CN-121996919-A - Magnetic resonance signal denoising method and device based on physical prior gating mechanism

CN121996919ACN 121996919 ACN121996919 ACN 121996919ACN-121996919-A

Abstract

The invention discloses a magnetic resonance signal denoising method and device based on a physical prior gating mechanism, which utilizes the physical principle of magnetic resonance, namely that real signals of biological tissues keep physical consistency (amplitude and phase are basically unchanged) in two acquisitions, and transient EMI noise in the environment has the characteristics of randomness and non-stationarity, and provides an adaptive dead zone gating mechanism based on amplitude energy and phase perception, namely that the identification of transient high-amplitude electromagnetic interference in two magnetic resonance signals acquired at the same k-space position is realized by utilizing the physical consistency among the plurality of acquisitions of the magnetic resonance signals, and the fusion weight of transient EMI interference signal points is dynamically adjusted, so that the nonlinear rejection of the non-stationary transient electromagnetic interference is realized; therefore, the method and the device can accurately identify and remove occasional transient high-amplitude electromagnetic interference, avoid the problem of denoising failure, and remarkably improve the purity of the image.

Inventors

ZHANG XIAOTONG
HUANG YIMAN
Tu Jinglian
ZHANG YI

Assignees

浙江大学

Dates

Publication Date: 20260508
Application Date: 20260126

Claims (10)

