Read Anytime Anywhere Easy Ebook Downloads at ebookmeta.

com

Artificial Intelligence, Machine Learning, and

Data Science Technologies: Future Impact and Well-
Being for Society 5.0 (Demystifying Technologies
for Computational Excellence) 1st Edition
https://ebookmeta.com/product/artificial-intelligence-
machine-learning-and-data-science-technologies-future-
impact-and-well-being-for-society-5-0-demystifying-
technologies-for-computational-excellence-1st-edition/

OR CLICK HERE

DOWLOAD EBOOK

Visit and Get More Ebook Downloads Instantly at https://ebookmeta.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Data Science and Innovations for Intelligent Systems:

Computational Excellence and Society 5.0 (Demystifying
Technologies for Computational Excellence) 1st Edition
Kavita Taneja (Editor)
https://ebookmeta.com/product/data-science-and-innovations-for-
intelligent-systems-computational-excellence-and-
society-5-0-demystifying-technologies-for-computational-
excellence-1st-edition-kavita-taneja-editor/
ebookmeta.com

Healthcare and Knowledge Management for Society 5.0:

Trends, Issues, and Innovations (Demystifying Technologies
for Computational Excellence) 1st Edition Vineet Kansal
(Editor)
https://ebookmeta.com/product/healthcare-and-knowledge-management-for-
society-5-0-trends-issues-and-innovations-demystifying-technologies-
for-computational-excellence-1st-edition-vineet-kansal-editor/
ebookmeta.com

Handbook of Machine Learning for Computational

Optimization: Applications and Case Studies (Demystifying
Technologies for Computational Excellence) 1st Edition
Vishal Jain (Editor)
https://ebookmeta.com/product/handbook-of-machine-learning-for-
computational-optimization-applications-and-case-studies-demystifying-
technologies-for-computational-excellence-1st-edition-vishal-jain-
editor/
ebookmeta.com

Semiconductors for Optoelectronics Basics and Applications

Naci Balkan

https://ebookmeta.com/product/semiconductors-for-optoelectronics-
basics-and-applications-naci-balkan/

ebookmeta.com
The Mind Brain Continuum Psychoneurointracrinology 1st
Edition Susan Gordon

https://ebookmeta.com/product/the-mind-brain-continuum-
psychoneurointracrinology-1st-edition-susan-gordon/

ebookmeta.com

Married to the Mountain Man (Courage County Curves) 1st

Edition Mia Brody

https://ebookmeta.com/product/married-to-the-mountain-man-courage-
county-curves-1st-edition-mia-brody/

ebookmeta.com

Hair Therapy Cures For Growing Your Beautiful Natural Hair

Tiffany Anderson

https://ebookmeta.com/product/hair-therapy-cures-for-growing-your-
beautiful-natural-hair-tiffany-anderson/

ebookmeta.com

A Ryan Revelation New York Ruthless short stories 7 1st

Edition Sadie Kincaid

https://ebookmeta.com/product/a-ryan-revelation-new-york-ruthless-
short-stories-7-1st-edition-sadie-kincaid/

ebookmeta.com

Our Ancestors A Trilogy Italo Calvino

https://ebookmeta.com/product/our-ancestors-a-trilogy-italo-calvino/

ebookmeta.com
Forbidden Alpha Bonds of Steele Omegaverse Book 6 1st
Edition S K Reign Laurel Night Ginna Moran

https://ebookmeta.com/product/forbidden-alpha-bonds-of-steele-
omegaverse-book-6-1st-edition-s-k-reign-laurel-night-ginna-moran/

ebookmeta.com
i

Artificial Intelligence,
Machine Learning, and
Data Science Technologies
ii

Demystifying Technologies for Computational Excellence:

Moving towards Society 5.0
Series Editors: Vikram Bali and Vishal Bhatnagar

This series encompasses research work in the field of Data Science, Edge Computing,
Deep Learning, Distributed Ledger Technology, Extended Reality, Quantum
Computing, Artificial Intelligence, and various other related areas, such as natural-​
language processing and technologies, high-​level computer vision, cognitive robotics,
automated reasoning, multivalent systems, symbolic learning theories and practice,
knowledge representation and the semantic web, intelligent tutoring systems, AI and
education.
The prime reason for developing and growing out this new book series is to focus on
the latest technological advancements -​their impact on the society, the challenges faced in
implementation, and the drawbacks or reverse impact on the society due to technological
innovations. With the technological advancements, every individual has personalized
access to all the services, all devices connected with each other communicating amongst
themselves, thanks to the technology for making our life simpler and easier. These aspects
will help us to overcome the drawbacks of the existing systems and help in building new
systems with latest technologies that will help the society in various ways proving Society
5.0 as one of the biggest revolutions in this era.

Artificial Intelligence, Machine Learning, and Data Science Technologies

Future Impact and Well-​Being for Society 5.0
Edited by Neeraj Mohan, Ruchi Singla, Priyanka Kaushal, and Seifedine Kadry

Transforming Higher Education through Digitalization

Insights, Tools, and Techniques
Edited by S. L. Gupta, Nawal Kishor, Niraj Mishra, Sonali Mathur, and Utkarsh Gupta

A Step towards Society 5.0

Research, Innovations, and Developments in Cloud-​Based Computing Technologies
Edited by Shahnawaz Khan, Thirunavukkarasu K., Ayman AlDmour, and Salam
Salameh Shreem

For more information on this series, please visit: www.routledge.com/​Demystifying-​

Technologies-​for-​Computational-​Excellence-​Moving-​Towards-​Society-​5.0/​book-​
series/​CRCDTCEMTS
iii

Artificial Intelligence,
Machine Learning, and Data
Science Technologies
Future Impact and Well-​Being for
Society 5.0

Edited by
Neeraj Mohan, Ruchi Singla, Priyanka Kaushal,
and Seifedine Kadry
iv

First edition published 2022

by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-​2742
and by CRC Press
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
© 2022 selection and editorial matter, Neeraj Mohan, Ruchi Singla, Priyanka Kaushal, and Seifedine
Kadry; individual chapters, the contributors
CRC Press is an imprint of Taylor & Francis Group, LLC
Reasonable efforts have been made to publish reliable data and information, but the author and publisher
cannot assume responsibility for the validity of all materials or the consequences of their use. The authors
and publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any future
reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter
invented, including photocopying, microfilming, and recording, or in any information storage or retrieval
system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, access www.copyright.com or
contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-​750-​
8400. For works that are not available on CCC please contact [email protected]
Trademark notice: Product or corporate names may be trademarks or registered trademarks and are used
only for identification and explanation without intent to infringe.
Library of Congress Cataloging‑in‑Publication Data
Names: Mohan, Neeraj, editor.
Title: Artificial intelligence, machine learning, and data science technologies:
future impact and well-being for society 5.0 / edited by
Neeraj Mohan, Ruchi Singla, Priyanka Kaushal, and Seifedine Kadry.
Description: First edition. | Boca Raton: CRC Press, 2022. |
Series: Demystifying technologies for computational excellence |
Includes bibliographical references and index.
Identifiers: LCCN 2021018852 (print) | LCCN 2021018853 (ebook) |
ISBN 9780367720919 (hbk) | ISBN 9780367720926 (pbk) |
ISBN 9781003153405 (ebk)
Subjects: LCSH: Artificial intelligence–Industrial applications. |
Artificial intelligence–Social aspects.
Classification: LCC TA347.A78 A86 2022 (print) |
LCC TA347.A78 (ebook) | DDC 006.3–dc23
LC record available at https://lccn.loc.gov/2021018852
LC ebook record available at https://lccn.loc.gov/2021018853
ISBN: 978-​0-​367-​72091-​9 (hbk)
ISBN: 978-​0-​367-​72092-​6 (pbk)
ISBN: 978-​1-​003-​15340-​5 (ebk)
DOI: 10.1201/​9781003153405
Typeset in Times
by Newgen Publishing UK
v

Contents
Preface.......................................................................................................................vii
Editor Biographies.....................................................................................................ix

Chapter 1 Breast Cancer Diagnosis Using Machine Learning and Fractal

Analysis of Malignancy-​Associated Changes in Buccal
Epithelium..............................................................................................1
Dmitriy Klyushin, Kateryna Golubeva, Natalia Boroday,
and Dmytro Shervarly

Chapter 2 Artificial Intelligence for Sustainable Health Care

Advancements......................................................................................19
Ayesha Banu

Chapter 3 Identification of Lung Cancer Malignancy Using Artificial

Intelligence...........................................................................................37
Vinod Kumar and Brijesh Bakariya

Chapter 4 Deep Learning in Content-​Based Medical Image Retrieval.................73

Harpal Singh, Priyanka Kaushal, Meenakshi Garg, and
Gaurav Dhiman

Chapter 5 Implication of Image Pre-​Processing in Object Detection Using

Machine Learning Techniques.............................................................93
Neeru Mago, Jagmohan Mago, Sonia Mago, and
Rajeev Kumar Dang

Chapter 6 Forecasting Time Series Data Using Artificial Neural Network:

A Review............................................................................................113
Sarbjit Singh, Kulwinder Singh Parmar, Harpreet Kaur, and
Jatinder Kaur

Chapter 7 Internet of Things: An Emerging Paradigm for Social Safety

and Security........................................................................................131
Bhanu Sharma, Krishan Dutt Sharma, and Archana Mantri

v
vi

vi Contents

Chapter 8 Artificial Intelligence: Encouraging Students in Higher

Education to Seize Its Potential..........................................................155
Rajul Kumar, Karan Sehgal, and Ankit Lal Meena

Chapter 9 Thermal Performance of Natural Insulation Materials for Energy

Efficient Buildings..............................................................................173
R.K. Pal, Parveen Goyal, and Shankar Sehgal

Chapter 10 Maximum Power Point Tracking Techniques for PV Framework

under Partial Shaded Conditions: Towards Green Energy for
Smart Cities........................................................................................193
Mohammad Junaid Khan and Anurag Choudhary

Chapter 11 Forecasting Energy Consumption Using Deep Echo State

Networks Optimized with Genetic Algorithm....................................205
V. Dinesh Reddy, Kurinchi Nilavan, G.R. Gangadharan,
and Ugo Fiore

Chapter 12 State-​of-​the-​Art Natural Language Processing Techniques...............219

Tarun Jaiswal, Manju Pandey, and Priyanka Tripathi

Chapter 13 Modern Approaches in HR Analytics towards Predictive

Decision-​Making for Competitive Advantage....................................249
K. Guru, S. Raja, A. Umadevi, M. Ashok, and Kumar Ramasamy

Chapter 14 Digital Transformation: Utilization of Analytics and Machine

Learning in Smart Manufacturing......................................................269
Iman Raeesi Vanani and Akram Shaabani

Chapter 15 A Robust Cyber Security: Challenges and Opportunities..................283

Glimpse Salwan, Priyanka Kaushal, Surbhi Gupta, and
Padmalaya Nayak

Index....................................................................................................................... 299
vii

Preface
This book provides a comprehensive, conceptual, and detailed overview of the wide
range of applications of Artificial Intelligence, Machine Learning, and Data Science.
These technologies are very successful in bringing a revolution across the world.
These technologies have become ubiquitous and greatly contribute to change.
These technologies have an impact on various domains such as healthcare,
business, industry, security etc. Moreover, they are recognized as a tool to deal with
complexities.
The book aims at low-​cost solutions that could be implemented even in developing
countries. It highlights the significant impact these technologies have on various
industries and on us as humans. It provides a virtual picture of forthcoming better
human life shadowed by the new technologies and their applications. The impact of
Data Science on various business applications has been discussed. The book will also
include an overview of the different AI applications and their correlation between
each other.
The book is useful for graduate/​post graduate students, researchers, academicians,
institutions, and professionals who are interested in exploring latest technologies
and advancements in the area of Artificial Intelligence, Machine Learning, and Data
Science.

