Deep Learning Dynamic MRI Reconstruction from Undersampled Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Dynamic MRI imaging is a slow process due to the need for acceleration methods to reconstruct images from incomplete data, necessitating more efficient and rapid decoding and reconstruction of image sequences.

Innovation Solution

A method involving neural networks to process preliminary spatial weighting functions, extracting temporal and spatial basis functions from undersampled imaging data to generate artifact-free image sequences, utilizing training data to adjust network weights until a cost function is satisfied, and applying these functions to undersampled data for reconstruction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional acceleration methods are used to reconstruct images from incomplete data, then image reconstruction is achieved, but the process remains slow and inefficient

Engineering Contradiction:

Improveimage reconstruction speedVSAvoidreconstruction time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent replaces traditional mechanical/mathematical reconstruction algorithms with a deep learning-based system. The neural network model learns temporal and spatial basis functions from training data and uses them to rapidly reconstruct image sequences from undersampled k-space data, substituting conventional iterative reconstruction methods with a data-driven approach that achieves much faster processing speeds

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs preliminary extraction of temporal basis functions and spatial weighting functions during a training phase using complete image sequences and their corresponding undersampled data. These pre-computed basis functions are then stored and applied during the actual reconstruction process, eliminating the need to perform complex calculations in real-time and significantly accelerating image reconstruction

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If deep learning neural networks are used to process spatial weighting functions, then reconstruction accuracy is improved, but network training complexity increases

Engineering Contradiction:

Improveimage reconstruction accuracyVSAvoidnetwork training complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex neural network training process into distinct functional components: extracting temporal basis functions from time-varying data, extracting spatial weighting functions from spatial data, and training the network to map preliminary spatial weighting functions to final artifact-free versions. This segmentation allows each component to be optimized independently and simplifies the overall training process while maintaining high reconstruction accuracy

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The neural network training process incorporates feedback mechanisms where the network weights are adjusted based on the cost function that measures differences between estimated and training final spatial weighting functions. This feedback loop enables the network to iteratively improve its performance and converge on optimal weights that minimize reconstruction artifacts and maximize image quality

Inventive Principle:

Principle #23Feedback

3Productivity

If undersampled imaging data is processed, then imaging speed is increased, but image quality and completeness deteriorate

Engineering Contradiction:

Improveimaging speedVSAvoidimage quality

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent introduces temporal basis functions and spatial weighting functions as intermediary representations that bridge the gap between undersampled k-space data and final high-quality image sequences. These basis functions serve as mediators that capture the essential temporal and spatial patterns, allowing the neural network to reconstruct complete artifact-free images from incomplete data without losing quality

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20250216489A1Systems and methods of deep learning for large-scale dynamic magnetic resonance image reconstruction

Publication Date: 2025.07.03 CEDARS SINAI MEDICAL CENT

AI summary

A method for performing magnetic resonance imaging on a subject comprises obtaining undersampled imaging data, extracting one or more temporal basis functions from the imaging data, extracting one or more preliminary spatial weighting functions from the imaging data, inputting the one or more preliminary spatial weighting functions into a neural network to produce one or more final spatial weighting functions, and multiplying the one or more final spatial weighting functions by the one or more temporal basis functions to generate an image sequence. Each of the temporal basis functions corresponds to at least one time-varying dimension of the subject. Each of the preliminary spatial weighting functions corresponds to a spatially-varying dimension of the subject. Each of the final spatial weighting functions is an artifact-free estimation of the one of the one or more preliminary spatial weighting functions.