AI Health Events
Our goal is to share learning experiences with a broad community, both at Duke and beyond. Our events include the AI Health Seminar Series, the monthly Spark Imaging Seminar Series, and workshops and studios throughout the year. You can sign up for our mailing list to receive emails about upcoming events.
Machine Learning for Precise Diagnostics and Therapeutics
Presented by: Rohit Singh, Assistant Professor of Biostatistics & Bioinformatics with host Aditya Parekh
Drug discoveries have been instrumental in improving global health over the last century, but the median drug now takes about 10 years to bring to market and costs over a billion dollars to develop. The work in our lab aims to expedite the development of precise diagnostics and therapeutics by applying machine learning. In this talk, I will outline two recent research directions. In the first part, we use single-cell multiomics to discover regulatory mechanisms governing the interaction between the epigenome, transcription factors, and target genes. In the second part, I will introduce the application of large language models to model protein interaction and function. These protein language models enable powerful new approaches to predicting and understanding protein-protein and protein-drug interactions. I will conclude with a prospective look at how similar methods may help answer foundational questions in both basic and translational science.
Ushering in the Algorithmic Century: Duke’s Pathfinding Leadership in Ensuring that Health AI Is Safe, Effective & Equitable for All
Tuesday, April 30, 2024 | 12:00 PM – 1:00 PM (Eastern time)
Virtual seminar via Zoom, open to members internal and external to Duke
Register here: https://duke.zoom.us/webinar/register/WN_pbPaFO89R-GpfALN23Hbiw
Presented by:
In this seminar, Dr. Michael Pencina, Vice Dean for Data Science, Chief Data Scientist for Duke Health, and Director of Duke AI Health, will examine the ongoing revolution in health data science and the application of artificial intelligence tools to the field of medicine. He will trace the evolution of increasingly sophisticated AI-powered technologies now being leveraged for patient care and clinical research, surveying the challenges and opportunities presented by these algorithmic technologies. In particular, the seminar will focus on the increasingly urgent need for robust oversight of health AI tools that encompass the entire lifecycle of such technologies, from development through validation and deployment. He will explain how robust frameworks of review and continuous oversight – including ones being articulated by Duke and its partners – are critical to ensuring that current and future applications of AI in healthcare and clinical research are safe, accurate, trustworthy, and equitable.
This seminar will emphasize Duke University’s contributions to a field that is rapidly reshaping the practice of healthcare and biomedical science for the coming century. The presentation will touch on the history of applied healthdata science at Duke, understood as part of a tradition of innovation that dates back to the pioneering efforts of Duke Cardiology professor Eugene Stead in creating computer-based predictive tools for improving patient care. In addition, the seminar will also acknowledge the larger legacy of innovation and collaboration between the disciplines of medicine and computer science at the university. It will also explore how Duke continues to play a major role in the AI revolution as it transforms medicine and clinical research in the coming century, with particular emphasis on developing frameworks of governance, review, and oversight that ensure that health AI technologies are safe, trustworthy, and equitable.
Centennial Summit on Data Science in Surgery
April 18, 2024, 8:00 AM – 1:00 PM | In person at Trent Semans Center for Health Education, 6th Floor
Register here: https://duke.qualtrics.com/jfe/form/SV_abLQdJa7FI81G4e
The Department of Surgery will celebrate Duke’s Centennial with the Centennial Summit on Data Science in Surgery, featuring speakers from Duke Surgery and Duke AI Health, to examine the future of data science and innovation in surgical application. The keynote address will be given by Ozanan Meireles, Vice Chair for Innovation in the Department of Surgery and Director of Surgery for Duke AI Health. AI Health’s Director Michael Pencina will also be a featured presenter.
AI Health Virtual Seminar: Myth vs. Reality: Unraveling common misconceptions about heat-related illnesses
Thursday, April 18, 2024 | 12:00 PM – 1:00 PM (Eastern time)
Virtual seminar via Zoom, open to members internal and external to Duke
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=06e91f16-9ab9-4a34-9e89-b15a0116b3bb
Follow along with the slides here: https://duke.box.com/s/gei22rh16pxt71eq9ng4kw1lx4zzz74a
Presented by:
Ashley Ward, PhD
Director, Heat Policy Innovation Hub
Nicholas Institute for Energy, Environment & Sustainability
with moderator Jennifer M. Lawson, MD, MA, Assistant Professor of Pediatrics and Faculty Associate of the Trent Center for Bioethics, Humanities and History of Medicine at Duke University
Summer of 2023 served as a wakeup call for many regarding the impact of extreme heat. The toll on communities was severe. In regions of the US, planes were grounded, emergency departments and hospitals were at capacity, industry operations were interrupted, and the military cancelled training exercises. While it is not a secret, it is also not a well-known fact that heat kills more people in the US than any other weather-related event, and its impact on the health of the most vulnerable is severe. Moreover, as temperatures continue to rise, so do the risks associated with heat-related illnesses, making it crucial for health providers to stay informed and equipped with accurate knowledge. In this session, we will discuss extreme heat and its impact on public health, focusing on the prevalent myths and misconceptions surrounding extreme heat, shedding light on the realities that healthcare providers need to understand to effectively mitigate risks and provide optimal care.
