Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Domain-Adaptive Transformer for Data-Efficient Glioma Segmentation in Sub-Saharan MRI

Ilerioluwakiiye Abolade, Aniekan Udo, Augustine Ojo, Abdulbasit Oyetunji, Hammed Ajigbotosho, Aondana Iorumbur, Confidence Raymond, Maruf Adewole

TL;DR

This work tackles domain shift in glioma MRI segmentation for Sub-Saharan Africa by proposing SegFormer3D+, a domain-adaptive transformer that combines histogram-based intensity harmonization, radiomics-guided stratification, and a frequency-aware encoder with dual attention. Pretrained on BraTS 2023 and fine-tuned on BraTS-Africa, the approach achieves superior Dice scores and boundary accuracy across heterogeneous scans, outperforming baselines such as nnU-Net and Swin-UNETR. Key contributions include an explicit preprocessing-and-architecture integration for cross-domain robustness and an ablation analysis quantifying each component's impact. The results highlight significant potential for reliable, equitable tumor delineation in resource-limited settings, with future work targeting self-supervised learning and broader SSA cohorts.

Abstract

Glioma segmentation is critical for diagnosis and treatment planning, yet remains challenging in Sub-Saharan Africa due to limited MRI infrastructure and heterogeneous acquisition protocols that induce severe domain shift. We propose SegFormer3D-plus, a radiomics-guided transformer architecture designed for robust segmentation under domain variability. Our method combines: (1) histogram matching for intensity harmonization across scanners, (2) radiomic feature extraction with PCA-reduced k-means for domain-aware stratified sampling, (3) a dual-pathway encoder with frequency-aware feature extraction and spatial-channel attention, and (4) composite Dice-Cross-Entropy loss for boundary refinement. Pretrained on BraTS 2023 and fine-tuned on BraTS-Africa data, SegFormer3D-plus demonstrates improved tumor subregion delineation and boundary localization across heterogeneous African clinical scans, highlighting the value of radiomics-guided domain adaptation for resource-limited settings.

Domain-Adaptive Transformer for Data-Efficient Glioma Segmentation in Sub-Saharan MRI

TL;DR

This work tackles domain shift in glioma MRI segmentation for Sub-Saharan Africa by proposing SegFormer3D+, a domain-adaptive transformer that combines histogram-based intensity harmonization, radiomics-guided stratification, and a frequency-aware encoder with dual attention. Pretrained on BraTS 2023 and fine-tuned on BraTS-Africa, the approach achieves superior Dice scores and boundary accuracy across heterogeneous scans, outperforming baselines such as nnU-Net and Swin-UNETR. Key contributions include an explicit preprocessing-and-architecture integration for cross-domain robustness and an ablation analysis quantifying each component's impact. The results highlight significant potential for reliable, equitable tumor delineation in resource-limited settings, with future work targeting self-supervised learning and broader SSA cohorts.

Abstract

Glioma segmentation is critical for diagnosis and treatment planning, yet remains challenging in Sub-Saharan Africa due to limited MRI infrastructure and heterogeneous acquisition protocols that induce severe domain shift. We propose SegFormer3D-plus, a radiomics-guided transformer architecture designed for robust segmentation under domain variability. Our method combines: (1) histogram matching for intensity harmonization across scanners, (2) radiomic feature extraction with PCA-reduced k-means for domain-aware stratified sampling, (3) a dual-pathway encoder with frequency-aware feature extraction and spatial-channel attention, and (4) composite Dice-Cross-Entropy loss for boundary refinement. Pretrained on BraTS 2023 and fine-tuned on BraTS-Africa data, SegFormer3D-plus demonstrates improved tumor subregion delineation and boundary localization across heterogeneous African clinical scans, highlighting the value of radiomics-guided domain adaptation for resource-limited settings.

Paper Structure

This paper contains 16 sections, 6 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Datasets
  4. Intensity Harmonization
  5. Radiomics-Based Stratification
  6. Architecture
  7. Training
  8. Postprocessing and Evaluation
  9. Results
  10. Experimental Setup
  11. Comparison with Baselines
  12. Ablation Study
  13. Qualitative Analysis
  14. Discussion and Conclusion
  15. Compliance with Ethical Standards
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Overview of the proposed SegFormer3D+ segmentation pipeline.
  • Figure 2: SegFormer3D+ predictions on BraTS-Africa validation cases showing consistent delineation across multi-contrast MRI inputs and artifact conditions.