Artificial Intelligence in Cancer: Diagnostic to Tailored Treatment

By Smaranda Belciug

Artificial Intelligence in Cancer: Diagnostic to Tailored Treatment provides theoretical concepts and practical techniques of AI and its applications in cancer management, building a roadmap on how to use AI in cancer at different stages of healthcare. It discusses topics such as the impactful role of AI during diagnosis and how it can support clinicians to make better decisions, AI tools to help pathologists identify exact types of cancer, how AI supports tumor profiling and can assist surgeons, and the gains in precision for oncologists using AI tools. Additionally, it provides information on AI used for survival and remission/recurrence analysis.

The book is a valuable source for bioinformaticians, cancer researchers, oncologists, clinicians and members of the biomedical field who want to understand the promising field of AI applications in cancer management.

• Discusses over 20 real cancer examples, bringing state-of-the-art cancer cases in which AI was used to help the medical personnel
• Presents over 100 diagrams, making it easier to comprehend AI’s results on a specific problem through visual resources
• Explains AI algorithms in a friendly manner, thus helping the reader implement or use them in a specific cancer case

• Language: English
• Publisher: Academic Press
• Release date: Jun 18, 2020
• ISBN: 9780128204108

Smaranda Belciug

Smaranda Belciug, PhD received her B.S. and MSc degrees in Computer Science from the University of Craiova, Romania, in 2006, and 2008, respectively and MSc degree in Computer Science from the University of Paris 12, France in 2007. She received her Ph.D. degree in Computer Science from the University of Pitesti, Romania in 2010. She is now Associate Professor with the Department of Computer Science, Faculty of Sciences, University of Craiova, Romania. Her research includes Artificial Intelligence, Machine Learning, Data Mining and Statistics focusing on cancer research. She authored 5 books, 3 book chapters and more than 25 scientific papers published in prestigious international journals. She won a special mention at the Young Researchers in Science and Technology contest, from Prof. Rada Mihalcea, University of Michigan.

Book preview

Artificial Intelligence in Cancer: diagnostic to tailored treatment

First Edition

Smaranda Belciug

Table of Contents

Cover image

Title page

Copyright

Dedication

Foreword

Preface

Chapter 1: Life challenge. Cancer

Abstract

1.1 Cancer throughout history

1.2 Where are we now?

1.3 Hope is around the corner. Artificial Intelligence steps in

Chapter 2: The beginnings

Abstract

2.1 Doctor's suspicion. Doctor + artificial intelligence combo's diagnosis

2.2 World collapses. Making an informed decision

Chapter 3: Pathologist at work

Abstract

3.1 Building the tumor's pattern

3.2 Artificial Intelligence and histology

3.3 Artificial Intelligence and immunohistochemistry

3.4 Artificial Intelligence and genetics

Chapter 4: Surgeon at work

Abstract

4.1 Learning everything about the tumor: Tumor profiling

4.2 Making a clean cut with the help of Artificial Intelligence

Chapter 5: Oncologist at work

Abstract

5.1 Establishing a treatment plan. Oncological guides

5.2 Chemotherapy and Artificial Intelligence

Chapter 6: Radiotherapist at work

Abstract

6.1 Establishing a treatment plan

6.2 Radiotherapy and Artificial Intelligence

Chapter 7: Survival analysis

Abstract

7.1 Kaplan-Meier survival curve

7.2 Life tables

7.3 Survival regression

7.4 The exponential distribution

7.5 Poisson process

7.6 The log normal distribution

7.7 The Weibull distribution

7.8 The log logistic distribution or Fisk distribution

7.9 Gamma distribution

7.10 The Gompertz distribution

Chapter 8: Remission and recurrence. What do to next?

Abstract

8.1 Decision trees

8.2 Self organizing maps or Kohonen networks

8.3 Cluster network

Chapter 9: Artificial Intelligence in cancer: Dreams may come true someday

Abstract

Index

Copyright

Academic Press is an imprint of Elsevier

125 London Wall, London EC2Y 5AS, United Kingdom

525 B Street, Suite 1650, San Diego, CA 92101, United States

50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom

© 2020 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices

Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN 978-0-12-820201-2

For information on all Academic Press publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Masucci, Stacy

Senior Acquisitions Editor: Teixeira, Rafael

Editorial Project Manager: Young, Sam W.

