Page 262 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 9 Applications of deep learning in biomedical engineering 253

as a seed generative, and it has to be chosen randomly from
Gaussian distribution. It helps to provide underlying structure
of training data. It is used to synthesis the low-dimensional input
domain to high-dimensional target domain [13].

12.2 Discriminator network
It focuses on discrimination. It accesses the data from gener-
ative network and attempts to discriminate between them. It is a
normal classification model.
Critically, the generator can directly access the real images
only by the interaction with the discriminator. The output of
the discriminator only decides whether the two networks need
to be optimized or not. The discriminator can access both the
real images and the synthesis images [13].

13. Applications of generative adversarial
network in biomedicine
GANs are a special type of neural network model where two
networks are trained simultaneously, with one focused on image
generation and the other centered on discrimination.
GAN is a neural network model, which comprised two net-
works. These two networks process simultaneously in which
one aims to generate image samples and other aims to classify
them [14].
The applications of GAN are categorized into the following:
• Image reconstruction
• Image synthesis
• Image segmentation
• Image classification
• Image detection
• Image registration
The examples of GAN applications are as follows:
• Semantic image editing, data augmentation, and style transfer
• Image denoising and removing artifacts
• Image synthesis such as generating synthetic samples for three
classes of liver lesions
• Cross-modality synthesis such as brain CT image synthesis
from MR image [14]
• Computing segmentation map
• Chest abnormality classification, patch-based retinal vessel
classification, and cardiac disease diagnosis