Method for improving MPI image resolution based on physical model

Abstract

The invention belongs to the technical field of signal imaging. The invention provides a high-resolution MPI image generation method based on a physical model. According to the embodiment of the invention, the feed-through interference signal is counteracted based on the adaptive compensation method, and the fundamental frequency of the particle signal is reserved, so that a complete magnetic particle nonlinear response signal can be obtained without filtering and fundamental frequency recovery, and the quality and resolution of an X-space reconstructed image are improved. An X-space imaging physical model is established on the basis of a recovered X-space reconstruction image, magnetic particle concentration distribution is solved by adopting a space iteration deconvolution algorithm on the basis of the model, noise is constrained by adopting TV regularization, an iteration error of the model is reduced by adopting l1 sparse prior, and the generalization ability of the model is improved. Based on the method, the MPI reconstruction resolution can be improved by more than multiple times, and the spatial resolution superior to 1mm can also be realized in a low signal-to-noise ratio environment.

Description

Technical Field

[0001] The disclosed embodiments relate to the field of signal imaging technology, and in particular to a method for improving the resolution of MPI images based on a physical model. Background Art

[0002] Magnetic particle imaging (MPI) is an emerging medical imaging technique that can image the distribution of superparamagnetic iron oxide nanoparticles (SPIO). Since its introduction in 2005, MPI has been widely used in biological imaging. MPI images have positive contrast and good depth penetration, without background tissue signals. In addition, MPI does not involve ionizing radiation, and SPIO tracers are biocompatible and renal safe, making MPI very suitable for clinical molecular imaging applications.

[0003] MPI uses field-free points (FFP) or field-free lines to scan the entire field of view (FOV) along a specific trajectory, and uses receiving coils to collect nonlinear response signals of MNPS during the scanning process. The voltage signal is then mapped to the MNP concentration distribution using a reconstruction algorithm. At present, the commonly used MPI reconstruction methods are system matrix reconstruction and X-space reconstruction. First, image reconstruction based on the system matrix has higher image resolution than the X-space reconstruction method, but the SM calibration measurement is very time-consuming. In addition, if the characteristics of the tracer or the magnetic field environment change, the SM needs to be remeasured and calibrated, further increasing the complexity of the method. In addition, the system matrix measurement time will be immeasurable in large-field MPI devices, and there are also challenges in the storage of the system matrix. Although there are many sparse methods that can reduce the number of system matrix measurements and reduce storage pressure, they will also result in a loss of reconstruction quality.

[0004] The X-space reconstruction method is a physical model-based method. The X-space reconstruction algorithm maps the time domain signal of the nonlinear response of magnetic particles to the spatial domain according to the FFP position for imaging. It has the characteristics of real-time image reconstruction and omits the complex system function calibration measurement process. The X-reconstruction diagram is shown in Figure 1 (h) and Figure 1 (b) However, there are serious feedthrough interference signals in the MPI system, which will reduce the signal-to-noise ratio of the system and cause the X-space reconstruction resolution to decrease, as shown in Figure 1 (f) and Figure 1(c) As shown. At present, in order to remove the feedthrough interference signal, a high-pass filter is used, which will cause the fundamental frequency of the particle signal to be lost, thereby destroying the shift invariance of the system and causing X-space reconstruction errors, such as Figure 1 (g) and Figure 1 As shown in (d). In order to solve the problem of fundamental frequency loss, KuanLu estimated the lost fundamental frequency signal by maximizing the continuity between the pFOV scan and its previous scan in the overlapping area. SemihKurt reduced the reconstruction artifacts caused by fundamental frequency loss by adopting pFOV center imaging technology, and improved the reconstruction quality of X-space based on the deconvolution method. Although the methods based on fundamental frequency recovery are relatively mature at present, filtering and recovery will always lead to the loss of fundamental frequency signal, resulting in the loss of X-space reconstruction quality and resolution. More importantly, the physical model of X-space reconstruction can be expressed as the point spread function of the magnetic particle concentration distribution convolved with the MPI system. Because of the blurring effect of the point spread function, the resolution of the X-space reconstructed image is reduced, as shown in Figure 1 As shown in (b), the X-space reconstructed image cannot distinguish Figure 1 Line pair phantom in (a).

