A gradient coil model calibration method for ultra-low field magnetic resonance imaging

Abstract

The present invention provides a gradient coil model correction method for ultra-low field magnetic resonance imaging, which relates to the technical field of gradient coil control in medical imaging equipment. The method comprises: real-time acquisition of a real voltage signal d(n) and a real current signal x(n) of the gradient coil, where n represents a sampling time; constructing a gradient coil model, wherein the gradient coil model comprises a linear model constructed based on an adaptive filtering algorithm and a nonlinear model constructed based on a neural network algorithm, wherein a first predicted voltage signal #imgabs0# output by the linear model and a second predicted voltage signal #imgabs1# output by the nonlinear model are determined according to the real current signal x(n); a mixed predicted voltage signal #imgabs4# of the gradient coil model is determined according to the first predicted voltage signal #imgabs2# and the second predicted voltage signal #imgabs3#; and, according to the real voltage signal d(n) and the mixed predicted voltage signal #imgabs5#, the weights of the linear model and the weights of the nonlinear model are updated to adapt to gradient coil model correction under complex working conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of gradient coil control in medical imaging equipment, and in particular to a gradient coil model correction method for ultra-low field magnetic resonance imaging. Background Art

[0002] Ultra-low-field magnetic resonance imaging (ULFMRI) is a key branch of magnetic resonance imaging (MRI) and generally refers to MRI systems with magnetic field strengths below 0.5 Tesla. In ULFMRI, the performance requirements for gradient power amplifiers are even higher. During MRI scanning, gradient coils generate gradient magnetic fields by rapidly switching currents. However, under extreme operating conditions (such as high load, insufficient heat dissipation, and coil aging), insulation failure or mechanical deformation can cause arcing, threatening patient safety. Existing technologies pre-set a static load model in the gradient controller to predict the theoretical output voltage, compare it with the actual value, and terminate the scan if a threshold is exceeded. However, this traditional approach has the following limitations: 1) The fixed-parameter model cannot dynamically track load changes, and the error drifts larger with operating conditions. The load characteristics of the gradient coil can be represented as a combination of resistance, inductance, and mutual inductance transformers. The series resistance of the coil and the contact resistance at the terminals are both affected by coil temperature fluctuations. Furthermore, the inductance and mutual inductance transformer parameters are also affected by aging of the coil's copper wire and oxidation of the insulation material. However, the factory default linear model cannot adapt to changes in load characteristics caused by factors such as temperature changes and coil aging, which in turn leads to an increase in the error between the model prediction results and the actual measured voltage, resulting in false positives and false negatives in arc anomaly identification, affecting normal scanning imaging; 2) Nonlinearity neglect: Traditional linear models cannot represent the nonlinear response caused by coil aging and deformation, and the risk of false negatives is high. Summary of the Invention

[0003] In view of this, the object of the present invention is to provide a gradient coil model correction method for ultra-low field magnetic resonance imaging, which uses hybrid load modeling to design a gradient coil model for dynamic correction and compensation of a linear model constructed based on an adaptive filtering algorithm and a nonlinear model constructed based on a neural network algorithm, so as to adapt to the gradient coil model correction under complex working conditions.

[0004] An embodiment of the present invention provides a gradient coil model correction method for ultra-low field magnetic resonance imaging, comprising: real-time acquisition of a true voltage signal d(n) and a true current signal x(n) of a gradient coil, where n represents a sampling time; constructing a gradient coil model, wherein the gradient coil model includes a linear model constructed based on an adaptive filtering algorithm and a nonlinear model constructed based on a neural network algorithm, wherein a first predicted voltage signal output by the linear model is determined according to the true current signal x(n). and the second predicted voltage signal output by the nonlinear model According to the first predicted voltage signal and the second predicted voltage signal Determine the mixed predicted voltage signal of the gradient coil model According to the real voltage signal d (n) and the mixed predicted voltage signal , updating the weight of the linear model and the weight of the nonlinear model to achieve correction of the gradient coil model.

[0005] In some embodiments, the linear model constructed based on the adaptive filtering algorithm and the nonlinear model constructed based on the neural network algorithm include: the expression of the linear model constructed based on the LMS algorithm is , where k is the order, is the first weight; the expression of the nonlinear model constructed based on the neural network algorithm is , where k is the order, is the second weight.

