Learning Unbiased Image Segmentation: A Case Study with Plain Knee Radiographs

Ahmad P. Tafti; Nickolas Littlefield; Johannes F. Plate; Kurt R. Weiss; Ines Lohse; Avani Chhabra; Ismaeel A. Siddiqui; Zoe Menezes; George Mastorakos; Sakshi Mehul Thakar; Mehrnaz Abedian; Matthew F. Gong; Luke A. Carlson; Hamidreza Moradi; Soheyla Amirian

Abstract

Automatic segmentation of knee bony anatomy is essential in orthopedics, and it has been around for several years in both pre-operative and post-operative settings. While deep learning algorithms have demonstrated exceptional performance in medical image analysis, the assessment of fairness and potential biases within these models remains limited. Our study revisited implementation of deep learning in knee bony anatomy segmentation of plain radiographs to uncover gender and racial biases and implement strategies for bias mitigation. Within the multiple models we implemented, we found that different bias mitigation strategies present a compromise between fairness and accuracy of predicted knee anatomy segmentation. Optimizing a deep learning model that can fairly account for racial and gender bias in interpretation of knee plain radiographs has significant implications in several areas. The proposed mitigation strategies mitigate gender and racial biases, ensuring fair and unbiased segmentation results. We found that racial and gender bias can be present in knee bony anatomy segmentation models, but that bias mitigation strategies can be effectively implemented. The current contribution offers the potential to advance our understanding of biases, and it provides practical insights for researchers and practitioners in medical imaging. Furthermore, this work promotes equal access to accurate diagnoses and treatment outcomes for diverse patient populations, fostering equitable and inclusive healthcare provision.

Keywords: Image Segmentation; AI Fairness; Knee Radiographs; Unbiased Image Segmentation; Safe AI

Links
[Full text PDF][Bibtex]