AI Health Virtual Seminar Series: Evaluating Generative Large Language Models in Healthcare
Tuesday, April 16, 2024 | 12:00 PM – 1:00 PM (Eastern time)
Virtual seminar via Zoom, open to members internal and external to Duke
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=c927af71-1560-4181-b0db-b154012694bd
Presented by:
Chuan Hong, PhD; Assistant Professor of Biostatistics & Bioinformatics, Duke University School of Medicine
The rapid evolution of large language models (LLMs) has ushered in a new era of computational linguistics, yet a systematic approach to their evaluation, particularly in sensitive domains such as healthcare, remains nascent. This work bridges these gaps by offering a detailed and integrated review of qualitative evaluation, quantitative evaluation, and meta-evaluation. For quantitative evaluation, our review introduces a taxonomy of evaluation metrics, categorizing them based on essential dimensions such as human supervision, contextual data, and analytical depth. In addition to generic settings, our work distinctively emphasizes additional considerations vital in the healthcare sector. As a result, we propose an integrated cross-walk between qualitative and quantitative assessment methods. The proposed framework harmonizes qualitative insights, such as user-focused evaluations, with objective quantitative metrics. We present a detailed “go-to menu” of evaluation criteria, tailored to address specific healthcare applications and emphasize distinct aspects in both pre-deployment and post-deployment phases. Our findings underscore the need for evaluations that extend beyond mere technical accuracy, factoring in medical ethics, fairness, equity, and potential operational biases. Our work offers a summary of existing methods of LLM evaluation that can establish a baseline from which future evaluation methods can be developed to keep pace with the rapid advancements in the field.
AI Health Virtual Seminar Series: Using artificial intelligence to automate the screening of psychiatric distress in ophthalmology clinics using electronic health records
Thursday, April 11, 2024 | 4:00 PM – 5:00 PM (Eastern time)
Virtual seminar via Zoom, open to members internal and external to Duke
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=45c35073-afbc-4d45-b336-b150015b43a9
Presented by:
Samuel Berchuck Ph.D. Assistant Professor of Biostatistics & Bioinformatics, with host Ayush Jain
In patients with ophthalmic disorders, psychosocial risk factors play an important role in morbidity and mortality. Proper and early psychiatric screening can result in prompt intervention and mitigate its impact. Because screening is resource intensive, we developed a artificial intelligence (AI) infused framework for automating screening using electronic health record (EHR) data. EHR data in our study came from the Duke Ophthalmic Registry, a retrospective EHR database that contains medical and clinical records for all patients seen at the Duke University Eye Center since 1993. Our study demonstrates that prescreening for distress in ophthalmology patients can be automated using an AI algorithm trained on existing EHR data. The algorithm identified distress in patients already being treated, and in those with incident distress. These findings suggest that screening for distress in ophthalmology clinics is feasible and may reduce negative health outcomes in patients.
Duke Symposium on Fostering AI/ML Research for Health Equity and Learning Transformation (FAIR HEALTH™)
March 13-14, 2024 | In person at the Duke University School of Nursing
We are thrilled to invite you to our upcoming symposium, Fostering AI/ML Research for Health Equity and Learning Transformation (FAIR HEALTH™), scheduled for March 13-14, 2024, at the Duke University School of Nursing in Durham, NC. This two-day event is dedicated to advancing discussions on cutting-edge research and practices aimed at promoting equity and fairness in algorithmic systems.
AI Health Hands-on Workshop: Using Large Language Models (LLMs) while measuring their accuracy and reacting to improve quality
Friday, January 26, 2024 | 11am – 2pm EST
In person in CRTP 2 in Hock Plaza, Live virtual streaming also available
Registration is open to everyone, including both academia and industry partners
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=74b7575d-7dc0-49a2-b6d9-b1030165887b
You can view the slides here: https://docs.google.com/presentation/d/1_XKxJzapiImVpTUr6wdR5tZXOrA9dXR06NUthZO_nFg/edit#slide=id.g2b38dd3077e_0_0
Presented by:
Sage Arbor, PhD., Sr. Informaticist,
Technology & Data Solutions, Duke Clinical Research Institute
Please join Sage Arbor, PhD. for a dynamic and collaborative learning experience where participants will delve deep into the world of large language models. Attendees will engage in hands-on activities ranging from prompt engineering techniques, investigating how AIs can improve other AIs, exploring the quality of model outputs, and making insightful comparisons between different language models.
Embark on a journey to understand the nuances of quality metrics, distinguishing between those that are easily measurable and those that pose challenges. Explore the multifaceted aspects of evaluating AI-generated content, emphasizing the importance of human evaluation in refining model performance.
One of the highlights of the workshop is the discussion and exploration of various strategies to enhance AI models. AIs with personas of quality assurance inspectors which cycle to improve another AIs output will be explored. Attendees have the opportunity to gain practical insights into optimizing and customizing models to better suit specific tasks and domains, thus appreciating the potential for tailoring AI to their unique needs.
In this intellectually stimulating workshop, participants not only gain a deeper understanding of large language models but also acquire the skills and knowledge necessary to harness the full potential of AI in their respective fields.
Artificial intelligence and the ethics of use: Patient and provider perspectives on utilizing prediction models in medical decision-making
Wednesday, January 17, 2024 | 4:00 PM – 5:00 PM (Eastern time)
Virtual seminar via Zoom, open to members internal and external to Duke
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=feb399a5-8641-4e36-a08b-b0fb0130d465
Presented by:
Jessica Sperling (Smokoski), Ph.D.
