CA-GAN: Context-Aware Generative Adversarial Networks for Pathological Image Super-Resolution

High-quality pathology images are essential for accurate clinical diagnosis and treatment. However, acquiring high-resolution (HR) pathology images is often hindered by equipment limitations, limited expert availability, and complex slide preparation procedures. Image super-resolution (SR), which reconstructs HR images from low-resolution (LR) inputs, offers a practical solution. However, most existing SR methods are designed for natural images and often struggle to capture the distinct structural characteristics of pathology data. In this paper, we propose Context-Aware Generative Adversarial Network(CAGAN), a novel SR framework tailored for pathological images. It introduces a context path that effectively leverages the rich spatial context in whole slide images (WSIs) while maintaining computational efficiency. In addition, considering the significant differences in staining patterns and reconstruction difficulty between the nucleus and cytoplasm, we propose a Nucleus-Enhanced Hematoxylin Channel (NEHC) loss. This loss imposes targeted constraints on nuclei to better preserve morphological consistency. Experiments on two pathological datasets demonstrate that CA-GAN achieves state-of-the-art performance in both quantitative metrics and perceptual quality. Code will be available soon.