An ultra-low-field magnetic resonance center frequency adaptive compensation system and method

Abstract

The application provides a kind of ultra-low field magnetic resonance center frequency adaptive compensation system and method, the system includes: environmental magnetic field sensing unit, including three-axis fluxgate magnetometer and analog-digital converter, the three-axis fluxgate magnetometer noise base is not more than 10 pT / √Hz, bandwidth covers direct current frequency band and upper limit is not less than 3kHz, for real-time acquisition environmental magnetic field vector in the vicinity of magnet;Real-time processing unit is realized based on field programmable gate array FPGA, is configured to carry out digital low-pass filtering processing to the digital signal output by the environmental magnetic field sensing unit, and converts the filtered environmental magnetic field data into the frequency deviation Δf at the isocenter of magnet according to the pre-calibrated transfer function model;Spectrometer execution unit is electrically connected with the numerical control oscillator NCO of spectrometer, is configured to write the compensation amount corresponding to the frequency deviation Δf into the phase accumulator of the NCO in the form of frequency control word, dynamically adjusts center frequency in the process of pulse sequence execution.

Description

Technical Field

[0001] This invention belongs to the field of hardware control and signal processing of magnetic resonance imaging (MRI) systems. Specifically, it relates to a system and method for eliminating image artifacts, geometric distortions, and signal loss in complex electromagnetic interference environments, particularly in scenarios near DC traction systems such as subways and light rails. This is achieved by dynamically compensating for larmor frequency drift through real-time sampling of environmental magnetic field fluctuations. This invention is particularly applicable to ultra-low field (ULF, magnetic field strength less than 0.1 T) MRI systems in the vertical magnetic field direction. Background Technology

[0002] In recent years, ultra-low field magnetic resonance imaging (MRI) technology has attracted widespread attention due to its advantages such as extremely low manufacturing cost, no need for expensive active shielding rooms, open structure (alleviating claustrophobia), and high safety for metal implants. Taking a 0.08 T (80 mT) vertical field system as an example, its corresponding proton larmor frequency is approximately 3.4 MHz; the 0.2 T system corresponds to a larmor frequency of approximately 8.5 MHz.

[0003] However, ultra-low field magnetic resonance imaging (MRI) systems face more severe external interference challenges than high field MRI systems. Unlike traditional high field MRI systems that rely on heavily shielded magnetic chambers, ultra-low field systems typically operate in unshielded or lightly shielded environments to maintain their portability and low cost advantages. More importantly, although the absolute value of the external interference magnetic field is small, its proportion is significantly increased relative to the low-field main magnetic field of 0.08 T, meaning that even small fluctuations in the ambient magnetic field can lead to noticeable frequency drift.

[0004] According to the Larmor equation ω0 = γB0, the proton precession frequency is strictly proportional to the strength of the main magnetic field. At 0.08 T, the magnetic field bandwidth is narrow, and even small magnetic field fluctuations (such as those in the nT to μT range caused by subways) can lead to severe image ghosting or geometric distortion. In the site selection of urban hospitals, subway and light rail systems are among the largest sources of magnetic field interference. Measured data show that within a range of 50 to 200 meters from a subway line, the magnetic field fluctuation amplitude can reach 100 nT to 5 μT; for 0.08 T MRI, a 1 μT fluctuation corresponds to a frequency drift of approximately 42.6 Hz, which, in ultra-low field systems with an imaging bandwidth of only a few kHz, can cause a displacement of several pixels.

[0005] Existing frequency compensation schemes for ultra-low field unshielded environments have significant limitations: 1. Passive shielding (silicon steel sheet) is too expensive and bulky, which goes against the original intention of ultra-low field MRI to be portable and low-cost.

[0006] 2. Active Coil Cancellation: Although effective, it consumes a lot of power, is complex to install, and requires an additional large coil system.

[0007] 3. Pure software post-processing method: It cannot solve the phase loss and loss of signal during the acquisition process caused by frequency drift, and can only perform limited image restoration.

