A method for nonlinear suppression of electromagnetic noise of a magnetic resonance imaging device based on a kernel function
By using kernel functions for nonlinear mapping, the problem of electromagnetic noise suppression in open environments for magnetic resonance imaging equipment was solved, achieving efficient electromagnetic noise suppression and improving image quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV
- Filing Date
- 2023-03-02
- Publication Date
- 2026-05-01
AI Technical Summary
Existing magnetic resonance imaging (MRI) equipment struggles to effectively suppress electromagnetic noise in open environments, and traditional linear mapping methods cannot accurately describe electromagnetic noise in the MRI signal channel, resulting in insufficient noise reduction performance.
A kernel function is used for nonlinear mapping to map the electromagnetic reference channel noise to a high-dimensional feature space. By constructing a nonlinear transfer coefficient matrix and a calibration source matrix, nonlinear fitting and suppression of electromagnetic noise in the magnetic resonance signal channel are achieved.
Without modifying the imaging sequence or increasing the imaging time, the impact of electromagnetic interference on the image quality of open magnetic resonance imaging equipment can be effectively reduced or eliminated, ensuring normal operation of the equipment in various environments.
Smart Images
Figure CN116466280B_ABST
Abstract
Description
A Nonlinear Suppression Method for Electromagnetic Noise in Magnetic Resonance Imaging Equipment Based on Kernel Function Technical Field
[0001] This invention belongs to the field of nuclear magnetic resonance technology and relates to a method for nonlinear suppression of electromagnetic noise in magnetic resonance imaging equipment based on kernel functions. Background Technology
[0002] Magnetic resonance imaging (MRI) systems typically operate in complex electromagnetic environments with multiple noise sources and dynamic variations. Traditional MRI equipment requires specialized shielded environments (e.g., shielded rooms, shielded chambers, shielded cages). Currently, to apply MRI systems to bedside monitoring, pre-hospital diagnosis, and other fields, some active noise reduction systems have been proposed to replace traditional passive electromagnetic shielding methods. However, active noise reduction systems based on image post-processing, such as thresholding and image filtering, generally only target specific types of interference, and some important information in the original image may be lost after processing. Furthermore, active noise reduction systems represented by signal processing methods such as low-rank matrix factorization, model fitting, and deep learning-based reference channel methods still have drawbacks, including the limited types of interference they can suppress, the large amount of data required for model parameter training that cannot cover all interferences, and the long time required for training data acquisition and neural network training. Recently, signal processing methods such as Fourier transform, wavelet analysis, and time-domain analysis have been used to process synchronously acquired magnetic resonance imaging (MRI) signal channels and electromagnetic noise reference channels in the time, frequency, or time-frequency domains to achieve active noise reduction, effectively addressing the shortcomings of these algorithms. The essence of this algorithm is based on the similarity between the noise signals of the MRI signal channel and each electromagnetic noise reference channel, using linear fitting to linearly map the electromagnetic noise reference channel noise signal to the electromagnetic noise signal in the MRI signal channel. However, as the principle of MRI signal generation shows, even the same noise source can exhibit nonlinear relationships between signals generated in coils at different locations and with different structures. Furthermore, there are also nonlinear relationships between the channels in the acquisition circuit of an MRI system. Therefore, the linearly mapped reference channel noise in the above algorithms cannot accurately describe the electromagnetic noise in the MRI signal channel, leading to insufficient noise reduction performance. Therefore, we propose a kernel function-based nonlinear suppression method for electromagnetic noise in MRI equipment to address this problem. Summary of the Invention
[0003] In view of this, the purpose of the present invention is to provide a nonlinear suppression method for electromagnetic noise in magnetic resonance imaging equipment based on kernel functions.
