AIMED Issue #02V02 by Ezequias Rodrigues da Rocha

ISSN 2516-5690

772516

570407

V02#02 THE CARDIOLOGY & RADIOLOGY ISSUE

To appraise images with new lens and facilitate detection before onset

THE POWER OF AI IN TWO CLINICAL REALMS

P.08

Can Physician and Data Scientist work together? P.50

AI-driven method to address physician burnout

P.56

Sneak peek into one of Europeâ&#x20AC;&#x2122;s largest AI events

THE DEEP DIVE

Radiologists and cardiologists in AI?

P.62

P.11

The top 100 AIMed Articles - Part One

ai-med.io

CARDIOLOGY

EXPERIENCE THE FUTURE OF CARDIOLOGY & CARDIAC SURGERY RITZ CARLTON CHICAGO Anthony Chang Chief Intelligence Officer M13 CHOC and Founder AIMed

17-18 JUNE 2019 Key topics

Deep Learning and Cardiac Image Interpretation AI in Decision Support in Cardiology and Cardiac Surgery Precision Cardiovascular Medicine Big Data and Databases in Cardiology and Cardiac Surgery Wearable Technology and Embedded AI for Cardiac Care Robotics and Virtual Assistants Augmented/Virtual Reality in Cardiology and Cardiac Surgery Future of Heart Program Administration Using AI Medical Education and Training Using AI Blockchain and Cybersecurity in Heart Program

TOWARDS A MORE HUMAN ARTIFICIAL INTELLIGENCE A MESSAGE FROM AI MED CHAIR AND FOUNDER DR. ANTHONY CHANG

â&#x20AC;&#x153;I think that artificial intelligence is almost a humanities discipline. It is really an attempt to understand human intelligence and human cognition.â&#x20AC;? SEBASTIAN THRUN, PROFESSOR OF COMPUTER SCIENCE, STANFORD UNIVERSITY

odern technology has been incessantly debated in the news recently with the Boeing Max 8 fatal crashes. The automation failure is a stark reminder for us to respect a continual human to machine relationship and human cognition remains essential during this maturing process. The faulty sensors also remind us that the exponential increase in health and biomedical sensors will need a high level of scrutiny as mistakes cost lives, perhaps more than the airplane crashes. If the plane sensor had become vulnerable to errors, perhaps additional sensors or better, human cognitive elements should be in place to correct it. The other major news item is the publication of the FDA white paper. Traditional regulatory processes are becoming inadequate to oversight the ultrafast-

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

RADIOLOGY

EXPERIENCE THE FUTURE OF RADIOLOGY RITZ CARLTON CHICAGO Anthony Chang Chief Intelligence Officer M13 CHOC and Founder AIMed

18 -19 JUNE 2019 Key topics

The need for unbiased Ground Truth Data. Overhaul of IT infrastructure within the healthcare system at large. Patient privacy/protection issues. Develop comprehensive partnerships between clinicians with the vendor community to develop tools of clinical utility and usability. How do we speed up research without putting scientific integrity at risk. Is there a hybrid model?

GENERAL ENQUIRIES enquiries@ai-med.io www.ai-med.io EDITORIAL ENQUIRIES Charlie Editorial Advisor E: charlie@ai-med.io T: +44 7901 850 319 ADVERTISING ENQUIRIES Andrew Johnson Commercial Director E: aj@ai-med.io T: +44 7769 274 142 TALK TO DR CHANG Dr Anthony Chang Chair & Founder E: achang007@aol.com T: +1 (949) 547 8902

evolution of software that can change in real-time in a matter of seconds. The FDA has proposed new submission type and data requirements based on risk, in the form of 510(k) notification, De Novo, or premarket approval application (PMA) pathway. The paper reflects a more innovative strategy (including an algorithm change protocol, or ACP) to the total product lifecycle (TPLC) regulatory approach, that will be more appropriate for new devices. Alan Turing, the father of artificial intelligence, reminded us that machines should be created to engage machines. Therefore, perhaps we eventually will need to develop “regulatory” algorithms. We continually evolve the AIMed magazine based upon your feedback. This issue of AI Med will have a new type of theme, Subspecialties in Focus: Radiology and Cardiology. We wanted to focus on two clinical areas that are very AI-centric right now, especially with the robust medical image interpretation AI tools being developed. By focusing on various subspecialties, we hope to provide even more relevant AI information to clinicians and all those who engage with them. In addition, an exciting new feature of the AIMed magazine which reflects our aim to be even more academic will be AI in Medicine: The Top 100 Articles. These are impactful current and past articles that are particularly well suited to both clinicians and non-clinicians . We will list 5 articles at a time, each accompanied by a brief commentary, and with immediate relevance to the theme of the current issue. Although AlphaGo Zero and other programs have heralded the advent of high-level self-learning AI, artificial intelligence remains mostly human intelligence imprinted into machines that reflect human thinking. Interestingly, there are biological terms for this anthropocentric human to machine relationship. The first stage is symbiosis, or the living together of two dissimilar organisms. I believe that we are mostly at this stage currently. The second stage is synergy in which the total combined effect is usually greater than the sum of the parts. Muscles and nerves are often “synergistic”. We already have the early underpinnings for this level of relationship. The final stage is probably convolution, a term that describes the sinuous ridges of the brain biologically or a third function that is promulgated from two existing functions mathematically. In our lifetime, we will observe human and machine intelligences be harmoniously intertwined. We may even stop calling anything “artificial” intelligence.

Anthony Chang

MD, MBA, MPH, MS Chief Intelligence and Innovation Officer Medical Director, The Sharon Disney Lund Medical Intelligence and Innovation Institute (MI3) Children’s Hospital of Orange County

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

CONTENTS THE DEEP DIVE 12

Ethical and legal challenges of Artificial Intelligence in Cardiology AI is promising but at the same time, it also embeds challenges that are beyond the reach of the technology

AI in Radiology: Asking some hard questions Radiology has been a synonym of change, but AI takes the reform a little differently

Making data work for hearts at risk A dashboard which gives clinicians access to the data needed to change care

How can AI be used in Radiology â&#x20AC;&#x201C; an Outline for Algorithm Development Breaking down of the radiology workflow into the lifecycle of an examination

LEAD FEATURE Can Physician and Data Scientist work together? No one is able to clap with one hand, so can AI

08 12

TECH COMPANIES 40

To Build, Shape and Sustain: The goals of integrating AI in healthcare A hands-on, practical and step-by-step guide to developing or adopting AI solutions into your organization

Genomics + Radiomics = A step towards better healthcare Extracting data from an image and combining it with the genomic profile of a tumor will lead to a better cancer diagnosis and treatment

The man behind no-survey survey An innovative, AI-driven method to look into physician burnout is now on its way

PHYSICIANS 50

The warrior and his invisible battles This is the answer to how one man manages his medical condition while setting up a company at the same time

56 THE BIG PICTURE 56

Experience AIMed Cardiology and Radiology this June in Chicago AIMed is hosting subspecialty meetings on two AI-centric clinical areas for the very first time

One of Europe’s biggest AI and digital health events that you don’t want to miss HIMSS Europe & Health 2.0 conference taking place this June will showcase some of the finest AI innovations in the region

NOTE FROM THE EDITOR Finding the right balance: A lesson for AIMed and healthcare At AIMed, we always think out of the box. We formed a community of those who believe in the power of AI and related new technology in changing the field of medicine and healthcare. We have organized and live-streamed conferences, workshops, and webinars to share relevant knowledge with audiences around the World, and we publish an academic magazine outlining the latest AI news and developments. Some of the attempts may appear contradictory; such as encouraging experts with different expertise to work together (i.e., Lead feature: Can physician and data scientist work together?), introducing new workflow to healthcare professionals (i.e., Deep Dive: How can AI be used in Radiology – an outline for algorithm development), and featuring two subspecialties in the same issue (like the one you are reading now). How do we do it? Sometimes, it’s all about finding a balance; the right balance. Yes, there will be challenges along the way but if you keep an open mind, you will surely see a different World. Like the one AI is about to create in the World of medicine.

THE TEAM EDITORIAL

STRATEGIC TEAM

CONTRIBUTORS

Charlie Moloney Editorial Advisor

Anthony Chang, MD, MBA, MPH, MS AIMed Chair & Founder

Sara Gerke

Hiten Thaker Art Director Hazel Tang Staff Writer

CENTRAL TEAM

THE DIGEST

Emma Chitty General Manager

62 AIMed Top 100 Articles 64 Book of the Month 65 Executive Summaries

Suzy White Operations Manager

Freddy White CEO

Spyro Mousses, PhD Strategic Advisor Laura Beken Strategic Advisor Nathaniel Bischoff Strategic Advisor

Daniel B. Kramer I. Glenn Cohen Tirath Y. Patel Nina Kottler Dr. William Bradlow Tom Sullivan

Sam King, MPH, MBA Strategic Advisor

Priya Samant Partnerships & Ambassadors PUBLISHERS STATEMENT: AIMED EVENTS LTD, West Sussex, England

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

LEAD FEATURE

CAN PHYSICIAN AND DATA SCIENTIST WORK TOGETHER?

Friendships come in many forms. It can be as literary as Willkie Collins and Charles Dickens, as poetic as Robert Frost and Edward Thomas, as intellectual as Holmes and Watson or as carefree as Sal Paradise and Dean Moriaty. When it comes to artificial intelligence (AI) and medicine, their names are Rajesh Dash and Sushant Shankar. Dr. Dash is a practicing Cardiologist, Assistant Professor of Medicine at the Stanford University Medical Center. Shankar is a software engineer/data scientist with extensive experience in E-commerce, analytics and AdTech. The duo is not only worlds apart in specialties, they also have a near two decades of age difference. However, none of these deter them from cofounding HealthPals. A startup which screens patients at a population level to uncover gaps within the healthcare system that prevent individuals from receiving adequate care. HealthPals’ intelligence clinical platform – CLINT, uses machine learning to ingest data from patients’ medical records and other registries to generate personalized treatment pathways. The inspiration to begin HealthPals came four and a half years ago, when Dr. Dash opened a clinic, SSATHI (means partner in Hindi) based in Stanford which focused on researching and preventing heart disease in young South Asians; a population found to have the highest rate of heart disease globally.

Dr. Dash noticed when patients arrived at his clinic for the very first time, their lab test results often suggested that they should be on some form of medications to control their blood pressure, diabetes or high cholesterol for basic prevention. Yet, most of them were not. These gaps in care were concerning. Dr. Dash believes if our system is effective in improving outcomes, it should be capable of controlling onset, and this is where technology can come to the rescue.

TO PERSUADE A PHYSICIAN INTO AI That being said, Dr. Dash feels it still takes courage for medical professionals to accept AI. It also makes sense for them to shy away from the technology because in most parts of the world physicians are trained in evidencebased care that requires continuous validation and peer critiques to ascertain the standard. Whilst it is a good practice, machine learning is gradually changing the tradition. Technology makes one wonder if it is truly applicable to the

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

LEAD FEATURE

real World, its non-transparent nature is not something the healthcare community is familiar with, so many find it threatening. Shankar agrees. He observed that, most of the time, people are venturing into the building of a new technology but at the core of it, there is hardly any real adoption. It is challenging to change a culture, so what HealthPals does is to look at the task at a higher level. “For example, if you have high cholesterol and meet the guideline criteria, you will be put on a cholesterol-lowering medication called Statin. There are potentially seven types of Statin in the market, and probably one or two which work best for a patient like you. HealthPals is using real-World data to guide physicians to the most effective treatment,” Dr. Dash explained. As such, Shankar asserted, although HealthPals is based in the Bay Area, it is targeted at international health systems that are much more likely to adopt the approach. Thus far, HealthPals had found some early adopters and they have also formed an innovation partnership with the American College of Cardiology. Nevertheless, in the wake of recent medical and data scandals, HealthPals is taking additional measures. Shankar said they are willing to move things slower to accommodate more time to clinically validate their solutions.

PHYSICIAN-DATA SCIENTIST PARTNERSHIP Dr. Dash and Shankar found each other on Foundersdating.com, a networking site which matches founders with co-founders to turn different ideas into reality. The pair forms what AIMed has always been advocating: an alliance between healthcare professionals and technology experts. When asked what makes this partnership work, both Dr. Dash and Shankar humbly regard patience and frequent feedback as the key.

“I basically knew nothing (about data and building software system) when I started. I feel like I have learnt a lot just working with Sushant every day for the past four years. He has the patience with me to turn my ambition into something possible with engineering. The inevitable causecorrection for clinical relevance through getting feedback on a very frequent basis,” Dr. Dash said. “Generally, it takes months for data scientists to build a product because you want to get a product market fit or ideally, a validation from the end-users. So, the opportunity here is, we are able to combine the effort of two people. I can receive a physician feedback right away on whether I am heading in the right direction. We generate insights using both of our experiences” Shankar added. They count themselves lucky to be in a very good working relationship and being able to share that common language of building something together with the use of technology. They advised others to put collaboration as something high on the list. Once that is formed, be practically critical with one another’s work and arrive at a common ground where knowledge from both sides are duly valued. As Shankar put it, “what I appreciate Rajesh most is that he doesn’t only pinpoint, he is also trying to solve those pinpoints. Because he is doing so manually, that’s where I see an opportunity to bridge the gap using my expertise.” “Ultimately, it is the acceptance of: I didn’t have what I knew I needed and what I didn’t even know that I needed to build what I have imagined building, and I needed the resource to help me find someone so that I don’t travel in a circle,” Dr. Dash noted.