[0008] Therefore, there is an urgent need for a frequency tracking scheme based on real-time feedback from pure electronics, which can dynamically adjust the system operating frequency before and during data acquisition to suppress interference at the source. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides an ultra-low field magnetic resonance center frequency adaptive compensation system, comprising: an environmental magnetic field sensing unit, including a three-axis fluxgate magnetometer and an analog-to-digital converter, wherein the noise floor of the three-axis fluxgate magnetometer does not exceed 10 pT / √Hz, and its bandwidth covers the DC frequency band with an upper limit of not less than 3kHz, used to collect environmental magnetic field vectors in real time near the magnet; a real-time processing unit, implemented based on a field-programmable gate array (FPGA), configured to perform digital low-pass filtering on the digital signal output by the environmental magnetic field sensing unit, and convert the filtered environmental magnetic field data into a frequency deviation Δf at the magnetic center of the magnet according to a pre-calibrated transfer function model; a spectrometer execution unit, electrically connected to the numerically controlled oscillator (NCO) of the spectrometer, configured to write the compensation amount corresponding to the frequency deviation Δf into the phase accumulator of the NCO in the form of a frequency control word, and dynamically adjust the center frequency during pulse sequence execution; and a transfer function calibration module, used during the system installation and commissioning phase to map and model the frequency deviation between the external probe and the magnetic center of the magnet based on the minimum mean square error (LMS) algorithm, and solve for the optimal transfer function coefficients.

[0010] In one embodiment, the analog-to-digital converter is a 24-bit sigma-delta analog-to-digital converter with a sampling rate of 10 kS / s to 50 kS / s.

[0011] In one embodiment, the real-time processing unit further includes a phase integration submodule for calculating the cumulative phase error φ(t) = 2π∫Δf(t)dt in real time and sending the phase correction value to the digital downconversion (DDC) module of the spectrometer.

[0012] In one embodiment, the three-axis fluxgate magnetometer is fixedly installed near the magnet, within the linear operating region outside the 5 Gauss line of the main magnetic field.

[0013] In one embodiment, the cutoff frequency of the digital low-pass filter is set to 100 Hz to 200 Hz.

[0014] In one embodiment, the spectrometer execution unit is connected to the NCO via a high-speed digital interface, supporting dynamic modification of the frequency control word during pulse sequence operation.

[0015] In one implementation, the real-time processing unit is based on a Xilinx Zynq-7000 SoC or a Kintex series field-programmable gate array, utilizing its parallel processing capabilities to achieve microsecond-level low-latency processing.

[0016] In one embodiment, the pulse sequence is a spin echo (SE) sequence, including a 90° radio frequency excitation pulse, a 180° refocusing pulse, and a signal acquisition stage; the spectrometer execution unit adjusts the NCO frequency before the application of the 90° radio frequency excitation pulse and before the application of the 180° refocusing pulse, respectively.

[0017] In one implementation, the system operates fully automatically. After the transfer function calibration is completed during the system calibration phase, all monitoring and compensation operations require no manual intervention.

[0018] In one embodiment, during the calibration process, the transfer function calibration module places a standard nuclear magnetic resonance (NMR) probe or water phantom at the isomagnetic center of the magnet, simultaneously acquires the output signal of the external probe and the reference signal at the isomagnetic center, and trains the transfer function coefficient matrix using the LMS algorithm.

[0019] In one embodiment, the real-time processing delay of the frequency deviation Δf is on the order of microseconds, which meets the timing requirements of the ultra-low field magnetic resonance imaging sequence.

[0020] This invention also provides an adaptive compensation method for the center frequency of ultra-low field magnetic resonance imaging, applied to the aforementioned adaptive compensation system for the center frequency of ultra-low field magnetic resonance imaging, comprising the following steps: Calibration step: During the system installation and commissioning phase, a standard nuclear magnetic resonance probe or water phantom is placed at the isomagnetic center of the magnet, and the output signal of the external three-axis fluxgate magnetometer and the reference signal at the isomagnetic center are simultaneously acquired. The transfer function coefficient matrix between the external magnetic field and the internal frequency offset is trained based on the minimum mean square error (LMS) algorithm; Real-time monitoring step: During magnetic resonance imaging, the three-axis fluxgate magnetometer acquires the environmental magnetic field vector data near the magnet in real time, and after analog-to-digital conversion, it is sent to a field-programmable gate array (FPGA) for processing; Filtering step: The environmental magnetic field vector data is subjected to digital low-pass filtering with a cutoff frequency of 100 Hz to 200 Hz. Hz, filtering out high-frequency noise; Frequency deviation calculation step: using the transfer function model established in the calibration step, the filtered environmental magnetic field data is mapped to the real-time frequency deviation Δf at the magnetic center of the magnet; Compensation execution step: the frequency deviation Δf is converted into the corresponding NCO frequency control word, written to the spectrometer's digitally controlled oscillator through a high-speed digital interface, and the center frequency is updated before the radio frequency excitation of the next pulse sequence; Phase compensation step: the cumulative phase error φ(t) = 2π∫Δf(t)dt is calculated in real time, and the phase correction value is sent to the spectrometer's digital downconversion module to compensate for the phase error in the acquired data.

