The field of medical imaging is witnessing a significant shift towards the adoption of generative models, which are being leveraged to improve image segmentation, reconstruction, and generation. These models have shown great promise in enhancing the accuracy and reliability of medical imaging tasks, particularly in high-stakes clinical settings. The use of flow-based generative models, in particular, has emerged as a key area of research, with these models demonstrating the ability to estimate complex distributions and capture uncertainty in medical images. Furthermore, the development of hybrid models that combine generative and discriminative approaches has shown great potential in unifying classification, generation, and uncertainty quantification in medical imaging. Noteworthy papers in this area include:

• Flow Stochastic Segmentation Networks, which introduces a generative segmentation model family featuring discrete-time autoregressive and modern continuous-time flow variants, achieving state-of-the-art results on medical imaging benchmarks.
• MedSymmFlow, which presents a generative-discriminative hybrid model built on Symmetrical Flow Matching, delivering reliable uncertainty estimates and matching or exceeding the performance of established baselines in classification accuracy and AUC.