P38
Poster Presentations: Saturday, July 23, 2016 Medicine Rostock, Rostock, Germany; 5Technische Universitaet Muenchen, Munich, Germany; 6University Hospital Klinikum rechts der Isar, Technical University of Munich, Munich, Germany; 7Institute for Stroke and Dementia Research, Klinikum der Universitaet Muenchen, Ludwig-Maximilians-University Munich, Munich, Germany; 8Department of Psychiatry and Psychotherapy, Ludwig-Maximilian-University, M€unchen, Germany; 9Department of Psychiatry and Psychotherapy, University Medical Center, Freiburg, Germany; 10Leibniz Institute for Neurobiology, Magdeburg, Germany; 11University of Rostock, Rostock, Germany. Contact e-mail:
[email protected] Background: Previous monocentric studies reported alterations of
Figure 4. Classification performance for the different network methods (at different spatial resolutions of m) in discriminating AD from CN, under a rigorous 5x2 CV evaluation method.
Figure 5. Critical difference diagram comparing the different classification methods under study. In a rigorous 532 CV evaluation method. If the two classifiers are connected by the bold line (who length is decided by a statistic called critical difference [8] (that accounts for the multiple comparisons in a non-parametric setting), they are not statistically significantly different from each other.
ter MRI scans. Cerebral Cortex, 22(7), 1530-1541. http://doi.org/ 10.1093/cercor/bhr221 5. Wee, C.-Y., Yap, P.-T., Shen, D., for the Alzheimer’s Disease Neuroimaging Initiative. (2012). Prediction of Alzheimer’s disease and mild cognitive impairment using cortical morphological patterns. Human Brain Mapping, http://doi.org/10.1002/hbm.22156 6. B. Schauerte, R. Stiefelhagen, “Learning Robust Color Name Models from Web Images”. In Proceedings of the 21st International Conference on Pattern Recognition (ICPR), Tsukuba, Japan, November 11-15, 2012. 7. Dietterich, T. G. (1998). Approximate statistical tests for comparing supervised classification learning algorithms. Neural computation, 10(7), 1895-1923.
functional connectivity in the default mode network (DMN) in patients with mild cognitive impairment (MCI) and Alzheimer’s dementia (AD) compared to healthy elderly controls. With respect to group effects in individual brain regions, those studies reported heterogeneous results due to low effect sizes and the high vulnerability of resting-state functional connectivity to noise. We pooled resting-state functional MRI data across various clinical centers to assess the influence of the distributed acquisition on the stability and robustness of group effects. Methods: We used data from N¼256 subjects obtained from the framework of the “resting-state initiative for diagnostic biomarkers” study (psymri.org). The sample included 83 patients with MCI, 54 patients with AD, and 119 healthy control subjects scanned across four German dementia centers. Functional connectivity maps were calculated using an 8 mm sphere in the posterior cingulate cortex (PCC) as seed region. We used a pooled linear model across all scans with a center covariate and additionally included age, gender, and education as covariates in the statistical model. Results: Decreased PCC functional connectivity with other DMN nodes was observed within each center for MCI/AD patients compared to healthy controls. For the pooled sample, these effects were similar, but generally of a low effect size despite the large sample size: At a liberal threshold of p<0.01, uncorrected for multiple comparisons, decreased functional connectivity with the PCC was found for several bilateral regions of the DMN in AD patients vs. controls, including clusters in the precuneus, inferior parietal cortex, lateral temporal cortex and medial prefrontal cortex. MCI subjects showed similar, but less pronounced, differences in PCC functional connectivity when compared to controls. Conclusions: Using a large multicenter dataset, our results confirm the patterns of decreased functional connectivity in AD previously reported in monocentric studies. To explicitly model multicenter variability we will use univariate analyses, such as random effects models and a voxel-based meta-analysis across centers, as well as multivariate machine learning approaches. Evaluating the stability and robustness across sites is important to assess the potential diagnostic use of resting state fMRI acquisitions for future multicenter diagnostic studies in AD.
8. Demsar, J. (2006). Statistical comparisons of classifiers over multiple data sets. The Journal of Machine Learning Research, 7, 1-30. IC-P-045
FUNCTIONAL CONNECTIVITY IN ALZHEIMER’S DEMENTIA AND MILD COGNITIVE IMPAIRMENT: A LARGE-SCALE MULTICENTER RESTING-STATE FMRI STUDY
Stefan J. Teipel1,2, Alexandra Wohlert3, Christina Heine2, Michel J. Grothe2,4, Timo Grimmer5, Christian Sorg6, Michael Ewers7, Eva Meisenzahl8, Stefan Kl€oppel9, Viola Borchardt10, Martin Walter10, Martin Dyrba2,11, 1University Medicine Rostock, Rostock, Germany; 2 German Center for Neurodegenerative Diseases (DZNE), Rostock, Germany; 3Clinic for Psychosomatics and Psychotherapeutic Medicine, Rostock, Germany; 4Department of Psychosomatic Medicine, University
Figure 1. Clusters of reduced functional connectivity of the PCC in Alzheimer’s disease dementia compared to healthy elderly controls (p<0.01, uncorrected) as revealed by a pooled analysis of multicenter resting-state fMRI data.