Page 44 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

32 Chapter 2 Deep convolutional neural network in medical image processing

and dry eye [24]. Authors in Ref. [25] have used hybrid features for
detecting glaucoma in images of the fundus. Li et al. [26] proposed
a DL model for polyp detection, which focused that DL-based
CAD system could find out the presence of colorectal adenomas
from colonoscopy images. Multiclass classification is done for Alz-
heimer's disease using a hybrid clinical and image features [27].

2.3 Detection and classification of abnormality
Identification of a particular type of diseases such as tumors,
cancer, or polyp can be done through abnormality detection us-
ing medical images. In traditional process, medical professionals
are able to detect abnormal conditions in a patient, which in-
volves lots of human effort and is also time-consuming. Due to
this, researchers are taking interest to develop systems that could
automatically detect abnormalities in patients. There are several
works already done in this area. Many authors such as Brosch
and Tam [28], Plis et al. [29], Suk and Shen [30], and Suk et al.
[31] have applied various DL techniques to classify patients who
have Alzheimer's disease using brain MRI. Kobayashi et al. [32]
have proposed a method for the detection of abnormalities in
myocardial using cardiac MRI. Cabria and Gondra [33] have pro-
posed a method that is useful in the identification of brain tumors
using MRI segmentation fusion.

2.4 Registration
Registration or also known as a spatial alignment of images is
an image processing task that transforms the coordinate from
one image to another. Generally, this process is carried out in a
repetitive manner in which a particular type of conversion is ex-
pected and a predetermined metric is optimized [34]. Often, DL is
used for segmentation, detection, and classification of a medical
image, but in recent time, researchers have also found that DL
is also helpful in achieving good registration performance. Cheng
et al. [35] assessed the local similarity between MRI and CT images
of the head using two types of stacked autoencoders. Simonovsky
et al. [36] used DL networks to optimize registration algorithms by
comparing similarity measures of two images. Miao et al. [37] used
CNNs to achieve registration of a 3D model to 2D X-ray so that the
location and pose of an implanted object can be assessed during
surgery.
In the process of the aforementioned medical image analysis,
artificial intelligence based on ML and DL techniques acts as one
of the important components for effective diagnosis of different