Comparative study of interactive seed generation for growcut-based fast 3D MRI segmentation

This paper proposes a speed-enhanced growcut method and presents comparative study of seed setting methods for fast 3D medical image (MRI) segmentation. The processing time tends to be larger in 3D image segmentation because of the large number of neighboring voxels as well as the number of voxels themselves. In this paper, two seed setting methods are proposed for our fast growcut-based segmentation algorithm: sphere-based bounding box method and label transfer based method using SIFT flow. Experimental results demonstrate that the tumor segmentation for each patient can be done very quickly as compared to the previous works. The segmentation accuracy can also be made very high with only a few user interactions.