Fig. 4
From: Machine learning prediction of future amyloid beta positivity in amyloid-negative individuals
Predicting future A\(\beta\)-positivity from multimodal data including PET and CSF baseline measures: a Bar plots showing the average AUC and average correlation score (predicting the difference between future and baseline A\(\beta\)42/global SUVR measures) across 10 computation runs for CSF-based and PET-based models, with 95% confidence intervals error bars, CSF/PET stands for CSF baseline measures (A\(\beta\)42, pTau, Tau) for predicting CSF-based A\(\beta\)-positivity and PET measures (global and regional) for predicting PET-based A\(\beta\)-positivity. b, c Scatter plot for estimation of the difference between future and baseline A\(\beta\)42 (c) and the difference between future and baseline global SUVR (d) derived by ridge linear regression (with CSF/PET measures). The results are from 1 computation run with median performance. d, e Heatmap of coefficient values across 10 runs of 10-fold CV (100 models) for CSF-based classification model (d) and PET-based regression model (e), with the bar graphs showing the importance of each predictor calculated by the mean of the absolute value of regression coefficient. The results of experiments without PET and CSF baseline measures are shown in Fig. 3