[0021] In one embodiment, the calibration steps include: preparation: placing a standard NMR probe or uniform water phantom at the magnetic center of the magnet, ensuring the reference probe is in the optimal measurement position. Recording the system reference frequency f0 at this time; synchronous data acquisition: simultaneously acquiring the three outputs (Bx, By, Bz) of the external triaxial fluxgate magnetometer and the frequency reading f_ref of the NMR probe at the magnetic center under different environmental magnetic field conditions; LMS algorithm iteration: using the external probe data as the input vector x(n) = [Bx(n), By(n), Bz(n), 1] The desired response is determined by the frequency deviation at the isomagnetic center, Δf_target(n) = f_ref(n) - f0. The weight coefficient vector w(n) = [Gx, Gy, Gz, offset] is iteratively updated using the LMS algorithm. :w(n+1) = w(n) + μ·e(n)·x(n), where e(n) = Δf_target(n) - w(n) • x(n) is the estimation error, and μ is the step size factor; Convergence criterion: When the mean square value of the estimation error drops below the preset threshold, or the number of iterations reaches the preset upper limit, the iteration stops, and the weight coefficient vector at this time is the optimal transfer function coefficient; Verification: A portion of the collected data is retained for verification, and the verification data is substituted into the calibrated transfer function model to check whether the deviation between the model prediction value and the expected value meets the accuracy requirements; If not, the calibration process is repeated.

[0022] In one implementation, the sampling rate of the real-time monitoring step is not less than 10 kS / s.

[0023] In one implementation, in the spin echo (SE) sequence, the compensation execution step completes the first frequency adjustment before the application of the 90° RF pulse and the second frequency adjustment before the application of the 180° RF pulse.

[0024] In one embodiment, the compensation execution step performs center frequency compensation in the radio frequency transmission (TX) phase and the signal reception (RX) phase of the pulse sequence, respectively.

[0025] In one implementation, the frequency offset calculation step models the system as a linear time-invariant LTI system, implemented using a discrete-domain model in the form of an FIR filter.

[0026] In one implementation, the total delay for all processing steps, from sampling the ambient magnetic field to updating the NCO frequency, is on the order of microseconds.

[0027] In one embodiment, the digital low-pass filter in the filtering process is a finite-length unit impulse response filter.

[0028] In one implementation, all steps are completed in real time during the execution of the magnetic resonance pulse sequence, without occupying an additional independent scan time window.

[0029] In one implementation, after each compensation is performed, the compensation effect is continuously verified in the next pulse sequence cycle. If the frequency deviation still exceeds the preset range, a new round of compensation is triggered.

[0030] This invention samples ambient magnetic field fluctuations in real time and uses an FPGA real-time processing unit to map these changes into a spectrometer center frequency adjustment. This dynamically compensates for Larmor frequency drift during pulse sequence execution, eliminating image artifacts, geometric distortions, and signal loss. It also offers the following significant advantages: Millisecond to microsecond response: The real-time processing unit is implemented based on the parallel processing architecture of the field-programmable gate array (FPGA), and the processing latency reaches the millisecond or even microsecond level. It is fully adapted to the short TR and fast acquisition working mode of the ultra-low field magnetic resonance system, and achieves a qualitative leap in second-level response compared with the existing AFC system. Real-time phase compensation: The phase integration submodule calculates the accumulated phase error in real time and performs phase compensation during the digital downconversion stage, ensuring the relative stationarity of the spin system in the rotating coordinate system and effectively eliminating phase-related image artifacts. Pre-calibrated transfer function accurate mapping: The LMS algorithm is used to establish an accurate mapping relationship between the external probe and the internal frequency offset in the field, ensuring the accuracy of compensation and avoiding the error caused by directly inferring the internal frequency offset based solely on external magnetic field data; Fully automated operation: After the system completes the transfer function calibration during the calibration phase, all monitoring and compensation operations are performed automatically without operator intervention and are transparent to the operators. Wide frequency response coverage: The bandwidth of the environmental magnetic field sensing unit covers the DC frequency band and the upper limit is not less than 3kHz. Combined with a digital low-pass filter with a cutoff frequency of 100 Hz to 200 Hz, it can simultaneously track slow drift (low frequency) and fast fluctuation (medium frequency), covering the main frequency components of typical interference sources such as subways and light rails. Attached Figure Description

[0031] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 Flowchart for transfer function calibration; Figure 2 Imaging effect without the technical solution of this invention under external magnetic field interference environment; Figure 3 Imaging effect after applying the technical solution of this invention under external magnetic field interference environment. Detailed Implementation

[0032] The present disclosure will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present disclosure, but do not limit the present disclosure in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present disclosure. These all fall within the protection scope of the present disclosure.

[0033] Example 1: Overall System Structure The system comprises four core functional units: an environmental magnetic field sensing unit, a real-time processing unit, a spectrometer execution unit, and a transfer function calibration module.

