Synthetic Dataset of Parotid Segmentations

该数据集是由合成方法生成的，包含了具有真实轮廓误差的腮腺分割数据。该数据集不仅用于训练和验证所提出模型的前期预训练，还用于预测分割误差的任务，确保了跨验证的一致性。每个折层中，数据集分别包含了4800个训练结构、600个验证结构和1400个测试结构。该数据集的任务是预测放射治疗中风险器官分割的误差。

This dataset is generated via synthetic methods and contains parotid gland segmentation data with real contour errors. It is not only utilized for the early-stage pre-training, training and validation of the proposed model, but also employed for the task of predicting segmentation errors, which ensures the consistency across cross-validation folds. In each cross-validation fold, the dataset respectively includes 4800 training samples, 600 validation samples and 1400 test samples. The task of this dataset is to predict the segmentation errors of organs-at-risk (OARs) in radiotherapy.