Résumé : Immunohistochemistry (IHC) is commonly employed for diagnostic and prognostic purposes in histopathology as well as for biomarker validation in clinical research. Whole slide scanning and image analysis tools now enable objective and quantitative evaluation of IHC biomarkers in a whole tissue slide or a specific region of interest delineated by a pathologist. Compartmentalizing the quantitative evaluation of IHC biomarkers in a specific histological structure is often required to provide more relevant and informative measurements for clinical research. For this purpose, pathologists have to annotate thousands of structures present in histological slide series, a long, tedious and potentially biased task that would greatly benefit from automation.We developed an algorithm for automatically annotating glandular epithelium in slide images from colorectal tissue samples submitted to different staining techniques, including haematoxylin-eosin (H&E) as well as IHC. Our approach combines Deep Learning and a new method of data augmentation. The algorithm implements a convolutional neural network, which was first trained and evaluated with regard to the state-of-the-art on H&E images provided by the international GLaS (Gland Segmentation in Colon Histology Images) challenge contest. To apply our method in the context of IHC staining, we created a second dataset by using tissue microarray slides submitted to IHC to evidence the expression of different antigens on colorectal tumour samples. An expert manually annotated the images to delineate the glandular epithelium. We then quantified the IHC staining in and/or out of the glandular epithelium delineated on the basis of the manual or automatic annotations.Our method achieves state-of-the-art performances in epithelium segmentation on the H&E images and provides accurate segmentation on the IHC images, whatever the targeted antigen. Compartmentalized IHC quantification showed high concordance between measurements carried out using either manual or automatic segmentation. In addition to be efficient in terms of epithelium segmentation, our algorithm is very fast and thus relevant for quantitative IHC analysis performed on large series of whole (tissue or TMA) slides, as generally required in clinical research.