Page 202 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

192 Chapter 7 Early detection and diagnosis using deep learning

applications ranging from neuroimaging to early diagnosis of
certain diseases. These applications have transformed the health-
care industry and with sustained efforts in the right direction will
continue to benefit the world.

1.1 Introduction to deep learning
As the digital era has expanded at an unfathomable rate, it has
brought with it several challenges and developments. Data are
being produced at a rate faster than ever before. An estimated
2.5 quintillion bytes of data are produced every day, and nearly
90% of the world's total data were produced in the past 2 years
alone. With such a large amount of data being produced every
day and streaming in from social media, e-commerce websites,
search engines, and more, handling and using these data is an
important challenge the world faces. These exponentially growing
big data are usually unstructured and hard to deal with, and
hence, complex AI systems are needed to process the data and
extract valuable information from it.
Machine learning (ML) is one such AI technique that helps
discover patterns and forms analysis using a self-adaptive algo-
rithm. In the simplest terms, DL is a ML technique, which takes
inspiration from working of the human brain. This revolutionary
AI is able to learn from data that are both unlabeled and unstruc-
tured. It helps the computer process through layers and predicts
information. Observations can take up various forms such as
sound, text, or images. Simply put, it is a large artificial neural
network processing a huge amount of data. These brain simula-
tions can revolutionize the world of data science and make
learning algorithms easy to use. According to Andrew Ng,
cofounder Coursera, DL is the best shot we have toward AI.
DL tries to imitate the way the human brain works by
following the pattern set inside the brain. Our brains have more
than 100 billion neurons connected to each other with each
neuron having 100,000 neighbors. The signal received by one
neuron transfers down its axon to the dendrite of another neuron
through the synapse. In DL, the node(neuron) takes the input val-
ue(signal), which, after passing through it, gives the output.
The input from the observation taken becomes one layer,
which creates an output layer. This output layer, in turn, acts
as an input for the next layer and so on till the point the final
output signal is received. A hierarchy of concepts is formed where
each updated layer is more comprehensive than the previous
one. Thus, DL is a subset of ML, which utilizes hierarchical arti-
ficial neural networks to carry out ML efficiently (Fig. 7.1).