In this section
@inproceedings{Wang:2026:10.1109/ISBI61048.2026.11515678,
author = {Wang, Z and Yang, L and Wang, F and Wu, Y and Zhang, Z and Huang, L and Yang, G},
doi = {10.1109/ISBI61048.2026.11515678},
title = {Making 3D Diffusion Easier: Autocalibration-Signal-Conditioned Diffusion Model for Dynamic Mri Reconstruction},
url = {http://dx.doi.org/10.1109/ISBI61048.2026.11515678},
year = {2026}
}
TY - CPAPER
AB - Highly accelerated dynamic magnetic resonance imaging (MRI) reconstruction is urgently needed to enable timeefficient and patient-friendly imaging. Diffusion models show great promise with their flexibility and robustness at high acceleration factors. However, the complexity and instability of high-dimensional diffusion often force existing approaches to decompose the natural 3D spatiotemporal problem into multiple 2D sub-problems, limiting their ability to capture the full 3D data distribution. In this work, we propose DiffACS, an autocalibration-signal (ACS)conditioned 3D diffusion model that bridges this gap by incorporating ACS information into the diffusion process via the element-wise cross-attention. The corresponding ACS images, derived from the fully sampled k-space center, naturally preserve low-frequency-related contrast and structural cues, providing an effective condition that simplifies the 3D diffusion, enhances stability, and facilitates faithful recovery of high-frequency-related details. Extensive experiments on cardiac cine MRI datasets demonstrate that DiffACS not only achieves state-of-the-art and robust reconstructions under high acceleration scenarios, but also breakthroughs the performance bottleneck of vanilla 3D diffusion, highlighting the critical role of highquality condition in practical 3D diffusion models.
AU - Wang,Z
AU - Yang,L
AU - Wang,F
AU - Wu,Y
AU - Zhang,Z
AU - Huang,L
AU - Yang,G
DO - 10.1109/ISBI61048.2026.11515678
PY - 2026///
SN - 1945-7928
TI - Making 3D Diffusion Easier: Autocalibration-Signal-Conditioned Diffusion Model for Dynamic Mri Reconstruction
UR - http://dx.doi.org/10.1109/ISBI61048.2026.11515678
ER -
For enquiries about the MRI Physics Collective, please contact:
Mary Finnegan
Senior MR Physicist at the Imperial College Healthcare NHS Trust
Pete Lally
Assistant Professor in Magnetic Resonance (MR) Physics at Imperial College
Jan Sedlacik
MR Physicist at the Robert Steiner MR Unit, Hammersmith Hospital Campus