1. The magnetic resonance signal denoising method based on a physical prior gating mechanism is characterized by comprising the following steps of: acquiring a first magnetic resonance signal and a second magnetic resonance signal which are acquired repeatedly at least twice for the same k-space position in an unshielded environment, wherein the first magnetic resonance signal and the second magnetic resonance signal both comprise signals in an MRI coil and an EMI coil; constructing a signal denoising network, wherein the signal denoising network comprises a twin residual error network and a self-adaptive dead zone heuristic gating; Respectively inputting the first magnetic resonance signal and the second magnetic resonance signal into a twin residual error network in a signal denoising network to perform initial denoising processing to obtain a first initial denoising signal and a second initial denoising signal; Inputting the first initial denoising signal and the second initial denoising signal into the self-adaptive dead zone heuristic gating to perform local energy difference calculation and phase conflict detection, so as to determine the fusion weight of each signal point in the first initial denoising signal and the second initial denoising signal according to the local energy difference and the phase conflict detection result, wherein when the local energy difference is greater than or equal to an energy threshold value or the phase conflict detection result is that phase conflict exists, the fusion weight of the signal point with high energy in the two initial denoising signals is smaller than the fusion weight of the signal point with low energy; And carrying out fusion processing on the first initial denoising signal and the second initial denoising signal according to the fusion weight of each signal point in the first initial denoising signal and the second initial denoising signal so as to obtain a denoised magnetic resonance signal after the fusion processing.
2. The method of claim 1, wherein the twinning residual network comprises a first depth full convolution residual sub-network and a second depth full convolution residual sub-network, and wherein the first depth full convolution residual sub-network and the second depth full convolution residual sub-network share the same network structure and weights; The method for performing initial denoising processing on the first magnetic resonance signal and the second magnetic resonance signal by respectively inputting the first magnetic resonance signal and the second magnetic resonance signal into a twin residual error network in a signal denoising network comprises the following steps: Acquiring EMI prior signals corresponding to the first magnetic resonance signal and the second magnetic resonance signal respectively; And inputting the first magnetic resonance signal and the EMI prior signal corresponding to the first magnetic resonance signal into a first depth full convolution residual sub-network for initial denoising, and inputting the second magnetic resonance signal and the EMI prior signal corresponding to the second magnetic resonance signal into a second depth full convolution residual sub-network for initial denoising, so as to obtain the first initial denoising signal and the second initial denoising signal respectively.
3. The method of claim 2, wherein the first depth full convolution residual sub-network comprises a head convolution layer, a residual convolution layer, and a tail convolution layer connected in sequence, wherein the residual convolution layer comprises a plurality of residual convolution blocks, and wherein any residual convolution block comprises a first convolution layer, a Relu layer, and a second convolution layer connected in sequence.
4. The method of claim 1, wherein inputting the first initial de-noised signal and the second initial de-noised signal to the adaptive dead band heuristic gating for local energy difference computation and phase collision detection comprises: Performing module taking and smoothing on the first initial denoising signal and the second initial denoising signal to obtain a first envelope energy curve and a second envelope energy curve respectively; calculating local energy differences between signal points at the same positions in the first initial denoising signal and the second initial denoising signal based on the first envelope energy curve and the second envelope energy curve; Calculating complex dot products of signal points at the same positions in the first initial denoising signal and the second initial denoising signal, and obtaining phase conflict detection results of the signal points at the same positions in the first initial denoising signal and the second initial denoising signal according to the complex dot products of the signal points at the same positions.
5. The method of claim 4, wherein calculating a local energy difference between signal points at each same location in the first initial de-noised signal and the second initial de-noised signal based on the first envelope energy curve and the second envelope energy curve comprises: Calculating an energy difference between an ith signal point in the second envelope energy curve and an ith signal point in the first envelope energy curve, and summing energy sums of the ith signal point in the second envelope energy curve and the ith signal point in the first envelope energy curve, wherein i is a positive integer; Taking the ratio of the energy difference value and the energy sum as the local energy difference between the ith signal point in the first initial denoising signal and the second initial denoising signal; Adding 1 to i, and recalculating an energy difference between an ith signal point in the second envelope energy curve and an ith signal point in the first envelope energy curve until i is equal to n, so as to obtain local energy differences between the first initial denoising signal and signal points at the same positions in the second initial denoising signal, wherein the initial value of i is 1, and n is the signal length of the first initial denoising signal; Correspondingly, according to the complex dot product of the signal points at the same positions, obtaining the phase conflict detection result of the signal points at the same positions in the first initial denoising signal and the second initial denoising signal, wherein the phase conflict detection result comprises: for two signal points at any same position in the first initial denoising signal and the second initial denoising signal, judging whether the complex dot product of the two signal points is negative; if yes, judging that phase conflict exists between the two signal points at any same position.
6. The method of claim 4, wherein determining the fusion weights for each signal point in the first initial de-noised signal and the second initial de-noised signal based on the local energy difference and the phase conflict detection result comprises: For two signal points at any same position in the first initial denoising signal and the second initial denoising signal, judging whether the local energy difference between the two signal points is greater than or equal to an energy threshold value or whether a phase conflict detection result is that phase conflict exists; If yes, calculating a first initial weight according to the local energy difference between the two signal points, and taking the difference between the 1 and the first initial weight as a second initial weight; And taking the largest weight of the first initial weight and the second initial weight as the fusion weight of the signal point with the lowest energy value of the two signal points, taking the smallest weight of the first initial weight and the second initial weight as the fusion weight of the signal point with the highest energy value of the two signal points, and obtaining the fusion weights of all the signal points of the first initial denoising signal and the second initial denoising signal after all the signal points of the first initial denoising signal and the second initial denoising signal are polled.
7. The method of claim 6, wherein if the local energy difference between the two signal points is less than the energy threshold and the phase conflict detection result is that there is no phase conflict, the method further comprises: setting the same fusion weight for the two signal points; Wherein, according to the local energy difference between two signal points, calculate the first initial weight, include: and calculating the first initial weight based on the local energy difference between the two signal points by adopting a Sigmoid function.
8. The method of claim 1, wherein the signal denoising network is trained with magnetic resonance training data as input, denoising magnetic resonance signal output, wherein one piece of magnetic resonance training data comprises a first sample magnetic resonance signal and a second sample magnetic resonance signal, and wherein a loss function of the signal denoising network is: ; in the formula, As a function of the loss, A first sample initial denoising signal output after the first sample magnetic resonance signal is input to a twin residual network in a signal denoising network, A second sample initial denoising signal output after the second sample magnetic resonance signal is input to a twin residual error network in a signal denoising network, Is the denoising magnetic resonance signal corresponding to the magnetic resonance training data, Is the label data corresponding to the magnetic resonance training data, Is the L1 norm.
9. A magnetic resonance signal denoising device based on a physical prior gating mechanism, comprising: the acquisition unit is used for acquiring a first magnetic resonance signal and a second magnetic resonance signal which are acquired by repeatedly acquiring the same k-space position at least twice under an unshielded environment, wherein the first magnetic resonance signal and the second magnetic resonance signal both comprise signals in an MRI coil and an EMI coil; The network construction unit is used for constructing a signal denoising network, wherein the signal denoising network comprises a twin residual error network and a self-adaptive dead zone heuristic gating; the denoising unit is used for inputting the first magnetic resonance signal and the second magnetic resonance signal into a twin residual error network in the signal denoising network respectively to perform initial denoising processing to obtain a first initial denoising signal and a second initial denoising signal; The denoising unit is used for inputting the first initial denoising signal and the second initial denoising signal into the self-adaptive dead zone heuristic gating to perform local energy difference calculation and phase conflict detection so as to determine the fusion weight of each signal point in the first initial denoising signal and the second initial denoising signal according to the local energy difference and the phase conflict detection result, wherein when the local energy difference is greater than or equal to an energy threshold value or the phase conflict detection result is that the phase conflict exists, the fusion weight of the signal point with high energy in the two initial denoising signals is smaller than the fusion weight of the signal point with low energy; The denoising unit is further used for carrying out fusion processing on the first initial denoising signal and the second initial denoising signal according to the fusion weight of each signal point in the first initial denoising signal and the second initial denoising signal so as to obtain a denoised magnetic resonance signal after the fusion processing.
10. An electronic device comprising a memory, a processor and a transceiver in communication with each other in sequence, wherein the memory is configured to store a computer program, the transceiver is configured to transmit and receive a message, and the processor is configured to read the computer program and perform the magnetic resonance signal denoising method based on a physical prior gating mechanism according to any one of claims 1 to 8.