Production Project Manager: Raviraj, Selvaraj

Cover Designer: Hitchen, Miles

Typeset by SPi Global, India

Dedication

In loving memory of my father, Florin Gorunescu

If you remember me, then I don't care if everyone else forgets

Haruki Murakami—Kafka on the Shore

I will always remember you, Dad!

Foreword

Sally McClean, Portrush, Northern Ireland

This is an exceptional book. It takes the reader through the cancer journey of a patient, explaining at each stage what exactly it means for the patient and where artificial intelligence (AI) can and is helping all concerned to understand what is happening and ameliorate the journey and the path. As such it provides a very readable, yet technical, description of the whole area of AI for the cancer patient journey that can be understood and enjoyed by students and researchers from many areas of computer science, mathematics, statistics, physics, and related disciplines. Overall this is book which combines passion and emotion with a comprehensive treatment of the whole area, providing a very scholarly approach while communicating the basic ideas in a clear and understandable manner. Throughout there are lots of examples, web-links and illustrations that make it easy for the novice or lay-person to appreciate the concepts, while the more experienced reader can empathize with the patient, their family and the healthcare professionals while enjoying the background anecdotes, instances and explanations. Thus, the patient can learn to understand what is happening and how to interpret the evidence while the student or scientist learn about the human journey and the real potential of relevant technical methods to alleviate suffering.

As such the book is structured through the stages of the cancer patient journey starting with the initial presentation of symptoms and diagnosis through pathology and testing to treatment, involving the role of AI in surgery, oncology using Big Data and dose control in radiology. This is followed by an explanation of survival analysis, including lots of examples and case studies, so that the reader can better calculate and compute future likelihood of remission, recurrence, and life-expectancy of the patient, following their diagnosis and treatment. In the final chapter the future of cancer research is discussed, including the role of AI in palliative care.

Overall this is an outstanding book which grasps its topic with enthusiasm and compassion while showcasing the breadth, depth and applicability of AI in treating the cancer patient, as well as inspiring and motivating the reader to learn more about these exciting new developments and possibilities.

Preface

Smaranda Belciug, Craiova, Romania

Dear reader, by continuing reading this book you agree on embarking on a rollercoaster ride that will take you through the ups and downs of Artificial Intelligence applied in cancer research. You will pass every single stage that is lived by a cancer patient: finding out that she/he is suspected of having the disease, followed by all the tests, blood work and imaging, etc. The world collapses. But wait, a sparkle, a little light appears at the end of the tunnel. What is that? It is hope in the shape of Artificial Intelligence.

Step by step, we see how the doctor's suspicion rises, and how the diagnostic is set using Artificial Intelligence. After that we go into the pathology lab and watch the pathologist at work. We learn how Artificial Intelligence is applied in histology, immunohistochemistry, and genetics. One step further and we find ourselves in the operating room seeing the surgeon resecting tumors, after she/he preplanned the operation, learning everything about the tumor using Artificial Intelligence. The surgery is over, the patient lives, and now she/he is about to meet the rest of the oncology team. Using Big Data, Artificial Intelligence helps the oncologist establish a tailored treatment plan, taking into account everything she/he knows about the patient, from genetic data to her/his way of living. Maybe chemotherapy is not the only answer, maybe we need to follow the radiotherapist, and see how she/he can determine the right radiation dosage with the aid of Artificial Intelligence.

Through every step of the journey we have only one thing in mind: what are the odds of beating this monster? How long will the patient survive? We compute them with survival analysis. When the miracle phrase: you are in remission is said, we jump up and down with joy. But also we find out that the next 5 years are crucial. The word recurrence hits us right in the chest. What are the chances that the cancer will come back? Artificial Intelligence is here to help us compute the odds. Last but not least, in the final chapter, we learn what might come next in cancer research and how Artificial Intelligence has started to enter the palliative care also. Whether Artificial Intelligence can help our dreams become true or not, only time can tell.

This book is addressed to computer scientists, physicians, mathematicians, and the general public. I decided to write it in a friendly way, so that you will want to keep on reading it. I wanted you to understand that Artificial Intelligence is not voodoo, not magic, nor snake oil. It is math and super computational power. In some parts of cancer research, Artificial Intelligence is just in the beginning. I hope that in the future doctors and data scientists will continue to work together in discovering new ways of fighting this awful disease.

