luohwu/SpineDepth_segmented

SpineDepth (preprocessed) — CT ↔ RGB-D lumbar vertebrae

A preprocessed version of the SpineDepth dataset (Liebmann et al., 2021) for multi-modal CT ↔ intraoperative RGB-D bone-surface registration of lumbar vertebrae. Each vertebra is provided as a preoperative CT bone mesh together with two intraoperative RGB-D point clouds (one per depth-camera view). It accompanies the Medical Image Analysis 2026 paper (see Citation).

Dataset summary

• 8 cadaveric specimens (IDs 2–9), lumbar levels L1–L5.
• 40 preoperative CT bone meshes (.stl) — one vertebra model per (specimen, level).
• 80 intraoperative RGB-D point clouds (.xyz) — two camera views (camera0, camera1) per vertebra, ~30k–40k points each (XYZ in millimetres).
• Every vertebra is thus one CT mesh + two RGB-D clouds: a ready CT↔RGB-D registration set.

Excluded specimens

The original SpineDepth comprises ten cadaveric specimens (1–10). Specimens 1 and 10 are excluded from this release because they provide insufficient anatomical coverage for intraoperative surface registration. For specimen 1, the original recordings used a standard midline approach with the soft tissue left intact (pedicle screws instrumented from TH12–S1 prior to the experiment), exposing too little bone surface. The released set is therefore specimens 2–9.

Preprocessing

This is not the raw SpineDepth release. Starting from the original SpineDepth RGB-D recordings (tracked video sequences) and CT data, we provide level-wise segmentation for both modalities: each lumbar vertebra (L1–L5) is isolated into its own intraoperative RGB-D point cloud and its own preoperative CT bone mesh, rather than a whole-spine scan. Each RGB-D cloud is built by merging the tracked frames of a recording into one point cloud per camera view, downsampling it, and segmenting it per vertebral level; the matching preoperative mesh is that vertebra's CT bone surface. See our paper for the full procedure.

Directory structure

SpineDepth/
├── README.md
├── preoperative/                 # CT bone meshes
│   └── 2_L1.stl  2_L2.stl  ...  9_L5.stl          # <specimen>_<level>.stl              (40 files)
└── intraoperative/               # merged + downsampled RGB-D point clouds
    └── 2_L1_camera0.xyz  2_L1_camera1.xyz  ...     # <specimen>_<level>_camera<0|1>.xyz  (80 files)

specimen ∈ {2..9}, level ∈ {L1..L5}, camera ∈ {0,1}. Each .xyz is plain XYZ coordinates in millimetres, one point per line; each .stl is a triangle mesh of the CT-segmented vertebra.

Usage

Each preoperative/<s>_<L>.stl (CT model) pairs with the two intraoperative/<s>_<L>_camera{0,1}.xyz (RGB-D) clouds of the same vertebra for CT↔RGB-D bone-surface registration (e.g. with NeuralBoneReg).

Citation

If you use this dataset, please cite both our paper and the original SpineDepth:

@article{wu2026neuralbonereg,
  title   = {NeuralBoneReg: An instance-specific label-free point cloud-based method for multi-modal bone surface registration},
  author  = {Wu, Luohong and Seibold, Matthias and Cavalcanti, Nicola A. and Ao, Yunke and Flepp, Roman and Massalimova, Aidana and Calvet, Lilian and F{\"u}rnstahl, Philipp},
  journal = {Medical Image Analysis},
  year    = {2026},
  doi     = {10.1016/j.media.2026.104133}
}

@article{liebmann2021spinedepth,
  title   = {{SpineDepth}: A Multi-Modal Data Collection Approach for Automatic Labelling and Intraoperative Spinal Shape Reconstruction Based on {RGB-D} Data},
  author  = {Liebmann, Florentin and St{\"u}tz, David and Suter, Daniel and Jecklin, Sascha and Snedeker, Jess G. and Farshad, Mazda and F{\"u}rnstahl, Philipp and Esfandiari, Hooman},
  journal = {Journal of Imaging},
  volume  = {7},
  number  = {9},
  pages   = {164},
  year    = {2021},
  doi     = {10.3390/jimaging7090164}
}