A Review of Generative Adversarial-Based Networks of Machine Learning/Artificial Intelligence in Healthcare

Introduction

Deep Learning is known to be a promising way in all the eras of business and life. It is a promising way to discover rich and hierarchical models (Bengio,2009). Deep Learning is a concept that is inspired by the architecture of the human brain’s neural network and so it can also be called an Artificial neural network. Similar to the brain, the neurons in Deep Learning models are interconnected with each other in a layered pattern which means that there are multiple layers where the initial layer is known as the input layer and the last layer is known as the output layer. All the layers between the above mentioned are known as hidden layers. The layers are composed of n number of neurons where n can be anything and can be found out experimentally, the same applies to the number of layers as well, it can be found out experimentally. The process of experimenting is known as hyperparameter tuning where the model is trained (made to learn the pattern from existing data to predict new data) and is evaluated based on metrics which is also based on the use case and the model which performs the best amongst all the models is considered. The main process is data mining incorporated with the healthcare of Machine Learning (Sayeedakhanum Pathan,2020).

The most used Deep Learning technique in the last couple of years was the discriminative model which maps high dimensionality and rich sensory data as an input to a class label (Bengio,2009), (Bengio,2013a). but now researchers are using a technique known as Generative adversarial networks (GANs) where the input data is mapped to itself so that it can learn the features and create a similar image when fed with other images, then a classifier classifies the image as real or fake in the same pipeline (Zhang,2019). It is mostly used in the healthcare industry to generate fake or synthetic images which can’t be distinguished from real images. Then the synthetic images can be used as inputs in classifiers and to perform analysis of the human body. There are many variations of GANs which work on similar architecture but the model's changes and the way of learning and mapping images.