Page 280 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications
P. 280
10
Deep neural network in medical
image processing
Siddharth Verma, Rashmi Agrawal
Manav Rachna International Institute of Research and Studies (MRIIRS),
Manav Rachna Campus, Faridabad, Haryana, India
1. Literature review
Work on pattern recognition using deep learning specifically
neural network started in the early 1980s, which is one of the
earliest instances of using deep learning for the same. Fukush-
ima [2] proposed in his paper on neocognition. Usage of deep
learning in medical image processing came a few years later
when Lo et al. [3]. proposed use of convolutional neural networks
(CNNs) for diagnosis interpretation of biomedical images. But all
of these were in the theoretical realm. One of the earliest prac-
tical implementations of CNNs for image processing came
when LeCun et al. [4] used multilayer neural networks trained
using backpropagation algorithm for hand-written digits recog-
nition. Mead et al. [5]inhis book Analog VLSI Implementation
of Neural Systems reviewed various early usage of learning ma-
chine in the field of image processing.
Notwithstanding these early usages of CNNs for solving image
processing problems, CNNs were not considered as mainstream
first-choice solutions for such problems. Main reasons behind
the low adoption rates were lack of suitably powerful hardware
and algorithms. Turning point in the usage of CNNs came in
2012 when Krizhevsky et al. [6] implemented large multilayered
CNN with more than 650,000 neurons trained over 1.2 million
high-resolution images. On test data, it produces error rates that
were significantly better than the state-of-the-art at that time. A
few years later, Russakovsky et al. [7] implemented a similar but
even more powerful large-scale image classification model, which
produced an image classification error rate less than 7%, which
powered the ideas of bringing this technology to the mainstream
computing domain.
Handbook of Deep Learning in Biomedical Engineering. https://doi.org/10.1016/B978-0-12-823014-5.00002-8
Copyright © 2021 Elsevier Inc. All rights reserved. 271
