MRI Gradient Coil Noise Cancellation with Neural Network Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Magnetic resonance imaging (MRI) scans are associated with long scan times and loud acoustic noise, which can cause patient discomfort and potential hearing damage, and existing noise cancellation methods introduce time-lags or require real-time feedback.
Innovation Solution
A method and apparatus that utilize a neural network to predict noise cancellation signals based on MRI gradient coil currents, generating an acoustic noise cancellation signal in advance using acoustic transfer functions, and apply phase adjustments to optimize noise reduction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If real-time feedback algorithm is used for active noise cancelling, then acoustic noise is reduced, but time-lag is introduced to reach maximum cancelling
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing noise cancelling signals in a lookup table before actual MRI scans. The signal generation is performed in advance based on predicted gradient coil currents, eliminating the need for real-time feedback processing and thus removing the time-lag inherent in conventional real-time feedback algorithms while maintaining effective noise cancellation.
2Object-affected harmful factors
If conventional active noise cancelling is used, then acoustic noise is reduced, but device complexity increases due to real-time feedback system
Solution Approach 1:
The patent extracts the complex real-time feedback processing from the noise cancellation system and replaces it with a pre-computed lookup table approach. By separating the signal generation step (performed in advance) from the signal application step (during scanning), the system removes the need for complex real-time feedback algorithms while maintaining noise cancellation effectiveness, thus reducing device complexity.
3Object-affected harmful factors
If noise cancelling signal is generated in advance using neural network, then noise reduction is optimized, but computational complexity increases during calibration
Solution Approach 1:
The patent applies preliminary action by performing the computationally intensive neural network training and calibration process before actual MRI scans. The neural network is trained offline using calibration data to learn the relationship between gradient coil currents and acoustic noise. Once trained, the network generates pre-computed noise cancelling signals stored in a lookup table, shifting the computational burden to the calibration phase rather than the scanning phase, thus optimizing noise reduction during scans while managing computational complexity.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Reduces acoustic noise effectively and conveniently for patients by generating noise cancellation signals in advance, minimizing patient discomfort and potential hearing damage.
Implementation Method 1
converting, by an acoustic transducer, simultaneously with the image acquisition sequence, the predetermined noise cancelling signal to an acoustic noise cancelling signal at the first position
Implementation Method 2
magnetic fields gradients are generated by three gradient coils provided in the MRI apparatus configured for three orthogonal directions respectively
Data Source
AI summary
Method for magnetic resonance imaging, MRI, comprising obtaining an image acquisition sequence comprising RF pulses and magnetic field gradients configured to encode spatial information in a part of an object; obtaining an acoustic noise cancelling signal corresponding to the obtained image acquisition sequence at a first position; generating the image acquisition sequence; wherein the magnetic fields gradients are generated by gradient coils configured for three orthogonal directions respectively; and converting, by an acoustic transducer, simultaneously with the image acquisition sequence, the predetermined noise cancelling signal to an acoustic noise cancelling signal at the first position. The noise cancelling signal is obtained based on an acoustic transfer function and the magnetic field gradients of the image acquisition sequence, wherein the acoustic transfer function is obtained by training a neural network on a plurality of generic image acquisition sequences followed by transfer learning with calibration image acquisition sequences obtained during a calibration state before the generating of the image acquisition sequence.