THE DEEP DIV E CARDIOLOGY & RADIOLOGY

From computed tomography (CT), magnetic resonance imaging (MRI) to the introduction of picture archiving and communication system (PACS), radiology has always been at the forefront of technological progress. Now, AI-driven algorithms show promises in the detection, characterization and monitoring of medical conditions, as it is able to digest a large amount of imaging data to generate quantitative assessments of complex patterns. Something which can be performed by human radiologists only after many years of training and experiences. At the same time, cardiology is probably just as techie. Be it built-in intelligence which captures abnormal heart rates and revives a patient when in need or wearables which permit remote supervision of patients. AI is injecting much enthusiasm in these two clinical areas, but it has also brought out many significant challenges in safety, security and workflow integration etc. Nevertheless, in the long run, AI is considered a reliable partner, assisting both cardiologists and radiologists to provide more efficient care.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE

BY: SARA GERKE, DANIEL B. KRAMER AND I. GLENN COHEN

ETHICAL AND LEGAL CHALLENGES OF ARTIFICIAL INTELLIGENCE IN CARDIOLOGY

rtificial intelligence (AI) offers new opportunities to improve diagnosis and treatment across a wide spectrum of cardiovascular conditions. In theory, algorithms driven by AI can interpret the torrent of physiologic data emerging from implantable and wearable devices to refine the diagnosis of conditions such as atrial fibrillation (a common arrhythmia that can increase the risk for stroke) and congestive heart failure (i.e., when the heart muscle cannot pump effectively enough to meet the bodyâ&#x20AC;&#x2122;s metabolic demands), in which timely identification can lead to meaningful treatment changes. AI may also help mine new and existing data sources to improve the precision of treatment delivery, identifying patients most and least likely to benefit from medications designed to prevent blood clots and strokes, or devices such as implantable cardioverter-defibrillators (ICDs) and cardiac resynchronization therapy (CRT). However, applying AI to these areas of critical clinical need raises key questions for regulators, clinicians, and patients. Thus, this article discusses the promise, alongside the legal and ethical challenges of AI in cardiology.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

POTENTIAL OF AI IN CARDIOLOGY Pacemakers are implantable devices that treat abnormally slow heart rhythms. Wires are placed into the heart through the venous system and attached to a “generator” placed under the skin on the chest, which includes the battery and software that governs device function. ICDs are similar, but also have the ability to deliver highvoltage shocks to restore a normal heart rhythm if patients develop a life-threateningly fast heart rhythm. In some patients with heart failure, a pacemaker or ICD system can include an additional wire placed on the back of the heart in order to “resynchronize” the pumping function of the heart, also called “CRT”. The generators for each of these device types collect massive amounts of physiologic information.

This includes data central to the function of these devices, such as battery and wire measurements, as well as more tangential information about patients’ heart rate profile, physical activity, temperature, and ambient arrhythmias. Devices implanted in patients with heart failure may also illustrate transthoracic impedance (a way to approximate fluid retention) and other metrics of disease status. Complex analysis of these data sets have yielded mixed results: Algorithms designed to target early problems with ICD leads, for example, can successfully identify device malfunction before patients are harmed.1 By contrast, clinically meaningful heart failure interventions based on device data have not been widely integrated into patient care despite promising studies.2

AI could also be leveraged to resolve two intractable problems related to ICDs and CRT in particular. Clinical guidelines for implantation of both device types draw upon modestlysized clinical trials, with professional society recommendations largely driven by the average treatment effects realized in these pivotal studies.3 However, observational data demonstrates marked heterogeneity of treatment response to both ICDs and CRT.4, 5 Most ICD recipients never receive an appropriate therapy from their devices, yet incur the lifelong expense and risk for complications. Similarly, a quarter of CRT recipients do not respond to treatment, and no current algorithms meaningfully improve that yield.4 AI applied to existing data sources, or new areas such as electronic health records or pre-implantation wearable diagnostics, could potentially improve the selection of device recipients and the associated cost-effectiveness of these expensive therapies. Pacemakers and ICDs often also include software for diagnosing new cases of atrial fibrillation. The simplest of these records high atrial rates and displays corresponding electrical signals for manual review. More sophisticated approaches, including the proprietary algorithm underlying the Apple Watch atrial fibrillation diagnostics, rely heavily on the variability in heart rate. Millions of patients with both implantable and wearable devices evaluating these arrhythmia patterns could potentially benefit from an AI-driven approach to improve the sensitivity and specificity of device-adjudicated atrial fibrillation.

ICD PACEMAKER SHOWING IN CHEST X-RAY A HUMAN

Clinically meaningful heart failure interventions based on device data have not been widely integrated into patient care despite promising studies.

BIAS AND FAIRNESS AI applications to these and other cardiology projects will only be as good as the training data that is fed to them. While an algorithm can learn to identify new arrhythmias, guide selection for new device implants, or predict key outcomes if trained with the right data; the results may be limited for all or selected patients if the original derivation draws from biased data. Experts are concerned that AI could simply automate human biases, such as gender and racial biases, rather than remove them, especially if those biases infect the training data.6 For example, an algorithm intended to optimize device placement for cost-effectiveness may reinforce or amplify socioeconomic disparities if patient factors such as medication compliance, frequency of

THE DEEP DIVE

doctor visits, and insurance status markedly influence the clinical outcomes of interest. Algorithms whose training data come from homogenous populations may also be simply inaccurate when generalized to more diverse cohorts. For this reason, it is essential to diversify data for AI training intended to be broadly applicable, but ensuring safeguards against amplification of bias remains an important problem.

INFORMED CONSENT The use of AI to assist cardiologists in diagnosing or treating patients also brings new challenges for the patient’s right to informed consent. How can clinical and patient/consumerfacing health AI accommodate informed consent as currently conceived and practiced, and what alterations to either AI or informed consent might be necessary? It is time to squarely face the question or whether informed consent is the appropriate paradigm for the use of AI in cardiology or whether modifications to informed consent are needed for this use? FDA-cleared approaches to arrhythmia identification incorporate proprietary “black box” computation,7 which asks clinicians to consider the extent to which they need to understand not just whether a specific technology works, but how it achieves those results. How understandable should AI be when introduced to clinical practice, by either physicians or patients? How should its inclusion in therapeutics or decision-making be incorporated into informed consent? Does the use of AI alone require specific attention in an already-crowded informed consent process? A related but separate question is what information should AI be allowed access about the patient. In general, the more an algorithm knows about the patient, the more effective it is likely to be in achieving the outcome of interest. Patients might have a special sensitivity to AI access to particular kinds of data generated by the health care system or outside it (e.g., Google search results, geolocation, etc.). Some pacemakers and ICDs can connect via Bluetooth with an app on the patient’s smartphone or tablet.8 The use of apps to communicate with pacemakers or ICDs or on wearables, such as the Apple Watch, will likely mean the involvement of user agreements that most patients/users have difficulties to understand and usually do not read.9, 10 When can user agreements suffice as opposed to true informed consent? In addition, frequent updates will make it even more difficult for patients/users to keep track of what is being changed in the

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

software.11 The potential integration of AI into these data relationships makes confronting questions around consent even more urgent, as the inscrutability of these algorithms will make effective education for patients regarding risks and benefits particularly difficult.

DATA PRIVACY In general, usage of patients’ medical data demands assurance of appropriate privacy protection. In the U.S., the Health Insurance Portability and Accountability Act (HIPAA) is the main legal safeguard against unauthorized use and disclosure of health information12 (though in some instances the Federal Policy for the Protection of Human Subjects (Common Rule) will also apply). HIPAA restricts some uses and disclosures of individually identifiable health information generated by

Regulation (GDPR) is broader in its scope and applies to all personal data, including “data concerning health” such as non-health information that supports inference about health.12 Does the U.S. need to move to a privacy regime more like Europe’s? It is also worth noting that in the U.S., state privacy laws may also play an increasing role, especially the California Consumer Privacy Act of 2018 (CCPA) which will be operative from 1 January 2020 and will apply alongside HIPAA.

CYBERSECURITY The new technologies used in the AI space are also vulnerable to cyber-attacks. Consider, for example, the recall (in the form of a firmware update) of around 500,000 pacemakers in August 2017 by FDA due to fears that they could be hacked to alter the patient’s heartbeat

The new technologies used in the AI space are also vulnerable to cyber-attacks. “covered entities” such as insurance companies and health care providers and their “business associates.”13 Importantly, technological companies such as Google, Facebook, and Amazon are not HIPAA-covered entities and thus health data which is collected through wearables such as the Apple Watch are generally not protected under HIPAA.13 In contrast, the EU’s new General Data Protection

or to run the batteries down.14 FDA released the Medical Device Safety Action Plan in April 2018 that, among other things, aims to advance cybersecurity to promote safer innovative technologies.15 While these initiatives are laudable, it is also crucial to have an internationally enforceable cybersecurity framework in place, since cyber-attacks pay no heed to national frontiers.11 Machine learning

systems are particularly vulnerable to manipulation in health AI applications: one small alteration in how inputs are shown to a system can entirely change its output, thus, for example, classifying a mole as malignant with 100 percent confidence.16 In conclusion, AI has the potential to transform healthcare, including cardiology. However, these innovations will raise ethical and legal challenges such as bias and fairness, informed consent, data privacy and cybersecurity. Stakeholders, especially AI makers and health care providers, need to address such challenges at the earliest stage possible to contribute to successful implementation of AI in cardiology and thus build patient/consumer trust. -----------------------------------------Acknowledgements Sara Gerke’s and I. Glenn Cohen’s research is supported by a Novo Nordisk Foundation-grant for a Collaborative Research Programme (grant agreement number NNF17SA027784). Daniel B. Kramer’s research is supported by the Greenwall Faculty Scholars Program.

THE DEEP DIVE REFERENCES

Kallinen, L. M., Hauser, R. G., Tang, C., Melby, D. P., Almquist, A. K., Katsiyiannis, W. T. & Gornick, C. C. Lead integrity alert algorithm decreases inappropriate shocks in patients who have Sprint Fidelis pace-sense conductor fractures. Heart Rhythm 7, 1048–1055 (2010).

Weissmann, J. Amazon Created a Hiring Tool Using A.I. It Immediately Started Discriminating Against Women. https://slate.com/ business/2018/10/amazonartificial-intelligence-hiringdiscrimination-women.html (2018).

Klugman, C. M., Dunn, L. B., Schwartz, J. & Cohen, I. G. The Ethics of Smart Pills and SelfActing Devices: Autonomy, Truth-Telling, and Trust at the Dawn of Digital Medicine. AJOB 18, 38–47 (2018).

Price II, W. N. & Cohen, I. G. Privacy in the Age of Medical Big Data. Nature Medicine 25, 37–43 (2019).

02 Cowie, M. R., Sarkar, S., Koehler, J., Whellan, D. J., Crossley, G. H., Tang, W. H. W., Abraham, W. T., Sharma, V. & Santini, M. Development and validation of an integrated diagnostic algorithm derived from parameters monitored in implantable devices for identifying patients at risk for heart failure hospitalization in an ambulatory setting. European Heart Journal 34, 2472–2480 (2013). 03 Epstein, A. E., DiMarco, J. P., Ellenbogen, K. A., Estes, N. A. M., Freedman, R. A., Gettes, L. S., Gillinov, A. M., Gregoratos, G., Hammill, S. C., Hayes, D. L., Hlatky, M. A., Newby, L. K., Page, R. L., Schoenfeld, M. H., Silka, M. J., Stevenson, L. W., Sweeney, M. O., Tracy, C. M., Darbar, D., Dunbar, S. B., Ferguson, T. B., Karasik, P. E., Link, M. S., Marine, J. E., Shanker, A. J., Stevenson, W. G. & Varosy, P. D. 2012 ACCF/AHA/HRS focused update incorporated into the ACCF/ AHA/HRS 2008 guidelines for device-based therapy of cardiac rhythm abnormalities: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines and the Heart Rhythm Society. Journal of the American College of Cardiology 61, e6-75 (2013).

07 Turakhia, M. P., Desai, M., Hedlin, H., Rajmane, A., Talati, N., Ferris, T., Desai, S., Nag, D., Patel, M., Kowey, P., Rumsfeld, J. S., Russo, A. M., Hills, M. T., Granger, C. B., Mahaffey, K. W. & Perez, M. V. Rationale and design of a large-scale, app-based study to identify cardiac arrhythmias using a smartwatch: The Apple Heart Study. American Heart Journal 207, 66–75 (2019). 08 See e.g., Horwitz, J. Medtronic Debuts First Apps to Let Heart Patients Monitor Their Pacemakers. https://venturebeat. com/2019/01/16/medtronicdebuts-first-apps-to-letheart-patients-monitor-theirpacemakers/ (2019).

10 Cohen, G. & Pearlman, A. Smart pills can transmit data to your doctors, but what about privacy? https:// www.newscientist.com/ article/2180158-smart-pillscan-transmit-data-to-yourdoctors-but-what-aboutprivacy (2018). 11 Gerke, S., Minssen, T., Yu, H. & Cohen, I. G. A Smart Pill to Swallow: Ingestible Electronic Sensors, Cutting Edge Medicine, and Their Legal and Ethical Issues. Nature Electronics (submitted).

14 Hern, A. Hacking risk leads to recall of 500,000 pacemakers due to patient death fears. https://www.theguardian. com/technology/2017/ aug/31/hacking-risk-recallpacemakers-patient-deathfears-fda-firmware-update (2017). 15 FDA. Medical Device Safety Action Plan: Protecting Patients, Promoting Public Health. https:// www.fda.gov/downloads/ AboutFDA/CentersOffices/ edicalProductsandTobacco/ CDRH/CDRHReports/ UCM604690.pdf (2018). 16

12 Cohen, I. G. & Mello, Michelle M. HIPAA and Protecting Health Information in the 21st Century. JAMA 320, 231–232 (2018).

