AI for Radiology Oge Marques

Visit to download the full and correct content document:

https://ebookmass.com/product/ai-for-radiology-oge-marques/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

AI For Lawyers Noah Waisberg

https://ebookmass.com/product/ai-for-lawyers-noah-waisberg/

Pediatric Emergency Radiology Dietrich

https://ebookmass.com/product/pediatric-emergency-radiology-
dietrich/

Microsoft Azure AI Fundamentals Certification

Companion: Guide to Prepare for the AI-900 Exam Krunal
S. Trivedi

https://ebookmass.com/product/microsoft-azure-ai-fundamentals-
certification-companion-guide-to-prepare-for-the-ai-900-exam-
krunal-s-trivedi/

Microsoft Azure AI Fundamentals Certification

Companion: Guide to Prepare for the AI-900 Exam 1st
Edition Krunal S. Trivedi

https://ebookmass.com/product/microsoft-azure-ai-fundamentals-
certification-companion-guide-to-prepare-for-the-ai-900-exam-1st-
edition-krunal-s-trivedi/
Microsoft Azure AI Fundamentals Certification
Companion: Guide to Prepare for the AI-900 Exam 1st
Edition Krunal S. Trivedi

https://ebookmass.com/product/microsoft-azure-ai-fundamentals-
certification-companion-guide-to-prepare-for-the-ai-900-exam-1st-
edition-krunal-s-trivedi-2/

Radiology-Nuclear Medicine Diagnostic Imaging Ali

Gholamrezanezhad

https://ebookmass.com/product/radiology-nuclear-medicine-
diagnostic-imaging-ali-gholamrezanezhad/

Exam Ref AI-900 Microsoft Azure AI Fundamentals Julian

Sharp

https://ebookmass.com/product/exam-ref-ai-900-microsoft-azure-ai-
fundamentals-julian-sharp/

Introdução à Mineração de Dados 1st Edition Leandro

Augusto Da Silva & Sarajane Marques Peres & Clodis
Boscarioli

https://ebookmass.com/product/introducao-a-mineracao-de-
dados-1st-edition-leandro-augusto-da-silva-sarajane-marques-
peres-clodis-boscarioli/

Fundamentals of Musculoskeletal Ultrasound

(Fundamentals Radiology) 3rd Edition

https://ebookmass.com/product/fundamentals-of-musculoskeletal-
ultrasound-fundamentals-radiology-3rd-edition/
 AI for Radiology
Artifcial intelligence (AI) has revolutionized many areas
of medicine and is increasingly being embraced. This book
focuses on the integral role of AI in radiology, shedding
light on how this technology can enhance patient care and
streamline professional workfows.
This book reviews, explains, and contextualizes some
of the most current, practical, and relevant developments
in artifcial intelligence and deep learning in radiology
and medical image analysis. AI for Radiology presents a
balanced viewpoint of the impact of AI in these felds,
underscoring that AI technologies are not intended to
replace radiologists but rather to augment their capabilities,
freeing professionals to focus on more complex cases. This
book guides readers from the basic principles of AI to their
practical applications in radiology, moving from the role of
data in AI to the ethical and regulatory considerations
of using AI in radiology and concluding with a selection of
resources for further exploration.
This book has been crafted with a diverse readership in
mind. It is a valuable asset for medical professionals eager
to stay up to date with AI developments, computer scientists
curious about AI’s clinical applications, and anyone
interested in the intersection of healthcare and technology.

Oge Marques, PhD, is Professor of Computer Science and

Engineering in the College of Engineering and Computer
Science, Professor of Biomedical Science (Secondary) in the
Charles E. Schmidt College of Medicine, and Professor
of Information Technology (by courtesy), in the College
of Business at Florida Atlantic University (Boca Raton,
FL – USA).
He is the author of 12 technical books, one patent,
and more than 130 refereed scientifc articles on image
processing, medical image analysis, computer vision,
artifcial intelligence, and machine learning. He is a senior
member of both the Institute of Electrical and Electronics
Engineers and the Association for Computing Machinery,
Fellow of the National Institutes of Health AIM-AHEAD
Consortium, Fellow of the Leshner Leadership Institute
of the American Association for the Advancement of
Science, Tau Beta Pi Eminent Engineer, and member of the
honor societies of Sigma Xi, Phi Kappa Phi, and Upsilon Pi
Epsilon.
 AI for Everything

Artificial intelligence (AI) is all around us. From driverless

cars to game winning computers to fraud protection, AI is
already involved in many aspects of life, and its impact will
only continue to grow in future. Many of the world’s most
valuable companies are investing heavily in AI research
and development, and not a day goes by without news of
cutting-edge breakthroughs in AI and robotics.
The AI for Everything series explores the role of AI in con-
temporary life, from cars and aircraft to medicine, educa-
tion, fashion and beyond. Concise and accessible, each book
is written by an expert in the field and will bring the study
and reality of AI to a broad readership including interested
professionals, students, researchers, and lay readers.
AI for Radiology AI for Games
Oge Marques Ian Millington

AI for Immunology AI for Sports

Louis J. Catania
AI for Cars
Josep Aulinas & Hanky
Sjafrie
AI for Digital Warfare
Niklas Hageback & Daniel
Hedblom
AI for Arts
Niklas Hageback & Daniel
Hedblom
AI for Death and
Dying
Maggi Savin-Baden
AI for Creativity
Niklas Hageback
Chris Brady, Karl Tuyls,
Shayegan Omidshafiei
AI for Learning
Carmel Kent & Benedict du
Boulay
AI for the Sustainable
Development Goals
Henrik Skaug Sætra
AI for School Teachers
Rose Luckin. Karine George
& Mutlu Cukurova
AI for Healthcare
Robotics
Eduard Fosch-Villaronga &
Hadassah Drukarch
AI for Physics AI for Finance
Volker Knecht Edward P. K. Tsang

AI for Diversity AI for Scientific Discovery

Roger A. Søraa Janna Hastings

For more information about this series please visit: https://

www.routledge.com/AI-for-Everything/book-series/AIFE
AI for Radiology

Oge Marques, PhD

First edition published 2024
by CRC Press
2385 NW Executive Center Drive, Suite 320, Boca Raton FL 33431
and by CRC Press
4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
CRC Press is an imprint of Taylor & Francis Group, LLC
© 2024 Oge Marques
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, microflming,
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 mpkbookspermissions@tandf.co.uk
Trademark notice: Product or corporate names may be trademarks
or registered trademarks and are used only for identifcation and
explanation without intent to infringe.
ISBN: 978-0-367-62778-2 (hbk)
ISBN: 978-0-367-62725-6 (pbk)
ISBN: 978-1-003-11076-7 (ebk)
DOI: 10.1201/9781003110767
Typeset in Palatino

by Apex CoVantage, LLC

To my beloved Ingrid, with deep gratitude for her unconditional

love, unlimited patience, and constant encouragement

Contents

Foreword ...............................................................................xiii
Preface ..................................................................................... xv
Acknowledgments ............................................................... xix

1 Artificial Intelligence and Medicine:

The Big Picture .................................................................1
1.1 Introduction ............................................................1
1.2 AI: Foundations, Brief History, and
Current State of the Art .........................................2
1.3 AI in Healthcare ...................................................13

2 AI in Radiology: From Fear to Leadership ...............18

2.1 Introduction ..........................................................18
2.2 AI in Radiology: Opportunities and
Applications ..........................................................20
2.3 The AI in Radiology Community Today ..........21

3 Fundamentals of Machine Learning and

Deep Learning................................................................27
3.1 Introduction ..........................................................27
3.2 Machine Learning ................................................28
3.3 Deep Learning ......................................................32
3.4 Selected Deep-Learning Architectures ............36
3.5 Transfer Learning ................................................41
3.6 The Machine-Learning/Deep-Learning
Workflow ...............................................................42

4 Fundamentals of Medical Image Analysis...............46

4.1 Introduction ..........................................................46
4.2 Medical Imaging Modalities Used in
Radiology ..............................................................47
4.3 Medical Image Analysis: Overview ..................58
x Contents

4.4 The Medical Image Processing and

Analysis Pipeline .................................................60
4.5 The Size and Complexity of Medical
Image Analysis using AI.....................................71
4.6 The Future of Medical Image Analysis.............74

5 Data: The Essential Ingredient in AI Solutions ......77

5.1 Introduction ..........................................................77
5.2 Data Types and Sources in Radiology ..............78
5.3 Data Collection .....................................................80
5.4 Data Annotation ...................................................81
5.5 Data Preprocessing ..............................................82
5.6 Data Storage, Governance, and
Management .........................................................83
5.7 Data Quality and Variety ....................................85
5.8 Data Challenges and Considerations ................86
5.9 Interoperability and Standardization ...............89
5.10 Ethics of Data Use in AI ......................................91

6 Clinical Applications of AI in Radiology.................94

6.1 Introduction ..........................................................94
6.2 Interpretative Uses of AI in Radiology .............97
6.3 Noninterpretative Uses of
AI in Radiology ..................................................103

7 Harnessing AI in Radiology Education and

Training .........................................................................109
Michèle Retrouvey and Oge Marques
7.1 Introduction ........................................................109
7.2 Importance of Incorporating AI into
Medical Education ............................................. 110
7.3 Challenges of Incorporating AI into
Medical Education ............................................. 111
7.4 How to Implement AI as an Educator.............112
7.5 Undergraduate Radiology Education.............. 114
7.6 Graduate Radiology Education ........................ 119
Contents xi

