Xiaoqun Zhang: Flow based generative models for medical image synthesis

Abstract:

This talk explores advancements in flow-based generative models for medical image synthesis, which are crucial for enhancing diagnostics, treatment planning, and data augmentation. Two novel approaches for bi-modality transfer are introduced. First, SyMOT-Flow minimizes discrepancies between distributions using optimal transport, enabling stable and interpretable image generation. Second, Bi-DPM improves efficiency and quality in bi-modality synthesis by avoiding complex ODE solvers and ensuring consistency across discrete time steps. Both methods are validated on MRI and CT datasets, demonstrating superior image quality and anatomical accuracy.