Finlayson, S. G., Bowers, J. D., Ito, J., Zittrain, J. L., Beam, A. L. & Kohane, I. S. Adversarial attacks on medical machine learning. Science 363, 1287–1289 (2019).

04 Nassif, M. E., Tang, Y., Cleland, J. G., Abraham, W. T., Linde, C., Gold, M. R., Young, J. B., Daubert, J. C., Sherfesee, L., Schaber, D., Tang, A. S. L., Jones, P. G., Arnold, S. V. & Spertus, J. A. Precision Medicine for Cardiac Resynchronization: Predicting Quality of Life Benefits for Individual Patients-An Analysis From 5 Clinical Trials. Circulation: Heart Failure 10, e004111 (2017). 05 Barsheshet, A., Moss, A. J., Huang, D. T., McNitt, S., Zareba, W. & Goldenberg, I. Applicability of a risk score for prediction of the long-term (8-year) benefit of the implantable cardioverterdefibrillator. Journal of the American College of Cardiology 59, 2075–2079 (2012).

AUTHORS

SARA GERKE, PETRIE-FLOM CENTER, PMAIL, HARVARD LAW SCHOOL The Petrie-Flom Center for Health Law Policy, Biotechnology, and Bioethics at Harvard Law School; The Project on Precision Medicine, Artificial Intelligence, and the Law (PMAIL); 23 Everett Street, Cambridge, MA 02138, USA. E-mail: sgerke@law.harvard.edu. DANIEL B. KRAMER, THE RICHARD A. AND SUSAN F. SMITH CENTER FOR OUTCOMES RESEARCH IN CARDIOLOGY, BETH ISRAEL DEACONESS MEDICAL CENTER, HARVARD MEDICAL SCHOOL 375 Longwood Avenue, 4th Floor, Boston, MA 02215. E-mail: dkramer@bidmc.harvard.edu. I. GLENN COHEN, HARVARD LAW SCHOOL Harvard Law School, Griswold 503, Cambridge, MA 02138, USA. E-mail: igcohen@law.harvard.edu.

AI MED I THE CARDIOLOGY AIMED I THE & INAUGURAL RADIOLOGY ISSUE

THE DEEP DIVE

BY: TIRATH Y. PATEL, MD

AI IN RADIOLOGY: ASKING SOME HARD QUESTIONS IT HAS BEEN OFTEN SAID THAT CHANGE IS INEVITABLE. THE FIELD OF RADIOLOGY IS A TEXTBOOK CASE OF CHANGE SINCE ITS INCEPTION OVER 100 YEARS.

rom the first diagnostic radiology report in 1896 describing abdominal findings, to the ubiquity of radiography in hospitals in the mid-20th century, to the discovery and use of CT and MRI and implementation of crosssectional imaging, and most recently the digitization of images and image transfer through PACS, radiology has always been a part of change in healthcare. Artificial intelligence (AI) in radiology will no doubt bring additional changes to the profession. AI will lead to significant changes in the workforce and training of radiologists, as well as in the overall radiology market, and will pose questions that need to be answered.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE

WORKFORCE Speaking with radiologists in the United States, there are essentially two camps. There are a fair number who, rightly or wrongly, believe AI will be an adjunct to radiologists and not necessarily a replacement. There is another camp, however, that is far more pessimistic. These radiologists have a ‘sky is falling’ mentality and believe far fewer radiologists will be needed in the future.

In the same vein, radiologists spend majority of their time interpreting radiology studies. If a significant fraction of this is automated, what will the radiologist do with the additional time? One avenue is of course direct consultation with patients. There is obviously the possibility of change, but right now, however, the payment structures in the United States are not aligned to easily support such an endeavor.

By looking at colleagues in pathology, we can gleam some insight into the role of automation in repetitive interpretive tasks in medicine. Many leaders in the field are in the former camp and believe AI will be an adjunct. By looking at colleagues in pathology, we can gleam some insight into the role of automation in repetitive interpretive tasks in medicine. For a long time, pathologists would manually interpret blood samples as part of a complete blood cell count, or a CBC. However, in the mid-20th century, much of this process began to be automated through the development and use of a Coulter counter. This machine would automatically perform CBCs, with only a small number (510%) requiring manual interpretation. Pathologists today do not routinely perform CBCs and have focused their work on histochemical and immunohistological interpretations.

There are secondary implications of this as well. An often-overlooked aspect of AI in radiology is the issue of residency positions. Are we training too many radiologists in the United States? Right now, approximately 1100 new radiologists are minted in the US annually. Should there be fewer residents? Some leaders in the camp with the belief that AI will be an adjunct to radiologists have said that “AI won’t replace radiologists, but radiologists who use AI will replace those who don’t.” Even in that setting, the implication is that there could be fewer radiologists. Regardless, radiologists will need to gain new skills and work patterns [1]. This, of course, requires changing radiology training, which, today, is predominantly spent on image interpretation.

The workforce situation is different in other parts of the world. In the United Kingdom, the Royal College of Radiologists (RCR), through their workforce surveys, has said that there is a shortage of clinical radiologists, and trending toward a greater shortage [2]. The situation is starker in other nations. At RSNA 2018, RSNA president Dr. Vijay Rao of Thomas Jefferson University, described a visit to the radiologist department in a large hospital in Johannesburg, South Africa. She described seeing boxes of plain chest radiographs scattered about. The radiologist working in South Africa mentioned how there were not enough radiologists to interpret chest x-rays [3]. AI could help immensely with the shortfall in radiologists in these and similar countries. What is a cause for concern in the United States may be a blessing elsewhere.

CONSOLIDATION Consolidation in health care sector in the United States has been steadily ongoing for nearly two decades. Hospitals are buying out or merging with nearby hospitals to create multi-hospital organizations. Radiology has not been immune to this consolidation. A recent article in the Journal of the American College of Radiology showed that the percentage of radiologists working in 100-plus person practices has increased over the past four years, while those practicing in practices with 99-or-less radiologists has decreased over the same time period [4]. It is through this change through which AI will be implemented.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE REFERENCES

Right now, many of the initiatives to implement AI in radiology are performed in large academic medical centers (AMCs), who themselves have been consolidating with nearby facilities. Because of access to larger amounts of data (among a host reasons), large AMCs and large private practices have a better framework with which to create, test, and implement AI in radiology relative to their smaller competitors. Will smaller practices partner up with larger entities as they see the larger entities better able to leverage AI for financial gain, thereby only increasing consolidation? The jury is still out.

THE PATIENTS These are only some of the questions that need to be asked, debated, and answers researched when discussing the effect of AI in the practice of radiology, particularly as it relates to the radiology workforce and practice types within the United States. Whatever the outcome for radiologists, I am optimistic that AI in radiology will be a blessing for patients: It flattens the hierarchy between patient and doctor, allows doctors to focus on more complex cases, and makes imaging more accessible to more individuals.

01 Davenport TH, Dreyer KJ. AI will change radiology, but it wonâ&#x20AC;&#x2122;t replace radiologists. Harvard Business Review 27 March 2018. Available at: https://hbr.org/2018/03/ai-willchange-radiology-but-it-wont-replaceradiologists. 02 The Royal College of Radiologists. Clinical radiology UK workforce census 2018 report. RCR. April 2019. Available at: https://www.rcr. ac.uk/system/files/publication/field_ publication_files/clinical-radiology-ukworkforce-census-report-2018.pdf. 03 Pearson D. Q&A: 20 minutes with 2018 RSNA president Vijay Rao, MD. Radiology Business 25 November 2018. Available at: https://www.radiologybusiness.com/sponsored/9667/ topics/leadership/qa-20-minutes2018-rsna-president-vijay-rao-md. 04 Rosenkrantz AB, Fleishon HB, Silva E, Bender CE, Duszak R. Radiology practice consolidation: Few but bigger groups over time. Journal of the American College of Radiology [article in press]. Available at: https://doi. org/10.1016/j.jacr.2019.02.030.

19 9 1 0 2 R E B M E C E D 4 1 11 -

RITZ CARLUTEOL LAGUNA NIG

AI MED NORTH AMERICA 2019 IS FOR THOSE INTERESTED IN THE BIGGEST PARADIGM SHIFT IN HEALTHCARE AND MEDICINE: ARTIFICIAL INTELLIGENCE Register now at aimed.events

KEY SPEAKERS This is the No.1 event of its type in the USA. Three-days dedicated to the transformative impact that AI-inspired technology is having on healthcare. A platform designed by clinicians to showcase latest thinking and facilitate new ideas and partnerships. Join us at the Ritz Carlton, Dana Point, California from 11 - 14 December and make your mark.

JOHN MATTISON

DANIEL KRAFT

JACK HIDARY

Chief Medical Information Officer at Kaiser Permanente

Founder and chair of Exponential Medicine

Entrepreneur

ANTHONY CHANG MD, MBA, MPH, MS

FRANCES A. AYALASOMAYAJULA

AIMed, Chair & Founder

Global Healthcare Solutions, HP

THE DEEP DIVE

BY: NINA KOTTLER

HOW CAN AI BE USED IN RADIOLOGY – AN OUTLINE FOR ALGORITHM DEVELOPMENT ARTIFICIAL INTELLIGENCE (AI) HAS THE ABILITY TO TRANSFORM OUR PROFESSION, NOT BY REPLACING US, BUT BY AUGMENTING US.

e recognize that healthcare needs to provide greater value. However, we cannot do this at the expense of efficiency. To be successful at this dual purpose, we need to utilize the power of information technology (IT) to augment our processes. Over the past several years, the need for IT to assist in driving value in healthcare has been recognized, and with that there has been a substantial capital investment into medical AI applications. In radiology, many of the applications were developed using the available data and an incomplete understanding of what radiologists do. With the idea that a radiologist is an image interpreter and with some large image sets globally available, akin to creating a CAD v2, most AI algorithms targeted image interpretation – specifically pathology detection. However, image detection is only a small component of what a radiologist does.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE

To fully utilize the power of AI in radiology, we must determine where it can provide the most value. To do this we must first break down radiology workflow into individual components. Curt Langlotz, a Professor of Radiology and Biomedical Informatics at Stanford terms the depiction of this breakdown “the lifecycle of an examination.” While our components and graphics are different, the idea of breaking down radiology into discrete elements is an essential first step in identifying areas where AI can drive improvement (Figure 1). From the moment a clinician wants to use imaging to help answer a clinical question through reporting, communicating, and follow-up, each step of this imaging lifecycle can be optimized. With this more global perspective, most imaging informaticists and leaders in radiology agree that applying AI to the segments outside of image interpretation would provide greatest value.

FIGURE 1: Lifecycle of an Examination – one representation of the breakdown of radiology workflow. Additional segments could be added to suit the local workflow. For instance, for groups that run the radiology department within their (client) sites, segments for optimization of patient/machine scheduling, check-in, transportation, patient communication, etc.

Let’s review the segments of the imaging lifecycle and discuss how IT is, and could be applied to each…

ORDER

PROTOCOL

A number of IT solutions are already available in this space as a result of the Protecting Access to Medicare Act of 2014. This legislation requires clinicians to consult a qualified clinical decision support mechanism when ordering CT, MRI, PET and NM exams. While these solutions are not typically machine learning AI algorithms, they do use IT to grade the utility of an ordered exam based upon appropriate use criteria and the reason for which the examination is being ordered.

As residents it was our responsibility to protocol each examination. We did this by reviewing the order, examining prior studies, gathering relevant information from the EMR, and in some cases, speaking directly with the ordering physician. In essence it was the residents’ job to collect and integrate all of the necessary information from multiple disparate sources to ensure a proper protocol was provided. While components of this exercise provided some education to the resident, the majority of this

time could be better spent on other learning activities. IT systems, if connected to the disparate sources of information, could use Natural Language Processing (NLP) to glean relevant information needed to create the optimal protocol and provide that information to the radiologist. In fact, these systems could use the information to automatically protocol the majority of studies and only forward the more complicated patients to a radiologist. In addition, image detection AI algorithms embedded on the imaging machine could create on the fly

protocols, adding additional series based on identified pathology. For instance, if an indeterminate adrenal lesion is detected by an AI algorithm on a CT abdomen/pelvis with IVC, the algorithm could create a 15 minute delayed series through the adrenal glands. A washout calculation could then be performed which would allow the lesion to be better assessed when it is initially reviewed. No longer would the patient need to return for a repeat study and deal with the uncertainty of a non-diagnostic diagnosis until the repeat exam is interpreted and communicated.

IMAGE ACQUISITION AND RECONSTRUCTION The risks of the radiation associated with CT scan are a necessary evil inextricably associated with acquiring the relevant information. A certain level of radiation is needed to produce a diagnostic signal to noise ratio (S/N). While there are some dose lowering techniques, none significantly reduces radiation dose while maintaining quality and the validity of the measurements we perform (e.g., HU). There are, however, evolving AI algorithms that are working toward translating a low dose, low S/N exam into a high S/N exam while maintaining the validity of HU and other measurements, and reportedly increasing the conspicuity of pathology. If successful, the results could be transformative. Imagine decreasing the radiation dose of CT studies close to that of an X-Ray. Similarly, algorithms are being created to decrease the image acquisition time for MR. Both types of AI algorithms

will add value for patients, while also improving scanner turn-around-time and decreasing power output so machine components last longer.

WORKFLOW Many algorithms can alter the workflow of critical exams as a biproduct of image detection AI. If exams are processed by an algorithm prior to that study appearing on the worklist, the position of the exam on the worklist could be optimized based on the AIpredicted imaging findings. Unlike image detection algorithms used primarily to assist the radiologist in identifying pathology, workflow algorithms can trade off specificity for sensitivity and still be successful. Some of these algorithms are already FDA approved and being used in practice. Other areas of workflow improvement, however, are less frequently seen. For instance, algorithms could elevate an exam on our worklist if that study is delaying patient discharge or if the patient has an upcoming physician appointment. It would also be helpful for cases to be directed to certain subspecialties based on identified findings. As an example, a head CT containing preidentified findings (e.g., mass, edema, postoperative findings) may be better interpreted by and directed to a neuro-radiologist, whereas the neuro-radiologist may not be required for a less complicated head CT.

