Table 2 Diagnostic accuracy using different classification systems. Feature selection had a significant impact on classifier performance with Friedman test \(\chi ^2 = 35.3\), 5 d.f., \(p = 1.3\times 10^{-6}\). V: Wilcoxon signed rank statistic of performance compared to complete system; p: Associated Bonferroni-corrected p-value.
Sensitivity | Specificity | DSC | Accuracy | V | p | |
|---|---|---|---|---|---|---|
SVM-based CAD system | ||||||
LOSO | 86% ±3.48 | 94% ± 2.39 | 84% ± 3.49 | 90% ± 2.01 | 136 | 0.0024 |
Tenfold | 78% ± 9.19 | 97% ± 4.83 | 85.96% ± 6.06 | 87.50% ± 4.86 | ||
Fourfold | 85% ± 1.41 | 92% ± 5.65 | 88.11% ± 1.76 | 88.50% ± 2.20 | ||
Twofold | 83% ±3.82 | 91% ± 3.83 | 86.44% ± 1.29 | 87% ± 1.15 | ||
Random forest-based CAD system | ||||||
LOSO | 76% ±4.29 | 96% ± 1.97 | 75% ± 4.27 | 86% ± 2.37 | 118 | 0.0054 |
Tenfold | 74% ± 1.26 | 98% ± 4.21 | 83.61% ± 9.15 | 86% ± 7.37 | ||
Fourfold | 71% ± 8.28 | 98% ± 2.31 | 81.87% ± 5.05 | 84.50% ± 3.41 | ||
Twofold | 71% ±4.24 | 99% ± 1.41 | 80.87% ± 2.14 | 80.30% ± 1.41 | ||
Naive Bayes-based CAD system | ||||||
LOSO | 84% ±3.68 | 94% ± 2.38 | 82.33% ± 3.68 | 89% ± 2.19 | 136 | 0.0024 |
Tenfold | 80% ± 1.05 | 97% ± 4.83 | 87.13% ± 7.10 | 88.50% ± 5.79 | ||
Fourfold | 77% ± 6.00 | 97% ± 2.00 | 85.46% ± 4.36 | 87% ± 3.46 | ||
Twofold | 77% ±4.24 | 95% ± 1.41 | 84.58% ± 2.03 | 86% ± 1.41 | ||
KNN-based CAD system | ||||||
LOSO | 80% ±4.02 | 99% ± 1.00 | 79.66% ± 4.01 | 89.50% ± 2.04 | 127.5 | 0.0114 |
Tenfold | 75% ± 8.87 | 100% ± 0.00 | 85.49% ± 5.88 | 87.50% ± 4.43 | ||
Fourfold | 71% ± 1.10 | 100% ± 0.00 | 82.61% ± 7.43 | 85.50% ± 5.50 | ||
Twofold | 70% ±0.00 | 100% ± 0.00 | 82.35% ± 0.00 | 85% ± 0.00 | ||
Decision Trees-based CAD system | ||||||
LOSO | 80% ±4.02 | 99% ± 1.00 | 79.66% ± 4.01 | 89.50% ± 2.04 | 127.5 | 0.0114 |
Tenfold | 75% ± 8.87 | 100% ± 0.00 | 85.49% ± 5.88 | 87.50% ± 4.43 | ||
Fourfold | 71% ± 1.10 | 100% ± 0.00 | 82.61% ± 7.43 | 85.50% ± 5.50 | ||
Twofold | 70% ±0.00 | 100% ± 0.00 | 82.35% ± 0.00 | 85% ± 0.00 | ||
