Healthy-brain connectivity networks were recomputed using this new atlas for the purpose of seeding tracts. In order to perform statistically rigorous hypothesis testing, we adopted a simple correlation approach. The t-statistic of atrophy within each disease group and for all cortical BMN 673 in vitro ROIs was correlated with the absolute values of all hypothesized eigenmodes, and the R2 and p values of Pearson correlation coefficients were calculated. The statistical atrophy of each disease was plotted against each persistent mode. The prevalence rates of various dementias
were collected from literature survey. Unfortunately, prevalence estimates vary wildly among sources, age groups, and ethnicity, especially at low prevalence rates in younger populations. We grouped studies into decadal age ranges from 50 to 90+ and restricted ourselves to studies in advanced (OECD) nations. For each age range, we computed prevalence rate as a percentage of each dementia over prevalence of ALL dementias. These data were taken from the following studies: (Harvey, 2003, Ratnavalli et al., 2002, Kobayashi et al., 2009, Jellinger and Attems,
2010, Kukull et al., 2002 and Morrison, 2010; Di Carlo et al., 2002 and Plassman et al., 2007). To this published data we compared the theoretical prevalence RAD001 that would be predicted by our model, as described in the subsection titled Development of a Network Diffusion Model. Since the model has two parameters (age of onset and diffusivity constant β) whose true
values cannot be uniquely determined from the literature, we estimated them by fitting the model to published data using a simple minimization routine. Finally we wish to determine whether the most persistent eigenmodes second have utility for the purpose of diagnosing and classifying various dementias. Atrophy of each subject in the aged groups was normalized using the young healthy subjects, giving a z-score, zk , for the k th subject. We computed the dot product between zk and the n th eigenmode, giving d(k,n)=unTzk. In order to remove the effect of different overall extent of atrophy in different dementias, this figure was normalized to d¯(k,n) such that ∑nd¯(k,n)=1. The latter values were fed into a three-way (normal aging, AD, bvFTD) linear discriminate analysis (LDA) classifier. ROC curves were obtained after repeated leave-one-out analysis whereby each subject was classified based on training over all the other subjects. For comparison, we also implemented a conventional classifier based directly on atrophy z-scores, zk, after dimensionality reduction using PCA. This research was supported in part by the following grants from the National Institutes of Health: R01 NS075425, F32 EB012404-01, P41 RR023953-02, P41 RR023953-02S1, and R21 EB008138-02. Author Contributions: A.R. conceptualized this study and developed the mathematical model, performed all correlations and statistical tests, and wrote the manuscript.