Artifact And/or Noise Reduction for Magnetic Resonance Fingerprinting Data Based on a Neural Network

Simple SummaryContent extracted from patent full text and abstract with AI.

This patent presents a method and system for improving the quality of Magnetic Resonance Fingerprinting (MRF) data using a neural network. The approach involves acquiring MR data with MRF protocols, processing this data with an AI model to reduce noise and artifacts, and then using the cleaned data to create detailed quantitative MR parameter maps (such as T1 and T2 relaxation times). The neural network is trained to automatically remove undersampling/aliasing artifacts and noise from the MR data, resulting in more accurate and faster MR imaging.

Use CasesContent extracted from patent full text and abstract with AI.

• Hospital and clinical MRI scans for faster, artifact-reduced quantitative imaging.
• Research imaging where precise parameter mapping is essential (e.g., neuroscience, oncology).
• Pediatric or emergency settings where quick MRI scans are required.
• Optimizing or upgrading existing MRI hardware/software to enable AI-powered image reconstruction for better diagnostic outcomes.
• Remote and telemedicine MRI analysis, providing improved data quality to off-site specialists.

BenefitsContent extracted from patent full text and abstract with AI.

• Significant reduction of noise and sampling artifacts in MRF data, leading to clearer and more accurate images.
• Faster image acquisition and processing, reducing scan times and patient discomfort.
• Potential for automated, generalizable, and robust quantitative MRI without heavy manual parameter tuning.
• Scalable approach that can adapt to various MRI protocols, scanner types, and clinical settings.
• Improved diagnostic confidence due to higher quality quantitative maps, supporting better clinical decisions.
• Potential decrease in computational and memory demands versus traditional reconstruction methods.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

Methods and devices for fingerprinting magnetic resonance imaging are described. The method comprises acquiring a first sequence of MR data (2000) within a region of interest using a fingerprinting magnetic resonance pulse sequence; inputting the first sequence of MR data to a neural network (3000); outputting a second sequence of MR data from the neural network, wherein the second sequence of MR data have reduced undersampling/aliasing artifacts and/or noise if compared to the first sequence of MR data (4000); determining values of at least one quantitative parameter for the region of interest based on the second sequence of MR data (4000); and constructing a quantitative parameter map of the at least one quantitative parameter for the region of interest based on the determined values.