Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised Domain Adaptation for Pediatric Brain Tumor Segmentation

Jingru Fu, Simone Bendazzoli, Örjan Smedby, Rodrigo Moreno

TL;DR

This study tackles the domain gap between adult and pediatric brain tumor imaging by proposing DA-nnUNet, an unsupervised domain adaptation framework that adds a GRL-based domain classifier to a 3D nnU-Net. The model learns domain-invariant features while keeping source-domain segmentation accuracy, enabling effective pediatric glioma segmentation without pediatric annotations. Empirical results show the method achieving tumor core segmentation performance close to an upper-bound model that uses annotations from both domains, with substantial Dice and HD95 gains over adult-only training. The approach offers a practical route for pediatric tumor segmentation in data-scarce settings and is accompanied by publicly available code for replication and extension.

Abstract

Significant advances have been made toward building accurate automatic segmentation models for adult gliomas. However, the performance of these models often degrades when applied to pediatric glioma due to their imaging and clinical differences (domain shift). Obtaining sufficient annotated data for pediatric glioma is typically difficult because of its rare nature. Also, manual annotations are scarce and expensive. In this work, we propose Domain-Adapted nnU-Net (DA-nnUNet) to perform unsupervised domain adaptation from adult glioma (source domain) to pediatric glioma (target domain). Specifically, we add a domain classifier connected with a gradient reversal layer (GRL) to a backbone nnU-Net. Once the classifier reaches a very high accuracy, the GRL is activated with the goal of transferring domain-invariant features from the classifier to the segmentation model while preserving segmentation accuracy on the source domain. The accuracy of the classifier slowly degrades to chance levels. No annotations are used in the target domain. The method is compared to 8 different supervised models using BraTS-Adult glioma (N=1251) and BraTS-PED glioma data (N=99). The proposed method shows notable performance enhancements in the tumor core (TC) region compared to the model that only uses adult data: ~32% better Dice scores and ~20 better 95th percentile Hausdorff distances. Moreover, our unsupervised approach shows no statistically significant difference compared to the practical upper bound model using manual annotations from both datasets in TC region. The code is shared at https://github.com/Fjr9516/DA_nnUNet.

Unsupervised Domain Adaptation for Pediatric Brain Tumor Segmentation

TL;DR

This study tackles the domain gap between adult and pediatric brain tumor imaging by proposing DA-nnUNet, an unsupervised domain adaptation framework that adds a GRL-based domain classifier to a 3D nnU-Net. The model learns domain-invariant features while keeping source-domain segmentation accuracy, enabling effective pediatric glioma segmentation without pediatric annotations. Empirical results show the method achieving tumor core segmentation performance close to an upper-bound model that uses annotations from both domains, with substantial Dice and HD95 gains over adult-only training. The approach offers a practical route for pediatric tumor segmentation in data-scarce settings and is accompanied by publicly available code for replication and extension.

Abstract

Significant advances have been made toward building accurate automatic segmentation models for adult gliomas. However, the performance of these models often degrades when applied to pediatric glioma due to their imaging and clinical differences (domain shift). Obtaining sufficient annotated data for pediatric glioma is typically difficult because of its rare nature. Also, manual annotations are scarce and expensive. In this work, we propose Domain-Adapted nnU-Net (DA-nnUNet) to perform unsupervised domain adaptation from adult glioma (source domain) to pediatric glioma (target domain). Specifically, we add a domain classifier connected with a gradient reversal layer (GRL) to a backbone nnU-Net. Once the classifier reaches a very high accuracy, the GRL is activated with the goal of transferring domain-invariant features from the classifier to the segmentation model while preserving segmentation accuracy on the source domain. The accuracy of the classifier slowly degrades to chance levels. No annotations are used in the target domain. The method is compared to 8 different supervised models using BraTS-Adult glioma (N=1251) and BraTS-PED glioma data (N=99). The proposed method shows notable performance enhancements in the tumor core (TC) region compared to the model that only uses adult data: ~32% better Dice scores and ~20 better 95th percentile Hausdorff distances. Moreover, our unsupervised approach shows no statistically significant difference compared to the practical upper bound model using manual annotations from both datasets in TC region. The code is shared at https://github.com/Fjr9516/DA_nnUNet.
Paper Structure (11 sections, 1 equation, 4 figures, 3 tables)

This paper contains 11 sections, 1 equation, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Supervised Baseline Models
  4. Unsupervised Domain Adaptation for Pediatric Brain Tumor Segmentation
  5. Experimental Setting
  6. Datasets
  7. Implementation
  8. Results and Discussion
  9. Conclusion
  10. Acknowledgments.
  11. Disclosure of Interests.

Figures (4)

  • Figure 1: Different strategies to obtain upper-bound and baseline models when assuming fully-annotated pediatric (PED) data are provided.
  • Figure 2: The proposed DA-nnUNet comprises a 3D nnUNet-based backbone and a domain classifier. UDA is facilitated by integrating a gradient reversal layer (GRL) before the domain classifier.
  • Figure 3: Distribution of tumor sizes (left) and examples of the appearance of adult and pediatric tumors on FLAIR and T1w after contrast (T1CE) (right).
  • Figure 4: Distributions of DSC (left) and HD95 (right) for Table \ref{['tab:UDA_remove']}. Pairwise t-tests were used to assess the statistical significance of differences between methods on $\mathcal{D}_{\text{PED}}^{\text{target}}$.