vii
viii
ix

Editor Biographies
Neeraj Mohan is a passionate researcher and academician. His research interest
areas are Network Traffic Management and Digital Image Processing. He works as
Assistant Professor in Computer Science & Engineering Department in I.K. Gujral
Punjab Technical University, Kapurthala (Punjab) India. He has a rich and quantita-
tive academic experience of 19 years at various positions. He did his doctoral degree
at I.K. Gujral Punjab Technical University, Kapurthala (Punjab) India. He is an active
researcher with more than 50 research papers in reputable journals and conferences.
He has one published patent. He has guided one PhD and 17 M.Tech theses. Two
more Ph.D. thesis are in pipeline.
Ruchi Singla offers a unique skillset and expertise of a PhD in Wireless
Communication from Thapar University, Patiala, with a diploma in Entrepreneurship
from Ashton College, Vancouver, and a certification course in “Growth Strategies and
Management” from IIM, Kolkata. She is the Professor in Department of Electronics
and Communication and is also the Dean Research for the research division. She is
currently coordinating the start-​ups, paper publications, research projects, consult-
ancy, and events related to research and innovation. She has 18 years of experience in
teaching and research domains. Her areas of interest include antennas and biosensors.
She has approximately 35 research publications in reputable international and national
journals having good Impact factor. Additionally, she has three published patents. She
has also worked on various DST funded projects like a greenhouse monitoring system
and electrotiles.
Priyanka Kaushal is Associate Professor in Applied Sciences department at
Chandigarh Engineering College, Landran (Mohali), Punjab, India. She did her
master’s degree at Himachal Pradesh University, Shimla, India and Doctoral degree
from IK Gujral Punjab Technical University, Punjab, India. She has published 20+
research papers in reputable journals and conferences. She has delivered many talks in
seminars and workshops. She has acted as reviewer for many reputable journals such
as Advanced Materials Science (SCI Indexed), Mechanics of Advanced Materials and
Structures (Taylor & Francis, SCI indexed), Materials Today: Proceedings (Elsevier,
SCI Indexed) etc. She has two published patents.
Seifedine Kadry gained a Bachelor degree in Applied Mathematics in 1999 from
Lebanese University, MS degree in Computation in 2002 from Reims University
(France) and EPFL (Lausanne), PhD in 2007 from Blaise Pascal University (France),
HDR degree in Engineering Science in 2017 from Rouen University. At present
his research focuses on education using technology, system simulation, operation
research, system prognostics, stochastic systems, and probability and reliability ana-
lysis. He is an ABET program evaluator. He is the editor-​in-​chief of Research Journal
of Mathematics and Statistics and ARPN Journal of Systems and Software. He is the
associate editor of International Journal Of Applied Sciences (IJAS) and editorial

ix
newgenprepdf

x Editor Biographies

board member of International Journal of Mathematics and Statistics, Journal of

Theoretical and Applied Information Technology and International Journal of
Engineering and Applied Sciences. He has published more than 270 research items,
which have been cited 1700 times, his h-​index is 18 and i10-​index is 40.
newgenprepdf

1 Breast Cancer Diagnosis

Using Machine
Learning and Fractal
Analysis of Malignancy-​
Associated Changes
in Buccal Epithelium
Dmitriy Klyushin, Kateryna Golubeva,
Natalia Boroday, and Dmytro Shervarly

CONTENTS
1.1 Introduction........................................................................................................1
1.2 Malignant-​Associated Changes in Buccal Epithelium.......................................2
1.3 Materials and Methods of Morphometric Research and Image Analysis..........5
1.4 Fractal Analysis of Chromatin............................................................................8
1.4.1 The Overall Algorithm for the Screening of Breast Cancer.................12
1.5 Results and Discussion.....................................................................................13
1.6 Conclusion and Future Scope...........................................................................15
References..................................................................................................................15

1.1 
INTRODUCTION
Currently, there is a constant increase in the incidence and death of breast cancer in
women around the world. Consequently, the problem of early diagnosis and screening
of breast cancer is very urgent. Early diagnosis of breast cancer involves the examin-
ation of large populations, so the screening method must be highly sensitive, specific
and safe.
The currently accepted “gold standardˮ for breast cancer diagnostics includes clin-
ical examination, mammography and aspiration biopsy. It allows for high diagnostic
accuracy, but mammography implies radiation exposure, and aspiration biopsy is
associated with tumor injury. This is contrary to the requirement for screening to be

DOI: 10.1201/9781003153405-1 1
2

2 AI, Machine Learning, and Data Science

safe, so it is highly desirable to develop an effective screening method that is non-​

invasive and harmless. As such, we propose to use a non-​invasive study of malignancy-​
associated changes (MAC) in the interphase nuclei of the buccal epithelium.
The purpose of the chapter is to describe a novel effective method for the screening
of breast cancer based on the investigation of fractal properties of chromatin in
Feulgen-​stained nuclei of buccal epithelium using machine learning. The chapter
consists of six sections. Section 1.1 is the introduction where the aim of the chapter
and importance of the proposed method are described. Section 1.2 contains a short
survey of papers on the malignancy-​associated changes. In Section 1.3 we describe
the morphometric research and image analysis of MAC in buccal epithelium. Section
1.4 describes the fractal analysis of the chromatin. Section 1.5 contains the results of
diagnosis. Section 1.6 concludes our study and states some open problems.

1.2 
MALIGNANT-​ASSOCIATED CHANGES IN BUCCAL
EPITHELIUM
The first reports on malignancy-​associated changes (MAC) occurred in the 1960s,
when the content of X-​chromatin in somatic cells was widely studied and its lability
was revealed during various functional changes in the body and general somatic path-
ology. In the presence of a tumor in the body, there are significant changes in the
content of X-​chromatin in the buccal epithelium and neutrophils of peripheral blood.
It was shown that changes in the number of cells with X-​chromatin are caused by
disorders of the functional state of the heterocyclic X-​chromosome.
Of particular interest are works showing changes in the epitheliocytes of the buccal
epithelium in patients with tumors. Thus, in the 1960s H. Nieburgs and his co-​authors
(Nieburgs et al. 1962, Nieburgs 1968) reported a characteristic redistribution of chro-
matin masses in somatic cells in 77% of cancer patients and called these changes
tumor-​associated changes. The latter were characterized by an increase in the size of
the nuclei of epitheliocytes, an increase in the size of the zones of “boundedˮ chro-
matin, which were surrounded by light zones. The same changes were observed in
the cells of the liver, kidneys and other organs. In the paper (Obrapalska et al. 1973) it
was reported that MAC was observed in the buccal epithelium of 74% of patients with
malignant tumors. An increase in the content of DNA in the nuclei of epitheliocytes
in patients with malignant melanoma in comparison with almost healthy women has
been shown. At the same time, a decrease in the number of chromatin positive cells
(X-​chromatin) was found in patients with malignant melanoma compared with that
in patients with benign nevi and in controls. An increase in the content of DNA and
the size of the interphase nuclei of the buccal epithelium was found in patients with
breast cancer. But some authors in cytospectrophotometric detection of the amount
of DNA in the epitheliocytes of buccal epithelium in men with bronchial epithelioma
did not find a significant difference between this indicator in sick and almost healthy
men (Ogden et al. 1990).
Later, in the 1990s, in screening examinations of the population, in experimental
conditions, and in a medical clinic the buccal epithelium of the oral cavity was used
as a convenient object of study to detect early forms of disease (Rathbone et al. 1994,
Rosin et al. 1994, Prasad et al. 1995, etc.). This object reflects the general health
3

Breast Cancer Diagnosis: Machine Learning 3

status of a person. Some researchers on the percentage of electronegative nuclei of

this epithelium and the velocity of the nuclei during microelectrophoresis revealed the
functional state of man and his biological age, the development of fatigue, the action
of harmful environmental factors, the state of the periodontium. Other researchers
have evaluated the genetic effects of environmental pollution and the genotoxicity of
xenobiotics by accounting for micronuclei in the epitheliocytes of the oral mucosa.
Elevated levels of cells with micronuclei were found in the epithelial exfoliative
cells of the oral cavity of patients with various types of allergies. Under the action
of various genotoxic carcinogens (chewing tobacco in different mixtures, nasal and
betel, tobacco smoking), formaldehyde levels of exfoliative cells with micronuclei can
increase tenfold compared to the control (Nair et al. 1991, Rosin et al. 1992; Tolbert
et al. 1992 etc.). These data suggest that the increase in the level of micronuclei may
be a kind of “dosimeter” of various pathological conditions of the body. The level of
micronuclei in the oral cavity of cancer patients probably increases after antitumor
chemotherapy or ionizing radiation (Tolbert et al. 1992). In exfoliative cells in per-
sons, who consume different chewing gums, which are characterized by carcinogenic
activity (mava, tamol, masher, nuts nat), the level of micronuclei increases (Adhraryn
et al. 1991 etc.). It is very important that a high level of correlation was observed
between the level of exfoliative cells with micronuclei and the number of other cyto-
genetic disorders in peripheral blood lymphocytes (sister chromatid exchanges and
chromosomal aberrations) of subjects exposed to mutagens (Sarto et al. 1990).
The state of chromatin was used to assess the degree of differentiation of buccal
epithelium in gastric and duodenal ulcers. Indicators that characterize the quantita-
tive and functional state of buccal epitheliocytes are also used. This is the index of
maturation, intoxication, differentiation and karyopyknotic index. The change in the
nature of differentiation, which is inherent in the norm of a particular area of buccal
epithelium, indicates local or systemic disorders. The presence of signs of cellular
atypia with a high probability indicates the development of precancerous and tumor
changes in buccal epithelium and in 96% of cases allows you to reliably diagnose
these diseases by cytological methods. Changes in the differentiation of the epithe-
lium of buccal epithelium can also be the result of metabolic and hormonal disorders,
the action of mechanical factors and chemicals (Schonwetter 1995).
In the papers (Palcic et al. 1994 etc.) the authors reported that quantitative cyto-
logical studies of the DNA content and texture of chromatin in the nucleus can detect
changes associated with malignancy, which were called H. Nieburgs’ malignancy-​
associated changes (MAC). These changes were detected in normal cells of macro-
scopically unchanged areas located at some distance from the malignant tumor. They
probably arose as a reaction of normal cells to growth caused by malignant trans-
formation in a particular organ (lungs, cervix, breast). Based on the data obtained,
it was hypothesized that the changes associated with tumors are clearly expressed
near malignant tumors and weakly expressed or absent near tumors that do not have
progressive growth. With the removal of tumors, the changes associated with the
degree of malignancy disappear, while incomplete removal of the tumor does not
significantly affect these changes. The hypothesis was expressed that quantitative
cytospectrophotometry can detect changes associated with malignant growth to diag-
nose these tumors, including early cancers, and to assess the prognosis of this process.
4

