MRI Deep Learning Reconstruction With Tunable Data Consistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for medical imaging technologies have not effectively addressed the limitations of effectively addressing the issues of image resolution, noise distribution, and acquisition speed in medical imaging, particularly in magnetic resonance imaging (MRI), leading to suboptimal image quality and inefficiencies in deep learning reconstruction processes.
Innovation Solution
A tunable gradient scaling factor is introduced to adjust denoising behavior in medical imaging systems, allowing for variable denoising levels by modifying the gradient step size in unrolled iterative reconstruction networks, which are trained with end-to-end machine learning to account for noise levels and sampling patterns, enabling better generalization across different imaging conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If parallel imaging with high acceleration factors is used to increase acquisition speed, then acquisition speed is improved, but image noise increases and becomes inhomogeneous
Solution Approach 1:
The patent applies parameter changes by introducing a tunable denoising parameter that controls the strength of denoising applied during reconstruction. This parameter can be adjusted based on the acceleration factor used, allowing the system to optimize the balance between acquisition speed and noise levels. The denoising parameter is integrated into the reconstruction algorithm to modify how noise is handled at different acceleration levels.
2Manufacturing precision
If deep learning reconstruction is used to improve image resolution and SNR, then image quality is improved, but the desired denoising level varies by application and radiologist preference
Solution Approach 1:
The patent implements dynamics by making the denoising parameter可调 (adjustable) rather than fixed. The system allows dynamic adjustment of the denoising strength based on different applications, radiologist preferences, and imaging conditions. This enables the same deep learning reconstruction model to adapt to varying requirements without requiring multiple separate models.
Solution Approach 2:
The patent introduces a tunable parameter that controls the denoising behavior of the deep learning reconstruction network. By changing this parameter, users can adjust the trade-off between noise reduction and image detail preservation according to specific application needs and radiologist preferences, making the system versatile across different clinical scenarios.
3Adaptability or versatility
If post-processing denoising is added to achieve variable denoising levels, then denoising flexibility is improved, but reconstruction time increases and additional training is required
Solution Approach 1:
The patent merges the denoising function directly into the deep learning reconstruction network by integrating a denoising parameter into the reconstruction algorithm itself. This eliminates the need for separate post-processing denoising steps, as the denoising is performed concurrently with reconstruction. The unified approach maintains reconstruction speed while providing denoising flexibility through parameter adjustment.
Data Source
AI summary
Systems and methods for reconstruction for a medical imaging system. A scaling factor is used during the reconstruction process to adjust a step size of a gradient update. The adjustment of the step size of the gradient provides the ability to adjust a level of denoising by the reconstruction process.