深度学习使辅助诊断的软件能够更积极有效地开发和应用，但是组织病理学图像的颜色变化降低了这些算法的性能。染色归一化可以解决扫描仪效应、不同的染色方法、患者的疾病状态、染色时间等因素产生的图像异质性。虚拟染色可以摆脱载玻片染色，减少载玻片的制备步骤，为临床缩短样本的制备时间，节省大量的成本。在缺乏注释训练数据的情况下，病理图像数据增强可用于创建具有纹理和颜色、样式逼真的人工样本来促进网络训练。本文就组织学病理图像在深度学习病理分析中染色处理的染色归一化、虚拟染色和数据增强等方面展开综述，为组织学病理图像在临床上的应用和研究提供参考。

Deep learning allows software to assist in diagnosis to be developed and applied more aggressively and efficiently, whereas the color variability of histopathology images degrades the performance of these algorithms. Stain normalization can address image heterogeneity arising from scanner effects, different staining methods, patient’s disease states, staining times, and other factors. Virtual staining can eliminate slide staining and reduce slide preparation steps, reducing sample preparation time for the clinic and saving significant costs. In the absence of annotated training data, pathology image data augmentation is performed by creating artificial samples with realistic texture, color and style to facilitate network training. In this paper, we made a review on staining processing of histological pathology images in deep learning pathology analysis to provide a reference for histological pathology maps in clinical applications and research.