4 AI, Machine Learning, and Data Science

In the 1990s, an attempt was made to use changes in the buccal epithelium to char-
acterize the effect of the tumor on its condition. For example, Ogden et al. tried to
characterize and substantiate the possibility of tumor influence on the functional state
of the buccal epithelium in order to use the obtained data to characterize the course of
processes that occurred in organs far from the tumor and identify patterns that char-
acterize the course of these processes. Disturbances expressed by changes in nuclear
material, heterogeneity of chromatin substances and changes in nuclear membranes
were observed in 77% of patients with tumors of different localizations (carcinomas,
lymphomas, seminomas). The criteria for assessing tumor-​associated changes (MAC)
were cytophotometric studies of DNA content, nucleus size and cytoplasm of tumor
cells, and the nature of the distribution of chromatin in the nucleus. But the authors
failed to identify clear patterns inherent in the tumor process, in addition to increasing
the size of the nuclei of tumor cells and changes in the nuclear-​cytoplasmic relation-
ship. However, they did not deny that the detected disorders were related to the effect
of tumors on the functional state of the oral mucosa.
Blood has been considered the traditional method of studying genomic DNA for
genetic analysis, but recently the method using DNA isolated from cells of the oral
mucosa has become more widespread. In benign hyperplastic processes, the number
of Langerhans cells increases significantly compared to their number in the normal
mucous membrane. In malignant tumors, it decreases the more the lower the level of
their differentiation.
Thus, an analysis of the literature showed that the relationship between tumor and
organism is very complex and due to the many connections that form between the
tumor and the body’s control systems (nervous, endocrine and immune) under the
influence of exogenous and endogenous factors, which cause appropriate reactions
from the body. From the general set of indicators of homeostasis of a cancer patient,
important information about the effect of the tumor can be obtained by studying
the functional state of the buccal epithelium, which has been shown to be in close
anatomical and physiological relationship with various organs and systems. This is
confirmed by the fact that a number of diseases of the internal organs are accom-
panied by changes in the mucous membrane, which may appear even earlier than
other clinical symptoms in this pathology.
In 2009 it was discovered that the DNA packing in the cell nuclei has fractal prop-
erties, i.e. the DNA is twisted like a three-​dimensional Peano curve (Lieberman-​Aiden
et al., 2009). The fractal analysis of cells became the field of intensive investigations.
The fractal dimension is considered an effective measure of heterogeneity of cells
of complex endometrial hyperplasia and well-​differentiated endometrioid carcinoma
(Bikou et al, 2016), and also as a prognostic factor for survival in melanoma (Bedin
et al. 2010), leucemia (Adam et al, 2006 etc.) and other diseases (Losa 2012, Metze
2010, 2013). Note also that numerous investigations (Nikolaou and Papamarkos
2002, Ohri et al. 2004, Losa and Castelli 2005 etc.) have shown the significant poten-
tial of fractal dimension for estimation of the morphological data. However, these
investigations were focused on tumor cells but not cells of buccal epithelium. Thus,
we can assume that the fractal properties are reflected in the distribution of the chro-
matin in nuclei of buccal epithelium and a tumor can affect this distribution causing
the malignancy-​associated changes. As we shall see further, our hypothesis holds.
5

Breast Cancer Diagnosis: Machine Learning 5

FIGURE 1.1 Distribution of chromatin in nuclei of buccal epithelium (1—​heterochromatin,

2—​euchromatin).

Registration of the characteristics of the interphase nucleus makes it possible to

assess functional changes in the genetic apparatus of cells, and indicators of the struc-
tural organization of chromatin substance can be used as markers of these disorders
in various pathological conditions, including tumors. There are also studies that indi-
cate that the study of the interphase nuclei of the buccal epithelium may answer the
question of the presence of a malignant tumor in the human body (Andrushkiw et al.
2007 etc.).
DNA cytophotometry is based on the established fact that the average amount of
DNA in the nucleus in normal cells corresponds to a certain number of chromosomes,
which is characteristic of this species. A number of researchers on various tumors
have shown that an increase in the average DNA content in the nucleus is a charac-
teristic feature of tumor cells (Boroday et al. 2016). Therefore, the increase in het-
erogeneity in DNA content can be considered as a diagnostic sign of the tumor and a
criterion for its malignancy.
It should be noted that in the nucleus chromatin can be in a condensed (compact,
tightly packed) and decondensed state. The degree of chromosome decondensation
can be different. Areas of complete decondensation are called euchromatin. Areas of
chromatin that are in a condensed state during the cell cycle are called heterochro-
matin. Due to the fact that heterochromatin is intensely stained with special dyes,
it can be easily observed under a light microscope. At the same time, decondensed
chromatin corresponds to poorly colored areas (Figure 1.1).
Changes in the nuclei are noticeable, but it is very difficult to make a visual
assessment, especially when it comes to dozens of cells. Therefore, digital analysis
should be used for objective quantification.

1.3 
MATERIALS AND METHODS OF MORPHOMETRIC RESEARCH
AND IMAGE ANALYSIS
We have studied the control group (29 people), the group of patients with breast cancer
of the second stage (68 patients) and the group of patients with fibroadenomatosis (33
patients). All diagnoses were verified histologically. The morphological dataset consists
of 20,256 images of interphase nuclei of buccal epithelium (6,752 nuclei scanned in
three variants: without filter, through a yellow filter and through a violet filter).
6

6 AI, Machine Learning, and Data Science

The morphological materials are smears of epitheliocytes of the oral mucosa

from the average depth of the spinous layer, dried at room temperature, fixated in
the Nikiforov mixture and Feulgen-​stained with cold hydrolysis in 5 n HCl for 15
min at t = 21–​22oC. The Feulgen-​stained chromatin was analyzed using the Olympus
analyzer, consisting of the Olympus BX microscope, Camedia C-​5050 digital zoom
camera and a computer. On average, every preparation consists of 52 cells in every
preparation. The DNA-​fuchsine content in the nuclei of the epitheliocytes was
computed as a product of the optical density by area. As a result of the first stage of
analysis, we obtained images of the chromatin distribution in the form of a matrix
containing 128×128 pixels.
Trying to reflect the fractal character of chromatin distribution and to provide the
invariance with respect to rotation of the image, we constructed a space-​filling curve
(Sagan, 1994) passing through every pixel of an image and sequentially read the RGB
values of the colors of the pixels of the scan not line by line. As a result, we can map a
matrix of pixels to three vectors corresponding to the three channels of the RGB color
model. As a space-​filling curve we used the Serpinski curve (Sagan 1994).
Before applying the methods of fractal image analysis, it must be pre-​processed.
To do this, we used the Otsu method. This method is used to perform threshold
binarization of halftone images. The algorithm assumes the presence of two classes
of pixels (main and background) in the image and looks for the optimal threshold that
divides them into two classes so that their intraclass variance was minimal.
This principle of the method was used to select kernels. Two clusters were
considered: kernel pixels and background pixels. We will explain the method in more
detail. Suppose there is an image, each point of which is represented by the value of
the intensity of gray. So we have N image points and L grayscales. Assume that all
points of the image can be divided into two classes, C1 and C2, which will distinguish
points with intensities. Then the optimal value of the threshold T will be the value of
k, at which the interclass variance reaches a maximum value:

T =k .
σC2 C max
1 2

The interclass variance σC2 C is computed as

1 2

σC2 C = ω1ω 2 (µ 2 − µ1 )2 ,
1 2

where ω1 is the probability to get the class C1 , ω 2 is the probability to get the
class C2 , µ1 is the mean brightness in the class C1 , µ 2 is the mean brightness in the
class C2 .
To calculate the interclass variance, you must first construct an image histogram,
i.e. a distribution that determines the number of ni points with intensity I, from the
total number of image points N. To calculate the probabilities and fall into classes C1
and C2 , accordingly, use the formulas:
7

Breast Cancer Diagnosis: Machine Learning 7

k
ω1 = ∑ pi , ω 2 = 1 − ω1 ,
i =1

where pi is the distribution pi = ni / N .

The mean brightness µ1 and µ 2 in C1 and C2 respectively are computed as

1 k 1 L
µ1 = ∑ ip , µ =
ω1 i =1 i 2 ω 2
∑ ip . i
i = k +1

Next, sequentially comparing the values of the intensities of each pixel of the
image with the selected threshold T determines the affiliation of the point to
the corresponding class. All points that belong to the class C2 will be painted white.
The selection of the kernel by means of adaptive threshold image processing proved
to be the best. The contour of the nucleus stands out quite clearly (Figure 1.2).
In the presence of any artifacts on the background of the image, the selection of
the kernel may not be quite correct. When we talk about the image of the nuclei of
the buccal epithelium, the artifacts’ mean, for example, contamination—​small spots
of black or just dark color, painted background—​occurs in cases where the drugs
were poorly washed under running water; is an area of different regions (from a few
pixels to several tens). The color brightness of such areas is similar to the brightness
of the nucleus itself, so in the process of selecting the nucleus, these areas remain in
the background. On the other hand, when selecting the contour of the nucleus, it is
necessary to take into account such a feature as the heterogeneity of the distribution
of chromatin in the nucleus. In practice, this means that in the core itself among the
dark areas can be very light, which in terms of its brightness may even coincide with
the average brightness of the general background.
Therefore, in the case when in the process of contour selection some areas in the
kernel image were initialized as a background and accordingly painted white, it is
necessary to restore the original color but only for the specified areas. This process is
meant by smoothing the selected nuclei.
The idea of algorithms in both described cases is very similar. For example, to
remove artifacts, the pixels of the image are sequentially considered first horizontally

FIGURE 1.2 Nuclei after adaptive threshold pre-​processing.

8

8 AI, Machine Learning, and Data Science

FIGURE 1.3 Artifact removing.

and then vertically, and there are areas whose area is less than a certain value of ε. If
there is a small area of non-​white color in the middle of the background, it means that
you have come across an artifact, so we paint it white. After viewing all the rows, and
then the columns of the image, we get rid of all the artifacts (Figure 1.3). It should be
noted that this algorithm is not designed to distinguish large area artifacts, in which
case you can use the selection of the kernel in manual mode.
Just like the artifact removal algorithm, the white spot search algorithm works
on the kernel. If there is a small area of white among the nucleus, and it is known in
advance that there can be no artifacts on the nucleus (this is one of the requirements
when choosing nuclei), then we restore the original color in this area. Thus, areas are
searched first in the rows of the image, and then in the columns (Figure 1.4).
After selecting the kernel contour, as well as after the operation of smoothing and
removing artifacts, the kernels are ready for further use.

1.4 
FRACTAL ANALYSIS OF CHROMATIN
There are several methods of computation of the fractal dimension of an image. We
selected the Hurst exponent because it is very suitable for analysis of sequential. The
9

Breast Cancer Diagnosis: Machine Learning 9

FIGURE 1.4 Nucleus smoothing.

Hurst exponent is connected with the fractal dimension D by the formula H = 2 –​D.
The algorithm of computation of the Hurst exponent is the following (Butakov and
Grakovskiy 2005).

1. Compute the deviation of the values of the data sequence from the mean in
current segment of a sequence:
m
δ m, N = ∑ ( xi − x N ), (1.1)
i =1

where N is the length of the segment (from 2 up to the end the sequence),
m is the upper limit of summing (from 1 to N −1), xi is a value of the
data sequence, x N is the mean of the segment. As a result, we have N −1
values δ 2, N ,..., δ N −1, N .

2. Compute the range of deviations using (1.1):

R = max δ m, N − min δ m, N (1.2)

m = 2 ,..., N m = 2 ,..., N
10

10 AI, Machine Learning, and Data Science

3. Normalize the range of deviation using (1.2):

R
Q= , (1.3)
s

where s is the standard deviation of the data sequence.

4. Compute lg Q and lg N using (1.3) and the linear graph of the dependence of
lg Q on lg N.
5. Compute the Hurst exponent as the tangent of the slope angle of the line
approximating the dependence of lg Q on lg N.

According to the Hurst exponent, we can classify data sequence by their chaotic
properties. If 0 < H < 0.5, a sequence is considered as ergodic, i.e. if the sequence
has increased in the previous segment, it is highly likely that it will decrease in
the next segment, and vice versa (Figure 1.5). If H = 0.5, then the sequence is
chaotic, i.e. their values do not influence on subsequent values (Figure 1.6). If
0.5 < H < 1.0, then the sequence is trend-​stable. If the sequence in the previous
segment increases or decreases in the previous segment, it is highly likely that it
will save this trend in the next segment (Figure 1.7). If H > 1, then a sequence is
a fractal random process having independent amplitude jumps, using Lévy distri-
bution (Figure 1.8).
We found the H values for each cell using the red, blue, and green channels. To
describe statistical properties of the Hurst exponent we calculated for every patient
the mean, the maximum and the minimum value of the Hurst exponents. Therefore,
we have minBH, minGH, minRH (minimums of Hurst exponents for three channels
among all patient cells), maxBH, maxGH, maxRH (maximums of Hurst exponents for
three channels among all patient cells), and averBH, averGH, averRH (means of Hurst
exponents for three channels among all patient cells).
As a result of analysis of the Hurst exponent, we have discovered that the most
informative is the blue channel. That is why in addition to the first group of features

FIGURE 1.5 An example of ergodic time series.

11

Breast Cancer Diagnosis: Machine Learning 11

FIGURE 1.6 An example of chaotic time series.

FIGURE 1.7 An example of trend-​stable time series.

FIGURE 1.8 An example of fractal time series.