[0005] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0006] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention

[0007] The purpose of the embodiments of the present disclosure is to provide a method for improving the resolution of MPI images based on a physical model, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.

[0008] According to an embodiment of the present disclosure, a method for improving the resolution of an MPI image based on a physical model is provided, the method comprising: Acquire a feedthrough interference signal, and use an adaptive compensation method to offset the feedthrough interference signal to obtain a complete magnetic particle nonlinear response signal; Processing the nonlinear response signal of the magnetic particles by using the MPI system to obtain an X-space reconstructed image; An X-space imaging physical model is constructed based on a TV regularized dual Richardson-Lucy deconvolution algorithm and a spatial iterative deconvolution algorithm, and the X-space reconstructed image is processed using the X-space imaging physical model to obtain an MPI image.

[0009] Furthermore, the step of obtaining a feedthrough interference signal and using an adaptive compensation method to offset the feedthrough interference signal to obtain a complete magnetic particle nonlinear response signal includes: Collect the feedthrough interference signal in the empty sampling state Amplitude and Phase ; According to the amplitude of the feedthrough interference signal and Phase Get compensation signal ; The compensation signal Interference signal with feedthrough Subtracting in LNA and compensating the phase of the compensation signal according to an adaptive compensation method to obtain an optimal low-noise signal; Based on the optimal low-noise signal, a magnetic particle response signal is collected to obtain the magnetic particle nonlinear response signal.

[0010] Further, the step of compensating the phase of the compensation signal according to the adaptive compensation method to obtain an optimal low-noise signal includes: The compensation signal is compensated, and the step length of the phase compensation each time is ; Determine the compensation signal after compensation Interference signal with feedthrough Whether the low noise signal subtracted in the LNA is stable; If it is stable, the phase change of the compensated signal after adaptive compensation is output and the corresponding low noise signal; where, , n represents the number of adaptive compensation; If it is unstable, continue to compensate the compensation signal.

[0011] Furthermore, the feedthrough interference signal The expression is:

[0012] in, is the magnetic field frequency, For low noise of MPI system, For time; The expression of the low noise signal is:

[0013]

[0014]

[0015] in, ; The expression of the nonlinear response signal of the magnetic particles is:

[0016] in, The magnetic particle response signal.

[0017] Furthermore, the expression of the X-space reconstructed image is:

[0018] in, represents the X-space reconstructed image, represents the spatial concentration distribution of magnetic particles, represents the spatial convolution kernel of the MPI system, Represents the spatial distribution function of the noise.

[0019] Furthermore, the step of constructing an X-space imaging physical model based on a TV regularized dual Richardson-Lucy deconvolution algorithm and a spatial iterative deconvolution algorithm, and processing the X-space reconstructed image using the X-space imaging physical model to obtain an MPI image includes: The X-space imaging physical model is constructed based on the double Richardson-Lucy deconvolution algorithm and the spatial iterative deconvolution algorithm based on TV regularization; The MPI system spatial convolution kernel is estimated using the TV regularized dual Richardson-Lucy deconvolution algorithm; The spatial concentration distribution of magnetic particles is solved by using a spatial iterative deconvolution algorithm to obtain the MPI image.

[0020] Furthermore, the expression of the TV regularized dual Richardson-Lucy deconvolution algorithm is:

[0021]

[0022] in, For the The spatial concentration distribution of magnetic particles in the iteration, For the The spatial concentration distribution of magnetic particles in the iteration, for, The first regularization parameter, is the second regularization parameter, is the convolution kernel estimated in one iteration, is the input for one iteration, () represents the divergence calculation function.

[0023] Furthermore, the spatial iterative deconvolution algorithm includes: According to the sparsity and smoothness characteristics of the MPI image, the minimized cost function is established as:

[0024] in, is the weight factor of fidelity; , represents the spatial distribution of magnetic particle concentration; , represents the spatial convolution kernel of the MPI system; , represents the X-space reconstructed image; is TV regularization, is the TV regularization parameter; is the l1 regularization term, is the regularization parameter of the sparse prior; beg The derivative of gets:

[0025] in, .

