A fully automated pipeline for dens morphometry based on semantic and geometric segmentation of an enhanced statistical shape model of the C2 vertebra

This paper introduces a fully automated pipeline for the morphometric analysis of the odontoid process (dens) of the second cervical vertebra, based on an enhanced Statistical Shape Model (eSSM). The proposed framework integrates semantic and geometric information into the statistical model, enabling automatic identification of dens and related anatomical entities, as well as the extraction of morphometric parameters from three-dimensional anatomical models. The pipeline includes automatic alignment of the vertebrae to a standardised Local Coordinate System, semantic segmentation of the dens based on skeleton-line analysis, automatic geometric detection of the inferior surface of the vertebral body, and a dedicated measurement protocol that relies on slicing and convex hull computation. Unlike deep learning approaches, the eSSM framework inherently preserves anatomical coherence and remains effective even when trained on relatively limited datasets. The model was trained on 25 C2 vertebrae from the VerSe dataset and subsequently evaluated on ten independent specimens. The resulting morphometric measurements were consistent across all test specimens and fell within the anatomical ranges reported in the literature, thus supporting the anatomical plausibility and methodological reliability of the proposed approach. While the current implementation should be regarded as a methodological proof of concept due to the limited training dataset, this framework demonstrates the potential of enhanced statistical shape models for automated, reproducible, and anatomically consistent morphometric analysis of complex vertebral structures.