Beyond Mean HU: Spatial Radiodensity Mapping of the Human Hyoid as a Baseline for Morphological and Biomechanical Investigation

Bone radiodensity is often reduced to global Hounsfield Unit (HU) summaries, although its spatial distribution may contain biologically relevant information. This study introduced a template-based framework for spatial radiodensity mapping of the adult human hyoid and tested whether spatial descriptors capture age- and sex-related variation better than whole-bone HU measures. Clinical head-and-neck CT scans from 300 adults, including 180 females and 120 males, were analyzed. Hyoids were segmented, normalized into a common anatomical template, and divided into 1500 supervoxels for spatial analysis. Scalar HU summaries were compared with PCA-based spatial descriptors, and regional effects were assessed using supervoxel-wise models with false-discovery-rate correction. The registered radiodensity maps showed spatial coherence, with neighboring supervoxels displaying more similar profiles than distant ones. In this cohort, spatial descriptors outperformed mean HU for age prediction, although performance remained limited (R2 = 0.255). Sex classification was moderate (ROC AUC = 0.810; balanced accuracy = 0.725). After the spatial features were adjusted for native volume, sex-classification AUC fell to 0.671, and none of the 51 unadjusted regional sex associations survived volume adjustment. Age-related associations involved the hyoid body, the body–greater horn transition, and the proximal-to-mid greater horns. These findings demonstrate spatial coherence in the registered maps but show that the sex-related pattern is strongly size-dependent.