7.7 Ethical Considerations.......................................133

7.8 Future Directions and Potential Impacts........134

8 Getting Started with Deep Learning in

Medical Imaging ..........................................................137
8.1 Introduction ........................................................137
8.2 The Machine-Learning/Deep-Learning
Workflow .............................................................138
8.3 Deep Learning for Medical Imaging: A
Recipe for Success ..............................................146

9 The Future of AI in Radiology..................................151

9.1 Introduction ........................................................151
9.2 Ethical and Regulatory Aspects of the Use
of AI in Radiology ..............................................152
9.3 Challenges, Controversies, and
Objections to the Use of AI and
Deep-Learning Models in Radiology .............155
9.4 Emerging Trends and Opportunities..............163

10 Resources for Further Learning................................167

References ...............................................................................179
Index .......................................................................................205
Foreword

In the realm of radiology, a nexus has emerged between

advanced machine learning methodologies and the intri-
cate art of medical imaging. It’s here, at this confluence,
that I had the pleasure of meeting a luminary in the field,
Dr. Oge Marques. Our paths first crossed at a SIIM (Society
for Imaging Informatics in Medicine) meeting, an occa-
sion that would mark the beginning of numerous fruitful
collaborations.
As a neuroradiologist deeply entrenched in the appli-
cations of AI in medicine, my journey has allowed me to
interface with various tools and innovations. But collabo-
rating with Oge has been an education unto itself. With
his robust background in computer science and a prolific
portfolio of publications, Oge doesn’t just utilize AI; he
advances it. His approach, underpinned by rigorous sci-
entific methodology, is a testament to his commitment to
both the precision of computer science and the compassion
intrinsic to healthcare.
The content of this book, judiciously curated by Oge, is
both comprehensive and enlightening. Beginning with
a broad overview of AI, especially its increasing impor-
tance in healthcare, it delves into the specific applications
and opportunities of AI in radiology. The fundamentals
of machine learning and deep learning are laid out with
clarity, making it accessible to both the novice and the sea-
soned practitioner.
What I find particularly commendable is Oge’s insight into
medical image analysis. The chapters are a journey, intro-
ducing readers to various imaging modalities and guid-
ing them through the intricate process of medical image
analysis using advanced AI techniques. But beyond the
technicalities, he acknowledges the core of radiology – the
xiv Foreword

data. With a dedicated segment on data – its collection,

annotation, preprocessing, and the ethical considerations
surrounding its use – Oge accentuates its centrality in the
world of AI in radiology.
This book is not just about the present state of affairs.
It offers a vision, exploring the future trajectories of AI in
radiology, addressing challenges, controversies, and the
endless possibilities on the horizon.
Having witnessed Oge’s dedication and forward-thinking
approach fi rsthand, I am confident that this book will
serve as an invaluable resource. For those stepping into the
realm of AI in radiology or seeking to deepen their knowl-
edge, this book provides a holistic, scientifically rigorous,
and practical guide.
In an era of digital transformation, with large health
organizations such as mine keenly looking towards
applied innovation and AI, Oge’s contributions become
ever more relevant. His ability to anticipate and address
pressing challenges has not only benefited our field but
also shaped its future direction.
It is both an honor and pleasure to introduce this book
to you. I wholeheartedly believe that it will stand as a cor-
nerstone for all enthusiasts eager to delve into the world of
AI in Radiology.
Felipe Kitamura, MD, PhD
Director of Applied Innovation and AI at Dasa
Affiliated Professor of Radiology at
Universidade Federal de São Paulo
São Paulo, Brazil, July 2023
Preface

We are currently witnessing an unprecedented surge in the

application of artificial intelligence (AI) within the medical
field, stretching across a broad range of specializations
and poised to reshape various facets of healthcare delivery.
As we look forward, we envision AI-based systems evolv-
ing to glean optimal therapeutic pathways directly from
electronic health records and other sources of data, usher-
ing us into a new era of individualized care and inching
closer to the realization of P4 Medicine – an approach to
medicine that’s Predictive, Preventive, Personalized, and
Participatory. AI’s ability to derive insights from vast data-
sets will enable us to predict disease risk and progression,
guide prevention strategies, customize treatment plans,
and foster patient involvement in their healthcare journey.
The dawn of AI in medicine is not just about advanced
technology; it’s about ushering in a new era of healthcare
that is truly patient centric and data driven.
For now, certain medical disciplines are poised to experi-
ence the immediate influence of these novel developments,
predominantly those that rely heavily on visual data inter-
pretation for predictions, diagnosis, and treatment decisions.
These include, but are not limited to, pathology, derma-
tology, ophthalmology, and radiology – the focus of this
book.
Radiology is a particularly promising field for AI, as
it is a discipline that is already heavily reliant on image
analysis. AI can be used to automate many of the tasks that
radiologists currently perform, such as image interpreta-
tion, diagnosis, and reporting. This can free up radiolo-
gists to focus on more complex cases and provide better
patient care.
xvi Preface

The radiology community has taken quick and effec-

tive action to embrace AI, moving from potential fear
and alarmism to a position where the consensus within
the radiology community is that AI will not replace radi-
ologists; instead, it will augment radiologists’ abilities and
assist them in some of their tasks. AI applications in radiol-
ogy can cover a vast spectrum, from the decision to request
an imaging study for a patient based on the patient’s health
record to the reporting and communication of results that
might inform further actions.
AI advancements in radiology have the potential to
bring considerable improvements to patient care, radi-
ologist workflows, and physician referrals. However, this
transformative path is paved with both high expectations
and substantial hype. The technical intricacies of AI tools
only represent a portion of the equation. For a complete
adoption, these tools must be demonstrably effective, reli-
able, and safe in a clinical environment. They should also
uphold the highest ethical standards and comply with reg-
ulatory norms to earn full trust.
Implementing AI into radiology practices demands a
thorough understanding and careful consideration of the
technical, legal, and ethical implications. While AI can pro-
vide valuable assistance to radiologists, it doesn’t serve to
replace human expertise. Instead, it amplifies it. I foresee a
future where the radiologist takes on a role analogous to a
symphony conductor, steering AI solutions – the orchestra
members – each contributing their unique input under the
guided coordination of the radiologist’s expertise.

This book reviews, explains, and contextualizes some of the

most current, practical, and relevant developments in AI and
deep learning (DL) in radiology and medical image analysis.
This book is structured as follows.
Chapter 1 introduces and motivates the use of AI and
DL in the medical domain and discusses how successful
Preface xvii

applications of AI in healthcare require a different rec-

ipe than the one often used to achieve success in AI in
general.
Chapter 2 provides concrete examples of how the radi-
ology community has taken quick and effective action to
embrace the new era of AI.
Chapter 3 covers the fundamental concepts and termi-
nology associated with machine learning and DL, the pre-
dominant paradigms of contemporary AI.
Chapter 4 provides an overview of the field of radiology
from the viewpoint of the medical image analysis pipeline,
including fundamental concepts and terminology and
main imaging modalities, and offers perspective on the
impact of AI and DL in this ever-growing research field.
Chapter 5 is devoted to the essential role of data in AI
solutions for radiology. It covers various aspects of data
collection, types and sources of data, preprocessing and
annotation techniques, data challenges, considerations,
and ethical considerations related to data sharing and
collaborations.
Chapter 6 shows examples of clinical applications of AI
in radiology – both interpretative and noninterpretative –
with a focus on commercially available solutions approved
for the European and/or American market.
In Chapter 7, Dr. Michèle Retrouvey and I discuss the
challenges and initiatives associated with integrating AI
content into the medical education curriculum, at under-
graduate and graduate levels, with emphasis on radiology.
For the most technically inclined, Chapter 8 offers a
learning path to starting with DL in medical imaging. It
covers some of the most popular tools, languages, and
frameworks for developing DL solutions for medical image
analysis today. It also describes an example of a typical DL
workflow in radiology.
Chapter 9 takes us into the future of AI in radiology
with a focus on three main aspects: (i) ethical, cultural, and
xviii Preface

regulatory aspects of the use of AI in radiology; (ii) chal-

lenges, controversies, and objections to the use of AI and
DL models in radiology; and (iii) emerging trends, chal-
lenges, and opportunities.
Finally, in Chapter 10, I present a list of curated resources
(books, online courses, scientific articles, software tools,
and more) to go deeper into some aspects discussed in this
book and assist you in the lifelong learning journey that
you are about to embark on.

I hope that this book will serve as a quick, sound, and

objective scientific reference for readers from both sides of
the equation: medical professionals who want to stay cur-
rent with the latest developments of AI and related technol-
ogies, as well as computer scientists and medical imaging
informatics professionals who want to learn more about
the clinical applications and implications of the use of AI
in radiological practice and integration into its workflow.

