Whole brain segmentation, which divides the entire brain volume into anatomically labeled regions of interest (ROIs), is a crucial step in brain image analysis. Traditional methods often rely on intricate pipelines that, while accurate, are time-consuming and require extensive manual intervention. Deep learning approaches have shown promise, but they often struggle with fine-grained boundary delineation, especially in regions with complex anatomical structures. In this paper, we propose a curvature-guided coarse-to-fine framework that leverages geometric information to enhance whole brain segmentation accuracy. Our method first performs a coarse segmentation to identify major brain regions, then uses curvature-based features to refine boundaries in anatomically challenging areas. The framework incorporates a multi-scale architecture that progressively refines segmentation results, with special attention to regions where traditional methods typically fail. Experimental results on multiple brain imaging datasets demonstrate that our approach achieves superior performance compared to state-of-the-art methods, particularly in boundary accuracy and computational efficiency.