Racial Bias Mitigation in Deep-Learning Assisted Prediction of Colorectal Cancer Onset

Principal Investigator

Name
Archana Gurudu

Institution
Paradise Valley High School

Position Title
High School Student Researcher

Email
argurudu14@gmail.com

About this CDAS Project

Study
PLCO (Learn more about this study)

Project ID
PLCO-1810

Initial CDAS Request Approval
Feb 6, 2025

Title
Racial Bias Mitigation in Deep-Learning Assisted Prediction of Colorectal Cancer Onset

Summary
A lack of racial heterogeneity in clinical data, from medical images to EHR data, can induce racial bias in machine learning models, in which the machine learning task is outperformed on majority groups in comparison to minority ethnicities. Modern diseases with existing systemic racial disparities remain of the greatest concern in relation to racially-biased machine learning models, as the large-scale implementation of these models may exacerbate existing ethnic discrimination in healthcare. Among these diseases are colorectal cancer, a disease known for having higher incidence and mortality rates in minority races, including African Americans and Native Americans. This study aims to compare different bias mitigation strategies to reduce racial bias in deep-learning assisted prediction of colorectal cancer onset across diverse racial groups.

Aims

- Determine whether racial bias is a guaranteed outcome of racial underrepresentation in training data
- Compare different bias mitigation strategies to reduce racial bias in deep-learning assisted prediction of colorectal cancer onset across diverse racial groups

Collaborators

None