Table 3.

Performance of different module settings on differentiating EGFR 19Del, 21L858R, and WT EGFR in two wises

	Lesion-wise							Patient-wise
Methods	19Del		21L858R		WT		Overall accuracy (95%CI)	19Del		21L858R		WT		Overall accuracy (95%CI)
	Accuracy	AUC	Accuracy	AUC	Accuracy	AUC		Accuracy	AUC	Accuracy	AUC	Accuracy	AUC
GCN Classifier	0.556 ± 0.047	0.555 ± 0.054	0.561 ± 0.046	0.552 ± 0.056	0.691 ± 0.044	0.606 ± 0.061	0.404 ± 0.045	0.634 ± 0.080	0.615 ± 0.103	0.525 ± 0.088	0.516 ± 0.112	0.702 ± 0.080	0.665 ± 0.107	0.441 ± 0.088
Feat. Stand. + GCN Classifier	0.784 ± 0.036	0.919 ± 0.027	0.902 ± 0.029	0.972 ± 0.014	0.719 ± 0.040	0.999 ± 0.001	0.703 ± 0.042	0.798 ± 0.067	0.940 ± 0.046	0.945 ± 0.038	0.986 ± 0.014	0.765± 0.076	1.000 ± 0.000	0.756 ± 0.076
Feat. Stand. + Feat. Fuse + GCN Classifier	0.877 ± 0.029	0.996 ± 0.004	0.901 ± 0.028	0.971 ± 0.013	0.859 ± 0.031	1.000 ± 0.000	0.818 ± 0.035	0.899 ± 0.050	1.000 ± 0.000	0.941 ± 0.042	0.991 ± 0.009	0.908± 0.504	1.000 ± 0.000	0.874 ± 0.059

Data are given as point estimation ± halfwidth of the 95% confidence interval. 19Del 19 deletion, 21L858R 21-point mutation, AUC Area under the curve, CI Confidence interval, EGFR Epidermal growth factor receptor, GCN Graph convolutional network, Feat. Stand. Feature standardization, Feat. Fuse Feature fusion, WT Wild-type