Purpose: In whole‑spine imaging vertebrae at the lumbosacral junction are named by counting caudally from C2, but lumbar‑only scans (T12‑S1) lack C2, making the count ambiguous when a lumbosacral transitional vertebra (LSTV) is present. Abdominopelvic CT captures the same range plus the lowest ribs and pelvis, offering a chance to resolve the ambiguity locally. CTSpinoPelvic1K investigates whether local morphology alone can identify a vertebra without a full‑spine count.
Data source and construction: Building on CTSpine1K (spine) and CTPelvic1K (pelvis), the bone‑richest series of each patient were paired and annotated under the VerSe native scheme for spine, ribs and femora. The release contains 802 CT studies with per‑level rib and femur labels, levels anchored on the lowest rib‑bearing vertebra and S1, and explicit classes for L6, T13, a separate S1 and lumbar ribs, so that anomalies are recorded as such.
Validation: All geometric invariants passed (802/802); rib‑vertebra incidence across 5,749 ribs showed a 0.035% offset; spinopelvic measurements matched published values.
Data format and usage: NIfTI image‑label pairs are provided, grouped by patient and stratified by LSTV into five‑fold splits, together with a loader. Download at https://doi.org/10.5281/zenodo.22139642.
Potential applications: vertebra classification from local features; updating cadaveric morphometry; spinopelvic assessment; opportunistic screening; and surgical planning research in the absence of a public pre‑operative lumbar cohort (377 prone‑position records). Limitations: thoracic ground truth limited by field‑of‑view; postural angles are supine; no held‑out test set; ribs are reviewer‑triaged pseudo‑labels; Castellvi grades achieved two‑reader consensus on only 33 records.
Review