Table 2.

LVSI prediction performance of SVM model.

Models	Task	AUC	95% CI	Accuracy	Sensitivity	Specificity	PPV	NPV	P
Habitat1	train	0.873	0.824 - 0.922	0.803	0.779	0.830	0.835	0.772	0.015
	test	0.683	0.577 - 0.789	0.686	0.963	0.375	0.634	0.900	0.346
Habitat2	train	0.869	0.821 - 0.917	0.798	0.913	0.670	0.754	0.875	0.023
	test	0.649	0.540 - 0.757	0.647	0.833	0.438	0.625	0.700	0.729
Habitat3	train	0.870	0.821 - 0.920	0.803	0.788	0.819	0.828	0.778	0.018
	test	0.780	0.692 - 0.869	0.745	0.741	0.750	0.769	0.720	0.006
Whole tumour	train	0.805	0.745 - 0.864	0.732	0.942	0.500	0.676	0.887	ref
	test	0.629	0.519 - 0.739	0.657	0.778	0.521	0.646	0.676	ref

AUC, area under the curve; PPV, positive predictive value; NPV, negative predictive value.

P values are derived from the DeLong’s test of AUCs where AUC of whole tumour is the reference standard for comparison.