CycleGAN for virtual stain transfer: Is seeing really believing?

作者：

Highlights：

• CycleGAN, often used for stain transfer, is sensitive to architectural changes.

• Changes do not affect visual plausibility but do affect pre-trained deep models.

• Stain transfer can perturb important diagnostic markers, limiting reliable use-cases.

摘要

•CycleGAN, often used for stain transfer, is sensitive to architectural changes.•Changes do not affect visual plausibility but do affect pre-trained deep models.•Stain transfer can perturb important diagnostic markers, limiting reliable use-cases.

论文关键词：CycleGAN,Stain transfer,Stain normalisation,Image-to-image translation,Digital histopathology

论文评审过程：Received 28 November 2021, Revised 16 March 2022, Accepted 2 October 2022, Available online 7 October 2022, Version of Record 17 October 2022.

论文官网地址：https://doi.org/10.1016/j.artmed.2022.102420