Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Kernel MedNeXt for Breast Tumor Segmentation and Self-Normalizing Network for pCR Classification in Magnetic Resonance Images

Toufiq Musah

TL;DR

This work tackles simultaneous breast tumor segmentation in DCE-MRI and pCR classification using the MAMA-MIA dataset. It introduces a large-kernel MedNeXt segmentation approach with an UpKern two-stage training to expand receptive fields from $3×3×3$ to $5×5×5$, achieving unseen Dice of about 0.67 and NormHD of 0.24 when ensembling. For pCR, radiomic features extracted from segmentations and first post-contrast MRI are fed into a Self-Normalizing Network to produce a balanced accuracy of 57% on unseen data, with notable subgroup variability. The results demonstrate the benefit of larger receptive fields for segmentation and radiomics-based classification while highlighting calibration and fairness considerations and suggesting directions for improved ensembling and clinical-variable integration.

Abstract

Accurate breast tumor segmentation in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is important for downstream tasks such as pathological complete response (pCR) assessment. In this work, we address both segmentation and pCR classification using the large-scale MAMA-MIA DCE-MRI dataset. We employ a large-kernel MedNeXt architecture with a two-stage training strategy that expands the receptive field from 3x3x3 to 5x5x5 kernels using the UpKern algorithm. This approach allows stable transfer of learned features to larger kernels, improving segmentation performance on the unseen validation set. An ensemble of large-kernel models achieved a Dice score of 0.67 and a normalized Hausdorff Distance (NormHD) of 0.24. For pCR classification, we trained a self-normalizing network (SNN) on radiomic features extracted from the predicted segmentations and first post-contrast DCE-MRI, reaching an average balanced accuracy of 57\%, and up to 75\% in some subgroups. Our findings highlight the benefits of combining larger receptive fields and radiomics-driven classification while motivating future work on advanced ensembling and the integration of clinical variables to further improve performance and generalization. Code: https://github.com/toufiqmusah/caladan-mama-mia.git

Large Kernel MedNeXt for Breast Tumor Segmentation and Self-Normalizing Network for pCR Classification in Magnetic Resonance Images

TL;DR

This work tackles simultaneous breast tumor segmentation in DCE-MRI and pCR classification using the MAMA-MIA dataset. It introduces a large-kernel MedNeXt segmentation approach with an UpKern two-stage training to expand receptive fields from to , achieving unseen Dice of about 0.67 and NormHD of 0.24 when ensembling. For pCR, radiomic features extracted from segmentations and first post-contrast MRI are fed into a Self-Normalizing Network to produce a balanced accuracy of 57% on unseen data, with notable subgroup variability. The results demonstrate the benefit of larger receptive fields for segmentation and radiomics-based classification while highlighting calibration and fairness considerations and suggesting directions for improved ensembling and clinical-variable integration.

Abstract

Accurate breast tumor segmentation in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is important for downstream tasks such as pathological complete response (pCR) assessment. In this work, we address both segmentation and pCR classification using the large-scale MAMA-MIA DCE-MRI dataset. We employ a large-kernel MedNeXt architecture with a two-stage training strategy that expands the receptive field from 3x3x3 to 5x5x5 kernels using the UpKern algorithm. This approach allows stable transfer of learned features to larger kernels, improving segmentation performance on the unseen validation set. An ensemble of large-kernel models achieved a Dice score of 0.67 and a normalized Hausdorff Distance (NormHD) of 0.24. For pCR classification, we trained a self-normalizing network (SNN) on radiomic features extracted from the predicted segmentations and first post-contrast DCE-MRI, reaching an average balanced accuracy of 57\%, and up to 75\% in some subgroups. Our findings highlight the benefits of combining larger receptive fields and radiomics-driven classification while motivating future work on advanced ensembling and the integration of clinical variables to further improve performance and generalization. Code: https://github.com/toufiqmusah/caladan-mama-mia.git

Paper Structure

This paper contains 10 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Dataset
  4. MedNeXt Segmentation Training
  5. pCR Classification
  6. Results
  7. Segmentation Performance
  8. pCR Classification Performance
  9. Discussion
  10. Conclusion

Figures (2)

  • Figure 1: Sample DCE-MRI data showing pre-contrast and post-contrast scans along with expert segmentation in green
  • Figure 2: Overview of the proposed pipeline. (a) MedNeXt performs tumor segmentation on DCE-MRI. (b) Radiomic features are extracted from the segmentations and first post-contrast MRI. (c) A Self-Normalizing Network (SNN) predicts pathological complete response (pCR) from the selected features.