[0006] In some embodiments, according to the first predicted voltage signal and the second predicted voltage signal Determine the mixed predicted voltage signal of the gradient coil model , comprising: a mixed predicted voltage signal of the gradient coil model The expression is ,in yes and The combined weight is the third weight.

[0007] In some embodiments, according to the real voltage signal d(n) and the mixed predicted voltage signal , updating the weight of the linear model and the weight of the nonlinear model to achieve correction of the gradient coil model, including: constructing a correction parameter model, the expression of the correction parameter model is ,in is the step size factor; is the instantaneous voltage error; the gradient coil model is corrected according to the correction parameter model.

[0008] In some embodiments, the order m of the nonlinear model is ≥4 and is a positive integer.

[0009] In some embodiments, after real-time acquisition of the real voltage signal d(n) and the real current signal x(n) of the gradient coil, the method includes: inputting the real voltage signal d(n) and the real current signal x(n) into a trained noise reduction model, outputting the noise-reduced voltage signal and the noise-reduced current signal, and the noise-reduced voltage signal and the noise-reduced current signal are used for training the gradient coil model.

[0010] In some embodiments, the step size factor is determined according to an autocorrelation matrix of the real current signal x(n).

[0011] In some embodiments, based on the instantaneous voltage error obtained at different sampling moments, the state data at the current sampling moment is determined, and the step size factor is determined based on the state data at the current sampling moment using the Q-learning algorithm. ,The current state data includes the instantaneous voltage error mean, instantaneous voltage error variance and load power, and the optimal step size is selected by maximizing the long-term reward.

[0012] In some embodiments, when the absolute value of the instantaneous voltage error is greater than a preset error threshold within T sampling periods, a warning message is issued, where T≥5 and is a positive integer.

[0013] In some embodiments, when the absolute value of the instantaneous voltage error is greater than a preset error threshold within T sampling periods, the method further includes: inputting the instantaneous voltage error into a trained classification model and outputting a fault probability; and issuing the warning information based on the fault probability and the preset probability threshold.

[0014] The embodiments of the present invention provide the following beneficial effects: 1) a gradient coil model that dynamically corrects and compensates for a linear model constructed using an adaptive filtering algorithm and a nonlinear model constructed using a neural network algorithm, making it suitable for gradient coil model correction under complex operating conditions (such as temperature drift and coil aging); 2) dynamic adjustment of the compensation factor through Q-learning reinforcement learning to balance convergence speed and stability; and 3) a joint decision-making process using primary threshold detection and a logistic regression classification model to reduce the false alarm rate.

[0015] Other features and advantages of the present disclosure will be set forth in the following description, or some features and advantages may be inferred or unambiguously determined from the description, or may be learned by practicing the above-mentioned technology of the present disclosure.

[0016] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A block diagram of an implementation of a gradient coil model correction method for ultra-low field magnetic resonance imaging provided by an embodiment of the present invention;

[0019] Figure 2 This is a flow chart of a gradient coil model correction method for ultra-low field magnetic resonance imaging provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0021] To facilitate understanding of this embodiment, a magnetic resonance image analysis system disclosed in an embodiment of the present invention is first introduced in detail.

[0022] Combine Figure 1 and Figure 2 The embodiments of the present invention are described in detail.

[0023] Specifically, the present invention constructs a gradient coil model, which includes a linear model constructed based on an adaptive filtering algorithm and a nonlinear model constructed based on a neural network algorithm.

[0024] Among them, the linear model constructed based on the adaptive filtering algorithm can be simplified to a series combination of equivalent inductance L(n) and equivalent resistance R(n), and the weight vector is defined as , where n represents the sampling time, and k is the order of the LMS (Least Mean Squares) model. The LMS model refers to a linear adaptive filter based on the least mean square algorithm. The process of calculating the linear model predicted voltage is as follows:

[0025] The ordinary differential expression of the linear part of the gradient coil model is:

[0026]

[0027] Where: t is a continuous time variable (unit: second), which represents the actual working time of the gradient coil;

[0028] v(t) is the coil terminal voltage at time t; i(t) is the coil current at time t; L is the equivalent inductance; R is the equivalent resistance.

