Krishnan Bharath Navalpakkam1, Harald Braun2, Susanne Zeigler2, Harald H.Quick2, Joachim Hornegger1, Torsten Kuwert3
1Pattern Recognition Lab, Friedrich-Alexander-Universitt Erlangen-Nrnberg, Erlangen, Germany; 2Institute of Medical Physics, Erlangen, Germany; 3Clinic of Nuclear Medicine, Friedrich-Alexander-Universitt Erlangen-Nrnberg, Erlangen, Germany
MR-based attenuation correction for PET images is till an open issue. An accurate attenuation correction is needed for PET quantification. In this work, we generate a continuous valued CT-like attenuation map from Ultrashort Echo Time and two-point DIXON VIBE sequences using ε-insensitive Support Vector Regression (ε-SVR). Further, we compare the PET quantification accuracy using attenuation maps from a segmented CT and predicted CT with respect to a patient CT as the gold standard. Results show that a continuous estimation of attenuation values outperforms a segmentation based attenuation map.