[0026] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiment of the present disclosure, the above-mentioned method for improving the resolution of MPI images based on the physical model, on the one hand, based on the adaptive compensation method, the feedthrough interference signal is offset, and the fundamental frequency of the particle signal is retained, so that the complete magnetic particle nonlinear response signal can be obtained without filtering and fundamental frequency recovery, thereby improving the quality and resolution of the X-space reconstructed image. And based on the restored X-space reconstructed image, an X-space imaging physical model is established, based on the model, a spatial iterative deconvolution algorithm is used to solve the magnetic particle concentration distribution, TV regularization is used to constrain the noise, and l1 sparse priors are used to reduce the iterative error of the model and increase its generalization ability. On the other hand, based on this method, the MPI reconstruction resolution can be improved by more than several times, and in a low signal-to-noise ratio environment, a spatial resolution better than 1mm can also be achieved. At the same time, this method does not require complex system function acquisition and calibration processes to achieve high-resolution magnetic particle concentration distribution reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0028] Figure 1 A schematic diagram showing a 1-dimensional MPI signal link in the prior art, using a double-point phantom for simulation; Figure 2 A diagram showing the steps of a method for improving the resolution of an MPI image based on a physical model in an exemplary embodiment of the present disclosure; Figure 3 A flowchart showing signal processing by an adaptive compensation method in an exemplary embodiment of the present disclosure is shown; Figure 4 A schematic diagram showing a single receiving channel MPI device in an exemplary embodiment of the present disclosure; Figure 5 A graph showing the effect of using different half-width full-width at half-maximum point diffusions to perform MPI image deconvolution in a spatial iterative deconvolution algorithm in an exemplary embodiment of the present disclosure; Figure 6 A comparison diagram showing the results of two-dimensional image reconstruction of line pair phantoms with spacings of 0.5 mm, 1 mm, 1.5 mm and 2 mm in the X direction using the RL method and the spatial iterative deconvolution method in an exemplary embodiment of the present disclosure; Figure 7 It is shown that the RL method and the spatial iterative deconvolution method are used in an exemplary embodiment of the present disclosure to reconstruct a two-dimensional image of a line pair phantom with spacings of 0.5 mm, 1 mm, 1.5 mm and 2 mm in the Y direction. DETAILED DESCRIPTION

[0029] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0030] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0031] The X-space reconstruction method is a method based on a physical model. The X-space reconstruction algorithm maps the time domain signal of the nonlinear response of magnetic particles to the spatial domain according to the FFP position for imaging. It has the characteristics of real-time image reconstruction and omits the complex system function calibration measurement process. The X-reconstruction diagram is shown in Figure 1 (h) and Figure 1 (b) However, there are serious feedthrough interference signals in the MPI system, which will reduce the signal-to-noise ratio of the system and cause the X-space reconstruction resolution to decrease, as shown in Figure 1 (f) and Figure 1 (c) As shown. At present, in order to remove the feedthrough interference signal, a high-pass filter is used, which will cause the fundamental frequency of the particle signal to be lost, thereby destroying the shift invariance of the system and causing X-space reconstruction errors, such as Figure 1 (g) and Figure 1 As shown in (d). In order to solve the problem of fundamental frequency loss, Kuan Lu estimated the lost fundamental frequency signal by maximizing the continuity between the pFOV scan and its previous scan in the overlapping area. Semih Kurt reduced the reconstruction artifacts caused by fundamental frequency loss by adopting pFOV center imaging technology, and improved the reconstruction quality of X-space based on the deconvolution method. Although the methods based on fundamental frequency recovery are relatively mature at present, filtering and recovery will always lead to the loss of fundamental frequency signal, resulting in the loss of X-space reconstruction quality and resolution. More importantly, the physical model of X-space reconstruction can be expressed as the point spread function of the magnetic particle concentration distribution convolved with the MPI system. Because of the blurring effect of the point spread function, the resolution of the X-space reconstructed image is reduced, as shown in Figure 1 As shown in (b), the X-space reconstructed image cannot distinguish Figure 1 Line pair phantom in (a).

