Figure 4.

F2 scores for TA-RNN-AE models predicting conversion to AD at the next second, third, and fourth visits ahead. (A–D) Models were tested on held-out samples in ADNI after training using two, three, five, and six visits in ADNI, respectively. (E–H) Models were tested on NACC after training using two, three, five, and six visits in ADNI, respectively. Error bars represent standard deviations. Statistical significance between TA-RNN-AE and other tools was assessed using t-test (see Supplementary Tables S13 and S14 for P-values).