Table 2.

Performance indices of the proposed system and baseline methods.

Value (95% CI)	The proposed system	Using ImageNet pre-trained model	Using multi-class MIL-trained model	CLAM
Accuracy (%)	77.5 (72.3–82.7)	72.6 (67.2–78.1)	75.6 (70.3–80.9)	67.3 (61.6–73.0)
κ	0.650 (0.570–0.730)	0.559 (0.471–0.647)	0.622 (0.540–0.703)	0.469 (0.376–0.562)
κquad	0.897 (0.815–0.979)	0.845 (0.746–0.945)	0.901 (0.819–0.982)	0.779 (0.658–0.900)

For each criteria, the maximum value is set bold-faced, and the second maximum is set italic.