DTNet: Dual-encoder generative adversarial network for generating breast cancer immunohistochemical images

In order to address the coloring problem in breast cancer, we propose an innovative generator network architecture paradigm and a pooling-based loss function to tackle the challenge of strict alignment between image pairs. Our network architecture primarily consists of a content encoder, a style encoder, a decoder, and a TSA skip-connection module. The content encoder ensures that content features are preserved during coloring, while the style encoder guarantees the accuracy of coloring styles. The TSS skipconnection layer based on a transformer architecture is employed to facilitate the flow of feature information. Experimental evaluations on four coloring datasets demonstrate that our model exhibits superior performance and robustness compared to existing models.