Automatic tooth segmentation in 2D panoramic X-ray images is crucial for various applications. Task 1 of the MICCAI STS 2024 Challenge aims to advance automated tooth segmentation techniques by providing datasets comprising labeled and unlabeled panoramic X-ray images. This paper addresses the challenge of limited labeled data by framing the problem as a semi-supervised learning task. We propose a two-stage deep learning model based on nnU-Net. The method first performs quadrant segmentation using one nnU-Net, followed by fine tooth segmentation using another nnU-Net within each quadrant. To effectively utilize unlabeled data, we implement a selective stability-based retraining strategy to generate reliable pseudo-labels. We further enhance model performance through post-processing methods such as connected domain analysis. Quantitative evaluation on the STS 2024 validation set demonstrates that our method achieves strong performance across several metrics (Dice_instance = \(79.82\%\) , Dice_image = \(94.02\%\) ). In the competition’s validation phase, the method was awarded second place, validating its efficacy in automated dental segmentation of panoramic X-ray images.