Improvement of kidney segmentation from volume data using shape constraint model and local deformation strategy

摘要

This paper introduces a volume segmentation technique for improvement of an initial kidney segmentation from volume data. Previously, a semiautomatic multi-stage approach for kidney (and other lower torso structure of interest) segmentation has been developed. That approach relies on the user selecting an initial seed point within the structure of interest but a bad seed point selection can produce a poor segmentation. The technique we present here uses a shape constraint model and a local deformation strategy to help overcome the poor segmentation problem of our previous segmentation method. A kidney shape constraint model is developed and combined with our previous segmentation approach to correct poor segmentations and generate an initial segmented kidney volume. This initial segmented volume is then locally deformed to improve the segmentation.