Color fourier ptychographic microscopy reconstruction based on double contrast learning

Abstract Fourier ptychographic microscopy (FPM), as an advanced digital imaging technique, holds significant potential in the pathology field. Color FPM images are essential for accurate pathological analysis. Currently, color FPM images are primarily acquired through modifications in imaging devices or virtual staining algorithms. However, the complexity of pathological slides often results in unclear image details during the imaging process, thereby impacting the subsequent colorization outcomes. To address these challenges, we propose an enhanced dual-contrast learning virtual staining algorithm to optimize the color imaging effect of Fourier ptychographic microscopy. We integrate an additional set of generators and discriminators into the ECS-GAN network for training. Furthermore, we introduce the HWD downsampling module to retain essential information in the images and extract more discriminative features to enhance the quality of virtual staining images. Comparative experimental results with other advanced methods demonstrate that our network achieves superior staining outcomes, outperforming existing methods in both subjective observations and objective quantitative evaluations.