This book is dedicated to my father, Prof. Florin Gorunescu, whom I lost in 2018 to a rare aggressive form of esophageal cancer. So, when I wrote every chapter I had him in mind and in my thoughts, and I relived everything. That is why you may feel my emotions floating around in the pages of the book. I am a data scientist, and still I did not know how to understand the numbers. I wanted a miracle, and unfortunately the miracle did not happen for us. So, I will say that this book is dedicated to my father, and to all the people that are struggling with this disease, may you succeed in your battle, and to those who have lost the battle, may you all rest in peace.

I want to express my gratitude to Rafael Teixeira and Samuel Young for offering me this great opportunity of writing this book, and who so warmly sustained this project. Many thanks go out to Prof. Sally McClean for always supporting me, and for taking the time to review this book before being published, and more especially for always being my friend. Last but not least, I would like to thank my family for their understanding, love, and unconditional support.

As a final note, thank you Dad for being my mentor, my friend, and for teaching me Statistics. Until we meet again, may God hold you gentle in the palm of His hand.

Chapter 1

Life challenge. Cancer

Abstract

The aim of this chapter is to establish a psychological connection with the reader, for her/him to truly understand the vast implication of cancer in a person’s life, whether her/his own health is involved, or a loved one’s. Dealing with cancer implies physical, psychological, and financial issues. At first, we shall take a walk down history lane and see how cancer was seen in the past, followed by a recap from nowadays. A statistical analysis handbook that needs to be performed anytime we are dealing with new clinical trials, new drugs, new oncological surgeries, new AI algorithms, is presented also. Last, but not least, we shall paint a glimpse of the future, in which a trusted friend, Artificial Intelligence, steps in to lend us a helping hand.

Keywords

Cancer history; Hypothesis testing; Statistical tests; Statistical tables; Statistical parameters; Power analysis; Bayes’ theorem; Genome project; Imagistics

1.1 Cancer throughout history

Cancer seems like a Russian roulette. It is like everyday you hear that someone you know or heard of has just been diagnosed with it. It is scary. You feel insecure and think that you might be next. You comfort yourself with thoughts that you or your family won’t get hit by it. You eat the right food, exercise, sleep well, take your minerals and vitamins, you do not smoke, nor drink, and maybe, just maybe, you have no family history records with it. Still, you might have heard of people that never smoked, never ate bad food, were fit, and Zen, and still got cancer. Alarmists that say that the cancer rates are going up fuel all these thoughts. Are they right, or are they just adding fire to the general paranoia? Should you live in fear? We believe not. Does our answer mean that you will live a long and healthy life, cancer free? Not quite. According to http://ourworldindata.org/cancer (Accessed May 3, 2019), cancer is the second leading cause of death, after cardiovascular diseases. The Institute for Health Metrics and Evaluation (IHME)—https://www.healthdata.org (Accessed May 3, 2019) states that in 2017 between 9.2 and 9.7 million deaths were caused by various forms of cancer. So, things are not looking great. Now, the alarmists step in and say that in the past the rates were lower. That may be true, but we should take into account the fact that nowadays through screening tests and new Artificial Intelligence (AI) technology we detect cancer more. Thus, before embarking on our cancer history trip, we suggest that you take a deep breath, relax, and enjoy the moment. Carpe diem! You never know what might come next, so do not waste your time fearing it, What must be, shall be—Juliet, from William Shakespeare’s Romeo and Juliet.

There are many types of cancer. Cancer can begin in any part of the body. The moment when a cell starts growing out of control and starts crowding out the healthy cells, that is where it all begins. Due to the fact that any cell can become cancerous implies the fact that cancer is not just a single disease. Cancer can start in any organ, or even in the blood. Cancer that starts in the same type of cells are somehow alike, but still they are different in what regards their speed of growth and spread. That is what makes each cancer unique, and tailored accordingly to the person that has it.

