Page 256 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 9 Applications of deep learning in biomedical engineering 247

• Medical imaging: It is the process of creating visualizations of
human internal parts via different acquisition tools such as X-
rays, magnetic resonance imaging (MRI), fundus photography,
and so on [3].
• Brain, body, and machine interface: It is a computer-based
system in which the brain signals (electroencephalogram
[EEG]) are analyzed and transmitted to an external device to
perform desired actions.
• Public and medical health management (PmHM): It deals
with the process of maintaining medical big data and deliv-
ering health services effectively.
The substantial growth of science and technology in biomed-
ical engineering leads to complexity in maintaining medical big
data. These data are heterogeneous and include a wide range
of clinically distinct biomarkers. Moreover, they are having
different dimensions as they are acquired using various imaging
modalities such as MRI, computed tomography (CT), ultrasound
(US) technology, positron emission tomography (PET), or two-
dimensional/three-dimensional X-ray.
Typically, biomedical data are unstructured and nonstatic
being characterized by a high complexity. However, the conven-
tional models are not efficient to deal with such problems. DL
surmounts these issues tremendously by accessing big medical
data for diagnosing diseases accurately [4].

3. Deep learning

DL is a function of AI that mimics the mechanism of human
brain in processing high-dimensional data for complex problems.
It is a subdivision of machine learning and consists of different
processing layers that can learn unlabeled or unstructured input
data. The role of DL in Artificial Intelligence is shown in Fig. 9.2.
The ability of dealing unstructured data makes DL very powerful
by processing large numbers of features. The DL model will be
overfitted for simple or incomplete data and thus fails to gener-
alize [5].
Deep neural networks can automatically extract the features,
which reduces the burden of medical experts for identifying
diseases and their progression. The input data are passed
through series of network processing layers. The high-level fea-
tures are extracted from the low-level features without human
intervention. It can solve complex problems even for the diverse
and unstructured data. The more they train, the better they
executed [2].