In recent years, numerous deep learning-based methods have been developed for the segmentation of nuclei in hematoxylin and eosin (H&E) images, with many achieving performance levels approaching those of pathologists. Nevertheless, the direct application of such methods to immunohistochemistry images may result in the omission of cell nuclei locations, which in turn affects the accuracy of nuclei identification and segmentation. The objective of this paper is to enhance the identification and segmentation precision of cell nuclei in immunohistochemical images, thereby furnishing more dependable data for pathology diagnosis and biomarker analysis. This paper developed a method called UPHGAN, which is based on Generative Adversarial Networks (GAN), to convert IHC images into H&E images while preserving the location and morphology of cell nuclei. Next, we applied a pre-trained cell nucleus segmentation model to the H&E images. The method uses the U-Net 512 generator deep learning architecture to capture cell nuclei's complex features effectively. The PatchGAN discriminator enhances the model's ability to recognize details. The Huber loss function balances the adversarial and cooperative relationship between the generator and discriminator. Our model efficiently segmented cell nuclei in two publicly available histopathological image datasets, surpassing existing technologies. It is capable of handling irregular cell nuclei and demonstrated excellent segmentation performance with Recall scores of 84% and 78%, PixAcc scores of 96% and 87%, and F1 scores of 63% and 56%, respectively.