Fig. 8.
Gaussian process model based z-scores of global and local volumes in the ADNI test sample. (A) First column shows z-scores of predictive probabilities δij of hippocampal voxel GMV at (24, − 12, − 18) mm MNI of 415 scans from study participants with ages 55–93 years. Separate boxplots for 100 NO, 95 stable MCI (sMCI), 92 progressive MCI (pMCI), and 128 AD subjects subgroups are shown. Columns 2–4 depict z-scores for the GP model of global volumes. (B) Potential of GP-based local and global volume z-scores and comparison to t-test based single case inference. The left column shows a characterization of the pMCI and NO (top) and AD and NO (bottom) subjects in 2 dimensional plot using z-scores from a single hippocampus voxel GMV and total gray matter volume respectively. Additional contours show predictive probabilities obtained by post hoc Gaussian process classification of the clinical vs. normal subjects using a squared exponential covariance and cumulative Gaussian likelihood function. Middle column shows receiver operating characteristic (ROC) curves of classification of pMCI vs. NO (top) and AD vs. NO (bottom) using a 2D Gaussian process classification with leave one out cross-validation. Colors indicate 2D classification based on local gray matter volume with tgmv (red), twmv (blue), and tcsf (green) respectively. The right column shows the same group classification based on 1d Gaussian process classification of uncorrected t-values (red), t-values after age correction using general linear model (GLM) (blue), and t-values after GLM based correction of all covariate effects.