VSDFullBodyBoneReconstruction
A forensic post-mortem CT from the forensic institutes of Bern and Zurich, from the lower lumbar spine to the toes, with its 21 bones segmented (Fischer 2023, on the Virtual Skeleton Database's CTs).
10 test caseslower limbsimage CTresearch only
Read before comparing (5)
- Images. Forensic post-mortem CTs from the forensic institutes of Bern and Zurich, the voxels as delivered inside the box of the two segmented blocks (pelvis and thighs, shanks and feet), in Hounsfield units: pixels 0.88 to 1.27 mm, slices 0.25 to 0.7 mm apart, 1,637 to 4,743 slices. The z subjects' headers put the feet at the top; the case is under the author's upside-down frame, so the head is up, nothing resampled.
- Truth. The author (Fischer 2023) thresholded bone above 200 HU, split it at every joint and corrected it by hand in 3D Slicer, closing the outer surface's holes; each bone's enclosed space is filled here, marrow included. One reader; no agreement between readers is published. 21 labels: the left and right femur, hip, patella, tibia, fibula, talus, calcaneus, other tarsals, metatarsals and toe phalanges, and the sacrum.
- Cases. On development case 002, 2,834 bone voxels of the iliac crest, drawn on its thorax CT, lie past the CT's cut-off and are dropped; the test cases have none.
- Overlap. No other dataset of the corpus holds these CTs.
- Start. A mask of the whole image scores a Dice of at most 0.011 for any bone at the median of the test cases, 0.012 on any case. TotalSegmentator (total) on test case z001 scores 0.95 to 0.97 on the five bones it draws, both femurs, both hips and the sacrum, and nothing on the other 16, which its total task does not segment: 0.23 over the 21.
No method has a result on every case of this dataset yet.
Case by case
Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- 10 test subjects drawn with seed 2026 among the 15 complete subjects of the author's table whose two blocks were cut from one CT, then the other 5 for development. Left out: 010, 015, 016 and 017, whose pelvis and legs are two CTs in two frames, and 019, whose zip holds no whole CT; and the ten subjects the author marks incomplete or inconsistent (z013, z036, z049, z050, z055, z057, z061, z063, z064, z066). The record has no test set.
- Truth
- contours by a program, verified by people the author's Reconstruction segmentations: bone above 200 HU, split at every joint and cleaned by hand in 3D Slicer, its outer surface's holes closed by hand; here each bone's enclosed space is filled, marrow included, as the author's bone volumes count it. Sesamoid bones sit in the metatarsals' label, the coccyx in the sacrum's; tarsal_other holds the tarsals other than the talus and the calcaneus (navicular, cuboid, cuneiforms)
- Source
- zenodo.org/records/8302449
doi.org/10.5281/zenodo.8302449
doi.org/10.1038/s41597-023-02669-z
zenodo.org/records/8270365
doi.org/10.2196/jmir.2930 - Licence
- CC BY-NC-SA 4.0 (segmentations); CC BY-NC-SA 3.0 CH (the Virtual Skeleton Database's CTs)
5 cases for tuning, never tested
z009, 002, z042, z019, 006