Professor, School of Health Information Science
Director, Global Laboratory for Digital Health Innovation
University of Victoria, Canada
Director, Global Laboratory for Digital Health Innovation
University of Victoria, Canada
TITLE: The Safety of AI in Healthcare: Human and Healthcare Organizational Considerations
ABSTRACT: Managing health technology risk is an important part of technology design, implementation and evaluation in a modern digital, healthcare system. Digital systems of care have the ability to improve care processes while at the same time introducing new types of errors that have significant impacts for human stakeholders. With the exponential rise in the use of AI applications and systems in varying health settings (i.e., hospital, community, and home), there is an even greater need to ensure our digital health ecosystems of care are safe. This keynote will review the evolution of the field of technology safety. Some of the issues and considerations associated with health technology risk management across the software development lifecycle will be discussed. In addition, methods will be presented that can be used to assess and ensure safety in health environments as part of a healthcare organizational strategy for using and improving AI.
BIO: Elizabeth Borycki is a Professor in the School of Health Information Science at the University of Victoria, and she leads the Global Laboratory for Digital Health Innovation. Elizabeth has a background in the public and private sectors focused on health technology design, implementation and evaluation. Elizabeth’s research focuses on safety in the field of health informatics and digital health.
ABSTRACT: Managing health technology risk is an important part of technology design, implementation and evaluation in a modern digital, healthcare system. Digital systems of care have the ability to improve care processes while at the same time introducing new types of errors that have significant impacts for human stakeholders. With the exponential rise in the use of AI applications and systems in varying health settings (i.e., hospital, community, and home), there is an even greater need to ensure our digital health ecosystems of care are safe. This keynote will review the evolution of the field of technology safety. Some of the issues and considerations associated with health technology risk management across the software development lifecycle will be discussed. In addition, methods will be presented that can be used to assess and ensure safety in health environments as part of a healthcare organizational strategy for using and improving AI.
BIO: Elizabeth Borycki is a Professor in the School of Health Information Science at the University of Victoria, and she leads the Global Laboratory for Digital Health Innovation. Elizabeth has a background in the public and private sectors focused on health technology design, implementation and evaluation. Elizabeth’s research focuses on safety in the field of health informatics and digital health.
Thomas Kannampallil, PhD
Associate Professor of Anesthesiology & Computer Science
Director for Acute Care Innovation Research
Washington University School of Medicine
Director for Acute Care Innovation Research
Washington University School of Medicine
TITLE: Implementing Real-time Clinical Decision Support for Perioperative Care
ABSTRACT: In this talk, I will describe the design, development and implementation of machine learning-based clinical decision support for perioperative settings. Specifically, I will discuss a range of implementation projects (and associated pragmatic clinical trials) for remote intraoperative telemedicine, predicting surgical duration, and predicting surgical transfusion risk. The focus will primarily be on characterizing the current challenges of EHR-based implementations including available data pipelines, data issues, performance drift, and considerations for running pragmatic trials at the point-of-care.
BIO: I am an Associate Professor of Anesthesiology at the Washington University School of Medicine. I also have affiliate appointments in the Department of Computer Science and Engineering and the Institute for Informatics. My research interests lie at the intersection of computer science, cognitive science, and clinical informatics. Specifically, my research focuses on developing and evaluating intelligent computational tools for improving clinical decision making and patient safety. My research is currently funded by 3 R01s from the National Library of Medicine (NLM), National Institute of Aging (NIA), and the Agency for Healthcare Research and Quality (AHRQ). I also serve as an Associate Editor for the Journal of Biomedical Informatics. I was elected as a Fellow of the American Medical Informatics Association in 2021.
ABSTRACT: In this talk, I will describe the design, development and implementation of machine learning-based clinical decision support for perioperative settings. Specifically, I will discuss a range of implementation projects (and associated pragmatic clinical trials) for remote intraoperative telemedicine, predicting surgical duration, and predicting surgical transfusion risk. The focus will primarily be on characterizing the current challenges of EHR-based implementations including available data pipelines, data issues, performance drift, and considerations for running pragmatic trials at the point-of-care.
