Fully Guided Neural Schr\"odinger bridge for Brain MR image synthesis

Multi-modal brain MRI provides essential complementary information for clinical diagnosis, but acquiring all modalities is often time-consuming and costly. Existing methods fall into two categories: paired methods that achieve high accuracy but require impractical large-scale paired datasets, and unpaired methods that are more scalable but often fail to preserve critical anatomical features like lesions. To address this, a team of researchers from multiple Korean institutions (including Hanyeol Yang, Sunggyu Kim, and Jong-Min Lee) introduces the Fully Guided Schrödinger Bridge (FGSB) in a paper on arXiv (2501.14171, revised May 2026). FGSB is a novel framework designed to enable high-fidelity generation with extremely limited paired data, while still preserving clinically relevant lesions when expert annotations or segmentation masks are available.

FGSB operates in two stages: (1) a generation stage that iteratively refines synthetic images using paired source images and Gaussian noise, and (2) a training stage that learns optimal transformation pathways by modeling intermediate states to ensure consistent, high-fidelity synthesis. Experimental results across multiple datasets show that FGSB achieves reliable synthesis performance across diverse imaging resolutions and data acquisition environments. Incorporating lesion-specific priors further enhances the preservation of clinically relevant features, making it particularly valuable for real-world clinical applications where data is scarce but diagnostic accuracy is paramount. The paper spans 33 pages with 6 figures.