[0032] Figure 1 middle, Figure 1 (a) represents the excitation magnetic field. Figure 1 (b) shows the 1D MPI receiving signal , including feedthrough interference signals and magnetic particle nonlinear response signal . Figure 1 Path 1, Path 2, and Path 3 on the right represent the received signal Three different processing procedures: Path 1 directly uses the received signal Perform X-space reconstruction, and the reconstructed image is shown in Figure (e). Through physical model analysis, IMG1 cannot be deconvoluted. Path 2 indicates that a high-pass filter is used to filter out feedthrough interference. The fundamental frequency component of the magnetic particle signal is shown in Figure (c). The X-space reconstruction result of the signal in Figure (c) is as follows: Figure 1 As shown in (f), IMG1 cannot be deconvoluted by physical model analysis. Path 3 shows the use of feedthrough interference suppression method to obtain the nonlinear response signal of magnetic particles. ,like Figure 1 As shown in (d), using Figure 1 (d) The signal X-space reconstruction result is as follows Figure 1 (g) As shown, Figure 1 (h) indicates according to Figure 1 (g) The IMG3 convolution physical model is used to solve the magnetic particle concentration distribution after deconvolution. .

[0033] This example embodiment provides a method for improving the resolution of MPI images based on a physical model. Figure 2 As shown in , the method for improving the resolution of MPI images based on the physical model may include: steps S101 to S103.

[0034] Step S101: obtaining a feedthrough interference signal, and using an adaptive compensation method to offset the feedthrough interference signal to obtain a complete magnetic particle nonlinear response signal; Step S102: using the MPI system to process the nonlinear response signal of the magnetic particles to obtain an X-space reconstructed image; Step S103: constructing an X-space imaging physical model based on the TV regularized dual Richardson-Lucy deconvolution algorithm and the spatial iterative deconvolution algorithm, and processing the X-space reconstructed image using the X-space imaging physical model to obtain an MPI image.

[0035] Through the above-mentioned method of improving the resolution of MPI images based on the physical model, on the one hand, the feedthrough interference signal is offset based on the adaptive compensation method, and the fundamental frequency of the particle signal is retained. Therefore, the complete magnetic particle nonlinear response signal can be obtained without filtering and fundamental frequency recovery, thereby improving the quality and resolution of the X-space reconstructed image. And based on the restored X-space reconstructed image, an X-space imaging physical model is established. Based on this model, a spatial iterative deconvolution algorithm is used to solve the magnetic particle concentration distribution, TV regularization is used to constrain the noise, and l1 sparse priors are used to reduce the iterative error of the model and increase its generalization ability. On the other hand, based on this method, the MPI reconstruction resolution can be improved by more than several times, and in a low signal-to-noise ratio environment, a spatial resolution better than 1mm can also be achieved. At the same time, this method does not require complex system function acquisition and calibration processes to achieve high-resolution reconstruction of magnetic particle concentration distribution.

[0036] Next, we will refer to Figures 2 to 7 Each step of the method for improving the resolution of an MPI image based on a physical model in this example implementation is described in more detail.

[0037] In step S101, in this application, an FPGA-based adaptive compensation system is used to cancel the feedthrough interference signal and obtain the complete particle signal. From formula (1), it can be seen that the signal collected by the receiving coil consists of the particle nonlinear response signal and the feedthrough interference signal (1)

[0038] in: is the sensitivity of the receiving coil, is the vacuum permeability, represents the applied magnetic field strength, Represents the total magnetization intensity of all magnetic particles in the field of view, represents the nonlinear response signal of magnetic particles, represents a direct feedthrough signal, represents the system white noise.

[0039] Feedthrough interference signal As an interference signal, its amplitude is the magnetic particle response signal several orders of magnitude, assuming that the magnetic field lines inside the excitation coil are parallel and uniformly distributed and the sensitivity distribution of the receiving coil is uniform ,but:

[0040] (2)

[0041] in, represents the magnetic field strength through the coil plane , represents the magnitude of the magnetic field, represents the frequency of the magnetic field, represents the phase of the magnetic field, S represents the area vector of the coil, Indicates the amplitude of the direct feedthrough signal.

