Fairness in Cardiac Magnetic Resonance Imaging: Assessing Sex and Racial Bias in Deep Learning-Based Segmentation

Esther Puyol-Antón; Bram Ruijsink; Jorge Mariscal Harana; Stefan K Piechnik; Stefan Neubauer; Steffen E Petersen; Reza Razavi; Phil Chowienczyk; Andrew P King

doi:10.3389/fcvm.2022.859310

Fairness in Cardiac Magnetic Resonance Imaging: Assessing Sex and Racial Bias in Deep Learning-Based Segmentation

Front Cardiovasc Med. 2022 Apr 7:9:859310. doi: 10.3389/fcvm.2022.859310. eCollection 2022.

Authors

Esther Puyol-Antón¹, Bram Ruijsink^{1

2

3}, Jorge Mariscal Harana¹, Stefan K Piechnik⁴, Stefan Neubauer⁴, Steffen E Petersen^{5

6

7

8}, Reza Razavi^{1

2}, Phil Chowienczyk^{1

9}, Andrew P King¹

Affiliations

¹ School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom.
² Department of Adult and Paediatric Cardiology, Guy's and St Thomas' NHS Foundation Trust, London, United Kingdom.
³ Division of Heart and Lungs, Department of Cardiology, University Medical Centre Utrecht, Utrecht, Netherlands.
⁴ Division of Cardiovascular Medicine, Radcliffe Department of Medicine, University of Oxford, Oxford, United Kingdom.
⁵ National Institute for Health Research (NIHR) Barts Biomedical Research Centre, William Harvey Research Institute, Queen Mary University London, London, United Kingdom.
⁶ Barts Heart Centre, St Bartholomew's Hospital, Barts Health NHS Trust, London, United Kingdom.
⁷ Health Data Research UK, London, United Kingdom.
⁸ Alan Turing Institute, London, United Kingdom.
⁹ British Heart Foundation Centre, King's College London, London, United Kingdom.

Abstract

Background: Artificial intelligence (AI) techniques have been proposed for automation of cine CMR segmentation for functional quantification. However, in other applications AI models have been shown to have potential for sex and/or racial bias. The objective of this paper is to perform the first analysis of sex/racial bias in AI-based cine CMR segmentation using a large-scale database.

Methods: A state-of-the-art deep learning (DL) model was used for automatic segmentation of both ventricles and the myocardium from cine short-axis CMR. The dataset consisted of end-diastole and end-systole short-axis cine CMR images of 5,903 subjects from the UK Biobank database (61.5 ± 7.1 years, 52% male, 81% white). To assess sex and racial bias, we compared Dice scores and errors in measurements of biventricular volumes and function between patients grouped by race and sex. To investigate whether segmentation bias could be explained by potential confounders, a multivariate linear regression and ANCOVA were performed.

Results: Results on the overall population showed an excellent agreement between the manual and automatic segmentations. We found statistically significant differences in Dice scores between races (white ∼94% vs. minority ethnic groups 86-89%) as well as in absolute/relative errors in volumetric and functional measures, showing that the AI model was biased against minority racial groups, even after correction for possible confounders. The results of a multivariate linear regression analysis showed that no covariate could explain the Dice score bias between racial groups. However, for the Mixed and Black race groups, sex showed a weak positive association with the Dice score. The results of an ANCOVA analysis showed that race was the main factor that can explain the overall difference in Dice scores between racial groups.

Conclusion: We have shown that racial bias can exist in DL-based cine CMR segmentation models when training with a database that is sex-balanced but not race-balanced such as the UK Biobank.

Keywords: cardiac magnetic resonance; deep learning; fair AI; inequality fairness in deep learning-based CMR segmentation; segmentation.

Abstract

Grants and funding