Classification of bones from MR images in torso PET-MR imaging using a statistical shape model

摘要: Abstract There have been exclusive features for hybrid PET/MRI systems in comparison with its PET/CT counterpart terms of reduction radiation exposure, improved soft-tissue contrast and truly simultaneous multi-parametric imaging capabilities. However, quantitative on PET/MR is challenged by attenuation annihilation photons through their pathway. The correction photon requires the availability patient-specific map, which accounts spatial distribution coefficients biological tissues. lack information electron density MR signal poses an inherent difficulty to derivation map from images. In other words, correlates proton densities tissue relaxation properties, rather than and, as such, it not directly related coefficients. order derive images at 511 keV, various strategies proposed implemented prototype commercial systems. Segmentation-based methods generate classification T1-weighted or high resolution Dixon sequences followed assignment predefined types. Intensity-based segmentation approaches fail include bones since conventional a difficult task. Most MR-guided techniques ignore owing difficulties associated bone unless specialized such ultra-short echo (UTE) sequence are utilized. this work, we introduce new technique based statistical shape modeling segment four-class map. Our approach torso model principle component analysis (PCA). A CT-based training set including clearly segmented region 20 clinical studies was designed. Using set, atlas trained taking advantage PCA analysis. active uses user defined initial seed points. algorithm evaluated using 10 datasets (aligned CT pairs). resulting maps were compared corresponding derived mean relative difference less 7%.