A computed tomography vertebral segmentation dataset with anatomical variations and multi-vendor scanner data

Hans Liebl,Katharina Grau,Malek El Husseini,Luca Malagutti,Claus Zimmer,Eva Niederreiter,Rickmer Braren,David Schinz,Thomas Baum,Maximilian T Löffler,Giles Tetteh,Amirhossein Bayat,Bjoern Menze,Jan S Kirschke,Anjany Sekuboyina,Benedikt Wiestler

doi:10.1038/s41597-021-01060-0

Abstract

With the advent of deep learning algorithms, fully automated radiological image analysis is within reach. In spine imaging, several atlas- and shape-based as well as deep learning segmentation algorithms have been proposed, allowing for subsequent automated analysis of morphology and pathology. The first “Large Scale Vertebrae Segmentation Challenge” (VerSe 2019) showed that these perform well on normal anatomy, but fail in variants not frequently present in the training dataset. Building on that experience, we report on the largely increased VerSe 2020 dataset and results from the second iteration of the VerSe challenge (MICCAI 2020, Lima, Peru). VerSe 2020 comprises annotated spine computed tomography (CT) images from 300 subjects with 4142 fully visualized and annotated vertebrae, collected across multiple centres from four different scanner manufacturers, enriched with cases that exhibit anatomical variants such as enumeration abnormalities (n = 77) and transitional vertebrae (n = 161). Metadata includes vertebral labelling information, voxel-level segmentation masks obtained with a human-machine hybrid algorithm and anatomical ratings, to enable the development and benchmarking of robust and accurate segmentation algorithms.

Highlights

Background & SummaryNumerous applications of computer-aided diagnostics (CADx) are currently being developed beginning to gradually reshape the future of radiological clinical practice and research[1,2,3,4,5,6,7]
Different deep learning approaches have been used for vertebral labelling and segmentation tasks in the form of convolutional neural networks (CNN), graph convolutional networks (GCN) or point clouds (PC) to analyse bone structures[8,9,10,11,12,13]
Experience, and learning from the VerSe 2019 challenge, we proposed to organise a second iteration of the vertebrae segmentation challenge at the MICCAI 2020 in Lima, Peru

Summary

Background & Summary

Numerous applications of computer-aided diagnostics (CADx) are currently being developed beginning to gradually reshape the future of radiological clinical practice and research[1,2,3,4,5,6,7]. The dataset was split into a training dataset, a public test dataset, and a private test dataset building on the preexisting VerSe 2019 dataset[20] published for the MICCAI conference in 2019, with an overlap of 105 CT image series comprising 319 image series of 300 subjects To date, this dataset represents the largest publicly available CT imaging dataset of the spine with corresponding metadata including labelling information, voxel-level segmentations of all fully visualized vertebrae and definition of enumeration abnormalities and transitional vertebrae. The successful segmentation challenges held at the MICCAI conferences in 2019 and 2020 based on these public datasets confirm, that reliable, fully-automated deep learning algorithms for segmentation of the spine can be trained and that algorithm performance benefits from large and diverse datasets. We are convinced that in the near future, patients will greatly benefit from CADx extracting even more relevant information from medical imaging than currently possible

Methods

Findings

Code availability