Table 2.
Top three models selected to fit for PH sub-group (PH subjects with preserved ejection fraction versus controls).
| Slices | Model | Feature selection | Mean | SD | Median | Min | Max |
|---|---|---|---|---|---|---|---|
| Original | |||||||
| RV mask | nnet | Lincomb | 0.885 | 0.097 | 0.906 | 0.688 | 1.000 |
| Combined | rf | Full | 0.878 | 0.092 | 0.906 | 0.609 | 1.000 |
| RV mask | ridge | Lincomb | 0.876 | 0.107 | 0.906 | 0.531 | 1.000 |
| First side study—interclass correlation with first two extractions (ICC2) | |||||||
| Combined mask | gbrm | Full | 0.808 | 0.151 | 0.844 | 0.313 | 1.000 |
| Combined mask | rf | Full | 0.798 | 0.118 | 0.797 | 0.594 | 1.000 |
| RV mask | rf | Full | 0.794 | 0.140 | 0.813 | 0.375 | 1.000 |
| Second side study—interclass correlation with all three extractions (ICC3) | |||||||
| Combined mask | nnet | Full | 0.815 | 0.119 | 0.813 | 0.563 | 1.000 |
| Combined mask | mlp | Full | 0.800 | 0.144 | 0.844 | 0.500 | 1.000 |
| Combined mask | lasso | Full | 0.785 | 0.161 | 0.750 | 0.500 | 1.000 |
| DAFIT without filtering | |||||||
| Combined mask | svmPoly | Full | 0.957 | 0.039 | 0.969 | 0.859 | 1.000 |
| Combined mask | svmPoly | pca | 0.947 | 0.036 | 0.945 | 0.891 | 1.000 |
| Combined mask | svmRad | Full | 0.926 | 0.043 | 0.930 | 0.836 | 0.984 |
| DAFIT with filtering ICC2 (DAFIT Filt2) | |||||||
| Combined mask | svmPoly | Corr | 0.908 | 0.095 | 0.930 | 0.617 | 1.000 |
| Combined mask | svmPoly | Full | 0.903 | 0.098 | 0.914 | 0.609 | 1.000 |
| Combined mask | svmRad | Full | 0.890 | 0.088 | 0.906 | 0.617 | 1.000 |
| DAFIT with filtering ICC 3 (DAFIT Filt3) | |||||||
| Combined mask | svmRad | Full | 0.887 | 0.100 | 0.906 | 0.656 | 0.992 |
| Combined mask | svmRad | Corr | 0.881 | 0.089 | 0.906 | 0.648 | 1.000 |
| Combined mask | linear | Corr | 0.863 | 0.082 | 0.875 | 0.703 | 0.992 |
LV left ventricle, RV right ventricle, combined combined RV and LV masks, original original data without inclusion of side experiments, ICC2 features with excellent intraclass correlation from first two extractions, ICC3 features with excellent intraclass correlation from all three extractions, DAFIT synthetic data creation using main and side study data, DAFIT Filt2 combining DAFIT with feature filtering from ICC2, DAFIT Filt3 combining DAFIT with feature filtering from ICC3, rf random forest, nnet neural network, gbrm gradient boost regression model, mlp multilayer perceptron, enet elastic net, lasso least absolute shrinkage and selection operator, svmPoly support vector machine (SVM) with a polynomial kernel, svmRad SVM with a radial kernel, full full feature set, corr high correlation filter, pca principal component analysis, lincomb linear combinations filter.