[0034] Ambient magnetic field sensing unit: A three-axis fluxgate magnetometer with a noise floor not exceeding 10 pT / √Hz and a bandwidth covering DC to 3 kHz is selected, with a measurement range covering environmental interference (typically ±100 μT is sufficient). The three-axis fluxgate magnetometer is fixedly installed near the scanning area, within the linear operating region outside the 5 Gauss line of the main magnetic field, to avoid saturation of the main magnetic field. The three-axis fluxgate magnetometer outputs three analog voltage signals, corresponding to the magnetic field components of the X, Y, and Z axes, respectively. A 24-bit sigma-delta analog-to-digital converter (such as AD7760) is selected, with a sampling rate set to 10 kS / s to 50 kS / s, converting the analog voltage signals output by the three-axis fluxgate magnetometer into 24-bit digital signals, which are then sent to the real-time processing unit.

[0035] Real-time processing unit: Utilizing an FPGA development board based on the Xilinx Zynq-7000 SoC or Kintex series, it achieves microsecond-level low-latency processing through its parallel processing capabilities. The FPGA internal logic includes the following processing modules: First stage, digital low-pass filter: Input data passes through a finite-length unit impulse response (FIR) digital low-pass filter with a cutoff frequency set between 100 Hz and 200 Hz to filter out high-frequency noise and environmental electromagnetic interference; Second stage, frequency offset calculation module: Applying a pre-calibrated transfer function model, it converts the filtered triaxial magnetic field data into a frequency deviation Δf at the isomagnetic center using a three-dimensional linear gain matrix; Third stage, phase integration module: Calculates the cumulative phase error φ(t) = 2π∫Δf(t)dt in real time and outputs this phase correction value in digital form to the spectrometer's digital down-conversion (DDC) module.

[0036] Spectrometer Execution Unit: The spectrometer must have an open NCO frequency update interface, supporting dynamic modification of the Frequency Tuning Word (FTW) during sequence runtime. The controller converts the calculated frequency deviation Δf into the corresponding FTW and sends it to the phase accumulator of the spectrometer's NCO via a high-speed digital interface (such as SPI or parallel bus). The NCO frequency update is completed before the RF excitation of the next pulse sequence, ensuring that the next 90° pulse and 180° pulse are applied at the corrected frequency.

[0037] Transfer function calibration module: During the calibration phase, a standard nuclear magnetic resonance (NMR) probe or water phantom is placed at the isomagnetic center of the magnet. Data is simultaneously acquired by an external triaxial fluxgate magnetometer and the reference probe at the isomagnetic center. Using the minimum mean square error (LMS) algorithm, with the external probe data as input and the reference frequency at the isomagnetic center as the desired output, the optimal transfer function coefficient matrix is ​​iteratively solved. This calibration process is completed in one step during the system installation and commissioning phase.

[0038] The entire compensation process is as follows: Real-time acquisition of the ambient magnetic field by a three-axis fluxgate magnetometer → ADC digitization → FPGA digital low-pass filtering → Δf calculation by transfer function model → Cumulative phase error calculation → NCO frequency control word update (spectrometer execution unit) → DDC phase compensation (real-time processing unit output) → Closed-loop verification.

[0039] Example 2: Transfer Function Calibration Procedure Reference Figure 1 The following details the process and algorithm for transfer function calibration.

[0040] The goal of transfer function calibration is to establish an accurate mapping relationship between the external probe measurements and the internal frequency deviation of the magnet caused by the ambient magnetic field. Due to the spatial inhomogeneity of the main magnetic field and the spatial difference between the probe installation position and the isomagnetic center, the magnetic field value measured by the external probe cannot be directly equated to the actual magnetic field value at the isomagnetic center. There exists a transfer function between the two that is closely related to the probe position and the magnet design.

[0041] Calibration Model: The system is modeled as a linear time-invariant (LTI) system. In the discrete domain, the triaxial transfer function can be represented as a multi-order FIR filter. In practical implementation, considering the simple structure and spatial uniformity of the low-field magnet, the above multi-order FIR model can be simplified to a single-order three-dimensional linear gain matrix: Δf = Gx·Bx + Gy·By + Gz·Bz + offset, where Gx, Gy, Gz, and offset are the transfer function gain coefficients and system bias in the three-axis directions, both of which are determined by the LMS algorithm during the system calibration phase.

[0042] Calibration steps: Step 1: Preparation. Place a standard NMR probe or uniform water phantom at the magnetic center of the magnet, ensuring the reference probe is in the optimal measurement position. Record the system reference frequency f0 at this point.