[0042] The electromagnetic coupling between the MPI excitation coil and the receiving coil causes the base frequency of the received signal to be contaminated. At present, there are three methods for suppressing feedthrough interference: filtering method, passive compensation, and active compensation. The filtering method uses a filter to filter out the base frequency of the magnetic particle response signal and the feedthrough interference signal, and then uses the base frequency recovery method to restore the particle signal base frequency. The passive compensation method is based on the electromagnetic mutual inductance theory, and usually uses a two-segment or three-segment gradient detection coil to suppress the feedthrough interference signal. However, in MPI, because there is a phase difference between the coupling signals of the receiving and compensation coils and the excitation field, the passive compensation effect is limited (Supplementary Note 2). The active compensation method outputs the compensation signal to offset the feedthrough interference signal through a subtractor, thereby retaining the complete particle nonlinear response signal. In this application, an FPGA-based adaptive compensation system is proposed to offset the feedthrough interference signal and obtain the complete particle signal. It can be seen from formula (1) that the signal collected by the receiving coil consists of the particle nonlinear response signal and the feedthrough interference signal. From formula (2), it can be seen that the feedthrough interference is a sinusoidal signal of a single frequency, so the compensation signal can be generated by obtaining the amplitude and phase information of the feedthrough interference signal. The adaptive compensation method of the present application is as follows: Figure 3 The method proposed in this application can cancel the feedthrough interference signal in real time and ensure real-time X-space reconstruction.

[0043] The FPGA-based adaptive compensation feedthrough interference suppression process consists of three processes, such as Figure 3 The figure shows the process flow chart of adaptive compensation signal processing. It is divided into three parts: 1) controlling the output of excitation signal and collecting feedthrough interference signal; 2) adaptive compensation; 3) collecting magnetic particle signal.

[0044] 1) Collecting feedthrough interference signals under background measurement status Amplitude and Phase .

[0045] (7) 2) Adaptive compensation. According to the amplitude obtained in 1) and Phase Output compensation signal = ,in express n The phase change of the compensation signal after the second adaptive compensation, , Indicates the phase compensation step size. Compensation signal Interference signal with feedthrough Subtract in LNA. As shown in equation (8), by adjusting make Less than the custom threshold voltage , lower noise than MPI system .

[0046]

[0047]

[0048] (8) in .

[0049] 3) Collect the magnetic particle response signal. Use adaptive compensation feedthrough interference suppression. When the magnetic particle sample is put in, the magnetic particle response signal is collected.

[0050] (9) Adaptive compensation method can suppress feedthrough interference signal, obtain complete nonlinear response signal of magnetic particles, retain LSI characteristics of MPI scanning, and improve the quality and resolution of X-space reconstructed image. The adaptive compensation algorithm is shown in Algorithm 1.

[0051]

[0052] In a specific embodiment, in the present application, in order to implement an adaptive compensation method. An adaptive compensation system suitable for the MPI system is developed based on FPGA. The Artix@7 XC7A35T FPGA chip is used to implement functions such as 4-channel synchronous acquisition, 4-channel asynchronous signal output, adaptive compensation and system control. The maximum sampling rate of the system ADC is 4MSPS (Sample / second), and the MAX5719 can achieve a single-channel 10MSPS sampling rate at 20 bits, with a maximum output frequency of 500kHz. Based on this hardware foundation, we obtain the phase and amplitude of the feedthrough interference signal through FFT in the FPGA, and the MAX5719 can stably output a 25kHz active compensation signal with the same frequency as the feedthrough interference, and offset the output compensation signal and the feedthrough interference in the LNA SR560. Finally, the amplitude of the feedthrough interference signal after subtraction is used as the judgment condition, and the adaptive compensation is completed when the amplitude of the feedthrough interference signal after subtraction is lower than the set threshold voltage. MPI integrated system such as Figure 4 As shown, the adaptive compensation system can realize device integration without increasing system complexity, thereby improving the robustness of the device.

[0053] like Figure 4 The figure shows a self-developed single-receive channel MPI device. Figure 4(a) represents the MPI adaptive compensation system based on FPGA. The AD7380 is a 4-channel 18-bit synchronous ADC chip that collects XY drive current signals, excitation current signals, and single-channel receiving coil signals. The MAX5719 is a 20-bit single-channel DAC chip. The three MAX5719 DACs output the XY drive signals and excitation signals respectively. The AD3541R is a 16-bit single-channel DAC chip that outputs compensation signals to offset feedthrough interference signals. Figure 4 (b) Represents the MPI system structure, including MPI power supply module, scanning equipment, and acquisition control system.

