IC‐P‐114: Longitudinal ADC changes in Alzheimer's disease in a multicenter clinical trial setting

tial tool to operator-independently determine both of these quantitative parameters. In this present work we evaluated the diagnostic performance of this tool when analyzing multi-center data. Methods: A BRASS database was generated from florbetaben PET scans of 93 cognitively normal, beta-amyloid-negative healthy volunteers (HVs). Using this normal database, 145 florbetaben PET scans (77 patients with probable AD, 68 HVs) obtained from themulti-center Phase 2 trial were analyzed. For that purpose, the datasets were compared with the normal database in a voxel-based and volume of interest (VOI)-based manner. The VOI analysis resulted in composite SUVRs which were compared to those obtained by a reference technique (grey matter vs. white matter / CSF segmentation on individual MRIs, VOIs from modified AAL template). Results: The BRASS analysis of the florbetaben datasets was possible without operator-interventions within 41 6 4 sec. The composite SUVRs as determined by BRASS correlated significantly with those determined by the referencemethod (r1⁄4 0.85, p< 0.001). Also, the composite SUVRs obtained with BRASS and the reference approach discriminated equally well between ADs and HVs (p < 0.001, Cohen’s d 1⁄4 1.37 for both approaches). In the ADs and HVs, 3.2 6 2.7 vs. 0.16 0.4 (p< 0.001, Cohen’s d1⁄4 1.61) neocortical regions were defined by BRASS as pathologic (z-score> 2.5). The total brain volume affected by beta-amyloid was 18.6 6 25.7 vs. 0.8 6 3.7ml for the ADs and HVs (p < 0.001). Conclusions: The BRASS tool customized for florbetaben PET demonstrated excellent ability in discriminating between ADs and HVs, both on a regional and a voxel-based level. Thus, this software has great potential in supporting the visual interpretation of florbetaben PET image data in a rapid, user friendly and operator-independent manner.