Description

Magnetic resonance signal denoising method and device based on physical prior gating mechanism Technical Field The invention belongs to the technical field of magnetic resonance imaging, and particularly relates to a magnetic resonance signal denoising method and device based on a physical priori gating mechanism. Background While conventional magnetic resonance requires placement in a shielded room to isolate electromagnetic interference noise in the environment, one purpose of portable (low-field) magnetic resonance imaging instruments is to achieve device portability, which requires no shielded room, magnetic resonance imaging in an open environment suffers from electromagnetic interference (electromagnetic interference), which results in severe electromagnetic interference noise (hereinafter referred to as EMI noise) in the resulting image, which severely affects image quality. The EMI noise is additive noise, that is, data affected by the EMI noise is composed of signals and the EMI noise, wherein the existing method for removing the EMI noise generally needs to be externally connected with a plurality of electromagnetic interference induction coils (EMI coils) so as to collect magnetic resonance signals (MRI coils) and receive the EMI noise in the environment at the same time, thereby removing the EMI noise in the imaging coils as a priori knowledge, and simultaneously, in order to fit the mapping relation of the EMI noise between the MRI coils and the EMI coils, a section of signal is needed to be used as a calibration signal for fitting the mapping relation (when no MRI signal and only the environment EMI noise exist). At present, deep-DSP mode is often used for EMI denoising (e.g. in the prior art with application number CN202310504030.5 and reference ：Robust EMI elimination for RF shielding-free MRI through deep learning direct MR signal prediction）;, wherein this mode is a Deep learning method, which maps an MRI signal with EMI noise and a signal of an EMI coil into a clean noise-free MRI signal, and its training input (input) is from calibration data and an additional MRI signal without EMI noise, and is constructed by superimposing an MRI signal without EMI noise and with a high signal-to-noise ratio into environmental EMI noise collected by the MRI coil in the calibration data, which is to simulate the MRI signal+emi noise in practical situations, and other training inputs are from the calibration signal of the EMI coil, and at the same time, the training tag (label) is the MRI signal with the aforementioned noise-free and high signal-to-noise ratio. Specifically, the training process is shown in fig. 1 (C), the training of the network essentially completes the mapping of all channels of data (including MRI and EMI coils) to a clean, EMI noise-free MRI signal, wherein one sample of the network training is a line acquired by the MRI and EMI coils, that is, a one-dimensional network, and the testing process is to input all data of a k-space line and an EMI coil acquired simultaneously into the network, so as to obtain a magnetic resonance k-space line from which EMI noise is removed, as shown in fig. 1 (D). However, the prior art has the following defects that (1) the method has good effect of inhibiting stable broadband background noise, but has poor effect of inhibiting transient strong EMI noise (sporadic and high amplitude, such as spike noise, sweep noise and the like), and can not effectively inhibit the type of noise, as shown in fig. 2, the image (a) in fig. 2 is a k-space (signal received by an MRI coil) containing EMI noise, a very short period of EMI noise besides broadband EMI noise can be seen, the image (b) in fig. 2 is a k-space after Deep-DSP denoising, the strong EMI noise can be seen to remain, the image (c) in fig. 2 is a reconstructed image, and the small EMI noise can be seen to generate serious interference on the image, so that the imaging quality is seriously influenced, and (2) in the magnetic resonance conventional scanning, the signal to noise ratio (especially the signal to noise ratio is low due to the fact that the signal to the noise ratio is low in a portable low-field) is generally utilized by repeatedly acquiring (NEX > 1) for many times, so that the noise to be more than the noise to be eliminated due to the fact that the noise to the noise is more than 1, however, the noise to be removed by the method can not have the normal noise due to the fact that the noise is high-noise is removed due to the fact that the amplitude is high, and the noise is not normally lost due to the noise. Therefore, based on the foregoing shortcomings, how to provide a magnetic resonance signal denoising method based on a physical prior gating mechanism, which can remove transient strong EMI noise, so as to improve the magnetic resonance imaging quality, has become a problem to be solved. Disclosure of Invention The invention aims to provide a magnetic resonance signal denoising metho