Prob-BBDM Synthesizes MRI Sequences with High Accuracy

In medical imaging, multi-modal image analysis is crucial for optimizing examination quality, but acquiring multiple MRI modalities is often resource-intensive and time-consuming. To address this, a new image-to-image translation model called Prob-BBDM (Probabilistic Brownian Bridge Diffusion Model) has been developed. This model is designed to synthesize magnetic resonance imaging (MRI) sequences directly from 2D axial slices. Prob-BBDM integrates a variational encoder-guided diffusion mechanism, leveraging probabilistic image distributions to significantly enhance synthesis quality. Evaluated on the BraTS 2021 dataset, the model demonstrated superior performance across various translation tasks, achieving high SSIM and PSNR scores, with consistent improvements over baseline methods. Notably, its diffusion process requires only four steps, making it computationally efficient while maintaining high-quality synthesis. Further validation on an external dataset confirmed its generalizability, and synthesized slices used as input for a pre-trained segmentation model yielded strong Dice scores, confirming the preservation of critical diagnostic information.