12

12 AI, Machine Learning, and Data Science

FIGURE 1.9 An example of the CART decision tree.

we found out for every patient the descriptive statistics of the Hurst exponent of all
the cells in the blue channel (let us call it the second group of features):

1. Variance.
2. Mean square.
3. 75th empirical quartile.
4. Median.
5. Harmonic mean.
6. Geometrical mean.
7. Average cut-​off value (including all but 5%).
8. Mean.
9. Excess values.

For classification of the data, we used the classification and regression tree
(CART). The accuracy of the classification was justified by the cross-​validation (one-​
leave-​out). We constructed a decision tree using all but one observation, then tested
this absent observation using the decision tree, and carried out this procedure for
every patient (see example at Figure 1.9).

1.4.1 The Overall Algorithm for the Screening of Breast Cancer

1. If the harmonic mean of the Hurst coefficient is less than or equal to 0.672
then go to 3 else go to 2.
2. If the median of the Hurst coefficient is less than or equal to 0.027 than go to
4 else go to 5.
3. Normal. Stop.
4. Breast cancer. Stop.
13

Breast Cancer Diagnosis: Machine Learning 13

5. If the median of the Hurst coefficient is less than or equal to 0.028 then go to
6 else go to 7.
6. Breast cancer. Stop.
7. If the mean square of the Hurst coefficient is less than or equal to 1.152 then
go to 4 else go to 8.
8. Fibroadenomatosis. Stop.

The complexity of the building of CART decision tree is O(mnlog2n) where m is the
number of the features and n is number the patients (Sani et al., 2018).

1.5 
RESULTS AND DISCUSSION
As far as we considered three channels, it is natural to compare their effectiveness from
the point of view of classification. One of the main properties of such data sequences
are their type. Thus, at the first stage it is useful to classify the color channels by their
Hurst exponents: ergodic, chaotic, trend stable and fractal (Tables 1.1–​1.3).

TABLE 1.1
Classification of data sequences by red component

Type/​group Healthy Breast cancer Fibroadenomatosis

Ergodic 0 0 0
Chaotic 0 0 0
Trend stable 29 54 27
Fractal 0 14 6

TABLE 1.2
Classification of data sequences by green component

Type/​group Healthy Breast cancer Fibroadenomatosis

Ergodic 0 0 0
Chaotic 0 0 0
Trend stable 2 6 6
Fractal 27 62 27

TABLE 1.3
Classification of data sequences by blue component

Type/​group Healthy Breast cancer Fibroadenomatosis

Ergodic 0 0 0
Chaotic 0 0 0
Trend stable 29 43 24
Fractal 0 25 9
14

14 AI, Machine Learning, and Data Science

TABLE 1.4
Pairwise specificity, sensitivity and accuracy of differential diagnosis by the
first group of features in the blue channel

Control vs breast Control vs Breast cancer vs

cancer fibroadenomatosis fibroadenomatosis
Specificity, % 82.76 96.55 63.63
Sensitivity, % 95.59 93.94 52.94
Accuracy, % 91.75 95.16 56.44

TABLE 1.5
Specificity, sensitivity and accuracy of screening by the first group of features
in the blue channel
Control vs breast cancer and fibroadenomatosis
Specificity, % 96.55
Sensitivity, % 94.00
Accuracy, % 94.94

TABLE 1.6
Pairwise specificity, sensitivity and accuracy of differential diagnosis by the
second group of features in the blue channel

Control vs breast Control vs Breast cancer vs

cancer fibroadenomatosis fibroadenomatosis
Specificity, % 93.10 93.10 72.72
Sensitivity, % 86.76 90.91 69.12
Accuracy, % 88.66 91.94 70.30

Tables 1.1 to 1.3 indicate the blue component potentially can allow separation
of the control group from the groups of patients suffering from breast cancer and
fibroadenomatosis, because it provides the most clear distinction between healthy
and breast cancer patients. Thus, further, we shall consider only the blue channel
(Table 1.4).
Since it was not possible to distinguish cancer from fibroadenomatosis using
the method described above, at the second stage the classes of patients with cancer
and fibroadenomatosis were combined into a general class and the classification of
healthy and sick patients was carried out (Table 1.5). This situation is typical for
screening, when among a large population of people it is necessary to recognize a
risk group without making a differential diagnosis of diseases. In this case, all (29)
healthy patients and half of the combined class (51 patients) were taken. The selection
15

Breast Cancer Diagnosis: Machine Learning 15

TABLE 1.7
Specificity, sensitivity and accuracy of screening by the second group of
features in the blue channel

Control vs breast cancer and

fibroadenomatosis
Specificity, % 96.55
Sensitivity, % 94.00
Accuracy, % 94.95

of sick patients was made at random. This was done for reasons of a large number of
sick patients and a small number of healthy ones.
In Tables 1.6 and 1.7 we present similar results obtained for descriptive statistics
of the distribution of Hurst exponent (the second group of the features).
The pairwise classification of “healthy –​cancer” and “healthy –​fibroadenomatosis”
allows us to build a very good model based on the Serpinsky curve. Classification of
all three cell types based on scanning of the Serpinsky curves led to the following
results.

1.6 
CONCLUSION AND FUTURE SCOPE
Our investigation has shown that fractal dimension of the Feulgen-​stained chro-
matin in interphase nuclei of buccal epithelium is an effective biomarker of breast
cancer. The classification method developed on this basis has the excellent accuracy
(94.95%), sensitivity (94%) and specificity (96.5%). For comparison, the sensi-
tivity of the mammography is about 87% (Breast Cancer Surveillance Consortium,
2017) and the specificity after the mammography varies from 93% to 88% (Nelson
et al., 2016) depending on the age (the younger the patient the higher the false
positive rate). Our method allows effective detecting of the presence of malignant
tumors in the human body during screening without invasive procedures. However,
the differential diagnosis of cancer and fibroadenomatosis still remains an open
problem.

REFERENCES
Adam R., Silva R., Pereira F. et al. 2006. The fractal dimension of nuclear chromatin as a
prognostic factor in acute precursor B lymphoblastic leukemia. Cellular Oncology. 28,
pp. 55–​59. doi: 10.1155/​2006/​409593.
Adhraryn S.G., Dave B.J., Trivedi A.H. 1991. Cytogenetic surveillance of tobacco-​areca
nut (mava) chewers, including patient with oral cancers and premalignant conditions.
Mutation Research. 261, 1, pp. 41–​49. doi: 10.1016/​0165-​1218(91)90096-​5.
Andrushkiw R.I., Boroday N.V., Klyushin D.A., Petunin Y.A. Computer-​aided cytogenetic
method of cancer diagnosis. New York: Nova Publishers, 2007. ISBN 10: 1-​59454-​882-​X.
Bedin V. et al. 2010. Fractal dimension of chromatin is an independent prognostic factor for
survival in melanoma. BMC Cancer. 10, 260. doi:10.1186/​1471-​2407-​10-​260.
16

16 AI, Machine Learning, and Data Science

Bikou O. et al. 2016. Fractal dimension as a diagnostic tool of complex endometrial hyper-
plasia and well-​differentiated endometrioid carcinoma. In Vivo. 30, pp. 681–​690.
Boroday N., Chekhun V., Golubeva E. and Klyushin D. 2016. In vitro and in vivo densito-
metric analysis of DNA content and chromatin texture in nuclei of tumor cells under
the influence of a nano composite and magnetic field. Advances in Cancer Research &
Treatment. 2016, 706183. pp. 1–​11, doi: 10.5171/​2016.706183.
Breast Cancer Surveillance Consortium (BCSC). 2017. Sensitivity, specificity, and false
negative rate for 1,682,504 screening mammography examinations from 2007–​
2013.      www.bcsc-​research.org/​statistics/​screening-​performance-​benchmarks/​screening-​
sens-​spec-​false-​negative.
Butakov V., Grakovskiy A. 2005. Evaluation of arbitrary time series stochastic level by Hurst
parameter. Computer Modelling and New Technologies. 9, 2, pp. 27–​32.
Lieberman-​Aiden E. et al. 2009. Comprehensive mapping of long-​range interactions reveals
folding principles of the human Genome. Science. 326, 5959, pp. 289–​193. doi: 10.1126/​
science.1181369.
Losa G., Castelli C. 2005. Nuclear patterns of human breast cancer cells during apoptosis: char-
acterization by fractal dimension and (GLCM) co-​occurrence matrix statistics. Cell and
Tissue Research. 322, pp. 257–​267. doi: 10.1007/​s00441-​005-​0030-​2.
Losa G. 2012. Fractals and their contribution to biology and medicine. Medicographia. 34,
pp. 365–​374.
Metze K. 2010. Fractal dimension of chromatin and cancer prognosis. Epigenomics. 2, 5,
pp. 601–​604. doi: 10.2217/​epi.10.50.
Metze K. 2013. Fractal dimension of chromatin: potential molecular diagnostic applications
for cancer prognosis. Expert Review of Molecular Diagnostics. 13, 7, pp. 719–​735.
doi: 10.1586/​14737159.2013.828889.
Nair U., Obe G., Nair J., Maru G.B. 1991. Evaluation of frequency of micronucleated oral
mucosa cells as a marker for genotoxic damage in chewers of betel quid with or without
tobacco. Mutation Research. 261, 2, pp. 163–​168.
Nelson H.D., Fu R., Cantor A., Pappas M., Daeges M., Humphrey L. 2016. Effectiveness of
breast cancer screening: systematic review and meta-​analysis to update the 2009 U.S.
Preventive Services Task Force Recommendation. Annals of Internal Medicine. 164, 4,
pp. 244–​255. doi: 10.7326/​M15-​0969.
Nieburgs H.E. 1968. Recent progress in the interpretation of malignancy associated changes
(MAC). Acta Cytologica. 12, pp. 445–​453.
Nieburgs H.F., Herman B.E., Reisman H. 1962. Buccal cell changes in patients with malignant
tumors. Laboratory Investigation. 11, 1, pp. 80–​88.
Nikolaou N., Papamarkos N. 2002. Color image retrieval using a fractal signature extrac-
tion technique. Engineering Applications of Artificial Intelligence. 15, 1, pp. 81–​96.
doi: 10.1016/​S0952-​1976(02)00028-​3.
Obrapalska E., Cadel Z., Kostyrka J. 1973. Ocena cytologic zna nablonka jamy ustney u
chorych na nowotwory zlosliwe. Nowotwory. 23, 1/​2, pp. 25–​29 (in Polish).
Ogden G.R., Cowpe J.G., Green M.W. 1990 The effect of distant malignancy upon quantitative
cytologic assessment of normal oral mucosa. Cancer. 65, pp. 477–​480. doi:/​10.1002/​
1097-​0142(19900201)65:3<477::AID-​CNCR2820650317>3.0.CO;2-​G.
Ohri S., Dey P., Nijhawan R. 2004. Fractal dimension in aspiration cytology smears of
breast and cervical lesions. Analytical and Quantitative Cytology and Histology. 26,
pp. 109–​112.
Palcic B. 1994. Nuclear texture: can it be used as a surrogate endpoint biomarker? Journal of
Cellular Biochemistry. 19, 1, pp. 40–​46.
17

