Structured patch model for a unified automatic and interactive segmentation framework

摘要: We present a novel interactive segmentation framework incorporating priori knowledge learned from training data. The is as structured patch model (StPM) comprising sets of corresponding local priors and their pairwise spatial distribution statistics which represent the shape appearance along its boundary global structure, respectively. When successive user annotations are given, StPM appropriately adjusted in target image used together with to guide segmentation. reduces dependency on placement quantity little increase complexity since time-consuming construction performed offline. Furthermore, seamless learning system can be established by directly adding results StPM. proposed method was evaluated three datasets, respectively, 2D chest CT, 3D knee MR, brain MR. experimental demonstrate that within an equal amount time, outperforms recent state-of-the-art methods terms accuracy, while it requires significantly less computing editing time obtain comparable accuracy.