Cancer develops in any living organism, whether it is a human being or an animal. Some of the first proofs of cancer have been found in Ancient Egypt among fossilized bone tumors. Osteosarcoma, or bone cancer, has been discovered in mummies (Hamada and Rida, 1972). In 1862, at Luxor, Edwin Smith, an American Egyptologist, came into custody of a medical papyrus that was named after his owner—the Edwin Smith Papyrus. The Edwin Smith Papyrus is the oldest known medical recording that dates back into 1700 BC, and it is thought to be a copy of the original manuscript dated in 3000 BC. It has 4.68 m in length and it is divided in 17 pages. In 1906, Smith’s daughter gave the papyrus to the New York Historical Society. The Edwin Smith Papyrus is a surgical document. In it, there are described eight cases of breast tumors or ulcerations that were removed through cauterization with the use of a tool called fire drill. Regarding this disease it is written, There is no treatment. Georg Moritz Ebers, a German Egyptologist and novelist, acquired another ancient papyrus, in Thebes in 1872. The Ebers papyrus dates back from 1500 BC, and included information regarding an incurable disease that produced tumors found in skin, uterus, stomach, and rectum. Compared to the Edwin Smith papyrus, this papyrus is medical, describing pharmacological, mechanical and magical treatments (Haas, 1999). An example of a cancer remedy is: when dealing with a tumor against the God Xenus the recommendation is do thou nothing there against (Hajdu, 2011).

Around 400 BC, Hippocrates named the non-ulcer forming and ulcer-forming tumors carcinos and carcinoma after the Greek word for crab. Remember that it was Karkinos, also known as Carcinus, the giant crab that helped Hydra in the legend battle with Herakles (or Hercules) at Lerna. Even if Herackles’ foot crushed Karkinos, the goddess Hera placed it amongst the stars, as the constellation cancer. There are some explanations to why he chose this name: (a) malignant tumors together with their blood vessels that have swollen, its adherents, resemble the picture of a crab with its legs spread in a circle; (b) malignant tumors have hard tissue that reminds of the hardness of the crab’s shell, and (c) the pain caused by the tumor can be associated with the pinch o a crab’s claw (Hajdu, 2011; Mitrus et al., 2012).

In what regards the term oncology, the name comes from the Ancient Greek word onkos a tragic mask worn in theater plays that symbolized the burdens carried by the person who wore it. On the other hand, oncos in Greek means swelling. Around the year 200 AD, the Greek physician Galen introduced the medical term onkos for all tumor types, whether they were malignant or benign.

As time passed by and Rome had fallen, Galen’s teachings regarding cancer started to reach physicians in Constantinople, Cairo, Alexandria, and Athens. While in the West, sorcery was used to cure diseases, in this part of the world scientists were trying to understand and explain cancer. They agreed with Hippocrates’ conclusions that cancer was a result of excess of black bile (a medieval belief that a humor secreted by kidneys or spleen that caused melancholy—according to Merriam Webster dictionary—https://www.merriam-webster.com/dictionary/black%20bile—Accessed May 2, 2019). Hippocrates hypothesized that cancer was correlated to the imbalance of the four body humors: blood, phlegm, yellow bile, and black bile. They also noted that cancer was curable only in its earliest stages. The treatment was based on arsenic.

Fast forwarding to the Middle Ages, due to religious beliefs that forbidded scientific improvement, cancer was supposedly be an infectious disease.

The excess of black bile idea continued to be present until the 16th century. During the Renaissance, scientists started performing autopsies (remember Rembrandt’s The Anatomy Lesson of Dr. Nicolaes Tulp). Andreas Vesalius, a 16th century anatomist, demonstrated that the black bile does not exist (Di Lonardo et al., 2015). Meanwhile, Gaspare Aselli, an Italian physician, discovered the lacteal vessels of the lymphatic system, and also advocated the fact that abnormalities of the lymph may cause cancer. It was in the same century, when Paracelsus, born Philippus Aureolus Theophrastus Bombastus von Hohenheim, a Swiss physician, started the medical revolution of the Renaissance. Paracelsus is the "father of toxicology," due to his studies concerning the tumors of mine workers. He suggested that the deposits of sulfur and arsenic salts found in their blood caused the cancer (Hajdu, 2011; Mitrus et al., 2012; Henschen, 1968; Di Lonardo et al., 2015). Another opinion came from Paris in the 1730s, when the physician Claude Gendron stated that cancer results from a locally hard, growing mass, that does not respond to drug treatment, and thus must be removed with all its adhesions.