Breast Cancer Diagnosis: Machine Learning 17

Prasad M.P., Mukundan M.A., Krishnaswamy K. 1995. Micronuclei and carcinogen DNA
adducts as intermediate end points in nutrient intervention trial of precancerous lesions
in the oral cavity. European Journal of Cancer. 31B(3), pp. 155–​160. doi: 10.1016/​
0964-​1955(95)00013-​8.
Rathbone M.J., Drummond B.K., Tucker I.G. 1994. The oral cavity as a site for systemic drug
delivery. Advanced Drug Delivery Reviews. 13, 1–​2, pp. 1–​23.
Rosin M. 1992. The use of the micronucleus test on exfoliated cells to identify anticlastogenic
action in humans. Mutation Research. 287, 2, pp. 265–​ 276. doi: 10.1016/​
0027-​5107(92)90071-​9.
Rosin M.P., Ragab N.F., Anwar W. et al. 1994. Localized induction of micronuclei in the
oral mucosa of xeroderma pigmentosum patients. Cancer Letters. 81, 1, pp. 39–​45.
doi: 10.1016/​0304-​3835(94)90162-​7.
Sagan H. 1994. Space-​filling curves. —​Springer-​ Verlag: New York–​ Berlin, ISBN
978-​0-​387-​94265-​0.
Sani H.M., Lei C. and Neagu D. 2018. Computational complexity analysis of decision tree
algorithms. In: Bramer M., Petridis M. (eds.) Artificial intelligence XXXV. SGAI 2018.
Lecture Notes in Computer Science. Springer: Cham. 11311: pp. 191–​197. doi: 10.1007/​
978-​3-​030-​04191-​5_​17.
Sarto F., Tomanin R., Giacomelli L. et al. 1990. The micronucleus assay in human exfoliated
cell of the nose and mouth: application to occupational exposures to chromic
acid and ethylene oxide. Mutation Research. 244, 2, pp. 345–​ 351. doi: 10.1016/​
0165-​7992(90)90083-​V.
Schonwetter B.S., Stolzenberg E.D., Zasloff M.A. 1995. Epithelial antibiotics induced
at sites of inflammation. Science. 257, 5204, pp. 1645–​ 1648. doi: 10.1111/​
j.1600-​0757.2010.00373.x.
Tolbert P.E,. Shy C.M., Allen J.W. 1992. Micronuclei and other nuclear anomalies in
buccal smears development. Mutation Research, 271, 1, pp. 69–​77. doi: 10.1016/​
0165-​1161(92)90033-​i.
18
newgenprepdf

19

2 Artificial Intelligence
for Sustainable Health
Care Advancements
Ayesha Banu

CONTENTS
2.1 Introduction......................................................................................................20
2.2 Health Care Data..............................................................................................21
2.2.1 Electronic Health Records....................................................................21
2.2.2 Disease Registries.................................................................................21
2.2.3 Health Surveys.....................................................................................22
2.3 Major AI Technologies Relevant to Health Care..............................................22
2.3.1 Machine Learning................................................................................22
2.3.2 Deep Learning......................................................................................23
2.3.3 Natural Language Processing...............................................................24
2.3.4 Robotics................................................................................................24
2.4 AI for Diagnosis and Treatment.......................................................................25
2.4.1 Recent Applications of AI in Medical Diagnostics..............................25
2.4.2 Chatbots................................................................................................25
2.4.3 Oncology..............................................................................................26
2.4.4 Pathology..............................................................................................26
2.4.5 Radiology.............................................................................................27
2.5 AI for Reducing Diagnostic Errors..................................................................27
2.5.1 Buoy Health..........................................................................................27
2.5.2 PathAI...................................................................................................28
2.5.3 Enlitic...................................................................................................28
2.6 AI for Drug Discovery.....................................................................................28
2.6.1 BioXcel Therapeutics...........................................................................29
2.6.2 BERG Health........................................................................................29
2.7 AI for Medical Transcription............................................................................29
2.8 AI for COVID-​19 Diagnosis............................................................................31
2.9 The Role of Corporations in AI in Health Care...............................................32
2.10 Benefits and Challenges of AI Services in Health Care...................................32
2.11 Looking Ahead—​Future of AI in Health Care.................................................33
References..................................................................................................................34

DOI: 10.1201/9781003153405-2 19
20

20 AI, Machine Learning, and Data Science

2.1 
INTRODUCTION
The possibility of machines to actually think and simulate human behavior was first
introduced by Alan Turing who differentiated humans from machines by developing
the Turing test. Later, John McCarthy coined the term artificial intelligence (AI).
The main focus of AI is to mimic human cognitive functions. [Yoav Mintz, Ronit
Brodie: 2019]. AI has no particular definition that is universally agreed. This term
mostly refers to the multiple technologies of computing such as learning, reasoning,
adaptation, interaction and sensory understanding that bear a resemblance to human
intelligence [https://​epsrc.ukri.org/​research/​ourportfolio/​researchareas/​ait/​]. AI is
sometimes used as the umbrella for machine learning (ML), deep learning (DL) and
robotics, which gives a system the power to learn and harness the experience and
performs some specific tasks faster, better and more efficiently than humans. ML is
also called the superset of DL and also subset of AI as shown in Figure 2.1.
In ML, the system learns iteratively based on its past experiences whereas DL is
the inspiration of human neural network where the system is given huge amounts
of data until it learns by example. This is a more complex form of AI. Robotics is
a machine performing any task beyond human capability with more accuracy and
precision [Ross 2019]. AI is one of the most interesting and upcoming research
areas of this decade, which has totally revolutionized the way humans live and
work today. AI is capable of learning from its own mistakes and performing suit-
able improvements and becomes more useful with use. Currently, there are many
applications of AI that are assisting specialists to increase their efficiency [Kumar
2020]. These applications are more general in education, banking, business, social
media and day-​to-​day life whereas they are more specific in the area of health and
medicine. Recent research in AI and the application of AI algorithms is gaining
increased interest and showing a positive impact on education. The advances in
AI systems like ICAI (intelligent computer-​assisted instruction) open new pos-
sibilities for teaching and learning in the field of education [Chen, Chen and Lin
2020]. In today’s world of e-​commerce, business management is in an epoch of
data, AI applications continuously help in improving efficiency in decision-​making
and overall business operations. AI aims at elevating the reality of extensive data,
promoting business intelligence using complicated algorithms and creating insight
on market trends and consumer behavior, which gives businesses a real competitive
edge [Al-​Zahrani and Marghalani 2018].
AI is now spreading its wings across health care to an extent where AI doctors
may possibly assist human physicians for improvisation in the future. The real

Machine Learning

Deep Learning

FIGURE 2.1 Artificial intelligence—​machine learning—​deep learning.

21

AI for Sustainable Health Care Advancements 21

motivation of AI in the health domain is to learn features and acquire knowledge from
huge volumes of medical data using complicated algorithms, and providing clinical
assistance from the obtained knowledge. AI also has the ability to learn and perform
self-​corrections in order to improve its accuracy on the basis of previous learning
feedback. An AI system provides assistance to physicians by giving updated medical
information from various sources and also helps reduce errors in diagnosis that may
be possible in clinical practice done by humans.
The primary focus of this chapter is on AI algorithms and its applications in health
care systems. The rest of the chapter is organized as follows. Section 2.2 gives an
insight of the different sources for health care data and Section 2.3 describes the
types of AI technologies in relevance to health care system. Section 2.4 focuses on the
application of AI for disease diagnosis and its treatment and section 2.5 emphasizes
on how the diagnostic errors can be reduces with AI. Sections 2.6 and 2.7 explain
the industrial revolution of AI in drug discovery and medical transcription. Section
2.8 gives an insight to how AI has helped to handle the COVID-​19 pandemic. The
benefits and challenges of AI in health care are discussed in Section 2.10 and the
following Section 2.11 puts forward the future of AI in health care.

2.2 
HEALTH CARE DATA
Before an AI system is deployed on any health care application, it must be trained
on medical data such that the system is able to learn and expertise itself. The data
pertaining to health care can be generated from but not limited to clinical activities,
electronic recordings and many other sources like www. fda. gov/​ downloads/​drugs/​
guidances/​ucm328691. pdf.

2.2.1 Electronic Health Records

The clinical data in an electronic form obtained from hospitals, health systems, medical
libraries or repositories is usually referred to as an Electronic Medical Record (EMR).
This data includes demographics, administrative information, medical prescriptions,
laboratory tests and diagnosis, drugs, medical notes, treatment, and doctor and patient
details. There also exist large collaborations like National Institutes of Health (NIH),
De-​identified Clinical Data Repository (DCDR) and the Stanford Center for Clinical
Informatics, which provides access to clinical data repositories.

2.2.2 Disease Registries
These registries are clinical information systems that keep a track of a very
narrow range and important data related to specific chronic disease conditions like
Alzheimer’s, asthma, diabetes, cancer, heart disease and many more. Disease regis-
tries also provide necessary information on patient conditions. The Global Alzheimer’s
Association Interactive Network (GAAIN) provides access to a vast data repository
of Alzheimer’s disease. The National Cardiovascular Data Registry (NCDR) is a suite
of data registries holding the data of cardiovascular care they provide. The National
Trauma Data Bank (NTDB) is the largest trauma registry data.
22

22 AI, Machine Learning, and Data Science

2.2.3 Health Surveys
These surveys are one of the important ways of collecting health care data especially
for the purpose of research. National health surveys are generally conducted to pro-
vide estimates of the most common chronic health conditions, i.e. National Health
and Nutrition Examination Survey (NHANES).

2.3 
MAJOR AI TECHNOLOGIES RELEVANT TO HEALTH CARE
AI is not a single technology, but is rather a combination of multiple technologies.
Many of these technologies are found to be relevant to the field of health care, but the
tasks they support may vary. The AI technologies that are considered highly important
to health care systems are described below.

2.3.1 Machine Learning
ML is a statistical technique that analyses structured data mostly used for identi-
fying patterns in the data or predicts its future. In medical terms these patterns can
either be used in identifying the risk factors for infection or predicting infected
patients in future. This is the most common technology of AI. Structured data is an
organized collection of information with some specific defined purpose. In health
care databases, the structured data is available in the form of patient details, lab inves-
tigation values, demographic data, imaging, genetic and financial information. In
health care applications, the ML algorithms attempt to form clusters from patients’
traits or estimate the probability of a particular disease outcome. There exist different
ML algorithms like logistic regression and decision trees, etc. to accomplish this task
[Jiang et al. 2017].
ML models can be classified into three types for understanding the inputs of data.

(i) The first and most straightforward model is the supervised learning model
(SLM). The inputs for this model are labeled, and can be trained to cor-
rectly map between the inputs and the labels using features and weights of
hidden layers. This model can be exposed to newly recorded data to make
predictions after being trained. The accuracy of this model can be measured
and refined. This learning can be applied for predictive modeling where
relationships can be built taking patient traits as input and the outcomes of
interest.
(ii) The second model is unsupervised learning, where there are no user-​defined
labels. This model has to discover the features on its own from the given
inputs in order to perform the mapping with the outputs. This model has much
less human intervention and is mostly used for extracting interesting features.
The two important learning methods are clustering and principal compo-
nent analysis (PCA). Clustering, groups the data with similar traits into one
cluster and gives cluster labels for the patients. PCA is primarily used for
dimensionality reduction, when the traits of the patient are saved with large
number of dimensions.
23

AI for Sustainable Health Care Advancements 23

Machine Learning

Classification Clustering
& Regression

Labeled Input Unlabeled Input

Supervised Learning Unsupervised Learning

Algorithms Algorithms

-- Decision Tree -- K means Clustering

-- Rule Based -- K mediods Clustering
-- Bayesian Semi Supervised Learning -- Hierarchical
Classification Clustering
-- Neural -- Density Based
Network Clustering
-- SVM -- Gaussain Mixture
-- Nearest -- Hidden Markov Model
Neighbor
-- Linear
Regression

FIGURE 2.2 Machine learning algorithms in health care.

(iii) The third model is Hybrid, which is also called semi-​supervised learning. The
inputs of this model consist of both labeled and unlabeled data. This model
is very close to the method in which human pathologists perform diagnosis
using labeled data, like tumor width or density, and unlabeled data i.e. pre-
vious records [Carpenter 2020].

The major ML algorithms especially used in health care [Jenni and Chris 2019, Jabbar
et.al. 2018, Smith 2020] can be shown in Figure 2.2.

2.3.2 Deep Learning
DL is an extension of neural network model and considered to be the most complex
form of ML. DL can also be called a neural network having multiple layers and
thousands of features hidden within it. This helps in exploring complex non-​linear
patterns of data. In recent days, data is rapidly increasing in its volume and com-
plexity. DL is gaining its popularity in handling such data. Recognizing potentially
cancerous lesions in radiology images is one of the most common applications of DL
in health care. Some other increasing applications of DL include radiomics, oncology-​
oriented image analysis and detection of relevant features in imaging data. In the field
of medical research, the application of DL started in the year 2013 and more than
doubled by 2016 and most of these algorithms are applied in image analysis because
24

24 AI, Machine Learning, and Data Science

images are complex with a high volume. The four major DL algorithms are Deep
Belief Network, Deep Neural Network, Convolutional Neural Network and Recurrent
Neural Network [Davenport and Kalakota 2019].

