Abstract
Longitudinal partial volume effects (z-axial PVE), which occur when an object partly occupies a slice, degrade image resolution and contrast in computed tomography (CT). Z-axial PVE is unavoidable for subslice objects and reduces their contrast according to their fraction contained within the slice. This effect can be countered using a smaller slice thickness, but at the cost of an increased image noise or radiation dose. The aim of this study is to offer a tool for optimizing the reconstruction parameters (slice thickness and slice spacing) in CT protocols in the case of partial volume effects. This optimization is based on the tradeoff between axial resolution and noise. For that purpose, we developed a simplified analytical model investigating the average statistical effect of z-axial PVE on contrast and contrast-to-noise ratio (CNR). A Catphan 500 phantom was scanned with various pitches and CTDI and reconstructed with different slice thicknesses to assess the visibility of subslice targets that simulate low contrast anatomical features present in CT exams. The detectability score of human observers was used to rank the perceptual image quality against the CNR. Contrast and CNR reduction due to z-axial PVE measured on experimental data were first compared to numerical calculations and then to the analytical model. Compared to numerical calculations, the simplified algebraic model slightly overestimated the contrast but the differences remained below 5%. It could determine the optimal reconstruction parameters that maximize the objects visibility for a given dose in the case of z-axial PVE. An optimal slice thickness equal to three-fourth of the object width was correctly proposed by the model for nonoverlapping slices. The tradeoff between detectability and dose is maximized for a slice spacing of half the slice thickness associated with a slice width equal to the characteristic object width.