Focal Liver Lesion Classification Based on Tensor Sparse Representations of Multi-phase CT Images

The bag-of-visual-words (BoVW) method has been proved to be an effective method for classification tasks in both natural imaging and medical imaging. In this paper, we propose a multilinear extension of the traditional BoVW method for classification of focal liver lesions using multi-phase CT images. In our approach, we form new volumes from the corresponding slices of multi-phase CT images and extract cubes from the volumes as local structures. Regard the high dimensional local structures as tensors, we propose a K-CP (CANDECOMP/PARAFAC) algorithm to learn a tensor dictionary in an iterative way. With the learned tensor dictionary, we can calculate sparse representations of each group of multi-phase CT images. The proposed tensor was evaluated in classification of focal liver lesions and achieved better results than conventional BoVW method.