We propose a segmentation framework for multi-class aortic segmentation that combines global context modeling using a Mamba-based module as the core component, with an additional gated attention mechanism explored for spatially adaptive feature refinement. To enhance anatomical plausibility, we incorporate a post-processing step based on anatomical location constraints. On the AortaSeg24 test set, our method achieved an average Dice score of 0.675 and NSD of 0.693. While the overall scores appear lower than the baseline, further analysis revealed that the near-zero scores for the last two anatomical classes were caused by a submission error. Excluding these, our model attained a Dice of 0.738 and NSD of 0.758, outperforming the baseline. Qualitative evaluation highlights strengths in major vessel segmentation but shows occasional boundary ambiguity in multi-segment regions and inconsistent predictions at distal vessel ends. These issues are likely due to limited patch-level context and the absence of clinical priors guiding the extent of vessel annotations. Future improvements will focus on context-aware modeling and clinical knowledge integration to enhance robustness and accuracy.