CONVERSION OF HE-STAINED IMAGES TO IHC IMAGES USING THE FEYNMAN-KAC DIFFUSION METHOD

Abstract

In this study, we propose the Feynman Diffusion Generative Model for converting HE tissue images into IHC images, a result that supports physicians in developing breast cancer treatment plans. This method leverages Feynman techniques by exploiting the relationship between stochastic differential equations (SDEs) and partial differential equations (PDEs), an approach that has seen significant success in previous research on SDEs. Specifically, we represent the reverse diffusion process in the form of the Feynman-Kac formula, which facilitates the transformation from an SDE model to a corresponding PDE system to describe the recovery process from noisy states to real data. The use of PDEs enhances both the stability and accuracy of inference, while also optimizing the flow normalizing objective function related to the data distribution flow. Experimental results on the proposed model achieved a PSNR of 19.25 and an SSIM of 0.569, outperforming previously published methods on the BCI dataset.