Abstract
Four-dimensional computed tomography (4DCT) is widely used for thoracic radiotherapy planning but is susceptible to respiratory motion artifacts that degrade image quality. Realistic motion artifacts were synthesized using ASTRA-based forward projection of breath-hold CT scans to create paired artifact-corrupted and artifact-free images. Using these paired data, we trained and compared a 3D Residual U-Net and a 3D Generative Adversarial Network (GAN) for motion artifact correction. Compared with the artifact-corrupted input, both frameworks improved agreement with the ground truth. The GAN improved SSIM by 5.6% and reduced MSE by 87%, while the 3D Residual U-Net improved SSIM by 6.6% and reduced MSE by 91% within artifact regions. Both models preserved image quality elsewhere in the lung.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/TIA_026.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{KelKar_Deep_MICCAISAT2026,
author = { Kelat, Karthika AND Bethard, William E. AND Carrizales, Joshua W. AND Gerard, Sarah E. AND Christensen, Gary E. AND Bayouth, John E. AND Reinhardt, Joseph M.},
title = { { Deep Learning Correction of Respiratory Motion Artifacts in 4DCT } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17269},
month = {pending},
page = {pending}
}