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

2025-11-04

AI Summary

Domain shift in glioma segmentation arises from MRI equipment heterogeneity and data scarcity across sub-Saharan African centers. Method: We propose a radiomics-guided domain-adaptive 3D Transformer. First, histogram matching and PCA-k-means enable domain-aware hierarchical sampling. Second, a dual-path frequency-domain-aware encoder integrates frequency-domain representations with spatial-channel co-attention. Third, a composite Dice-Cross-Entropy loss optimizes tumor boundary delineation. Results: Fine-tuned on the BraTS-Africa dataset, our model achieves significantly improved sub-region segmentation accuracy and boundary localization, outperforming state-of-the-art methods in cross-center generalization. Contribution: This work is the first to embed radiomic features into the domain-adaptive sampling pipeline and to introduce a frequency-enhanced Transformer architecture—establishing a novel paradigm for low-resource, multi-center medical image segmentation.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Segmentation
• Application Domains: Transportation

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Security and Privacy: Data transparency and provenance
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

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.

Problem

• Addresses glioma segmentation challenges in Sub-Saharan Africa's MRI data
• Mitigates domain shift from heterogeneous acquisition protocols and limited infrastructure
• Enables robust tumor delineation in resource-limited clinical settings

Innovation

• Histogram matching harmonizes scanner intensity variations
• Radiomic feature extraction enables domain-aware stratified sampling
• Dual-pathway encoder with attention refines boundary localization