Herman Boerhaave, a Dutch physician, first introduced the idea of cancer being hereditary. He implied that … cancer was most likely induced by viruses present in water or soil. Once acquired, the cancer viruses remained in the body, and they could be transferred by contagious infections or by heredity (Di Lonardo et al., 2015). We can easily see the similarities between the idea of an infectious disease present in the Middle Ages and the toxicity discovered by Paracelsus.

Tumors started to be better investigated when researchers started using the microscope. In 1838, Johannes Peter Muller, a German physiologist, anatomist, ichthyologist, and herpetologist, suggested that cells form cancer, while in 1869, Rudolf Ludwig Carl Virchow, a German physician, proposed that cancer is cell disease.

Giovanni Battista Morgagni, an Italian anatomist, founded in the 18th century the modern anatomical pathology. He taught pathology for 56 years at the University of Padua. Regarding cancer, he reported that it is the result of organ lesion. In the 18th century hospitals that specialized in cancer were opened. Between 1871 and 1874, it was the English surgeon Campbell de Morgan that defined the term metastatis, the cancer poison that spread from the primary tumor to the nearby lymphatic nodes. Surgery was the first course of treatment, but unfortunately, due to precarious hygiene few people would survive it. The Scottish surgeon Alexander Morno reported only two out of sixty breast cancer surgery survivors. The number of survivors went up when aseptic surgeries started to take place.

In the late 19th century, Marie Sklodowska Currie and Pierre Currie discovered radiation, thus making possible a non-surgical cancer treatment.

In 1911, Francis Petyon Rous, an American Nobel Prize—winning virologist, documented a viral cause of cancer in chickens. Being a pathologist, he diagnosed a large lump in the breast of a hen brought to him by a farmer as a sarcoma. He verified if the tumor could be transplanted into other hens that resembled the original one. After establishing that this was possible, he observed that with each passing the tumor became more and more aggressive. The next step was to test whether a virus caused the cancer, thus Rous shredded a sample tumor in saline solution, and passed it afterwards through a filter that removed the bacteria and tumor cells. This extract was injected into healthy hens and curiously enough it produced new tumors (Rous, 1911a,b). His findings were very controversial, so he abandoned cancer research until the 1930s, when a colleague of his, Richard Shope, revealed that another virus, papilloma, caused tumors in rabbits. In the 1960s, scientists discovered the src gene that produces a protein that leads to tumors. After 50 years since the discovery of the sarcoma virus, today known as a retrovirus, Rous won a Nobel Prize. It was Rous’ study that inspired other scientists hence new examples of tumor-induced viruses in other nonhuman primates (rabbits, cats, mice, etc.) were performed (Shope and Hurst, 1933; Bittner, 1942; Gross, 1951; Sweet and Hilleman, 1960). In 1964, the Epstein-Barr virus, a human herpesvirus that has been associated with lymphoid and epithelial tumors, was observed (Epstein et al., 1964).

However, it was not Rous who described the first tumor virus. The Danish physician-veterinarian duo of Vilhelm Elerman and Oluf Bang proved in 1908 that a filterable extract could transmit leukemia among chickens (Ellerman and Bang, 1908). You may ask yourself why was this important finding overlooked, and the answer is that leukemia was no recognized as a neoplastic disease up until 1930 (Van Epps, 2005). In Rous’ Nobel lecture the work of Ellerman and Bang was mentioned.

David Paul von Hansemann, a German pathologist, described in his works, published between 1890 and 1919, the multipolar mitoses in animal cells. Theodor Boveri, a German biologist, continued the Hansemann theory and proposed in 1914 the chromosomal theory of cancer, with respect to the abnormal numbers of chromosomes arising in cells by multipolar mitoses in adult cells. Boveri’s theory was based on the observation of some cells in early sea urchin embryos that were having abnormal chromosome complements wandering from their normal growth paths (Balmain, 2001; Boveri, 2008). Boveri wrote in his paper Zur frage der entstehung maligner tumoren (Boveri, 1914), that tumor growth is based on … a particular, incorrect chromosome combination which is the cause of the abnormal growth characteristics passed on to daughter cells. Boveri’s theory is often recognized as the first chromosomal theory of cancer, when in fact it was Hansemann’s contribution that is the original and the more significant hypothesis (Bignold et al., 2006).

In 1907, the American Association for Cancer Research was founded (Creech, 1979). In 1913, with all these discoveries people were still uninformed regarding cancer and a wave of public fear started. Hence, a major breakthrough took place: an article regarding cancer’s warning signs was published in a woman’s magazine, Ladies’ Home Journal (Adams, 1913).

