Abstract
In this study we propose a method to generate probability maps for material identification and quantification in spectral computed tomography (CT). Our goal in producing these maps is to provide spectral CT users, such as clinicians and researchers, per-voxel confidence levels on the materials and concentrations identified, especially for complex biological specimens with unknown materials and unknown concentrations. Our method is based on a likelihood ratio approach that modifies prior (pre-test) probabilities to produce posterior (post-test) probabilities. Prior probabilities are calculated from the per-voxel material decomposition error conditioned on non-negative quantifications. To evaluate our method, we scanned a calibration phantom containing vials of water, lipid, hydroxyapatite (HA) and gadolinium (Gd) in various concentrations using a MARS Spectral Photon Counting CT scanner. In addition, for a validation dataset we also scanned a bovine patella sample incubated in Gd contrast agent solution with the same scanner and scan protocols. Material decomposition was performed on both datasets and likelihood ratios were calculated from the calibration phantom for each material at their respective concentrations. Posterior probability maps for HA and Gd were generated for both datasets using the calculated likelihood ratios on the prior probabilities. The results show that these maps provide additional information on the degrees of confidence for each voxel’s material identification and quantification. We also demonstrate that our method can be applied to a complex biological sample.