ALT Schematic of the workflow: each stent is imaged by placing it in an X-ray beam from a synchrotron storage ring and rotating it during acquisition of 2D radiographic projections. The stack of projections then undergoes a phase retrieval step and is thus converted into a stack of 2D phase maps. These are then subjected to tomographic reconstruction, yielding a 3D volume data set in which each voxel is represented by a floating-point value indicating its electron density, modulo image artifacts and noise. To obtain a thematic label for each voxel in the volume (i.e., to "segment" the image), a convolutional neural network ("U-net") is used. The segmented data are then used for visualization and quantification of the degradation.