Liver disease is one of the main sources of malignancy demise. To help specialists in hepatocellular carcinoma conclusion and treatment arranging, a precise and programmed liver and tumor division strategy is very requested in the clinical practice. As of late, completely convolutional neural systems (FCNs), including 2D and 3D FCNs, fill in as the spine in numerous volumetric picture division. Be that as it may, 2D convolutions can not completely use the spatial data along the third measurement while 3D convolutions experience the ill effects of high computational expense and GPU memory utilization.
To address these issues, we propose a novel cross breed thickly associated UNet (H-DenseUNet), which comprises of a 2D DenseUNet for proficiently separating intra-cut highlights and a 3D partner for progressively collecting volumetric settings under the soul of the auto-setting calculation for liver and tumor division. We plan the learning procedure of H-DenseUNet in a conclusion to-end way, where the intra-cut portrayals and between cut highlights can be together enhanced through a half and half element combination (HFF) layer. We broadly assessed our technique on the dataset of MICCAI 2017 Liver Tumor Segmentation (LiTS) Challenge and 3DIRCADb Dataset. Our strategy beat other condition of expressions of the human experience on the division aftereffects of tumors and accomplished extremely focused execution for liver division even with a solitary model.