How to Read This Book

This book has been written with a diverse readership in
mind. Therefore, your journey through the chapters may
differ based on your specific background and objectives.
For those unfamiliar with the landscape of AI in radiol-
ogy, the first two chapters will serve as a helpful primer.
Following that, based on your expertise, you may con-
sider bypassing Chapter 3 (if you already have a strong
understanding of AI) and/or Chapter 4 (if you are well
versed in radiology and medical image analysis).
Chapters 5 through 9 can be read independently and in
any order that aligns with your priorities.
Last, Chapter 10 is highly recommended for all readers,
and it would be beneficial to earmark some of the resources
compiled there for future reference.
Acknowledgments

I was trained as an engineer but have been engaged in

projects, grants, and publications related to biomedical sci-
ences since my early years in engineering school. I have
been working with AI in some form or another for 35 years.
The impetus for this book came, curiously enough, from
my work in dermatology. In 2017, I attended my first Society
for Imaging Informatics in Medicine (SIIM) Annual Meeting
to present research results in skin lesion detection, seg-
mentation, and classification. During that event, I had the
privilege of attending numerous keynotes, technical ses-
sions, and hands-on labs in radiology and was welcomed
by this amazingly friendly and inclusive community of
medical imaging informatics experts. From that moment
on, I intensified my efforts in Medical AI, became a SIIM
member – as well as a member of the European Society of
Medical Imaging Informatics (EuSoMII), European Society of
Radiology (ESR), and Radiological Society of North America
(RSNA) – and along the way had the immense privilege of
working closely with some of the brightest minds in this
space.
I am immensely grateful to many colleagues in the AI,
radiology, and medical imaging informatics communities
for their encouragement and valuable lessons throughout
these years.
Many thanks to Luciano M. Prevedello, Kevin Mader,
Barbaros Selnur Erdal, Igor R. dos Santos, Ian Pan, Felipe
C. Kitamura, Bradley J. Erickson, Katherine P. Andriole,
George L. Shih, and Peter Chang for their excellent hands-
on sessions and labs during several SIIM and RSNA
meetings.
xx Acknowledgments

I am grateful to Sameer Antani and his group at the

National Library of Medicine (NLM) at the National
Institutes of Health (NIH), where I had the privilege of
spending a sabbatical semester in 2018.
A very special thank you to my dear colleague Les Folio,
whom I first met in 2018 and have since then collaborated
in several publications and projects, including a very suc-
cessful SIIM Hackathon in 2019.
Many thanks to my European colleagues, particularly
Erik Ranschaert, Peter van Ooijen, Daniel Pinto dos Santos,
Elmar Kotter, Angel Alberich-Bayarri, Bettina Baessler,
Sergey Morozov, Benoit Rizk, Luis Martí-Bonmatí, and
Federica Zanca, for the numerous discussions and exchanges
of ideas over the past few years and for welcoming me into
the EuSoMII community.
My gratitude to colleagues in the Florida Atlantic
University Charles E. Schmidt College of Medicine, where
I have had the honor of serving as Professor of Biomedical
Science (Secondary), particularly Janet Robishaw, Julie
Pilitsis, Adam Wyatt, and Phillip Boiselle.
I am very grateful to Matt Lungren for opening the door
to successful collaborations with Stanford University in
publications, research, and educational initiatives.
I am also indebted to Pranav Rajpurkar, Tessa Cook,
Michael Do, Raym Geis, Judy Wawira Gichoya, Hugh
Harvey, and Stephen Borstelmann. I have learned a lot
from you and appreciate your willingness to explore excit-
ing paths of scientific discovery.
Special thanks to my colleague Michèle Retrouvey,
board-certified radiologist and Director of Radiology
Education at the Charles E. Schmidt College of Medicine
at Florida Atlantic University in Boca Raton, Florida, for
agreeing to co-author Chapter 7 and sharing her insights
on why and how AI content should be taught to future
doctors and radiologists.
Acknowledgments xxi

I would like to express my gratitude to Christian Garbin

and Mario Taschwer for their comments and suggestions
during the preparation of this manuscript and to Humyra
Chowdhury, Roshni Merugu, and Jennifer Gogova for
their assistance with diagrams, figures, and supporting
tasks.
A very special note of gratitude to Randi (Cohen) Slack
and her team at CRC Press / Taylor & Francis for their sup-
port throughout this project.
Oge Marques
Boca Raton, Florida
July 2023
1
Artificial Intelligence and
Medicine: The Big Picture

1.1 Introduction
Artifcial intelligence (AI) has been one of the most sig-
nifcant technological developments of our era. In the last
ten years, thanks to the emergence of deep learning (DL),1
many high-profle problems that seemed beyond the reach
of computer-based solutions have been solved, including
better-than-human performance in games such as chess,
Go, and Shogi [1]. In parallel with headline-grabbing
advancements, AI has become a fundamental tool in the
business world, where managers are busy fguring out
how to “add AI” to their products and processes to get an
advantage over their competitors. AI has impacted every
sector of the economy and every feld of human activity,
from education to transportation, games to communica-
tions, shopping to healthcare.
The growing adoption of AI products and solutions in
healthcare will affect patients, insurance providers, drug
companies, researchers, and medical professionals in every
area of medicine. It is expected that in the near future,
every medical student will have a signifcant amount
of AI-related content in their curriculum [2], preparing
1 Deep learning (DL) is a term used to refer to AI solutions that use
neural networks and are capable of learning from (vast amounts of)
data. Since the early 2010s, most AI advancements have been possible
thanks to the use of DL approaches. See Chapter 3 for more details.

DOI: 10.1201/9781003110767-1
2 AI for Radiology

themselves for a world in which doctors who understand

AI will be more valuable than doctors who don’t. This is
particularly true for radiology, a specialty that, for a brief
while, might have felt threatened by the advancements in
DL AI and eventually became the driver of progress in the
feld (see Chapter 2).
Ideally, the future of healthcare will be built by combining
the talent of data scientists, machine-learning engineers,
statisticians, software developers, and AI researchers on
one end with the subject matter expertise of every specialty
in medicine on the other end. We should be prepared to
build a pathway that will ultimately improve patients’
(and doctors’) well-being [3].
In this chapter, we introduce and motivate the use of AI
in the medical domain and discuss how successful appli-
cations of AI in healthcare require a different equation
than the one often used to achieve success in AI in general.
We start by providing an overview of AI to better under-
stand its foundations, revisit some of its historical ups and
downs, standardize the terminology, and appreciate what
contemporary AI solutions can and cannot do.

1.2 AI: Foundations, Brief History, and

Current State of the Art
1.2.1 Defining AI
Defning AI is surprisingly diffcult, in signifcant part
because we don’t have a good defnition for intelligence to
begin with, and the use of the word artifcial in this context
is not always precisely clear either. In general, artifcial usu-
ally means “by means of a computer or robot,” which sug-
gests that some hardware and software components enable
an electronic device to act as if it had cognitive capabilities
comparable to the ones exhibited by humans.
Artificial Intelligence and Medicine: The Big Picture 3

A dictionary (Merriam-Webster) defnes AI as “a branch of

computer science dealing with the simulation of intelligent
behavior in computers” or “the capability of a machine
to imitate intelligent human behavior.” An encyclopedia
(Encyclopedia Britannica) may phrase it as “the ability of a
digital computer or computer-controlled robot to perform
tasks commonly associated with intelligent beings.” These
are fne defnitions that grab the general sense of the term AI.
In more technical texts [4], there are four groups of def-
nitions, refecting some of the most prominent approaches
to AI in its history. We follow these book authors in choos-
ing the “rational agents” approach, which sets the goals
of AI to be “the study and construction of agents that do
the right thing” (emphasis added) [4].

1.2.2 Agents and Environments

According to Russell and Norvig,

[A]n agent is anything that can be viewed as

perceiving its environment through sensors and
acting upon that environment through actuators.2
For each possible percept sequence, a rational agent
should: select an action that is expected to maximize
its performance measure, based on the evidence
provided by the percept sequence and whatever
built-in knowledge the agent has. [4]

Figure 1.1 shows how rational agents interact with their

environments. The box with the question mark is where
“the magic happens”: it represents the intelligent algo-
rithms capable of processing sensor data (raw input) and
producing decisions that will eventually result in actions

2 A sensor is a device that detects and measures physical or environmental

properties, converting them into electrical signals or other forms of
output. An actuator is a component that receives a control signal and
initiates a physical action or movement, typically to manipulate or con-
trol a system or its environment.
4 AI for Radiology

Figure 1.1
Agents and environment (adapted and redrawn from [4]).

in the environment. The diagram is general enough to be

used for signifcantly different scenarios, from the simple
game of tic-tac-toe (where the environment is fully observ-
able, static, and discrete; the input is the current state of the
board; the action is playing an “X” or “O” in an empty slot;
and the algorithm can be as simple as a game tree), to the
complex world of self-driving vehicles (where the environ-
ment is partially observable, dynamic, and continuous; the
input consists of a representation of a rich and vibrant three-
dimensional (3D) world of cars, pedestrians, traffc signs,
lane marks, and much more, through multiple sensors that
produce vast amounts of raw data in different modalities –
such as radar, lidar, camera pixels; the action might include
braking, turning, etc.; and the algorithm must be incred-
ibly complex, requiring multiple sub-blocks to handle
Artificial Intelligence and Medicine: The Big Picture 5

specialized tasks – from 3D reconstruction of the world

to path planning).
In the context of healthcare, the diagram in Figure 1.1
can also be helpful to understand the recently proposed
model for a “virtual medical coach” that takes as a per-
cept a vast amount of data associated with a patient – from
genomics to medical history, from physical activity and
sleep indicators captured by smartwatches or equivalent
devices to advances reported in the medical literature that
might be relevant to the case at hand – feeds this into a
DL AI solution, and produces personalized “virtual health
guidance” at the output [5].

