Large-Scale Pretraining for Improving Deep Learning-Based Geometric Distortion Correction of Diffusion-Weighted Imaging
Published in arXiv preprint arXiv:2609.06437, 2026
Recommended citation: Khanal, Saroj, Yashawant Kumar Yadav, Kritam Bhattarai, Jeevan Neupane, Shristi Subedi, Saship Gwachha, Manish Kumar Tiwari, Dong Zhang, Confidence Raymond, Aondona Moses Iorumbur, Udunna Anazodo, Surendra Maharjan, Bishesh Khanal, Mahesh Shakya, and Pralhad Kumar Shrestha. "Large-Scale Pretraining for Improving Deep Learning-Based Geometric Distortion Correction of Diffusion-Weighted Imaging." arXiv:2609.06437 (2026) https://arxiv.org/abs/2609.06437
We study whether large-scale pretraining (a self-supervised model and a generative model) improves deep learning-based correction of geometric distortion in single-phase-encoding diffusion-weighted MRI, compared with a non-pretrained baseline. Pretrained models do better, but applying the best one to data from a low-resource setting shows transferability problems, which registration to a common standard space helps to reduce.