Whole slide images (WSIs) are rich in prognostic information that can be quantified by means of deep learning and computer image analysis algorithms. In this study, we train a tissue classification network and a nuclei segmentation network. Then we use these models to process colon cancer patient cohort in order to evaluate the prognostic value of a risk index computed using morphological features for epithelium nuclei found in a tumour region to patient overall survival (OS). First, Convolutional Neural Networks (CNNs) were used to segment tissues in WSIs in a patch-wise approach. A tumour region is then localised in order to segment and classify the nuclei using U-Net segmentation network. Finally, we extract morphological features from the epithelium nuclei to predict patient OS. The generated risk index was shown to be significant to patient survival in a univariate and multivariate Cox analysis.