Machine Learning and Bias in Medical Imaging: Opportunities and Challenges

Abstract

Bias in health care has been well documented and results in disparate and worsened outcomes for at-risk groups. Medical imaging plays a critical role in facilitating patient diagnoses but involves multiple sources of bias including factors related to access to imaging modalities, acquisition of images, and assessment (ie, interpretation) of imaging data. Machine learning (ML) applied to diagnostic imaging has demonstrated the potential to improve the quality of imaging-based diagnosis and the precision of measuring imaging-based traits. Algorithms can leverage subtle information not visible to the human eye to detect underdiagnosed conditions or derive new disease phenotypes by linking imaging features with clinical outcomes, all while mitigating cognitive bias in interpretation. Importantly, however, the application of ML to diagnostic imaging has the potential to either reduce or propagate bias. Understanding the potential gain as well as the potential risks requires an understanding of how and what ML models learn. Common risks of propagating bias can arise from unbalanced training, suboptimal architecture design or selection, and uneven application of models. Notwithstanding these risks, ML may yet be applied to improve gain from imaging across all 3A's (access, acquisition, and assessment) for all patients. In this review, we present a framework for understanding the balance of opportunities and challenges for minimizing bias in medical imaging, how ML may improve current approaches to imaging, and what specific design considerations should be made as part of efforts to maximize the quality of health care for all.

Keywords: artificial intelligence; bias; diagnostic imaging; health equity; machine learning.

Figure 1.. Opportunities for Bias in Clinical Workflows.

In addition to the multiple opportunities for bias that are recognized to exist throughout the clinical workflow, from initial patient presentation to decision making and intervention, there are additional and often overlooked opportunities for bias to be introduced as well as mitigated in the process of obtaining medical imaging.

Figure 2.. Imbalanced training data can lead to biased models.

Although it is unclear whether any under-representation in training data or under-representation relative to the population proportion of a demographic group is a driver, imbalanced training datasets (as shown in Model A) have led to models that perform worse for select population subgroups compared to models trained where demographic groups are better balanced (Model B). For this reason, investigating bias in model predictions and imbalance in training data is key to creating equitable models.

Figure 3.. Challenges and Opportunities Inherent to Machine Learning Applications in Imaging.

Machine learning can both mitigate (Section 1) and propagate (Section 2) bias in medical imaging. With appropriate care, providers can create models that are not only clinically useful but that can also be equitable (Section 3).