[0043] Step 2: Synchronous data acquisition. Under different environmental magnetic field conditions, simultaneously acquire the three outputs (Bx, By, Bz) of the external three-axis fluxgate magnetometer and the frequency reading f_ref of the NMR probe at the isomagnetic center.

[0044] Step 3: LMS algorithm iteration. External probe data is used as the input vector x(n) = [Bx(n), By(n), Bz(n), 1] The desired response is determined by the frequency deviation at the isomagnetic center, Δf_target(n) = f_ref(n) - f0. The weight coefficient vector w(n) = [Gx, Gy, Gz, offset] is iteratively updated using the LMS algorithm. :w(n+1) = w(n) + μ·e(n)·x(n), where e(n) = Δf_target(n) - w(n) ·x(n) is the estimation error, and μ is the step size factor (which needs to be chosen appropriately to balance the convergence speed and steady-state error).

[0045] Step 4: Convergence Determination. When the mean square value of the estimation error drops below a preset threshold, or the number of iterations reaches a preset upper limit, the iteration stops. The weight coefficient vector at this point is the optimal transfer function coefficient.

[0046] Step 5: Verification. Reserve a portion of the collected data for verification. Substitute the verification data into the calibrated transfer function model and check if the deviation between the model's predicted values ​​and the expected values ​​meets the accuracy requirements. If not, repeat the calibration process.

[0047] Example 3: Signal Processing Link of Real-Time Processing Unit The following details the complete signal processing chain of the real-time processing unit from data acquisition to frequency offset output.

[0048] Signal Acquisition: The three-axis fluxgate magnetometer outputs three analog voltage signals, corresponding to the magnetic field components in three orthogonal directions in space. These three signals are fed into a 24-bit sigma-delta ADC after passing through their respective independent analog front-ends (including instrumentation amplifiers and anti-aliasing filters). The ADC sampling rate is set from 10 kS / s to 50 kS / s, outputting three 24-bit digital signals Bx(n), By(n), and Bz(n), where n is the sampling time index.

[0049] Digital Low-Pass Filtering: The three digital signals are each passed through a digital low-pass filter within the FPGA. This filter employs a finite-length unit impulse response (FIR) structure, with an order ranging from 32 to 128 (selectable based on real-time requirements and resource trade-offs), and a cutoff frequency set between 100 Hz and 200 Hz. The filter design must meet the requirement of a group delay of less than 1 ms to satisfy the low-latency constraint of the entire processing chain. The filtered signals are denoted as Bx_f(n), By_f(n), and Bz_f(n).

[0050] Frequency offset calculation: The filtered three-axis data is transformed using a pre-calibrated three-dimensional linear gain matrix: Δf(n) = Gx·Bx_f(n) + Gy·By_f(n) + Gz·Bz_f(n) + offset, where Gx, Gy, Gz, and offset are the transfer function coefficients determined during the calibration phase. The calculated Δf(n) is a scalar representing the estimated frequency offset at the current moment, in Hz.

[0051] Phase integration: The cumulative phase error is calculated through discrete integration: φ(n) = φ(n-1) + 2π·Δf(n)·Δt, where Δt is the sampling interval (the reciprocal of the sampling rate), and φ(n) is the cumulative phase error in radians. This integration process continues throughout the entire execution of the pulse sequence.

[0052] Data Output: The real-time processing unit outputs two digital signals: one is the NCO frequency control word increment ΔFTW, corresponding to the current frequency deviation Δf(n); the other is the phase correction value φ(n), which is sent to the spectrometer's DDC module. Both are output in a pipelined manner with delays controlled in the microsecond range.

[0053] Example 4: Pulse Sequence Timing Integration and Compensation Operation The following details the frequency adjustment and phase compensation operations at key time points during the execution of the spin echo (SE) sequence.

[0054] Spin echo (SE) sequences are the most basic and commonly used pulse sequences in magnetic resonance imaging (MRI). Their basic timing is as follows: 1. A 90° excitation pulse is applied at the beginning of the repetition time (TR), flipping the longitudinal magnetization vector to the transverse plane; 2. The TE / 2 astigmatism period, from after the 90° pulse to before the 180° pulse, during which the spins become astigmatic due to the inhomogeneity of the main magnetic field; 3. A 180° refocusing pulse is applied at TE / 2, gradually refocusing the astigmatic spins; 4. The TE / 2 convergence period, from after the 180° pulse to echo formation, during which the spins gradually converge; 5. Echo signal acquisition (ADC sampling), where the spin echo formed at TE is digitally acquired by the ADC.