RELEVANT PATIENT INFORMATION “Evaluate trauma” – although not ICD-10 compliant, this type of history is not infrequent. Even when a more complete characterization of presenting symptoms is available, some of the relevant data (e.g., cancer history, surgical history, lesions being tracked) is often absent. Similar to the work required to determine the proper protocol, gathering patient information relevant to the exam requires the radiologist to collect and integrate data from different sources. Unfortunately, in our current fast paced environment, this time-intensive exercise is usually bypassed, sometimes affecting the utility of the resulting interpretation. An IT system using NLP, however, could gather this disparate but essential information and provide it to the radiologist as a part of their workflow.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE

IMAGE INTERPRETATION (DETECTION AND DIAGNOSIS) Most of the available AI algorithms involve detection of pathology, a vital component of image interpretation. However, radiologists are skilled at image review and pathology identification. This skill is in part related to our many years of training, but also because the process is reliant on pattern recognition â&#x20AC;&#x201C; a capability that is highly evolved in

humans. However, there are other AI use cases in this imaging sector. For instance, AI tools could be used to segment the anatomy so the muscles, organs, and other structures could easily be defined. Based on this knowledge, once pathology is identified by the radiologist, an algorithm could volumetrically measure the size, evaluate for interval growth, identify the characteristics on each series/sequence, compare to a library of similar pathology, provide a differential diagnosis and pass this

information to the VR system to be placed directly into our report. Names of structures can be automatically reported by the software (e.g., segment 4 of the liver, gracilis muscle, clinoid ICA) and the PACS viewer toolset could be optimized based on the structure being evaluated. As more radiomic data becomes available, algorithms could utilize this data within pathologic structures to help predict treatment outcomes.

REPORT This segment of radiology is rich with potential AI use cases. While not the totality, the radiology report is the visible product of our work as radiologists. With that in mind, we should spend time ensuring it is optimized for best patient care and to best suit the needs of our referring clinicians (the consumers of our report). From a population health perspective, NLP tools can help ensure we are utilizing evidence-based medicine by providing the appropriate follow-up recommendation based on the reported findings and the patient metadata. A similar system could be used to ensure billing and MIPS data is completed during the reporting process thereby limiting follow-up radiologist requests for additional information. This system could also help the radiologist customize the report to the desires of the referring clinician by alerting the radiologist of clinician-specific needs based on exam type and identified pathology. A patient friendly report could also be generated from the physician report using AI as a mechanism to improve patient communication and awareness of the role of the radiologist.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE

COMMUNICATION OF FINDINGS A phone call is one mechanism to ensure important information is collected, but this type of communication is neither permanent nor necessarily performed when it is best suited for the referring clinician and radiologist. An automated system that records and provides relevant findings to the referring clinician

our referring clinicians disagree with our recommendations but rather because our hand off system is flawed. Instead of relying on a piece of paper or a phone call to ensure short and long-term followup is performed, NLP and reminder/ notification systems can be used. Such systems are being applied in a minority of locations. Utilizing this information in our reports, these programs identify necessary

Information technology, including AI, has the ability to transform our profession by augmenting the radiologist. In order to identify the areas of greatest need, it is helpful to break down radiology workflow into its individual components, the lifecycle of an exam. could replace our current outdated mechanism of communication. This system allows the referring clinician to acknowledge the result at a time that it is convenient for them, closing the communication loop and assisting with proper documentation.

FOLLOW-UP Surprisingly the majority of followup recommendations we make as radiologists are either not performed or not performed in the recommended time frame. This lack of continuity is not usually because

follow-up based on evidence-based medicine and additional radiologist generated recommendations, store this information in a database and produce reports for the client site to manage the required follow-up. As these systems continuously evaluate interpreted exams, they can acknowledge when the proper followup has been performed. In addition, user interfaces can be developed for direct tracking and management of these patients, generating automatic notifications and updates to stakeholders (e.g., nurse navigators, patients, or primary care physicians), as necessary.

PEER LEARNING Radiology peer review is starting a long needed and welcomed transition to peer learning. Reviewing potentially missed findings is one of the most effective ways to learn. Frequently this opportunity is lost because of the stigma associated with making a mistake and the limited radiology capacity to perform these reviews. AI programs, however, have capacity and could be used for educational purposes as a second read. Additional programs could be created to identify areas for targeted learning based on the categorization of a radiologistâ&#x20AC;&#x2122;s potential variances. In addition, AI systems could create case-based teaching materials by collecting and cataloguing exams with different findings. If given access to a pathology database, a definitive diagnosis could be applied to these educational cases making the resource and even more valuable feedback loop for learning. Information technology, including AI, has the ability to transform our profession by augmenting the radiologist. In order to identify the areas of greatest need, it is helpful to break down radiology workflow into its individual components, the lifecycle of an exam. Once identified, these components can be used to direct the growth of AI toward those use cases that are the most impactful. Surprisingly to some, many of the most impactful use cases lie outside of image interpretation. Hopefully we will begin to see an expansion of noninterpretive AI algorithms in radiology and with that produce a greater impact on driving value in radiology and healthcare.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE

BY: DR WILLIAM BRADLOW

MAKING DATA WORK FOR HEARTS AT RISK HYPERTROPHIC CARDIOMYOPATHY, DEFINED BY UNEXPLAINED HEART MUSCLE THICKENING, IS THE MOST COMMON FORM OF INHERITED HEART DISEASE, AND HAS A NOTORIOUS REPUTATION DUE TO THE FACT IT CAUSES SUDDEN DEATH IN YOUNG, PREVIOUSLY WELL INDIVIDUALS. THE CONDITION CAN BE COMPLICATED BY HEART FAILURE, ATRIAL FIBRILLATION AND STROKE. HOWEVER, IN CONTEMPORARY PRACTICE MOST PATIENTS WILL LIVE LONG, HEALTHY LIVES.

anagement involves estimating the risk of complications, intervening when indicated (for example by implanting a defibrillator to prevent sudden death) and screening first degree family members. For a service to do this effectively, the identity, disease severity and genotype of every diagnosed individual under its care needs to be known. If you were to ask for this information in most UK hospital outpatient departments, it would be difficult to readily obtain. This is because routine hospital statistics focus on inpatient episodes. So despite being in the setting best suited to deliver precision medicine - and to prevent admissions â&#x20AC;&#x201C; capabilities to do so may be limited by lack of information.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE DEEP DIVE

At University Hospital Birmingham NHS Foundation Trust, we have created a disease management platform to try to address this. It allows relevant data for every diagnosed patient in a hospital to be brought together and areas of need highlighted. Its key components are a clinico-genomic registry, dashboard and an automated means of extracting data from unstructured clinical text. This work involved the creation of a new interdisciplinary team of clinical, IT, clinical informatics and

genetics department. This mainstream approach has attracted 2 years of funding for a specialist nurse from the BHF Miles Frost Fund. Patients can view this information securely via a web portal. The dashboard gives clinicians access to the data needed to change care in a single, easily absorbed picture. For the first time, the population can be visualized and segmented in real time. This has allowed care to be targeted at those with increased risk of sudden death, heart failure and stroke (6%, 20%

We envisage a very different service emerging over the next 5 years; one in which low risk patients are followed remotely, with focus placed on new patients and those at increased risk. bioinformatic specialists. As many members were doing this alongside their usual commitments, in its early stages the project relied on their enthusiasm alongside support from senior staff and stakeholders. More recently, it has benefitted from being incorporated into the Midlands Health Data Research UK programme. We have piloted this work in our Heart Muscle Disease Service. Detailed data has now been collected during routine care for almost 1000 patients over 3 years. This includes results for genetic tests, over half of which were performed by a cardiologist, without requiring referral to the

and 23% of the cohort respectively), and led to the detection of 1 in 10 patients who had not been gene tested. As we develop a clearer picture of our entire population, we are in a much better position to plan services. The data has also been exploited to drive recruitment for the 100,000 Genomes Project â&#x20AC;&#x201C; the aim of which is to bring the predicted benefits of genomics to the NHS - and the National Institutes of Health funded HCMR study. Whilst demonstrating the value of this approach, we are also simplifying data collection. The amount of work

involved is significant; data has been extracted from the outset by hand from seven different sources within the hospital IT infrastructure, and several externally. We have recently automated integration of blood results, medications and clinic codes into the dashboard and are preparing to do the same for as many other elements in the data set as is possible. To identify and characterise diagnosed patients from clinical documents, academic bioinformaticians from the University of Birmingham, working alongside a clinician, have created and validated a natural language processing algorithm. This algorithm has identified a sizeable group of patients across the electronic patient record who were not present in either the registry or hospital coding records. We are planning to extend the platform through a clinical network in the West Midlands. It will then become increasingly useful for clinicians screening family members as they will know in advance whether to offer cardiac or genetic tests, and in the case of genetic testing, precisely what genetic test to offer. This will be particularly useful in the West Midlands as the region is home to one of the largest non-transient populations in Europe. The registry will also reach a suitable size for research, a benefit highlighted in the UKâ&#x20AC;&#x2122;s Life Sciences Strategy. As improved management results in a growing population of patients, the standard practice of seeing patients in person every year will become untenable. To address this, we are taking advantage of new technologies by integrating smartphone generated data and video clinics

into the platform, as well as risk prediction algorithms which will be developed from the curated data we are collecting. We envisage a very different service emerging over the next 5 years; one in which low risk patients are followed remotely, with focus placed on new patients and those at increased risk. Whilst risk prediction will never be perfect, improving current practice will be a step forward for a condition which remains defined by its potentially fatal nature.

DR WILLIAM BRADLOW is a cardiology consultant in cardiac imaging and heart muscle disease at the Queen Elizabeth Hospital Birmingham. He undertook echocardiography in Auckland and cardiovascular magnetic resonance at the Royal Brompton Hospital, London before training in Oxford. He manages a large cohort of patients with hypertrophic cardiomyopathy and has a specific interest in how data science and technology can be used to better manage chronic disease. @wbradlow1

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

STATS

Researchers from North Carolina State University had developed a technique which reduces training time for deep learning networks by more than 60% without compromising on accuracy. Instead of treating all data chunks as novel, the new method makes a deep learning network translates results from one computational filter to other similar data. This believe will not only saves computing power but also accelerate AI development.

MEDICAL DEVICE COMPANY, EKO, HAD DEVELOPED THE WORLDâ&#x20AC;&#x2122;S FIRST COMBINED ECG AND DIGITAL STETHOSCOPE, WHICH IS NOW AVAILABLE IN THE MARKET FOR PURCHASE.

THE DUO COMBINATION IS ABLE TO CONNECT TO SMARTPHONE OR TABLE FOR PHYSICIAN TO RECORD HEART SOUND AND ECG ALL AT ONCE. THE CAPTURED DATA CAN ALSO BE CONVENIENTLY SHARED WITH OTHER PHYSICIANS FOR A SECOND OPINION.

1.5M

ELECTROCARDIOGRAMS (ECGS) WERE USED TO TRAIN A DEEP LEARNING MODEL WHICH NOW PROVED TO BE RELIABLE IN DETECTING HYPERKALEMIA OR HIGH LEVEL OF POTASSIUM IN BLOOD IN PATIENT WITH CHRONIC KIDNEY DISEASE. THE DATA, DATED BETWEEN 1994 TO 2017, WERE OBTAINED FROM NEARLY 450,000 PATIENTS VISITED THE MAYO CLINIC IN MINNESOTA.

According to a recent study published in Radiology. AI system is able to distinguish abnormalities with high accuracy and speed up radiologists’ opinion from an average of 11.2 days to about 2.7 days.

Cardio Ex, is the 4th specialty game created by the medical video games company Level Ex. The game, available on App Store for free, aims to present interventional cardiologists with medical scenarios that challenge their reasoning and decision-making skills. The risk-free and engaging nature of the exercise believes to provide physicians, an alternative form of training.

BEIJING LAUNCHES A $4.4 BILLION PLAN BY 2022 TO BUILD 5G NETWORK WHICH ALLOWS THE CITY TO TAKE LEAD IN COMMERCIALIZING TECHNOLOGY AND EXPLORE PIONEERING APPLICATIONS, WHICH INCLUDE REMOTE HEALTHCARE SERVICES, SMART SENSORS, BIG DATA AND MEDICAL EQUIPMENT.

MORE MEDICAL SCHOOLS HAD AGREED TO JOIN THE ACCELERATING CHANGE IN MEDICAL EDUCATION CONSORTIUM, AN EFFORT BY THE AMERICAN MEDICAL ASSOCIATION’S (AMA) TO MODERNIZE MEDICAL CURRICULUM IN PREPARING PHYSICIANS FOR A TECHNOLOGY-DRIVEN HEALTHCARE ENVIRONMENT.

AIMED I THE CARDIOLOGY & RADIOLOGY ISSUE

02 AI CONFERENCE UPDATES ITS PAPER SUBMISSION POLICY One of the World’s largest annual AI research conferences, the Neural Information Processing Systems Conference (NeurIPS) has updated its paper submission policy which requires a reproducibility checklist. The move believes to echo with the recent “reproducibility crisis” in science, where research findings often failed to be replicated by other researchers. This has injected uncertainties on validation of early discoveries.

