Table 2.
Summary of the classification performances on the training set. Data are reported as the mean (SD). Classification metrics were computed on the held-out set for 15 resamples (5-fold cross-validation repeated 3 times) during training. Abbreviations: CE = contrast enhancement; NEC = necrosis; HYP = hyperintensity in FLAIR; TUM = tumor core (union of CE and NEC); WHOLE = whole lesion (union of CE, NEC, and HYP); SVM = support vector machine; RF = random forest; SD = standard deviation; AUC = area under the curve.
| Classification Metric | WHOLE | CE | NEC | HYP | TUM | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| SVM | RF | SVM | RF | SVM | RF | SVM | RF | SVM | RF | |
| Accuracy, % | 61.4 | 56.6 | 63.1 | 58.5 | 67.6 | 58.5 | 60.9 | 55.9 | 61.6 | 56.6 |
| (SD) | (6.9) | (6.0) | (10.0) | (6.9) | (9.0) | (7.2) | (6.9) | (10.3) | (5.0) | (7.4) |
| Sensitivity, % | 30.2 | 29.0 | 54.1 | 37.3 | 59.6 | 43.5 | 22.4 | 26.3 | 36.5 | 32.9 |
| (SD) | (19.0) | (12.7) | (15.6) | (12.3) | (9.9) | (8.6) | (18.0) | (10.9) | (18.6) | (11.9) |
| Specificity, % | 85.3 | 77.7 | 70.1 | 74.8 | 73.8 | 69.8 | 90.5 | 78.7 | 80.7 | 74.8 |
| (SD) | (13.3) | (7.4) | (16.5) | (9.8) | (14.0) | (11.5) | (8.8) | (14.4) | (15.2) | (10.1) |
| AUC | 0.661 | 0.535 | 0.688 | 0.622 | 0.759 | 0.615 | 0.610 | 0.549 | 0.669 | 0.588 |
| (SD) | (0.120) | (0.100) | (0.104) | (0.090) | (0.069) | (0.087) | (0.124) | (0.100) | (0.112) | (0.081) |