Deep Learning MRI Reconstruction for Multi-Slice Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional MRI acceleration techniques, such as compressed sensing and parallel imaging, are unsuitable for simultaneous multi-slice data collection due to requirements for good coil configurations and iterative calculations, making them inefficient for clinical practices.

Innovation Solution

An artificial neural network (ANN) is trained to process under-sampled MRI data from simultaneous multi-slice datasets, using sub-networks with shared structures and parameters to disentangle and reconstruct MRI images, with a data consistency component for k-space data estimation and inverse Fourier transform application.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If conventional MRI acceleration techniques (compressed sensing, parallel imaging) are used, then imaging speed is improved, but device complexity and requirement for good coil configurations increase

Engineering Contradiction:

Improveimaging speedVSAvoidcoil configuration requirements

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The patent replaces conventional iterative reconstruction methods with a deep learning-based neural network that directly maps undersampled k-space data to reconstructed images. This substitution eliminates the need for complex iterative calculations and good coil configurations, achieving fast imaging speed through trained network parameters rather than mechanical system complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Speed

If conventional MRI acceleration techniques are used, then imaging speed is improved, but iterative calculations are required making them unsuitable for multi-slice data collection

Engineering Contradiction:

Improveimaging speedVSAvoidsuitability for multi-slice data collection

Core Design Contradiction:

SpeedVSProductivity

Solution Approach 1:

The patent substitutes iterative reconstruction algorithms with a neural network that performs direct transformation of undersampled multi-slice k-space data into reconstructed images. This replacement eliminates iterative calculations, enabling the system to handle multi-slice data collection efficiently and produce results suitable for clinical practice.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Loss of time

If under-sampling is applied to accelerate data collection, then scanning time is reduced, but image quality deteriorates due to artifacts

Engineering Contradiction:

Improvescanning timeVSAvoidimage quality

Core Design Contradiction:

Loss of timeVSManufacturing precision

Solution Approach 1:

The patent employs a neural network trained with feedback from fully-sampled ground truth images to reconstruct undersampled data. The network learns to predict and remove artifacts by comparing reconstructed images with ground truth during training, enabling high-quality image reconstruction from undersampled data without the typical quality deterioration.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent uses a neural network that copies and adapts patterns from fully-sampled ground truth images during training to reconstruct undersampled images. By learning the mapping from undersampled to fully-sampled characteristics, the network can generate high-quality images that replicate the quality of fully-sampled images despite the reduced sampling density.

Inventive Principle:

Principle #26Copying

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 ANN effectively reconstructs MRI images from under-sampled data, removing artifacts and achieving quality similar to fully-sampled images, suitable for clinical use cases involving multi-slice data collection.

Implementation Method 1

an artificial neural network (ANN) may be trained and used to obtain (e.g., receive) the SMS dataset and generate a first reconstructed MRI image corresponding to the first MRI slice and a second reconstructed MRI image corresponding to the second MRI slice

Methodology Applied

Scientific EffectDeep learning:

Implementation Method 2

The first reconstructed MRI image and the second reconstructed MRI image may then be generated by applying an inverse Fourier transform (e.g., a 3D fast Fourier transform (FFT)) to the estimated k-space data

Methodology Applied

Scientific EffectFourier transform:

Data Source

PatentUS11941732B2Multi-slice MRI data processing using deep learning techniques

Publication Date: 2024.03.26 SHANGHAI UNITED IMAGING INTELLIGENCE CO LTD

AI summary

Disclosed herein are systems, methods, and instrumentalities associated with reconstructing magnetic resonance (MR) images based on multi-slice, under-sampled MRI data (e.g., k-space data). The multi-slice MRI data may be acquired using a simultaneous multi-slice (SMS) technique and MRI information associated with multiple MRI slices may be entangled in the multi-slice MRI data. A neural network may be trained and used to disentangle the MRI information and reconstruct MRI images for the different slices. A data consistency component may be used to estimate k-space data based on estimates made by the neural network, from which respective MRI images associated with multiple MRI slices may be obtained by applying a Fourier transform to the k-space data.