Oral structure segmentation plays a crucial role in facilitating computer-aided orthodontic treatment. While numerous studies have concentrated on specific categories of structures, such as the inferior alveolar nerves, jawbones, and teeth, a comprehensive method for segmenting all structures within the oral cavity has been lacking. The MICCAI ToothFairy2 Challenge seeks to enhance the development of deep learning frameworks and expand the availability of publicly accessible 3D-annotated CBCT datasets. In response to this challenging task, we employed a segmentation network to segment multiple oral structures. Subsequently, we introduced an Adaptive Structure Optimization technique to mitigate the trade-off between missed and false segmentations. Our method was rigorously evaluated on the ToothFairy2 test dataset, yielding a Dice Similarity Coefficient (DSC) of 0.9167 and a 95% Hausdorff Distance (HD95) of 17.5809 mm, winning second place in the competition. Code is available at Oculins/Multi_Oral_Structure .