Fig 1.
The longitudinal MRI recordings (orig.mgz) and the corresponding Freesurfer segmentations (aseg.mgz) from one of the participants at each of the three study waves.
The age at the MRI examinations and corresponding left and right lateral ventricle volumes are given along the time-line.
Fig 2.
Illustration (left hemisphere) of the subject-specific measures (b1iL, b1iR, VdevL, VdevR) of LVV trajectories obtained from the LME analysis.
Table 1.
Definitions and characteristics of fast, medium, and slow performers.
Fig 3.
Subject-specific longitudinal lateral ventricle volumes versus age in left (a) and right (b) hemisphere shown as color-coded spaghetti plots across the three study waves. For left and right hemisphere the random effects, estimated from the linear mixed-effect model Volij = β0 + β1 Ageij + (b0i + b1i Ageij) + ϵij, are depicted as thin line segments in black superimposed on the color-coded line plots. The thick regression line in black represents the estimated fixed effect, and the broken line represents ordinary linear least squares regression (OLS) line. Subject-specific longitudinal eTIV-normalized lateral ventricle volumes versus age in left (c) and right (d) hemisphere, respectively, are shown as color-coded spaghetti plots across the three study waves. Here, a linear mixed-effect model was applied and fitted to the eTIV-normalized data.
Fig 4.
Generalized pairs plot depicting the kernel density estimated empirical distributions of each of the six variables and Pearson correlations between age, the four LVV trajectory measures and response inhibition.
(a) Non-normalized LVVs. (b) eTIV-normalized LVVs. The graphs and correlations are given separately for females (in red) and males (in green). Age3 = age of participant at study wave 3; b1iL = LVV steepness measure, left hemisphere; b1iR = LVV steepness measure, right hemisphere; VdevL = LVV deviance measure, left hemisphere; VdevR = LVV deviance measure, right hemisphere; RI3 = response inhibition reaction time at study wave 3.
Fig 5.
Result from the simulation experiments assessing the significance of a 5-fold cross-validated score (f1) with 500 permutations using multinomial logistic regression.
The predictors are X = {b1iL, b1iR, VdevL, VdevR} and the classes are the three levels of RI reaction times, y = {slow, medium, fast}.
Table 2.
Predictions from each split of cross-validation, generating cross-validated estimates for each input data point using multinomial logistic regression.
Fig 6.
Plots showing the observed RI labels (leftmost two panels, for left (a) and right hemisphere (c), respectively) and the predicted RI labels (rightmost two panels, for left (b) and right hemisphere (d), respectively) for each of the 74 subjects in the cohort. When a given trajectory in (a) or (c) changes its color as it occur in (b) or (d), that subject is misclassified; otherwise he or she is correctly classified with respect to RI performance.
Fig 7.
The 3 × 3 confusion matrix computed for the slow, medium and fast RI labels returned from the cross validation prediction with our multinomial logistic regression model compared with the co-occurrences of the true (observed) RI labels.
The diagonal cells are those representing correctly classified subjects (number of occurrences in each cells are given as N), and these cells are shaded in blue. Off-diagonal cells represents various events of misclassification. Observed/predicted co-occurrences are also accompanied, for each cell, with corresponding information about gender ratio (F/M), confirmed age at inclusion larger than 65 years (Age1 > 65), and volume means in microliters of left and right lateral ventricle (Vol1L and Vol1R), respectively, at time of subject inclusion in the study.