1.2.3 AI and Deep Learning

Before we delve deeper into DL, it seems appropriate to
set the stage by briefy introducing DL within the broader
framework of modern AI. The ongoing AI revolution is
being propelled by vast amounts of data and increasing
computational power, with DL playing a central role in
this dynamic environment.
In the simplest terms, DL is a technique that trains com-
puters to mimic the human brain’s decision-making pro-
cess. DL is a subset of machine learning, a branch of AI
that involves training algorithms, or set instructions, to
learn from data and make predictions or decisions without
being explicitly programmed. DL takes this concept even
further by creating deep neural networks – multilayered
“artifcial brains” – that learn to identify patterns and
interpret data by training on vast information.
Imagine a deep neural network as an incredibly complex
web of interconnected points, similar to the neurons in
our brain. Data inputs travel through this network and
are processed at multiple layers, each learning to recog-
nize increasingly complex features. Eventually, the net-
work generates an output, which could be anything from a
6 AI for Radiology

recommendation for a product to a diagnosis of a medical

condition.
In the modern context, AI has evolved into gathering
and harnessing an extensive dataset to train and fne-tune
these deep neural networks. The goal is to enable these
networks to make predictions, solve problems, and learn
independently.
Chapter 3 expands upon these concepts and shows how
they translate into creating DL solutions.

1.2.4 Promises, Predictions, and Winters

The history of AI can be traced back to a seminal event
in the summer of 1956 at Dartmouth College. During this
summer workshop, luminaries such as John McCarthy,
Marvin Minsky, Nathaniel Rochester, Claude Shannon,
and colleagues laid the groundwork for AI as an academic
feld and explored the potential of computational systems
to exhibit intelligent behavior.
Over the next 50 years, the feld of AI underwent two
complete cycles characterized by periods of enthusiasm,
optimism, and ambitious assertions, followed by disil-
lusionment and unmet expectations. These downturns
in the cycles are commonly referred to as “AI winters.”
The general interest in AI, including publications, fund-
ing levels, and more, experienced signifcant declines
during these intervals. To illustrate the extent of the
decline, the mere mention of terms like AI or neural net-
works could be viewed unfavorably in the context of a
scientifc paper.
The frst AI winter occurred from the mid-1970s to the
early 1980s [6]. This period was marked by a disconnect
between the computational demands of certain theoretical
AI solutions and the limitations of the available hardware,
software, and network infrastructure required for their
implementation. The second AI winter, which took place
from the late 1980s to the early 1990s [6], was characterized
Artificial Intelligence and Medicine: The Big Picture 7

by the increasing cost, sluggish performance, and diff-

culty in updating expert system computers compared to
the more advanced desktop computers of that time. As
a result, funding and resources were redirected toward
more promising projects.

1.2.5 AI Today
Since the early 2010s, AI has experienced a signifcant
resurgence, mainly due to several critical factors described
next.
One of the primary drivers has been the advances
in computer hardware. The development of powerful
architectures specifcally designed for processing high-
dimensional data has been instrumental. Graphics pro-
cessing units (GPUs), tensor processing units (TPUs),
and data processing units (DPUs) have signifcantly
accelerated AI computations, enabling more effcient
training and inference processes.
Another crucial factor has been the availability of mas-
sive volumes of data. The internet has provided a wealth of
information through public pages and social media, and
the increasing adoption of electronic records across vari-
ous businesses, including healthcare, has further contrib-
uted to this data pool. This wealth of data has provided the
necessary input for AI systems to learn and improve.
The feld has also benefted from continuous advance-
ments in algorithms. The development of novel neural
network architectures and the optimization of existing
algorithms used for training these networks have been
vital in enhancing the performance and capabilities of AI
systems.
The publicity generated by solving highly visible and
challenging problems using machine learning approaches
has also played a role in AI’s resurgence. Notable examples
include AlphaGo and its successor AlphaZero, which dem-
onstrated exceptional performance in the games of Go,
8 AI for Radiology

chess, and Shogi. These achievements have highlighted

the potential of machine learning techniques in solving
complex problems.
Finally, the proliferation of better, smaller, cheaper, and
widely distributed sensors, often referred to as the Internet
of Things (IoT), has been a signifcant factor. These sen-
sors, ranging from smartwatches to cameras, generate
vast amounts of data and enable the collection of real-time
information. This has facilitated various AI applications in
healthcare, transportation, and environmental monitoring.

1.2.5.1 Comparing AI Solutions

The hype surrounding AI, particularly in mass media,
often obscures the actual capabilities and limitations of AI
systems. Recognizing that not all AI systems are created equal
is crucial. When assessing the claims made about an AI
solution, we suggest considering the following fve dimen-
sions along with their associated questions [7]:

1. Strength: How intelligent is it?

2. Breadth: Does it solve a narrowly defned problem,
or is it general?
3. Training: How does it learn?
4. Capabilities: What kinds of problems are we asking it
to solve?
5. Autonomy: Are these assistive technologies, or do
they act on their own?

Using AlphaGo and AlphaZero as examples, this is how

the previous questions could be answered:

1. For games such as chess, Go, and Shogi, it is possible

to benchmark AI solutions against one another
(as well as against the human-level performance
of chess grandmasters, for example). This is done
Artificial Intelligence and Medicine: The Big Picture 9

using the Elo metric.3 According to that metric,

AlphaZero is superior to every other human or AI
player in all of these games [8].
2. Both AlphaGo and AlphaZero solve narrowly
defned problems: the former learned how to play
the game of Go, whereas the latter is slightly less
narrow since it also learned how to play chess and
Shogi. While AlphaZero demonstrates exceptional
performance across different games and exhib-
its impressive learning capabilities, it cannot be
considered as achieving artifcial general intelli-
gence (AGI) due to its domain-specifc nature. AGI
refers to a system with human-like intelligence
that can excel in a wide range of tasks, including
those outside the scope of specifc domains, which
AlphaZero does not encompass.
3. Both AlphaGo and AlphaZero use an approach
known as “deep reinforcement learning.” The
main difference between them, however, is that
while AlphaGo learned from a dataset of games
of Go played by human experts, AlphaZero
essentially learned from scratch (hence the Zero
in its name) by playing millions of games (of Go,
chess, or Shogi) against itself and fguring out
which strategies worked out best. This has led to
AlphaZero being capable of choosing movements
that were somewhat unusual, surprisingly effec-
tive, and downright puzzling to human experts in
those games.
4. AlphaZero is designed to solve complex problems
related to strategy, decision-making, and optimal
play in board games. Specifcally, it aims to tackle

3 Elo metric is a rating system used to evaluate the relative skill levels
of players in competitive games, initially designed for chess but now
applied to various other sports and games as well.
10 AI for Radiology

challenges in games such as Go, chess, and Shogi,

where the primary objective is to make intelligent
moves, outmaneuver opponents, and ultimately
win the game. AlphaZero focuses on mastering
the intricacies and strategies involved in these
specifc games rather than addressing broader
problem domains outside the realm of board
game playing.
5. AlphaGo and AlphaZero are truly autonomous
technologies; the only “assistance” they might
need is to place physical pieces on the respective
physical board.

To conclude this brief discussion about the fve dimensions

of AI solutions, it seems useful to mention that

• Not all AI solutions can be benchmarked against

each other – or compared to humans performing
the same task. This is why open challenges and
competitions such as ImageNet LSVRC, Kaggle
challenges, and radiology-specifc challenges (see
Chapter 2) have become essential factors in driving
the evolution of AI in recent years. In other words,
for most AI applications, it is impossible to rank
solutions based on their “strength”; the universe
of chess, go, and Shogi is a notable exception.
• Most AI solutions available today, however impres-
sive, are examples of “narrow AI” (although
some are narrower than others, as the AlphaGo/
AlphaZero example demonstrated). This is
quickly changing, though, since late 2022, with
the popularization of advanced large language
models (LLMs), such as OpenAI’s GPT-3.5 and
GPT-4 (the models behind the – now ubiquitous –
ChatGPT). The search for the elusive goal of arti-
fcial general intelligence (AGI) (also referred to
as “strong AI”) continues, and the debate on how
Artificial Intelligence and Medicine: The Big Picture 11

soon we might be reaching that stage is heating

again.4
• Discussions about how an AI solution learns before
a model is built and deployed (e.g., how much data
are needed, where the data come from, how can
we ensure that the data are representative of the
population, etc.) as well as its ability to continue to
learn and improve its performance after it has been
deployed, are fundamentally important. We revisit
some of these issues in this context of radiology later
in this book.
• When developing AI solutions, it is essential to
start by asking meaningful questions that will
promote a clear understanding of which prob-
lems we expect to solve, leading to better user
and system specifcations, a smoother software
development process, better documentation, and
improved workfow. Failure to do so, on the other
hand, may lead to inaccurate evaluation, mislead-
ing claims of success, and overall distrust.
• For most applications, especially in radiology/
medicine/healthcare, it is common (and convenient)
to think of AI as “augmented intelligence,”
removing the cumbersome word “artifcial” from
the equation and suggesting a collaborative
ecosystem where AI and humans work together
toward common goals – much like medicine itself,
where multiple professionals apply their collec-
tive expertise for the beneft of the patient.

