High-resolution 3D abdominal segmentation with random patch network fusion

•Patch-based method plus random shifting could boost high resolution 3D segmentation performance.•Random patch enables prediction of original space segmentation in computed tomography (CT) without resampling.•Performance of a multiple stages present positive result of 3D abdominal segmentation in variety of structures and datasets.•Performance of hierarchical labeling design paves the road for computational efficient medical image analysis.