# GDCNet: Calibrationless geometric distortion correction of echo planar imaging data using deep learning

arXiv: https://arxiv.org/abs/2402.18777 · HTML: https://mlsanity.com/paper/2402.18777/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 5 of its 51 references that are in the map.
- First posted on arXiv 29 Feb 2024.

## Builds on (most cited first, 5 of 5)

- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems](https://mlsanity.com/paper/1603.04467/) (2016)
- [VoxelMorph: A Learning Framework for Deformable Medical Image Registration](https://mlsanity.com/paper/1809.05231/) (2018)
- [FD-Net: An Unsupervised Deep Forward-Distortion Model for Susceptibility Artifact Correction in EPI](https://mlsanity.com/paper/2303.10436/) (2023)
- [Off-resonance artifact correction for magnetic resonance imaging: a review](https://mlsanity.com/paper/2205.01028/) (2022)

## Built on by (most cited first, 1 of 1)

- [Unsupervised Susceptibility Distortion Correction of EPI without Calibration Scans via Image Translation-Based Registration](https://mlsanity.com/paper/2606.21588/) (2026)

## Abstract (arXiv)

Functional magnetic resonance imaging techniques benefit from echo-planar imaging's fast image acquisition but are susceptible to inhomogeneities in the main magnetic field, resulting in geometric distortion and signal loss artifacts in the images. Traditional methods leverage a field map or voxel displacement map for distortion correction. However, voxel displacement map estimation requires additional sequence acquisitions, and the accuracy of the estimation influences correction performance. This work implements a novel approach called GDCNet, which estimates a geometric distortion map by non-linear registration to T1-weighted anatomical images and applies it for distortion correction. GDCNet demonstrated fast distortion correction of functional images in retrospectively and prospectively acquired datasets. Among the compared models, the 2D self-supervised configuration resulted in a statistically significant improvement to normalized mutual information between distortion-corrected functional and T1-weighted images compared to the benchmark methods FUGUE and TOPUP. Furthermore, GDCNet models achieved processing speeds 14 times faster than TOPUP in the prospective dataset.