Page 283 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

274 Chapter 10 Deep neural network in medical image processing

which is now available as pytorch for python developers. Caffe
[36] is another powerful deep learning framework available as a
python library. Mxnet [37] is also a deep learning open-source
framework from Apache available for multiple languages. Tensor-
flow [38] is a very popular multipurpose deep learning, which is
being used for all kinds of deep learning tasks. Theano [39]isa
very powerful versatile deep learning framework, which uses
numpy-like syntax and has GPU capabilities.

2. Digital image and computer vision

2.1 Introduction
Computer vision is a discipline of computer science, which
deals with the study of working of human eye and brain subsys-
tem dealing with interpreting images. That is how a human eye
reads an image and converts it into signal, and that is how the
brain (the part that is responsible for vision) understands and
interprets those signals to understand the content and context
of the image. Until recently, computer vision was thought of
as a field with limited potential fit for doing petty task like autoi-
mage enhancement (fixing contrast and brightness issues) or
trivial stuff like that. But in recent times, various researchers
have successfully shown various use cases such as object identi-
fication and facial recognition, among others. This change in
perspective has been largely attributed to two major factors of
which the first and most important factor is the exponential
growth in processing power of modern-day computers. To put
things in perspective, let us take an example Intel that released
the first processor in the pentium line code named Pentium 1 in
1993, the much celebrated processor that takes a special place in
personal computing history operated in the 50e60 MHz fre-
quency range. It was sold at a price north of 800USD. Now, it is
compared with Intel 10th Gen Core X chips, which were
announced at its fall hardware event 2019. It boasts 18 CPU cores
further enhanced with 36 virtual cores operating at frequency of
3 GHz and is available at around the same price point (around
800e1000 USD). This example highlights the exponential growth
in processing power; this fact coupled with the vast amount of
investment that is being poured in the field of data science.

2.2 What is an image?
Each image can be described as an ordered set of small units
called pixels. An image can be thought of as two-dimensional