“…Several segmentation methods for axon and myelin have been proposed which are based on traditional image processing algorithms including thresholding and morphological operations 8,9 , axon shape-based morphological discrimination 10 , watershed 11,12 , region growing 13 , active contours without 14,15 and with discriminant analysis 15 . However, few limitations can be reported from the previous work: (i) traditional image-based methods are designed to work on specific imaging modalities and often fail if another contrast is used (e.g., optical image instead of electron microscopy); (ii) previous methods are not fully-automatic as they typically require either preprocessing, hand-selected features for axon discrimination and/or postprocessing; (iii) traditional image-based methods do not make full use of the contextual information of the image (i.e., multi-scale representation of axons, average shape of axons, etc.)…”