[0029] By discretizing it, we can get:

[0030]

[0031] The discrete time variable n represents the nth sampling moment (such as n=0,1,2,...), corresponding to time t=n Ts, where Ts is the sampling period; represents the predicted voltage output of the linear model at time n.

[0032] By parameterizing the time-varying inductance L(n) and resistance R(n), and recording the discrete current signal as x(n), we can obtain

[0033]

[0034] Finally, the parameters are converted into weight vectors To express it, we get:

[0035]

[0036] in, Mainly reflects the resistance component, Reflects the inductance component;

[0037] Rewrite the above expression into the expression form of FIR model as follows:

[0038]

[0039] Through the above modeling process, the expression of the linear model is obtained.

[0040] For example, the recursive least squares (RLS) algorithm may be used to replace the LMS algorithm for linear model construction, which has fast convergence but high computational complexity.

[0041] In addition, the nonlinear part of the gradient coil is manifested in eddy currents, c-crosstalk between different axes (X / Y / Z), and other characteristics. They are generally modeled using a mutual inductance transformer, and mathematically, an IIR filter is usually used instead of an FIR filter. However, considering that the B0 field of low-field MRI equipment is much smaller than that of conventional MRI equipment, and the gradient coil voltage is also much smaller, the contribution of the nonlinear part can be further characterized by a high-order FIR (finite impulse response) filter. Specifically, the expression of the nonlinear model is , where k is the order, is the weight, where the order m≥4 and is a positive integer. In this embodiment, m=4 is used as an example for explanation. The nonlinear model of this embodiment adopts a micro neural network with a 4th-order FIR structure, which is trained by offline pre-training (using the real current data of historical monitoring) and online fine-tuning (using the instantaneous voltage error). ) learns the nonlinear characteristics of the gradient coil (such as eddy currents, crosstalk) and updates the weights through backpropagation.

[0042] Both linear and nonlinear models are constructed using FIR models, so the weights can be combined. yes and Therefore, the hybrid prediction voltage signal of the gradient coil model constructed by the present invention is The expression is .

[0043] In this embodiment, the mixed prediction voltage signal The expression is .

[0044] An embodiment of the present invention provides a method for accelerating the calculation of a hybrid model (a linear model built based on the LMS algorithm and a nonlinear model built based on the neural network algorithm) using an FPGA coprocessor, achieving a calculation delay of <50μs. At the same time, 8-bit fixed-point quantization technology is used to compress the neural network weights from 32-bit floating points to 8-bit integers, and the memory usage is reduced from 128KB to <50KB. Intelligent deployment is achieved through hardware acceleration and resource optimization. Therefore, even in a low-computing-power controller of ultra-low-field magnetic resonance imaging, the gradient coil model provided by the present invention can respond and operate quickly.

[0045] The present invention provides an embodiment in which a real voltage signal d(n) and a real current signal x(n) are input into a trained denoising model, and the denoised voltage signal and the denoised current signal are output for training a nonlinear model. The denoised voltage signal and the denoised current signal are used for training the gradient coil model, and adaptive denoising is performed on the collected signal to retain the transient characteristics of the signal and improve the signal-to-noise ratio.

[0046] In the embodiment provided by the present invention, the training process of the nonlinear model constructed based on the neural network includes: taking the real current signal x(n) collected at different sampling times as the input of the model, taking the real nonlinear voltage distortion component =d(n)- As a label, use and Construct the MSE (mean square error) loss function. The training of the model is affected by the step size factor, which affects the stability of the model training.

[0047] According to the real voltage signal d(n) and the hybrid predicted voltage signal , updating the weights of the linear model and the nonlinear model to achieve correction of the gradient coil model, including: constructing a correction parameter model, the expression of the correction parameter model is ,in is the step size factor; is the instantaneous voltage error; the gradient coil model is corrected according to the correction parameter model. Specifically, the Q-learning algorithm is used to determine the step size factor according to the state data at the current sampling moment. The state data includes the instantaneous voltage error mean, instantaneous voltage error variance, and load power. For example, an affine projection algorithm (APA) can be used instead of Q-learning to simplify step size adjustment, but this requires additional storage space.

