Table 2.

Comparison of the proposed multistage transfer learning method with state-of-the-art ultrasound breast cancer classification methods. SVM: support vector machine; ANN: artificial neural network; AUC: area under ROC curve.

Paper	CNN	Application	Image Dataset	Train-/Validation/Test Size	Performance
Acevedo et al. [17]	SVM	Classification	Mendeley	250 images (150 malignant and 100 benign)	Accuracy = 94%
Zeebaree et al. [18]	ANN	Segmentation Classification	Mendeley	50 images for training (25 from each class)	Accuracy = 95.4%
Guldogan et al. [19]	AlexNet	Classification	Mendeley	250 images (150 malignant and 100 benign): 85% training; 15% test	Specificity = 1 Sensitivity = 0.957 Accuracy = 97.4%
The proposed MSTL method	ResNet50	Classification	Mendeley	200 images (100 from each class): 60% training; 20% validation; 20% test	AUC = 0.999 F1 measure = 0.989 Specificity = 0.98 Sensitivity = 1 Accuracy = 99%