Comparison of Transferred Deep Neural Networks in Ultrasonic Breast Masses Discrimination

Table 2

Quantitative classification results based on traditional approaches and CNN model.

Model	Specificity	Sensitivity	Accuracy	AUC	F1
First-order features + AdaBoost	71.00%	52.02%	67.35%	0.66	0.38
Texture features + SVM	66.80%	48.69%	66.52%	0.52	0.04
Morphological features + AdaBoost	75.18%	57.22%	70.41%	0.72	0.51
First-order features + Morphological features + AdaBoost	74.73%	54.95%	69.29%	0.73	0.49
Texture features + First-order features + AdaBoost	70.38%	49.87%	66.52%	0.65	0.36
Texture features + Morphological features + AdaBoost	74.85%	55.42%	69.53%	0.72	0.50
Texture features + Morphological features + First-order features + AdaBoost	75.13%	55.57%	69.67%	0.72	0.50
Texture features + Morphological features + First-order features + AdaBoost with LDA	74.61%	58.10%	70.55%	0.73	0.49
Texture features + Morphological features + First-order features + SVM	66.93%	37.50%	64.53%	0.53	0.15
Texture features + Morphological features + First-order features + SVM with LDA	77.00%	58.96%	71.77%	0.68	0.55
CNN3	79.22%	63.19%	74.44%	0.78	0.60