Traumatic Brain Injury (TBI) is regarded as “the most complex disease in our most complex organ”[1]. Clinical outcome is unpredictable, especially in repetitive mild TBI, in terms of behavior, cognition, emotion and associated long-term effects such as dementia etc. Today’s workhorse in diagnosing TBI is CT, which is widely available, fast and cost-effective. However, CT may not be the optimal diagnostic tool in TBI since the full extent of structural abnormalities is not detected and there may be a mismatch between CT findings and neurological examination. For instance, traumatic axonal injury may remain undetected but can lead to significant behavioral or cognitive problems. Dedicated MR examination is likely more sensitive to TBI, though there is no consensus about the the most effective MR protocol. The Collaborative European NeuroTrauma Effectiveness Research in TBI (CENTER-TBI) study [1] is a pan-European prospective longitudinal observational study aiming to improve care for TBI patients. One of the key goals is to improve the quality of imaging-derived data by the application of a standardized MR imaging protocol across up to 25 clinical sites in a large, heterogeneous sample of TBI patients. To be able to make inferences about disease progression and outcome, as shown in figure 1, MR data, MR acquisition and MR quality needs to be consistent across sites.A 3T MR protocol consisting of 3D T2, 3D FLAIR, 3D T1, SWI, DTI and rs-fMRI is devised [2] for assessment of structure and function of the injured brain as described in figure 1. The protocol needs to be fit to be used in a clinical setting, with a scan time limit of around 45 minutes, using clinically available protocols. Harmonization is achieved on three scanner models of the main vendors with vendor assistance. The protocol is distributed electronically to participating sites and adapted to the local scanner configuration. For quality control, DWI [3] and DTI phantoms [4], [5] are sent to the sites along with a dedicated scan protocol. A site can start patient enrollment after central reading and approval of a healthy volunteer dataset. Patient data is centrally collected and inspected visually for quality by an imaging contract research organization.Protocol implementation was challenging and time-consuming because of large heterogeneities in scanner hard- and software configuration, see table 1. Full protocol harmonization could not be achieved, especially for T1, DTI and SWI because of differences in sequence implementation between the 3 main vendors, scanner model, differences in software versions and licenses. Scan time for the entire protocol is variable between scanners, ranging from 42-55 minutes. Data is uploaded to a central XNAT (www.xnat.org) server: https://neuro-imaging.center-tbi.eu. Visual quality control is performed on 265 patient datasets from 11 sites and classified into a) interpretable for a radiologist, b) usable for automated analyses or c) unusable. For T2, 3.4% of scans were classified as unusable, 92.1% interpretable, 4.5% was not acquired. For T1: 1.9/95.8/2.3%, FLAIR: 4.1/92.5/3.4%, SWI: 3.4,92.1,4.5%, DTI (unusable/usable/no data): 29.6/56.6/13.8%, fMRI:13.6/79.2/7.2%. Figures 2 and 3 show the quality metrics. None of the datasets was completely artifact-free. Site performance is variable, minimum performance is 87.5% (of 40 patients in that site) usable anatomical data, 9.1% (of 11) usable DTI data, 0% (of 11) usable fMRI data. Fig 3 shows quality differences across vendors for the T1, T2, FLAIR and SWI data.Protocol harmonization is a challenging task because of lack of standardization of sequences across vendors. Our experience is that within a vendor, sequences are not standardized between software releases. Secondly, best data quality can not be realized because compromises need to made in order to obtain matching protocols between scanners and to achieve protocol that works in a clinical setting. This aim has been achieved in other studies (e.g. [6]) which involve small numbers of research active sites. The solutions achieved in these settings may not generalizable to larger studies which involve clinically busy sites that do not have an active imaging research program. Regardless of successful harmonization, it is essential to establish protocols that work in a clinical setting. Our experiences tell us that TBI patients can be un-cooperative so the long scan time results in decreased data quality, especially for DTI.In order for such large-scale observational studies to be really effective, sequence harmonization and standardization is of key importance, but lacking at the moment. For traumatic brain injury patients, where patients may be confused or disoriented, an effort should be made to reduce scan time to avoid data corruption due to motion.