Task‐based evaluation of axial and multiplanar reconstructions in computed tomography: A volumetric analysis of spatial resolution, noise, and detectability

Abstract

Abstract Background Multiplanar reconstructions (MPRs) are an essential clinical tool for interpreting complex 3D anatomy, offering improved diagnostic accuracy in computed tomography (CT). However, image quality can be impaired by the inappropriate use of reconstruction parameters for primary axial slices. Purpose This study aimed to characterize the fundamental aspects of image quality of axial CT slices and MPRs associated with different reconstruction parameters for three recent CT systems and to propose optimal reconstruction parameters for MPRs. Methods A dedicated cubic test‐object was designed to assess spatial resolution and noise in the primary axial slices, and in the coronal and sagittal MPRs. Two inserts with contrast levels of 120 and 1000 HU were used to measure the task‐based transfer function (TTF) in all three planes. Three‐dimensional noise power spectra (NPS) were measured in the homogeneous water volume of the cubic phantom. The TTF and NPS were used in the non‐prewhitening observer model with eye filter (NPWE) to calculate the detectability index of spheric objects with diameters from 0.5 to 5.0 mm. Three different kernels and slice thicknesses from 0.6 to 3.0 mm were tested on three recent CT systems using common iterative reconstruction algorithms. Results MPRs are generated by reslicing axial CT images rather than by direct reconstruction from raw data. The spatial resolution was significantly lower in MPRs than in axial images, especially in the longitudinal direction when thick axial slices were used. High‐resolution and edge‐enhancing kernels produced MPRs with highly anisotropic image quality and lower detectability, while standard smooth kernels provided higher isotropy and detectability. Conclusion MPR image quality is typically inferior to axial images and deteriorates further when derived from thicker axial slices. Therefore, appropriate selection of the reconstruction kernel, pixel size and slice thickness are essential to maintain diagnostic image quality.