Abstract
The high-b-value DWI (HBV) and the apparent diffusion coefficient (ADC) map are essential sequences, along with T2, for determining PI-RADS score to assess prostate cancer risk. They are also the ones most often degraded or missing, and replacing them requires an additional patient visit for rescanning. We propose a Latent Diffusion Model (LDM) for prostate MR image synthesis that generates HBV and ADC jointly from T2, conditioned on anatomical segmentations, in a single shared latent space. Our model generates a whole 3D volume, which can be helpful for having consistency between slices compared to 2D or 2.5D based models. Diffusion-based approach was employed as it captures the full distribution of textures consistent with a given T2 volume, rather than collapsing to their average, as regression-based models tend to do. We compare against a deterministic ResUNet and a pix2pix cGAN built on the same backbone to show our approach produces more consistent and anatomically faithful synthesis. In the axial plane, our LDM and the cGAN showed similar quantitative performance, but the LDM outperformed in through-plane slices, achieving values comparable to the real-to-real floor. We also show that conditioning on the segmentation improves ADC synthesis, and appending synthesized images to T2 raises AUC across every csPCa classifier we trained. These findings support diffusion-based synthesis as a potential strategy for recovering missing prostate MRI sequences while preserving diagnostic utility for csPCa detection.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/cdmri_017.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=WkveIUOFef
BibTex
@InProceedings{ParKan_SegmentationConditioned_MICCAISAT2026,
author = { Park, Kangwoo AND Noh, Yung-Kyun AND Kim, Heejong},
title = { { Segmentation-Conditioned Latent Diffusion for Prostate High-b-value DWI and ADC Synthesis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}