Table 6.

Performance evaluation of the model trained on the largest slice using tumor ROI and data augmentation.

	Mean 3-Fold CV			Ensemble 3-Fold CV
	Train	Validation	Test	Test (95% CI)
LGG vs. HGG
Loss	0.375	0.384	0.335
Accuracy	0.841	0.844	0.870	0.883 (0.845–0.920)
AUC-ROC	0.913	0.905	0.928	0.932 (0.902–0.961)
Precision
LGG	0.680	0.685	0.732	0.758 (0.708–0.808)
HGG	0.927	0.931	0.940	0.942 (0.915–0.970)
Recall
LGG	0.834	0.842	0.858	0.863 (0.822–0.903)
HGG	0.844	0.846	0.874	0.891 (0.855–0.927)
F1
LGG	0.749	0.755	0.759	0.807 (0.761–0.853)
HGG	0.883	0.886	0.890	0.916 (0.884–0.948)
Grade (2/3/4)
Loss	0.722	0.712	0.617
Accuracy	0.780	0.798	0.818	0.824 (0.779–0.869)
G.2	0.847	0.854	0.869	0.874 (0.835–0.913)
G.3	0.852	0.883	0.885	0.892 (0.856–0.929)
G.4	0.847	0.858	0.881	0.881 (0.843–0.919)
AUC-ROC	0.847	0.837	0.886	0.893 (0.857–0.930)
G.2	0.915	0.909	0.920	0.921 (0.889–0.952)
G.3	0.724	0.701	0.809	0.825 (0.780–0.869)
G.4	0.903	0.901	0.929	0.935 (0.780–0.869)
Precision
G.2	0.554	0.534	0.570	0.580 (0.522–0.638)
G.3	0.241	0.400	0.381	0.500 (0.441–0.559)
G.4	0.911	0.905	0.920	0.916 (0.884–0.949)
Recall
G.2	0.781	0.837	0.855	0.870 (0.830–0.909)
G.3	0.154	0.076	0.100	0.100 (0.065–0.135)
G.4	0.875	0.899	0.916	0.921 (0.889–0.953)
Specificity
G.2	0.877	0.858	0.872	0.875 (0.836–0.914)
G.3	0.939	0.984	0.980	0.988 (0.975–1.000)
G.4	0.772	0.749	0.789	0.776 (0.727–0.825)
F1
G.2	0.647	0.651	0.663	0.696 (0.642–0.750)
G.3	0.187	0.125	0.122	0.167 (0.123–0.210)
G.4	0.893	0.902	0.907	0.919 (0.886–0.951)