Hybrid classifiers ensemble with an undersampling scheme for liver tumor segmentation

In this paper, we propose a new framework, namely hybrid classifiers ensemble with random undersampling for liver tumor segmentation. Essentially, the proposed framework is working on computed tomography images in which each pixel is represented by a rich feature vector. To handle the class imbalance problem, those pixels which correspond to non-tumor region are randomly subsampled. Outcomes of three types of classifiers are then combined in a decision level for performance enhancement. Our empirical results on 19 tumor images from 11 patients show promising segmentation performance.