One-shot medical landmark detection effectively reduces annotation effort by relying on a single labeled template; however, landmark ambiguity frequently arises when multiple anatomical structures exist near the target landmark. In such cases, the model struggles to distinguish which structures truly correspond to the landmark, as mere pixel-level similarity often fails to capture essential semantic cues, leading to reduced detection accuracy. To tackle this ambiguity, we propose a novel mask drawing method for one-shot landmark detection. Specifically, we manually annotate and mask out regions unrelated to each landmark’s semantic features, then replace these masked areas with various fill strategies. By replacing irrelevant features in the template image with random pixels, our approach enables the network to recognize that these features are insignificant for landmark detection, thereby mitigating ambiguity. Experimental results on a cephalometric X-ray dataset confirm that our method not only boosts precision for ambiguous landmarks, but also augments data diversity and generalization, all while maintaining minimal annotation overhead. This mask-based framework effectively resolves landmark ambiguity and enhances the accuracy of one-shot medical landmark detection.