Table 1.
Main accuracy metrics of the biological age estimates described here.
| Tissue type | Input features | ML algorithm | N (training:test) | Population | Pearson r | R2 | MAE (years) | Reference |
|---|---|---|---|---|---|---|---|---|
| Blood | 41 haemochrome markers | DNN | 62,419 (90:10) | Russian | 0.9 | 0.8 | 6.07 | (20) |
| 19 haemochrome markers | DNN | 65,760 (80:20) | South Korean | 0.7 | 0.49 | 5.59 | (2) | |
| 55,920 (80:20) | Eastern European | 0.84 | 0.69 | 6.25 | ||||
| 20,699 (80:20) | Canadian | 0.7 | 0.52 | 6.36 | ||||
| 142,379 (80:20) | All | 0.8 | 0.65 | 5.94 | ||||
| Brain | Structural MRI (normalized GM volumes) | GPR | 2,001 (90:10) | different ancestries | 0.95 | 0.89 | 4.66 | (23) |
| Structural MRI (normalized WM volumes) | 0.92 | 0.84 | 5.88 | |||||
| Structural MRI (normalized GM + WM volumes) | 0.96 | 0.91 | 4.41 | |||||
| Structural MRI (raw data) | 0.57 | 0.32 | 11.81 | |||||
| Structural MRI (normalized GM volumes) | CNN | 0.96 | 0.92 | 4.16 | ||||
| Structural MRI (normalized WM volumes) | 0.94 | 0.88 | 5.14 | |||||
| Structural MRI (normalized GM + WM volumes) | 0.96 | 0.91 | 4.34 | |||||
| Structural MRI (raw data) | 0.94 | 0.88 | 4.65 | |||||
| Structural MRI (normalized GM + WM volumes) | GPR | 2,001 (80:10:10) | different ancestries | 0.94 | 0.88 | 5.02 | (5) |
ML, Machine Learning; MAE, Mean Absolute Error; DNN, Deep Neural Networks; GPR, Gaussian Process Regressions; CNN, Convolutional Neural Networks; MRI, Magnetic Resonance Imaging; GM/WM, gray/white matter.