Publication

Abstract : This letter presents a classification model for focal liver lesions (FLLs) based on statistical variations of discrete Haar wavelet transform subbands. The statistical perturbations among the frames of FLL subbands are modeled using eigenvector and diagonal matrices of the singular value decomposition (SVD). Further, the maximum value across the columns of SVD matrices is computed to obtain the frame-level statistical attributes of the FLLs. Subsequently, the frame level cues are averaged over the number of FLL video frames and given to a trained decision tree (DT) classifier for final classification. Experimental results on the SYSU-FLL-CEUS dataset demarcate that the proposed model achieved best performance results compared to conventional machine learning approaches, showcasing the superiority of the proposed classification model for FLLs.