AbdomenCT-1K

我们提供了一个大型且多样化的腹部 CT 器官分割数据集，称为 AbdomenCT-1K，其中包含来自 12 个医疗中心的超过 1000 (1K) 次 CT 扫描，包括多阶段、多供应商和多疾病病例。此外，我们对肝脏、肾脏、脾脏和胰腺的分割进行了大规模研究，并揭示了 SOTA 方法尚未解决的分割问题，例如对不同医疗中心、阶段和未见疾病的泛化能力有限。为了推进未解决的问题，我们进一步建立了四个器官分割基准，用于全监督、半监督、弱监督和持续学习，这些都是目前具有挑战性和活跃的研究课题。因此，我们为每个基准开发了一种简单有效的方法，可以用作开箱即用的方法和强大的基线。我们相信 AbdomenCT-1K 数据集将促进未来对临床适用的腹部器官分割方法的深入研究。

We present a large-scale and diverse abdominal CT organ segmentation dataset named AbdomenCT-1K, which contains over 1000 (1K) CT scans from 12 medical centers, covering multi-phase, multi-vendor, and multi-disease cases. Furthermore, we conducted a large-scale study on the segmentation of liver, kidney, spleen, and pancreas, and uncovered unsolved segmentation challenges faced by state-of-the-art (SOTA) methods, such as limited generalization ability across different medical centers, scanning phases, and unseen diseases. To advance the research on these unsolved challenges, we further established four organ segmentation benchmarks for fully supervised, semi-supervised, weakly-supervised, and continual learning, which are all challenging and active research topics nowadays. Consequently, we developed a simple yet effective method for each benchmark, which can serve as an out-of-the-box solution and a strong baseline. We believe that the AbdomenCT-1K dataset will facilitate in-depth future research on clinically applicable abdominal organ segmentation methods.

搜集汇总

背景与挑战

背景概述

AbdomenCT-1K是一个大型腹部CT器官分割数据集，包含来自12个医疗中心的1000多次CT扫描，覆盖多阶段、多供应商和多疾病病例，专注于肝脏、肾脏、脾脏和胰腺的分割。该数据集旨在解决SOTA方法在泛化能力方面的局限性，并提供了全监督、半监督、弱监督和持续学习四个基准，以促进临床适用的腹部器官分割方法研究。