a potential approach for the treatment of AD and a BACE radioligand could be valuable for assessing target occupancy and therapeutic responses to BACE in- hibition. Methods: Mdr1a/b (P-gp 3 and 1 deficient, termed P-gp KO), FVB wild type (wt), BACE1 knockout (KO), and BACE1 wt mice, were purchased from Taconic. Mdr1a/b and FVB WT mice were 8 weeks old, while BACE1 KO and BACE1 WT mice were two years old when used in the studies. The unla- beled compounds and [3 H]Compound-A (75.3Ci/mmol) were synthesized in house. Autoradiography was performed using frozen brain slices. In vitro bind- ing assays were done with brain membranes or homogenates. In vivo brain oc- cupancy was performed in Mdr1a/b, FVB and BACE1 KO mice by i.p. administration of BACE inhibitor or vehicle and then followed by i.v. dosing of the radioligand. Results: In mouse brain homogenates, [3 H]Compound-A showed binding site densities (B max) from 6.5 nM to 8.4 nM and high binding affinity (K d ¼ 0.5 nM), yielding good binding potentials (B max/K d). In auto- radiography of BACE1 wt mice, [3 H]Compound-A binding sites were widely distributed across the brain with different binding densities in various regions, and the non-displaceable binding was low and homogenous among all the brain slices. In contrast, [3 H]Compound-A failed to show specific binding in the brain slices of BACE1 KO mice. Furthermore, the in vivo brain occupancy studies demonstrated that [3 H]Compound-A binding to Mdr1a/b brains was dose-dependent and displaceable by pre-administration of a selective, brain penetrant BACE inhibitor. In FVB mouse brains, [3 H]Compound-A did not show specific brain uptake following i.v. administration of [3 H]Compound- A. Conclusions: These results demonstrate that [3 H]Compound-A binding is specific to BACE which is widely distributed in mouse brain. In vivo brain uptake of [3 H]-Compound-A only occurred in P-gp KO mice, not in wild type and BACE1 KO mice. Together, these data suggest that a compound with no P-gp liability could be developed into a PET tracer for BACE. IC-P-196 ADVANCING THE ACCURACY OF AUTOMATED FDG-PET MEASUREMENTS USING HIGH- DIMENSIONAL IMAGE NORMALIZATION Michel J. Grothe 1 , Marina Boccardi 2 , Martina Bocchetta 3 , Giovanni Frisoni 4 , Stefan Teipel 5 , 1 German Center for Neurodegenerative Diseases (DZNE), Rostock, Germany; 2 IRCCS S.Giovanni di Dio - Fatebenefratelli, Brescia, Italy; 3 IRCCS Centro S. Giovanni di Dio, Fatebenefratelli, Brescia, Italy; 4 IRCCS, Centro San Giovanni di Dio, Fatebenefratelli, Brescia, Italy; 5 University Medicine Rostock and DZNE Rostock, Rostock, Germany. Contact e-mail: michel.grothe@dzne.de Background: Fluorodeoxyglucose (FDG)-PET imaging is among the most important imaging modalities for in-vivo research on Alzheimer’s disease. The gold standard for regional quantification of PET signal relies on manual delineation of anatomical regions-of-interest (ROIs) on coregistered high-res- olution structural MRI scans of the same subject. Analysis of large-scale im- aging datasets has been made possible through the development of automated image analysis techniques employing spatial normalization of the PET images to a standardized reference space with associated atlases of anatomical ROIs. Novel high-dimensional image warping algorithms for MRI allow matching individual images to a given reference template with much higher accuracy compared to earlier low-dimensional implementations. The potential benefits of these novel normalization procedures for the accuracy of automated FDG- PET measurements have not yet been investigated. Methods: Hand-drawn hippocampus labels based on a harmonized delineation protocol were recently published for a subset of 100 subjects from the Alzheimer’s Disease Neuroi- maging Initiative. Coincident FDG-PET scans were available for 38 of these subjects and were rigidly coregistered to their corresponding MRIs. Quantifi- cation of hippocampal tracer-uptake was carried out using SPM8 software and the VBM8-toolbox and followed 4 different analysis approaches: (i) the "gold standard" of direct signal extraction within individual