1 Administration, Department of Computer Science, Faculty of Science, Københavns Universitet2 The Image Section, Department of Computer Science, Faculty of Science, Københavns Universitet3 Copenhagen University Hospital, Rigshospitalet4 Medizinische Universität Wien5 Administration, Department of Computer Science, Faculty of Science, Københavns Universitet
reducing image artifacts and increasing accuracy through inpainting
Aim: In combined PET/MR, attenuation correction (AC) is performed indirectly based on the available MR image information. Implant-induced susceptibility artifacts and subsequent signal voids challenge MR-based AC (MR-AC). We evaluate the accuracy of MR-AC in PET/MR in patients with metallic dental implants or braces, and propose a clinically feasible correction method. Materials and Methods: This study includes subjects selected retrospectively from our routine PET/MR referral base of patients with neurological disorders. Seven patients with metallic implants and implant-induced signal voids > 100 mL were included. In all patients simultaneous PET/MR imaging (mMR, Siemens Healthcare) of the head was performed at 40 min (n=5) and 80 min (n=2) p.i. of 200 MBq [18F]-FDG. MR-AC was performed using (A) standard Dixon water-fat segmentation (DWFS), and (B) as (A) but with the implant-induced signal voids semi-automatically filled with soft tissue (Soft). Following MR-AC, both PET emission images were reconstructed on 344x344 matrices using AW-OSEM (3iter, 21sub, 4mm Gauss). We report the volume of the inpainted area and the relative difference (Δ%) of PET(/MR)Soft to PET(/MR)DWFS for reference regions in the inpainted region, tongue and cerebellum. Furthermore, PET ratio images are computed for visual inspection. Results: The mean volume of implant-induced signal voids was (232±68) mL (min: 152mL, max: 348mL). The relative difference in mean SUV following inpainting as part of MR-AC (PET(/MR)Soft) was (217±65)% and (340±121)% in the inpainting region and tongue, respectively. Of note, comparatively large relative differences were noted also in the cerebellum: (6±3)% (max: 11%). Visual inspection of the ratio images indicated a marked regional variation of PET uptake depending on the shape and size of the signal void. Conclusion: Metallic dental work causes severe MR signal voids and PET/MR artifacts that exceed the actual implant volume. The resulting bias in AC-PET is severe in regions in and near the signal voids. Notably, the bias is present also in areas further away from the implants. In selected cases this bias may markedly affect regions used commonly as reference for kinetic modeling. Artifacts and bias can be corrected to a first degree by inpainting with soft tissue prior to MR-AC.
European Journal of Nuclear Medicine and Molecular Imaging, 2013, Vol 40, Issue Supplement 2, p. 488-489
Main Research Area:
Annual Congress of the Europen Association of Nuclear Medicine, 2013