[0004] To achieve the above objectives, the present invention provides the following technical solution:
[0005] A nonlinear suppression method for electromagnetic noise in magnetic resonance imaging equipment based on kernel functions is proposed. The method is as follows:
[0006] The kernel function and the size of the K-space neighborhood are selected to perform nonlinear mapping on the noise of the electromagnetic reference channel, and to construct the target matrix and calibration source matrix in the calibration data for calculating the nonlinear transfer coefficient matrix, respectively. In the entire K-space range, the reconstructed source matrix constructed from the electromagnetic noise reference channel data in the feature space is multiplied by the nonlinear transfer coefficient matrix to obtain the fitted value of the electromagnetic noise in the nonlinearly fitted magnetic resonance signal channel. The obtained fitted value is subtracted from the K-space data of the contaminated magnetic resonance signal channel to achieve the purpose of electromagnetic noise suppression.
[0007] Optionally, when acquiring contaminated magnetic resonance signals in the magnetic resonance signal channel, electromagnetic interference is simultaneously acquired through an electromagnetic noise reference channel; calibration data is additionally acquired using the magnetic resonance signal channel and the electromagnetic noise reference channel, or a portion of the K-space data from each channel is selected as calibration data; a suitable kernel function is selected to map the calibration data from a linearly inseparable low-dimensional space to a linearly separable high-dimensional feature space using a nonlinear mapping method, in which the calibration data of the magnetic resonance signal channel is linearly represented by the calibration data of the electromagnetic noise reference channel.
[0008] Optionally, a suitable neighborhood size is selected along the frequency encoding and phase encoding directions in the calibration data, and this neighborhood is moved along the frequency encoding and phase encoding directions within the calibration data range to traverse the entire calibration data area. A calibration source matrix for calculating the nonlinear transfer coefficients is constructed from the calibration data of each electromagnetic noise reference channel in the feature space according to the neighborhood movement. At the same time, a target matrix for calculating the nonlinear transfer coefficients is also constructed from the calibration data of the magnetic resonance signal channel according to the neighborhood movement. The nonlinear transfer coefficients that minimize the difference between the target matrix and the source matrix are determined by the least squares method and gradient descent optimization algorithm. Then, the same nonlinear mapping is performed on the data in the entire K-space range of each channel, and a source matrix for noise reduction reconstruction is constructed from the data of each electromagnetic noise reference channel. The reconstructed source matrix is multiplied by the obtained nonlinear transfer coefficient matrix to obtain the fitted value of electromagnetic noise in the magnetic resonance signal channel obtained by the nonlinear mapping of the signals of each electromagnetic noise reference channel. Finally, the contaminated magnetic resonance signal is subtracted from the transferred reference noise signal to achieve the purpose of suppressing or eliminating electromagnetic noise.
[0009] The beneficial effects of this invention are that it reduces or eliminates the impact of electromagnetic interference noise on the image quality of open magnetic resonance imaging equipment without modifying the imaging sequence, without adding extra imaging time, and without passive shielding, thereby ensuring that the open magnetic resonance imaging equipment can be used in various required environments.
[0010] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0011] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:
[0012] Figure 1 is a schematic diagram of the sensor placement position in the electromagnetic noise acquisition system;
[0013] Figure 2 is a schematic diagram of a sleeve-type dual-inductive electrode differential output electric field coupling sensor; Figure 2(a) is a structural diagram of the dual-inductive electrode differential output electric field coupling sensor; Figure 2(b) is a schematic diagram of electric field coupling sensor measurement.
[0014] Figure 3 is a flowchart comparing the calibration process of linear and nonlinear active noise reduction algorithms;
[0015] Figure 4 is a flowchart of the multi-channel electromagnetic noise nonlinear active noise reduction algorithm;
[0016] Figure 5 shows the results of data collection from experimental volunteers in an open environment; Figure 5(a) is a comparison of the electromagnetic noise suppression of the contaminated one-dimensional magnetic resonance signal before and after; Figure 5(b) is an image of the experimental volunteer's head before electromagnetic noise suppression; Figure 5(c) is an image of the experimental volunteer's head after electromagnetic noise suppression. Detailed Implementation
[0017] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.
