Bayesian-based Saliency Model for Liver Tumor Enhancement

Automatic tumor enhancement and detection has an essential role for the computer-aided diagnosis of liver tumor in CT volume data. This paper proposes a novel tumor enhancement strategy by extracting a tumor saliency map, which represents the uncommon or tumor tissue compared to the liver and vessel ones in CT volumes. The saliency map can be constructed by exploring the existing probability of tumor in any voxel. However, the tumor prototypes in a test liver volume from a specific patient or common tumor prototypes are extremely difficult to achieve due to requirement of full-searching and large variation of tumor tissues in different liver volumes. Therefore, this paper investigates a tumor-training-data free strategy by only constructing the common healthy liver and vessel prototypes, which can be extracted from any slice of a liver volume, and then applies a nonparametric Bayesian framework for calculating the existing probability of liver or vessel. Finally, the existing probability of tumor can be deduced from that of liver or vessel. The advantages of our proposed strategy mainly include three aspects: (1) it only needs to construct the prototypes of common tissue such as liver or vessel region, which are easily obtained in any liver volume; (2) it proposes an adaptive non-parametric framework for tumor enhancement, which does not need to learn a common classification model for all liver volumes; (3) dispensable to remove the other different structure such as vessel in liver volume as a pre-processing step. Experiments validate that the proposed Bayesian-based saliency model for liver tumor enhancement can perform much better than the conventional approaches such as EM, EM/MPM tumor segmentation methods.