[0054] In step S102 and step S103, according to the magnetic particle response signal , and the magnetic particles and the magnetic particle concentration distribution are obtained The relationship between the response signal and the

[0055] (3) in, The nonlinear response process of magnetic particles can be expressed in convolution form: (4) in represents the derivative of the excitation field, and in equation (4) it can represent the speed at which the FFP point moves. represents the magnetic particle concentration distribution. Represents the point spread function (convolution kernel) of the MPI system. It represents the position of the FFP point at time t. After the FFP moving speed is normalized and the particle nonlinear response signal is mapped to the image domain according to the FFP position, the X-space spatial reconstructed image can be obtained.

[0056] (5) From equation (5), we can see that x-space reconstruction can be expressed as the convolution of the magnetic particle concentration distribution and the point spread function of the MPI system. Therefore, FWHM can be used to represent the imaging resolution of X-space. The FWHM of the point spread function of the MPI imaging system can be expressed as: (6) in, Related to the properties of magnetic particles, Represents the selection field gradient.

[0057] MPI imaging is the imaging of the concentration distribution of magnetic particles, but the X-space has a reduced imaging resolution due to the presence of the convolution kernel (blur kernel). In order to improve the MPI imaging resolution, a deconvolution operation can be used to improve the imaging resolution.

[0058] According to formula (9), the signal after feedthrough interference suppression is composed of the nonlinear response signal of magnetic particles and noise. Formula (5) represents the one-dimensional X-space imaging convolution model.

[0059] According to formula (5), X-space reconstruction can be expressed as the convolution of the magnetic particle concentration distribution and the point spread function of the MPI system, and the resolution of the X-space reconstructed image is limited by the point spread function, as shown in formula (6). In the MPI system, in order to more clearly express the X-space imaging properties, we use a more vivid mathematical analytical expression. The X-space imaging process can be modeled as the convolution of the blur function and the magnetic particle concentration distribution, and then the noise term is added, as shown in formula (10).

[0060] (10) in, represents the X-space reconstructed image, represents the spatial concentration distribution of magnetic particles, which needs to be solved. represents the spatial convolution kernel of the MPI system, an unknown quantity, represents the spatial distribution function of the noise. In equation (10), to solve , a deconvolution method is needed.

[0061] (1) Estimating the spatial convolution kernel of the MPI system When solving the spatial concentration distribution of magnetic particles, it is necessary to know and , It can be solved by x-space reconstruction, MPI system point spread function Unknown. At present, the MPI system uses a point phantom to measure the spatial convolution kernel, but the measurement result is the convolution result of the real point spread function of the MPI system and the point phantom, which cannot be used as the system spatial convolution kernel to solve the spatial concentration distribution of magnetic particles. Therefore, this application uses a blind deconvolution method (TV regularized double Richardson-Lucy deconvolution) to estimate the MPI system spatial convolution kernel.

[0062] (11) (12) The spatial convolution kernel estimation of the MPI system is shown in Algorithm 2. This method requires the X-space image measured by the point phantom as the initialization of the algorithm, thereby accelerating the convergence of the algorithm.

[0063]

[0064] (2) Estimation of spatial concentration distribution of magnetic particles (spatial iterative deconvolution) Based on the estimation of the spatial convolution kernel, the spatial concentration distribution of magnetic particles can be estimated. In this application, the deconvolution model of the spatial concentration distribution of magnetic particles can be solved by an optimization problem, and according to the sparsity and smoothness characteristics of the MPI image, the minimization cost function is established as: (13) The first term is the fidelity term, which represents the estimated The convolution result and input image The distance is the weight factor of fidelity, which is used to balance the fidelity of the image during the iteration process. The second term is TV regularization. The introduction of this regularization term can suppress more noise and retain better edge information. is the l1 regularization term, providing sparse prior information of the MPI image. The regularization parameter of the sparse prior is used to balance the sparse features of the image during the iteration process.

[0065] beg The derivative of is shown in formula (14): (14) in , the first term in equation (14) is consistent with the derivative of the likelihood estimate under the standard Richardson-Lucy deconvolution Gaussian distribution. This application uses AdamW to solve this convolution model. AdamW is an adaptive solution method with a smaller generalization error than Adam. The spatial iterative deconvolution method proposed in this application is shown in Algorithm 3.

