Gamma Corrected Pyramid Pix2pix – Breast Cancer HE to IHC Image Generation

Breast cancer is the second biggest cause of cancer death in American women. Among the available screening methods for the detection of blood cancer, mammography is the most widely used test. HE (Hematoxylin and Eosin) stained histological analysis, on the other hand, is required to determine the malignancy of the tumor since it analyzes the image at the cellular level. However, Human pidermal Growth Factor Receptor 2 (HER2) expression is critical in the proper therapy of breast cancer. Typically, assessment of HER2 is directed with Immuno-Histo-Chemical techniques (IHC) is a cost-effective method. In developing country like India, the ability to bear the cost for such advanced tests is very difficult from the perspective of patient. In this research work, we present gamma corrected HE images with advanced GAN model employed to generate IHC data in consistent with the paired HE stained images. There are 4870 pairs of images which are registered depicting a variety of HER2 expression levels. Experiments are designed to investigate the influence of pre-processed (gamma corrected) HE image in the process of IHC image generation as well to study the performance of proposed model in comparison with latest reported solutions. Results are quite significantly better as compared to existing solutions. Proposed method; gamma corrected- Pyramid pix-to-pix has recorded promising value of PSNR of 16.024 with SSIM of 0.431. In present scenario, it is cost and time effective to conduct HER2 evaluation through preprocessing and curating IHC-stained slice. There is need for synthesizing the stained image of HE and IHC on Whole Slide Image (WSI). This proposed project is one such attempt, where our solution will complement medical experts to carry out HER2 expression evaluation directly based on the synthesized IHC-stained slices.