Table 3.

The classification performance on the test set (with/without bad sample removal from the training and validation set).

	Acc	Recall	Spec.	Prec.	F1	MCC	Kappa
Using original train and validation sets
ResNeSt50	0.742	0.724	0.914	0.734	0.728	0.642	0.650
ResNet18	0.733	0.710	0.910	0.726	0.715	0.628	0.637
Swin-B	0.756	0.744	0.919	0.749	0.746	0.665	0.671
Ensemble	0.766	0.750	0.921	0.760	0.754	0.676	0.682
Using cleaned train and validation sets
ResNeSt50	0.752	0.734	0.917	0.751	0.740	0.659	0.664
ResNet18	0.746	0.722	0.914	0.747	0.730	0.648	0.654
Swin-B	0.759	0.746	0.920	0.759	0.751	0.671	0.674
Ensemble	0.769	0.752	0.923	0.771	0.759	0.683	0.687