This study used 26,275 standardised whole‑body magnetic resonance imaging (MRI) examinations drawn from the population‑based German National Cohort (NAKO). Participants were adults aged 19–74 years. The dataset provided broad population representation compared with earlier rib‑cage shape work that relied on smaller clinical computed tomography cohorts and therefore over‑represented injury and disease.
Ribs were segmented with a deep‑learning pipeline described as a rib‑extended SPINEPS model. Each of the 24 ribs was reconstructed as a per‑rib surface mesh. The pipeline produced the surface geometry used as input for dense correspondence and statistical modelling.
Per‑rib surface meshes were brought into dense vertex‑wise correspondence using Gaussian‑process morphable registration implemented in Scalismo. The aligned ensemble of rib‑cage surfaces was summarised by generalised Procrustes analysis to remove pose and scale differences, followed by principal component analysis (PCA) to characterise the dominant modes of geometric variation across the cohort. The final model is a surface‑based statistical shape model (SSM) spanning the complete 24‑rib cage.
Shape variation was strongly concentrated in a modest number of PCA modes: 28 modes captured 95% of total variance. The first three principal components alone accounted for 69.4% of variance (PC1 42.6%, PC2 16.3%, PC3 10.5%). Anatomical readings provided for the leading modes were:
These PCA modes thus admit interpretable anatomical descriptions linking abstract statistical axes to recognizable shape features.
To bridge PCA modes with anatomically named features, the authors computed 14 per‑rib geometric descriptors. These descriptors provided a quantitative cross‑walk between the abstract PCA space and explicit rib features (for example, local size, curvature, and angular relationships). The descriptor set enabled subsequent multivariable modelling of associations between shape and participant characteristics.
Associations between rib‑cage shape and participant covariates were estimated by multivariable regression modelling with Benjamini‑Hochberg false‑discovery‑rate control. Key findings reported from these analyses include:
The source reports these directionally and comparatively but does not provide additional numeric effect sizes for all covariates beyond the variance explained and Cohen’s d for sex separation noted above.
The modelling workflow combined dense correspondence registration, Procrustes alignment, and PCA for dimensionality reduction, followed by multivariable regression to relate PCA loadings and geometric descriptors to participant phenotypes. Multiple testing was controlled using the Benjamini‑Hochberg false‑discovery‑rate procedure. The authors emphasise that shape variation is concentrated in a relatively small number of modes, simplifying downstream association testing and interpretation.
The authors released the SSM assets and analysis resources: full source code, the canonical run configuration, and the figure bundle are available on GitHub (NAKO‑Ribcage‑SSM). The mean shape, principal modes, and metadata loadings are archived on Zenodo. An interactive manuscript with a 3D model viewer is also provided.
Individual‑level NAKO imaging and phenotype data are not publicly shareable due to data‑protection regulations; researchers may request access through the NAKO application portal subject to data‑access policies.
Ethical approvals were in place: the NAKO study and whole‑body MRI examinations were approved by the Ethics Committee of the Bavarian State Chamber of Physicians and the ethics committees of the 18 NAKO study centres. The present analysis used pseudonymised NAKO imaging and baseline data under the established approval and data‑use regulations; written informed consent was obtained from all participants.
The released population‑representative geometric reference can serve as a benchmark for rib‑cage morphology in population studies, enable morphing of donor‑derived finite‑element human‑body models, and support further large‑cohort shape analyses. The authors position the SSM as a resource for reproducible, population‑anchored work on thoracic biomechanics, ventilation, and injury response, while noting that individual NAKO data access requires standard application procedures.