Investigating Gender Bias in Lymph-node Segmentation with Anatomical Priors

TL;DR

This paper addresses the challenge of accurate and fair automated segmentation of the Clinical Target Volume (CTV) in radiotherapy, where gender bias has been observed in deep-learning models. The authors compare multiple Anatomical Prior encoding strategies—applied as additional input channels to an nnU-Net—using 45 full-body CT scans and region-based fairness analyses (AGD, MGD, QD) along with DSC and HD metrics. They report that incorporating Anatomical Priors improves segmentation quality for female patients and reduces gender disparities, particularly in the abdomen (ABDM) region, with MI-Z and MI-TS showing notable improvements. The work demonstrates a practical path to fairer, faster auto-contouring and motivates validation on larger datasets and with clinician-guided AP selection, as well as exploring other biases beyond gender.

Abstract

Radiotherapy requires precise segmentation of organs at risk (OARs) and of the Clinical Target Volume (CTV) to maximize treatment efficacy and minimize toxicity. While deep learning (DL) has significantly advanced automatic contouring, complex targets like CTVs remain challenging. This study explores the use of simpler, well-segmented structures (e.g., OARs) as Anatomical Prior (AP) information to improve CTV segmentation. We investigate gender bias in segmentation models and the mitigation effect of the prior information. Findings indicate that incorporating prior knowledge with the discussed strategies enhances segmentation quality in female patients and reduces gender bias, particularly in the abdomen region. This research provides a comparative analysis of new encoding strategies and highlights the potential of using AP to achieve fairer segmentation outcomes.

Figures (1)

• Figure 1: Examples of improvements in the CTV segmentation predictions due to anatomical prior addition in the ABDM region.