The Robustness Evaluation of Advanced Models in CT Image Segmentation: From Multi-organ to Multi-organ

Medical image segmentation models are often tested on the same dataset used for training, which limits real-world applicability. This paper evaluates the robustness of several 3D and 2D models by comparing their performance on BTCV and AbdomenCT-1K datasets. The study explores the effects of model architecture, dimensionality, organ characteristics, and dataset differences on robustness through visualizations and various metrics, providing insights and recommendations for improving robustness and generalization.