Social Science Research Institute (SSRI)
Clinical & Translational Science Institute (CTSI)
Clinical predictions are a critical component of delivering quality healthcare. As clinical data grows rapidly in volume and quality, a large amount of research has been dedicated to developing machine learning (ML)-based clinical predictive models (CPMs), which can leverage large amounts of patient information to predict life expectancy, disease progression, and other health outcomes. In conjunction with attention to these models’ development, we must also consider ethics of how these models can be best utilized and understood by patients and providers, including considering interpretability and implications for medical decision-making. In this seminar, we discuss to a current qualitative study examining these factors as understood by four key end-user groups: clinical providers; support providers such as dialysis staff and social workers; patients; and patients’ caregivers (e.g., family members). As a use case, this study specifically examines a mortality prediction model for patients undergoing hemodialysis. We identify key factors related to trustworthiness, interpretability and use, and we provide suggestions of use-focused considerations to examine and prioritize ongoing.
FAIR HEALTH (Fostering AI/ML Research for Health Equity and Learning Transformation) ™
January 8, 2024, 9:00 AM – 12:00 PM
In person at Sarah Duke Gardens in Kirby Horton Hall
Register to reserve your spot: https://duke.qualtrics.com/jfe/form/SV_6sCG5CMTCryDuGG
This event is open to anyone who is passionate about advancing healthcare through innovation while ensuring health equity and fairness in clinical algorithms.
This workshop will delve into the critical issue of algorithmic bias and clinical decision making. This is an opportunity to gain essential insights and practical strategies to identify, mitigate, and evaluate bias in clinical algorithms. We will also explore the legal and ethical implications of algorithmic bias in healthcare, an aspect that is gaining increasing importance in today’s dynamic healthcare landscape. To enrich the workshop and encourage active participation, we have incorporated a combination of lectures and an interactive case study discussion, making it an even more rewarding experience for all attendees. By engaging in this workshop, and learning from one another, we can pave the way for a future where healthcare algorithms enhance patient care, minimize bias, and prioritize equity.
Duke Health Data Science Poster Showcase
Friday, December 8, 2023 | 11:00 AM – 1:00 PM (Eastern time)
In person at the Sarah Duke Gardens, Kirby Horton Hall
We’re looking forward to holding our next Duke Health Data Science Poster Showcase in December 2023. Read more about the previous spring showcase here, and watch the highlights video on YouTube here.
More details will be announced in early fall 2023. Please join our mailing list if you would like to receive updates.
All are welcome, and a light lunch and refreshments will be served.
Electronic Health Records Study Design Workshop (EHR-SDW) 2023
Duke AI Health is pleased to announce the Duke Electronic Health Records Study Design Workshop (EHR-SDW) 2023. The workshop will be offered in December as a virtual five-day class (December 4 – 8, 2023) that provides foundational lectures and hands-on studios on the fundamentals of working with and designing EHR based studies.
The EHR-SDW is targeted toward individuals interested in learning about how to work with and conduct studies using electronic health records (EHR) data. EHR data are a widely available form of real-world data that have become standard in studies ranging from clinical trials, comparative effectiveness, risk prediction, and population health. The EHR-SDW will introduce the components of EHR data and introduce considerations for design of effective studies. In addition to didactic lectures, participants will get hands-on experience in working with publicly available tools to facilitate EHR studies (e.g., RxNorm, CCS codes, geocoding) as well as feedback on effective study designs that they will work on.
AI Health Virtual Seminar: AI Designing, Measuring, and Acting on misAlignment
Tuesday, October 24, 2023 | 12:00 – 1:00 PM (Eastern time)
Virtual seminar via Zoom, open to members internal and external to Duke
Presented by:
- Sage Arbor PhD , Sr Informaticist , DCRI
- with host Elliot Hill, AI Health Fellow
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=8d38aee7-46ac-4cf0-85c6-b0a5012f7c93
View the slides here: https://duke.box.com/s/x077ibujr8411u1me517ghzfn5xlsc2j
Description:
Artificial Intelligence (AI) capabilities are increasing faster than AI security and alignment to human values. Unintended detrimental use of AI can fall into two categories: purposeful execution by bad actors (e.g. theft, terrorism), or unintended outcomes without human malicious intent (misalignment). Large Language Models (LLMs , e.g. Chat GPT) are making changes to address these security and alignment issues, but failures already being seen in deployed AIs will be detailed. Steps to detect these safety failures and improve the alignment landscape will be discussed.
Please join us for this lunchtime virtual seminar. The presentation will be accessible to a broad audience, including those with no prior background in health data science or artificial intelligence, but will also delve into technical details in the final third of the talk.
AI Health Virtual Seminar: Introduction to Machine Learning and Artificial Intelligence in Health
Thursday, October 5, 2023 | 4:00 – 5:00 PM (Eastern time)
Virtual seminar via Zoom, open to members internal and external to Duke
Presented by:
- Matthew Engelhard, MD, PhD; Assistant Professor of Biostatistics and Bioinformatics; Director of the Duke AI Health Data Science Fellowship Program
- Shelley Rusincovitch, MMCi; Managing Director, Duke AI Health
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=67c45bbb-dc56-40bf-adea-b0930179457d
You can follow along with the slides here: https://duke.box.com/s/6ze0pw50163chun4ocwp5zcfa6zisiuu
Description:
The transformation in artificial intelligence (AI) and machine learning (ML) is impacting clinical medicine with potential for benefit but also harm. In this one-hour virtual seminar, we’ll introduce AI and ML, highlight capabilities of current methods for images, text, and other big data in clinical settings, and discuss considerations in how to use these powerful methods responsibly. This session will be a high-level introduction accessible to people with no prior knowledge of artificial intelligence.
