Slice-Interleaved Diffusion Encoding MRI Acceleration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current diffusion MRI techniques require lengthy acquisition times, especially problematic for pediatric and uncooperative patients, as they necessitate subjects to remain still, leading to increased discomfort and motion artifacts.
Innovation Solution
The method employs slice-interleaved diffusion encoding (SIDE) with simultaneous multislice (SMS) excitations and graph convolutional neural networks (GCNNs) to reconstruct full diffusion-weighted volumetric images from highly undersampled data, reducing the need for extensive data acquisition and minimizing motion artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional diffusion MRI acquisition is used, then comprehensive diffusion data is obtained, but acquisition time becomes excessively long
Solution Approach 1:
The patent segments the diffusion MRI acquisition by dividing the volume into multiple slices that can be excited and acquired simultaneously using SMS techniques. Each slice is assigned a different diffusion encoding, allowing parallel acquisition of multiple diffusion-weighted images across different spatial locations, thereby reducing total acquisition time while maintaining comprehensive diffusion coverage.
Solution Approach 2:
The patent introduces slice dimension as an additional degree of freedom by assigning different diffusion encodings to different slices. This slice-interleaved approach transforms the traditional single-volume acquisition into a multi-slice parallel acquisition scheme, enabling acceleration factor proportional to the number of slices acquired simultaneously.
2Measurement precision
If longer acquisition time is used, then image quality improves, but subject motion increases
Solution Approach 1:
The patent rushes through the acquisition process by acquiring multiple slices simultaneously with different diffusion encodings in a single excitation cycle. This skips the sequential acquisition of each slice individually, dramatically reducing the time the subject must remain stationary and minimizing motion artifacts while maintaining image quality through advanced reconstruction.
Solution Approach 2:
The patent performs preliminary assignment of different diffusion encodings to different slices before acquisition begins. This preliminary organization allows the system to efficiently parallelize the acquisition across slices, reducing total scan time and thereby reducing the window during which subject motion can occur.
3Productivity
If simultaneous multislice excitations are used, then acquisition speed increases, but image reconstruction complexity increases
Solution Approach 1:
The patent introduces GCNN reconstruction algorithms as an intermediary that bridges the gap between the simplified undersampled k-space data and the desired high-quality diffusion-weighted images. The neural network learns to reconstruct full images from the accelerated acquisition data, handling the complexity of reversing the undersampling and slice-interleaving effects while maintaining image quality.
Solution Approach 2:
The patent changes the reconstruction approach from traditional iterative methods to deep learning-based reconstruction. By training a GCNN on pairs of undersampled and corresponding full images, the system transforms the reconstruction problem into a parameter optimization task where the network learns optimal transformation parameters, simplifying the reconstruction process while handling the complexity of simultaneous multislice data.
4Loss of time
If slice undersampling is applied, then acquisition time reduces, but data completeness decreases
Solution Approach 1:
The patent incorporates feedback mechanisms where the GCNN reconstruction algorithm iteratively refines the image reconstruction by comparing the reconstructed image against the acquired undersampled data and adjusting to minimize errors. This feedback loop ensures that even though slice undersampling reduces raw data completeness, the reconstruction process recovers missing information to produce complete diffusion-weighted images.
Solution Approach 2:
The patent uses the acquired undersampled slice data as a template or copy that the GCNN learns to transform into complete images. By training the neural network on pairs of undersampled copies and corresponding full images, the system learns to generate complete diffusion-weighted images from the undersampled copies, effectively recovering the missing information through learned patterns.
Data Source
AI summary
A method for accelerating diffusion magnetic resonance imaging (MRI) acquisition via slice interleaved diffusion encoding (SIDE) includes conducting a plurality of simultaneous multislice (SMS) excitations for each of a plurality of SIDE diffusion-weighted volumes to obtain SMS images of an MRI subject at different diffusion orientations, regrouping the images into slice groups with different orientations, generating a plurality of slice-undersampled diffusion weighted volumetric images of the subject, wherein each of the plurality of slice-undersampled diffusion weighted volumetric images is produced by cyclically interleaving the slice groups, such that each slice group is associated with a different diffusion wavevector, and reconstructing a full diffusion-weighted volumetric image of the subject by providing the plurality of slice-undersampled diffusion weighted volumetric images to a neural network trained to produce full diffusion-weighted volumetric versions of diffusion magnetic resonance images from undersampled versions of the diffusion magnetic resonance images.