Segmentation of bone in grey-level tomographs from micro-CT imaging is critical in determining the accuracy of morphometric analysis. The degree of variability in image segmentation between and within multiple operators will be quantified and compared with automated image segmentation. Three cubes of cancellous bone were cut from T12, L1, L3 and L4 human vertebral bodies (n=12). Micro-CT imaging was performed and a global threshold was determined by 3 operators independently and automatically using Otsu′s algorithm. Bone volume, trabecular thickness, trabecular separation, trabecular number, trabecular bone pattern factor, structure model index and degree of anisotropy were calculated. Percent bias and percent random error were calculated between all operators and Otsu’s method. For BV/TV, the maximum percent bias and percent random error were 22.0% and 11.3%, respectively, which constitutes differences in individual measurements between operators of up to 0.07. For Tb.Th, the maximum percent bias and percent random error were 13.1% and 6.4%, respectively, which constitutes differences in individual measurements between operators of up to 35μm. These data highlight to users of micro-CT imaging that morphometric analysis is highly sensitive to operating parameters. The effect on measurements of cancellous bone structure of different operators can be greater than experimental differences, which can lead to erroneous interpretation of results.
Keywords: image segmentation; micro-CT; tomographs; bias; random error; cancellous bone