Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DMD-augmented Unpaired Neural Schrödinger Bridge for Ultra-Low Field MRI Enhancement

Youngmin Kim, Jaeyun Shin, Jeongchan Kim, Taehoon Lee, Jaemin Kim, Peter Hsu, Jelle Veraart, Jong Chul Ye

TL;DR

The proposed framework achieves an improved realism structure trade-off, enhancing distribution level realism on unpaired benchmarks while increasing structural fidelity on the paired cohort compared to unpaired baselines.

Abstract

Ultra Low Field (64 mT) brain MRI improves accessibility but suffers from reduced image quality compared to 3 T. As paired 64 mT - 3 T scans are scarce, we propose an unpaired 64 mT $\rightarrow$ 3 T translation framework that enhances realism while preserving anatomy. Our method builds upon the Unpaired Neural Schrödinge Bridge (UNSB) with multi-step refinement. To strengthen target distribution alignment, we augment the adversarial objective with DMD2-style diffusion-guided distribution matching using a frozen 3T diffusion teacher. To explicitly constrain global structure beyond patch-level correspondence, we combine PatchNCE with an Anatomical Structure Preservation (ASP) regularizer that enforces soft foreground background consistency and boundary aware constraints. Evaluated on two disjoint cohorts, the proposed framework achieves an improved realism structure trade-off, enhancing distribution level realism on unpaired benchmarks while increasing structural fidelity on the paired cohort compared to unpaired baselines.

DMD-augmented Unpaired Neural Schrödinger Bridge for Ultra-Low Field MRI Enhancement

TL;DR

The proposed framework achieves an improved realism structure trade-off, enhancing distribution level realism on unpaired benchmarks while increasing structural fidelity on the paired cohort compared to unpaired baselines.

Abstract

Ultra Low Field (64 mT) brain MRI improves accessibility but suffers from reduced image quality compared to 3 T. As paired 64 mT - 3 T scans are scarce, we propose an unpaired 64 mT 3 T translation framework that enhances realism while preserving anatomy. Our method builds upon the Unpaired Neural Schrödinge Bridge (UNSB) with multi-step refinement. To strengthen target distribution alignment, we augment the adversarial objective with DMD2-style diffusion-guided distribution matching using a frozen 3T diffusion teacher. To explicitly constrain global structure beyond patch-level correspondence, we combine PatchNCE with an Anatomical Structure Preservation (ASP) regularizer that enforces soft foreground background consistency and boundary aware constraints. Evaluated on two disjoint cohorts, the proposed framework achieves an improved realism structure trade-off, enhancing distribution level realism on unpaired benchmarks while increasing structural fidelity on the paired cohort compared to unpaired baselines.
Paper Structure (17 sections, 12 equations, 2 figures, 1 table)

This paper contains 17 sections, 12 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method: DMD2-augmented UNSB
  4. UNSB Formulation.
  5. DMD2 Augmented Adversarial Loss.
  6. Regularization for Anatomical Structure Preservation.
  7. Overall Objective.
  8. Experiments
  9. Datasets.
  10. Protocol.
  11. Baselines.
  12. Unpaired evaluation.
  13. Paired evaluation.
  14. Experimental Results
  15. Quantitative Results.
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Overview. The proposed framework builds upon UNSB to perform multi-step 64 mT$\rightarrow$3 T translation. During training, DMD2-based diffusion-guided distribution alignment using a frozen 3T teacher and a trainable fake critic augments adversarial supervision. Structural fidelity is enforced through combined PatchNCE and ASP regularization, which constrains anatomical consistency.
  • Figure 2: Qualitative comparison on the test set. The figure shows two different slices extracted from distinct subjects within the test set. Rows: 64mT input, CUT, CycleGAN, UNSB, Ours, and Target (3T). Columns: T1/T2 and T1/T2 .