Towards Reliable Stain Transfer: An Iterative Data-Model Co-Optimization Framework Based on Multimodal Expert-Guided Assessment

Researchers introduced DMCoStain, an iterative data-model co-optimization framework for computational stain transfer in histopathology.

DMCoStain uses Multimodal Expert-Guided Finer Selection powered by an IHC-positive-expression vision-language model to refine training data and improve performance.

The team also released ImmunoInstruction, a new 150K-sample instruction dataset to support better stain-transfer learning.

The paper reports state-of-the-art results across multiple tissues and biomarkers, suggesting more reliable H&E-to-IHC image generation for pathology workflows.