Comparison of Transferred Deep Neural Networks in Ultrasonic Breast Masses Discrimination

Table 4

Comparison of quantitative classification results with feature combination.

Features	Specificity	Sensitivity	Accuracy	AUC	F1
Transferred features based on ResNet50	83.05%	87.10%	84.44%	0.91	0.79
Transferred features based on Xception	80.53%	77.61%	79.44%	0.87	0.74
Transferred features based on InceptionV3	82.20%	85.48%	83.33%	0.89	0.78
Transferred features based on InceptionV3 and Xception	80.99%	86.44%	82.78%	0.89	0.77
Transferred features based on ResNet50 and InceptionV3	86.49%	85.51%	86.11%	0.92	0.83
Transferred features based on ResNet50 and Xception	87.39%	86.96%	87.22%	0.92	0.84
Transferred features based on ResNet50, Xception, and InceptionV3	89.91%	88.73%	89.44%	0.93	0.87