Page 263 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

254 Chapter 9 Applications of deep learning in biomedical engineering

Figure 9.6 Application of GAN in medical images. GAN, Generative Adversarial Network. From https://commons.
wikimedia.org/wiki/File:Computerized_tomography_of_the_chest_of_a_patient_with_congenital_cystic_adenomatoid_
malformation.jpg.
• Lesion classification
• Image registration such as prostate MR to transrectal ultra-
sound (TRUS) [13].
The application of GAN in medical images is illustrated in
Fig. 9.6.

14. Deep belief network
DBN is a generative probabilistic model composed of
numerous layers of stochastic and latent variables. These vari-
ables form the restricted Boltzmann machine network in which
each layer communicates with the previous layer.
The first two layers are undirected and accomplish symmetric
connections to create associative memory. The bottom layers
receive connections in a top-down manner. DBN complies with
two-stage training practices:
1. Pretraining stage
2. Fine-tuning stage

15. Pretraining stage

The unsupervised training is executed in the layers one by one
in the upward direction. The main aim of this stage is feature
extraction.