[0018] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0019] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.
[0020] Magnetic resonance imaging (MRI) equipment typically operates in a complex electromagnetic environment with dynamically changing noise sources. If passive electromagnetic interference (EMI) shielding measures are not feasible, active EMI suppression methods are required. For unshielded MRI systems, various electromagnetic noise acquisition sensors are used to obtain EMI reference channel signals. Figure 1 shows their placement within the MRI system. The EMI acquisition coil can be a solenoid coil (single-turn, multi-turn, Helmholtz coil, irregularly shaped coil), a saddle-shaped coil, a planar coil, a flexible coil etched onto a flexible PCB, or an array coil. The optimal structure of the noise acquisition coil is determined using finite element method simulations, and its center frequency and bandwidth are matched using appropriate matching methods. The EMI acquisition coil is placed at suitable angles, such as 0 degrees, 30 degrees, or 45 degrees from the central axis of the RF receiving coil, near the center of the magnetic field or around the magnet and the MRI machine's electronic equipment. Simultaneously, current sensors are placed on the internal connection lines of the MRI equipment, including between the gradient power amplifier and gradient coil, the spectrometer and gradient power amplifier, the spectrometer and RF power amplifier, and the RF front-end box and spectrometer. Current sensors (magnetic sensors) can be resistive shunts, electromagnetic current transformers, Hall effect current sensors (open-loop and closed-loop), fluxgate current sensors, Rogowski coils, magnetoresistive current sensors, fiber optic current sensors, etc. In addition, multiple inductive electrode electric field coupling sensors, spherical electric field coupling sensors, ring electric field coupling sensors, dual-inductive electrode electric field coupling sensors, and differential output electric field coupling sensors are placed on three-phase power lines. For example, Figure 2(a) is a schematic diagram of a sleeve-type dual-inductive electrode differential output electric field coupling sensor, mainly including an inner copper ring, an outer copper ring, and an insulating support. The inner and outer copper rings are concentric rings with different radii and are both fixed on the insulating support, which has through holes for passing through and fixing the conductor being measured. As shown in Figure 2(b), this electric field coupling sensor is used by fitting it onto the transmission line for measurement.
[0021] A method of synchronous sampling of the electromagnetic noise reference channel and the magnetic resonance signal channel is employed to suppress dynamically changing electromagnetic noise in real time. The signal acquisition circuit controls the synchronous sampling of each electromagnetic noise acquisition sensor in the electromagnetic noise reference channel and the RF receiving coil in the magnetic resonance signal channel. The data acquired by each channel is amplified by its respective preamplifier for low-noise, high-gain amplification, and then input to the signal conditioning circuit for further amplification, low-pass, and high-pass signal processing. Finally, the sampling card and its peripheral circuitry acquire the signal, thereby obtaining the contaminated magnetic resonance signal and the synchronously acquired reference electromagnetic noise signals.
[0022] The nonlinear active noise reduction algorithm can be divided into two steps: calibrating the nonlinear transfer coefficients and reconstructing the K-space data of the denoised magnetic resonance signal channels. The algorithm operates on the K-space data acquired from each channel. The specific implementation plan is as follows:
[0023] (1) Calibrate the transfer coefficient
[0024] Calibration measurements are performed on the magnetic resonance signal channel and the electromagnetic noise reference channel, and calibration data is collected. Alternatively, a portion of the K-space data can be directly used as calibration data to calculate the nonlinear transfer coefficients of each electromagnetic noise reference channel. To make the description more precise, the symbol definitions are reiterated as follows: Let S... RF (k x ,k y ) = M RF (k x ,k y )+N RF (k x ,k y ) represents the K-space data acquired by the magnetic resonance signal channel, where M RF (k x ,k y ) represents noise-free magnetic resonance signal K-space data, N RF (k x ,k y Let N represent the K-space data of electromagnetic noise; Refl (k x ,k y () represents the K-space data of the electromagnetic noise acquired by the l-th electromagnetic noise reference channel. As shown in Figure 3, existing linear active noise reduction algorithms, based on the similarity of electromagnetic noise in each channel, believe that the electromagnetic noise in the magnetic resonance signal channel can be represented by a linear combination of the noise from each reference channel, i.e. Where w Refl This represents the transfer coefficient corresponding to each electromagnetic noise reference channel obtained through linear fitting. However, the signal induced by the RF coil is as shown in equation (1):
[0025]
[0026] in, For the receiving coil, the unit current at point The radio frequency magnetic field strength generated at the location, V s Let M be the sample volume and M be the magnetization vector.
