nnUNet for Brain Tumor Segmentation in Sub-Saharan Africa Patient Population

Glioma is one of the most dangerous and aggressive brain tumor types. These tumors are difficult to diagnose, and they require careful and precise identification and delineation of different tumor subtypes. Currently, magnetic resonance imaging (MRI) is the gold standard for glioma detection. However, not all the territories are equipped with up-to-date medical devices and highly skilled professionals, and it is important to develop glioma segmentation methods for the images acquired in such low-resource settings. Therefore, BraTS-Africa challenge was announced in 2023 to provide researchers the opportunity to develop segmentation algorithms using brain MRI glioma cases from Sub-Saharan Africa population. In this paper, we present our submission to this challenge. We based our approach on the well-known nnUNet model, which won original BraTS 2020 and 2021 challenges. Within our work, we studied the impact of using pretrained and fine-tuned models, different image input modalities, ensemble of different models, and the application of specific region based strategies. Obtained results on the unseen testing data showed promising results, having good Dice values for all 3 classes and a small HD95 distance.