Automatic volumetric localization of the liver in abdominal CT scans using low level processing and shape priors

In this paper we present an automatic volumetric liver localization method as an approach for liver segmentation. In the proposed method the aim is to localise a mean shape model of the liver in the target CT scan. The framework consists of three main steps: shape model construction, low level processing and shape model registration. We evaluated our method on the MICCAI 2007 liver segmentation challenge dataset. The Leave-one-out validation results demonstrate the effectiveness of the proposed method. The average volume overlap between our method and the ground truth, using the Jaccard index, is 0.64±0.11 which is acceptable for an initial localisation of the liver prior to further refinement.