Deep Learning-Inferred Multiplex ImmunoFluorescence for Lamina-associated polypeptide 2 and Ki-67

Whole Slide Imaging (WSI) is a widely used pathological technology to digitize slide images and improve their navigation. Dyeing is an important part of this process, with two major options—Immunohistochemistry (IHC) and ImmunoFluorescence (IF). IF provides noticeably higher quality and contrast, as well as multiplexing capabilities, allowing scans of multiple targets simultaneously. However, IF is considerably more expensive than IHC, putting pathologists who use WSI for applications such as cancer treatment at a disadvantage. In this paper, I propose a novel solution, generating IF images from IHC images using a style transfer network. The proposed system is based on a Generative Adversarial Network (GAN), which utilizes a generator and a discriminator. Specifically, the system is trained to stain Lamina-associated polypeptide 2 and Ki-67. Additionally, I introduce multiplex staining results to demonstrate the system's effectiveness in a broader context. The proposed system achieved a Peak Signal-to-Noise Ratio (PSNR) of 31.38 and a Structural Similarity Index Measure (SSIM) of 0.8764 on a public immunofluorescence dataset.