Medical image lesion segmentation is crucial for disease diagnosis and treatment planning but faces significant challenges, including the randomness of lesion morphology and location, variations in imaging equipment, ambiguous boundaries, difficulties in identifying early-stage small lesions, scarcity of medical data, and high annotation costs. To address these issues, this paper proposes an innovative dynamic feature fusion mechanism based on the Kolmogorov-Arnold (KA) formula, termed the KA Dynamic (KAD) module. This module dynamically generates channel weights using the KA formula, adaptively adjusting the importance distribution of cross-scale features to achieve efficient feature fusion between the encoder and decoder. Experimental results show that the proposed method performs exceptionally well in brain tumor segmentation tasks, achieving improved training performance with minimal increases in computational time while demonstrating strong generalization capabilities. The study indicates that this approach holds significant advantages in handling complex medical image segmentation tasks.