Sandra Baum
Winner of travel grant of DKK 30.000.
Introducing the human element: Using the Universal Design paradigm to mitigate Selection Bias in Medical AI datasets
Selection bias poses a significant threat to the effectiveness and fairness of AI models, potentially leading to misdiagnoses and delayed treatments for underrepresented groups, such as individuals with medical disabilities. To address this problem, there is an urgent need for research and innovation in developing more inclusive algorithms and sampling techniques, with the goal of ensuring that medical AI systems serve all members of society.
Sandra Baum plans to investigate this problem by 1) Investigating causes, impact and possible ways to address selection bias by reviewing current academic literature. 2) Evaluating issues in the subject area against the goals of Universal Design. 3) Synthesize a practical framework that matches issues to principles of Universal Design. 4) Applying Universal Design principles to help mitigate selection bias in AI enabled medical devices.
Sandra plans to conduct her investigation at University of Maryland in the summer of 2024.
Watch the pitch below.
Follow
Sandra Baum
's investigation
Sandra has started her investigation in USA. Below you can read posts along the way from her journey where she describes the experiences she has had and what she has learned so far.
April 2024
How are underrepresented groups, such as people with disabilities represented in medical data?
Scholarship Travel grant winner Sandra Baum is currently visiting University of Maryland, she is pictured here with her host Professor Paul H Yi.
Sandra reached out to Professor Yi because of his impressive work on the topic of selection bias. The Professors medical background also makes him ideal as a mentor in her investigation of how to reduce bias in AI models used in healthcare, for a more accessible and equitable world for all.
Right now Sandra is investigating how different historically underrepresented groups, such as people with disabilities are represented in the data. This involves investigating various medical information datasets to see what kind of information is available on the participants. Sandra is analyzing large sets of data, using tools such as web scraping as well as studying literature on people with disabilities in medical datasets.
May 2024
June 2024
Sandra Baum: We need to integrate Universal Design more broadly
Marginalized groups are challenged by selection bias in medical AI systems, as it can lead to misdiagnoses. By visiting the University of Maryland, IT student Sandra Baum investigated how to address this problem.
Main findings
“My project draws attention to the need to integrate Universal Design more broadly. Not only would it be mentioned in policy documents, but it would also be included in research where it could affect the actual practices of creating and evaluating models in the medical field.
Additionally, my project shows that voluntary disclosure programs, including transparency on the groups used to train and evaluate AI algorithms, can be effective. Furthermore, non-technical approaches can be effective in mitigating bias.”
The use of Universal Design
“My project focused on showing how Universal Design as a paradigm can be instrumentalized to reintroduce the human focus into otherwise impersonal, abstracted, and profit-driven systems that shape our way of life in less visible ways.
More practically, this will benefit and protect the health and safety of many data-underrepresented groups, such as older people not being included in training data for a device meant to predict the risk of heart disease. This shows the real-world challenges caused by short-sighted design approaches.”
Why others should apply
“The program is uniquely flexible. People can choose their knowledge space and country freely, and the choice of project topics is also free, as long as it focuses on Universal Design.”