2.3.3 Natural Language Processing

It comprises several applications like text analysis, speech recognition, translation
and many others for making sense of human language. A very large proportion of
health care data is generated from narrative text like laboratory reports, medical
notes, discharge summaries etc. This data is mostly unstructured and also beyond
understandability for any computer program. In this context, the primary goal of NLP
is to extract meaningful information from these narrative texts and give assistance in
clinical decision-​making. In the domain of health care, the important applications of
NLP include creating, understanding and classifying clinical documents, analyzing
the unstructured medical notes, generating reports for examinations and transcribing
patient interactions. Most of the NLP systems learn repeatedly by reabsorbing the
previous interaction results as feedback to determine the accurate results and the
results that did not meet the required expectations. The two major components of
the NLP system are classification and text processing. Text processing is used to iden-
tify all the disease-​related keywords from the medical notes. A subset of keywords
are further selected and their effect on classification is examined for both normal
and abnormal cases. After applying classification, a set of validated keywords are
generated to improve the structured data and support the process of decision-​making
[Jiang et al. 2017].

2.3.4 Robotics
Robotics has become an important and emerging research field from the last two
decades. Robots are also called machines that are capable of performing various tasks
autonomously and with a greater degree of freedom when compared to humans. The
rapid development of ML and robotics has opened a new means to integrate these
technologies and transform the way of living. Robotics handles the machines specially
designed and programmed to perform labor intensive works. ML makes functioning
possible even if machines are not being programmed to do so. The integration of
these two gives robots the capability to perform tasks on their own and allow ML
tasks to start in supervised fashion and move forward to semi-​supervised and com-
plete with unsupervised learning. In the field of health care and medical industries,
the integrated duties done by robots may include patients’ care, monitoring, e-​health,
rehabilitation, medical interventions and artificial prosthetics [Patel et al. 2017]. As
per the report of “TNO Quality of Life on Robotics for Healthcare” medical robotics
has an enormous value in health care with respect to health, economic and societal
benefits. The solutions offered by robotics are very significant especially for patient
groups. The important application areas of medical robotics include:

• Developing smart medical capsules.

• Surgical robotics for prosthetics.
25

AI for Sustainable Health Care Advancements 25

• Treatment by robots.
• Mental and social therapy assisted by robots.
• Patient monitoring systems using robots.

2.4 
AI FOR DIAGNOSIS AND TREATMENT
Diagnosis of disease and suggestion for treatment has been a focus of AI for the
past five decades. In 1972 Stanford University in California started working on
MYCIN, an early AI program to treat blood infections. MYCIN made an attempt
to diagnose patients on the basis of the symptoms reported and results of medical
tests. This program also suggests extra laboratory tests and gives a probable diag-
nosis. After diagnosing the disease, this program suggests a suitable treatment
explaining the reasons leading to the diagnosis and suggestion of treatment.
This program was found to operate at the same level of proficiency as medical
specialists to identify and treat blood infections [Copeland: 2018]. Practically,
MYCIN was never used in medicine due to legal and ethical issues not because
of its performance weakness. In the mid-​1980s, another medical expert system
named CADUCEUS was launched by the University of Pittsburgh, which worked
on inference engine similar to MYCIN and was intended to improve MYCIN
and focus on blood-​borne infectious bacteria. The CADUCEUS system could
diagnose nearly 1000 different diseases [Banks 1986]. Several such AI medical
systems evolved subsequently, out of which some gained importance and a few
were not clinically practiced.

2.4.1 Recent Applications of AI in Medical Diagnostics

The process of determining the condition of the disease is termed medical diagnosis.
The information required for this medical diagnosis is taken from a patient’s history,
physical examinations and medical tests. AI plays an integral role in evolution of
medical diagnosis. Currently, many of the applications of AI in medical diagnosis
come under the following major categories [Faggella 2020].

2.4.2 Chatbots
Chatbots are computer programs specifically designed to interact with users through
natural language. Chatbots basically work on the principles of AI. These chatbots
have inbuilt conversational agent programs that ask the user a series of questions
regarding their health issues and obtain the symptoms of disease. They also give
recommendations to users regarding different symptoms for clarity. These obtained
symptoms may be potentially ambiguous. Therefore, the chatbots map them with the
documented symptoms and their respective codes from the medical databases. This
helps the chatbots to diagnose the disease. Now the chatbots personalize the diagnosis
for future reference and refer the patient to an appropriate doctor for consultation
in case of any major disease. The chatbots already available for medical diagnosis
are Florence, Your.MD and Babylon. Many surveys done in the area of chatbots for
26

26 AI, Machine Learning, and Data Science

medical diagnosis have proved that they are very good health care providers that are
considerably low in cost [Divya et al. 2018].

2.4.3 Oncology
Oncology is the field of medicine that diagnoses and treats cancer. AI contributes in
resolving several biomedical problems. DL, a subset of AI, extracts features automat-
ically and is more flexible to be applied in different research areas including cancer.
It plays a very important role especially in the early detection of cancer. It was stated
by Jeremy Howard, the CEO and founder of Enlitic that “If cancer can be detected
early, then the probability of survival of the patients is 10 times higher.” DL achieved
highest accuracy in early diagnosis compared to many domain experts. The current
DL application in oncology includes detecting cancer from gene expression data,
which is considered as very complex data due to its high dimensionality. DL extracts
meaningful features from this data and classifies the breast cancer cells. This tech-
nology also extracts genes helpful for cancer prediction and cancer biomarkers, to
detect breast cancer.
DL can also be used for segmenting brain tumors in MR images, with much more
stable results compared to segmenting brain tumors manually by physicians, since it
is prone to errors due to motion and vision. This can also be used to measure tumor
sizes during treatment and also detect new metastases if there are any that may have
been overlooked. The algorithm is made to read more CT and MRI scans of patients
to get more accurate results. Prognosis gives an approximation of how advanced the
cancer is and what the chances of survival are. DL develops a prediction model for
prognosis detection of patients suffering from cancer and receiving treatment. It was
found that the survival predictive power of DL is superior to other prediction models
[Ali 2019].

2.4.4 Pathology
Pathology is a study related to the different causes of disease and their respective
effects. The disease diagnosis is based on laboratory analysis done on body fluids
like urine, blood and also tissues. Pathology plays a vital role in advanced medicine
and suggests suitable treatments to fight viruses. The initial way of pathologists to
diagnose a disease involves images observation under a microscope done manually.
The gold standard for diagnosis in pathology is microscopic morphology, but diag-
nostic variability is the main limitation bearing errors among pathologists. In this case
AI can be introduced in the domain of pathology for consistent and more accurate
diagnosis. To improve the speed and accuracy in diagnoses, DL is used to train the
algorithm and make it capable of image recognition to diagnose tumors. The research
was first started using hundreds of labeled images showing noncancerous as well
as cancerous cells. The results extracted from the labeled regions were used as the
model to train the algorithm. The popular AI algorithms used in pathology are CNN-​
Convolutional Neural Network, SVM-​Support Vector Machine and KNN-​K nearest
neighbors [Faggella: 2020; Chang et al. 2018].
27

AI for Sustainable Health Care Advancements 27

2.4.5 Radiology
AI techniques or algorithms, DL in particular, have shown significant advancement
in image-​recognition. Different methods starting from CNN to auto encoders have
initiated numerous applications in the field of medical image analysis. Traditionally
in radiology the trained physicians evaluate medical images visually in order to
detect and characterize the diseases. Radiological imaging data is continuously
growing at a faster rate compared to number of trained professionals and this has
increased radiologists’ workloads dramatically. Some corporate studies reported that
a radiologist must interpret at least one image for every five seconds. As radiology
involves visual perception and decision-​making, such constrained conditions makes
errors inevitable. Today, AI methods have excelled at recognizing complex patterns
in image data and producing qualitative assessment of radiographic characteristics.
Integration of an AI system with the imaging workflow definitely increases the effi-
ciency, reduces diagnostics errors and achieves the expected objectives by providing
radiologists with all pre-​screened images and identified features. Recently, many clin-
ical tasks have been automated and powered by AI, which shifted radiology from a
perceptual domain to a computable domain evolving radiology in parallel to other
application areas [Hosny et al. 2018].

2.5 
AI FOR REDUCING DIAGNOSTIC ERRORS
In today’s health care systems, the total volume of medical data associated with
every patient is increasing tremendously. Whenever a patient visits any doctor or is
admitted into any hospital or goes for any medical test, the size of data repository
starts increasing. The ability to access this real-​time data and produce accurate diag-
noses without human errors has really become critical. A misdiagnosis of any illness
or other health problems may lead to adverse effects and even patients’ death in some
cases. In this sensitive situation of health care, the application of AI is of great help
to clinicians to lower the error rate. AI changes the way in which data is analyzed and
results are obtained. Today AI is contributing to accurate diagnoses at a faster rate,
and also at a reduced cost. Some popular ways in which AI reduce diagnostics errors
are as follows [Daley 2019].

2.5.1 Buoy Health
This is a symptom checker that uses sophisticated algorithms to diagnose the disease
and suggest treatment to cure the illness. Buoy works in five processes: (1) There is
a chatbot that talks to the patient about the symptoms and difficulties concerned to
health. (2) Later, it gets feedback on the symptoms from cause to severity. (3) Buoy
then chooses the best option for care and guides the patient based on its diagnosis.
(4) After this, with the permission of the patient Buoy can follow up the process and
progress of the treatment. (5) Finally, Buoy adds the patient to a community group
to get in touch with people who have experienced similar symptoms and also help
others from their own experience [www.buoyhealth.com/​]. Harvard Medical School
is the hospital using Buoy’s AI for diagnosis and treatment of patients more quickly.
28

28 AI, Machine Learning, and Data Science

2.5.2 PathAI
PathAI develops ML technology that assists pathologists to make rapid and accurate
diagnoses of cancer. The present goal of PathAI is error reduction in cancer diag-
nosis. It is also providing solutions and methods for individualized medical treatment
to patients. It helps patients to get the benefits of novel therapies. PathAI is currently
collaborating with Bristol-​Myers Squibb the drug developers and the Bill & Melinda
Gates Foundation to share the technology globally and expand AI into other health
care industries [www.pathai.com/​].

2.5.3 Enlitic
Enlitic is a company that is bridging the gap between human and AI to bring the
latest advancements in medical diagnostics. It is developing DL medical tools espe-
cially for radiology. World class radiologists have collaborated with data scientists
to analyze the most comprehensive clinical data. The DL platform makes it more
convenient to analyze unstructured medical data like blood tests reports, radiology
images, patient medical history and genomics. Enlitic is also pioneering new medical
software that gives doctors a better insight to the data and enables them to perform
diagnoses sooner and with more accuracy [www.enlitic.com/​].
Freenome is a biotechnology company that uses AI in screenings and diag-
nostics. It is a pioneer in early cancer detection and also the multiomics platform
[www.freenome.com/​]. Zebra Medical Vision is working on transformation of
patient care using the power of AI. It provides AI-​enabled assistance to radiologists
by receiving image scans and analyzing them automatically. It also produces sev-
eral clinical findings. These findings are considered by the radiologists while doing
diagnosis [www.zebra-​med.com/​]. BIDMC-​Beth Israel Deaconess Medical Center
is a teaching hospital of Harvard University, which is working on diagnosing toxic
blood diseases at an early stage using AI. The doctors use sophisticated AI-​enhanced
microscopes to scan the blood samples and identify the harmful bacteria. The entire
process completes faster than manual scanning. The machine is trained to search
for bacteria using 25,000 sample blood images. The machines were found to learn
identifying and predicting harmful bacteria in blood with a 95% accuracy rate
[www.bidmc.org/​].

