A Computed Tomography Vertebral Segmentation Dataset with Anatomical\n Variations and Multi-Vendor Scanner Data

Abstract

With the advent of deep learning algorithms, fully automated radiological\nimage analysis is within reach. In spine imaging, several atlas- and\nshape-based as well as deep learning segmentation algorithms have been\nproposed, allowing for subsequent automated analysis of morphology and\npathology. The first Large Scale Vertebrae Segmentation Challenge (VerSe 2019)\nshowed that these perform well on normal anatomy, but fail in variants not\nfrequently present in the training dataset. Building on that experience, we\nreport on the largely increased VerSe 2020 dataset and results from the second\niteration of the VerSe challenge (MICCAI 2020, Lima, Peru). VerSe 2020\ncomprises annotated spine computed tomography (CT) images from 300 subjects\nwith 4142 fully visualized and annotated vertebrae, collected across multiple\ncentres from four different scanner manufacturers, enriched with cases that\nexhibit anatomical variants such as enumeration abnormalities (n=77) and\ntransitional vertebrae (n=161). Metadata includes vertebral labelling\ninformation, voxel-level segmentation masks obtained with a human-machine\nhybrid algorithm and anatomical ratings, to enable the development and\nbenchmarking of robust and accurate segmentation algorithms.\n

Ähnliche Arbeiten