[0055] The compensation operation of this invention is precisely integrated with the aforementioned SE timing: T0 (before the 90° pulse is applied): The FPGA reads the current triaxial magnetic field data and calculates the current frequency deviation Δf(n). The spectrometer execution unit writes the compensation amount into the NCO frequency control word to ensure that the 90° pulse is applied at the correct correct frequency to excite the correct spin resonance.

[0056] From T0 to T1 (TE / 2 dephasing period): The phase error φ(t) caused by the continuous integral magnetic field fluctuation inside the FPGA is 2π∫Δf(t)dt. At time T1 (before the 180° pulse is applied), the FPGA updates the NCO frequency control word again to ensure that the 180° pulse is applied at the correct refocusing frequency.

[0057] From T1 to TE (TE / 2 phase integration period to echo acquisition): Phase integration continues. The FPGA sends the accumulated phase correction φ(n) to the spectrometer's DDC module. The DDC module adds the phase rotation exp(-j·φ) to the received raw sampled data to complete the phase demodulation correction.

[0058] ADC sampling: The receiving NCO frequency has been adjusted to the compensated value. The phase error of the sampled data is compensated by the DDC module to eliminate the phase accumulation effect caused by frequency drift.

[0059] The entire process is completed in real time during the pulse sequence execution, without occupying an additional independent scan time window. The total delay is in the microsecond range, far less than the typical TR value (50 ms to 200 ms), ensuring the timeliness of the compensation.

[0060] Example 5: System Performance Indicators and Typical Application Scenarios Based on the above embodiments, the ultra-low field magnetic resonance center frequency adaptive compensation system of the present invention can achieve the following performance indicators: Environmental sensing bandwidth: DC to 3 kHz, covering the main frequency components of typical low-frequency interference sources such as subways and light rails.

[0061] Magnetic field noise floor: The noise floor of the three-axis fluxgate magnetometer does not exceed 10 pT / √Hz and can sense nT level magnetic field changes.

[0062] ADC resolution: 24-bit, corresponding to a theoretical resolution of approximately 0.06 nT (within the ±100 μT range).

[0063] Frequency offset calculation accuracy: microsecond-level processing delay, meeting the timing requirements of ultra-low field imaging sequences.

[0064] Phase compensation accuracy: Real-time integration accumulates phase error, and precise phase rotation compensation is performed during the DDC stage.

[0065] Transfer function calibration accuracy: Using the LMS algorithm and a standard NMR probe, a calibration residual of less than 0.1 ppm can be achieved.

[0066] Explanation of the technical effects of this invention: The technical problem this invention aims to solve is how to reduce the impact of external magnetic field interference on image quality during ultra-low field magnetic resonance imaging (ULMRI) scanning under such interference. Part of the Shanghai Sixth People's Hospital Xuhui Branch is located above the Shanghai Metro Line 9 tunnel. In an unshielded environment, the imaging quality of ULMRI is severely affected by electromagnetic interference from the metro. To verify the technical effectiveness of this invention, tests were conducted near the location above the metro tunnel without any shielding facilities.

[0067] As shown in Figure 2, a baseline scan of a standard square water model was performed in an ultra-low field environment without activating the center frequency adaptive compensation system described in this invention. This scan employed a head coil, with a scanning field of view set to 270 mm × 270 mm, and used a T1-weighted three-dimensional fast field echo sequence (T1-FFE3D), which is highly sensitive to the inhomogeneity and drift of the main magnetic field. Due to magnetic field fluctuations generated by external DC traction systems such as subways, significant Larmor frequency drift occurred during the pulse sequence execution. During RF excitation and data reception, the uncompensated frequency deviation was converted into a continuously accumulating phase error. This uncontrolled phase accumulation led to severe intravoxel spin astigmatism. Simultaneously, the overall frequency shift also caused errors in the spatial readout gradient localization, macroscopically manifested as large-area signal loss inside the water model and geometric distortion (edge ​​bending) at the boundaries, as shown in Figure 2, resulting in a severely degraded overall signal-to-noise ratio.

[0068] As shown in Figure 3, after activating the compensation system of this invention, the system ensures the relative stillness of the spin system in the rotating coordinate system through this millisecond-level feedback control. The corrected image successfully eliminates the phase error caused by frequency drift, restores a high degree of consistency in the signal inside the water model, eliminates uneven brightness, and effectively restores the signal-to-noise ratio of ultra-low field imaging.

[0069] Typical application scenarios of this invention include: Hospitals along subway lines: Installed in hospitals along urban subway lines, the ultra-low field magnetic resonance imaging equipment directly uses the compensation system of this invention to counteract the magnetic field interference caused by the subway DC traction system, achieving stable imaging without the need for expensive shielding engineering modifications. Emergency room and ICU bedside: In the emergency room and ICU environment, medical equipment is dense and the electromagnetic environment is complex. The millisecond-level adaptive compensation capability of this invention ensures that image quality is maintained in a dynamically changing electromagnetic environment. Mobile medical vehicles: In mobile scenarios such as mobile stroke units or field hospitals, the external electromagnetic environment is variable and unpredictable. The transfer function calibration and real-time compensation features of this invention enable it to automatically adapt to different environments without the need for manual recalibration.

