This presentation aims at introducing a novel approach to segmentation providing a structural guarantee, resulting from a study carried out in collaboration with Oréal research. In this study, consisting in the segmentation of different skin layers within histological images of reconstructed human epidermis, we combined a U-Net type architecture with a « cumulative layer » in order to enforce a layered structure of the output. Moreover, we include an alternate sequential filter within the network to further constrain the shape of the layer contours. The resulting model is trained end-to-end and is fully convolutional.