Table 2.

Performance comparison of LightGBM models and XGBoost models

Comparison	Predictive models		AUC [95% CI]	Accuracy	Precision	Sensitivity	Specificity	p value
LGB	T2WI	Training set	0.974 [0.973–0.974]	0.897	0.903	0.890	0.891
		Test set	0.872 [0.871–0.875]	0.806	0.75	0.917	0.893	0.023
	CE-T1WI	Training set	0.899 [0.894–0.899]	0.831	0.821	0.846	0.841
		Test set	0.848 [0.846–0.857]	0.750	0.821	0.639	0.705	0.030
XGB	T2WI	Training set	0.951 [0.947–0.955]	0.886	0.852	0.934	0.927
		Test set	0.838 [0.829–0.842]	0.750	0.701	0.861	0.82	0.023
	CE-T1WI	Training set	0.872 [0.864–0.873]	0.783	0.789	0.772	0.777
		Test set	0.843 [0.841–0.849]	0.750	0.846	0.610	0.700	0.030

p values were obtained by performing DeLong test between LightGBM and XGBoost models constructed using the same features