TABLE 2.

Comparison of overall performance with state-of-the-art WSI classification methods across different metrics on the test set. Here, SSC denotes single input scale (10). MSC denotes multiple input scales (7.5, 10, 12.5). MSC* denotes multiple input scales (10, 20).

Row #	Method	Accuracy	Fl	Sensitivity	Specificity	AUC
R1	Patch-based (SSC)	0.35	0.35	0.35	0.79	0.67
R2	Patch-based (MSC)	0.40	0.40	0.40	0.80	0.68
R3	Penultimate-weighted (SSC)	0.44	0.44	0.44	0.81	0.67
R4	Hypercolumn-weighted (SSC)	0.43	0.43	0.43	0.81	0.67
R5	Streaming CNN (SSC)	0.32	0.32	0.32	0.77	0.58
R6	ChikonMIL (SSC)	0.56	0.56	0.56	0.85	0.74
R7	MS-DA-MIL (SSC)	0.49	0.49	0.49	0.83	0.68
R8	MS-DA-MIL (MSC*)	0.58	0.58	0.58	0.86	0.75
R9	ScATNet (SSC)	0.60	0.60	0.60	0.87	0.77
R10	ScATNet (MSC)	0.64	6.64	0.64	0.88	0.79