The goal is to perform attenuation correction (AC) for MR-PET scanners using the background activity from LSO (Cerium-doped Lutetium Oxyorthosilicate) scintillator used in PET scanners¹, which are the most commonly used scintillator crystals in PET. To obtain quantifiable PET images AC is the most important correction. Furthermore, AC is not only required for the object being imaged (phantom or subject), but also for the MR coils normally placed inside the PET field-of-view (FoV). In MR-PET imaging, although the object being imaged can usually be visualized with MR images, the MR coils are not generally visible. Despite the low attenuation materials used for the fabrication of MR coils, it is still necessary to perform AC, especially if thorough PET image quantification is required. The AC factors depend highly on the material used and this is not precisely known for MR coils. Instead of using computed tomography (CT) or transmission scans, which require an additional separate acquisition in another scanner, intrinsic LSO radiation can be used to obtain an AC map in the same scanner if sufficiently long transmission scan times can be realised. This has the advantage of a geometric alignment of the generated AC map with the PET emission scan.In this work we performed the experiments on a Siemens 3T MR-BrainPET², which allows simultaneous acquisition of MR and PET. The BrainPET consists of 32 detector cassettes covering an axial FoV of 19.2 cm and a diameter of 31.4 cm. Each detector module is a 12×12 matrix of 2.5×2.5×20 mm3 LSO crystals coupled to a 3×3 array of avalanche photodiodes (APDs). Two types of measurements were carried out: with object (Tx/Rx 8-channel head coil or 3-rod phantom) and without object (blank scan). Figure 1 depicts the 3-rod phantom. Since natural activity of LSO due to presence of 176Lu is low (225 decays s-1cm-3), scan times must be long to minimize statistical errors. This was accomplished by concatenation of data from LSO scans during idle periods. Temperature and background activity were checked for stability. Temperature was simultaneously recorded for each detector cassette for gain stability control. Furthermore, the energy window for acquisition was widened so that all gamma rays coming from LSO would be detected (307 and 202 KeV). The detected counts were corrected for random events estimated using the delayed window events and applying a variance reduction method for noise reduction. Calculations were then performed to obtain quantities that can be reconstructed: the object scan is divided by the blank scan and the logarithm is applied (ACF=-log(I/I₀), where I is the object scan and I0 is the blank scan). Maximum likelihood expectation maximisation (MLEM) with median root prior regularization was used to reconstruct images in both line-of-response (LOR) and sinogram space. As usually done for CT, AC values obtained from LSO background need to be extrapolated from 202 and 307 keV to 511 keV (energy of gamma rays used in PET imaging). Furthermore, to assess reconstruction using different AC maps, a 3-rod phantom activity measurement was performed with and without coil. Simultaneously, two MR images were acquired: an MPRAGE and a SPACE. These MR images can be used to obtain AC maps of the 3-rod phantom through segmentation methods.Mean temperatures of all cassettes remained stable with maximum difference to reference temperature 1.7°C ± 0.1 (Figure 2 shows example of four cassettes) and only negligible effect on APD gain is expected. Data acquired in transmission measurements (64 h each) allowed for the generation of AC maps from the reconstructed image. In Figure 3, the reconstructed maps of the head coil and 3-rod phantom can be seen. Since the coil goes beyond the PET reconstructed FoV, truncation artifacts were observed (Figure 3 (a)). We also found that sinogram-based calculations outperformed LOR-based calculations, which resulted in better AC maps. This is mainly due to statistics, since in sinogram space several LORs are grouped together, thus increasing the statistics per bin (Figure 4). Obtaining AC maps for MR coils from LSO background seems feasible if the PET reconstruction FoV is adapted to the coil size or, alternatively, if AC factors for each sinogram entry are computed directly. More complex reconstruction algorithms can be used to obtain better reconstruction³. Furthermore, the use of MR surface coils, in which the exact location depends on the placement during acquisition, is also a problem that might be tackled using this approach.