Jointly Trained MRI Subsampling and Reconstruction Models

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional MRI subsampling models are arbitrary or manually set, leading to low reconstruction accuracy and the use of a single model for different machine learning techniques, resulting in inefficient sampling and artifacts in reconstructed images.

Innovation Solution

A system and method for generating a subsampling model corresponding to an MRI reconstruction model by jointly training a preliminary subsampling model and a preliminary MRI reconstruction model using full MRI data, with iterative updates to improve accuracy and reliability, and utilizing specific models for each MRI sequence to enhance reconstruction accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If conventional arbitrary or manually set subsampling models are used, then device complexity is reduced, but manufacturing precision (reconstruction accuracy) deteriorates

Engineering Contradiction:

Improvesubsampling model complexityVSAvoidreconstruction accuracy

Core Design Contradiction:

Device complexityVSManufacturing precision

Solution Approach 1:

The system performs preliminary training of the subsampling model using full k-space data before actual MRI scanning. Training samples are pre-processed and stored, allowing the model to learn optimal subsampling patterns in advance. This preliminary action enables the model to achieve high reconstruction accuracy without increasing the complexity of the scanning process itself.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements an iterative feedback mechanism where the subsampling model is trained using ground truth full k-space data, then evaluated on reconstruction accuracy. The model parameters are updated based on the difference between reconstructed images and ground truth, creating a closed-loop feedback system that continuously improves reconstruction accuracy while maintaining simple scanning protocols.

Inventive Principle:

Principle #23Feedback

2Device complexity

If a single subsampling model is used for different machine learning techniques, then device complexity is reduced, but productivity (sampling efficiency) deteriorates

Engineering Contradiction:

Improvemodel management complexityVSAvoidsampling efficiency

Core Design Contradiction:

Device complexityVSProductivity

Solution Approach 1:

The system implements different subsampling models tailored to specific MRI sequences and reconstruction algorithms. Each model is optimized for its specific application context (e.g., T1-weighted imaging, T2-weighted imaging, different acceleration factors), allowing each local instance to achieve optimal sampling efficiency for its particular use case rather than using a one-size-fits-all approach.

Inventive Principle:

Principle #3Local quality

3Ease of operation

If arbitrary subsampling models are used, then ease of operation is improved, but measurement precision (reconstruction accuracy) deteriorates

Engineering Contradiction:

Improvemodel selection simplicityVSAvoidreconstruction accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The system automatically selects and applies the appropriate subsampling model based on the MRI sequence and reconstruction parameters without requiring manual intervention. The model selection and configuration is handled autonomously by the system, maintaining ease of operation while ensuring that the most accurate model is used for each specific application.

Inventive Principle:

Principle #25Self-service

4Manufacturing precision

If jointly trained subsampling and reconstruction models are used, then manufacturing precision (reconstruction accuracy) is improved, but device complexity increases

Engineering Contradiction:

Improvereconstruction accuracyVSAvoidmodel training complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The joint training process is segmented into distinct phases: (1) preparing training samples with full k-space data, (2) training the subsampling model to generate optimal sampling patterns, (3) training the reconstruction model to recover images from subsampled data, and (4) iterative refinement of both models together. This segmentation makes the complex joint training process manageable and implementable in practice.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS11880915B2Systems and methods for magnetic resonance imaging

Publication Date: 2024.01.23 SHANGHAI UNITED IMAGING INTELLIGENCE CO LTD

AI summary

A system for Magnetic Resonance Imaging (MRI) is provided. The system may obtain at least one training sample each of which includes full MRI data. The system may also obtain a preliminary subsampling model and a preliminary MRI reconstruction model. The system may further generate a subsampling model corresponding to an MRI reconstruction model by jointly training the preliminary subsampling model and the preliminary MRI reconstruction model using the at least one training sample. The subsampling model may be the trained preliminary subsampling model, and the MRI reconstruction model may be at least a portion of the trained preliminary MRI reconstruction model.