MRI Intensity Inhomogeneity Correction Using Reference-Guided CNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing magnetic resonance imaging (MRI) technologies suffer from instrumentation-induced intensity inhomogeneities due to varying sensitivities of receive antennas, which hinder accurate diagnosis.
Innovation Solution
A computer-implemented method using machine learning, specifically a convolutional neural network (CNN), to generate an intensity-corrected MRI representation by applying a trained function to reference and image measurement data, generating correction data to compensate for instrumentation-induced inhomogeneities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional intensity correction methods are used, then processing speed is improved, but correction accuracy deteriorates due to inability to distinguish instrumentation-induced inhomogeneities from anatomically induced ones
Solution Approach 1:
The patent segments the intensity inhomogeneity problem into two distinct components: instrumentation-induced inhomogeneities and anatomically induced inhomogeneities. The machine learning model is trained to separately identify and correct only the instrumentation-induced component, while preserving the anatomically induced variations. This segmentation enables accurate differentiation between correctable artifacts and meaningful anatomical information.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the raw MRI data and the corrected output. This intermediary learns the complex mapping between input images and correction factors, enabling accurate distinction between different sources of inhomogeneity without requiring explicit mathematical modeling of each source.
2Measurement precision
If machine learning methods are applied to distinguish and correct only instrumentation-induced inhomogeneities, then correction accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary training of the machine learning model using a large dataset of MRI images with known ground truth corrections. This pre-training phase captures the complex patterns of instrumentation-induced inhomogeneities, enabling the model to make accurate corrections during actual clinical use without requiring complex real-time calculations.
Solution Approach 2:
The patent uses synthetic training data generated by simulating various MRI acquisition scenarios and applying known correction factors. These synthetic copies of real-world scenarios allow the model to learn from diverse examples without requiring extensive collection and manual annotation of clinical data, reducing both time and resource requirements.
3Reliability
If conventional correction algorithms are used, then processing speed is maintained, but diagnostic accuracy deteriorates due to over-correction of anatomically significant features
Solution Approach 1:
The patent applies local quality by allowing different regions of the image to receive different correction factors based on their specific characteristics. The machine learning model analyzes local patterns in the image data and applies tailored corrections to each region, preserving anatomically significant intensity variations while correcting localized instrumentation artifacts.
Solution Approach 2:
The patent dynamically adjusts correction parameters based on the specific characteristics of each image and region. Rather than applying a uniform correction factor, the system modifies local intensity parameters selectively, changing only those parameters that correspond to instrumentation-induced variations while maintaining parameters that reflect true anatomical properties.
Data Source
AI summary
A method for generating am intensity-corrected magnetic resonance representation. In this method, reference measurement data is provided, wherein at least part of the reference measurement data is generated from magnetic resonance signals acquired by at least one receive antenna of at least one local coil by a magnetic resonance apparatus in accordance with a reference magnetic resonance sequence. In addition, image measurement data is provided, which is generated from magnetic resonance signals acquired by the at least one receive antenna of the least one local coil by the magnetic resonance apparatus in accordance with an image magnetic resonance sequence. A function trained by a machine learning algorithm is applied to the reference measurement data and the image measurement data as input data of the trained function. Correction data is provided from output data of the trained function, wherein the correction data describes an instrumentation-induced intensity inhomogeneity of the image measurement data. An intensity-corrected magnetic resonance representation is generated on the basis of the image measurement data and the correction data.