Accuracy of CT‐based thickness measurement of thin structures: Modeling of limited spatial resolution in all three dimensions

Abstract

Measurement of the width of thin structures such as the cortical shell of the vertebral body or femoral neck with computed tomography (CT) is limited by the spatial resolution of the CT system. Limited spatial resolution exists both within the CT image plane and perpendicular to it and can be described by the in-plane point spread function (PSF) and the across-plane slice sensitivity profile (SSP), respectively. The goal of this study was to confirm that errors of thickness measurement of thin structures critically depend on the spatial positioning of the object and the spatial resolution limitations of CT in all three dimensions, and to assess the size of the errors themselves. We compared computer models that incorporated both effects to experimentally assessed cortical thicknesses of the European Spine Phantom. Analysis included varying CT slice width, the orientation of measurement and angle beta of misalignment of longitudinal scanner and phantom axes. Agreement of models with measurements was good in all configurations with an overall error of 0.17 mm. This showed that PSF and SSP are adequate system characteristics to predict deviation of measured values from true widths. Errors between measurements and true cortical thickness values delta(true) averaged to 1.5 mm were strongly positively correlated with slice width d and beta. When the across-plane partial volume effect was eliminated, limited in-plane resolution still accounted for overestimation of delta(true) by 0.68 (137%), 0.27 (27%), and 0.06 mm (4%) for delta(true)=0.5, 1.0, and 1.5 mm, respectively. For delta(true) of 1.0 mm and above, it was shown that although the absolute cortical thickness values might not be accurately measurable, relative differences between two values are reflected in measurement. Implications for cortical thickness measurement are that the spinal cortical shell is too thin, whereas accurate assessment at locations of the femoral neck exhibiting a thicker cortical shell of both difference and absolute values should be possible with CT even for larger misalignment angles, especially when a smaller CT slice width is chosen.

A quasi-brittle continuum damage finite element model of the human proximal femur based on element deletion

“…While the QCT measurement of human femur distinguishes between cortical and trabecular bone, partition of the femur into trabecular and cortical regions based on the empirical separation threshold (HU>600) may generate a certain inaccuracy in the cortical bone thickness, which can play an important role in bone resistance to fracture [9]. The accuracy of the cortical thickness measurement with QCT depends on the spatial resolution of the CT system, image noise, the positioning of specimens and the image processing software [59]. Also, Lee et al [41] and Popescu et al [60] reported that the spatial registration between CT scans of a patient's organ and the computer model of the organ is still a critical step in most computer-aided models accuracy.…”

An analysis algorithm for accurate determination of articular cartilage thickness of hip joint from MR images

“…This may be caused by the finite width of PSF (25,28,29). Thus, we propose a new measurement method that greatly reduces PSF induced bias by incorporating PSF directly into thickness measurement.…”

Accuracy of lung nodule density on HRCT: analysis by PSF‐based image simulation

“…It can be assumed that a PSF is separable into a two‐dimensional (2D) PSF in the xy scan plane and a slice sensitivity profile (SSP) in the z direction perpendicular to the scan plane 12 , 14 . Then, the three‐dimensional (3D) CT image I(x, y, z) can be expressed as follows: $Ifalse(x,y,zfalse)=false(Ofalse(x,y,zfalse)\ast \ast \mathrm{P}\mathrm{S}\mathrm{F}false(x,yfalse)false)\ast \mathrm{S}\mathrm{S}\mathrm{P}false(zfalse)$ …”

“…Several studies 12 , 14 have performed the above computer simulations. In those simulations, the PSFs, SSPs, and object functions were treated as discrete (digital) data; however, the interval of the discrete data was not specified.…”

“…The object functions used in those image simulations were designed to emulate cortical bone, 11 , 12 small high‐density structures (calcifications and stented vessels), (13) and lung nodules (14) . To the best of our knowledge, such simulated images were computed with a fine digital sampling pitch (the interval of the discrete data); one such example is illustrated in Fig.…”