The field of medical imaging is witnessing significant advancements with the integration of synthetic data and vision-language models. Researchers are exploring the potential of synthetic data to mitigate issues such as dataset diversity and class imbalance, leading to improved performance in tasks like brain tumor segmentation. Vision-language models are also being developed to effectively leverage volumetric medical images and associated clinical narratives, enhancing the generalization ability of learned encoders. These innovative approaches are paving the way for more accurate and realistic medical image synthesis, with applications in data augmentation, privacy-preserving synthesis, and diagnostic signal preservation. Noteworthy papers include:

• A study on using synthetic data in brain tumor segmentation, which demonstrated the feasibility of synthetic data as an augmentation strategy.
• The introduction of VELVET-Med, a novel vision-language pre-training framework for volumetric imaging tasks in medicine, which achieved state-of-the-art performance across various downstream tasks.
• The proposal of a two-stage Posterior-Mean Rectified Flow pipeline for synthesizing volumetric contrast-enhanced brain MRI from non-contrast inputs, which restored lesion margins and vascular details realistically.
• The development of CTFlow, a latent flow matching transformer model for 3D CT synthesis conditioned on clinical reports, which demonstrated superiority in terms of temporal coherence, image diversity, and text-image alignment.