01 FDA ANNOUNCED NEW REGULATORY FRAMEWORK FOR AI-DRIVEN MEDICAL DEVICES US Food and Drug Administration’s (FDA) Commissioner Scott Gottlieb, announced steps to be taken towards creating a new regulatory framework for medical devices driven by artificial intelligence (AI) algorithms. In a statement released on 2 April, Gottlieb said FDA is aware of the impact AI has on healthcare and they are making new attempts to keep up to the nature of these promising technologies.

To adhere to the change, researchers will have to detail out their algorithms, data collection and keeping processes, and the simulation environment which they had used for training. NeurIPS assures the change is to enforce better practice of transparency for researchers to lay out the processes that lead them to their individual conclusions.

As of last year, the regulatory body had already authorized two AI based devices for detecting diabetic retinopathy and alerting patients of a potential onset of stroke. Next, FDA is publishing a discussion paper to gather feedback. A draft guidance will be issued on a later date. Eventually, the agency hopes to monitor software products from its premarket research stage to post-market performance, to ascertain safety and effectiveness.

03 A NEW SOFTWARE PLATFORM FOR RADIOLOGISTS INTERESTED IN AI The American College of Radiology’s Data Science Institute has set up a new software platform - the ACR AI-LAB, which aims to empower radiologists to be more involved in the “creation, validation and use of healthcare AI”. Announced on 5 April, the initiative provides support and tools for radiologists to develop AI algorithms within their own institutional facilities. Clinicians are free to employ their collected data to generate solutions which meet their clinical needs, and connect to the ACR member network and community for collaborations. ACR believes the free, open and vendor-natured of the platform facilitates democratization of AI while not succumbing to possible data risks as radiologists are performing their respective research behind institutional firewalls.

TECH MOVERS & SHAKERS

06 A MALWARE WHICH SET OUT TO DECEIVE RADIOLOGISTS

A malware developed by researchers from the Ben-Gurion University Cyber Security Research Center in Israel tricks politicians or presidential candidates into believing that they are suffering from adverse medical condition which make them to withdraw from a race to seek treatment. The result of a recent blind study supported the claim.

04 ALGORITHM IDENTIFIES NO SHOW PATIENTS

05 LIGHTSTORE, THE DATA STORAGE FUTURE

University College Hospital in London has created an algorithm developed from more than 20,000 appointments for MRI scans, to identify patients who are likely to miss their medical appointments. This project is part of UK’s National Health Service’s (NHS) initiative to introduce machine learning into the healthcare system.

Researchers from Massachusetts Institute of Technology (MIT) had designed a new flash-storage system which could cut down energy and physical space, the two most expensive components making up of a data center, for data storage. The new system, LightStore, modifies traditional storage servers which make use of solid-state drives (SSDs) or electronically programmable memory microchips.

Partnering with the Turing Institute in London, the hospital hope that in near future, AI is able to take over mundane tasks to relieve the burdens of healthcare professionals and administrative staff as well as to prioritize resources for patients at risks. At the moment, the algorithm is far from perfect as it falsely highlighted about half of patients attending appointments but bear a higher tendency of not showing up.

70 CT lung scans were tainted by the malware and they managed to deceive three skilled radiologists into misdiagnosing conditions almost every time. In the case involving fictitious cancerous nodules, the radiologists made diagnostic errors 99% of the time. Even when told of the presence of the malware, radiologists continued to make mistakes. Although the study focused on lung cancer only, other facets are equally susceptible.

Instead, researchers now connect SSDs directly to the data center’s network and do away with the need for other components. Together with the accompanying software “adapter”, which ensure data access, storage operations become more efficient and it also lowers energy consumptions.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE TECH PERSPECTIVE

TO BUILD, SHAPE AND SUSTAIN:

THE GOALS OF INTEGRATING AI IN HEALTHCARE I

n the latest AIMed webinar, that took place on March 26, 2019, Molly K. McCarthy, National Director, US Provider Industry and Chief Nursing Officer at Microsoft and John Frownfelter, Chief Medical Information Officer at Jvion, spoke about how to shape and set artificial intelligence (AI) goals within healthcare. The hour-long session was facilitated by Dr. Anthony Chang, AIMed Founder and Chairman, Chief Intelligence and Innovation Officer of Children’s Hospital of Orange County (CHOC). AI presents a tremendous opportunity to impact patient lives and improve outcomes. But, as with any new technology, there is risk involved. Hospitals may direct their AI resources to the wrong areas. They may misalign AI investments to the needs of the organization. They may invest in solutions that aren’t proven or don’t work. Frownfelter and McCarthy shared how to avoid these pitfalls, what to expect from AI within healthcare over the next five years, how to define and build a successful AI strategy that aligns to hospital goals, and how to operationalize an AI strategy effectively.

KEY COMPONENTS THAT ENABLE AI’S SUCCESS Frownfelter explained that for an organization to build a successful AI strategy, leadership should start by understanding where the organization is in its analytics maturity. They must look at whether there is executive sponsorship for an AI-related initiative and understand who is going to drive implementation and adoption at the highest level.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

AI is both disruptive and threatening especially to physicians and nurses, who were taught and trained to reject anything they cannot see, touch, feel and understand. AI doesn’t easily fit into the current technology adoption framework. Also, it is crucial to look at existing resources within the organization. Things like analytics architecture and the accessibility of data are critical to AI regardless of the approach to building or buying a solution. Newer approaches that used cloud-based solutions can leapfrog the traditional limitations seen with analytics and should be kept in mind. That isn’t enough however – implementing AI begins with a problem that needs to be solved, rather than having a “solution that is looking for a problem.” Frownfelter stated that an organization might be in a disadvantaged position if they come forward and say, “we want to invest in AI, but we are not sure where.” McCarthy agreed. She encouraged conversations between departments as they decide on an AI strategy. This dialogue should include competency building to ensure the right technical aptitude of individuals who are supporting and driving the success of the project within the organization. McCarthy also emphasized the need to change the way organizations frame an AI project. Thinking about AI and driving the success of these programs requires a different approach to management and messaging.

McCarthy added that healthcare organizations should look outside of their four walls for best practices. “Not just (within the) organization… but potentially looking at your consumer and patient base, as well as… folks outside of healthcare, in retail or other industries where AI has been implemented, adopted and successful,” she said. The speakers also touched on the cultural requirement to assimilate AI into healthcare. As Frownfelter highlighted, AI is both disruptive and threatening especially to physicians and nurses, who were taught and trained to reject anything they cannot see, touch, feel and understand. AI doesn’t easily fit into the current technology adoption framework. The validation approach to AI outputs requires a different perspective. Leadership has to understand and own what is needed and determine how to communicate success in an agile and meaningful way.

AI ISN’T A SILVER BULLET, IT IS A TOOL Both speakers warned of the danger in underestimating the complexity of healthcare and building AI solutions

that target it. Often, when organizations are trapped in the “build it or buy it” dilemma, they would choose the former, without realizing the work effort required to establish an AI solution. The risk of misjudging work effort is rather prominent for organizations with robust teams of data scientists. They are brought into the false perception that they can create anything. “Most of the time, they can’t. Then there is a gap between what’s been promised and what’s been delivered, and the clinical community gets frustrated. So, creating some boundaries and focus for who does what is very important,” Frownfelter said. A useful exercise to illustrate this, would be to do a cost and time analysis for an internal team to develop a working predictive model, and then further, to operationalize and adopt that model. Likewise, it is too simplistic to think that AI is going to do everything or that it has the potential in the near-term to replace clinicians. As Frownfelter suggested, we should think of AI as a “lab test,” something that physicians and nurses use to interpret what they are seeing and make more informed decisions that help to focus resources, clinical actions, and communications to ensure the right steps are taken on the right patients at the right time. Ultimately, if an organization is willing to invest in their own AI solutions, they will have to be mindful of their target and the timeline to address it. The building process may be extended when AI is built internally. It may take years to develop something novel—a

THE TECH PERSPECTIVE

technology to look into a challenging topic: discharge optimization. The goal is to reduce avoidable extended inpatient stays that are the result of preventable complications, adverse events, and avoidable delays in discharge. The Jvion Machine’s outputs integrate into the provider’s Discharge Planning Integrated Model. The provider transmits real-time patient data to Jvion. This data is analyzed, and individual patient risk levels and interventions are delivered every 24-hours through the system’s Electronic Health Record (EHR). Patient risk and intervention information is integrated into existing physician and staff workflow.

process that repeats during testing and implementation. Frownfelter noted that most of the time, organizations embark on building something on their own until they realize the associated effort and the risk. There is a place for data science teams within hospitals, he noted. It comes down to strategically incorporating existing solutions with internal capabilities to make the most out of AI investments.

A USE CASE EXAMPLE To illustrate the successful operationalization of an AI strategy, Frownfelter highlighted a large, integrated delivery network based in the Pacific Northwest. The organization is using Jvion’s

Frownfelter explained that the Machine’s interventions supplement existing efforts to identify and communicate target discharge dates and support the timely discharge of hospital patients across care management, RNs, and hospitalists. As a result, the provider is ensuring the appropriate care progression, optimizing discharge activities, and providing timely and appropriate communication to families and patients.

WHAT CAN’T AI DO? Dr. Chang concluded the session with a thought-provoking question. What do you think is not going to be good for AI? McCarthy focused on the human touch. “I think ultimately, at the end of the day, healthcare is about people and truly technology has its place, but you don’t want to lose that human connection,” she explained. Frownfelter echoed a similar sentiment. He expressed that if he were to be a patient one day or in the near future, he would not look to a machine for empathy. Frownfelter explained that he wants a physician or a nurse at his bedside because “empathy will never be replaced by AI.”

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE TECH PERSPECTIVE

GENOMICS + RADIOMICS = A STEP TOWARDS BETTER HEALTHCARE I

f you have physically or digitally (via live-streaming) attended the AIMed Breakfast Briefing: Experience the future of AI in Radiology, held respectively in Chicago, Boston and Toronto, between 9 and 11 April, you probably saw our sponsor: SOPHiA GENETICS.

The tech company is tapping into artificial intelligence (AI) to “Democratize Data-Driven Medicine”. Its’ AI technology, SOPHiA, analyzes genomic profiles to assist clinicians in giving improved diagnosese and treatment plans. A year ago SOPHiA gained radiomics capabilities. The added feature is further empowering medicine, particularly in the areas of oncology. AIMed spoke with SOPHiA GENETICS’ Senior Vice-President Radiomics and renowned mathematician, Thierry Colin, to find out more about SOPHiA Radiomics as well as their views towards AI in medicine.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

AI MED: For the benefits of readers who do not know, what is the significance of combining genomics and radiomics? THIERRY COLIN: In order to determine the best possible cancer treatment or care plan for an individual patient, the tumor board needs to gather all available information (clinical, radiology, biology, pathology) it can get. Oncologists then use their expertise and this information to assess diagnosis and prognosis, and to monitor the progression of the disease as well as the efficiency of treatment. SOPHiA Radiomics already simplifies the tumor board’s task by centralizing all useful information in a unique place. But the scope of this application doesn’t stop there; by extracting valuable data from the image and truly combining it with the genomic profile of the tumor, SOPHiA Radiomics provides a new kind of quantitative information for a better outcome. It means that Physicians are now armed with even more information to decide on the kind of treatment that best fits the patient.

AI MED: Radiomics studies tend to extract dozens or even hundreds of image features. Machine Learning studies also require large datasets which may not always be available. How do you address this challenge and deliver statistically relevant and accurate Radiomics results? TC: First of all, yes; Extracting hundreds of features might sound very tempting but could indeed end up being counterproductive. This is why we developed SOPHiA Radiomics in close collaboration with healthcare professionals, as a way to identify the most critical sets of features and extract only the most valuable ones. Even though we can still adapt features extraction to the customer’s needs. Secondly, SOPHiA Radiomics aims to work first with very specific diseases in order to minimize variability. We may not underline each and every tumor for the moment, but we are working on certain specific sub-types and dealing with actual clinical data. This is the same kind of bottom up approach

that we applied from inception, and that allowed us to develop highly performant genomic applications that now cover a full range of disorders. Finally, the vision of SOPHiA GENETICS, is to create a community of users enabling improved patient care worldwide. Our genomic community already groups over 930 healthcare institutions in 80 countries. In the radiomics field, connecting professionals from around the world through SOPHiA’s community will help overcome this lack of available local cases. AI MED: You mentioned something about “user-community”. So, while setting up a “user-community” may be the key to defining accurate Radiomics-based predictive models, how do you cope with variability in terms of imaging protocols? TC: Technically, there is a need for reproductible segmentation and features extraction tools. The Image Biomarkers Standardization Initiative (IBSI) is grouping major healthcare institutions all over the world in order to implement a set of recommendations for automation and standardization of features extraction. We are working so that, everything that we compute from the image will comply with this set of guidelines. AI MED: There has been a lot of talk about AI diversity these days. SOPHiA AI platform is being employed in many hospitals in different continents, but different countries have different resources and AI adoption rate, how does SOPHiA address those cultural differences?

THE TECH PERSPECTIVE

TC: The vision of SOPHiA GENETICS is the democratization of Data-Driven Medicine worldwide. We are building advanced technologies to help healthcare institutions provide faster and more accurate diagnosis and treatment to patients. We’re also growing a community that promotes the sharing and accessibility of valuable knowledge. These actions pave the way to a more performant, more sustainable, hence more accessible healthcare system. The global adoption of our technology, trusted by over 930 hospitals to date, shows that Institutions all over the world understand the benefits they can get from our innovative approach and SOPHiA’s top analytical performance. AI MED: Perhaps to some people, “democratizing data-driven medicine” is still a novel concept, do you mind explaining more? TC: Of course. As explained before, we bring tech recipes to the healthcare industry to better analyze genomic and radiomic data; feeding healthcare professionals with knowledge and valuable information they didn’t have before, to improve medicine. That’s the Data-Driven Medicine part. Beside this technology aspect, the three other pillars of our action are Community, Universality and Support. The community makes this knowledge available to all our users. The universality of our platform, that adapts to the needs and means of the users, allows anyone around the world to access our technology. Finally, our local teams of experts are there to understand the needs of institutions and support them in the implementation of Data-Driven Medicine’s applications. This is what “democratizing” means, this is our mission.

