Related Experiment Video
Updated: Jun 6, 2026
04:44
Inter-Brain Synchrony in Open-Ended Collaborative Learning: An fNIRS-Hyperscanning Study
Published on: July 21, 2021
FMRI 3D registration based on Fourier space subsets using neural networks
Luis C Freire1, Ana R Gouveia, Fernando M Godinho
1Escola Superior de Tecnologia da Saúde de Lisboa, Instituto Politécnico de Lisboa, 1990-096, Portugal. luis.freire@estesl.ipl.pt
Summary
This study introduces a fast neural network (NN) method for aligning functional MRI (fMRI) images using Fourier coefficients. The technique achieves high accuracy in 3D rigid-body registration, offering potential for real-time fMRI applications.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Machine Learning
Background:
- Functional Magnetic Resonance Imaging (fMRI) is crucial for neuroscience research.
- Accurate image registration is essential for analyzing fMRI data, especially correcting for head motion.
- Existing registration methods can be computationally intensive, limiting real-time applications.
Purpose of the Study:
- To develop and evaluate a novel neural network (NN) based method for 3D rigid-body registration of fMRI time series.
- To leverage a limited set of Fourier coefficients for efficient image alignment.
- To assess the accuracy and speed of the proposed NN approach for fMRI registration.
Main Methods:
- A neural network (NN) model was designed to process Fourier coefficients from 3D fMRI images.
- Six NNs were employed, each estimating a specific rigid-body registration parameter (translations and rotations).
- The method utilizes a small cubic neighborhood in the first octant of 3D Fourier space, including the DC component.
Main Results:
- The NN-based method achieved mean absolute registration errors of approximately 0.030 mm for translations and 0.030 degrees for rotations.
- The training set construction and learning stages were rapid, taking 90 seconds and 1-12 seconds, respectively.
- The method demonstrated effectiveness across various DC neighborhood sizes and typical fMRI motion amplitudes.
Conclusions:
- Neural network-based approaches show significant promise for advancing fMRI registration techniques.
- The proposed method offers a fast and accurate solution for 3D rigid-body registration of fMRI data.
- This approach could be valuable for real-time (in-frame) fMRI registration, potentially using limited K-space data.