[0027] As shown in formula (1), even the same electromagnetic noise source will exhibit a nonlinear relationship when generated by coils placed in different locations with different structures. Furthermore, the electromagnetic noise signals acquired by different types of electromagnetic noise detection sensors will also exhibit a nonlinear relationship with the noise signals detected by the magnetic resonance signal channel due to differences in physical structure and spatial location. Moreover, the sampling circuits in each acquisition channel of the magnetic resonance imaging system will also exhibit a nonlinear relationship. Therefore, algorithms that simply consider the electromagnetic noise acquired by the magnetic resonance signal channel as a linear combination of the noise acquired by each reference channel contain errors, leading to a decrease in the noise reduction efficiency of the active noise reduction system. The nonlinear active noise reduction algorithm proposed in this invention can nonlinearly map the reference channel noise signal to a high-dimensional feature space, making the electromagnetic noise of the magnetic resonance signal channel and the reference noise, which are linearly inseparable in the original space, linearly separable in this feature space.
[0028] The flowchart of the nonlinear active noise reduction algorithm is shown in Figure 4. During the calibration process, the noise of each reference channel is nonlinearly mapped to a high-dimensional feature space using methods such as nonlinear mapping functions, kernel functions (linear kernel functions, polynomial kernel functions, Gaussian kernel functions, exponential kernel functions, etc.), and kernel tricks. For example, a polynomial kernel function can be used for nonlinear mapping. The expression for the polynomial kernel function is: Its key feature is that it has an explicit solution given the variables in the expression, while preserving linear relationships in nonlinear relationships, thus covering existing linear active denoising algorithms as a special solution when d=0. When γ=r=1 and d=2, the kernel function has an explicit solution: Under these conditions, the noise signal N in the K-space data of the magnetic resonance signal channel RF (k x ,k y ) can be made by N Refl (k x ,k y ), and N Refl (k x ,k y )×N Refl (k x +1,k yThe linear combination of the values is then used. Next, a suitable neighborhood size, such as 3×3 or 4×4, is given along the phase encoding and frequency encoding directions in K-space. This selected neighborhood is moved along the phase encoding and frequency encoding directions in K-space to traverse the entire calibration data, thus obtaining the target matrix composed of noise signals from the magnetic resonance signal channels and the calibration source matrix composed of noise from each reference channel in high-dimensional space. Then, optimization algorithms such as least squares and gradient descent are used to calculate the nonlinear transition coefficient matrix that minimizes the difference between the target matrix and the calibration source matrix. This concludes the calibration process of the nonlinear active noise reduction algorithm.
