Heartz00/MAPS-Glioma

Modality-Specific Augmentation and Tissue-Adaptive Postprocessing for Robust Glioma Segmentation (BraTS-Africa 2025) # MICCAI 2025 BraTS-Africa Challenge: Team Tanzania ## MAPS-Glioma: Modality-Specific Augmentation and Tissue-Adaptive Postprocessing for Robust Glioma Segmentation ## Using this Repository This repository contains code developed for the BraTS-Africa 2025 Challenge, implementing a deep learning framework that integrates **modality-specific augmentation** and **tissue-adaptive postprocessing** on an optimized 3D U-Net architecture. The framework is specifically designed to address the unique challenges of glioma segmentation in Sub-Saharan African (SSA) populations, where lower-quality MRI scans and distinct tumor characteristics require specialized approaches. ### Key Features - **Enhanced 3D U-Net architecture** optimized for SSA medical imaging data - **Modality-specific augmentation** tailored for T1, T1ce, T2, and FLAIR sequences - **Tissue-adaptive postprocessing** with region-specific refinement - **Low-resource training strategies** suitable for limited computational infrastructure - **Comprehensive evaluation metrics** including Dice scores and Hausdorff distances ### BraTS-Africa 2025 Submission Our final submission achieved the following performance on the validation set: - **Enhancing Tumor (ET)**: Dice 0.75 ± 0.22, Hausdorff95 11.62 ± 13.68 mm - **Tumor Core (TC)**: Dice 0.73 ± 0.25, Hausdorff95 13.97 ± 13.12 mm - **Whole Tumor (WT)**: Dice 0.872 ± 0.17, Hausdorff95 8.86 ± 8.04 mm Multiple training strategies were explored: 1. Training with only BraTS-Africa SSA data 2. Training with BraTS-Global data and fine-tuning with SSA data 3. Training with combined BraTS-Global and SSA data with modality-specific augmentation 4. Multi-stage training with progressive augmentation strategies --- ## Citation **Please reference this article if you use this code and its scripts in your research:** Ayomide B. Oladele, Helena Machibya, Mariam Kaoneka, Frederick Lyimo, Debora Hoza, Immaculata Kafumu, Idris Olalekan, Jeremiah Fadugba, Dong Zhang, Aondona Iorumb …