Machine Learning-Based Multiplex Immunofluorescence Staining from Immunohistochemistry with Generative Adversarial Networks

Abstract

Immunohistochemistry (IHC) and Immunofluorescence (IF) are the two most commonly employed methods of cell staining, which aims to enhance visibility of cellular components for medical research and diagnosis. Immunofluorescence which uses fluorescently labeled antibodies to detect specific proteins is preferred over immunohistochemistry for multiplexing and colocalization which analyze protein distributions due to its transparent nature. However access is often barred due to high costs of reagents and equipment, slide quality degradation overtime, and technical limitations such as autofluorescence, and background staining from non-specific binding or cross reactivity antibodies. To address the aforementioned issue, we propose a novel machine learning-based immunofluorescence staining approach utilizing Generative Adversarial Network (GAN). The model intakes a digital pathology image of IHC and converts it to an IF image through feature maps. The proposed model achieved state-of-art performance with PSNR value of 30.67 and SSIM value of 0.8992. Additionally, the qualitative experimental results demonstrate the efficacy of the proposed method in enhancing generating high-quality IF stained images.