MAPS-Glioma: Modality-Specific Augmentation and Tissue-Adaptive Postprocessing for Robust Glioma Segmentation in Resource-Limited Settings

Abstract

Gliomas are the most common primary brain tumors and pose significant diagnostic challenges in Sub-Saharan Africa due to limited access to advanced imaging and late-stage patient presentations. We propose MAPS-Glioma, a deep learning framework built on an optimized 3D U-Net enhanced with modality-specific data augmentation (elastic deformations, Rician noise injection) and tissue-adaptive, topology-preserving postprocessing. Trained and validated on a diverse dataset of African MRI scans from the BraTS-Africa challenge, the framework achieves Dice scores of 0.74 (enhancing tumor), 0.75 (tumor core), and 0.85 (whole tumor), representing a step toward reducing global health disparities in precision medicine for glioma patients in Sub-Saharan Africa.

Jeremiah Fadugba (Jerofad)

Data Scientist | Machine Learning Engineer

My current research interests include Medical Imaging, Machine Learning, Deep Learning, and Trustworthy ML.