Deep Interactive Segmentation of Uncertain Regions with Shadowed Sets
Haiyan Zheng, Yufei Chen, Xiaodong Yue, Chao Ma · 2019
Pancreas segmentation is a challenging task in medical image analysis because of its large variations in texture, location, shape and size and the high similarity to the surrounding tissues especially around the boundary regions, which leads to the high segmentation uncertainty and makes the results inaccurate. Existing fully automatic segmentation methods rarely achieve sufficiently accurate and robust results. To tackle this problem, we propose a deep learning based interactive uncertain segmentation method which can involve the domain knowledge in the process of segmentation in an interactive and iterative way. Specially, the proposed method describes the uncertain regions of pancreatic CT images based on shadowed sets theory which are further corrected through interaction. The proposed method is evaluated on a challenging 3D pancreatic CT images dataset collected from the Changhai Hospital. The experimental results demonstrate that our proposed method outperforms the existing methods in terms of both the Dice similarity coefficient of 78% and the pixel-wise accuracy of 96%, which reveals the effectiveness and the potential of our method in clinical settings.