Automatic Liver Lesion Segmentation in CT Combining Fully Convolutional Networks and Non-negative Matrix Factorization

Automatic liver tumor segmentation is an important step towards digital medical research, clinical diagnosis and therapy planning. However, the existence of noise, low contrast and heterogeneity make the automatic liver tumor segmentation remaining an open challenge. In this work, we focus on a novel automatic method to segment liver tumor in abdomen images from CT scans by using fully convolutional networks (FCN) and non-negative matrix factorization (NMF). We train the FCN for semantic liver and tumor segmentation. The segmented liver and tumor regions of FCN are used as ROI and initialization for the NMF based tumor refinement, respectively. We refine the surfaces of tumors using a 3D deformable model which derived from NMF and driven by local cumulative spectral histograms (LCSH). The refinement is designed to obtain a smoother, more accurate and natural liver tumor surface. Experiments demonstrated that the proposed segmentation method achieves satisfactory results. Likewise, it has been notably observed that the computing time of the segmentation method is no more than one minute for each CT volume.