Table 4.

Evaluation of the explainable MIL strategy (bag size=100).

Method	ACC($\text{\%}$)	Precision($\text{\%}$)	Recall($\text{\%}$)	F1 Score($\text{\%}$)	AUC($\text{\%}$)
ResNet18-Voting	88.97 $[85.52,92.76]$	85.85 $[80.49,92.03]$	90.37 $[85.14,94.85]$	88.05 $[83.74,92.02]$	90.08 $[86.30,92.18]$
Gated-Attention	94.48 $[92.07,97.24]$	95.89 $[92.28,98.73]$	93.24 $[91.61,97.08]$	94.55 $[83.74,92.02]$	94.63 $[91.65,97.08]$
EM-Cls	96.73 $[95.00,98.80]$	97.22 $[95.09,100.0]$	94.59 $[90.21,98.14]$	95.89 $[93.06,98.17]$	96.98 $[93.80,98.99]$

The upper and lower bounds of 95% confidence interval are shown in [$\cdot$].