2.6 
AI FOR DRUG DISCOVERY
AI is one of the prime technologies in the era of the industrial revolution and it is
expected to have a tremendous effect in reducing the cost and time required to dis-
cover new drugs. Many pharmaceutical industries perform the rational drug discovery
through omics and structure-​based drug development. Even though traditional drug
discovery brings huge benefits if successful, it is still costly with high risk and also
time consuming. At present, great changes are observed with the fusion of AI and
previous technologies for drug discovery [Son 2018].
The primary goal in drug discovery is identifying the medicines that can help in
preventing or treating any particular disease. Many of the drugs are small chemically
29

AI for Sustainable Health Care Advancements 29

synthesized molecules. In order to identify these molecules, the traditional approach

is to screen the libraries of molecules to recognize the one which can potentially
become a drug. Later, several rounds of tests are performed to develop the challen-
ging compound. Currently rational structure-​based drug design is becoming the most
common approach which avoids the initial screening phase.
The chemists create new drugs by synthesizing and evaluating several compounds.
In spite of this, it is still unknown which chemical structure has the properties to
become an effective drug and show the desired biological effects. The process of
refining a chemical compound into an effective drug is costly and time consuming
as well. Now experts are showing inclination towards the data processing cap-
abilities of AI to accelerate the process and reduce the cost of new drug discovery
[www.merckgroup.com/​en/​]. Different companies depending on AI for developing
next generation medicines are as follows:

2.6.1 BioXcel Therapeutics
This is a clinical-​stage biopharmaceutical company using AI to identify enhanced
therapies in the field of neuroscience and immuno-​oncology and new drug develop-
ment. Their primary focus is to utilize novel technologies and innovative research for
developing valuable therapeutics that aim at transforming patients’ lives. BioXcel
Therapeutics works on unique AI platforms to reduce the therapeutic development
costs. Several big data and ML algorithms are considered to identify new therapeutic
indices and reduce the time taken for drug development. Additionally, the com-
pany employs AI for drug re-​innovation and identifying the new applications for the
existing drugs [www.bioxceltherapeutics.com/​].

2.6.2 BERG Health
BERG is a clinical-​stage biotechnology company powered by AI using a “Back
to Biology” approach to health care. Their prime focus areas of research include
oncology, neurology and rare diseases. By using the interrogative biology intelli-
gence platform, the goal of BERG is to map diseases and develop treatments. By
combining the AI technology with patient biology, this company aims at picking
up the pace of clinical identification and searching for capable therapeutic targets
to treat disease. Combining the approaches of interrogative biology with the trad-
itional research methods, BERG can develop strong drugs that can fight rare diseases
[www.berghealth.com/​]. Having understood the power of AI in discovering new
medicines, 43 pharma companies are currently using AI [Smith 2020].

2.7 
AI FOR MEDICAL TRANSCRIPTION
Today most of the medical documentation across the globe is stored in Electronic
Health Records (EHR) and online databases are helping clinicians in accessing
patient information quickly. The doctors and surgeons mostly use traditional methods
for documenting the treatment details, patient history or the procedure carried out
for surgeries either writing on a paper or typing in any text editor. The doctors and
Exploring the Variety of Random
Documents with Different Content
Garfield taking a poor half-starved, half-frozen dog from the streets
of Washington to his comfortable home.
General Porter says that he never saw General Grant really angry but
twice in his life—and one of those times was when he saw an army
teamster beating a poor horse. He ordered the teamster to be tied
up and severely punished.
The great Duke of Wellington, who won the battle of Waterloo, was
so kind to the lower creatures that he ordered that special protection
be given a toad in his garden.
It may be worth a thousand dollars to you some day, if you
remember what I am now going to tell you.
It is this: if the time ever comes when you feel as though you hadn’t
a friend in the world and wish that you were dead, go and get some
pet that you can talk to and love and care for—if it is only a little
bird. You will be astonished to find the relief and happiness it will
bring into your life.
—George T. Angell—Adapted.

Blessed are the merciful:

For they shall obtain mercy.—Bible.
 NELLIE’S DOG
I

He was a lonely looking little beggar with a wistful look in his eyes.
Shaggy-haired, with a limp in one leg, and the scars of many stones
on his small body, he was a miserable looking dog as he trotted
down the dusty road. His tongue lolled out and his sides heaved
from panting.
But his rough looks hid a heart of gold. Any one could see that by
looking into his eyes, which were pleading and trustful.
But no one looked into his eyes; they looked only at his shaggy coat
and rough appearance, then shouted and threw stones and clubs at
him. The stones hurt cruelly, and it was a club that put the limp in
his leg, for he was a stray dog and unwelcome everywhere.
He was hungry for some one to love, to live for. His eyes told that
every time he met a stranger, or when, with drooping tail and with
fawning side-steps, he presented himself at some new farmhouse.
But rebuffs and kicks had brought a faint light of distrust and caution
to his eyes, and he began to crawl into the weeds along the roadside
when he saw any one approaching; and when he came to a
farmhouse he would stop at the gate, ready to run at the first hostile
move.
Sometimes people set well-fed home dogs on him. When these were
his size, or smaller, he would back away with teeth half-bared
defensively. He made no move to fight. It was not in his nature to
fight unless he had some one to defend. When the dogs were larger
than himself, he would run as well as the limp in his leg would allow.
Twice he had been overtaken by dogs—huge, fierce fellows that
mauled him without mercy, while their owners encouraged them. But
always they had allowed the little dog to go with his life. Even dogs
have codes of honor.

It was just at sunset one evening when he limped into the yard
where little Nellie was playing. He gazed into her eyes with a
pleading, homesick look, and she smiled. Then she threw her arms
around his neck and caressed him tenderly. He fawned on her in a
very ecstasy of joy, and his scarred, thin little body wagged from end
to end. And so the pact was sealed. He was Nellie’s dog and she was
his mistress.
There was just a trace of Airedale blood in his veins, and an Airedale
dog always selects some one person as the idol of his undying love
and faithfulness. Nellie was to him the one person in the world.

II

Nellie’s father was a big man, and abrupt. He became excited when
he saw her playing with the little dog, and dragged her away. He
declared that she might have been bitten by the cur.
In spite of the little girl’s protests, he kicked the dog from the yard
and stoned him, sending him, a whimpering, heart-broken little piece
of misery, limping down the road.
Nellie cried and declared that she had always wanted a doggy and
that no doggy but the little stray dog would do. But her father was
firm; he would have no stray dogs about the place; there was no
telling what the dog was, or had been—he might be dangerous; for
the father had not looked into the little dog’s eyes as had Nellie.
One day Nellie was taken sick. A raging fever colored her face and
sent her pulses bounding. For many anxious hours her tearful father
watched by her bedside. Then Nellie began calling for her “doggy.”
The doctor, who was already grave, became graver. He told the
father that if Nellie’s doggy was not found he feared that she might
not get better.
 Then She Lay Back on Her Pillow
A great pain came to the father’s heart, and his face twitched in
misery. He would have given all he possessed to have back the little
stray cur to save her—the little dog that he had stoned and sent
whimpering down the road.
Evening came. The doctor, who had been holding little Nellie’s wrist
in his hand, laid it very gently on the bed, a misty look in his eyes.
Suddenly he turned toward the door. There, just within the
threshold, with drooping tail and a loving, pleading look fastened on
the little figure in the bed, stood a stray dog.
The doctor looked into the little dog’s eyes, and understood. He
knew it was Nellie’s doggy. Swiftly he caught the dog up in his arms
and placed him on the bed.
With a glad cry little Nellie half-raised herself from her pillow, as her
hands found the dog’s shaggy hair and felt the warm touch of his
tongue. Then she lay back on her pillow, a new color in her cheeks
and a new light in her eyes. She breathed easily and sighed
contentedly. The doctor smiled tenderly and her father cried tears of
joy.
The little dog curled himself against Nellie and licked her hand
lovingly, for of all the people in the world she alone was his mistress,
and neither kicks nor stones could keep him away.
—Robert E. Hewes.

There is joy in caring

For helpless little things.
 WHO SAID RATS?
If every rat costs the public one dollar per year, what do you
suppose a kitten is worth?
Cats are nature’s destroyers of rats. Rats devour much good food
and carry dreadful diseases.
No trap or poison bait can do pussy’s work, because rats are very
wise and cunning, and after a few are caught they avoid coming
near a trap. As for poisons, it isn’t very pleasant to have a poisoned
rat die in or near a house.
Rats have very large families and become great-grandfathers in a
short time; so you see the cats have plenty of work.
It is foolish to think, though, that cats can live upon mice and rats.
They need other food, and the better fed cats are generally the best
“mousers.”
Cats are kept in large postoffices to protect the mail from the rats
and mice. They are so valuable for this purpose that Uncle Sam sets
money aside to be used in feeding them.
In a large city in China there is a law which says that in every house
one or more cats must be kept; for there the rats carry a dreadful
disease or plague, and the cats kill the rats.—H. H. Jacobs.

The idea that cats should be poorly fed, in order that they may be
good mousers and ratters, is a very cruel and ignorant one. A cat
catches rats and mice because instinct tells him to do so, and he will
do his work better if strong and well fed.—Mary Craige Yarrow.
 A BRAVE MOTHER
At the burning of an apartment house in Kansas City, early one
morning, the firemen and onlookers were astonished to see a cat
leap in at the door, though she must pass through fire and water to
enter.
Some one called out, “Look at that cat—she must have gone crazy.”
While they watched she returned, bringing a kitten held up as high
as she could lift it by throwing back her head.
She hurried through the crowd, and after a few moments again
appeared, and dashed once more into the flames. Soon the brave
creature came back with another kitten in her mouth. By this time
the people were watching to see what she would do next, for she
was giving a wonderful exhibition of mother love.
When she tried to enter once more there were many cries of “Stop
her—don’t let her go in—it’s sure death!” But she would not be
stopped; she slipped through the crowd and went in again.
The firemen turned their attention now almost wholly to the part of
the building where she was; but the walls fell, and the noble little
self-forgetting mother was buried beneath them.
A search was made for her kittens; they were found in a place of
safety. There were four of them.
The janitor of the building remembered that there had been five.
How well the mother cat knew the number! and how bravely she
had saved them—all but one! Do you suppose she was with it, to
cover it and guard it to the last moment?
The motherless kittens were taken to the central fire station and
tenderly cared for. The firemen had been very eager to own an
Angora kitten; but when one was offered to them, they decided not
to take it.
“It might put these little chaps out if we brought in another cat,”
they said; “and we feel that we ought to take care of them—for their
mother’s sake, you know.”
The little mother had been lifted up to a place of honor with these
men, who knew so well how to value true courage.—H. H. Jacobs.

How far that little candle throws his beams!

So shines a good deed in a naughty world.
—Shakespeare.

QUESTIONS

Is the cat ever a brave animal?