[0070] The specific embodiments of this disclosure have been described above. It should be understood that this disclosure is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of this disclosure, which do not affect the substantive content of this disclosure. The above-described preferred features can be used in any combination without conflict.

Claims

1. An adaptive compensation system for the center frequency of ultra-low field magnetic resonance, characterized in that, include: An environmental magnetic field sensing unit includes a three-axis fluxgate magnetometer and an analog-to-digital converter. The noise floor of the three-axis fluxgate magnetometer is no more than 10 pT / √Hz, and its bandwidth covers the DC frequency band with an upper limit of no less than 3kHz. It is used to collect environmental magnetic field vectors in real time near the magnet. The real-time processing unit, implemented based on a field-programmable gate array (FPGA), is configured to perform digital low-pass filtering on the digital signal output by the environmental magnetic field sensing unit, and convert the filtered environmental magnetic field data into a frequency deviation Δf at the magnetic center of the magnet according to a pre-calibrated transfer function model. The real-time processing unit also includes a phase integration submodule, which is used to calculate the cumulative phase error φ(t) = 2π∫Δf(t)dt in real time, and send the corresponding phase correction value to the digital down-conversion (DDC) module of the spectrometer. The spectrometer execution unit, electrically connected to the digitally controlled oscillator (NCO) of the spectrometer, is configured to write the compensation amount corresponding to the frequency deviation Δf into the phase accumulator of the NCO in the form of a frequency control word, and dynamically adjust the center frequency during the execution of the pulse sequence. The transfer function calibration module is used during the system installation and commissioning phase to place a standard NMR probe or water phantom at the isomagnetic center of the magnet, simultaneously acquire the output signal of the external probe and the reference signal at the isomagnetic center, train the transfer function coefficient matrix using the LMS algorithm, model the frequency deviation between the external probe and the isomagnetic center of the magnet, and solve for the optimal transfer function coefficients. During the calibration process, the transfer function calibration module uses the external probe data as the input vector x(n) = [Bx(n), By(n), Bz(n), 1]. The desired response is defined as the frequency deviation at the isomagnetic center, Δf_target(n) = f_ref(n) - f0, where f_ref(n) is the frequency reading of the NMR probe at the isomagnetic center, and f0 is the system reference frequency. The weighting coefficient vector w(n) = [Gx, Gy, Gz, offset] is iteratively updated using the LMS algorithm. ;w(n+1) = w(n)+ μ·e(n)·x(n), where e(n) = Δf_target(n) - w(n) ·x(n) represents the estimation error, and μ is the step size factor. When the mean square value of the estimation error drops below the preset threshold, or the number of iterations reaches the preset upper limit, the iteration stops, and the weight coefficient vector at this time is the optimal transfer function coefficient.

2. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The analog-to-digital converter is a 24-bit sigma-delta type analog-to-digital converter with a sampling rate of 10 kS / s to 50 kS / s.

3. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The three-axis fluxgate magnetometer is fixedly installed near the magnet, within the linear working region outside the 5 Gauss line of the main magnetic field.

4. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The cutoff frequency of the digital low-pass filter is set to 100 Hz to 200 Hz.

5. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The spectrometer execution unit is connected to the NCO via a high-speed digital interface, supporting dynamic modification of the frequency control word during pulse sequence operation.

6. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The real-time processing unit is implemented based on Xilinx Zynq-7000 SoC or Kintex series field-programmable gate array, and utilizes its parallel processing capability to achieve microsecond-level low-latency processing.

7. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The pulse sequence is a spin echo (SE) sequence, including a 90° radio frequency excitation pulse, a 180° refocusing pulse, and a signal acquisition stage; the spectrometer execution unit adjusts the NCO frequency before the application of the 90° radio frequency excitation pulse and before the application of the 180° refocusing pulse, respectively.

8. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The system operates fully automatically. After the transfer function calibration is completed during the system calibration phase, all monitoring and compensation operations require no manual intervention.

9. The ultra-low field magnetic resonance center frequency adaptive compensation system according to claim 1, characterized in that, The real-time processing delay of the frequency deviation Δf is on the order of microseconds, which meets the timing requirements of ultra-low field magnetic resonance imaging sequences.