[0066]

[0067] In this application, based on the adaptively compensated X-space reconstructed image, the TV regularized double Richardson-Lucy deconvolution method is used to estimate the point spread function of the MPI system. Finally, the spatial deconvolution model proposed in this application is used to solve the spatial distribution of magnetic particles. .

[0068] In a specific embodiment, the deconvolution method is used to solve the magnetic particle concentration distribution based on the X-space reconstructed image, which requires accurate estimation of the point spread function of the MPI system. A 1mm point phantom is used for testing. First, the X-space reconstruction method is used to obtain a 1mm phantom reconstructed image. Then, the double Richardson-Lucy deconvolution method is used to predict the point spread of the system. Based on different prediction results of the point spread function, the deconvolution method of this application (γ=0.0011, α=0.0022, β=0.002) is used to test the effect of the point spread function on the deconvolution. The results are as follows: Figure 5 shown.

[0069] In order to further illustrate the effectiveness of the spatial iterative deconvolution method in improving resolution, experiments were conducted using line pair phantoms. Four line pair distances were used for the experiments, and the line pair distances were 0.5 mm, 1 mm, 1.5 mm, and 2 mm, respectively. Figure 5 The PSF-2 estimated in the Y direction is used to improve the resolution. Figure 5 The estimated PSF-3 in the X-direction is used to improve the resolution. Experiments show that this method has a significant effect on improving the resolution compared to the RL (Richardson-Lucy) deconvolution method, and the spatial resolution of the X-space reconstructed image is improved by 4 times.

[0070] like Figure 5 As shown, Figure 5 MPI image deconvolution was performed using point spreads with different full width at half maximum. Figure 5 (a) Point spread function estimated by the dual RL+TV iterative method based on the measured point spread (1 mm dot phantom, SNR = 3 dB) (the number of iterations are (5, 10, 15, 20) respectively). Figure 5 (b) The figure shows the 1 mm point pair phantom in the X and Y directions. Figure 5 (c) Figure 2. Deconvolution based on the spatial iterative deconvolution method using different point spread functions.

[0071] like Figure 6 As shown, Figure 6 The RL and spatial iterative deconvolution methods were used to reconstruct the two-dimensional images of the line pair phantoms with spacing of 0.5 mm, 1 mm, 1.5 mm, and 2 mm in the X direction. The signal-to-noise ratio was set to 3 dB, and the gradient value in the X direction was set to 1.25 mT / mm. Figure 6 As shown, the spatial iterative deconvolution method is easier to distinguish line pair phantoms than RL.

[0072] like Figure 7 As shown, Figure 7The RL and spatial iterative deconvolution methods were used to reconstruct the two-dimensional images of the line pair phantoms with spacing of 0.5 mm, 1 mm, 1.5 mm, and 2 mm in the Y direction. The signal-to-noise ratio was set to 3 dB, and the gradient value in the Y direction was set to 2.5 mT / mm. Figure 7 As shown, the spatial iterative deconvolution method is easier to distinguish line pair phantoms than RL.

[0073] Through the above-mentioned method of improving the resolution of MPI images based on the physical model, on the one hand, the feedthrough interference signal is offset based on the adaptive compensation method, and the fundamental frequency of the particle signal is retained. Therefore, the complete magnetic particle nonlinear response signal can be obtained without filtering and fundamental frequency recovery, thereby improving the quality and resolution of the X-space reconstructed image. And based on the restored X-space reconstructed image, an X-space imaging physical model is established. Based on this model, a spatial iterative deconvolution algorithm is used to solve the magnetic particle concentration distribution, TV regularization is used to constrain the noise, and l1 sparse priors are used to reduce the iterative error of the model and increase its generalization ability. On the other hand, based on this method, the MPI reconstruction resolution can be improved by more than several times, and in a low signal-to-noise ratio environment, a spatial resolution better than 1mm can also be achieved. At the same time, this method does not require complex system function acquisition and calibration processes to achieve high-resolution reconstruction of magnetic particle concentration distribution.

[0074] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0075] 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, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials 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.

