TABLE 4.
The accuracies and AUROCs of five ML algorithms for each static and dynamic rsfMRI measures, divided into two atlases.
|
ALFF |
fALFF |
FC |
REHO |
Concatenated |
||||||
| AAL 116 Static | With PCA | Without PCA | With PCA | Without PCA | With PCA | Without PCA | With PCA | Without PCA | With PCA | Without PCA |
| GB | 55%/0.6 | 55%/0.62 | 65%/0.74* | 40%/0.25 | 65%/0.65 | 55%/0.47 | 45%/0.46 | 45%/0.43 | 55%/0.62 | 55%/0.49 |
| LASSO | 55%/0.41 | 50%/0.41 | 55%/0.54 | 65%/0.61 | 50%/0.58 | 50%/0.65 | 65%/0.54 | 50%/0.6 | 50%/0.27 | 45%/0.55 |
| LR | 60%/0.55 | 45%/0.5 | 35%/0.29 | 35%/0.34 | 35%/0.34 | 35%/0.32 | 35%/0.36 | 35%/0.32 | 45%/0.43 | 35%/0.29 |
| RF | 45%/0.57 | 50%/0.55 | 65%/0.66 | 30%/0.34 | 50%/0.43 | 60%/0.44 | 30%/0.32 | 25%/0.24 | 55%/0.67 | 50%/0.36 |
| SVM | 50%/0.46 | 50%/0.42 | 50%/0.55 | 40%/0.29 | 50%/0.14 | 40%/0.37 | 50%/0.38 | 30%/0.34 | 45%/0.38 | 60%/0.48 |
| AAL 116 Dynamic | ||||||||||
| GB | 45%/0.47 | 60%/0.5 | 40%/0.42 | 50%/0.5 | – | – | 60%/0.5 | 40%/0.29 | – | – |
| LASSO | 40%/0.45 | 50%/0.39 | 60%/0.57 | 50%/0.45 | – | – | 40%/0.43 | 55%/0.49 | – | – |
| LR | 50%/0.65 | 65%/0.75* | 60%/0.56 | 35%/0.39 | 35%/0.27 | – | 50%/0.51 | 35%/0.41 | 45%/0.44 | – |
| RF | 45%/0.49 | 50%/0.57 | 35%/0.48 | 45%/0.47 | 50%/0.43 | – | 35%/0.38 | 40%/0.41 | 70%/0.6 | – |
| SVM | 45%/0.46 | 45%/0.42 | 40%/0.42 | 40%/0.41 | – | – | 45%/0.32 | 55%/0.39 | – | – |
| Power 264 Static | ||||||||||
| GB | 60%/0.55 | 40%/0.5 | 40%/0.32 | 55%/0.56 | 45%/0.67 | 50%/0.5 | 40%/0.4 | 30%/0.33 | 50%/0.62 | 50%/0.5 |
| LASSO | 45%/0.42 | 25%/0.42 | 70%/0.65 | 55%/0.31 | 50%/0.41 | 70%/0.76 | 35%/0.34 | 35%/0.36 | 45%/0.51 | 50%/0.41 |
| LR | 55%/0.61 | 60%/0.48 | 45%/0.51 | 40%/0.44 | 45%/0.47 | 45%/0.43 | 50%/0.51 | 40%/0.45 | 45%/0.48 | 45%/0.4 |
| RF | 60%/0.65 | 40%/0.41 | 45%/0.47 | 55%/0.46 | 60%/0.69 | 55%/0.38 | 55%/0.68 | 70%/0.57 | 55%/0.48 | 50%/0.49 |
| SVM | 50%/0.34 | 50%/0.39 | 50%/0.45 | 50%/0.33 | 50%/0.68 | 50%/0.46 | 45%/0.24 | 40%/0.43 | 50%/0.69 | 50%/0.46 |
| Power 264 Dynamic | ||||||||||
| GB | 55%/0.58 | 45%/0.6 | 50%/0.5 | 60%/0.57 | – | – | 55%/0.6 | 55%/0.64 | – | – |
| LASSO | 60%/0.5 | 55%/0.65 | 45%/0.62 | 45%/0.38 | – | – | 45%/0.58 | 45%/0.5 | – | – |
| LR | 60%/0.57 | 70%/0.66 | 55%/0.51 | 50%/0.44 | 50%/0.46 | – | 50%/0.47 | 50%/0.5 | 50%/0.48 | – |
| RF | 45%/0.44 | 55%/0.53 | 50%/0.34 | 50%/0.48 | 50%/0.41 | – | 40%/0.61 | 60%/0.55 | 50%/0.49 | – |
| SVM | 45%/0.46 | 45%/0.6 | 40%/0.47 | 55%/0.42 | – | – | 50%/0.55 | 50%/0.51 | – | – |
GB, gradient boosting; LASSO, least absolute shrinkage and selection operator; LR, logistic regression; RF, random forest; SVM, support vector machine. Significant results (p < 0.05), estimated using the permutation method, are marked with an asterisk and bolded. The binary classification was conducted using NSR and SR groups.