency domain signal to obtain a filtered time domain signal, and then perform magnetic particle imaging based on the filtered time domain signal, characterized in that: The system includes: a signal acquisition module and a filtering processing module; The signal acquisition module is configured to acquire the original frequency domain signal in the magnetic particle imaging system as an input signal; The filtering processing module is configured to filter the input signal through a pre-built multi-scale feature filtering neural network model to obtain a filtered time domain signal; The multi-scale feature filtering neural network model includes four filtering submodules; the four filtering submodules are connected in sequence, wherein the frequency domain data passes through the first two filtering submodules, is inversely Fourier transformed into a time domain signal, and then is processed by the last two filtering submodules; The filtering submodule is constructed based on sequentially connected convolutional layers, feature filtering structures, and two fully connected layers; the feature filtering structure includes a multi-channel signal feature encoding module and an attention-based filtering processing module; The multi-channel signal feature encoding module includes two multi-scale convolution units; the first multi-scale convolution unit is constructed based on four parallel convolution layers, and the features processed by the four parallel convolution layers are spliced ​​and input into the second multi-scale convolution unit; the second multi-scale convolution unit is constructed based on two parallel convolution layers; The attention-based filtering processing module is constructed based on a global average pooling layer, two fully connected layers, and a Sigmoid activation function connected in sequence; the input of the attention-based filtering processing module is the concatenated features of the outputs of the two parallel convolutional layers in the second multi-scale convolutional unit; The output of the filtering submodule is the product of the input of the attention-based filtering processing module and the features processed by the Sigmoid activation function.

7. An electronic device, characterized in that: include: at least one processor; and a memory communicatively coupled to at least one of said processors; Wherein, the memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by a computer to implement the magnetic particle imaging time-frequency domain signal filtering method based on multi-scale feature fusion according to any one of claims 1 to 5.

Images