Variational Network for Low-Latency Cine MRI Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for reconstructing cine MRI images face challenges in achieving low-latency, high-quality image reconstruction, particularly for interactive real-time cardiac imaging, where high spatial and temporal resolution requirements are coupled with stringent latency constraints.
Innovation Solution
A variational reconstruction network with a cascade of serially arranged cascade-modules is employed, along with a selection unit to choose a single frame for output, enabling efficient low-latency reconstruction of cine MRI images. This network is trained using a deep learning-based method that optimizes image quality and reduces reconstruction latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If high acceleration rates are used to achieve desired spatial and temporal resolution, then image quality is improved, but reconstruction time increases
Solution Approach 1:
The reconstruction process is divided into multiple cascade modules, each performing a partial reconstruction step. This segmentation allows the system to process complex high-resolution reconstruction tasks in smaller, more manageable steps, reducing overall reconstruction time while maintaining image quality.
Solution Approach 2:
The network uses a stack of N frames as input, performing preliminary processing on multiple frames simultaneously. This preliminary action on stacked frames enables the model to predict the current frame based on temporal context, significantly accelerating reconstruction compared to processing frames individually.
2Manufacturing precision
If iterative gradient descent-based reconstruction techniques are used, then image quality is improved, but reconstruction latency increases
Solution Approach 1:
The invention creates a neural network model that learns from training data and produces reconstructed frames without requiring iterative gradient descent during inference. The trained model directly predicts reconstructed frames from input stacks, eliminating the time-consuming iterative optimization step while maintaining image quality.
Solution Approach 2:
The patent replaces the mechanical iterative gradient descent optimization process with a learned neural network model. This substitution transforms the reconstruction process from a computational optimization problem into a direct prediction task, dramatically reducing latency while preserving reconstruction quality.
3Manufacturing precision
If more cascade-modules are added to improve reconstruction quality, then image quality is improved, but device complexity and processing time increase
Solution Approach 1:
The cascade modules are designed with dynamic characteristics where each module processes the output of the previous one in a sequential manner. This dynamic architecture allows the system to achieve high reconstruction quality through multiple processing stages without requiring all modules to be fully complex, as earlier modules provide foundational processing that reduces the burden on subsequent modules.
Data Source
AI summary
The reconstruction network is a variational network configured to reconstruct images from cine MRI data. The reconstruction network comprises an architecture with a cascade of cascade modules. The input of the first cascade module is an input-stack of a plurality of N frames and the input of each following cascade module is the input-stack and an output-stack of the preceding cascade module. The reconstruction network further includes a selection unit configured to select a single frame being processed by the cascade modules that corresponds to the i-th frame of the input stack as the basis for the output dataset.