MRI Magnet Shimming Using Machine Learning for Field Homogeneity

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for correcting magnetic field inhomogeneities in MRI magnets are complex, time-consuming, and user-dependent, requiring specialized skills and manual adjustments, with no objective protocol for ensuring repeatable outcomes.

Innovation Solution

A machine learning algorithm is used to predict the distribution of correction elements on a grid based on measured magnetic field data, trained on a database of known cases, to simplify and accelerate the shimming process while maintaining precision.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If analytical methods are used to correct magnetic field inhomogeneities, then correction precision can be achieved, but the process becomes extremely complex and time-consuming

Engineering Contradiction:
Improvemagnetic field homogeneity correction precisionVSAvoidshimming process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces complex analytical mathematical methods with a machine learning model that has been pre-trained on magnetic field correction data. The model automatically predicts correction element distributions without requiring manual analytical calculations, thereby reducing process complexity while maintaining correction precision.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The machine learning model is pre-trained offline on a database of magnetic field measurements and corresponding optimal correction solutions. This preliminary training phase allows the model to quickly provide accurate correction predictions during actual shimming operations without requiring complex real-time analytical computations.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If traditional shimming methods are used, then magnetic field inhomogeneities can be corrected, but the process requires highly specialized personnel and manual adjustments

Engineering Contradiction:
Improvemagnetic field homogeneityVSAvoidoperational simplicity
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The machine learning model performs the correction prediction automatically based on input magnetic field measurements. The system self-determines the optimal distribution of correction elements without requiring specialized human operators to perform manual calculations or adjustments, making the process easier to operate while maintaining precision.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces manual expert judgment and manual adjustment procedures with an automated machine learning system. The model processes measurements and generates correction predictions automatically, eliminating the need for highly specialized personnel to perform manual shimming operations.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Measurement precision

If repeated corrections are performed to achieve optimal magnetic field homogeneity, then field uniformity improves, but the time required increases significantly

Engineering Contradiction:
Improvemagnetic field uniformityVSAvoidshimming process time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The machine learning model is pre-trained on a comprehensive database of magnetic field configurations and optimal corrections. This preliminary training allows the model to provide accurate correction predictions in a single pass without requiring multiple iterative corrections, significantly reducing the time required while maintaining field uniformity.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system uses measured magnetic field data as input to the machine learning model, which then predicts the optimal correction element distribution. This feedback-based approach allows the system to determine the correct correction in one step rather than requiring repeated trial-and-error adjustments, reducing time while achieving the desired field uniformity.

Inventive Principle:
Principle #23Feedback

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

The method allows for efficient and precise reduction of magnetic field inhomogeneities in a fraction of the time required by traditional methods, without the need for analytical equations, and provides repeatable outcomes.

Implementation Method 1

measuring the magnetic field and sampling it in a plurality of locations

Methodology Applied
Scientific EffectMagnetic field measurement: Magnetic Field

Implementation Method 2

calculating the position and magnitude parameters of one or more correction elements to obtain predetermined target values of the field characteristics

Methodology Applied
Scientific EffectMagnetic field correction: Magnetic Field

Data Source

PatentEP3985409B1Method for correcting inhomogeneity of the static magnetic field particularly of the static magnetic field generated by the magnetic structure of a machine for acquiring nuclear magnetic resonance images and MRI system for carrying out such method
Publication Date: 2025.08.20 ESAOTE
  • EP3985409B1 patent drawingFigure 1A~1B
  • EP3985409B1 patent drawingFigure 2
  • EP3985409B1 patent drawingFigure 3~4

AI summary

A method for shimming a magnetic field of an MRI magnet structure is provided comprising the following steps: a) measuring the magnetic field and sampling it in a plurality of locations, with a predetermined space distribution within a predetermined volume of space permeated by the said magnetic field and delimited by a surface boundary and/or also on the said surface boundary; b) defining a grid for positioning the correction elements, depending on the magnet structure and on the correlation thereof with the field structure; c) calculating the position and magnitude parameters of one or more correction elements to obtain predetermined target values of the field characteristics, in which the step c) is carried out by a machine learning algorithm or combinations thereof, which algorithm has been trained by a database of known cases in which each record links a certain initial magnetic field of a magnetic structure to the pattern of correction elements on a positioning grid for these correction elements in the magnet structure, thereby delivering as an output a pattern of correction elements on the positioning grid which contributions to the magnetic field of the magnet structure generate a magnetic field which at least best approximates or meets the said predetermined target values of the field characteristics.