Background: In the clinical practice of pathology, Hematoxylin-and-Eosin (H&E) staining isthe gold standard for basal cell carcinoma (BCC) diagnosis. However, standard H&E stainingof high-quality tissue sections requires lengthy and laborious tissue preparation. Fluorescenceconfocal microscopy (FCM) of fresh or frozen tissue enables fluorescence detection, rapid andhigh-resolution imaging, with minimal tissue preparation. Despite these advantages, gray-levelFCM images are not easy to read for pathologists due to the lack of diagnosis specificity comparedwith H&E-stained images. To correlate the color of the FCM images with H&E images, thiswork proposes a deep learning model for the computational staining of FCM images based on anunsupervised approach.Methods: In this study, we investigated the utility of FCM on BCC tumor sections, stained withAcridine Orange and computationally colored to simulate H&E dyes. We adopted an unsupervisedCycleGAN framework to computationally stain FCM images. The dataset consists of unpairedand unannotated thousands of FCM and H&E patches from whole slide images (WSI)of BCC tissue sections. CycleGAN includes two forward and backward GANs which completea cycle to ensure a reliable transformation between the two domains. The network was trainedunder adversarial, cycle-consistency, and saliency constraint learning scheme mapping betweenFCM and H&E images while avoiding distortions of the image content. The coupling of forwardand backward mapping together ensures that a generated image is close to its original. The structuralsimilarity index (SSIM) scores were computed between source and reconstructed images toshow information preservation for each cycle. We evaluate the quality of the generated imagescompared to the original images using similarity measures.Results: We assessed the quality of the images with subtyping BCC and skin tissue characteristicsqualitatively. The generated H&E-like images from FCM through this CycleGAN modelwere qualitatively and quantitatively similar to real H&E images. We also achieved a high cycleconsistency for the generator networks by obtaining similarity indices greater than ۰.۹۲. Highlyresembling H&E staining allow the pathologist’s easy adaptation. The application of the FréchetInception Distance (FID) was used to measure the quality of generated H&E images and indicatedthat the similarity improves up to ۰.۷۳ after the transformation of images from source to target.Conclusion: Our results suggest that combining the FCM and computational staining using theCycleGAN model can eliminate the need for BCC tissue preparation steps (except staining). Theproposed method has the potential to expand the application of rapid analysis of tissue which iscomparable to the standard histopathology images. Thus, using the CycleGAN model for computationalstaining is beneficial for diagnostic applications with simplifying laboratory stainingprocedures. We believe that our approach has significant potential in clinical computational stainingand advances the progress of computer-aided histology image analysis. Keywords: Basal cellcarcinoma (BCC), Fluorescence confocal microscopy, pathology, hematoxylin-and-eosin (H&E),cycle-consistent generative adversarial network (CycleGAN), deep learning