Table 5 Comparison of the performance of different backbone architectures.
From: DLA-Net: dual lesion attention network for classification of pneumoconiosis using chest X-ray images
Models | Multi-class classification | ||||
|---|---|---|---|---|---|
ACC | SEN | SPE | F1 | AUC | |
DenseNet121 | 0.8258 ± 1.88 | 0.7188 ± 1.46 | 0.8766 ± 1.98 | 0.7260 ± 1.69 | 0.7977 ± 1.60 |
DLA-DenseNet121 | 0.8458 ± 1.97 | 0.7525 ± 2.80 | 0.8858 ± 1.65 | 0.7575 ± 2.73 | 0.8191 ± 2.22 |
ResNet50 | 0.8248 ± 2.40 | 0.7229 ± 2.81 | 0.8689 ± 2.71 | 0.7293 ± 3.06 | 0.7959 ± 2.63 |
DLA-ResNet50 | 0.8348 ± 1.25 | 0.7388 ± 1.79 | 0.8738 ± 2.05 | 0.7415 ± 2.36 | 0.8063 ± 1.85 |
Inceptionv3 | 0.8288 ± 0.77 | 0.7204 ± 1.20 | 0.8762 ± 1.12 | 0.7278 ± 1.31 | 0.7983 ± 1.06 |
DLA-Inceptionv3 | 0.8302 ± 1.04 | 0.7188 ± 1.53 | 0.8807 ± 0.59 | 0.7269 ± 1.44 | 0.7998 ± 0.89 |
EfficientNetB4 | 0.8323 ± 1.01 | 0.7293 ± 1.23 | 0.8860 ± 1.00 | 0.7394 ± 0.92 | 0.8076 ± 0.71 |
DLA-EfficientNetB4 | 0.8489 ± 1.62 | 0.7468 ± 2.91 | 0.8966 ± 1.07 | 0.7550 ± 2.41 | 0.8217 ± 1.75 |
Xception | 0.8279 ± 2.55 | 0.7189 ± 3.04 | 0.8838 ± 1.93 | 0.7294 ± 2.92 | 0.8013 ± 2.23 |
DLA-Xception | 0.8562 ± 1.61 | 0.7643 ± 1.44 | 0.8941 ± 1.81 | 0.7687 ± 1.53 | 0.8292 ± 1.53 |
