Comparison of Transferred Deep Neural Networks in Ultrasonic Breast Masses Discrimination

Figure 1

The architecture of CNN3. The network has 3 convolutional layers followed by 2 fully connected layers. 32-64-128-256-2 is the number of feature maps generated in each layer. 74-36-17 is the size of the feature maps. Global average pooling is used to reduce the total number of parameters. There are 256 neurons in the first fully connected layer and 2 neurons in the output layer indicating class scores of benign and malignant masses.