TABLE 2.
Training dataset (OASIS) | Testing dataset (AVID) | |||
---|---|---|---|---|
ML method | R | p‐value | R | p‐value |
CL (ref) | 0.9047 | N/A | 0.9274 | N/A |
ER | 0.9862 | <1e‐31 | 0.9536 | 0.04 |
PLSR | 0.9651 | 1e‐08 | 0.9634 | 10e‐05 |
RVR | 0.9623 | 10e‐08 | 0.9728 | 2e‐06 |
ANN (3) | 0.9794 | 2e‐10 | 0.9745 | 4e‐07 |
ANN (2) | 0.9905 | <1e‐31 | 0.9847 | 8e‐08 |
Note: The Steiger test was used to compare the PiB/FBP correlation coefficient (R) between the Centiloid (as reference) and each ML method. The comparison was performed separately for the training and testing datasets.
Abbreviations: ANN(n), artificial neural network with n hidden layers; AVID, the Centiloid project dataset downloaded from the GAAIN website (http://www.gaain.org/centiloid‐project); CL (ref), Centiloid reference; ER, ensemble regression; FBP, florbetapir; ML, machine learning; OASIS, Open Access Series of Imaging Studies; PiB, Pittsburgh compound‐B; PLSR, partial least square regression; ROI, region of interest; RVR, relevance vector regression;.