AbdomenCT-1K: Is Abdominal Organ Segmentation a Solved Problem?

• DOI: 10.1109/tpami.2021.3100536
• Publication Date: 2021-07-27T20:48:15Z
• Authors: Jun Ma; Yao Zhang; Song Gu; Cheng Zhu; Cheng Ge; Yichi Zhang; Xingle An; Congcong Wang; Qiyuan Wang; Xin Liu; Shucheng Cao; Qi Zhang; Shangqing Liu; Yunpeng Wang; Yuhui Li; Jian He; Xiaoping Yang
• Keywords: Benchmark (surveying)

With the unprecedented developments in deep learning, automatic segmentation of main abdominal organs seems to be a solved problem as state-of-the-art (SOTA) methods have achieved comparable results with inter-rater variability on many benchmark datasets. However, most existing datasets only contain single-center, single-phase, single-vendor, or single-disease cases, and it is unclear whether excellent performance can generalize diverse This paper presents large CT organ dataset, termed AbdomenCT-1K, more than 1000 (1K) scans from 12 medical centers, including multi-phase, multi-vendor, multi-disease cases. Furthermore, we conduct large-scale study for liver, kidney, spleen, pancreas reveal unsolved problems SOTA methods, such limited generalization ability distinct phases, unseen diseases. To advance problems, further build four benchmarks fully supervised, semi-supervised, weakly continual which are currently challenging active research topics. Accordingly, develop simple effective method each benchmark, used out-of-the-box strong baselines. We believe AbdomenCT-1K dataset will promote future in-depth towards clinical applicable methods.