UngLong/openm3chest-labels
收藏资源简介:
OpenM3Chest是一个医学多模态多任务数据集,用于诊断胸部异常,重点是肺癌筛查。原始数据来自NLST(国家肺癌筛查试验)和MIDRC。该存储库仅包含标签文件(JSON格式)。每个样本包括:keys(DICOM序列实例UID,用于从IDC下载影像数据)、pids(患者ID)、lung_bboxes(左右肺区域边界框)、labels(真实标签,0=否,1=是)、pixel_size(体素间距)、clinical_data(患者人口统计和临床历史,如年龄、吸烟等)、questions(用于视觉问答任务的自然语言问题列表)和data_name(任务标识符)。数据集支持多个任务配置,例如胸部异常检测(如肺不张、胸膜增厚或积液)、心血管疾病诊断和死亡率预测、肺癌风险评估,以及肺结节相关特征分类(如衰减、位置、边缘、存在性和大小)。
OpenM3Chest is a medical multimodal multitask dataset for diagnosing chest abnormalities with a focus on lung cancer screening. The original raw data comes from NLST (National Lung Screening Trial) and MIDRC. This repository contains the label files only (JSON format). Each sample includes: keys — DICOM Series Instance UID (use with IDC to download imaging data), pids — Patient ID, lung_bboxes — Bounding boxes for left/right lung regions, labels — Ground truth label (0 = No, 1 = Yes), pixel_size — Voxel spacing, clinical_data — Patient demographics and clinical history (age, smoking, etc.), questions — List of natural language questions for VQA tasks, and data_name — Task identifier.



