For each patient, the HN dataset provides CT and PET scans and Gross Tumor V olume (GTV) mask, preprocessed according to the pipeline in Fig. 1. Several clinical variables are included, in particular the Locoregional Recurrence (LR) within the follow-up period (median: 43 months; range: 6-112 months), and T-stage at diagnosis. Data are gathered from four different hospitals, each one representing a single cohort. No- tably, each hospital has its own image acquisition equipment and acquisition settings, which is a cause of heterogeneity in image characteristics, in particular resolution of the PET images. Moreover, The HN dataset is highly unbalanced for the LR prognosis, with 15.8% of recurrence (Table I). 解释

这段话是在描述Head-Neck-PET-CT（HN）数据集1中提供的数据。对于每个患者，该数据集提供了经过预处理的CT和PET扫描以及Gross Tumor Volume（GTV）掩模，预处理的管线如图1所示。数据集中还包括几个临床变量，特别是随访期内（中位数为43个月，范围为6-112个月）的局部复发（LR）和诊断时的T分期。数据收集自四个不同的医院，每个医院代表一个单独的队列。值得注意的是，每个医院都有自己的图像采集设备和采集设置，这是图像特征异质性的原因，特别是PET图像的分辨率。此外，HN数据集在LR预后方面高度不平衡，复发率为15.8%（表I）。