Table 4.

Performance comparison.

Model	Precision	Sensitivity	Specificity	F1 score	Parameters
ResNet-34 (baseline/student)	0.812	0.845	0.992	0.820	21.3M
ResNeXt-50	0.817	0.830	0.992	0.822	23.0M
Densenet-161	0.865	0.826	0.993	0.838	26.5M
ResNeSt-200 (teacher)	0.856	0.856	0.994	0.854	68.2M
Our model (w/o weight pre-training)	0.833	0.856	0.993	0.841	21.3M
Our model* (w/ weight pre-training)	0.848	0.865	0.994	0.853	21.3M

Bold value indicates the best performance in the corresponding criteria.

^*indicates the proposed method.