Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Motion and Eddy-Current Correction in Diffusion MRI

Muqari, Kinda LU (2025) BMEM01 20261
Division for Biomedical Engineering
Abstract
Subject motion and eddy-current distortions represent significant challenges in diffusion MRI, particularly for clinical protocols utilizing "half-sphere" sampling. While the current gold standard, FSL’s eddy, employs Gaussian Process (GP) predictions to correct these artifacts, its performance can degrade under specific acquisition constraints, most notably when sampling lacks antipodal symmetry. This limitation compromises the sensitivity required to detect subtle microstructural changes in neurodegenerative diseases.

This thesis presents ECHO (Eddy Current and Head motion cOrrection), a novel hierarchical framework optimized for such challenging acquisitions. Unlike existing methods, ECHO utilizes a biophysically informed GP based on... (More)
Subject motion and eddy-current distortions represent significant challenges in diffusion MRI, particularly for clinical protocols utilizing "half-sphere" sampling. While the current gold standard, FSL’s eddy, employs Gaussian Process (GP) predictions to correct these artifacts, its performance can degrade under specific acquisition constraints, most notably when sampling lacks antipodal symmetry. This limitation compromises the sensitivity required to detect subtle microstructural changes in neurodegenerative diseases.

This thesis presents ECHO (Eddy Current and Head motion cOrrection), a novel hierarchical framework optimized for such challenging acquisitions. Unlike existing methods, ECHO utilizes a biophysically informed GP based on a synthetic tissue basis to predict robust reference signals. The correction model progressively refines estimates through global polynomial, volume-wise rigid-body, and slice-level affine transformations, optimized via Levenberg–Marquardt.

We evaluated ECHO against eddy and the extrapolation-based method (EB-MEC) across three cohorts (N = 30), ranging from legacy clinical data to high-angular-resolution diffusion imaging (HARDI). Performance was assessed using blinded visual scoring and a novel quantitative metric, Rim Standard Deviation, designed to measure signal stability at the brain's periphery. ECHO consistently outperformed EB-MEC and matched or exceeded eddy's across all datasets. Notably, ECHO demonstrated superior robustness in high-b-value regimes where eddy exhibited instability, while simultaneously achieving a 61% reduction in computational runtime.

Our proposed algorithm ECHO achieves as accurate or more accurate corrections than eddy in the tested datasets, while requiring substantially less processing time and relying solely on CPU resources. The newly developed evaluation metric aligns closely with visual assessments, providing an objective and efficient means to compare motion and eddy-current correction algorithms. Together, these advances enable a more robust and accessible correction pipeline for diffusion MRI. (Less)
Popular Abstract
New algorithm to Correct Brain Scans
MRI distortions limit early disease diagnosis. This study evaluates a new algorithm designed to fix these images, demonstrating results that match or even surpass current industry standards.

Medical imaging can be affected by distortions. A complex MRI technique, called Diffusion MRI, is used to map the brain's microscopic "wiring", for which patients must lie perfectly still while the scanner works. But in reality, patients move. On top of that, the scanner’s own powerful magnets create interference known as "eddy currents." Together, these forces stretch and blur the images, potentially hiding early signs of diseases like Alzheimer's.

This thesis evaluates and further develops a new solution... (More)
New algorithm to Correct Brain Scans
MRI distortions limit early disease diagnosis. This study evaluates a new algorithm designed to fix these images, demonstrating results that match or even surpass current industry standards.

Medical imaging can be affected by distortions. A complex MRI technique, called Diffusion MRI, is used to map the brain's microscopic "wiring", for which patients must lie perfectly still while the scanner works. But in reality, patients move. On top of that, the scanner’s own powerful magnets create interference known as "eddy currents." Together, these forces stretch and blur the images, potentially hiding early signs of diseases like Alzheimer's.

This thesis evaluates and further develops a new solution called ECHO, designed to correct these distortions. The current industry standard is good, but often struggles when the data is incomplete or of poor quality. Unlike current methods, which correct distortions using a guess based entirely on raw data, which, when poor, can lead to bad results, ECHO uses a "physics-based blueprint" for guessing. It simulates how water moves in healthy tissue to create a perfect reference.

ECHO matched the current gold standard in most cases and sometimes even outperformed it, particularly in the lower parts of the brain where interference is worst. Just as importantly, it did this much faster, cutting the processing time by over 60%. This makes large-scale research faster and more accessible. (Less)
Please use this url to cite or link to this publication:
author
Muqari, Kinda LU
supervisor
organization
course
BMEM01 20261
year
type
H2 - Master's Degree (Two Years)
subject
keywords
MRI, Diffusion MRI, Eddy-currents, Correction algorithm, Image registration
language
English
additional info
2026-01
id
9218374
date added to LUP
2026-01-30 13:13:57
date last changed
2026-01-30 13:13:57
@misc{9218374,
  abstract     = {{Subject motion and eddy-current distortions represent significant challenges in diffusion MRI, particularly for clinical protocols utilizing "half-sphere" sampling. While the current gold standard, FSL’s eddy, employs Gaussian Process (GP) predictions to correct these artifacts, its performance can degrade under specific acquisition constraints, most notably when sampling lacks antipodal symmetry. This limitation compromises the sensitivity required to detect subtle microstructural changes in neurodegenerative diseases.

This thesis presents ECHO (Eddy Current and Head motion cOrrection), a novel hierarchical framework optimized for such challenging acquisitions. Unlike existing methods, ECHO utilizes a biophysically informed GP based on a synthetic tissue basis to predict robust reference signals. The correction model progressively refines estimates through global polynomial, volume-wise rigid-body, and slice-level affine transformations, optimized via Levenberg–Marquardt.

We evaluated ECHO against eddy and the extrapolation-based method (EB-MEC) across three cohorts (N = 30), ranging from legacy clinical data to high-angular-resolution diffusion imaging (HARDI). Performance was assessed using blinded visual scoring and a novel quantitative metric, Rim Standard Deviation, designed to measure signal stability at the brain's periphery. ECHO consistently outperformed EB-MEC and matched or exceeded eddy's across all datasets. Notably, ECHO demonstrated superior robustness in high-b-value regimes where eddy exhibited instability, while simultaneously achieving a 61% reduction in computational runtime.

Our proposed algorithm ECHO achieves as accurate or more accurate corrections than eddy in the tested datasets, while requiring substantially less processing time and relying solely on CPU resources. The newly developed evaluation metric aligns closely with visual assessments, providing an objective and efficient means to compare motion and eddy-current correction algorithms. Together, these advances enable a more robust and accessible correction pipeline for diffusion MRI.}},
  author       = {{Muqari, Kinda}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Motion and Eddy-Current Correction in Diffusion MRI}},
  year         = {{2025}},
}