BIO: I am an Associate Professor of Anesthesiology at the Washington University School of Medicine. I also have affiliate appointments in the Department of Computer Science and Engineering and the Institute for Informatics. My research interests lie at the intersection of computer science, cognitive science, and clinical informatics. Specifically, my research focuses on developing and evaluating intelligent computational tools for improving clinical decision making and patient safety. My research is currently funded by 3 R01s from the National Library of Medicine (NLM), National Institute of Aging (NIA), and the Agency for Healthcare Research and Quality (AHRQ). I also serve as an Associate Editor for the Journal of Biomedical Informatics. I was elected as a Fellow of the American Medical Informatics Association in 2021.
Senior AI Research Scientist, Nokia Bell Labs
Visiting Researcher, University of Cambridge
Visiting Researcher, University of Cambridge
TITLE: Multimodal AI for Real-World Signals and the Role of Language
ABSTRACT: The limited availability of labels for machine learning on multimodal data hampers progress in the field. In this talk, I will discuss our recent efforts to address that, building on the paradigm of self-supervised multimodal learning. With models such as CroSSL, Step2Heart, and SelfHAR, we put forward principled ways to learn generalizable representations from high-resolution physiological and behavioural signals and show how these models can be applied to various high-stakes tasks in health and wellbeing. At the same time, due to data constraints, these models are limited in size and generalization capabilities compared to popular generative models such as GPT. What if we could use Large Language Models (LLMs) as data-agnostic pre-trained models? I will close the talk by highlighting LLMs' challenges in processing signals like text and some ideas on how to address this critical “modality gap”.
BIO: Dimitris Spathis, PhD is a senior research scientist at Nokia Bell Labs and a visiting researcher at the University of Cambridge, where he completed his doctoral degree. His research enables machine learning to handle complex real-world data efficiently, with a particular interest in health sensing. His work studies various topics in AI including data-efficiency, multimodality, model robustness/fairness, and signal processing. He has previously worked at Microsoft Research, Telefonica, and Ocado. In 2020, he helped start one of the largest studies in audio AI for health (covid-19-sounds.org). His research has been featured in international media outlets such as the BBC, CNN, Guardian, Washington Post, Forbes, and Financial Times. He serves on the program committees of top AI conferences such as AAAI , IJCAI, and KDD, and the editorial board of Nature Digital Medicine.
ABSTRACT: The limited availability of labels for machine learning on multimodal data hampers progress in the field. In this talk, I will discuss our recent efforts to address that, building on the paradigm of self-supervised multimodal learning. With models such as CroSSL, Step2Heart, and SelfHAR, we put forward principled ways to learn generalizable representations from high-resolution physiological and behavioural signals and show how these models can be applied to various high-stakes tasks in health and wellbeing. At the same time, due to data constraints, these models are limited in size and generalization capabilities compared to popular generative models such as GPT. What if we could use Large Language Models (LLMs) as data-agnostic pre-trained models? I will close the talk by highlighting LLMs' challenges in processing signals like text and some ideas on how to address this critical “modality gap”.
BIO: Dimitris Spathis, PhD is a senior research scientist at Nokia Bell Labs and a visiting researcher at the University of Cambridge, where he completed his doctoral degree. His research enables machine learning to handle complex real-world data efficiently, with a particular interest in health sensing. His work studies various topics in AI including data-efficiency, multimodality, model robustness/fairness, and signal processing. He has previously worked at Microsoft Research, Telefonica, and Ocado. In 2020, he helped start one of the largest studies in audio AI for health (covid-19-sounds.org). His research has been featured in international media outlets such as the BBC, CNN, Guardian, Washington Post, Forbes, and Financial Times. He serves on the program committees of top AI conferences such as AAAI , IJCAI, and KDD, and the editorial board of Nature Digital Medicine.