SpineWeb Dataset 10 (multi-modality vertebra recognition)

SpineWeb Dataset 10 pairs spine MR and CT images with manual vertebra annotations. It was released on SpineWeb, an online platform for spinal imaging research, as reference data for vertebra recognition: finding where each vertebra is, how it is oriented and which level it is. The data accompanies the 2015 IEEE Transactions on Medical Imaging paper by Cai and colleagues, which matches a 3D deformable hierarchical spine model to MR and CT images taken in arbitrary views.

Composition

The SpineWeb entry lists MR and CT images from 20 subjects. It does not say how many scans each subject has, which spine regions are covered, or whether every subject has both modalities. No sex, age, diagnosis or split breakdown is published. The SpineWeb news feed dates the release to 5 April 2015.

Acquisition

The SpineWeb entry gives no sites, scanners, MR sequences, resolution or file format. The paper evaluates its method on T1 and T2 weighted MR and on CT from several sources, including scans collected in the Ontario area of Canada, but it does not state which of its evaluation images make up the released set.

Annotations

For each vertebra, the 3D center location, placed at the spinal cord, and the 3D orientation of the vertebra were annotated by hand. In the paper, annotators drew a planar bounding box over each vertebra in the sagittal, axial and coronal views and assigned a vertebra label to each box.

Known limitations

• The only dataset documentation is a short entry on the SpineWeb datasets page. The paper describes the method and its evaluation data, not the released files.
• No license or terms of use are published. Access starts with a request form that asks for the applicant's institution, a short research description and the principal investigator's contact details.
• On 2026-10-09 the SpineWeb host name did not resolve, so the facts here come from archived copies of the official pages. Whether the data can still be requested is unknown.