Page 40 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

28 Chapter 2 Deep convolutional neural network in medical image processing

presented comprehensively. Lastly, in Sections 5 and 6, a critical
discussion with several concluding remarks is presented.

2. Medical image analysis

Medical imaging, in general, provides visual knowledge about
the internal of the human system. This helps the clinicians and
medical professionals to give a better diagnostic result and effi-
cient treatment to the patients. Medical imaging of the human
system can be carried out by using different medical imaging mo-
dalities. There are several imaging modalities available in the
market to recognize the human organ, such as magnetic reso-
nance imaging (MRI), endoscopy, computed tomography (CT),
X-ray, positron emission tomography (PET), ultrasound,
narrow-band imaging, elastography, optical imaging, infrared
thermography, and terahertz [10e12]. Table 2.1 gives a compara-
tive study of these imaging modalities. These techniques have an
important role in providing vital anatomical and functional infor-
mation of the different parts of the human body. Medical imaging
is of great support for the modern healthcare system. Fig. 2.3 pro-
vides an idea about the different organs for which the individual
modality can be used. There are various areas of medical imaging
in which DL methods were implemented successfully. Fig. 2.4
shows these areas of application. A brief description of the various
domains of clinical image analysis is stated in the following.

2.1 Segmentation
Segmentation is a process in which the image is divided into
several nonoverlapping areas by taking a set of rules, which can
be a set of features such as color, texture, and contrast or set of
similar pixels [13]. This method helps to reduce the search region
in an image by separating it into two groups such as the back-
ground and the object. An important feature of segmentation is
to represent the image in a form that is more meaningful and
can be easily analyzed. This helps to extract several significant in-
formation such as shape, position, volume, and different abnor-
malities related to the organ [14,15]. Authors in Ref. [16] have
presented a 3D clinical image segmentation approach in which
the system has the ability to evaluate and compare the quality
of segmentation. Feng et al. [17] have proposed a 3D multiscale
Otsu thresholding algorithm, which is iterative in nature, for the
segmentation of medical images. In this algorithm, the image is
represented in multiple levels so as to remove the effect of weak