摘要Abdominal organ segmentation is an essential and fundamental medical procedure with many clinical and research applications.There is extensive variability in the size,location,and shape of the abdominal organs among individuals,and neighboring organs and structures exhibit similar textures and levels of intensity,which contribute to the difficulties encountered when developing robust,accurate,and automated seg-mentation approaches.In the past decade,deep learning(DL)-based methods have shown promising results based on a large amount of labeled data.However,acquiring large-scale images with high-quality annota-tions is both difficult and impractical in clinical practice.Furthermore,the images obtained from multi-centers contain domain shift,which degenerates the model's performance on any new and unseen dataset.There have been extensive efforts to develop limited-supervised segmentation methods that can tackle this imperfect data problem.At the same time,prior knowledge from the medical domain may improve model performance while also constraining the results to an anatomically plausible range.In this paper,we provide a review of DL-based methods for abdominal organ segmentation that covers supervised and limited-su-pervised segmentation techniques,as well as the utilization of prior knowledge of abdominal organ and strategies in DL models.We present a categorized methodological overview of these approaches and sum-marize the relevant benchmarks and evaluation metrics used in this research area.Finally,we discuss the challenges and potential trends that may emerge in abdominal segmentation.Accordingly,this review systematically synthesizes advancements in DL for abdominal organ segmentation,provides relevant re-ferences for researchers in this field,and promotes the transformation of DL techniques into more precise clinical tools for this domain.