1.2.5.2 What AI Can (and Cannot) Do Well

The enormous hype around AI in the media – both when
its successes are artifcially magnifed as well as when the

4 See this “AI Apocalypse Scorecard” recently compiled by IEEE

Spectrum magazine [9] for a quick overview.
12 AI for Radiology

prospects of job losses and other societal impacts due to AI are

exaggerated – makes it diffcult to draw a precise line between
what contemporary AI solutions can do well and what they
cannot. The lists that follow (adapted from [10]) should help.
Today’s AI can:

• Perform some (often impressive) well-defned tasks

as well as or better than humans. Examples include
imageclassifcation,speechrecognition,handwriting
transcription, and autonomous driving.
• Find and act upon patterns in data, including
patterns invisible to humans.
• Get better at performing certain tasks when given
lots of labeled, well-organized data from which to
learn.

Today’s AI however typically cannot:

• Perform any entire job better than humans can.

• Explain its mechanism for fnding patterns in
information or what those patterns mean.
• Understand the context that surrounds a given
task.
• Perform tasks that require creativity, empathy, or
complex judgment.

In summary, looking at what today’s AI can and cannot do,

we start to realize that, as AI technology advances and some
of its limitations become apparent, it would be advisable to
rethink the meaning of the “A” in “AI”: rather than artif-
cial intelligence (and the associated negative implications
in many people’s minds), today’s AI is being used to assist,
augment, and automate processes in a wide variety of areas.
Perhaps what we have today – and will continue to see more
Artificial Intelligence and Medicine: The Big Picture 13

of in the near future – is akin to “automated intelligence,”

“augmented intelligence,” or “assistive intelligence,” all of
which are excellent examples of putting technology to good
use and (hopefully) improving human lives and optimizing
the use of exhaustible resources along the way.

1.3 AI in Healthcare
1.3.1 AI in Healthcare: The Growing Potential
There has been explosive growth in the feld of AI in
healthcare in recent years, whether it is measured by the
number of scientifc publications, mentions in the mass
media, fled patents, or venture capital (VC) deals (and dol-
lar amounts) for funding start-up companies in this space.
The editorial of the 25th anniversary edition of the pres-
tigious Nature Medicine journal whose central theme was
“Medicine in the Digital Age” states that “Digital medi-
cine . . . holds promise in revolutionizing healthcare and
well-being. . . . AI, and in particular deep learning (DL),
is among the leading technological tools beginning to be
used in the interpretation of medical images and electronic
health records” [11].
In short, AI in healthcare is here to stay. There are numerous
references and vast amounts of data to support this claim.
They range from scientifc discoveries (publications, grants,
citations, patents, etc.) to entrepreneurial activity (start-up
companies and associated VC amounts and the number of
deals, etc.) to deployment, clinical validation, and US Food
and Drug Administration (FDA) clearance of AI solutions.
It should be noted that AI in healthcare has multiple facets
and dimensions, many of which are beyond the scope of
this book (e.g., surgical robotics, clinical trial participation,
14 AI for Radiology

or fraud detection, to name but a few). The adoption of AI and

DL solutions will vary among different medical specialties
and will impact different aspects of the healthcare system at
different stages, and the prevalence of AI methods will span
human life, from embryo selection for in vitro fertilization
[12] to prediction of 24-hour mortality in palliative care [13].
Eventually, smart healthcare systems will emerge, in
which “the best treatment decisions are computationally
learned from electronic health record data by deep-
learning methodologies” [14] Until then, certain areas of
medicine are more likely than others to be impacted by
the latest developments in DL AI, particularly those areas
in which predictions, prognoses, and decisions are made
based primarily on visual data, notably pathology, derma-
tology, ophthalmology, and radiology.

1.3.2 AI in Healthcare: A New Formula

The recipe for success in AI (in general) includes ingredients
such as the availability of large amounts of good quality
data, access to powerful computational resources, devel-
opment and validation of robust algorithms, and massive
investments to make it all come together. We can think of
it as an equation5:

Success in AI = Big Data + Powerful Compute + Robust

Algorithms + Massive Investments

These four ingredients are necessary, but not suffcient,

for achieving successful results in AI in healthcare. AI in
healthcare requires a new equation, where the AI efforts and
investments are aligned with the goals, vision, knowledge,

5 This equation, and the ones that follow it, were inspired by a slide
by Dr. Keith Dreyer presented during his keynote at the Society for
Imaging Informatics in Medicine 2017 Annual Meeting, Pittsburgh,
PA.
Artificial Intelligence and Medicine: The Big Picture 15

Figure 1.2
AI in healthcare requires a new equation (adapted and redrawn from a
slide used by Dr. Keith Dryer in his keynote speech at SIIM 2017 Annual
Meeting, Pittsburgh, PA).

and environments for which those AI solutions might be

useful (“Domain Expertise” in Figure 1.2) and eventually
result in products that successfully meet the requirements
for clinical translation, workflow integration, and eventual
commercialization (“AI Productization” in Figure 1.2).

1. 3.3 AI in Healthcare: Encouraging Early Results

for Medical Professionals
In the discourse surrounding AI’s role in healthcare, the
focus primarily lies on new technological breakthroughs.
16 AI for Radiology

This includes patents for health-monitoring methods, sen-

sors, or devices, as well as the impressive diagnostic capa-
bilities of DL AI for specifc diseases, such as diabetic
retinopathy [15,16]. These advances draw signifcant atten-
tion in both academic circles and the general media.
However, there’s another aspect that receives less cover-
age: the impact of AI implementation on the daily routines
of healthcare professionals. A notable study shedding light
on this issue was published in late 2019 [17]. The research
included 908 professionals6 working at healthcare institu-
tions in the US (70%) and the UK (30%), including medical
professionals and business and administrative profession-
als who are involved in the purchasing or who infuence
the selection of AI, big data analytics, or medical equip-
ment and technology.
The survey results were positive. Over 82% of profession-
als reported that AI deployment had led to operational and
administrative improvements in their workfows. This, in
turn, allowed medical staff more time to focus on patient care.

In summary, the outlook is promising for AI in health-

care. AI is expected to serve as a valuable aid in medical
procedures while always necessitating human supervi-
sion and oversight. There is no need to harbor fear toward
AI; instead, we should focus on being prepared. The
Radiology community is an outstanding example of how
this preparedness can be accomplished, as we explore in
Chapter 2.

6 Of the total, 17% are medical doctors and specialists, 5% are nurses or
nurse practitioners, 26% are senior management, 16% are in information
technology, 16% are in research and development, 9% are in legal or
regulatory departments, and 9% are in fnance or accounting.
Artificial Intelligence and Medicine: The Big Picture 17

Key Takeaways
• Despite the hype, AI is poised to impact numer-
ous aspects of healthcare and several medical
specializations, notably radiology.
• AI in healthcare will likely play an assistive role
in medical routine and will always require human
supervision and oversight.
• AI has the potential to impact every aspect of the
healthcare system.
• This is a highly active research area, but the
application of research results to the real world is
just beginning.
• Success in AI in healthcare requires a differ-
ent equation than the one often used to achieve
success in AI in general.
2
AI in Radiology: From
Fear to Leadership

2.1 Introduction
In 2016, at the Machine Learning and the Market for Intelligence
Conference in Toronto, the prominent researcher Geoffrey
Hinton1 declared that “We should stop training radiologists
now. It’s quite obvious that in fve years deep learning
will do better than radiologists” [18]. He was speaking
about the potential of deep learning (DL) to revolutionize
radiology, and he argued that radiologists were at risk of
being replaced by artifcial intelligence (AI) systems within
fve years. Hinton’s statement was based on the success of
DL algorithms in other medical imaging tasks, such as clas-
sifying skin cancer and detecting diabetic retinopathy. He
argued that these algorithms were already better than radi-
ologists at identifying specifc types of diseases and that
they would only get better with time.
This bombastic statement came at a time when other
news outlets were also speculating about the rise of “robot
radiologists” [19], and even scholarly journals in the feld
contemplated the threat of AI, machine learning (ML), and
DL to the radiology profession (see, e.g., [20]).

1 Geoffrey Hinton received the Association for Computing Machinery

A.M. Turing Award – the computer science equivalent of a Nobel Prize –
in 2018, together with Yoshua Bengio and Yann LeCun, for their contribu-
tions to the advancement of DL AI.

DOI: 10.1201/9781003110767-2
AI in Radiology: From Fear to Leadership 19

More than fve years have passed, and Geoffrey Hinton’s

predictions have not come true: radiologists are still as busy
and essential to the healthcare ecosystem as ever. The adoption
of AI in radiology has been strengthened, refned, discussed,
and expanded, largely thanks to the radiology community.
Radiologists quickly and decisively moved from a place of
potential fear and alarmism to a position of leadership. Today,
there is a consensus within the radiology community that AI
will not replace radiologists; instead, it will augment radiolo-
gists’ abilities and assist them in some of their tasks.
Indeed, the workload for radiologists has dramatically
increased, with a staggering 300% surge in imaging volume
in recent years. However, the number of trained radiology
residents is not proportionately growing to meet this
heightened demand. The resulting job market dynamics
are exceptionally intense. This situation underscores the
urgency and necessity of AI implementation in radiology.
Through AI, we can enhance the effciency of existing
radiologists, enabling them to manage the escalating
workload effectively and accurately.
Beyond merely utilizing the latest AI tools, the radiology
community has further expanded its engagement with AI and
DL. This includes involvement in areas such as

• Developing AI-powered tools: Radiologists have devel-

oped AI tools that can help them diagnose diseases
more accurately and effciently.
• Training radiologists on AI: Radiologists are now
being trained on AI concepts, techniques, algo-
rithms, and frameworks. This training helps radi-
ologists to understand how AI works and how to
use it to improve their diagnostic skills.
• Collaborating with developers: Radiologists collabo-
rate with software developers to develop new
AI-powered solutions. This collaboration helps to
ensure that these solutions are developed in a way
that meets the needs of radiologists.
20 AI for Radiology

This chapter showcases initiatives by the radiology and

imaging informatics communities. They are actively
addressing the need for medical professionals to “know
enough AI to strip away the mystery” of AI and DL solu-
tions for radiology [21].