AI Health Virtual Seminar: 7 Problems Facing Artificial Intelligence for Health (and what’s being done about them)
Thursday, September 28, 2023 from 12:00 PM to 1:00 PM (Eastern time)
Presented by:
- Jonathan McCall, MS, Communications Director, Duke AI Health
- Shelley Rusincovitch, MMCi, Managing Director, Duke AI Health
- with host Matthew Elmore, ThD, AI Ethics & Evaluation Specialist, Duke AI Health
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=a74de10f-3dd8-4a19-bd09-b08b014ef55f
View the slides here: https://duke.box.com/s/sxv22denfckuelzbmd0pwqnxr6g6spci
A link to the complementary reading list can be found here: https://duke.app.box.com/s/ebajl1ycp7jdjwfk8ohb2arvqnbbogbn
Would you like to receive AI Health’s weekly roundup in your inbox? Sign up for the dedicated listserv (just one message per week) at https://lists.duke.edu/sympa/subscribe/aihealthroundup
Description:
Artificial Intelligence has been promoted as the key to solving key problems in healthcare, but the challenges facing us are complex and sometimes intractable – with potential even to bring harm. In this 1-hour seminar, we’ll discuss this landscape at a high level, including a history of unrealistic expectations, overhype that too often permeates this space, serious issues of bias and fairness, and other areas. We’ll discuss illustrations and case studies drawn from the current scientific literature, informed by AI Health’s longstanding Weekly Roundup of news and publications in AI. We’ll also talk about the researchers and approaches that are informing responses to these problems, and the opportunities that exist to actively develop ethical and equitable data science.
This session is free and open to anyone in the world.
Lunch and Learn: Introduction to Large Language Models
Tuesday, September 19, 2023 | 1:00-2:00 PM (Eastern time)
Presented by Larry Carin with introductory remarks by Jeff Ferranti and Michael Pencina
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=166075b1-0903-43db-9930-b085011e6440
Description:
Large language models (LLM) are powering amazing recent innovations in generative AI such as ChatGPT. Although their capabilities may seem like magic, behind these technologies are concepts that anyone can understand.
In this lunch and learn, Larry Carin will provide a math-free, intuitive explanation of how LLMs work. Please join us for an engaging session as Dr. Carin introduces participants to the deep-learning technology that has revolutionized the capacity of machines to perform language translation, to answer questions posed for given text, and to generate (synthesize) text that is near human-generated quality.
The session will be introduced by remarks by Jeff Ferranti, Senior Vice President and Chief Digital Officer for Duke Health, and Michael Pencina, Vice Dean for Data Science and Director of Duke AI Health. They will discuss the potential for LLMs to improve the lives and clinicians and patients, and how Duke Health is approaching this area of strategic importance.
AI Health Virtual Seminar: Real-world Applications of AI Chatbots with ChatGPT
Thursday, September 14, 2023 from 12:00 PM to 1:00 PM (Eastern time)
Presented by:
- Stephen Toback, Media Architect/Senior Producer, Academic Media Engineering & Production, Duke Office of Information Technology
- with host Shelley Rusincovitch, MMCi, Managing Director, Duke AI Health
View the recording here: https://duke.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=4137a817-de1b-49d6-b906-b07d016917af
View the slides here: https://duke.box.com/s/lagky15oat5rjnanxo16gt6bzarug73v
Description:
Join us for an engaging presentation on the transformative power of Generative Artificial Intelligence (AI) in the form of Chatbots. Designed for individuals with little to no experience in using AI Chatbots, this event aims to share valuable insights with a broad community, both at Duke and beyond. Our talk will focus on leveraging ChatGPT, a cutting-edge language model developed by OpenAI, to automate mundane tasks. Stephen Toback, Assistant Director of Academic Media Production and Engineering will showcase real-world applications of ChatGPT, including prompt engineering and refinement, practical implications, and security and privacy considerations. Don’t miss this opportunity to learn about the fascinating world of AI Chatbots and connect with a vibrant community of innovators, researchers, and enthusiasts. And yes, this description was drafted by ChatGPT!
This session is free and open to anyone in the world.
AI Health Spark Seminar Series: Towards solving breast cancer diagnosis with deep learning
Tuesday, June 6, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
Krzysztof Geras, PhD; NYU Grossman School of Medicine and NYU Center for Data Science and at Courant Institute for Mathematical Sciences
with host Maciej Mazurowski, PhD; Associate Professor in Radiology, Duke University
Register here: https://duke.zoom.us/webinar/register/WN_kXC15T_6Rg6_ADxI2cy1Fg
Although deep learning has made stunning progress in the last few years, both in terms of engineering and theory, its real-life applications in medicine remain rather limited. One of the fields that has been anticipated to be revolutionized by deep learning for some time, yet proved to be much harder than many expected, is medical imaging. In this talk I will shed some light on my 7-year long journey in developing deep learning methods for medical imaging, in particular, for breast cancer screening. I will explain how we created a deep learning model that can perform a diagnosis with an accuracy comparable to experienced radiologists. To achieve this goal we needed a lot of perseverance, novel neural network architectures and training methods specific to medical imaging. I will also discuss the limitations of our work and what can likely be achieved in the next few years.
This session is a part of the monthly seminar series organized by Spark: AI Health Initiative for Medical Imaging. The seminar will highlight outstanding work in medical imaging at Duke and beyond. The seminar recordings will be publicly available.
The Spark initiative focuses on development, validation, and clinical implementation of artificial intelligence algorithms for broadly understood medical imaging by bringing together the technical and clinical expertise across Duke campus. For more information please contact Dr. Maciej Mazurowski (maciej.mazurowski@duke.edu).
