<p>Due to the inherent noise and artifacts of 3D medical imaging, issues such as low contrast and signal interference make segmentation still highly challenging. Existing methods mainly focus on modeling spatial information, while the intrinsic texture and directional cues embedded in frequency-domain representations remain underexplored. To address this, we propose a Frequency-Domain Modulation Network (FDM-Net), which incorporates frequency-domain cues into the segmentation process. To differentiate between the global morphological components of primary features and the fine-grained texture components, we develop a Frequency Decomposition (FD) module that separates images into heterogeneous frequency components, thereby highlighting overall contours and microscopic edges. To capture the characteristics of both high-frequency and low-frequency features, we introduce a Frequency-domain Adaptive Convolution (FAConv), which dynamically models convolutional kernel weights to enhance the dual-branch responsiveness to different frequencies. Furthermore, to effectively integrate multi-scale and multi-branch representations, we design a Frequency-Guided Hybrid Attention (FGHA) module that aggregates dimension-specific information. Through visualization and experimental analysis, extensive results on three different imaging modality datasets demonstrate that the proposed FDM-Net achieves superior performance and robustness compared to state-of-the-art methods, while maintaining low computational cost.</p>