[0076] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A method for generating a high-resolution MPI image based on a physical model, characterized in that: The method includes: Acquire a feedthrough interference signal, and use an adaptive compensation method to offset the feedthrough interference signal to obtain a complete magnetic particle nonlinear response signal; Processing the nonlinear response signal of the magnetic particles by using the MPI system to obtain an X-space reconstructed image; An X-space imaging physical model is constructed based on a TV regularized dual Richardson-Lucy deconvolution algorithm and a spatial iterative deconvolution algorithm, and the X-space reconstructed image is processed using the X-space imaging physical model to obtain an MPI image.

2. The method for improving MPI image resolution based on a physical model according to claim 1, characterized in that: The step of obtaining a feedthrough interference signal and using an adaptive compensation method to offset the feedthrough interference signal to obtain a complete magnetic particle nonlinear response signal includes: Collect the feedthrough interference signal in the empty sampling state Amplitude and Phase ; According to the amplitude of the feedthrough interference signal and Phase Get compensation signal ; The compensation signal Interference signal with feedthrough Subtracting in LNA and compensating the phase of the compensation signal according to an adaptive compensation method to obtain an optimal low-noise signal; Based on the optimal low-noise signal, a magnetic particle response signal is collected to obtain the magnetic particle nonlinear response signal.

3. The method for improving MPI image resolution based on a physical model according to claim 2, characterized in that: The step of compensating the phase of the compensation signal according to the adaptive compensation method to obtain an optimal low-noise signal includes: The compensation signal is compensated, and the step length of the phase compensation each time is ; Determine the compensation signal after compensation Interference signal with feedthrough Whether the low noise signal subtracted in the LNA is stable; If it is stable, the phase change of the compensated signal after adaptive compensation is output and the corresponding low noise signal; where, , n represents the number of adaptive compensation; If it is unstable, continue to compensate the compensation signal.

4. The method for improving MPI image resolution based on a physical model according to claim 3, characterized in that: The feedthrough interference signal The expression is: in, is the magnetic field frequency, For low noise of MPI system, For time; The expression of the low noise signal is: in, ; The expression of the nonlinear response signal of the magnetic particles is: in, The magnetic particle response signal.

5. The method for improving MPI image resolution based on a physical model according to claim 4, characterized in that: The expression of the X-space reconstructed image is: in, represents the X-space reconstructed image, represents the spatial concentration distribution of magnetic particles, represents the spatial convolution kernel of the MPI system, Represents the spatial distribution function of the noise.

6. The method for improving MPI image resolution based on a physical model according to claim 4, characterized in that: The step of constructing an X-space imaging physical model based on a TV regularized dual Richardson-Lucy deconvolution algorithm and a spatial iterative deconvolution algorithm, and processing the X-space reconstructed image using the X-space imaging physical model to obtain an MPI image includes: The X-space imaging physical model is constructed based on the double Richardson-Lucy deconvolution algorithm and the spatial iterative deconvolution algorithm based on TV regularization; The MPI system spatial convolution kernel is estimated using the TV regularized dual Richardson-Lucy deconvolution algorithm; The spatial concentration distribution of magnetic particles is solved by using a spatial iterative deconvolution algorithm to obtain the MPI image.

7. The method for improving MPI image resolution based on a physical model according to claim 6, characterized in that: The expression of the TV regularized dual Richardson-Lucy deconvolution algorithm is: in, For the The spatial concentration distribution of magnetic particles in the iteration, For the The spatial concentration distribution of magnetic particles in the iteration, for, The first regularization parameter, is the second regularization parameter, is the convolution kernel estimated in one iteration, is the input for one iteration, () represents the divergence calculation function.

8. The method for improving MPI image resolution based on a physical model according to claim 7, characterized in that: The spatial iterative deconvolution algorithm comprises: According to the sparsity and smoothness characteristics of the MPI image, the minimized cost function is established as: in, is the weight factor of fidelity; , represents the spatial distribution of magnetic particle concentration; , represents the spatial convolution kernel of the MPI system; , represents the X-space reconstructed image; is TV regularization, is the TV regularization parameter; is the l1 regularization term, is the regularization parameter of the sparse prior; beg The derivative of gets: in, .

Images