Challenges in Bring-Your-Own-Device Design-based Digital Health Studies to Develop Reliable and Generalizable Artificial Intelligence Tools for Healthcare Applications
Thursday, May 10, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
- Md Mobashir Hasan Shandhi, PhD; American Heart Association Postdoctoral Fellow, Department of Biomedical Engineering, Duke University
- with host Andrew Olson, MPP; Duke AI Health
Register here: https://duke.zoom.us/webinar/register/WN_roxld4mLSQCxz-WnPEu3CA
Longitudinal digital health studies combine information from digital devices, such as commercial wearable devices, and patient-reported data, such as surveys, from participants. While the ubiquitous adoption of smartphones and access to the internet supports the development of large-scale and distributed digital health studies, there are challenges in collecting representative data as a result of low adherence to, engagement with, and regularity of performing study tasks such as filling out surveys and charging and wearing devices. These challenges may result in a study population that is not representative of the general population or the population group of interest. Artificial Intelligence tools developed based on a non-representative population have a higher chance to fail to generalize in the real-world deployment of such technologies and may not work for underrepresented and underserved communities. In this seminar, the speaker will share his research group’s experience in conducting longitudinal digital health studies for COVID-19 monitoring, the challenges the researchers faced to collect data from a representative population, and how his team developed a guideline to mitigate demographic imbalance in bring-your-own-device (BYOD) design-based digital health studies. Furthermore, the speaker will also share how his team developed a machine learning method based intelligent allocation method for COVID-19 diagnostic testing in a resource-limited setting (when we have limited diagnostic tests, like the earlier phase of the pandemic and onset of new variants) using wearable and survey data collected during the longitudinal CovIdentify study.
Please join us for this lunchtime virtual seminar. The presentation will be accessible to a broad audience, including those with no prior background in health data science or artificial intelligence.
Data bias is the roadblock to realizing the promise of AI in healthcare
Thursday, May 4, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
- Leo Anthony Celi, MD, MPH, MSc; Principal Research Scientist Massachusetts Institute of Technology; Associate Professor (Part Time) Harvard Medical School; Instructor, Harvard T.H. Chan School of Public Health; Associate Program Director, Department of Medicine, Beth Israel Deaconess
- with host Nicoleta Economou, PhD; Duke AI Health
Register here: https://duke.zoom.us/webinar/register/WN_hvcS3mW0QYmfkTrGaY5wOg
The application of artificial intelligence in healthcare requires a team science approach. A diverse set of expertise, perspectives and lived experiences are required to understand the various ways bias lurks in the data – from bias introduced by sampling selection (who made it to the database, who didn’t, and what’s the impact on downstream models), variation in the frequency of measurement that is not explained by the disease or patient phenotype (aka “shortcut” features in medical images), technology that performs differently across patient subgroups (e.g. pulse oximetry, wearable sensors optimized around fit individuals), etc. Data bias is the roadblock to realizing the promise of machine learning. Algorithmic bias is not just about evaluating model performance across patient subgroups post hoc. The goal is to ascertain that the model does not learn from features that should not affect decision making. Offering chemotherapy should not depend on whether a patient is on Medicaid or has a private insurance, predicting job performance should not be informed by the gender of the applicant, optimizing treatment for sepsis should be not be confounded by the use of infrared sensing technology. This is much easier said than done because of the discovery that computers can easily learn sensitive attributes that the human eye does not see. Using real world data to evaluate the models makes this extremely challenging. Excellent model accuracy means existing outcome disparities are fully encoded in the algorithms
Please join us for this lunchtime virtual seminar. The presentation will be accessible to a broad audience, including those with no prior background in health data science or artificial intelligence.
AI Health Spark Seminar Series: A Transparency- and Trust-Centric Design Approach to AI for Healthcare
Tuesday, May 2, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
- Alexander Wong, PhD, PEng, SMIEEE, FIET, Professor, Department of Systems Design Engineering, University of Waterloo
- with host Maciej Mazurowski, PhD; Associate Professor in Radiology, Duke University
Register here: https://duke.zoom.us/webinar/register/WN_WDnYq-mnRjO6ACXsfVOfHw
Artificial Intelligence (AI), particularly thanks to the advances of machine learning (ML) and deep learning (DL) in recent years, holds tremendous promise and potential for enhancing clinical decision support in healthcare. From disease diagnosis to prognosis to treatment planning to patient triaging, AI-driven clinical decision support across the entire clinical workflow can great improve the accuracy, consistency, and speed with which clinicians can better serve their patients for greater quality of care. However, widespread adoption of AI in healthcare has remained limited despite these advances, with one of the biggest challenges being trust in such AI systems. Thankfully, significant progress and advances have been made towards transparency- and trust-centric design, and in this seminar I will be discussing recent developments in quantitative explainable AI, trust quantification, improvements in data, and best practices and ethical considerations for trusted AI, as well as example use uses of how this enables the building of trusted AI solutions for healthcare.
This session is a part of the monthly seminar series organized by Spark: AI Health Initiative for Medical Imaging. The seminar will highlight outstanding work in medical imaging at Duke and beyond. The seminar recordings will be publicly available.
The Spark initiative focuses on development, validation, and clinical implementation of artificial intelligence algorithms for broadly understood medical imaging by bringing together the technical and clinical expertise across Duke campus. For more information please contact Dr. Maciej Mazurowski (maciej.mazurowski@duke.edu).
A Guide to Conducting a Successful Computer Vision Project: Insights from Duke AI Health Fellows
Wednesday, April 26, 2023 | 3:00 – 4:00 PM (Eastern time)
Presented by:
- Irene Tanner, MS
- Akhil Ambekar, MS
- Hyeon Ki Jeong, PhD
- Gavin Karr, MS
- with host Silvana Lawvere de Moreno, PhD
Register here: https://duke.zoom.us/webinar/register/WN_LsXTvWLBQO6FlJR58IYEaA
This seminar provides a practical guide to initiating and executing a computer vision analysis project, with examples from clinical imaging projects in pathology, orthopedics, and dermatology. This topic may be especially pertinent for data scientists, researchers, and practicing clinicians in specialties that use imaging data such as x-rays, whole slide images, and CT scans.
