The development of deep learning methodologies has significantly impacted medical image segmentation, leading to highly accurate models for critical applications such as brain tumor delineation in MRI volumes. Precise tumor segmentation is indispensable for quantitative analysis, surgical planning, radiation treatment delivery, and disease monitoring. This paper presents a novel context-aware segmentation pipeline, conceptually rooted in anomaly detection, by integrating brain tissues as an auxiliary segmentation class. This inclusion enhances the discriminability between pathological and healthy tissues. To mitigate the class imbalance inherent in this added segmentation scheme, we incorporated a class-adaptive loss function within the nnU-Net and MedNeXt frameworks. The efficacy of this approach was rigorously evaluated on five tasks within the Brain Tumor Segmentation (BraTS) 2025 MICCAI Lighthouse challenge: adult glioma (GLI), pediatric glioma (PED), brain metastasis (MET), meningioma in pre-operative (MENpre), and meningioma in treatment planning (MENrt) segmentation. Our method demonstrated promising overall whole tumor segmentation performance on the validation set, yielding lesion-wise Dice scores of 0.873 (GLI), 0.945 (PED), 0.861 (MENpre), 0.845 (MENrt), and 0.679 (MET). The final phase of testing demonstrated that the proposed solutions performed among the top-ranked algorithms for the MET, MENpre, MENrt, and GLI challenges.