Table 6 -.
Comparison between mean Dice Kappa and detection/outline error rate (DER/OER) values of different classifiers for segmentation of WMHs using T1w and FLAIR data in the ADC, NACC, and ADNI2/GO datasets. Blue color indicates the best performance in terms of SI.
| Dataset | ADC | NACC | ADNI2/GO | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Measure | SI | DER | OER | SI | DER | OER | SI | DER | OER |
| Naïve Bayes | 0.42±0.25* | 0.34±0.27 | 0.82±0.30 | 0.50±0.21 | 0.43±0.26 | 0.74±0.34 | 0.50±0.29 | 0.43±0.30 | 0.73±0.39 |
| Logistic | 0.56±0.18 | 0.27±0.24 | 0.61±0.19 | 0.65±0.13 | 0.27±0.39 | 0.55±0.23 | 0.64±0.20 | 0.30±0.40 | 0.53±0.25 |
| LDA | 0.58±0.19 | 0.35±0.33 | 0.49±0.17 | 0.69±0.13 | 0.21±0.26 | 0.57±0.36 | 0.60±0.23 | 0.21±0.27 | 0.66±0.41 |
| QDA | 0.42±0.23* | 0.44±0.32 | 0.73±0.22 | 0.54±0.21 | 0.48±0.32 | 0.62±0.27 | 0.51±0.29 | 0.48±0.34 | 0.64±0.33 |
| KNN | 0.65±0.16 | 0.18±0.18 | 0.51±0.18 | 0.71±0.13 | 0.15±0.17 | 0.50±0.23 | 0.72±0.18 | 0.17±0.20 | 0.48±0.24 |
| Decision Trees | 0.58±0.16 | 0.25±0.25 | 0.58±0.14 | 0.65±0.12 | 0.23±0.25 | 0.55±0.18 | 0.65±0.22 | 0.25±0.28 | 0.54±0.21 |
| Random Forests | 0.66±0.14 | 0.18±0.18 | 0.50±0.16 | 0.72±0.10 | 0.14±0.16 | 0.48±0.20 | 0.72±0.19 | 0.15±0.18 | 0.47±0.23 |
| Bagging | 0.14±0.16* | 0.27±0.28 | 0.63±0.27 | 0.69±0.13 | 0.19±0.31 | 0.53±0.25 | 0.69±0.17 | 0.21±0.33 | 0.51±0.26 |
| SVM | 0.56±0.24 | 0.31±0.37 | 0.56±0.26 | 0.67±0.13 | 0.22±0.37 | 0.53±0.25 | 0.68±0.22 | 0.26±0.40 | 0.48±0.28 |
| AdaBoost | 0.65±0.15 | 0.18±0.18 | 0.50±0.17 | 0.72±0.11 | 0.16±0.22 | 0.49±0.23 | 0.71±0.20 | 0.18±0.25 | 0.49±0.27 |