Deep Learning-Based Segmentation of Locally Advanced Breast Cancer on MRI in Relation to Residual Cancer Burden: A Multi-Institutional Cohort Study

Mark Janse, Liselore Maria Janssen, Bas H.M. van der Velden, Maaike Moman, Elian Wolters-van der Ben, Marc Kock, Max A. Viergever, Paul van Diest, Kenneth Gilhuijs

Keywords: Breast MRI, response monitoring, segmentation, nnU-Net

TL;DR: nnU-Net based segmentations of locally advanced breast cancer generalize well and are predictive for outcome after chemotherapy

Abstract: While several methods have been proposed for automated assessment of breast-cancer response to neoadjuvant chemotherapy on breast MRI, limited information is available about their performance across multiple institutions. In this paper, we assess the value and robustness of nnU-Net-derived volumes of locally advanced breast cancer (LABC) on MRI to infer the presence of residual disease after neoadjuvant chemotherapy. An nnU-Net was trained to segment LABC on a single-institution training set and validated on a multi-center independent testing cohort. Based on resulting tumor volumes, an extremely randomized tree model was trained to assess residual cancer burden (RCB)-0/I vs. RCB-II/III. An independent model was developed using functional tumor volume (FTV) for comparison to an established method. Models were tested on an independent testing cohort, response assessment performance and robustness across multiple institutions were assessed.Results show that nnU-Net accurately estimate changes in tumor load on DCE-MRI, that these changes associated with RCB after NAC, and that they are robust against variations between institutions.