Measuring Marrow Density and Area Using Peripheral Quantitative Computed Tomography at the Tibia: Precision in Young and Older Adults and Individuals With Spinal Cord Injury

The objective of this study was to compare the test-retest precision error for peripheral quantitative computed tomography (pQCT)-derived marrow density and marrow area segmentation at the tibia using 3 software packages. A secondary analysis of pQCT data in young adults (n = 18, mean ± standard deviation 25.4 ± 3.2 yr), older adults (n = 47, 71.8 ± 8.2 yr), and individuals with spinal cord injury (C1-T12 American Spinal Injury Association Impairment Scale, classes A-C; n = 19, 43.5 ± 8.6 yr) was conducted. Repeat scans of the tibial shaft (66%) were performed using pQCT (Stratec XCT2000). Test-retest precision errors (root mean square standard deviation and root mean square coefficient of variation [RMSCV%]) for marrow density (mg/cm³) and marrow area (mm²) were reported for the watershed-guided manual segmentation method (SliceOmatic version 4.3 [Sliceo-WS]) and the 2 threshold-based edge detection methods (Stratec version 6.0 [Stratec-TB] and BoneJ version 1.3.14 [BoneJ-TB]). Bland-Altman plots and 95% limits of agreement were computed to evaluate test-retest discrepancies within and between methods of analysis and subgroups. RMSCV% for marrow density segmentation was >5% for all methods across subgroups (Stratec-TB: 12.2%-28.5%, BoneJ-TB: 14.5%-25.2%, and Sliceo-WS: 10.9%-23.0%). RMSCV% for marrow area segmentation was within 5% for all methods across subgroups (Stratec-TB: 1.9%-4.4%, BoneJ-TB: 2.6%-5.1%, and Sliceo-WS: 2.4%-4.5%), except using BoneJ-TB in older adults. Intermethod discrepancies in marrow density appeared to be present across the range of marrow density values and did not differ by subgroup. Intermethod discrepancies varied to a greater extent for marrow area and were found to be more frequently at mid- to higher-range values for those with spinal cord injury. Precision error for pQCT-derived marrow density segmentation exceeded 5% for all methods of analysis across a range of bone mineral densities and fat infiltration, whereas precision error for marrow area segmentation ranged from 2% to 5%. Further investigation is necessary to determine alternative acquisition and analysis methods for pQCT-derived marrow segmentation.