Table 3.

Performance comparison of different feature extraction modules using capsule network.

Model	Params. (M)	Accuracy	Precision	Sensitivity	Specificity	AUC
ResNet-50	9.63	0.849	0.931	0.771	0.935	0.910
Inception-V3	9.92	0.864	0.933	0.800	0.935	0.883
DenseNet-121	8.04	0.970	0.971	0.971	0.968	0.986

* Bold font indicates the best value among the three models.