Table 7.

Performance evaluation of the model trained on 20 consecutive slices using tumor ROI with data augmentation.

	Mean 3-Fold CV			Ensemble 3-Fold CV
	Train	Validation	Test	Test (95% CI)
LGG vs. HGG
Loss	0.354	0.336	0.371
Accuracy	0.894	0.883	0.882	0.883 (0.845–0.920)
AUC-ROC	0.923	0.922	0.921	0.927 (0.896–0.957)
Precision
LGG	0.772	0.767	0.752	0.747 (0.697–0.798)
HGG	0.952	0.935	0.945	0.952 (0.927–0.977)
Recall
LGG	0.888	0.843	0.871	0.887 (0.851–0.924)
HGG	0.896	0.898	0.886	0.881 (0.843–0.919)
F1
LGG	0.825	0.804	0.807	0.811 (0.766–0.857)
HGG	0.923	0.916	0.915	0.915 (0.883–0.948)
Grade (2/3/4)
Loss	0.696	0.590	0.603
Accuracy	0.839	0.818	0.810	0.835 (0.791–0.878)
G.2	0.895	0.870	0.863	0.878 (0.839–0.916)
G.3	0.896	0.877	0.872	0.896 (0.860–0.932)
G.4	0.889	0.888	0.884	0.896 (0.860–0.932)
AUC-ROC	0.860	0.846	0.862	0.873 (0.834–0.912)
G.2	0.918	0.904	0.914	0.920 (0.889–0.952)
G.3	0.758	0.720	0.752	0.772 (0.722–0.821)
G.4	0.908	0.915	0.921	0.927 (0.896–0.957)
Precision
G.2	0.617	0.576	0.560	0.583 (0.525–0.641)
G.3	0.684	0.258	0.287	0.600 (0.542–0.658)
G.4	0.914	0.911	0.928	0.930 (0.900–0.960)
Recall
G.2	0.916	0.791	0.812	0.913 (0.880–0.946)
G.3	0.106	0.05	0.122	0.100 (0.065–0.135)
G.4	0.936	0.938	0.911	0.926 (0.895–0.957)
Specificity
G.2	0.890	0.885	0.874	0.871 (0.831–0.910)
G.3	0.995	0.978	0.963	0.992 (0.981–1.000)
G.4	0.766	0.755	0.812	0.816 (0.770–0.861)
F1
G.2	0.737	0.667	0.663	0.712 (0.659–0.765)
G.3	0.179	0.078	0.171	0.171 (0.127–0.216)
G.4	0.925	0.924	0.919	0.928 (0.898–0.958)