Table 1.
Calculated MAE for each cohort for each of the trained models.
| Model | MAE per cohort (years) | |||||
|---|---|---|---|---|---|---|
| ADNI | AIBL | GENIC | UK Biobank | AddNeuroMed | J-ADNI | |
| CNN11 | 2.56 | 2.85 | 4.20 | 2.75 | 4.03 | 5.63 |
| CNN22 | 2.40 | 2.69 | 3.96 | 2.66 | 3.67 | 5.30 |
| CNN32 | 2.41 | 2.69 | 3.98 | 2.67 | 3.02 | 2.25 |
| CNN41 | 2.98 | 3.33 | 4.42 | 3.03 | 3.64 | 4.14 |
The numbers in italics (AddNeuroMed and J-ADNI cohorts for models CNN1, 2, and 4) were calculated as external datasets for the model. 1Statistical differences were found (value of p < 0.05) comparing the CNN absolute error; 2No statistical differences were found in the comparison between the absolute error of CNN2 and CNN3 (value of p = 0.37). Statistical differences between the models were calculated using Friedman’s test with Conover’s post hoc test, adjusting for multiple comparisons with the Bonferroni method.