Segmentation of pathology images proves to be a challenging problem because of the high complexity of the data: complex tissue morphology, imprecision of the expert annotation, inherent multi-resolution nature of the target concept, to name just a few aspects. The deep learning (DL) approach led to the development of several successful models for tackling specific segmentation problems. This thesis will further explore the use of deep neural networks for the semantic segmentation of pathology images. The regions to be recognized are defined based on a pathologist's annotation of "important" regions - some are carcinoma/tumor, but others refer to vessels, nerves, connective tissue and the problem is not limited to cancer vs normal segmentation. The resulting approach should work generalize to other classes, provided that the annotation exists. The outcome of the thesis will be (i) a short review of the available (DL) methods for pathology image analysis; (ii) a practical implementation of a method for segmenting images into a set of predefined classes based on DL models; (iii) a method for quantitatively evaluating the results and (iv) analysis of the results and comparison with some baseline models. The student will be provided with in-house training and validation sets (colon cancer cases) and will be allowed to use any of the public data sets deemed suitable. The implementation may use any available libraries and toolkits that may help in achieving the goals.