Please join us for this afternoon virtual seminar featuring members of the AI Health Data Science Fellowship Program. The presentation will be accessible to a broad audience, including those with no prior background in health data science or artificial intelligence.
Spring Health Data Science Poster Showcase
Monday, April 24, 2023 | 12:00 – 2:00 PM (Eastern time)
In person at the Trent Semens Center, 6th floor
The Health Data Science poster showcase will be hosted by the Duke AI Health, the Laboratory for Transformative Administration, and the Center for Computational Thinking.
More than 25 posters will be presented, including Duke participants from: AI Health Fellowship Program; Biomedical Engineering; Clinical and Translational Science Institute (CTSI); Computer Science; Department of Biostatistics and Bioinformatics; Department of Internal Medicine; Department of Neurosurgery; Department of Surgery; Division of Geriatrics, Department of Medicine; Division of Hematology, Department of Medicine; Duke Clinical Research Institute (DCRI); Electrical and Computer Engineering (ECE); Duke Health Technology Solutions (DHTS); Laboratory for Transformative Administrative (LTA); Master of Management in Clinical Informatics (MMCi); OB-GYN; and Trinity College of Arts & Sciences.
Poster awards will include Best Computational Thinking Poster and the Good DEEDS Award (Ethical and Equitable Data Science).
What’s important in the ethics and safety of artificial intelligence in health?
Tuesday, April 25, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
- David Carlson, PhD; Assistant Professor of Civil and Environmental Engineering, Assistant Professor in Biostatistics & Bioinformatics, Assistant Professor in the Department of Electrical and Computer Engineering, Assistant Professor of Computer Science
- Jessica Sperling, PhD; Director, Office of Evaluation and Applied Research Partnership (Director, AREE, SSRI; Director, Evaluation & Strategic Planning, CTSI)
- James Tcheng, MD; Professor of Medicine, Assistant Dean for Academic Appointments, Professor of Family Medicine and Community Health
- Shelley Rusincovitch, MMCi; Managing Director, Duke AI Health
Register here: https://duke.zoom.us/webinar/register/WN_wyM9ikvlSrSnSEneQstGGw
The application of artificial intelligence in health settings holds great potential, but there are many considerations and concerns for the safe and responsible integration of these powerful methods. In this one-hour virtual seminar, we’ll discuss these considerations with 3 experts: Dr. Tcheng, who is an interventional cardiologist and clinical informaticist; Dr. Carlson, who is a quantitative and machine learning expert; and Dr. Sperling, who is an expert in evaluation and applied/community-engaged research methods. Through their complementary areas of expertise, we’ll explore perspectives on the current state and future development of ethics and safety in cutting-edge AI methods and applications, with a particular focus on health and clinical settings.
Using Multimodal Data and AI to study Post-Traumatic Epilepsy
Thursday, April 20, 2023 | 4:00 – 5:00 PM (Eastern time)
Presented by:
Dominique Duncan, PhD, USC Stevens Neuroimaging and Informatics Institute, University of Southern California
with host Nicoleta J Economou, PhD; Duke AI Health
Register here: https://duke.zoom.us/webinar/register/WN_SywJvEBmSLKsocQ8MRQE_w
The Epilepsy Bioinformatics Study for Antiepileptogenic Therapy (EpiBioS4Rx) is a multi-site, international collaboration that aims to investigate the multifactorial processes underlying the development of epilepsy after traumatic brain injury (TBI). The study involves a parallel investigation in both humans and animal models and includes the collection of multimodal data such as MRI, EEG, and blood samples. The development of epilepsy after TBI is a complex phenomenon that spans multiple modalities, and a comprehensive understanding of its underlying biological mechanisms is essential for the development of effective treatments. To this end, we have established a centralized data archive that standardizes data and provides tools for searching, viewing, annotating, and analyzing them. This archive includes data generated from multicenter preclinical trials, clinical sites, and various laboratories in different formats. We use machine learning (ML) and artificial intelligence (AI) techniques to analyze large-scale, multimodal datasets and identify complex patterns that may not be apparent through traditional statistical methods. By integrating these cutting-edge analytical approaches with our data archive, we aim to accelerate progress towards the development of effective treatments for post-traumatic epilepsy. In addition to this EpiBioS4Rx archive, we have also developed the Data Archive for the BRAIN Initiative (DABI) and the COVID-19 Data Archive (COVID-ARC) to facilitate research in related areas and foster collaboration among researchers worldwide.
Design and implementation of a novel platform for COVID-19 contact tracing and testing
Wednesday, April 12, 2023 | 4:00 – 5:00 PM (Eastern time)
Presented by:
- Andrew Olson, MPP; Associate Director, Policy Strategy and Solutions for Health Data Science, Duke AI Health
- with host Hyeon Ki Jeong, PhD
Register here: https://duke.zoom.us/webinar/register/WN_oXddPm7bRXSJWzZI4aXDgQ
In July, 2020, just a few months into the COVID-19 pandemic, the CDC funded the Snowball Study to test an approach for more efficiently identifying and contacting people in Durham, NC who may be at risk of infection. The study deployed respondent-driven sampling methods, in which individuals infected with COVID-19 serve as seed cases and recruit their own social contacts for testing and enrollment in the study. To operationalize this approach, we built a cloud-based platform that integrates with the Duke Health electronic health record for case ascertainment and with the Research Electronic Data Capture (REDCap) application for electronic consent and collection of a social contact survey. The platform generates unique coupons that participants validate to enroll in the study and that are used to track all of the links between seeds and multiple waves of peer participants. The Snowball study tested novel methods of contact tracing as well as conducting clinical research and in this seminar we will describe the design of the platform and share results from the study. This presentation will be accessible to a broad audience, including those with no prior background in health data science or artificial intelligence.
