DEMIST: Decoupled Multi-stream latent diffusion for Quantitative Myelin Map Synthesis

TL;DR

DEMIST introduces a novel multi-stream latent diffusion approach to synthesize quantitative myelin maps from standard MRI scans, significantly reducing scan time while maintaining high accuracy and boundary sharpness.

Contribution

It proposes a decoupled multi-stream latent diffusion model with innovative conditioning mechanisms for efficient and accurate PSR map synthesis from common MRI images.

Findings

Outperforms VAE, GAN, and diffusion baselines on multiple metrics.

Produces sharper boundaries and better quantitative agreement.

Validated on 163 scans from 99 subjects with 5-fold cross-validation.

Abstract

Quantitative magnetization transfer (qMT) imaging provides myelin-sensitive biomarkers, such as the pool size ratio (PSR), which is valuable for multiple sclerosis (MS) assessment. However, qMT requires specialized 20-30 minute scans. We propose DEMIST to synthesize PSR maps from standard T1w and FLAIR images using a 3D latent diffusion model with three complementary conditioning mechanisms. Our approach has two stages: first, we train separate autoencoders for PSR and anatomical images to learn aligned latent representations. Second, we train a conditional diffusion model in this latent space on top of a frozen diffusion foundation backbone. Conditioning is decoupled into: (i) \textbf{semantic} tokens via cross-attention, (ii) \textbf{spatial} per-scale residual hints via a 3D ControlNet branch, and (iii) \textbf{adaptive} LoRA-modulated attention. We include edge-aware loss terms to preserve lesion boundaries and alignment losses to maintain quantitative consistency, while keeping the number of trainable parameters low and retaining the inductive bias of the pretrained model. We evaluate on 163 scans from 99 subjects using 5-fold cross-validation. Our method outperforms VAE, GAN and diffusion baselines on multiple metrics, producing sharper boundaries and better quantitative agreement with ground truth. Our code is publicly available at https://github.com/MedICL-VU/MS-Synthesis-3DcLDM.