AI MED: That’s brilliant! But how does SOPHiA uphold democratization of Data-Driven Medicine and respect patients’ privacy at the same time? TC: Obviously as the champion of Data-Driven Medicine, it is our duty to secure data privacy and security. We are ISO27001 certified for Information Security Management and audited regularly. SOPHiA GENETICS thus respects international and national privacy regulations, meaning that we only act as a processor of anonymized and encrypted data. In short, the healthcare institutions that we work with have full control over the data that remains in the ownership of individuals, and we don’t have access to patient’s personal data under any circumstances. AI MED: SOPHiA received a $77 million Series E funding at the turn of the year. How will you use these resources and what will be the focus for SOPHiA GENETICS in the future? TC: We’re expanding our presence dramatically in the US in order to face the growing demand in the region. In parallel, we continue to develop SOPHiA by adding analytical capabilities and combining different sources of medical data. SOPHiA Radiomics is the first example of that. We’ll also continue to support and grow our large network all over the world. This is why we are talking about the democratization of Data-Driven Medicine worldwide. We encourage every institution to adopt such tech applications, not only for the direct benefit for their patients but also because it allows for a more sustainable global healthcare system where the information used to help patients today will help those of tomorrow.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

SPOTLIGHT

CORY LINTON FOUNDER & CEO Established by

2018 Founded

SALT LAKE CITY, UTAH Headquarters

THE MAN BEHIND NO-SURVEY SURVEY I O

used to work in a cognitive science laboratory. Our team collaborated with researchers from the University of Washington and Université de Neuchâtel to look at the cross cultural differences in attitudes towards mathematics and group learning. The projects went well and funding was sufficient. For years, we were pretty much independent and happy. Until one day, the Dean of college stormed in. Our junior staff almost barred him from entering because we never had visitors. All the more, the Dean looked at least a decade dissimilar from the profile picture we were shown on the first day of work. The laboratory was told it had been given the worst feedback ever across the whole campus for three years consecutively. So the Dean was eager to find out why. We were appalled by the comment as we were never dissatisfied, let alone hearing anyone express this view out loud. After a series of clarifications, it turned out that something had gone wrong with the survey responsible for capturing our feedback. Cory Linton, chief executive officer of Edify.ai, was not at all surprised when I recalled my story during our interview. In fact, Linton said most healthcare systems will have at least one annual survey per year. He agreed that whilst these surveys tend to yield good data, they are ineffectiveat applying it in context. At a glance, a person may seem to be doing well in his/ her job at a particular time. However, if one managed to

track his/her performance over a period, the story may tell otherwise. So what Edify.ai does is to leverage the massive amount of real-time data that exists in the system to help employees to be more engaged and effective; creating what Linton called, a “no-survey survey”. “Think about email in an organization,” Linton explained “the average employee sends about 50 messages a day. A large healthcare system we are working with sends 300 million emails per year, so that’s a lot of data. With that, we analyze the tone of communications; whether it’s positive or negative and when does it begin to change.” Linton believes unlike survey, it’s harder for these data sets to lie. He said when people decide which email they respond to immediately and which they do not, they are making a judgement. Structured questions, on the other hand, may not be sensitive enough to pick up this information. Previously, Linton and his eminent technical team had successfully generated algorithms based on enterprise data to help companies like ShipEx to better engage truck drivers, and JT Thorpe, US’s largest and oldest refractory construction company, to improve safety and awareness. Now, Edify.ai is setting their eyes on healthcare. They are employing artificial intelligence (AI) to predict when healthcare professionals are at risk of burnout. According to a report put forward by the Massachusetts medical society, Massachusetts health and hospital association, Harvard T.H. Chan school of public health

and Harvard global health institute, nearly half of all physicians in US experience burnout in some form. As high as 78% of physicians expressed they are feeling the pinch. By 2025, an estimated 90,000 healthcare professionals will be lost due to general shortage and continuous work pressure. On average, the cost of recruiting and replacing a physician can range between $500,000 to $1 million. “Sadly, what’s going on right now is that we will never be able to find out if a doctor has the desire to quit until they are ready to do so” Linton remarked. “Our system typically learns whether a doctor is showing burnout in more or less than a week’s time. But they will not leave immediately, they may go on for job searching in the next few months”. So, the sooner a burnout is detected, doctors can be moved into a less stressful position. Apart from emails and messages, Edify.ai is also making use of other actionable data like electronic health records (EHRs), to gain an insight into healthcare professionals’ workflow. Ultimately, the company hopes that the eventual application, will not only reflect burnout rates,

but also whether healthcare professionals are spending too much time interacting with EHRs, whether diversity and inclusion were considered in the workplace, and the overall dynamics within the team. “Instead of actual percentages, users will see shades of green, yellow and red. Green represents the statistical norm; yellow means outside of it, and red means statistically different” Linton said. As real-time data keeps streaming in and results change every day, even if it says 10% abnormal it may not be statistically significant, so color codes will prevent people from reacting when they do not need to. In the coming year, Edify.ai is working with several companies in the US to strengthen its neural network. They are also debating on the level of transparency the end-product shares. Linton foresees their pioneering clientage to be cutting edge. At the same time, he also wishes to have more likeminded on board. Oh yes, my old Dean of college has been informed.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE PATIENT PERSPECTIVE

SEAN HAMILTON

THE WARRIOR AND HIS INVISIBLE BATTLES Based on statistics published by the World Health Organization this February, more than 50 million people around the World have epilepsy. Sean Hamilton is one of them. His added heart condition often makes him confused as to whether he is having a seizure or a cardiologic-related episode. Despite this, he has taken a step forward to start War on Epilepsy, pledging to educate the public more about his invisible battle. AI MED: What prompted you to start War on Epilepsy? SEAN HAMILTON: War on Epilepsy was born unofficially about nine years ago when my epilepsy condition first took off while I was on my way to completing a business course. I had two epileptic seizures within a couple of weeks and then things started to snowball a little. Nevertheless, if you would really like to trace the cause, it probably goes back to when I was much younger. I had many different medical conditions when I was a child. They kind of planted seeds

inside me to embark on different missions, one of them is to be a paramedic. Sadly, I was not at all academic back in my school days, so I would not be able to become a paramedic. My medical conditions will also make things difficult. The turning point came when my school recommended me for a public service course through the London Fire Brigade. After the training, I became one of the first under 18 volunteers, and I was able to provide my service at any operational fire stations. From there, I took up different responsibilities and also helped out at other volunteering organizations.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE PATIENT PERSPECTIVE

It was an amazing experience but once I reached 18, I couldn’t help but begin to think again, what is going to be my next challenge? While I was assessing myself, I realized at the end of the day, all I want is to help. Since my best subject in school was information technology (IT), I thought how wonderful it would be if I could make a connection between the two. AI MED: In general, do you think people are still unsure of what epilepsy or seizure is?

As such, I believe some form of education is important. At the very least, people should be more comfortable with how to deal with a person when a seizure takes place. For example, I had a seizure once on the train and I was shocked to see transport police surrounding me when I came around. It was only much later that I was told I had been hostile and become aggressive when I had the seizure, and that’s why passengers pulled me off and threw me onto the platform.

SEAN: Yes. We have got 10,000 people with epilepsy living in London but we only have one specialist hospital catered to them. Some patients may

I have also heard stories such as someone having their fingers literally bitten off when trying to put something into the patient’s

People should be more comfortable with how to deal with a person when a seizure takes place. have to travel 500-600 miles just to visit a specialist. It is frustrating, especially considering we are in a digital age. Besides, the UK’s National Health Service (NHS) long term plan has highlighted so many medical conditions but not one of them was neurological related, and epilepsy was not even in the list. Often it occurs to me that no one is doing anything and no one is setting a plan for us.

mouth when a seizure broke out. It is our natural instinct to protect ourselves, even when we have a seizure and we are unsure of what is going on. So, never ever try to put a person in the recovery position when he or she is having a seizure. Because that physical contact may trigger a muscle response and those who are giving help may get hurt as a result.

AI MED: Is that why you name your organization “War on Epilepsy”? Because it is an invisible battle that all epilepsy patients are fighting for? SEAN: Not exactly. The name was inspired by a friend who was diagnosed with cancer. He named his organization “War on Cancer” and he won the battle. AI MED: You appear to be extremely positive in spite of your medical conditions and challenges. Have you ever had any doubts about yourself or moments when you feel like “no, I don’t think I can do this”? SEAN: I try to say no. Usually when my dad is around, he will throw me some visual cues but let’s be honest, I would

50M people around the world have epilepsy

10k

people in London living with epilepsy

say yes, about six years ago, I kind of went through a period of depression. In those days, I pretty much confined myself at home, I got to the stage where I could not even leave the house anymore. Because if I ever do that, due to my cardiological condition and epilepsy, I would probably end up in the back of an ambulance. I lost my grandad through Alzheimer’s around that time too and my uncle died of a heart attack right outside his front door. Things like these took a toll on me, they impacted upon me, and motivated meto want to do more. That’s when I started to pick up, but again my medical conditions are creating a limit on what I can actually offer. I learnt that I cannot always go through the traditional route and have to be creative about it. I believe if a problem froze in your way, there is always going to be a solution to it. If you cannot go under it, then find a way around it, find a way through it, over it or try to knock it out. Yes, it may not be easy but life is not supposed to be easy. The best things in life are always going to be difficult. AI MED: From the view of a cardiology patient with epileptic seizures, how do you think you are benefitting from technology? SEAN: One of my many medical conditions is POT’s or Postural Orthostatic Tachycardia Syndrome. The simplest way to explain it, is in a healthy person, the blood flows freely throughout the body. For someone with POT’s, the blood will get stuck to the legs when he or she sit or stand over a long period of time. As such, the blood will not flow back to the heart to resuscitate it.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE PATIENT PERSPECTIVE

When this occurs, the body’s natural instinct is to drop itself to a levelling position or passing out. I have had an implant inside my chest to monitor my heart condition for about four years. It was removed last year. It is through this piece of technology that I began to learn how to interpret electrocardiogram (ECG). From there, I can better monitor my routine and balance my life between my cardiac condition and my seizures. I began to use wearable after my implant was removed. Although it gives me instant updates, I find more can be done. This is why “War on Epilepsy” is tapping into artificial intelligence (AI). I would like to design a device which not only keeps track of my routine and raise emergency calls, but also give patients more independence. This would allow me to go out onto the street at any time without someone accompanying me. When a seizure takes place, the AI which had learnt my caregiver’s voice will begin to deliver instructions: to reassure me and also to advise bystanders on what to do. The camera on the device will record the onset, so that physicians can become better informed of where and when seizures are likely to occur for this patient. Most of the time, patients do not know what is going on during a seizure, so recalling the episode back to the physician is nearly impossible. Another feature I would like to include in our wearable device is the ability to

aggregate data from different sources, be it implants or other devices. I also hope that it can stand alone without internet, so patients are free to go anywhere without the worry of a proper connection. AI MED: Some people may still be skeptical about what technology can offer to patients, how do you build trust? SEAN: This is an interesting question. Indeed, it is getting harder to build trust these days especially after so many data breaches and scandals. Personally, a big message from me is always to create a culture of openness and transparency. For example, we will always inform patients what we are using their data for, be it to train our AI or machine learning algorithm or channeling it to our medical team. There is always a communication between us and the patients. At the end of the day, the result is owned by both us and the patients. Likewise, when we released the research result to, let’s say, a

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

charity or a pharmaceutical company or anyone, patients will automatically be consulted and they will have to give their respective consents. AI MED: How do you balance managing your medical conditions while running War on Epilepsy at the same time? SEAN: I admit it can be somewhat challenging at times. Other times, I thought it actually helps my health because I am so focused on my work. I have had occasions where I pushed myself and delayed a seizure a little, but once I got home it hit again and it was much harder than normal. I believe things will change; I am hopeful and grateful with what technology is doing now. I think people should not view technology or AI as something that is going to destroy the World, especially in medicine. I see a convergence, where AI and human intelligence will all meet one day. AI will support what we do instead of taking over what we do.

9 1 0 2 R E B M E T P E S 9 1 -

TOWN SHOREDITCH LONDON

HALL

AI MED EUROPE 2019 IS FOR THOSE INTERESTED IN THE BIGGEST PARADIGM SHIFT IN HEALTHCARE AND MEDICINE: ARTIFICIAL INTELLIGENCE Register now at aimed.events

KEY SPEAKERS This is the No.1 event of its type in the Europe. Three-days dedicated to the transformative impact that AI-inspired technology is having on healthcare. A platform designed by clinicians to showcase latest thinking and facilitate new ideas and partnerships. Join us at Shoreditch Town Hall, London from 17 - 19 September and make your mark.

NOEL GORDON Chairman, NHS Digital (HSCIC)

ANTHONY CHANG MD, MBA, MPH, MS Founder AIMed, Chief AI Officer, CHOC

DR. INDRA JOSHI

JACKIE HUNTER

PROF. TONY YOUNG

Digital Health & AI Clinical Lead, NHS England

Chief Executive Officer, Benevolent Bio

National Clinical Lead for Innovation, NHS England

THE BIG PICTURE

EXPERIENCE AI MED CARDIOLOGY & RADIOLOGY THIS JUNE IN CHICAGO EVENT HIGHLIGHTS

CARDIOLOGY

RADIOLOGY

EXPERIENCE THE FUTURE OF RADIOL OG

RITZ CARLTON CHICAGO Anthony Chang Chief Intelligence Officer M13 CHOC and Founder AIMed

18 -19 JUNE 2019 Key topics

The need for unbiased Ground Truth Data. Overhaul of IT infrastruc ture within the healthcar e system at large. Patient privacy/protection issues. Develop comprehensive partnerships between clinicians with the vendor community to develop tools of clinical utility and usability. How do we speed up research without putting scientific integrity Is there a hybrid model? at risk.

