Deep Learning Reconstruction of Diffusion Metrics from Sparse MRI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional imaging protocols for generating reliable maps of neurite orientation dispersion and density imaging (NODDI) or other advanced diffusion models require extensive sampling in q-space, making them time-consuming and clinically impractical, while deep learning has shown promise in sparse acquisition schemes but faces challenges in jointly modeling k-q space redundancy.
Innovation Solution
A method using a trained neural network to generate diffusion metric maps from magnetic resonance data by optimizing k-space and q-space sampling, allowing for simultaneous estimation of multiple diffusion metrics from a reduced dataset, leveraging the redundancy in k-q joint space through high b-value and high resolution integrated diffusion (HIBRID) acquisition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive sampling in q-space is used to generate reliable diffusion metric maps, then measurement precision is improved, but acquisition time increases significantly
Solution Approach 1:
The patent applies partial action by acquiring only a subset of k-q space data points rather than exhaustive sampling. Specifically, it acquires data at selected k-space locations (e.g., center and outer regions) and selected q-values, then uses deep learning to reconstruct the complete diffusion metric maps from this reduced dataset, achieving clinical feasibility while maintaining accuracy
Solution Approach 2:
The patent introduces a deep learning model as an intermediary between the reduced k-q space sampling data and the final diffusion metric maps. This intermediary network learns the mapping from sparse measurements to complete diffusion metrics, enabling accurate reconstruction without requiring extensive direct sampling
2Measurement precision
If high q-space sampling is used to improve diffusion model accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the q-space sampling into discrete shells (e.g., b=0, b=1000, b=2000, b=3000 s/mm²), with each shell containing a specific number of diffusion directions. This segmentation allows systematic optimization of sampling density at different q-values while managing overall protocol complexity
Solution Approach 2:
The patent optimizes specific acquisition parameters including the number of diffusion directions per shell (e.g., 6, 18, 30, or 42 directions), the b-value shells selected, and the k-space sampling patterns. These parameter changes enable tailored sampling strategies that balance accuracy requirements with protocol simplicity
3Productivity
If reduced k-q space sampling is used to decrease acquisition time, then productivity is improved, but loss of information increases
Solution Approach 1:
The patent performs preliminary action by pre-training the deep learning model on extensively sampled diffusion data to learn the relationships between k-q space patterns and diffusion metrics. This preliminary training enables the model to accurately predict diffusion metrics from reduced sampling during actual clinical acquisition, preventing information loss
Solution Approach 2:
The patent uses copying by training the deep learning model to copy the information content of fully sampled diffusion metric maps from the training dataset. The model learns to generate accurate diffusion metric maps that replicate the information quality of exhaustive sampling even when given only reduced input data
Data Source
AI summary
Diffusion metric maps are generated from a limited input of magnetic resonance data to a suitably trained machine learning algorithm, such as a suitably trained neural network. In general, a downsampling strategy is implemented in the joint k-q space to enable the simultaneous estimation of multiple different diffusion metrics from a more limited set of input diffusion-weighted images.