Abstract

Accurate and volumetrically consistent synthetic CT generation remains challenging across heterogeneous anatomy. Slice-wise 2D models are efficient but can introduce inter-slice discontinuities, whereas full 3D models incur substantially greater computational cost. We propose an anatomy-guided, slice-consistent latent diffusion framework for multi-anatomy MR-to-CT synthesis. It combines 2.5D MRI conditioning, slice-relative noise, anatomy-aware guidance, and lightweight 3D refinement to model local anatomical correlations and suppress z-axis artifacts without full 3D diffusion. On 1,129 paired MRI/CT volumes from head-and-neck, thorax, abdomen, pelvis, and spine, our method achieved a PSNR of 31.40 and an SSIM of 0.9018, outperforming representative CNN-, GAN-, and diffusion-based baselines. Qualitative and ablation results further showed improved sagittal and coronal continuity and fewer inter-slice artifacts. Code is available at https://github.com/allenem/LSRN.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMMCA2026_006.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=o7uljEcZNm

BibTex

@InProceedings{PuYao_Bridging_MICCAISAT2026,
        author = { Pu, Yao AND Zhang, Zhiqing AND Yu, Peixin AND Zhao, Mayang AND Zhuang, Yitao AND Zhang, Zhenxi AND Cai, Jing AND Ren, Ge},
        title = { { Bridging Slices to Volumes: Anatomy-Guided Slice-Consistent Latent Diffusion with Local Correlated Noise for MR-to-CT 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}
}


back to top