Chabert, SterenSterenChabertGustavo UlloaRodrigo NaranjoHéctor Allende‐CidHéctor Allende2025-08-252025-08-252019-01-0110.1007/978-3-030-13469-3_782-s2.0-85063039270https://cris-uv-2.scimago.es/handle/123456789/5408WOS:001416966000078Convolutional Neural Networks (CNN) have been obtaining successful results in the task of image segmentation in recent years. These methods use as input the sampling obtained using square uniform patches centered on each voxel of the image, which could not be the optimal approach since there is a very limited use of global context. In this work we present a new construction method for the patches by means of a circular non-uniform sampling of the neighborhood of the voxels. This allows a greater global context with a radial extension with respect to the central voxel. This approach was applied on the 2015 Longitudinal MS Lesion Segmentation Challenge dataset, obtaining better results than approaches using square uniform and non-uniform patches with the same computational cost of the CNN models.enacceso restringidoTheoretical Computer ScienceComputer Scienceconference paper