Trainee Rounds: Kristina Kupferschmidt and Min Li Chen
August 2, 2023, 12:00 pm to 1:00 pm
Hear from two UofT trainees researching new ways to use artificial intelligence in medicine.
PhD student Kristina Kupferschmidt from the University of Guelph School of Engineering and Vector Institute as well as MSc in Medical Biophysics student Min Li Chen from the University of Toronto Temerty Faculty of Medicine will be presenting their research.
Kristina Kupferschmidt, PhD Student
School of Engineering, University of Guelph
TITLE: Back to the building blocks: Designing human-centric-AI for health through data-centric practices
ABSTRACT: AI's potential in controlled settings has been well-established, but its application in real-world clinical workflows remains challenging. Limited by training data, AI systems struggle to generalize well beyond examples. Concerns like algorithmic trust, interpretability, and biased predictions hinder real-world adoption. Human-centered AI (HCAI) addresses these issues by involving end-users in design and evaluation, tailoring models to specific needs. While efforts have focused on interpretability and fairness evaluation, our research explores training data composition, providing insight on model workings and limitations. By emphasizing data-centric approaches and providing context to clinicians, we aim to build trust and improve understanding of model predictions for better decision-making in healthcare.
Min Li Chen, MSc Student
Temerty Faculty of Medicine, University of Toronto
TITLE: Utilization of unsupervised image feature-based clustering to scale classifier design in histopathology
ABSTRACT: Histopathological analysis of patient tissue is a powerful clinical tool for diagnosis and study of human disease but is challenged by inter-subjective variation and need for sub-specialized pathologists. While there is much excitement surrounding use of artificial intelligence (AI) to automate and increase objectivity of microscopic examination, current use cases represent largely proof-of-concepts limiting widespread adoption. To address this, I describe a prototype AI workflow that empowers microscopists to design and share tissue classifiers without need for any sophisticated coding or complex collaborations. Scaling this solution through crowdsourcing could allow for large-scale development of custom classifiers across all relevant histopathology.
WEDNESDAY, AUGUST 2nd, 2023
12:00-1:00pm ET
ZOOM