The AI Experts Focus Sessions cover frequently-requested topics to enhance knowledge and insights into AI.
AI Learning & Resources
Whether you are an AI Expert or not, you can still view the recorded focus session portion from the meetings.
In February of 2022, the efforts of Xavier Health were assumed by the AFDO/RAPS Healthcare Products Collaborative. Because of the important work done before this transition, the Collaborative has chosen to retain documents that have Xavier branding and continue to provide them to the communities through this website. If you have questions, please contact Timothy Hsu, Director of Health Technology Initiatives, at thsu@healthcareproducts.org.
December 3, 2021: Issues in Reference Standard Determination for Performance Evaluation Studies of AI/ML-based Medical Devices
Featuring Alexej Gossmann, Staff Fellow at the Division of Imaging, Diagnostics, and Software Reliability, FDA-CDRH
Evaluation of an AI/ML-based medical device is typically performed by comparing the system output to a reference standard. A poorly chosen reference standard that is a poor proxy to the targeted real-world clinical task may result in the measurement and reporting of misleading performance results for the AI/ML algorithm, which do not adequately represent the real-world performance or utility of this algorithm. This is particularly concerning when it applies only to specific socio-economically disadvantaged groups or racial minorities.
A more appropriately chosen reference standard can still be subject to several distinct types of bias that can lead to systematic flaws in algorithm training and testing. This includes, for instance, verification bias, measurement bias, and imperfect reference bias, which need to be assessed and accounted for when evaluating the performance of medical AI/ML algorithms. In addition, the reference standard can be partially missing, which can be accounted for with appropriate statistical analysis methods in some settings. This presentation provides a discussion of these and other reference standard issues in the assessment of AI/ML-based medical devices.
October 8, 2021: Incorporating AI Techniques for Medical Imaging
Featuring Lars Bielak, a researcher at the University Medical Center in Freiburg, Germany
Tumor segmentation, or segmentation using CNNs in general, is a well-developed field in AI research. Unfortunately, the data acquisition and the data post-processing are often two completely separate topics, leaving computer scientists to refine their models based only on large, publicly available, and fixed datasets. In this session, Lars Bielak presents a strategy for prospective input data optimization for CNN head and neck tumor segmentation based on seven unique MRI channels.
August 6, 2021: Pharma Manufacturing – When Complexity Cannot be Linearly Managed or Solved via a Singular AI Model
Featuring Toni Manzano, Co-Founder and CSO, Aizon
Toni Manzano presents two use cases, one of AI application in biopharmaceutical operations and one of biotech research in which Xavier AI was involved, to explain the hard reality behind amazing AI outputs.
June 25, 2021: Developing AI-Enabled Smart Medical Devices Using Model-Based Design
Featuring Kirthi Devleker, Mathworks, and Akhilesh Mishra, Mathworks
In this talk, Kirthi Devleker and Akhilesh Mishra provide an overview of how AI algorithms can be developed and deployed in Medical Devices using Model Based Design approach – an approach adopted by large medical device organizations globally to develop next generation medical devices.
April 16, 2021: GxP AI Workflow: An NLP Case Study
Featuring Erez Kaminski, Researcher/Graduate Student, MIT
In this second AI Learning Session covering the topics most frequently requested by our AI Experts Network, Erez Kaminski provides an overview of a systematic way to develop GxP artificial intelligence applications focused on code reusability and streamlined development operations. The talk includes a case study on the development of a Natural Language Processing application for adverse event detection, which is used to reduce processing costs and intake error rates.
February 5, 2021: Cybersecurity Compliance and AI
Featuring Emily Luvison, Cybersecurity Compliance Lead for Digital Health Technologies, Genentech
In this inaugural AI Learning Session, Emily Luvison (Genentech) will be providing an overview of cybersecurity considerations as they apply to AI.