EXPERIENCE THE FUTURE conferences taking place in North he mission of AIMed is to OF CARDIOLOGY T bring healthcare professionals, America,&Europe and Asia, AIMed is also hosting subspecialties meetings technology experts, business CARDIAC SURGERY FOLLOW US

17-18 JUNE 2019

in various areas. partners enthusiasts, on board and RITZ and CARLTON CHICAGO discuss the opportunities and feasibilities To begin, there are AIMed Cardiology of incorporating artificial intelligence Anthony Chang Key topics Intelligence Officer Deep Learning and Cardiac Image Interpretation and AIMed Radiology, the two clinical (AI) Chief and related new technologies into M13 CHOC and AI in Decision Support in Cardiology and Cardiac Surgery Founder AIMed regions which show a relatively high medicine, with the goal of providing Precision Cardiovascular Medicine Big Data and Databases in Cardiology and Cardiac Surgery adoption of AI at the moment because more holistic and efficient care. Wearable Technology and Embedded AI for Cardiac Care Robotics and Virtual Assistants of the advanced imaging tools that are Augmented/Virtual Reality in Cardiology and Cardiac Surgery Future of Heart Program robustly Administration Using AI employed. Both events will AIMed had its inaugural one-off Medical Education and Training Using AI be 1.5 days long over 17 and 18 June satellite meeting back in 2016. Forand theCybersecurity Blockchain in Heart Program for AIMed Cardiology, and 18 and 19 past few years, we have witnessed June for AIMed Radiology, at The Ritza growth of attention and audience. Carlton, Chicago, Illinois. This year, apart from the annual FOLLOW US

Attendees may find themselves immersed in practical sessions, learning new ways to examine medical imaging with the use of data science and programming through live demonstrations. There are also workshops for attendees to learn more about certain aspects of AI such as Machine Learning, Deep Learning and Convoluted Neural Network etc. and explore ways to integrate them into their present workflow. In open seminars, distinguished guest speakers will share with fellow audience their insights into AI and Cardiology or Radiology. Specifically, tips on how to make AI less cumbersome to work with, ethical queries revolving around AI adoption and patientsâ&#x20AC;&#x2122; safety/privacy, differentiating between AI hypes and real trends, and how the industry will change in the next couple of years.

LIST OF SPEAKERS (NON-FINALIZED) The actual agenda and guest list have yet to be finalized but here are the snippets of who is going to share what in the upcoming AIMed Cardiology and Radiology.

FOR AI MED CARDIOLOGY: Dr. John Rumsfeld The Chief Innovation Officer of the American College of Cardiology and Professor of Medicine at the University of Colorado School of Medicine is going to explore whether AI will disrupt cardiovascular (CV) care. He will delve into the present landscape of AI in relation to CV care, potential areas where AI can be adopted and also efforts to advance evidence-based… with the use of AI. Dr. G. Hamilton Baker The duo board certified in pediatric and adult congenital cardiologist is going to present the challenges faced by physicians to select the right AI tools and maximize return of investment in the era with an exploding number of vendors. Dr. Baker will share strategies for clinical implementation, choosing between AI tools and offthe-shelf solutions, as well as gauging the effectiveness, consistency and potential bias of AI algorithms. Dr. Jai Nahar The Pediatric Cardiologist of Children’s National Health System’s talk will focus on Conversational AI, Virtual Voice Assistants and their potential uses in augmenting cardiovascular care and challenges in terms of adoption. Dr. David M. Axelrod The Pediatric Cardiologist and Critical Care Physician at Lucile Packard Children’s Hospital Stanford is using congenital heart disease as a

springboard to discuss virtual reality, augmented reality and immersive technologies. Attendees are expected to learn more about the infrastructure required to kick-start or to maintain these clinical and educational programs and possibilities of using extended or mixed realities as future research tools.

FOR AI MED RADIOLOGY: Sonia Gupta Director of Ultrasound, Beth Israel Deaconess Medical Center will discuss the “New Power Dynamics” brought about by AI in Radiology. Specifically, the expected power shift from physicians to patients as knowledge becomes free-flow and the importance of collaborations between physicians, industry, engineers and patients to shape AI in medicine. Neil Tenenholtz Director of Machine Learning at MGH & BWH Center for Clinical Data Science will share the collaboration between his organization and Massachusetts General Hospital and Brigham & Women’s Hospital, which focuses on developing AI algorithms to improve patient care, translating them into clinical practice and commercialization to maximize patient benefits. Dr. Ross Filice The Associate Professor and Chief of Imaging Informatics in the Department of Radiology at MedStar Georgetown University Hospital and Chief of Imaging Informatics at

MedStar Medical Group Radiology will talk about AI in Radiology without images. While most AI tools target image-based use cases, there are also other equally powerful deep learning techniques which aim to improve quality and efficiency. Dr. Filice will be elaborating on some of them. Ben Panter Chief Executive Officer and Founder of Blackford Analysis, a medical imaging company which uses its Blackford platform to streamline the procurement, deployment, use and support of clinical application and AI products, and will touch upon solutions that make AI work for the Radiology majority. Specifically, he will assess the different value propositions being tested in the market by AI vendors and traction in various setting. -----------------------------------

HOW TO GET THE MOST OUT OF AI MED CARDIOLOGY AND RADIOLOGY “Come with an open mind” is something which AIMed always advocates. As mentioned, both events will attract delegates and speakers from all over the World with different expertise. Do not be afraid to start discussing your curiosity if you are intending to adopt AI into your workflow. Likewise, share your stories and seek advice if you are already using AI in your work. AIMed tries to create a friendly community for everyone, regardless of their clinical or technical background. No one should feel left out, unless they are not part of the crowd to begin with.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE BIG PICTURE

HIMSS EUROPE & HEALTHCARE 2.0 One of Europe’s biggest AI and digital health events that you don’t want to miss

fter the success of last year’s inaugural HIMSS Europe & Healthcare 2.0 conference in Barcelona, the Healthcare Information and Management Systems Society (HIMSS) is hosting the event again in June, this time in Helsinki, Finland. This year’s focus is massive, as highlighted on the official website. Basically, they can be spilt into five board areas: to assimilate health and social care, facilitate a shift from acute to community-based care, encourage a secure, ethical and actionable flow of data, ponder of the pros and cons of artificial intelligence (AI) and explore opportunities and challenges brought about by new innovations. AIMed caught up with Sean Roberts, Vice-President of HIMSS EMEA as he brings us through the details and insights and where should attendees set their eyes on, when they are at this mega European event of the year. AI MED: In terms of scale and focus, how does HIMSS Europe & Health 2.0 conference differ from HIMSS2019 took place this February in Florida? SEAN ROBERTS: HIMSS2019 was a global conference, it was meant to be on a global platform. We try to incorporate the best practices and top leaders from all over the World to

come together, to have a voice and to share their expertise. We are glad to see an increase in international representatives over the years. I will say by about 10% increase annually. In spite so, because of the location and the presence of local digital health communities, the conference was still a rather North American-centric. I believe the main difference will be both events are trying to engage regionally and locally. HIMSS Europe & Health 2.0, in turn, will be much more Euro-centric. We are trying to attract the European community and around the perimeter of that. Perhaps we will see delegates coming from North Africa, Eurasia, the Middle East and Russia. There will still

be guests coming from North America and Canada but the general interests will be the communities in Europe. AI MED: In your opinion, how similar/ difference are the AI and digital health culture in Europe and the US? SR: That’s a very good question. I would say very different. Most countries in Europe is offering socialized medicine but, in the US, it’s not. I think they are using more technology in the US and they are seeing it as a wow factor. In Europe, it’s more about embracing technology and there are still some exciting things that we can do. But I believe culturally, in Europe, once the technology is there, it will be more impactful on the healthcare system.

AI MED: Why is Helsinki chosen for this year’s conference? SR: Generally, we look at locations from the community itself; who we are engaging with and so on. Finland was an easy choice because they are doing so much on innovations. At HIMSS, our mission is to showcase the best practices and to provide a global driving force for leaders and entire digital health ecosystem. I am not saying that innovation is not happening elsewhere. It’s just so that HIMSS has a more traditional approach while Finland is introducing cutting-edge reforms. We are trying to transform but at the same time, keeping our mission. So, it makes sense to partner with the Finnish government and other nearby regions who are rich in working across borders and demonstrating their new technologies. AI MED: AI resources and buy-in can be quite diverge among countries, we

wonder if HIMSS is doing anything to bring everyone on board at the same starting point to look at AI together?

think that’s why when HIMSS acquire Health 2.0, that was one of the valueadd for us as a global organization.

SR: HIMSS has a special interest in creating a community. Conventionally, AI communities can be quite diverse, depending on each specialty, they can vary from the clinical side, to the technology side and the innovative side. So, we are working closely with everyone, to create a special interest kind of community.

As mentioned, HIMSS has a more traditional approach and we have been working with so many companies, hospitals, regions and government for many years so the addition of Health 2.0 brought a change. It brought in technology and other ways of working, with lots of focus on social media and applications, in a way that people are communicating these days.

At the same time, we are also bringing in top leaders, to the different organization and partners, so there is a kind of merge. We also play a role in showcasing best practices from different parts of the World, to have this whole knowledge share and driving healthcare to the next level.

When we marry the two models together, I believe that’s when we are putting values into healthcare. Indeed, it can be disruptive but at the same time, it’s actually a plateau where we can see the strength in it. AI MED: How do attendees get the most out of HIMSS Europe & Health 2.0 conference?

AI MED: What are some of the challenges to achieve that? SR: There are always challenges in making sure that we remain relevant to the ecosystem. After all, healthcare is a broad topic and there are certain groups or special areas that people will pay particular attention to. Hence, finding a balance, keeping our mission and being consistent and respond to the needs of the healthcare community, is a challenge. AI MED: There has been a growing number of startup in the recent years. Do you feel these startups have changed the dynamics of HIMSS conference? SR: Exactly and definitely it has, in a positive way. The emergence of startups is a big game changer and I

SR: There are some off-site tools and infrastructures like the two new hospitals that had just been finalized in Finland. The facilities and technology are incredible. The two of them are actually public-funded as well, it was not just tax but investments made by the citizens of Finland. The Finns put in a lot of effort and details on their patients so apart from the conference, the talks and meeting other delegates, one can actually be immersed in these effort and details. We are making sure that attendees can capture some of these local essence during the conference. This is the first time we are doing with a partner who is so engaged. So, we are very grateful.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

THE BIG PICTURE

WHAT TO EXPECT AT THE HIMSS MACHINE LEARNING AND AI FOR HEALTHCARE EVENT Taking place in Boston in mid-June, the conference will offer a holistic view of the skills needed to deploy and scale analytics, AI and machine learning.

ealthcare organizations can’t succeed at machine learning without a solid foundation of analytics and data. That reality is the starting point for our HIMSS Machine Learning and AI for Healthcare event, scheduled for June 13-14, 2019, in Boston. The conference will focus on building analytics maturity and, beyond that, the requisite skills for moving into the age of artificial intelligence. Providence St. Joseph Chief Clinical Officer Dr. Amy Compton-Phillips will kick things off with an opening keynote titled, “No Data Without Stories, No Stories Without Data,” where she’ll discuss the drive to convert mountains of data into information, then use AI and ML to simplify it so physicians can put that data to work. Next up, Steven Horng and Michael Schwarz will chart a roadmap from analytics to algorithms; Horng is associate director in the Emergency Medicine Informatics division at Beth

Israel Deaconess Medical Center and Schwarz is the executive director of IS at Indiana University Health. The morning sessions will also feature a deep dive into HIMSS Adoption Model for Analytics Maturity with Duke University Health System Chief Analytics Officer Stephen Blackwelder – Duke is the first hospital to achieve Stage 7 of AMAM – and James Gaston, senior director at HIMSS Analytics. Other sessions will include a look at mission-critical data quality and governance from Tina Esposito, chief health information officer at Advocate Aurora, insights about developing an analytics team from Northshore University HealthSystem Assistant Vice President of Clinical Analytics Chad Konchak and Rush University Medical Center Chief Analytics Officer Bala Hota will share expertise about maximizing AI’s impact. That’s just the morning. Afternoon sessions will feature UPMC CIO Srinivasan Suresh on scaling

analytics in clinical operations and UNC Academy of Population Health Director Dr. Michael Dulin discussing ways to avoid bias in analytics, AI and population health work. Day Two opens with Children’s Hospital of Orange County Medical Director Dr. Anthony Chang’s morning keynote. During “Common misconceptions of AI in healthcare,” Chang will deliver a reality check on the chatter about AI replacing physicians. He’ll also explore deep learning for decision support, AI’s promise to return the joy to medicine and future challenges. Following Chang, a panel discussion will look at strategies for incorporating AI into clinical practice, and then Harvard Medical School Assistant Professor Len D’Avolio will share tactics for analyzing, optimizing and ultimately customizing AI and ML deployments. Vikas Chowdry, chief analytics and information officer at Parkland Center for Clinical Innovation, will demonstrate

BY: TOM SULLIVAN, EDITOR-IN-CHIEF, HEALTHCARE IT NEWS

the value of making ML actionable, while other morning speakers will discuss ways to overcome challenges and use ML to reduce risk. In the final morning session, a few speakers will come back to the stage to outline a Success Checklist for artificial intelligence and machine learning. The afternoon sessions will dive into more tech topics, including blockchain and natural language processing. Mount Sinai Health Systemâ&#x20AC;&#x2122;s Director of Data Engineering Varun Gupta, in fact, will relate the systemâ&#x20AC;&#x2122;s experience using AI and NPL to uncover social determinants of health in unstructured EHR data.

