Accurate segmentation of glioblastoma’s distinct regions is vital for treatment planning but remains challenging, especially in datasets from low-resource settings. This paper presents our submission to the BraTS-Lighthouse 2025 Challenge (Task 5: Sub-Saharan Africa Adult Glioma Segmentation), which addresses heterogeneous multi-institutional mpMRI scans from the sub-Saharan African population. We extend the U-Net architecture with three dedicated decoders to segment non-overlapping tumor regions: surrounding non-enhancing FLAIR hyperintensity (SNFH), non-enhancing tumor core (NETC), and enhancing tumor (ET). Each decoder specializes in its region, supported by an attention mechanism that emphasizes the most relevant MRI sequences—for example, T2 and FLAIR for SNFH, T1 and T1ce for NETC and ET—enhancing region-specific feature learning. Evaluated on the BraTS-Africa 2025 validation set, our approach outperforms state-of-the-art baselines, including optimized and tuned U-Net achieving an average lesion-wise dice of 0.846, demonstrating robust and precise tumor delineation in this underserved and diverse population.