Reference | Machine learning method | Features | Feature selection | Number of cases | Classification performance |
---|---|---|---|---|---|
Chang et al. 2005 [22] | SVM | 6 morphological features | No | 210 (B: 120, M: 90) | Acc.: 0.909, Sens.: 0.888, Spec.:0.925 |
Nascimento et al. 2016 [24] | Non-linear SVM | 5 morphological features, 39 texture features | Yes | 120 (B: 70, M: 50) | Acc.: 0.958, Sens.: 0.960, Spec.: 0.957 |
Wei et al. 2019 [27] | SVM | 4 morphological features, 3 texture features | No | 1061 (B: 472, M: 589) | Acc.: 0.873, Sens.: 0.870, Spec.: 0.876 |
Daoud et al. 2020 [28] | SVM | 18 morphological features, 800 texture features, VGG features | Yes | 643 (B: 327, M: 216) | Acc.: 0.961, Sens.: 0.957, Spec.: 0.949 |
Fujioka et al. 2019 [32] | CNN (GoogLeNet [35]) | – | – | 360 (B: 144, M: 216) | Acc.: 0.925, Sens.: 0.958, Spec.: 0.875, AUROC: 0.913 |
Lazo et al. 2020 [31] | CNN (VGG-16 [36]) | – | – | 947 (B: 587, M: 360) | Acc.: 0.919, AUROC: 0.934 |
Luo et al. 2023 [33] | Spatial attention CNN (for images) MLP (for BIRADS descriptors) | 25 BIRADS descriptors (Manually annotated) | – | 596 (B: 291, M: 305) | Acc.: 0.910, Sens.: 0.928, Spec.: 0.890, AUROC: 0.945 |