In the 1900s, cancer research was marching at full speed. Some of the most important concepts and accomplishments are presented in Table 1.1.

Table 1.1

In the last two decades cancer research produced massive amounts of data. Knowledge on human genomes along with protein expressions (mass-spectrometry—MS), metabolomic (metabolomics is the comprehensive analysis of metabolites—small molecule that intermediates and produces the metabolism, which is a formidable tool in precision medicine; Clish, 2015), and clinical data, needs to be processed fast and accurate in order to identify cancer in novel patients and to tailor therapy and monitoring. So, as we let go of the past, we shall commence our next section where the present state-of-the-art on cancer research is discussed.

1.2 Where are we now?

A great part of this book will regard Statistics. Statistics provides the perfect compass for traveling along AI through the depths of cancer research. Still, one must keep in mind the fact that any measure, any result, that we obtain using Statistics is not a certainty. In this day of age the majority of us use Google to obtain our diagnosis before talking to a healthcare professional. It is a well known fact that if you Google any symptom, whether we are talking about a toothache or bunion, cancer will appear on our search. Obviously, after seeing a physician, one shall Google the diagnosis. If that person has been diagnosed with cancer she/he will start looking at those numbers that appear before her/his eyes. What is the 5-year survival rate? What is the overall survival rate? What is the morbidity or mortality regarding a certain cancer surgery? People Google forums, support groups, trying to see whether they or their loved ones will fall under the win category. Seeing the percentages some build up confidence, others fall in depression.

Statistics is a good measure, but in order to trust it we need to know how to interpret it. Because some numbers without proper analysis might lead to wrong conclusions. For example, let us look at the data provided by the World Cancer Research Fund—American Institute for Cancer Research—https://www.wcrf.org/dietandcancer/cancer-trends/data-cancer-frequency-country (Accessed May 4, 2019) in Table 1.2. The data is age-standardized rate per 100,000.

Table 1.2

From Table 1.2, we depict that the highest cancer rate for both men and women was in Australia, and also that the top 12 countries come from Europe, North America, and Oceania. We could draw the conclusion that the developed countries (Australia, New Zealand, Ireland, United States, Belgium, metropolitan France, Denmark, Norway, etc.) have higher rates of cancer, thus implying that some factors related to the lifestyle of their citizens lead to cancer. We strongly believe this is not the case. Our explanation is that in these developed countries the screening procedures and state-of-the-art diagnosing techniques detect cancer at a faster pace than in the other countries. Having screening programs and easy access to medical care without no doubt lead to discovering more cancer cases. The fact that other countries do not report cancer cases does not imply that those are safer. More on data and statistics will be covered in what follows.

Statistical assessment of research

Statistical analysis is essential. Most of the biomedical research regards designing and implementing new algorithms. Sadly enough, the most important part, the evaluation of their efficiency, is still overlooked. Medical residents and computer science MSc or PhD students lack the know-how of understanding common statistics that are found in clinical journals (Windish et al., 2007). If one fails in comprehending crucial information in those research papers, then it is clear that she/he cannot apply the new found knowledge in practice, ultimately stopping the advancement in patient treatment and subsidiarity, in science. There are several organisms that support and argue about the importance of statistics in medicine, including the International Medical Informatics Association (IMIA) and the National Library Medicine (NLM) degree-granting programs in biomedical informatics. IMIA includes statistics in their recommendation regarding medical informatics education, and NLM requires introductory course in biostatistics (Johnson, 2003; IMIA, 2000; NLM, 2007).

In 2010, a review of the use of statistical analysis in biomedical informatics literature has been done, and the authors’ findings are that the main statistical tools that are used in research paper are: descriptive statistics, elementary statistics, multivariable statistics, and regression analysis (Scotch et al., 2010). If we look at the numbers presented in the paper, we discover some disturbing facts, such as:

•18% of the papers published in the Journal of American Medical Informatics Association (JAMIA) and 36% paper published in the International Journal of Medical Informatics (IJMI) between 2000 and 2007 have no statistics whatsoever in them;

•71% from JAMIA and 62% from IJMI contain basic descriptive statistics;

•42% from JAMIA and 22% from IJMI contain elementary statistics;

•12% from JAMIA and 6% form IJMI contain multivariable statistics;

•9% from JAMIA and 6% from IJMI contain machine learning.

