Specification and estimation of sources of bias affecting neurological studies in PET/MR with an anatomical brain phantom J. Teuho a,n , J. Johansson a , J. Linden a,b , V. Saunavaara a , T. Tolvanen a , M. Teräs a a Turku PET Centre, Turku, Finland b Department of Mathematics and Statistics, University of Turku, Turku, Finland article info Available online 10 September 2013 Keywords: PET/MR Phantom Attenuation Reconstruction RAMLA abstract Selection of reconstruction parameters has an effect on the image quantification in PET, with an additional contribution from a scanner-specific attenuation correction method. For achieving comparable results in inter- and intra-center comparisons, any existing quantitative differences should be identified and compensated for. In this study, a comparison between PET, PET/CT and PET/MR is performed by using an anatomical brain phantom, to identify and measure the amount of bias caused due to differences in reconstruction and attenuation correction methods especially in PET/MR. Differences were estimated by using visual, qualitative and quantitative analysis. The qualitative analysis consisted of a line profile analysis for measuring the reproduction of anatomical structures and the contribution of the amount of iterations to image contrast. The quantitative analysis consisted of measurement and comparison of 10 anatomical VOIs, where the HRRT was considered as the reference. All scanners reproduced the main anatomical structures of the phantom adequately, although the image contrast on the PET/MR was inferior when using a default clinical brain protocol. Image contrast was improved by increasing the amount of iterations from 2 to 5 while using 33 subsets. Furthermore, a PET/MR-specific bias was detected, which resulted in underestimation of the activity values in anatomical structures closest to the skull, due to the MR-derived attenuation map that ignores the bone. Thus, further improvements for the PET/MR reconstruction and attenuation correction could be achieved by optimization of RAMLA-specific reconstruction parameters and implementation of bone to the attenuation template. & 2013 Elsevier B.V. All rights reserved. 1. Introduction In our institution, a Philips Ingenuity TF PET/MR, Siemens HRRT Brain PET (HRRT) and GE Discovery 690 (D690) PET/CT are currently used to conduct neurological studies. The transverse spatial resolution at 1 cm and 10 cm off axis is 4.7 mm and 5.0 mm for the PET/MR, 4.7 mm and 5.1 mm for the D690 and 2.3 mm and 2.5 mm for the HRRT [1,13, 20]. The coincidence timing window for the PET/MR and the HRRT is 6 ns, while the D690 has a coincidence timing window of 4.9 ns [1, 13, 20]. The timing resolution for the PET/MR is 525.0 ps and 544.3 ps for the D690 [1, 20]. The energy resolutions are 12.0% for the PET/MR and 12.4% for the D690 [1,20]. The attenuation correction method was a transmission-based attenuation correction (TXAC) in the HRRT, a CT-based attenuation correction (CTAC) in the PET/CT and a MR-based attenuation correction (MRAC) in the PET/MR. The reconstruction methods were 3D-OP-OSEM for the HRRT, 3D-OSEM for the PET/CT and LOR-RAMLA for the PET/MR. Therefore, the PET/MR employs a different methodology for both attenuation correction and recon- struction. If any scanner-specific biases exist due to these differ- ences, they should be specified and minimized to reduce any scanner-specific bias on neurological studies. As MRAC is a measure of tissue proton density contrary to electron density measured by TXAC and CTAC, the attenuation map has to be derived by image segmentation and specification of discrete tissue classes [10,14,19,21]. In clinical use, a uniform attenuation value of soft tissue class is assigned for the entire head contour. Attenuation map with a single value ignores tissue heterogeneity and the attenuation effect of the skull, which could contribute to additional bias. The bias introduced from ignoring the attenuation effect of the skull has been reported on previous studies, although the exact magnitude of the effect has been reported to vary [4,9,18]. Furthermore, as LOR-RAMLA differs from the traditionally used 3D-OSEM-derivates in terms of image filtering and reconstruction parameter selection [5,6,11], its effect on the visual quality and also quantitative values should be determined. The reconstruction parameters in intra- and inter-center studies should be selected to minimize any significant differences [2,3]. For this purpose, standardization programs such as EARL FDG-PET/CT accreditation Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/nima Nuclear Instruments and Methods in Physics Research A 0168-9002/$ - see front matter & 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.nima.2013.09.002 n Corresponding author. Tel.: þ358 407741366. E-mail address: jarmo.teuho@tyks.fi (J. Teuho). Nuclear Instruments and Methods in Physics Research A 734 (2014) 179–184