Page 259 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

250 Chapter 9 Applications of deep learning in biomedical engineering

Figure 9.4 Architecture of CNN. CNN, convolutional neural network. Based on https://commons.wikimedia.org/wiki/File:
Molumen_zaz_965.svg.

1. It performs the dot product between the filters and the recep-
tive field of input image.
2. The receptive fields are shifted step by step across the width
and height of the input image [7].
To highly reduce the number of hyperparameters, this layer
shares the same parameters in every portion of the image [9].

7. Pooling layer

It performs subsampling or downsampling images by
reducing the dimensionality of the feature maps. It computes
the mean or maximum value of the feature maps. This layer helps
to generate more abstract features.

8. Fully convolutional layer

In this layer, neighboring neurons are connected together to
flatten the matrix. It classifies the features extracted in the previ-
ous layers to produce the final output [7].
The most popular CNNs used in the machine learning appli-
cations are AlexNet, U-Net, and GoogleNet [2].

9. Applications of convolutional neural

network in biomedicine
CNN attains incredible success in various problems such as
image classification and object recognition. In addition, CNN is
the most promising architecture in the biomedical field for auto-
mated disease diagnosis.