UniStain Dataset

The hematoxylin and eosin (H&E) stain is widely recognized as the gold standard for routine pathological examinations, yet the conventional workflow entails the use of chemical reagents and intricate staining protocols. The emergence of virtual staining presents a promising opportunity to revolutionize traditional chemistry-based staining techniques. However, existing research has primarily focused on single organs, necessitating the development of multiple models for the analysis of diverse organ types, which undoubtedly brings more workload to model training and adaptation. In this paper, we introduce UniStain, a prompt-based organ-control approach to enable versatile multi-organ virtual staining from grayscale autofluorescence (AF) images, allowing a single model to effectively handle diverse organ types. To evaluate the model performance, we curated a new multi-organ pathology image dataset along with microscopic image data from human samples.