10. An adaptive compensation method for the center frequency of ultra-low field magnetic resonance, applied to the adaptive compensation system for the center frequency of ultra-low field magnetic resonance as described in any one of claims 1 to 9, characterized in that, Includes the following steps: Calibration steps: During the system installation and commissioning phase, place a standard nuclear magnetic resonance probe or water phantom at the isomagnetic center of the magnet, and simultaneously acquire the output signal of the external triaxial fluxgate magnetometer and the reference signal at the isomagnetic center. Train the transfer function coefficient matrix between the external magnetic field and the internal frequency offset based on the minimum mean square error (LMS) algorithm. The calibration steps include: Preparation: Place the standard NMR probe or uniform water phantom at the magnetic center of the magnet, ensuring that the reference probe is in the optimal measurement position, and record the system reference frequency f0 at this time; Synchronous data acquisition: Under different environmental magnetic field conditions, the three outputs (Bx, By, Bz) of the external three-axis fluxgate magnetometer and the frequency reading f_ref of the NMR probe at the isomagnetic center are acquired simultaneously; LMS algorithm iteration: Using external probe data as input vector x(n) = [Bx(n), By(n), Bz(n), 1] The desired response is determined by the frequency deviation at the isomagnetic center, Δf_target(n) = f_ref(n) - f0. The weight coefficient vector w(n) = [Gx, Gy, Gz, offset] is iteratively updated using the LMS algorithm. ;w(n+1) = w(n) + μ·e(n)·x(n), where e(n) = Δf_target(n) - w(n) ·x(n) represents the estimation error, and μ is the step size factor; Convergence criterion: When the mean square value of the estimation error drops below the preset threshold, or when the number of iterations reaches the preset upper limit, the iteration stops, and the weight coefficient vector at this time is the optimal transfer function coefficient; Verification: A portion of the collected data is retained for verification. The verification data is then substituted into the calibrated transfer function model to check whether the deviation between the model's predicted value and the expected value meets the accuracy requirements. If not, the calibration process is repeated. Real-time monitoring steps: During magnetic resonance imaging, the triaxial fluxgate magnetometer collects environmental magnetic field vector data near the magnet in real time, and sends it to the field programmable gate array (FPGA) for processing after analog-to-digital conversion; Filtering steps: Perform digital low-pass filtering on the environmental magnetic field vector data, with a cutoff frequency of 100 Hz to 200 Hz, to filter out high-frequency noise; Frequency offset calculation steps: Using the transfer function model established in the calibration steps, the filtered environmental magnetic field data is mapped to the real-time frequency deviation Δf at the magnetic center of the magnet; Compensation execution steps: The frequency deviation Δf is converted into the corresponding NCO frequency control word and written to the spectrometer digital oscillator through a high-speed digital interface to complete the center frequency update before the radio frequency excitation of the next pulse sequence; Phase compensation steps: Calculate the cumulative phase error φ(t) = 2π∫Δf(t)dt in real time, and send the corresponding phase correction value to the digital downconversion module of the spectrometer to compensate for the phase error in the acquired data.

11. The adaptive compensation method for the center frequency of ultra-low field magnetic resonance according to claim 10, characterized in that, In the spin echo (SE) sequence, the compensation execution step completes the first frequency adjustment before the application of the 90° RF pulse and the second frequency adjustment before the application of the 180° RF pulse.

12. The adaptive compensation method for the center frequency of ultra-low field magnetic resonance according to claim 10, characterized in that, The compensation execution step performs center frequency compensation in the radio frequency transmission (TX) phase and the signal reception (RX) phase of the pulse sequence, respectively.

13. The adaptive compensation method for the center frequency of ultra-low field magnetic resonance according to claim 10, characterized in that, The frequency offset calculation step models the system as a linear time-invariant LTI system, which is implemented using a discrete-domain model in the form of an FIR filter.

14. The adaptive compensation method for the center frequency of ultra-low field magnetic resonance according to claim 10, characterized in that, The total delay for all processing steps, from sampling the ambient magnetic field to updating the NCO frequency, is in the microsecond range.

15. The adaptive compensation method for the center frequency of ultra-low field magnetic resonance according to claim 10, characterized in that, In the filtering process, the digital low-pass filter is a finite-length unit impulse response filter.

16. The adaptive compensation method for the center frequency of ultra-low field magnetic resonance according to any one of claims 10 to 15, characterized in that, All steps are completed in real time during the execution of the magnetic resonance pulse sequence, without occupying an additional independent scan time window.

17. The adaptive compensation method for the center frequency of ultra-low field magnetic resonance according to claim 10, characterized in that, After each compensation is performed, the compensation effect is continuously verified in the next pulse sequence cycle. If the frequency deviation still exceeds the preset range, a new round of compensation is triggered.

Images