Do animals know when people are kind? How can you be kind
to cats?
Have cats any value?
How does it help birds for you to feed cats?
If you cannot find a home for a stray cat, what should you do?
Do you know where the S. P. C. A. office is?
If you were a cat would you like to be left to starve when the
family moved away?
How do you suppose a mother who loses all her children feels?
How do you suppose a mother dog who loses all her puppies
feels? A mother cat who loses all her kittens?
Wouldn’t it be better to keep one?
Oh, how dare we ask a just God to bestow
The mercy we grant not to creatures below!
 ABOUT THOREAU
Henry D. Thoreau, a famous man who lived for some time in a little
cabin in the woods near Concord, Massachusetts, was noted for his
kindness to all God’s harmless creatures.
It is said that even the fishes came into his hand when he dipped it
into the stream.
The little mice would come and playfully eat from his fingers, and
the very moles paid him friendly visits.
Sparrows alighted on his shoulders when he called them; phœbe
birds built their nests in his shed; and the wild partridge with her
brood, came and fed quietly beneath his window, as he sat and
looked at them.
After he had been two or three months in the woods the wild birds
ceased to be afraid of him, and would come and perch on his
shoulder, and sometimes on his spade when he was digging.—
George T. Angell.
 FAIR PLAY FOR OUR WILD ANIMALS
Did you ever see a buffalo in the park?
Do you know that there are only a few of them in this country
because years ago people hunted them for sport, and killed
them by thousands?
Do you know that there is danger that the wild duck, reindeer,
and mountain sheep will disappear in the same way?
Do you think it is a fair game to hunt animals with a gun? Can
they take their part?
Don’t you think that boys and girls who live a hundred years
from now will be glad if we all try to protect our wild animals?
Will you bring some of your favorite pictures of animals to
school?
 THE TRUE STORY OF PEDRO
“Three balls for five cents, Mister. Have a shot at the monk.”
Young Mr. Williams wondered what the man meant as he held out
three hard balls painted in bright colors. Mr. Williams had gone to
Woodlyn Park with his little niece and nephew for an afternoon of
pleasure.
“Oh, look, uncle!” cried the little boy; “don’t you see the monkey?
There! see his face through the hole in that sheet? The men throw
the balls at him.”
Mr. Williams did, indeed, see Pedro’s poor little scared face. Just as
he caught sight of it, bang! a man threw a ball that hit the monkey
on the head.
“Oh, I am afraid they will kill the poor little monkey,” cried the little
girl. “Can’t we make them stop, uncle?”
“There isn’t much use in talking to these men,” said Mr. Williams.
“The best thing to do is to notify the ‘cruelty lady.’ We will do that as
soon as we get home; shall we?”
“Oh, yes, indeed,” cried the children. “Let us go right away.”
The “cruelty lady” started immediately for the park. When she saw
how the little monkey was being abused, she had the two men who
owned him arrested. Although they were very angry, she was not
afraid of them.
The poor monkey was taken away from his cruel masters and carried
to the home of a good woman, who cared for the sick little animal.
For days Pedro lay exhausted in the nice soft bed she made for him.
She bathed the many, many bruises on his poor little body, and fed
him good wholesome food.
In about ten days, Pedro began to feel better, and showed how
much he appreciated the kindness of his new mistress by following
her everywhere he could.
He got into mischief, too, by trying to do everything he saw people
do. One day, when his mistress had stepped out into the garden, he
turned the key on the inside of the door, and locked her out.
It was a good thing that one of the second-story windows was open,
so that a young man could climb up and get inside and unlock the
door.
The last time I heard of Pedro, he was living happily with other
monkeys in the Zoological Garden.

Let us speak for those

That cannot speak for themselves.

Even the smallest kind act is never lost. It isn’t always the size of the
good deed that counts.
 WHAT CHILDREN CAN DO
They can speak pleasantly to boys or men who are whipping their
horses and ask them to stop.
They can ask men and boys to blanket their horses in cold weather,
to put them in the shade in warm weather, and to loosen the tight
check rein.
They can keep fresh water where their own horses, dogs, cats, and
hens can get it, not once, but at any time during the day or night.
They can feed their dog or cat morning and night, remembering that
all animals and fowls have as good appetites and suffer as much
when hungry and thirsty as boys and girls do.
They can see that all their animals are made comfortable at night
and never turned out in the cold.
They can feed homeless dogs and cats and try to find shelter for
them. They can be careful never to lose a pet animal, or to leave
one behind if they move away, unless they have arranged for some
one to take care of it.
Every kind act that children do, not only makes the world better, but
helps to make them better and happier men and women.—Animal
Rescue League of Boston.
A HORSE’S PETITION TO HIS DRIVER
Up the hill whip me not,
Down the hill hurry me not,
In the stable forget me not,
Of hay and grain rob me not,
Of clean water stint me not,
With sponge and brush neglect me not,
Of soft, dry bed deprive me not,
When sick or cold chill me not,
With bit or rein jerk me not,
And, when angry, strike me not.
—Selected.
 THE HORSE’S POINT OF VIEW
If a horse could talk, he would have many things to say, especially
when winter comes.
He would tell his driver how a frosty bit stings and sears his lips and
tongue when it is thrust into his mouth without first being warmed.
He would tell how it feels to have nothing but ice-cold water to
drink, when he is already shivering from the cold.
He would tell of the bitter winds that frost his sides when he halts,
steaming from exertion, and is tied for hours in an exposed place
without a blanket.
He would talk of slippery streets, and the fear of falling on cruel city
paving-stones. He would tell of the bruised knees and wrenched
joints, the tightened straps and the pain of the driver’s lash, and the
horrible fright of it all.
Yes, the horse would say a good many things if he had the power of
speech. And having horse-sense, he would urge his driver to “play
fair,” not merely for the sake of kindness, but for the sake of keeping
a faithful servant in good condition.
—The American Humane Education Society.

MEMORY GEM

They are slaves who fear to speak

For the helpless and the weak;
They are slaves who dare not be
In the right with two or three.
—Lowell.
Sketched from the seal of the Massachusetts S. P. C. A.
 A MAN WHO KNEW HOW
There was a big team, a mighty load, and a long hill, and we
watched to see what the driver would do.
He was brawny and strong, and he had a kind face. He did not use
the whip. He talked to his horses in a friendly way. He said: “Come
on, Jim,” and “Hi, there, Bill,” and when he reached a steep place he
jumped off the load and walked. Once he put on the brake and gave
the panting animals a much-needed rest. At the top of the hill he
patted the noses of his faithful friends, allowed them to breathe a
bit, and then the big load moved along as easily as you please.—
Selected.
 HOW TO TREAT A HORSE
The only charm I use, boys, is the Golden Rule. Treat a horse as you
would like to be treated yourself. There is never any need for any
one to beat or abuse a horse, for there is no creature living more
faithful and loving, if you are only kind and patient with him. Teach
him to love and have confidence in you, and give him time to find
out what you want, then he will serve you not only willingly but
gladly and proudly. The best charm any man can use with a horse is
kindness. Be kind, gentle, and considerate, and you will soon win his
confidence and can do anything you like with him.
—Uncle Dan.
 THE HORSE’S PRAYER
To you, My Master, I offer my prayer: Feed, water, and care for me,
and, when the day’s work is done, provide me with shelter, a clean,
dry bed, and a stall wide enough for me to lie down in comfort.
Always be kind to me. Talk to me. Your voice often means as much
to me as do the reins. Pet me sometimes, that I may serve you the
more gladly and learn to love you.
Do not jerk the reins, and do not whip me when going up hill. Never
strike, beat, or kick me when I do not understand what you want,
but give me a chance to understand you.
Watch me, and if I fail to do your bidding, see if something is not
wrong with my harness or my feet.
Do not check me so that I cannot have the free use of my head. If
you insist that I wear blinders, so that I cannot see behind me as it
was intended that I should, I pray you be careful that the blinders
stand well out from my eyes.
Do not overload me, or hitch me where water will drip on me. Keep
me well shod.
Examine my teeth when I do not eat; I may have an ulcerated tooth,
and that, you know, is very painful.
Do not tie my head in an unnatural position, or take away my best
defense against flies and mosquitoes by cutting off my tail.
I cannot tell you when I am thirsty, so give me clean cool water
often. I cannot tell you in words when I am sick, so watch me, that
by signs you may know my condition.
Give me all possible shelter from the hot sun, and put a blanket on
me, not when I am working, but when I am standing in the cold.
Never put a frosty bit in my mouth; first warm it by holding it a
moment in your hands.
I try to carry you and your burdens without a murmur, and wait
patiently for you during long hours of the day and night. Without the
power to choose my shoes or path, I sometimes fall on the hard
pavements, and I must be ready at any moment to lose my life in
your service.
And finally, O My Master, when my useful strength is gone, do not
turn me out to starve or freeze, or sell me to some cruel owner, to
be slowly tortured and starved to death; but take my life in the
kindest way, and your God will reward you here and hereafter.
Amen.—Issued by the Ohio Humane Society.

Suppose this Horse Takes Cold and Suffers and Dies—Who is to

Blame?

QUESTIONS

What animal has been one of mankind’s most faithful servants

and one of his very best friends?
Can you explain the difference between the work horse, the
saddle horse, the race horse, the fire horse?
Should we have as fine a city if there had never been any
horses?
Can you think of some ways of repaying horses for the work
they do for us?
When you grow old, how do you want to be treated, especially
if you have worked hard all your life?
How do you think old horses should be treated?
Did you ever have a hard fall on the ice?
How did you walk afterward?

Could you have walked carefully if you had been going down hill
holding back a heavy load?
How can we help horses in slippery weather?
How can we help them up when they fall?
Why is it harder to start to move a load than it is to keep it
moving after it is started?
Do you know that many horses are made blind by being over-
driven?
Tell a story showing how intelligent horses are.
Have you ever seen the non-slip chains which drivers can put on
horses’ feet to keep them from slipping?
Have you ever read “Black Beauty,” or “Beautiful Joe,” or “Our
Goldmine at Hollyhurst,” or “The Strike at Shane’s”?

II

What are the names of some of the great societies founded to

protect animals?
 How can children help in such work?
Do children who are kind to animals turn out better than those
who are cruel to them? Why?
If you had your choice as to what animal you would be, would
you choose to be a horse?
Does it make any difference to you whether any one else is cold
and hungry and tired and suffering? Whether an animal is
suffering?
Will you tell your teacher the next time you help some animal?
Tell some of the things you think a horse would say if he could
talk. A dog. A cat. A monkey. A bird.
Where does a horse get his strength? Should he not be well
fed?

Children should never feel that their hands are too small and weak
to help toward making the world a happier place for all to live in, for
the world needs their work quite as much as it does that of the older
people.—M. C. Yarrow.
 BIRDS AS THE FRIENDS OF PLANTS
I

“Just listen, Mary Frances!” said Billy, pulling a paper out of his
pocket.
“‘One robin has been known to feed his family five yards of worms a
day.
“‘A chickadee will dispose of 5,500 eggs of the canker-worm moth in
one day.
“‘A flicker eats no less than 9,000 ants a day.
“‘A pair of wrens have been seen to carry one hundred insects to
their young in an hour. They are especially fond of plant-lice and
cutworms.
“‘Little humming-birds lick plant-lice off foliage with lightning
rapidity.
“‘The yellow-billed cuckoo eats hundreds of tent caterpillars in a day.
“‘Seed-eating birds destroy millions of seeds of troublesome weeds—
actually eating hundreds of tons of seeds.’
“How do people know what the different birds eat?” asked Mary
Frances. “Did some one watch to see what each different bird took
for a meal?”
“No;” Billy referred to his clipping. “Scientists have examined the
contents of the stomachs of the birds, and have learned what food
each kind of bird uses. There was a time when people imagined that
robins stole so many cherries and berries that it was a good deed to
kill them. Now they have found that they destroy so many injurious
insects that they do not begrudge them a few cherries. Besides, if
mulberry trees are planted nearby they will prefer their fruit to the
cherries.”
“Oh, Billy,” cried Mary Frances, “isn’t it wonderful! Not only do birds
help us by destroying harmful insects and seeds, but they help us by
their beauty. I believe they are the most beautiful of living things!
They could have helped us just as much and have been as ugly as—
cutworms.”
“Yes,” replied Billy, “I believe that is so; but it takes a girl to think
such things out. The strangest thing to me, however, is that without
birds we should die of starvation. This paper says that if the birds
disappeared entirely, agriculture and farming would be impossible
within a few years.”

II

“Bees and birds,” said Mary Frances softly, “keep us from starving.
How wonderful it all seems. Why, Billy, it must have all been planned
out when God made the world!”
“I have thought of that myself, Mary Frances,” said Billy; “it’s one of
those thoughts a fellow doesn’t often speak out loud. I don’t know
why.”
“Everybody ought to take care of birds,” went on Mary Frances.
“Surely the reason they don’t is because they do not understand
how wonderfully they help us. Birds and bees keep us from starving.
Oh, Billy, let’s have lots of birds in our garden!”
“Why, how?” asked Billy. “Perhaps we could put food out for them.”
“Yes, but I wasn’t thinking of that. I thought maybe we could put
houses where they would build their nests.”
“Of course,” replied Billy; “and we could keep a small bathtub full of
water for them.”
“What fun!” cried Mary Frances. “Billy, do you know how to build the
right kind of houses for each different kind of bird?”
“No, I do not,” answered Billy; “I know of only a few. They are the
ones our manual training teacher showed us. I have some pictures