hippocampus labels in native space, as well as automated extraction within harmonized hippocampus labels in standard reference space using (ii) direct low-dimensional normali- zation of the PET scans to SPM’s default H2O-PET template ("standard direct"), (iii) indirect normalization using parameters from standard low- dimensional normalization of the coregistered MRI ("standard indirect"), and (iii) indirect normalization using parameters from high-dimensional normalization of the coregistered MRI ("DARTEL"). Results: Automatically estimated values for hippocampal FDG-uptake using "DARTEL" were highly correlated with the values extracted using the "gold standard" method (R 2 (left/ right) ¼ 0.94/0.93), with a mean percentage divergence below 3%. The cor- relations were significantly higher compared to those using "standard direct" (R 2 (left/right) ¼ 0.85/0.87, p(difference) < 0.001/0.05) and "standard indi- rect" (R 2 (left/right) ¼ 0.91/0.84, p(difference) < 0.01/0.0001) normalization strategies. Conclusions: Even though PET images are generally limited by relatively low spatial resolution, novel high-dimensional image normalization techniques may still significantly improve the accuracy of automated regional signal measurements compared to more traditional normalization strategies. IC-P-197 MICROHEMORRHAGE FINDINGS ON BASELINE ROSAS MRI DATA Hubert Basselerie 1 , Luc Bracoud 2 , Hans-Martin Schneble 3 , Sylvain Gouttard 2 , Isabelle Guignot 3 , Audrey Istace 2 , Florent Roche 2 , Joel Schaerer 2 , Maria Pueyo 3 , Chahin Pachai 2 , Bruno Vellas 1 , Fabrice Bonneville 1 , 1 CHU Casselardit, Toulouse, France; 2 BioClinica, Lyon, France; 3 IRIS, Suresnes, France. Contact e-mail: luc.bracoud@ bioclinica.com Background: Cerebral microhemorrhages, although common among the elderly, are known to be related to the presence of amyloid in the brain, cogni- tive impairment and dementia severity. This work assesses the prevalence of microhemorrhages in ROSAS, a monocentric observational study which in- cludes normal controls as well as MCI and AD subjects. Methods: 408 sub- jects 65 years and older were enrolled, including 110 Normal Controls (NC, no memory complaints, MMSE26 and CDR¼0), 100 Mild Cognitive Impairment (MCI, MMSE24 and CDR¼0.5, memory impairment based on RAVLT and that did not meet DSM IV criteria for AD dementia) and 198 AD (12MMSE26 and CDR0.5 and meeting DSM IV criteria). Sus- ceptibility-Weighted (SWI) MRI scans were collected up to 3 times between Baseline and Month 48, at one site using a Philips Achieva 3T scanner, for consenting subjects (n¼153, including 51 NC, 41 MCI and 61 AD). Baseline SWI data were reviewed by one rater, looking at native 0.5 mm thick images along with minimum Intensity Projection (minIP) reconstructed slices (6 mm thick, with 3 mm overlap). Bleeds were categorized as deep, lobar and infra- tentorial, and further subdivided as <5 mm, between 5 and 10 mm (so-called large microhemorrhages), and >10 mm in size. Descriptive statistics were computed for each group. Pearson Chi-Square tests were used to assess differ- ences between groups. Results: Microhemorrhages (10mm) were reported on 56.9% NC, 68.3% MCI and 65.6% AD. Only 2.0% NC had more than 4 microhemorrhages, as compared to 22.0% MCI and 11.5% AD. Significant differences between groups were found for the presence of intratentorial mi- crohemorrhages (p¼0.035) and of >4 microhemorrhages (p¼0.01), and be- tween NC and non-NC for the presence of large microhemorrhages (p¼0.045) and >4 lobar microhemorrhages (p¼0.029). Bleeds >10 mm were found on only 2 MCI and 2 AD. Conclusions: A majority of ROSAS subjects from all clinical categories had >1 microhemorrhage. Multiple Table 1 Proportion of microhemorrhages by group, classified by number # Microhemorrhages (10 mm) NC (%) MCI (%) AD (%) 1 56.9 68.3 65.6 0 43.1 31.7 34.4 [1-4] 54.9 46.3 54.1 >4 2.0 22.0 11.5 >4 lobar 2.0 17.1 9.8 Table 2 Proportion of microhemorrhages by group, classified by location Microhemorrhage location NC (%) MCI (%) AD (%) Deep 9.8 24.4 18.0 Lobar 54.9 63.4 57.4 Infratentorial 7.8 26.8 13.1 Poster Presentations: IC-P P109