[0048] It can also be determined based on the autocorrelation matrix of the real current signal x(n) collected in real time, where n represents the sampling time, n≥1, and is a positive integer. Calculate the maximum eigenvalue of the autocorrelation matrix , step factor Need to meet , to ensure the balance between the convergence speed and stability of the algorithm.

[0049] This embodiment also provides a warning judgment scheme. When the absolute value of the instantaneous voltage error is greater than the preset error threshold within T sampling periods, a warning message is issued. T≥5 and is a positive integer to achieve primary threshold detection. At the same time, a logistic regression model is constructed to analyze the input error. The sequence is substituted into the calculation to obtain the probability of arc (fault) occurrence. The logistic regression model is based on a large amount of instantaneous voltage error data during ignition. The model is trained to perform joint early warning of risks through the above-mentioned multi-level arc judgment to improve the accuracy of the early warning.

[0050] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0051] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not limiting, and thus other examples of the exemplary embodiments may have different values.

[0052] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A gradient coil model calibration method for ultra-low field magnetic resonance imaging, characterized in that: include: The real voltage signal d(n) and real current signal x(n) of the gradient coil are collected in real time, where n represents the sampling time. Constructing a gradient coil model, the gradient coil model includes a linear model constructed based on an adaptive filtering algorithm and a nonlinear model constructed based on a neural network algorithm, wherein the first predicted voltage signal output by the linear model is determined according to the real current signal x(n) and the second predicted voltage signal output by the nonlinear model ; The linear model constructed based on the adaptive filtering algorithm and the nonlinear model constructed based on the neural network algorithm include: the expression of the linear model constructed based on the LMS algorithm is: , where k is the order, is the first weight; the expression of the nonlinear model constructed based on the neural network algorithm is , where k is the order, is the second weight; According to the first predicted voltage signal and the second predicted voltage signal Determine the mixed predicted voltage signal of the gradient coil model ; The gradient coil model predicts a mixed voltage signal The expression is ,in yes and The combined weight is the third weight; According to the real voltage signal d(n) and the hybrid predicted voltage signal , updating the weight of the linear model and the weight of the nonlinear model to achieve correction of the gradient coil model.

2. The method according to claim 1, characterized in that After the real voltage signal d(n) and the real current signal x(n) of the gradient coil are acquired in real time, the method further comprises: The real voltage signal d(n) and the real current signal x(n) are input into the trained denoising model, and the denoised voltage signal and the denoised current signal are output. The denoised voltage signal and the denoised current signal are used for training the gradient coil model.

3. The method according to claim 1, characterized in that include: The order m of the nonlinear model is ≥4 and is a positive integer.

4. The method according to claim 1, wherein The method according to the real voltage signal d(n) and the mixed predicted voltage signal , updating the weight of the linear model and the weight of the nonlinear model to achieve correction of the gradient coil model, including: Construct a correction parameter model, the expression of the correction parameter model is ,in is the step size factor; is the instantaneous voltage error; The gradient coil model is corrected according to the correction parameter model.

5. The method according to claim 4, characterized in that include: The step size factor is determined according to the autocorrelation matrix of the real current signal x(n).

6. The method according to claim 4, characterized in that include: Determine the state data at the current sampling moment based on the instantaneous voltage errors obtained at different sampling moments, wherein the state data includes the instantaneous voltage error mean, the instantaneous voltage error variance, and the load power; The step size factor is determined based on the state data at the current sampling moment using the Q-learning algorithm. .

7. The method according to claim 4, characterized in that Also includes: When the absolute value of the instantaneous voltage error is greater than a preset error threshold within T sampling periods, an early warning message is issued, where T is greater than or equal to 5 and is a positive integer.

8. The method according to claim 7, characterized in that When the absolute value of the instantaneous voltage error is greater than a preset error threshold within T sampling periods, the method further includes: Inputting the instantaneous voltage error into the trained classification model and outputting the fault probability; The warning information is issued according to the failure probability and a preset probability threshold.

Images