. Author manuscript; available in PMC: 2025 Oct 15.

Published in final edited form as: IEEE Access. 2025 Jul 28;13:133351–133369. doi: 10.1109/access.2025.3593420

TABLE 3.

T-test results comparing the ood auc in the (panda,cam16) experiment using uni features.

(a) Statistical comparison for vaeabmil
Model	OoD/auroc	t_stat	p_value	Significant
vaeabmil	1.0000 ± 0.0000	-	-	-
daeabmil	1.0000 ± 0.0000	2.0000	0.1835	False
dsmil	0.7926 ± 0.0306	11.7303	0.0072	True
clam	0.6966 ± 0.0516	10.1895	0.0095	True
dftdmil	0.9436 ± 0.0298	3.2799	0.0817	False
transmit	0.9106 ± 0.0223	6.9357	0.0202	True
abmil	0.9590 ± 0.0069	10.3725	0.0092	True
(b) Statistical comparison for daeabmil
Model	OoD/auroc	t_stat	p_value	Significant

daeabmil	1.0000 ± 0.0000	-	-	-
vaeabmil	1.0000 ± 0.0000	−2.0000	0.1835	False
dsmil	0.7926 ± 0.0306	11.7294	0.0072	True
clam	0.6966 ± 0.0516	10.1893	0.0095	True
dftdmil	0.9436 ± 0.0298	3.2798	0.0817	False
transmit	0.9106 ± 0.0223	6.9359	0.0202	True
abmil	0.9590 ± 0.0069	10.3694	0.0092	True