Page 60 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

48 Chapter 2 Deep convolutional neural network in medical image processing

4.2 Eye
Over the past few years, ophthalmic imaging has developed
rapidly. Even then, it is very recently that DL algorithms are being
useful in understanding the eye images. For the examination of
color fundus imaging (CFI), most of the works have used simple
CNNs. It can be observed from Table 2.2 that many of the studies
use fundus photography for better examining the eye. Fundus
photography is a noninvasive process that captures images of
the retina, optic disc, and macula using retinal cameras. It can
be helpful in detecting and monitoring diseases such as diabetic
retinopathy, neoplasms of the retina, glaucoma, and age-related
macular degeneration, and also it plays an important role in iden-
tifying causes of preventable blindness. A wide variety of applica-
tions such as detection and segmentation of retinal diseases,
diagnosis of eye abnormalities, segmentation of different anatom-
ical parts, and image quality assessment have been addressed in
the recent research works.

4.3 Breast
CNN applications were initially used on breast imaging
around the late 1990s. Recently, interest has resumed in this
domain, which has led to important developments over the
state-of-the-art and accomplishing the efficiency. As most of
the breast imaging methodologies are 2D in nature, techniques
that were fruitful in processing natural images can be transferred
with ease. But the lacuna here is that the only problem that is
handled is the detection of breast lesions, which is subdivided
into three tasks. One is the detection and classification of mass-
like lesions, second is the detection and classification of microcal-
cifications, and the third is the risk scoring of breast cancer
images. One of the most commonly used modalities for examina-
tion of the breast is mammography and therefore has gained the
attraction of most researchers. Work related to tomography, ul-
trasound, and shear wave elastography is not much in use, and
thus, these modalities have the possibility of getting more atten-
tion in the coming years. Table 2.2 summarizes the recent
research work done in breast image analysis. As most of the coun-
tries have screening facilities for breast cancer, there should be a
huge amount of data available, mostly for mammography, and so
there is enough scope for DL models to flourish.