 |
    |
Data-Driven Interactive 3D Medical Image Segmentation Based on Structured Patch Model
Sang Hyun Park1, Il Dong Yun2, and Sang Uk Lee1
1Electrical Engineering, ASRI, INMC, Seoul National University, Korea
shpark@cvl.snu.ac.kr
sanguk@snu.ac.kr
2Digital Information Engineering, Hankuk University of Foreign Studies, Korea
yun@hufs.ac.kr
Abstract. In this paper, we present a novel three dimensional interactive medical image segmentation method based on high level knowledge of training set. Since the interactive system should provide intermediate results to an user quickly, insufficient low level models are used for most of previous methods. To exploit the high level knowledge within a short time, we construct a structured patch model that consists of multiple corresponding patch sets. The structured patch model includes the spatial relationships between neighboring patch sets and the prior knowledge of the corresponding patch set on each local region. The spatial relationships accelerate the search of corresponding patch in test time, while the prior knowledge improves the segmentation accuracy. The proposed framework provides not only fast editing tool, but the incremental learning system through adding the segmentation result to the training set. Experiments demonstrate that the proposed method is useful for fast and accurate segmentation of target objects from the multiple medical images.
Keywords: interactive segmentation, 3D medical image, structured patch model, localized classifier
LNCS 7917, p. 196 ff.
lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2013