Governance practices for fairness and equity: anticipating, identifying, and mitigating algorithmic bias
Thursday, April 6, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
Sophia Bessias, MPH, MSA; Evaluation Lead, Algorithm-Based Clinical Decision Support (ABCDS) Oversight, and Michael Cary, PhD, RN; Duke AI Health Equity Scholar
Register here: https://duke.zoom.us/webinar/register/WN_j9WrGE3tSXOrqgaorKi59A
With the explosion of interest in healthcare applications of AI and machine learning, health systems are increasingly implementing clinical algorithms intended to benefit patient care and reduce costs. Integration with clinical care also brings risks, including the potential for biased algorithms to automate and entrench existing health disparities. In this 1-hour seminar, we will share examples of bias in healthcare algorithms, discuss ways to assess bias potential across the modeling lifecycle, and offer strategies for mitigation. We will also consider how bias assessment fits into the algorithmic governance framework at Duke Health.
Please join us for this lunchtime virtual seminar. The presentation will be accessible to a broad audience, including those with no prior background in health data science or artificial intelligence.
CHAI Blueprint for Trustworthy AI Webinar
Wednesday, April 5, 2023 | 12:00-1:00 PM (Eastern time)
The use of artificial intelligence (AI) in healthcare offers enormous potential for accelerating clinical research and improving quality and delivery. But concerns with some AI-based algorithms have raised questions about their safety, efficacy, and equity.
As part of the Coalition for Health AI (CHAI), Duke AI Health has been helping to build and refine a Blueprint for Trustworthy AI in partnership with member organizations. This blueprint explores the parameters for the guidance, the guardrails, the best practices, and the governance to help ensure ethical AI use in healthcare.
On April 5, 2023, from 12:00-1:00 ET, we will hold a webinar to discuss the Blueprint, how it aligns with other AI frameworks, and our next steps to evolve the Blueprint and encourage wide adoption.
Speakers include:
- Keynote: Alan E. Mislove, Assistant Director for Data and Democracy, White House Office of Science and Technology Policy
- Laura Adams, National Academy of Medicine
- Eric Horvitz, Chief Scientific Officer, Microsoft
- CHAI Steering Committee Members
We hope you will join us to learn how you can apply this Blueprint to your organization and help institutionalize the responsible use of AI in healthcare.
Register online: https://mitre.zoomgov.com/webinar/register/WN_7I8p4jwgS8yJakgr4299SQ
To learn more about CHAI, visit our website: www.coalitionforhealthai.org
AI Health Spark Seminar Series: Driving AI Innovation with Synthetic Data in Longitudinal Imaging, Continual Leaning, and Federated Learning for Healthcare
Tuesday, April 4, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by: Xiaoxiao Li, PhD; Assistant Professor at the University of British Columbia, Assistant Professor Adjunct at Yale University, and Faculty Member at Vector Institute with host Maciej Mazurowski, PhD; Associate Professor in Radiology, Duke University
Register here: https://duke.zoom.us/webinar/register/WN_2AGtDoheSYChmTf3v4F3tg
Description: In this presentation, we explore the power of synthetic data in advancing healthcare AI applications, particularly focusing on longitudinal imaging, continual learning, and federated learning. The talk delves into two data synthesis methods: the diffusion model, a fine-grained approach, and datasets distillation, a coarse approach. First, I will begin by introducing our novel diffusion model-based method for longitudinal image synthesis, which has shown great promise in understanding brain aging through the generation of spatial-temporal data. This innovative technique allows for the efficient analysis of brain development and degeneration over time. Next, I will discuss the use of data distillation to enhance the performance of continual learning and federated learning. These learning paradigms address critical challenges in modern healthcare, such as data privacy constraints. Data distillation helps mitigate the catastrophic forgetting issue in continual learning and tackles the heterogeneity issue in federated learning, enabling more effective and adaptive AI solutions in healthcare. Through the use of synthetic data and cutting-edge AI methodologies, this presentation demonstrates the potential to advance healthcare by providing improved understanding, and diagnosis for patients, fostering collaboration, and ultimately improving patient outcomes.
This session is a part of the monthly seminar series organized by Spark: AI Health Initiative for Medical Imaging. The seminar will highlight outstanding work in medical imaging at Duke and beyond. The seminar recordings will be publicly available.
The Spark initiative focuses on development, validation, and clinical implementation of artificial intelligence algorithms for broadly understood medical imaging by bringing together the technical and clinical expertise across Duke campus. For more information please contact Dr. Maciej Mazurowski (maciej.mazurowski@duke.edu).
What are informatics and data science, and why do they matter in health?
