MRI-based pseudo-CT generation using sorted atlas images in whole-body PET/MRI

In this work, we propose a novel approach for MRI-based generation of pseudo-CT images in whole-body PET/MRI based on Hofmann's pattern recognition and atlas registration approach. The major improvement emanates from sorting registered atlas images based on voxelwise local normalized cross-correlation and choosing the most similar atlas image for Gaussian process regression (GPR) analysis. Furthermore, prior knowledge derived from the correlation between lung volume and attenuation coefficients was embedded in the GPR kernel for accurate patient-specific prediction of lung attenuation coefficients. Modifying the GPR algorithm improved the similarity index of bone extraction from 0.55 to 0.61 and enabled significant bias reduction of tracer uptake (SUV) in bony regions. Incorporating prior knowledge about lung volume in the GPR algorithm resulted in SUVmean bias reduction from 8.9% to 4.1% in the whole lung region. Overall, the proposed algorithm provided more accurate PET quantification in the lungs and bony regions.