2.2 AI in Radiology: Opportunities

and Applications
AI is rapidly transforming the feld of radiology. AI-powered
tools are being developed to help radiologists with various
tasks, from image interpretation to diagnosis to patient
management.
AI applications in radiology span a vast spectrum. They
start when an imaging study is requested, where AI can
help radiologists decide which imaging studies to order
for their patients. For example, AI can analyze a patient’s
electronic health record and identify risk factors for certain
diseases. This information can then be used to recommend
specifc imaging studies most likely to help diagnose the
patient’s condition.
Once the images have been acquired, AI comes into play
again to support image analysis.2
DL algorithms can identify and highlight potential abnor-
malities, aiding radiologists in their interpretations. AI algo-
rithms can also quantify specifc image features that can be
challenging for the human eye to detect, adding an extra
layer of precision to the diagnostic process.
AI tools can also facilitate triaging cases, automatically
prioritizing those that contain urgent fndings and require
immediate attention. This ensures that time-sensitive
diagnoses are not missed in a busy workfow.

2 See Chapter 4.
AI in Radiology: From Fear to Leadership 21

Finally, once the analysis is complete, AI can assist in gener-

ating preliminary reports, streamlining the communication
process to referring physicians. These reports can help syn-
thesize complex imaging fndings into concise, clinically
relevant information, helping to guide patient management
decisions. In this way, AI applications are woven into each
step of the radiology workfow, augmenting the abilities of
radiologists and improving patient care.

2.3 The AI in Radiology Community Today

In this section, we review some examples of active
involvement of the radiology community3 in advancing AI
applications and solutions. We discuss how (and why) some
medical doctors took up programming, engaged a commu-
nity of medical professionals and software developers in
search of practical, concrete solutions to meaningful prob-
lems, and along the way created numerous mechanisms for
the dissemination of scientifc knowledge and preparation
of the next generations of AI in radiology professionals.

2.3.1 Doctors Who Write Code

Many radiologists took the challenge of writing code and
building their solutions to challenging AI problems. They
did so for several reasons, among them

• Satisfying their intellectual curiosity: Many radiology

professionals have an educational background
that extends far beyond their MD training and
subsequent specializations and certifcations in
radiology. It is not uncommon to fnd prominent
3 When we refer to radiology community, we implicitly include many other
groups of professionals in the medical imaging informatics community.
22 AI for Radiology

radiologists with advanced degrees in engineer-

ing, computer science, physics, statistics, and other
scientifc felds.
• Facing the threat allegedly posed by AI objectively and
assessing if it deserves all the hype: This is brilliantly
summarized in a cartoon by Felipe Kitamura,
MD, PhD (Figure 2.1), where he indicates that
his fear (as a radiologist) of being replaced by AI
dropped to zero the day he trained his frst AI
model. Understanding frst hand how brittle and
limited DL AI solutions can be provides extremely
valuable insight into the hype and ideas on how
to leverage this new skill set to build solutions to
specifc problems.
• Digging into the inner working of DL AI solutions: To
truly grasp the capabilities and limitations of AI, a
deep practical understanding can only be acquired
through hands-on experience — writing, running,
and modifying code. While a book like this one
provides valuable conceptual insights, it is through

Figure 2.1
“Fear of AI” (by Dr. Felipe Kitamura, used with permission).
Another random document with
no related content on Scribd:
 "They come to one's door," continued the enraged
father, "many hundreds of them. The man with a broken
fiddle, the woman with a screeching voice, both clothed in
rags. That's the end of them all! Sir, if you had an only son,
would you make a musician of him?"

"Surely not one like you describe," answered Mr.

Delrick, "but if I had a son with the talent which makes a
great composer, I know nothing would hinder me from
helping him fulfil his desire."

"My lad does not have such a gift, that is sure,"

declared the father obstinately. "Such genius is not often
found. Believe me, sir, when Vinzi comes to his senses, he
will be glad enough that he lives on a fine farm and is its
owner instead of being a wandering musician."

Mr. Delrick confessed he did not know how much talent

for music Vinzi possessed; he only knew of the boy's
intense delight in it. He also felt that Vinzenz Lesa's opinion
about a musician's life was not to be changed with words.
One question had arisen, however. How could he decide
whether he would be justified in trying to overcome the
father's aversion or whether he ought to give him his
support and aid in bringing his son into the right path?

"Mr. Lesa," he said as he rose and held out his hand,

"we will talk no more about it now, for it seems we can't
come to any understanding today. But we will discuss it
again, when I hope we will agree, for we have always
gotten along well with one another until now."

"We have that," replied Vinzenz Lesa, shaking the

outstretched hand. "And when we do not hold the same
opinion, still I know that you mean well."
 On the following day, the last before Mr. Delrick
departed, it was so silent and sad in the house that one
might have supposed a great misfortune was impending.

Stefeli had thought that when Vinzi was once more at

home there would be nothing but happiness in the house.
Now it was just the opposite and the only one who could
help them was leaving.

Mr. Delrick had told Mrs. Lesa of his conversation with

her husband, but declared in spite of it, he still hoped to
find a way out for Vinzi. However, she could entertain no
such view and saw only disaster ahead. Even if her husband
yielded to Mr. Delrick's persuasion, he would never be
reconciled and there would always be dissension between
father and son. Only one person could avoid this and he
was going away.

Vinzi thought that if he only could talk everything over

again with Mr. Delrick, joy and confidence would be
restored, but this was impossible since he was departing.

Mr. Lesa was under the impression that his wife and son
did not understand what was necessary for the boy's
welfare. The only one who would be likely to have such
insight and bring the others to his viewpoint was now
leaving them.

Mr. Delrick still had a kind word for one and all, but he
could not lift them out of their depression.

As he withdrew to his own room on the last evening,

there came a knock at his door and Vinzi stepped in. He
was carrying two books and a little package, and asked
timidly if Mr. Delrick would take them to his cousins. The
little packet was for Russli; he had promised faithfully he
would send it to him. The books were for Jos and Faz, for
the boys had told him how glad they were to read on the
long winter evenings, and as they had only a few books,
they had to read them over and over again. Vinzi wanted
him to carry his best wishes to all the cousins, and to the
grandfather and Father Silvanus. Would he thank them for
all they had done and say how much he would like to be
with them, how glad he—but Vinzi could get no further. He
said a hurried good-night and went off.

Mr. Delrick intended to return to Germany from the

Italian lakes by another route. He hoped to return the next
summer, which news was received by the Lesa family with
much rejoicing, though Stefeli thought it was a very long
time to wait. Early the next morning Mr. Delrick was on his
way up to the Simplon.
 CHAPTER IX
SURPRISES, NOT FOR RUSSLI ALONE

SEPTEMBER and October were months of heavy work

for Vinzenz Lesa, and he was busy from dawn to dark.
Usually he was in a happy mood at this season because of
the blessings of a bounteous harvest. However, this autumn
he went around in silence and often stood lost in thought,
gazing into space. It was evident he was pondering some
difficulty. Indeed, the problem of Vinzi's future filled his
mind night and day and left him no peace, for he loved his
son and was as proud of him as only a father can be. But
the boy must be brought to his senses and realize his good
fortune.

After many days spent in weighing the matter, he came

to a definite decision, and went into the room where his
wife sat mending his shirt. She was just as much disturbed
as he, and for the same reason.

"I'll take the boy away on Sunday," said he directly he

came in. "I'll take him to my brother in Freiburg. There will
be plenty of work until winter, and Vinzi will be glad of it for
he will find no amusements there."

Mrs. Lesa's sewing slipped out of her fingers into her lap
as she looked up at her husband pale with dread.

"Have you considered the state your brother is in,

Vinzenz? Do you not remember the name they have given
him?" she asked. She was filled with anxiety for she could
picture their Vinzi staring ahead as was his habit and her
brother-in-law, sad figure that he was, alongside him.

"That's nothing," replied her husband. "He is not

vicious; he just doesn't take to work and can't manage
things; but he has sense enough to know there should be a
master on the place as well as a servant. That is why he
insists I ought to go back or send my boy to him. Vinzi is
not stupid, and when he finds he can issue orders, he will
like it. After that comes knowledge. This is just the thing for
him, believe me! I have thought it all over, and we are off
on Sunday."

Plenty of objections occurred to Mrs. Lesa, but since it

seemed everything she had said only strengthened her
husband's determination, she remained silent and he went
off.

