The impact of segmentation approach on HR-pQCT microarchitectural and biomechanical metrics depends on bone structure.

Abstract

High-resolution peripheral quantitative computed tomography (HR-pQCT), combined with micro-finite element (μFE) models, provide a powerful clinical research tool for evaluating bone structure-function relationships with musculoskeletal disorders and bone-targeting treatments. Based on ex vivo cadaveric phantom scans, the Laplace-Hamming (LH) segmentation approach, compared to the standard segmentation approach, improves the accuracy and precision of microarchitecture evaluation using second-generation HR-pQCT scanners. However, the effect of LH segmentation on in vivo scans acquired from clinically relevant cohorts with disease-specific bone microarchitecture (eg, patients with end-stage kidney disease) has not been investigated. Additionally, the effect of LH segmentation on μFE biomechanical estimations remains unexplored, defining the objectives of the current study. Findings from the current study demonstrated that LH segmentation, compared to standard segmentation, reduced structure-dependent bias in HR-pQCT microarchitectural and μFE biomechanical metrics. Specifically, trabecular bone volume fraction (Tb.BV/TV), trabecular thickness (Tb.Th), and cortical pore diameter (Ct.Po.Dm) were particularly sensitive to segmentation strategy. Due to the structure dependence of the standard segmentation approach, applying LH segmentation can alter the results of between-cohort comparisons, potentially leading to different clinical interpretations. For example, differences in cortical porosity (Ct.Po) between healthy participants and patients with end-stage kidney disease were only significant when evaluated using the standard segmentation approach. Thus, it is important that investigators consider the segmentation approach utilized when interpreting HR-pQCT outcome metrics for disease progression or drug effects assessments. Additionally, a structure-based parameter (Tb.Th$\times$Ct.Po.Dm) that robustly predicted the effect of LH segmentation on μFE biomechanical estimations was established. The predictive power of this parameter highlights the importance of incorporating LH segmentation when evaluating cohorts with documented disease-specific alterations in bone microstructure (eg, changes in Tb.Th and Ct.Po.Dm), such as patients with type 2 diabetes.