Buckets:
| license: cc-by-nc-4.0 | |
| pretty_name: "SpineDepth (preprocessed) — CT & RGB-D lumbar vertebrae" | |
| tags: | |
| - medical-imaging | |
| - rgbd | |
| - depth-camera | |
| - computed-tomography | |
| - spine | |
| - point-cloud | |
| - registration | |
| - orthopedics | |
| size_categories: | |
| - n<1K | |
| # 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](#citation)). | |
| ## Dataset summary | |
| - **7 cadaveric specimens** (IDs `2`–`8`), lumbar levels **L1–L5**. | |
| - **35 preoperative CT bone meshes** (`.stl`) — one vertebra model per (specimen, level). | |
| - **70 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, 9, and | |
| 10 are excluded** from this release because they provide **insufficient anatomical | |
| coverage** for intraoperative surface registration (see [Liebmann et al., 2021](https://doi.org/10.3390/jimaging7090164) and [Massalimova et al., 2025](https://doi.org/10.1080/24699322.2025.2511126)). 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; for specimen 9 the depth views observe too little of each vertebra to | |
| disambiguate its near-symmetric pose. The released set is therefore specimens 2–8. | |
| For completeness, the preprocessed data for the excluded specimens is still provided under | |
| `excluded_specimens/` (mirroring the main `preoperative/` + `intraoperative/` layout): CT bone | |
| meshes for specimens 1, 9, and 10, plus RGB-D clouds for specimens 9 and 10 (specimen 1's bone | |
| surface is too occluded to reconstruct usable intraoperative clouds). This data is **not** part of | |
| the released registration set and is **not** loaded by the default pipeline. | |
| ## 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 ... 8_L5.stl # <specimen>_<level>.stl (35 files) | |
| ├── intraoperative/ # merged + downsampled RGB-D point clouds | |
| │ └── 2_L1_camera0.xyz 2_L1_camera1.xyz ... # <specimen>_<level>_camera<0|1>.xyz (70 files) | |
| └── excluded_specimens/ # specimens 1, 9, 10 (excluded above) — provided for completeness | |
| ├── preoperative/ # CT bone meshes for 1, 9, 10 | |
| └── intraoperative/ # RGB-D clouds for 9 and 10 (specimen 1: CT only) | |
| ``` | |
| `specimen ∈ {2..8}`, `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: | |
| ```bibtex | |
| @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} | |
| } | |
| ``` | |
Xet Storage Details
- Size:
- 5.21 kB
- Xet hash:
- 007ccbff3be851d1aabd5a0eb67472628470ee5701a7e4e20426aafd36559cc5
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.