The recent developments in the field of vision-language pre-training and its applications in medical imaging and pathology have shown significant advancements. Researchers are increasingly focusing on zero-shot and few-shot learning scenarios, leveraging the power of multi-modal models to tackle complex tasks such as lesion segmentation, nuclei detection, and camouflaged object segmentation without the need for extensive annotated datasets. These approaches are particularly innovative as they bridge the gap between visual and textual data, enabling models to generalize better to unseen data. The integration of cross-modal knowledge injection and auto-prompting techniques is proving to be a game-changer, enhancing the performance of models in label-free environments. Additionally, the adaptation of foundation models to various downstream tasks in pathology, through benchmarking and parameter-efficient fine-tuning, is providing valuable insights into the deployment of these models in clinical settings. Notably, the field is also exploring the potential of large-scale visual-language pre-trained models for tasks in the medical field, demonstrating their versatility and efficacy. These trends indicate a shift towards more adaptable and efficient models that can operate in diverse and data-limited environments, pushing the boundaries of what is possible in medical image analysis.