Research Article
[Retracted] A Data-Efficient Framework for the Identification of Vaginitis Based on Deep Learning
Table 1
Results of the informative sample selection method and the all samples selection approach under different pretrained CNNs.
| CNN type | Annotation method | Precision (%) | Recall (%) | F1 score (%) | Accuracy (%) | AUC (%) |
| AlexNet | All samples | 94.35 ± 2.66 | 89.12% ± 2.16 | 91.62 ± 1.01 | 92.85 ± 0.91 | 97.75% ± 0.59 | Informative samples | 93.05 ± 2.71 | 90.18% ± 2.46 | 91.54 ± 1.29 | 92.69 ± 1.16 | 97.88 ± 0.26 |
| VGG16 | All samples | 91.49 ± 1.49 | 90.18 ± 1.12 | 90.81 ± 0.51 | 92.00 ± 0.49 | 98.15 ± 0.36 | Informative samples | 92.03 ± 1.13 | 91.05 ± 0.38 | 91.54 ± 0.60 | 92.62 ± 0.56 | 97.25 ± 0.26 |
| ResNet50 | All samples | 95.29 ± 2.64 | 89.47 ± 5.73 | 92.16 ± 2.57 | 93.38 ± 2.04 | 98.86 ± 0.56 | Informative samples | 94.09 ± 3.34 | 95.09 ± 2.24 | 94.52 ± 0.87 | 95.15 ± 0.87 | 98.93 ± 0.31 |
| Se_ResNet50 | All samples | 96.06 ± 1.54 | 93.51 ± 0.98 | 94.76 ± 0.75 | 95.46 ± 0.67 | 99.06 ± 0.16 | Informative samples | 96.99 ± 0.75 | 90.18 ± 2.46 | 93.44 ± 1.36 | 95.06 ± 1.09 | 98.94 ± 0.28 |
|
|