Our goal is to try and change this fact. So buckle your seatbelts, because in what follows we shall present step-by-step statistical tools, so that the reader gains this very important skill in biomedical informatics.

An old saying is medicine is not mathematics, which is true, because in medicine nothing is 100% accurate. On an endoscopy, for instance, an esophageal tumor may appear to be only stage II, when in fact it is stage IV. On MRI small tumors, metastasis, may not be seen with the naked eye, due to the fact that they are measured in microns. A surgeon knows that she/he cannot rely 100% on the scans that she/he reads before going into surgery. She/he knows that things may not be as they seem on the scans or on their interpretation. An oncologist knows that each person is unique, thus hers/his cancer is unique and it may respond different from others to a certain course of treatment.

Accuracy in medical diagnosis is still not perfect, producing false positives or false negatives.

False positive and false negative

A false positive is when a person receives a positive result when she/he should have received a negative result. For instance a woman is being told after a routine mammogram that she has breast cancer, when in fact it was just a misread on the scan. In statistics, this represents a Type I error, or α. A Type I error implies the incorrect rejection of the true null hypothesis. The null hypothesis, notated H0, is the generally accepted hypothesis, while its opposite, notated H1, is the alternative hypothesis. Theoretically, in order to test a statistic hypothesis, the whole population should be examined. Due to the fact, that this is practically impossible, a random sample of the population is chosen, just as Jules Henri Poincaré stated: Our weakness does not allow us to embrace the entire universe, and obliges us to decompose it into slices. Please keep in mind that the random sample should match the characteristics of the whole population (e.g. if we have 54% women and 46% men in the population, then in the sample population the percentage should remain the same, 54% vs 46%). After this step, a hypothesis is formulated. An example of such a hypothesis is: a new drug produces better outcome, thus being more effective than the drug that is in use today for a certain cancer treatment. In this case we have:

•H0: states that there is no significant difference between the two drugs, the difference reported by the drug company simply being the result of hazard. The null hypothesis refers to the general accepted hypothesis, which the statisticians are trying to reject taking into account new discoveries. Thus, we can say that the true research hypothesis is H1.

•H1: states that there are significant differences between the new drug and the classical one, differences that are not the result of hazard.

A false negative is the opposite concept, when a person receives a negative result on a medical test when she/he should have received a positive result.

When in a private conversation a person uses the terms statistical significance, many of us do not comprehend it’s meaning, and we just agree and nod our heads, considering her/him as being some sort of a statistic guru that has just performed complex computations which proved that her/his point cannot be questioned. Now, it is time to show that this statistical significant subject is not rocket science, and absolutely everyone should grasp it. Statistical significance implies know-how on three simple concepts: hypothesis testing, the Normal distribution, and the famous p-level.

How do we determine which hypothesis is correct, based on the evidence we have? There are two types of statistical tests: parametric and non-parametric.

•The parametric test refers to statistical hypothesis, which regard statistical parameters such as the mean, standard deviation or dispersion, as well as the distributions, which govern the data. These types of test include: t-test and ANOVA. Both tests assume that the data has a Normal distribution.

•The non-parametric test does not make presumptions on the population parameters or distributions. They are used on data that is not governed by the Normal distribution. These type of tests include: chi-square and Mann Whitney U test.

Each parametric test has its non-parametric equivalent. For instance, if you have parametric data from two independent groups then you simply apply the t-test for mean comparison, otherwise if you do not have parametric data, you apply the Wilcoxon rank-sum test for mean comparison.

Before we proceed with the presentation of some of the most used statistical tests, we shall discuss the p-level and Normal distribution.

p-Level

p-level is a number between 0 and 1 that defines the significance of the obtained results. We can interpret the p-level values as it follows:

•if the p-level has a small value, usually ≤ 0.05, this indicates that there exists strong evidence against the null hypothesis, thus the results are significant and one must reject H0.

•if the p-level value is large (> 0.05), this indicates that the evidence is not sufficient to reject the null hypothesis, the results not being statistically significant enough.

We would like to underline the fact that above-mentioned cut-off values are arbitrary and offer just an assumption. Let us suppose that our p-value is 0.051. Does this mean that the obtained results are not significant