Page 13 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

P. 13

Congruence of deep learning in

biomedical engineering: future

prospects and challenges

Aradhana Behura
Veer Surendra Sai University of Technology, Burla, Sambalpur, Odisha, India

1. Introduction

Death from breast cancer [7,8,11,14,34,35] may be avoided by
detecting the risks of medical patients and discussing them
effectively [3]. One of the known risks for tumor development
besides age, gender, gene mutations, and family history is the
comparative sum of radiodense tissue in the female breast, called
mammographic density [26e30]. By using a stacked autoencoder
we can more accurately predict brain tumors. C-means,
K-means, and DBSCAN clustering techniques are used to detect
affected areas in medical images. There are various types of
nature-inspired algorithms used to optimize the performance of
clustering that provide better results. Segmentation of brain [31]
and liver tumors provides [9,10,12,13,23] important biomarkers
for medical diagnosis [24,25]. Here, we present and authenticate
a procedure to integrate an improved edge pointer and derive an
initial curve for magnetic resonance imaging (MRI)-based disease
segmentation from the dataset. At the preprocessing step, the
computed tomography (CT) image intensity values were truncated
to lie in a ﬁxed range to enhance the image contrast surrounding
the organ and the disease-affected area. To eliminate nonliver tis-
sues for the following segmentation of the disease, the liver is
segmented by two convolutional neural networks (CNNs)
[15,21e23] in a coarse-to-ﬁne manner. Here, we present a new pro-
cedure for combining high-resolution photorealistic medical im-
ages from the semantic label charts with conditional generative
adversarial networks (GANs). A GAN [32,33] is a type of deep
learning method made up of two neural networks in conﬂict with
each other in a zero-sum game outline.

8 9 10 11 12 13 14 15 16 17 18