DeepLesion dataset contains 32,735 diverse lesions in 32,120 CT slices from 10,594 studies of 4,427 unique patients. Each lesion has a bounding box annotation on the key slice, which is derived from the longest diameter and longest perpendicular diameter. We annotated 5000 lesions with MedSAM2 in a human-in-the-loop pipeline.
DeepLesion 数据集包含来自 4,427 名独特患者的 10,594 项研究的 32,120 个 CT 切片中的 32,735 个不同病灶。每个病变在关键切片上都有一个边界框注释，该注释源自最长直径和最长垂直直径。我们在人机回环管道中用 MedSAM2 注释了 5000 个病灶。