Tuesday, March 28, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
- W. Edward Hammond, PhD; Professor in Family Medicine and Community Health, Research Professor in the School of Nursing, Professor of Biostatistics and Bioinformatics
- Warren Kibbe, PhD; Professor in Biostatistics & Bioinformatics
- Shelley Rusincovitch, MMCi; Managing Director, Duke AI Health
Register here: https://duke.zoom.us/webinar/register/WN_j9WrGE3tSXOrqgaorKi59A
The disciplines of informatics and data science are critical to the use of technology in clinical research and healthcare. In this one-hour virtual seminar, we’ll discuss the foundations of these fields with 2 experts: Dr. Hammond, who is one of the great pioneers in medical informatics and a founding member of HL7®; and Dr. Kibbe, who is chief for Translational Biomedical Informatics in the Department of Biostatistics and Bioinformatics and Chief Data Officer for the Duke Cancer Institute. Our conversation will include what the COVID-19 global pandemic has taught us about challenges and opportunities in data infrastructure, a debate on the future and direction of the HL7® FHIR® standard, and a broad-ranging perspective into the evolution and future of informatics and data science.
Trailblazing a path to simpler, more transparent and equitable risk algorithms
Thursday, March 23, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
- Michael Pencina, PhD; Professor of Biostatistics and Bioinformatics; Director of Duke AI Health; Vice Dean of Data Science, Duke University School of Medicine
- Chuan Hong, PhD; Assistant Professor of Biostatistics & Bioinformatics
- with host Marcella Dalrymple, MPH, MBA; Duke AI Health
Register here: https://duke.zoom.us/webinar/register/WN_unts9Z2rR2-sFt09NWAIfg
Join us for a live virtual seminar with Dr. Pencina and Dr. Hong, who will discuss the need for a careful evaluation of risk prediction algorithms to avoid bias and propagation of health inequities. The problem is illustrated in the most recent publication by AI Health and collaborators, “Predictive Accuracy of Stroke Risk Prediction Models Across Black and White Race, Sex, and Age Groups,” demonstrating how currently proposed risk prediction algorithms perform markedly worse in Black individuals and are not improved when employing advanced machine learning techniques. Our expert presenters will discuss the important implications of these findings, as well as an overview of the methods, implementation, and opportunities for improvement in risk prediction. Register now to connect with Duke AI Health, and join us as we work to redefine algorithmic standards for clinical practice.
Neuroimage Analysis in Autism: from Model-Based Estimation to Data-driven Learning
Tuesday March 7th, 2023 | 12:00-1:00 PM (Eastern time)
Presented by:
- James S. Duncan, Ph.D., Ebenezer K. Hunt Professor and Department Chair of Biomedical Engineering, Professor of Radiology & Biomedical Imaging Professor of Electrical Engineering, Professor of Statistics & Data Science; Yale University
- with host Maciej Mazurowski, PhD; Associate Professor in Radiology, Duke University
Register here: https://duke.zoom.us/webinar/register/WN_NmXBWQZURnaFzbwkjkfWvg
Functional magnetic resonance imaging (fMRI) has been shown to be helpful for the study of autism spectrum disorders (ASD). This talk will describe the evolution of efforts in this area within our group that carry promise for producing objective biomarkers for ASD, as well as predicting patient response to a behavioral therapy known as Pivotal Response Treatment (PRT), using task-based fMRI. Such biomarkers would provide an important step for better understanding the underlying pathophysiology of ASD that could help with objective and personalized diagnosis, provide new targets for development of new treatments, and provide a way to monitor patient progress. Initially a robust, group-wise unified Bayesian framework to detect both hyper and hypo-active communities from connectivity maps will be described. Next, more recent work will be presented that has focused on deriving ASD biomarkers from individual subject’s time-series data, based on the classification of individual subjects (into ASD or typical control) and identifying spatially-specific key regions using convolutional neural networks and ablation analysis of regions. In addition, a strategy based on recurrent neural networks (using long-short-term memories or LSTMs) will be presented that predicts patient response to PRT behavioral therapy from baseline imaging while incorporating subject-specific phenotypic information for network initialization. Finally, early work on the use of effective connectivity based on whole brain dynamic causal modeling will be discussed as an alternative or an adjunct to functional connectivity for classification and biomarker analysis.
This session is a part of the monthly seminar series organized by Spark: AI Health Initiative for Medical Imaging. The seminar series highlights outstanding work in medical imaging at Duke and beyond.
What are machine learning and artificial intelligence, and why do they matter in health?
Tuesday, February 28, 2023 | 12:00 – 1:00 PM (Eastern time)
Presented by:
- Matthew Engelhard, MD, PhD; Assistant Professor of Biostatistics and Bioinformatics
- Shelley Rusincovitch, MMCi; Managing Director, Duke AI Health
Register here: https://duke.zoom.us/webinar/register/WN_oXcW126lRVaQHU9tCBb9BQ
The transformation in artificial intelligence (AI) and machine learning (ML) is impacting clinical medicine with potential for benefit but also harm. In this one-hour virtual seminar, we’ll introduce AI and ML, highlight capabilities of current methods for images, text, and other big data in clinical settings, and discuss considerations in how to use these powerful methods responsibly. This session will be a high-level introduction accessible to people with no prior knowledge of artificial intelligence.
Algorithm-Based Clinical Decision Support (ABCDS) Oversight: A framework for the governance and evaluation of algorithms to be deployed at Duke Health
Tuesday, February 14, 2023 12:00-1:00 PM (Eastern time)
Presented by:
- Armando Bedoya, MD, MMCi; Duke Health Technology Solutions
- Nicoleta J Economou, PhD; Duke AI Health
In 2021, Duke Health launched the Algorithm-Based Clinical Decision Support (ABCDS) Oversight program, a collaborative effort between the Duke University School of Medicine and the Duke University Health System to ensure high-quality care, patient safety, and ownership are maintained for algorithms and related tools. In this session, Drs. Economou & Bedoya will introduce you to the ABCDS Oversight framework. This session is open to members outside of Duke.
