Overview
Respiratory and bulk motion displace the spinal cord between volumes of a functional MRI (fMRI) time series. Current correction methods such as the Spinal Cord Toolbox (SCT) require several minutes per time series, and general-purpose brain tools register whole volumes—whereas the cord moves primarily slice by slice.
To address this, the authors propose a lightweight convolutional neural network (CNN) that regresses the in-plane translation aligning each axial slice to a reference. Two models, trained without motion labels, correct large motion first and then residual motion.
Results
Evaluated on a 3T dataset (120 scans) and a 7T dataset (59 scans), the proposed method:
- Improves on SCT on every metric (tSNR +10% to +25%; DVARS −20% to −38%)
- Leaves almost no slice displaced by more than one voxel (<0.1%)
- Performs on par with a slice-wise AFNI variant and better than slice-wise FSL FLIRT
- Runs in approximately 10 seconds per scan—about 30 times faster than SCT
Availability
The code is available at github.com/ivadomed/moco-dl, and the method is integrated into SCT v7.4.
Publication Details
- arXiv: 2611.14677 [cs.CV]
- Submitted: 10 November 2026
- Subject: Computer Vision and Pattern Recognition (cs.CV)
- Format: 4 pages, 3 figures
- DOI: https://doi.org/10.48550/arXiv.2611.14677
Authors
Thomas Dagonneau, Sandrine Bédard, Nontharat Tucksinapinunchai, Uten Yarach, Loïc Le Folgoc, Pietro Gori, Julien Cohen-Adad
via ArXiv CV