Table 6.

Comparison results of different transfer learning strategies

Method	Backbone model	ACC	AUC	SENS	SPEC
Train from scratch	ResNet-50	0.746	0.835	0.719	0.752
Transfer learning	VGGNet-16	0.768	0.840	0.781	0.765
	MobileNetV2	0.757	0.835	0.750	0.758
	DenseNet-121	0.774	0.845	0.750	0.779
	AlexNet	0.762	0.828	0.75	0.765
	Inception-v3	0.780	0.836	0.750	0.785
	ResNet-18	0.774	0.854	0.813	0.765
	ResNet-50 (proposed)	0.796	0.844	0.813	0.792