Table 3.

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

	ResNet50			ResNeXt101			EfficientNetB2
Class	Precision	Recall	$F1$-score	Precision	Recall	$F1$-score	Precision	Recall	$F1$-score
Framewise internal test
CA	0.933 ± 0.002	0.934 ± 0.002	0.934 ± 0.002	0.922 ± 0.002	0.922 ± 0.002	0.922 ± 0.002	0.933 ± 0.002	0.934 ± 0.002	0.933 ± 0.002
CP	0.916 ± 0.002	0.916 ± 0.002	0.916 ± 0.002	0.951 ± 0.002	0.951 ± 0.002	0.951 ± 0.002	0.912 ± 0.002	0.912 ± 0.002	0.912 ± 0.002
Normal	0.931 ± 0.002	0.931 ± 0.002	0.931 ± 0.002	0.929 ± 0.002	0.927 ± 0.002	0.927 ± 0.002	0.879 ± 0.002	0.874 ± 0.003	0.875 ± 0.003
Studywise internal test
CA	0.914 ± 0.054	0.898 ± 0.046	0.905 ± 0.038	0.918 ± 0.051	0.902 ± 0.043	0.909 ± 0.034	0.911 ± 0.055	0.850 ± 0.053	0.878 ± 0.040
CP	0.969 ± 0.030	0.922 ± 0.041	0.944 ± 0.027	0.940 ± 0.042	0.921 ± 0.042	0.930 ± 0.031	0.911 ± 0.056	0.894 ± 0.048	0.901 ± 0.040
Normal	0.938 ± 0.035	0.987 ± 0.013	0.961 ± 0.020	0.920 ± 0.043	0.957 ± 0.026	0.938 ± 0.026	0.906 ± 0.049	0.971 ± 0.020	0.937 ± 0.029