When she was alone with her thoughts, she

remembered what grief she had suffered when Vinzi went
away before, and how much better things had gone with
him than she had feared. She had not trusted in the dear
God. He had led her lad to kindly people, and surely to
begin grieving and doubting now, as though she knew
better than He what was best for Vinzi, displayed
ingratitude. She would put everything in His hands, in full
confidence that the Father in Heaven meant well with all His
children and would guide hers as was best. She grew calm,
and longed to talk with Vinzi, who did not know anything
about this new plan.

That evening she heard Vinzi and his father as they

returned from the woods where they had been all day and
called the boy to her, for she knew that her husband would
be busy in the stable and barn and would not miss him.
 Stefeli came running in as soon as she heard Vinzi, but
her mother sent her out to the barn on an errand, expecting
she would remain there quite a while with her animal
friends.

But Stefeli had noticed that her mother had something

special on her mind, and thought it was absolutely
necessary she should be present. She flew on her errand
and was back again. But her mother did not share her view,
and said, "Go and search every corner in the hen-house.
You know the hens lay in unexpected places. Hunt
everywhere before you bring in the eggs."

Stefeli ran as fast as she could, but she was hardly out
of the house before she came rushing back. Flinging open
the door, she shouted, "He is coming back! He is coming
back!"

With that she was off again. The mother and Vinzi
gazed at one another. Both had the same thought, but it
seemed so impossible that neither expressed it.

The door opened again and what they thought

impossible was really a fact. Stefeli stepped in triumphantly,
holding Mr. Delrick by the hand. The surprise was so great
that neither Mrs. Lesa nor Vinzi could utter a word though
both their faces beamed with joy.
 "I have changed my plans," explained Mr. Delrick after
first greetings. "My friends are returning to Germany
another way, so I have come back. Really I could not do
otherwise for I carry such heaps of greeting for Vinzi from
the people on the mountain that I had to unload them. They
treated me as an old friend because I brought Vinzi's
messages."

Vinzi's eyes sparkled as he asked in eager anticipation,

"Did you see them all? Grandfather and Father Silvanus
too?"

"Everybody! And they all seem to love you, Vinzi,"

replied Mr. Delrick. "Your good Cousin Lorenz and his wife
could not talk enough about the happy times you had
together."
 Then he launched into an account of the happiness the
gifts had given the three boys. Out of sheer ecstasy Russli
would not let his present leave his hands. Wherever he
went, he carried the red silk bag filled with the glittering
agate marbles. Mr. Delrick said he brought a special
message from Russli, and though he did not quite
understand it, Vinzi would likely do so. Russli wished to say
he would never tickle them again and had not expected his
surprise to be so lovely.

Mr. Lesa now stepped into the room, and in his surprise
at seeing Mr. Delrick, he stood stock still. Then shaking the
proffered hand with all his might, his eyes shining with
pleasure, he said, "So it is actually true? I could not help
thinking it would be this way, life has been so empty since
you left. Welcome back again!" he said, emphasizing his
words with another handshake. "And now let's sit down to
supper. My wife is sure to have something special for this
guest. She is certainly just as glad as I am that you are
back."

Mrs. Lesa had already disappeared to prepare the

supper, and the meal was eaten with high satisfaction for
the pleasure of the reunion had banished all sad thoughts.

When they rose from the table, Mr. Delrick said, "May I
have a little talk with you out on the bench, Mr. Lesa, after
our old manner? Don't forget your pipe."

Mrs. Lesa understood that Mr. Delrick wished to have a

private conversation with her husband, so she kept the
children indoors with her.

Mr. Delrick lost no time in beginning, and as soon as the

two men had sat down, he said, "You may well suppose, Mr.
Lesa, that I had a reason for altering my plans and
returning here."

"You surely would not do so otherwise," was the

prudent reply.

"I have some news I think is so important that I wished

to lose no time in giving it," continued Mr. Delrick. "I spent
a day up on the mountain in order to deliver Vinzi's
messages in person and to look up his friends there. First I
called on your cousin, the worthy Lorenz Lesa, and his wife.
These good people could not tell me enough about Vinzi,
how beautiful he had made the summer with his music and
songs, and how hard it is for both old and young to get
along without him. That ought to please a father, hadn't it?"

The latter nodded assent.

"Then I visited the grandfather in the Tower. His joy was

really touching when I gave him Vinzi's greetings. He said
Vinzi had given him the most delightful hours he had had
for many years with his playing and singing, and that it was
his dearest wish to have the youngster sing as he departed
on his last journey. His music must have been remarkable,
not just frivolous wandering minstrel music, to have made
such an impression on the aged man. What do you think?"

Vinzenz Lesa again nodded silently.

"When I asked him whether Father Silvanus knew much

about music, my question roused the old man and he told
me that Father Silvanus is really a great musician. He
studied many years in a college at Rome. However, he
sought solitude, and came up on the mountain to find it,
and has lived there a number of years, doing much good.
That convinced me I had discovered just the man I was
searching for, one who could judge intelligently Vinzi's
talent. So I went to Father Silvanus. As soon as he heard I
came from Vinzi, he took a great interest in me. He asked if
the boy was taking lessons and if his talent was being
developed. I confessed it was on that account I had come to
him, to hear what he thought about his talent. The good
Father showed real enthusiasm, and exclaimed:

"'The boy is full of music! I never tried to teach him

anything, only to entice it out of him.'

"He said he wanted to call my attention to two things;

then I could judge for myself. The first melody that Vinzi
composed, or at least the first that he had composed and
played, had been so original and so beautiful that he
himself often played it for his own pleasure. Next, Vinzi had
set a tune to some words, and it had won all hearts. It was
being sung by all the herd boys on the pastures, by the girls
at their spinning-wheels, by the boys in stable and barn.

"'It is whistled and hummed by the people on the roads,

who call it Our Song. No one knows whence it came, but it
is the beloved property of the entire mountain. Isn't that
sufficient?' asked the Father.

"I do not doubt any longer, Mr. Lesa, that the boy has
genius. Even you must be convinced it is worth while to
open the way so his gift can be developed, and of course
you will do so."

For a while Mr. Lesa blew out clouds of smoke in silence,

then said cautiously: "And what then? To develop it means
the boy will want to do nothing else than make music.
However, Vinzenz Lesa does not wish to make a musician of
his only son. Musicians are vagrants; they have no real
homes. If Vinzi once starts the roaming life, he will be
ruined. And I, who know this, am I to start him on it? No,
sir, you cannot ask it!"

Only after considerable reflection did Mr. Delrick speak,

for he had not expected such a reply. Then he said with
rising emotion, "It seems you cling to the idea that Vinzi
can be nothing more than a wandering musician. But let me
make a proposition. You surely have faith in me, Mr. Lesa?"

"I certainly have," rejoined the latter.

"Good! Then I propose that you let me have your boy

for a year or a little longer. I'll do with him just what I would
with my own son. When he comes back, if you think just
the same as today, a year abroad will not have done Vinzi
any harm. If he is to spend his life as a farmer, it cannot be
very bad to have learned something. That does everybody
good, whatever his occupation."

Mr. Lesa seemed to be weighing the matter, but

suddenly and firmly said, "That cannot be, sir," and when
asked his reason continued, "I do not care to send my son
to your house for a whole year as a visitor when you have
paid more than what was necessary for just a few weeks in
mine."

Mr. Delrick smiled as he explained, "I live all alone in my

big old house, which is quiet and often empty, so you can
understand your son would be a blessing and put life into
the old house. However, if this does not satisfy you, I will
promise to come back to your house as your guest so often
that it will balance any difference. Now shake hands on this,
Mr. Lesa; I hope you will never regret it."

Though many objections crowded into Mr. Lesa's mind,

he could not conquer the thought that a year abroad could
really do no harm to a boy like Vinzi; he had seen so little of
other people. Seeing how many boys of his age had to
struggle for a living might teach him to be grateful for his
pleasant home. Taking him from his free life in the country
to the paved streets of the city might make him homesick
and he would be glad to return even before the year was
out.

"I'll shake hands on that!" and Vinzenz Lesa pressed the

offered hand to seal the compact. "I will only add, sir, that
should Vinzi want to come home sooner than the time set,
you are to let him do so."

Mr. Delrick gave the promise unconditionally, and then

rose to take the news to Mrs. Lesa, while the farmer went
about his evening tasks in barn and stable. She could
scarce find words to express her joy and gratitude over the
turn of events. How wonderful it was for Vinzi to escape
living with his morose uncle! And although she did not know
just what Mr. Delrick had in view for Vinzi, she was
overjoyed at the prospect of her son spending a whole year
with such a man.

Vinzi knew no more about the plans than his mother,

but Mr. Delrick had told him he would follow the wishes of
Father Silvanus so he was sure all would be well with him.

Three days later was a time of farewell, but now no one

was sad. Stefeli alone was upset, for she thought no one in
the world had so lonesome a life as she. As a matter of fact,
she was much alone during the winter that followed, for she
had no companion on her way to and from school, and
sociable Stefeli sighed many times because of this
misfortune.

Now and again a letter came from Mr. Delrick bringing

news of Vinzi. Each one had to be read aloud by the mother.
The reports were always very favorable: the boy was in
good health and busy with his work; he sent hearty
greetings to all of them, and Mr. Delrick added a few
friendly words, reiterating what a pleasure it was to have
the boy with him and to watch his gratifying development.

