Beyond mean trabecular separation: decomposing metaphyseal marrow space into lower- and higher-diameter components

Micro-computed tomography (μCT) is widely used to assess trabecular bone microarchitecture, with trabecular separation (Tb.Sp) among the core parameters recommended for reporting. Tb.Sp is typically expressed as a single volume-weighted mean derived from maximal sphere fitting, although the underlying distribution of local separation values is rarely examined. Here, we show that Tb.Sp distributions in metaphyseal trabecular bone are frequently non-Gaussian and bimodal or multimodal. Using μCT datasets from three established models of osteoporosis, spinal cord injury (SCI), ovariectomy (OVX), and ageing (AGE), we demonstrate that this behaviour is most evident in metaphyseal trabecular bone and is less apparent in epiphyseal trabecular bone or trabecular thickness distributions. We further show that multimodal metaphyseal Tb.Sp distributions contain lower- and higher-diameter components with different predominant spatial localisations. Lower-diameter values were predominantly associated with marrow spaces within the residual trabecular network, whereas higher-diameter values were predominantly associated with larger contiguous marrow cavities. Based on this observation, we introduce a simple extension to standard morphometric analysis in which Tb.Sp is decomposed into lower-diameter trabecular separation (Tb.Sp L ) and higher-diameter trabecular separation (Tb.Sp H ). Tb.Sp decomposition revealed model-specific patterns of trabecular deterioration. SCI showed increases in both components, dominated by a marked redistribution toward Tb.Sp H , consistent with expansion of larger marrow cavities, whereas OVX showed increased Tb.Sp L without a significant Tb.Sp H difference. AGE showed progressive increases in Tb.Sp L and greater representation of Tb.Sp H in older animals. Together, these findings demonstrate that mean Tb.Sp can mask structurally distinct forms of metaphyseal marrow-space organisation, and support reporting distributional descriptors, and where appropriate, Tb.Sp L and Tb.Sp H , alongside conventional Tb.Sp.