One of the known potential benefits of integrated positron emission and magnetic resonance (PET/MR) scanners is the possibility of using MR data to monitor patient motion. However, motion monitoring in currently available MR sequences is accessory to their imaging function. Even when MR imaging is not required, the existing sequences with motion monitoring capability will not lend themselves to run uninterrupted for long periods of time, as often required by PET studies. Problems such as the generation of large amounts of imaging data, unnecessary radio-frequency exposure and patient discomfort due to continuous noise conditions will have to be dealt with. The present work evaluates the performance of a new pulse sequence designed exclusively for PET/MR head tracking. By assuming that all the required imaging data is provided by the PET subsystem, the MR sequence can be optimized exclusively for the task of providing accurate motion estimates over long periods of time (e.g. 30-60min). Data management, SAR and noise concerns are therefore minimized.The proposed sequence borrows on the Prospective Motion Correction approach developed by White et al1. This approach uses orthogonal spiral navigator acquisitions (TE 1ms, TR 16ms, FA 8deg, BW 125kHz, FOV 32cm, matrix 2048x1, image size 128x128, ST 13mm) as the input of an extended Kalman filter algorithm in charge of real-time motion estimation2. Contrary to the typical implementation of sequences based on this correction method, where the navigator pulses are integrated in a 3D SPGR or FSE framework, the proposed sequence includes no imaging generation pulses, instead providing the user with full control over the navigator timing properties. In-vivo measurements of the motion estimation performance were carried out on a SIGNA PET/MR system (GE Healthcare). These included: long-term stability measurements, both under static and piecewise static conditions; and accuracy measurements, using as reference 3D FSPGR volumes (TE 1.2ms, TR 5.5ms, FA 15deg, BW 31.25kHz, FOV 24cm, matrix 120x120, pixel size 2x2mm, ST 2mm) acquired immediately before and after each motion estimation sequence. The Integrated Registration software (Advantage Workstation, GE Healthcare) was used to compute the rigid transformation between these volumes.The accuracy measurements show good agreement between the motion estimates provided by the sequence and the ground truth obtained by registering the FSPGR datasets. The measured differences between these methods are summarized in Table 1. The motion estimation algorithm was found to diverge in two series where a sudden large motion caused the head tracking to lose lock. These series were excluded from further analysis. Short-term stability measurements on the static intervals before and after patient motion show standard deviations below 0.2mm for the translation parameters and below 0.3deg for the rotation parameters. Long-term stability measurements yielded similar results after de-trending, with standard deviations below 0.6mm for the translation parameters and below 0.5° for the rotation parameters.The obtained accuracy and stability values fall well within the acceptable range for PET motion correction (the intrinsic spatial resolution of clinical PET systems being typically >4mm). Trends and oscillatory patterns were patient-dependent, which suggests they are not caused by the motion tracking algorithm. Long acquisition times were possible without any kind of data management issues. SAR values were negligible throughout the examination. Ongoing work is aimed at increasing the lock and re-capture range of the tracking algorithm. While this was not a relevant issue in the MR imaging scenarios for which the algorithm was originally developed, sudden large movement is likely to occur during long PET acquisitions.The results obtained in this study indicate that the proposed non-imaging sequence is suitable for MR-driven PET motion correction in hybrid scanners. Further work is underway to improve the tracking range.