Table 2.

Tabular data showing the overall performance of the various model architectures and augmentation methods on the internal test set in terms of precision, recall, and $F1$-score. Optimal performance is highlighted in bold. 95% confidence interval added for framewise results.

Average internal test performance
	ResNet50			ResNeXt101			EfficientNetB2
Class	Precision	Recall	$F1$-score	Precision	Recall	$F1$-score	Precision	Recall	$F1$-score
No augmentation	0.843 ± 0.008	0.836 ± 0.012	0.836 ± 0.012	0.876 ± 0.008	0.873 ± 0.010	0.873 ± 0.010	0.856 ± 0.009	0.851 ± 0.013	0.851 ± 0.013
Traditional augmentation	0.879 ± 0.007	0.874 ± 0.011	0.874 ± 0.010	0.873 ± 0.007	0.869 ± 0.010	0.869 ± 0.009	0.862 ± 0.010	0.855 ± 0.015	0.855 ± 0.014
Proposed method	0.906 ± 0.007	0.903 ± 0.008	0.904 ± 0.008	0.892 ± 0.006	0.890 ± 0.006	0.890 ± 0.006	0.868 ± 0.010	0.864 ± 0.013	0.864 ± 0.013