Related Experiment Video
Updated: Aug 15, 2025
09:33
Diffusion Tensor Magnetic Resonance Imaging in the Analysis of Neurodegenerative Diseases
Published on: July 28, 2013
28.5K
Affine transformation edited and refined deep neural network for quantitative susceptibility mapping
Zhuang Xiong1, Yang Gao2, Feng Liu1
1School of Information Technology and Electrical Engineering, University of Queensland, Brisbane, Australia.
Neuroimage
|December 31, 2022
Summary
We developed the AFTER-QSM deep neural network for robust Quantitative Susceptibility Mapping (QSM). It overcomes limitations of existing methods, offering high accuracy across various acquisition parameters and significantly faster reconstruction times.
Area of Science:
- Medical Imaging
- Neuroimaging
- Artificial Intelligence in Medicine
Background:
- Deep neural networks show promise for Quantitative Susceptibility Mapping (QSM).
- Existing deep learning methods for QSM struggle with variations in acquisition parameters like orientation and resolution.
- This performance degradation limits the clinical applicability of current QSM techniques.
Purpose of the Study:
- To introduce an end-to-end deep neural network, AFTER-QSM, for robust dipole inversion in QSM.
- To enhance the generalizability of QSM reconstruction across diverse acquisition orientations and spatial resolutions.
- To improve the speed and accuracy of QSM compared to conventional and existing deep learning methods.
Main Methods:
- Proposed an end-to-end deep neural network architecture named AFfine Transformation Edited and Refined (AFTER) for QSM.
- The network incorporates a forward affine transformation, a U-Net for dipole inversion, an inverse affine transformation, and a Residual Dense Network (RDN) for refinement.
- Validated the method using both simulated data and in-vivo experiments, assessing robustness against varying acquisition parameters.
Main Results:
- AFTER-QSM demonstrated excellent generalizability, successfully reconstructing susceptibility maps from highly oblique and anisotropic scans.
- Achieved superior image quality in simulations and reduced streaking artifacts and noise in in-vivo experiments compared to other methods.
- Ablation studies confirmed the RDN component significantly mitigated blurring and underestimation caused by affine transformations.
Conclusions:
- The AFTER-QSM network provides a robust and generalizable solution for Quantitative Susceptibility Mapping.
- This method significantly improves QSM reconstruction quality and reduces artifacts, especially under challenging acquisition conditions.
- AFTER-QSM drastically reduces reconstruction time from minutes to seconds, enhancing clinical workflow efficiency.