[0029] (2) Reconstructing the noise-reduced data
[0030] The reference noise channel K-space data, excluding calibration data, are used with the same kernel function and a neighborhood of the same size to construct a source matrix for reconstruction. Multiplying the reconstructed source matrix by the nonlinear transfer coefficient matrix obtained in the first step yields the electromagnetic interference noise signal of the magnetic resonance channel, obtained by nonlinear fitting using the reference channel noise. Finally, subtracting the nonlinearly mapped reference noise channel K-space data from the contaminated magnetic resonance signal K-space data yields the magnetic resonance signal K-space data with suppressed or eliminated electromagnetic interference noise, achieving nonlinear active noise reduction. The electromagnetic noise acquisition sensor was configured in an open, portable ultra-low field magnetic resonance imaging system according to the method described above. The data acquisition results from experimental volunteers in an open environment (without electromagnetic shielding) are shown in Figure 5. Figure 5(a) shows the results of one-dimensional magnetic resonance signals before and after nonlinear active denoising, as well as the synthesized noise of the electromagnetic noise reference channel. This experimental data demonstrates that the proposed nonlinear active denoising system effectively suppresses electromagnetic noise. Figures 5(b) and 5(c) show the head imaging results of volunteers in an open environment before and after nonlinear active denoising, respectively. The signal-to-noise ratio (SNR) of the images improved from 2.1 before denoising to 21.8 after denoising. The one-dimensional and image results in Figure 5 prove that the kernel-function-based nonlinear electromagnetic noise suppression algorithm for magnetic resonance imaging devices proposed in this invention effectively suppresses electromagnetic noise encountered by the magnetic resonance imaging system in an open environment.
[0031] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for nonlinear suppression of electromagnetic noise in magnetic resonance imaging equipment based on kernel functions, characterized in that: The method is as follows: A kernel function and the size of the K-space neighborhood are selected to perform nonlinear mapping on the noise of the electromagnetic reference channel, and a target matrix and a calibration source matrix are constructed in the calibration data to calculate the nonlinear transfer coefficient matrix. Within the entire K-space range, the reconstructed source matrix constructed from the electromagnetic noise reference channel data in the feature space is multiplied by the nonlinear transfer coefficient matrix to obtain the fitted value of the electromagnetic noise in the nonlinearly fitted magnetic resonance signal channel. The obtained fitted value is subtracted from the K-space data of the contaminated magnetic resonance signal channel to achieve electromagnetic noise suppression. In the calibration data, a suitable neighborhood size is selected along the frequency encoding and phase encoding directions, and the neighborhood is moved along the frequency encoding and phase encoding directions within the calibration data range to traverse the entire calibration data area. A calibration source matrix for calculating nonlinear transfer coefficients is constructed by shifting the calibration data of each electromagnetic noise reference channel in the feature space according to the neighborhood. At the same time, a target matrix for calculating nonlinear transfer coefficients is also constructed by shifting the calibration data of the magnetic resonance signal channel according to the neighborhood. The nonlinear transfer coefficients that minimize the difference between the target matrix and the source matrix are determined by the least squares method and the gradient descent method optimization algorithm. Then, the same nonlinear mapping is performed on the data in the entire K-space range of each channel, and a source matrix for noise reduction reconstruction is constructed from the data of each electromagnetic noise reference channel. The reconstructed source matrix is multiplied by the obtained nonlinear transfer coefficient matrix to obtain the fitted value of electromagnetic noise in the magnetic resonance signal channel obtained by the nonlinear mapping of the electromagnetic noise reference channel signal. Finally, the contaminated magnetic resonance signal is subtracted from the transferred reference noise signal to achieve the purpose of suppressing or eliminating electromagnetic noise.
2. The method for suppressing nonlinear electromagnetic noise in magnetic resonance imaging equipment based on kernel functions according to claim 1, characterized in that: When acquiring contaminated magnetic resonance signals in the magnetic resonance signal channel, electromagnetic interference is simultaneously acquired through the electromagnetic noise reference channel; additional calibration data is acquired using the magnetic resonance signal channel and the electromagnetic noise reference channel, or a portion of the K-space data from each channel is selected as calibration data. By selecting an appropriate kernel function, the calibration data is mapped from a linearly inseparable low-dimensional space to a linearly separable high-dimensional feature space through a nonlinear mapping method. In this feature space, the calibration data of the magnetic resonance signal channel is linearly represented by the calibration data of the electromagnetic noise reference channel.