Biases in machine learning models for medical imaging and sociotechnical impacts
Artificial intelligence (AI) plays an increasingly important role in transforming large volumes of data into actionable insights, and healthcare is no exception. Advances in diagnostic technologies have led to a rapid growth in the volume and complexity of medical data. Medical imaging data are among the most information-rich clinical sources. However, their scale and complexity, especially when combined with other clinical variables, can overwhelm clinicians and hinder accurate, timely interpretation. Machine learning (ML) methods offer powerful tools to support clinical decision-making, reduce cognitive burden, and enable clinicians to devote more attention to complex cases and patient interaction, thereby supporting precision and equity in healthcare.
Despite this potential, clinical adoption of ML in radiology remains limited. A key barrier is increasing evidence that ML models for medical image analysis can exhibit biased or discriminatory behavior, especially in computer‑aided diagnosis. Although these concerns are widely acknowledged, systematic research on how biases embedded in imaging data influence model behavior is still lacking. Moreover, ML is not a value‑neutral technology; it is shaped by data, deployment contexts, and societal norms.
This presentation presents recent work on identifying and analyzing bias in medical imaging and ML models. It introduces a synthetic data simulation framework for controlled evaluation of imaging biases and mitigation strategies, examines how neural networks encode bias and shortcut learning, and presents methods for uncovering unfair behavior. Finally, it emphasizes the need to link technical fairness metrics with broader sociotechnical impacts to develop equitable and trustworthy medical AI.
Short Bio
Dr. Nils Daniel Forkert is a Professor at the University of Calgary in the Departments of Radiology and Clinical Neurosciences. He received his German diploma in Computer Science in 2009 from the University of Hamburg, his master’s degree in medical physics in 2012 from the Technical University of Kaiserslautern, his PhD in computer science in 2013 from the University of Hamburg, and completed a postdoctoral fellowship at Stanford University before joining the faculty of the University of Calgary in 2014. He is an imaging and machine learning scientist who develops new image processing methods, predictive algorithms, and software tools for the analysis of medical data. This includes the extraction of clinically relevant parameters and biomarkers from medical data describing the morphology and function of organs with the aim of supporting clinical studies and preclinical research as well as developing computer-aided diagnosis and patient-specific, precision-medicine, prediction models using machine learning based on multi-modal medical data. Dr. Forkert is a Canada Research Chair (Tier 2) in Medical Image Analysis, and Director of the Child Health Data Science Program of the Alberta Children’s Hospital Research Institute as well as the Theme Lead for Machine Learning in Neuroscience of the Hotchkiss Brain Institute at the University of Calgary. He has published over 230 peer-reviewed manuscripts, over 100 full-length proceedings papers, 1 book, and 2 book chapters. He has received major funding from the Canadian Institutes of Health Research (CIHR), Natural Sciences and Engineering Research Council, the Heart and Stroke Foundation, Calgary Foundation, and the National Institutes of Health as a PI or co-PI.