Clinical validation of automatic phantom-less quantitative computed tomography for osteoporosis screening: fat region of interest comparison and multidevice validation

Abstract

Background Osteoporosis significantly increases fracture risk and healthcare burden. Traditional bone mineral density screening requires dedicated equipment, limiting opportunistic screening. Phantom-less quantitative computed tomography (PL-QCT) enables bone density assessment during routine chest CT examinations but requires comprehensive validation. This study uses fat region of interest (ROI) comparison and multidevice validation to verify the feasibility and accuracy of automatic PL-QCT for osteoporosis screening. Methods This prospective study included 252 participants who underwent routine chest CT examinations. An automatic PL-QCT system was used to measure the volumetric bone mineral density at the T11–L1 vertebrae. Clinical validation compared measurements with phantom-based quantitative computed tomography (PB-QCT) using the system’s standard pipeline with automatically selected subcutaneous fat ROIs and, for methodological comparison, manually selected visceral fat ROIs. Diagnostic performance was evaluated via receiver operating characteristic curve analysis with the DeLong test, whereas accuracy was assessed via linear regression, ICC, and Bland‒Altman analysis. Multidevice validation was conducted across five CT scanners. Results The automated PL-QCT system under validation (using automatically selected subcutaneous fat ROI) demonstrated excellent diagnostic performance, with an area under the curve (AUC) of 0.962 for osteoporosis and 0.794 for osteopenia, and showed strong correlation with PB-QCT (r = 0.867, ICC = 0.926). In the methodological comparison, the PL-QCT values derived from the manually selected visceral fat ROI showed significantly superior performance to those from the automated system (subcutaneous fat ROI) for the diagnosis of osteopenia (P = 0.029). Multidevice validation demonstrated consistency across the five CT scanners. Gender analysis revealed greater accuracy in females with distinct age-related bone density patterns. Conclusions This study validates that the automated PL-QCT system (utilizing its standard pipeline with automatically selected subcutaneous fat ROI) provides a reliable approach for opportunistic osteoporosis screening during routine chest CT examinations. A key methodological finding is that the manually selected visceral fat ROI demonstrated superior performance to the system’s automatically selected subcutaneous fat ROI specifically in osteopenia detection, offering important evidence for optimizing future calibration strategies. The technology’s cross-platform consistency, confirmed by multidevice validation, supports its broad clinical applicability.