2.7. Convolutional Neural Networks (CNNs)

CNNs are applied to explore patterns in an image. This is done by convoluting over an image and looking for patterns [27]. The network can detect lines and corners in the few front layers of CNNs. Via our neural net, however, we can then transfer these patterns down and begin to identify more complex characteristics as we get deeper. This property ensures that CNNs are very effective at detecting objects in images [26]. The proposed system uses CNNs to detect breast cancer from breast tissue images.

The architecture of a CNN has 3 main layers, the convolutional layer, pooling layer, and fully connected layer, as shown in Figure 5. The first layer calculates the output of neurons which are linked with local regions. Each one is calculated by a dot product of weights and the region. For image inputs, typical filters are small in area such as 3 × 3, 5 × 5, or 8 × 8. These filters scan the image by a sliding window on the image, while learning the recurrent patterns which arise in any area of the image. The interval between filters is known as the stride. The convolution is extended to overlapping windows if the stride hyperparameter is smaller than the filter dimension. A detailed visual explanation of neural networks (NNs) is shown in Figure 6.

Typical CNN architecture for automatic detection of IDC breast cancer.

Detailed process of a neural network (NN).