The father always listened breathlessly to these reports

about his son, but they often seemed to disappoint him in
some way, and he would ask, "Is there no more?" as if he
hoped the letter would contain some other message. When
his wife assured him she had read it all, he would walk
silently away. She well knew he was expecting something
that did not come, and she guessed that he would have
been better pleased if Vinzi were not doing so well, if he
wanted to come home.

She began to worry once more. How would things go

when he did return? Would he do as his father wished? If
so, he would never be really happy. And if his father allowed
him to do the thing he loved most, that would make
dissension between the two, for the father would never
become reconciled to the thought that his son would be a
wandering musician. She remembered how intensely happy
her husband had been when a son was born, how he had
worked untiringly, how nothing had been too good for him.
When he looked on the baby, he would say, "He shall have
everything he wants." Such sympathy for her husband filled
her heart, that she felt she must call the boy home at once,
but the next moment she said to herself: "Oh, but if I do,
poor Vinzi will never be happy!"

In her anxiety, she was glad to remember she had

nothing to decide in the matter, and consoled herself with
the fact that, after all, the One who decides all things saw
further than any of them and He alone knew what was best
for one and all.
 CHAPTER X
OLD FRIENDS AND NEW LIFE

IT was spring again, and the trees and hedges were in

bloom. The grass was so fresh and so green that Stefeli
feasted her eyes on it as she walked through the meadows
with her schoolbag on her back.

Stefeli was coming home on the last day of school. No

more study until winter. How lovely that day had been the
year before when she had walked home with Vinzi, talking
over the days ahead. Then there lay before them the whole
beautiful summer with its long days on the sunny pasture.
But what would happen this summer? Stefeli foresaw many
long hot days in the house with the tiresome knitting, and
not a single day on the pasture. When she thought about it
she sat down in the meadow and sobbed aloud.

But Stefeli never cried very long, and when she

remembered that two days before she had seen half-ripe
strawberries behind the barn, she jumped up quickly; she
would toss her schoolbag in the hall and then gather the
berries. But when she flung open the door, she stood
riveted to the spot in surprise. Her mother sat talking quite
confidentially with a stranger, and beside him sat a boy the
size of Vinzi, who was taking a lively part in the
conversation.
 "Of course that is the little daughter," the man said,
glancing toward the door. "Come here, Stefeli; we are not
strangers. I am Cousin Lorenz and this is Jos, Vinzi's good
friend."

The delighted Stefeli went forward to greet them both;

anyone whom Vinzi loved was particularly welcome when
she was feeling so forlorn. She shook hands with Cousin
Lorenz for he looked at her with such kindly eyes, and then
stepped up to Jos, who smiled at her as though to say,
"We'll get along nicely together."

Stefeli turned to face her Cousin Lorenz and asked, "Jos

is going to stay with us the whole summer; isn't that so?
Just as long as Vinzi stayed with you."

Her cousin laughed heartily, and said, "That is surely a

fine welcome! We will see what your father thinks. I wish
you would take Jos outdoors for a while."

Stefeli did not wait for him to ask twice, but took Jos by
the hand and drew him happily along. He must see
everything in stable and barn, in garden and hen-house.

While they were gone, Vinzenz Lesa came in from his

work and greeted his guest with evident pleasure; it was
good to see his happy, contented cousin just when he was
so depressed.
 After the first greetings, Lorenz said, "We were worried
because we did not hear anything more from you, and my
wife gave me no rest. She feared we might not see your
Vinzi until late in the summer, and we want him with us all
summer. It occurred to me that I would bring our Jos to you
for a while, as we had arranged, and then later the two
boys could come to us until winter. But your wife now tells
me Vinzi is away, that I cannot even see him, and as to his
coming up to us, that is out of the question. My wife will be
bitterly disappointed; you cannot conceive how she loves
the boy, but he deserves it."

Hospitable Stefeli soon thought Jos ought certainly to

have something to eat, and came leading him into the room
like an old friend.

As Jos went up to greet him, Vinzenz Lesa looked into

the lad's open countenance with both pain and pleasure.

"He will be as big as you are, Lorenz," he declared after

looking Jos over critically again. "He must be a great help to
you already."

Mrs. Lesa had withdrawn some time before, and now

came to the door to signal to Stefeli, who immediately
began to lay the table.

"Your wife has a good helper, too," said Cousin Lorenz,

who approved the way the little girl went about her work.

The mother now came in to spread her table with the

best larder and cellar provided, for to feast those who had
shown such kindness to her Vinzi was a keen delight.

"You must stay with us a few days, cousin," said she as

she sat opposite him at table and saw to it that their plates
were replenished with second servings of the ham which
looked tempting indeed against the green leaves of the
tender lettuce her garden had supplied. "You must leave Jos
with us for several weeks at least."

"You make it easy for one to wish to stay, cousin," said

Lorenz. "I chose to come on Saturday in order to spend
Sunday with you, so I will gladly remain if you say so, but I
must return home on Monday. As to the boy, Cousin Vinzenz
shall settle that; I leave it to him."

"There's plenty of time for that," said the latter

deliberately. "We will take a walk through the fields
tomorrow morning. You would like to look over the farm,
wouldn't you. We can talk over things then."

"Look at the cow stable, father, above everything else,"

cried Jos with enthusiasm, who had remained silent out of
respect for his Cousin Vinzenz. But the impression he had
received in the stable was so powerful, he must speak now.
"There certainly are no finer cows than those in his stable,
and they are as clean as though they had just been
washed."

"I thought the cattle would please you," remarked his

father.

Early the next afternoon the two men wandered off

through the verdant fields that belonged to the Lesa farm,
and Mrs. Lesa took the children up the sunny slope where
the first strawberries were ripening. She knew that walk
would please them, and besides, the men wished to be
alone to talk. Looking at the blossoming trees, the lush
grass, the fields promising bountiful harvests, Vinzenz and
Lorenz reached the high ground where the woods began.
Before they stepped into the woodland path, Lorenz paused
to gaze down on the dwelling that looked so inviting among
the tall walnut trees.

"Vinzenz, you are a lucky man!" he exclaimed. "Peace

and happiness at home, and surrounded by acres that could
not be more beautiful, all of them your own."

"Yes, and over in Freiburg another place with twice as

many cattle as here, and a grass crop to fill the haymow to
the roof," but the furrows on Vinzenz's forehead grew
deeper and deeper as though each thing he mentioned was
worse than the last. "Twelve cheeses a year are made from
that milk."

"Vinzenz, you have no reason to be downhearted," said

his cousin with laughter in his eyes. "I never knew the
paternal estate belonged to you. Two such farms for your
own! Certainly God has showered you with blessings. Yet
you look as though you had nothing but bad weather for
your share."
 "It's easy for you to talk," said Vinzenz savagely. "You
have three strong sons who are happy in their work. Joy
and success are ahead of you. But after all my effort, I
must look on while a beautiful estate goes to ruin. I cannot
be in two places at one time, and my only son won't open
his eyes to see the fine career awaiting him. Hundreds
would envy him. When I inherited this farm, I left my
father's homestead, where every tree was a comrade and
every head of cattle had grown up under my eyes. I was not
happy to leave it, but everything here had gone to ruin. No
stranger would have undertaken to restore it, but I said to
myself, 'You will do it for the sake of your son. In a few
years he will be old enough to manage it and you can go
home again.' Well, the farm has been put into shape quicker
than I expected. You yourself say it looks like a blossoming
garden from one end to the other. Must I see it go to ruin,
or shall I let my homestead run down? Now tell me, what
do you think? Do you think singing and piping can take the
place of caring for an estate? You see how everything
stands with me!"

"Matters are not half as bad as you think," rejoined

Lorenz cheerfully. "You have a boy who will amount to
something some day, rest assured of that, cousin. And you
have a girl besides, of whom any father could be proud. Let
six or seven years pass. You are such a robust man you can
keep the two farms in condition until then, with some help.
By that time you can settle your daughter here; she will
know how to manage it, and you can return to your
homestead. I wouldn't wonder if someone would pop up
who would be willing to share the work and the
management with your daughter. Then your farm will have
the right care."

Lorenz had started to walk on, but now he suddenly

paused to say, "But I have forgotten to ask the principal
question. Do you want to keep Jos or shall I take him
home? He is fairly quick to learn."

"I can see that," remarked Vinzenz. "You will miss him,
and I am already in your debt for Vinzi was of no assistance
to you in any way."

But Lorenz remonstrated. Vinzenz ought to hear what

his wife would have to say about that; she would tell him
quite a different story, and with reason. It was she who had
urged him to bring Jos to them. She had never allowed any
of her boys to stay away even over night, but she was sure
Jos could learn only good things with the parents of such a
boy as Vinzi.

"Now tell me frankly," concluded Lorenz, "is there any

other reason you hesitate to keep Jos?"

"That is my only reason," was the decided answer.

"Then he will stay with you and you can send him home
whenever it suits you best."

Lorenz now quickened his pace for he wished to have

time to have a pleasant chat with Stefeli and her mother,
with both of whom he had established a close friendship.

When he bade Stefeli farewell, as he was leaving before

she would be up in the morning, he gave her his hand, but
she refused to say good-bye, and in the morning she was
standing at the door long before sunrise to meet him as he
came downstairs. She had grown so fond of him it was no
task to get up early to see him off. Then, too, she had
something on her mind and when he came down stairs, she
asked eagerly, "Can Jos stay? Can he stay all summer?"