And in the final session of the event, William Paiva, executive director of the Center for Health Systems Innovation at Oklahoma State University, will discuss using machine learning to address rural healthcare challenges, such as improving Native American health with innovative care delivery and technology models. The HIMSS Machine Learning and AI for Healthcare event takes place June 13-14, 2019, at the Westin Copley Place in Boston. Learn more at healthcaremachinelearningai.com and register today!

CONTACT THE WRITER: Twitter: @SullyHIT Email: tom.sullivan@himssmedia.com This article originally appears on Healthcare IT News, a HIMSS Media publication.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

REVIEW

Top 100A

FOR AI IN MEDICINE

PART 01

#01 CONVOLUTIONAL NEURAL NETWORKS: AN OVERVIEW AND APPLICATION IN RADIOLOGY An excellent review (with outstanding diagrams) of the relatively abstruse concept of convolutional neural networks and deep learning that can be mostly understood by AI enthusiasts without a data science background. In addition to elucidating difficult concepts like convolutional and pooling layers as well as loss function and gradient descent, the authors even explain the challenges of CNN such as overfitting and small datasets. This paper is a must-read for any sub specialist interested in AI and especially for image-intensive subspecialties like radiology, pathology, dermatology, ophthalmology, and cardiology.

#02 ARTIFICIAL INTELLIGENCE IN CARDIOLOGY An excellent overview of artificial intelligence concepts such as predictive modeling concepts as well as supervised and unsupervised learning, including new concepts such as â&#x20AC;&#x153;dichotomaniaâ&#x20AC;?. What is valuable about this paper is the relevant discussions on pitfalls of predictive modeling as well as the various methods in machine learning (with excellent illustrations). Given that cardiology is an image intensive field, there is surprising little on convolutional neural network but the authors did include reinforcement learning. Finally, the background AI material is complemented by a good discussion of its application in cardiovascular medicine.

Articles #03

ARTIFICIAL INTELLIGENCE IN HEALTHCARE

This is a timely comprehensive review of main principles of artificial intelligence in healthcare. The authors start with a historical perspective of AI in medicine and then discusses the current and potential AI applications in medicine. Sections include image-based diagnosis, genome interpretation, and others. The most helpful section of the paper explains the models of information flow in a myriad of situations from clinical practice to fully automated clinical systems. There is a short but helpful glossary of key terms included in this paper.

#04 PRECISION RADIOLOGY: PREDICTING LONGEVITY USING FEATURE ENGINEERING AND DEEP LEARNING METHODS IN A RADIOMICS FRAMEWORK A very well thought out introductory proof-of-principle of precision radiology by a dually trained radiologist-data scientist. This work focuses on how deep learning with convolutional neural networks can be applied to radiomics research and how this strategy helps to develop new methods to assess phenotype variations (as clinical and genomic data often are not adequate).

#05 DEEP LEARNING The ultimate review by the godfathers of deep learning on key concepts such as convolutional and recurrent neural networks as these relate to image interpretation as well as language processing. There will be no other reviews of this high intellectual caliber for quite some time (unless these three DL experts decide to update this review). While the data science is at times very esoteric, anyone who have the patience and fortitude to read this outstanding treatise will be far ahead of others in their group.

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

BOOK

REVIEW

THE ALGORITHMIC LEADER MIKE WALSH

“The greatest threat we face is not robots replacing us, but our reluctance to reinvent ourselves” – Mike Walsh, author of The Algorithmic Leader Mike Walsh, the CEO of Tomorrow, heads a global consultancy on designing companies for the future. In his latest book, The Algorithmic Leader: How to be Smart When Machines are Smarter than You (Page Two Books, Canada, 2019, 256 pages), he writes on his prescient vision of how future leaders will need to understand, manage, and leverage machine intelligence and artificial intelligence. The algorithmic leader (not a great term), as he defines it, is someone who has successfully adapted their decision making, management style and creative output to the complexities of the machine age. Among the ten principles that he espouses in the book, a few stand out as being more relevant for healthcare and medicine. His first principle, Work Backward from the Future, describes how the algorithmic leader starts with a strong vision for the future and then works backwards by using their imagination to leverage machine intelligence. Any healthcare leader will need to appreciate this principle with the exponential convergence of future technologies in medicine, led by artificial intelligence. We have much to learn from Masayoshi Son, the President of SoftBank, as well as our children for their

capability to see the future. In addition, Walsh elucidates the principle Think Computationally as a leadership strategy to reason not by analogy, but by taking a problem apart and looking at it again from the perspective of fundamental truths. In the principle Automate and Elevate for leaders of the future, Walsh advises that they themselves invest in this new AI capability as well as encourage and inspire others to reconfigure their jobs to maintain relevancy. As automation removes repetitive parts of our jobs in healthcare and medicine, managing exceptions, as well as finding nonlinear solutions to complex problems, will be the reconfigured jobs for most people in healthcare and medicine. Lastly, the principle When in Doubt, Ask a Human reminds us that human cognition remains essential, especially in the machine age, and that the algorithmic leader uses humancentered design for machine learning. This very readable and timely guide, while not particularly insightful in the modern nuances of AI, does have a myriad of important principles that have a high degree of relevance for healthcare and medicine leaders of the future. Perhaps this practical guide is simply a necessary read for anyone.

REVIEWER

DR ANTHONY CHANG, MD, MBA, MPH, MS is the Chief Intelligence and Innovation Officer and Medical Director of the Sharon Disney Lund Medical Intelligence and Innovation Institute (MI3) at CHOC Children’s. He is the Chairman and Founder of AIMed.

DIGEST

EXECUTIVE SUMMARIES Our longer features broken down into takeaways P 08 - 10

P 18 - 22

CAN PHYSICIAN AND DATA SCIENTIST WORK TOGETHER?

AI IN RADIOLOGY: ASKING SOME HARD QUESTIONS

By Hazel Tang

By Tirath Y. Patel, MD

•

There are two camps of radiologists: one who believes AI will be an adjunct to radiology and not necessarily a replacement, the other is far more pessimistic, foreseeing a future with fewer radiologists.

•

AI is likely to take up manual tasks, as in the case of the Coulter counter, which is used to interpret complete blood cell count in pathology. Nevertheless, healthcare leaders do warn that radiologists who use AI may replace those who don’t in the near future.

•

Consolidation of healthcare drives many of the initiatives being performed in large academic medical centers to implement AI in radiology.. Questions remain whether smaller practitioners will partner up with large entities to leverage AI for gain.

HealthPals, a startup which screens patients at a population level with the aim to uncover gaps in the healthcare system that prevent individuals from receiving adequate care, was set up by data scientist Sushant Shankar and cardiologist Rajesh Dash.

•

Dr. Dash believes it takes courage for medical professionals to accept AI because physicians are trained in evidence-based care that requires continuous validation and peer critiques to ascertain the standard.

•

The duo expressed patience and sharing as the common language for building something with the use of technology as the key for professionals with different formal training to work together.

assess the cost-effectiveness of expensive therapies in cardiology.

•

P 12 - 17 ETHICAL AND LEGAL CHALLENGES OF ARTIFICIAL INTELLIGENCE IN CARDIOLOGY By Sara Gerke, Daniel B. Kramer, and I. Glenn Cohen

•

AI could potentially improve the selection of device recipients and

•

However, AI applications will only be as good as the training data that is fed to them. Using AI to assist physicians also brings new challenges for the patients’ right to informed consent, particularly how their medical data will be used. Overall, the use of AI remains vulnerable to cybersecurity. AI has the potential to transform healthcare but that comes only when related ethical and legal challenges are duly addressed.

P 24 - 31 HOW CAN AI BE USED IN RADIOLOGY – AN OUTLINE FOR ALGORITHM DEVELOPMENT By Nina Kottler

•

In radiology, many applications were developed using the available data and an incomplete understanding of what

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

DIGEST

radiologists do. The idea of an image interpreter propels AI algorithms to be mostly targeting image interpretation and pathology detection.

•

To fully utilize the power of AI in radiology, there is a need to determine where it can provide the most value. To do so, the radiology workflow needs to be broken down into individual components.

•

Over here, a radiology workflow includes order, protocol, image acquisition and reconstruction, workflow, relevant patient information, image interpretation (detection and diagnosis), report, communication of findings, follow-up and peer learning.

P 44 - 47 GENOMICS + RADIOMICS = A STEP TOWARDS BETTER HEALTHCARE By Hazel Tang and Erwann Vieu

•

The new SOPHiA Radiomics is an added feature to its existing AIdriven, genomics focused platform. The combination has the ability to enhance present cancer diagnosis and treatment.

•

The tech company’s mission is to democratize data-driven medicine by growing a user community which promotes sharing and accessibility of valuable knowledge.

•

It has been awarded a $77 million Series E funding at the turn of the year and they will be expanding their presence in US, as well as adding new analytical capabilities and combining different sources of medical data into the platform.

P 40 - 43 TO BUILD, SHAPE AND SUSTAIN: THE GOALS OF INTEGRATING AI IN HEALTHCARE By Hazel Tang and Allison Kavanagh

•

Molly K. McCarthy, National Director, US Provider Industry and Chief Nursing Officer at Microsoft and John Frownfelter, Chief Medical Information Officer at Jvion were invited to speak at the recent AIMed webinar on how to shape and set artificial intelligence (AI) goals within healthcare.

P 48 - 49

•

A new startup company employs realtime work data to predict burnout, one of the main reasons that is causing doctors to leave their career.

•

Speakers agreed it is crucial to examine existing resources within the organization, facilitate dialogue between departments and draft out a timeline. Do not underestimate the effort required to develop or adopt an AI solution.

•

The company do away with traditional survey as it believes data from work will provide a better insight into how employees interact with one another in the team and how physicians make use of electronic health records.

•

A use case example was shared and speakers believe, at the end of the day, healthcare is about people, so AI should not lose its human touch. Things like empathy will never be replaced by machine.

•

The company hopes that the deep learning algorithm that they are training will, not only highlight those who are at risks of burnout, but also, create a better working environment for everyone.

P 50 - 54 THE WARRIOR AND HIS INVISIBLE BATTLES By Hazel Tang

•

Sean Hamilton’s seizure and heart condition had put an abrupt end to his dream of becoming a paramedic. However, it opens a new door for him to start his own company “War on Epilepsy”, which promotes the awareness of epilepsy and encourages patients to use technology for better care management.

•

He cited that often people panic when they witness an ongoing seizure. The government is cutting down funding for epileptic patients, but he believes most of the challenges faced can be overcome with public education and technology.

•

His fun-loving and never-give-up character drives his desire to be in control of his seizures while accomplishing what he wants to achieve for the wider population.

THE MAN BEHIND NO-SURVEY SURVEY By Hazel Tang

AI MED I THE CARDIOLOGY & RADIOLOGY ISSUE

June 13-14, 2019 | Westin Copley Place | Boston, MA

Wherever your organization is on its analytics, machine learning or AI journey, this event is programmed to help you move your initiatives forward.

Save $200 using the code AIMed200 at registration

OUR KEYNOTE PRESENTERS

AMY COMPTON-PHILLIPS Executive Vice President and Chief Clinical Officer Providence St. Joseph Health

ANTHONY CHANG Chief Intelligence and Innovation Officer CHOC Children

ATTEND the Machine Learning and AI in Healthcare event focused on implementation! HealthcareMachineLearningAI.com | #smartHIT

“

I’m confidently living on my own. HP PHIT Solutions for home-based monitoring Safer, smarter, and secured for healthcare

HP PHIT Solutions, a portfolio of population health IT products and services, is helping build patient and provider confidence in remote patient monitoring by: •

Offering support for Apple® products—including iPad®, iPhone®, and Apple Watch® —as well as Chrome OS™, Android™, and Windows devices

•

Managing the packaging of software, hardware, and peripheral devices

•

Deploying and supporting device bundles through optional HP DaaS

•

Providing affordable and flexible financing

Safer and smarter based on HP Healthcare Edition products compared to other HP products as of August 2017; based on ability to be cleaned with sanitizing agents listed in the user guide, EN60601-1-2 testing, and DICOM calibration.

HP DaaS plans and/or included components may vary by region or by Authorized HP DaaS Service Partner. Please contact your local HP Representative or Authorized DaaS Partner for specific details in your location. HP services are governed by the applicable HP terms and conditions of service provided or indicated to customer at the time of purchase. Customer may have additional statutory rights according to applicable local laws, and such rights are not in any way affected by the HP terms and conditions of service or the HP Limited Warranty provided with your HP product.

Financing and service offerings available through Hewlett-Packard Financial Services Company and its subsidiaries and affiliates (collectively HPFSC) in certain countries are subject to credit approval and execution of standard HPFSC documentation. Rates and terms are based on customer’s credit rating, offering types, services and/or equipment type, and options. Not all customers may qualify. Not all services or offers are available in all countries. Other restrictions may apply. HPFSC reserves the right to change or cancel this program at any time without notice.

© Copyright 2019 HP Development Company, L.P. The information contained herein is subject to change without notice. Apple, iPad, iPhone, and Apple Watch are trademarks of Apple Inc., registered in the U.S. and other countries and regions. Windows is a registered trademark of Microsoft Corporation in the United States and/or other countries. Android and Chrome OS are trademarks of Google, Inc. Printed in the United States.