Evaluation of viable tumor burden is an important step in the assessment of response rates for chemotherapy and it involves segmentation of tissue regions still affected by liver cancer from the whole slide images (WSIs). Recent developments in deep convolutional neural networks (CNNs) have shown state-of-the-art performance for various medical image analysis problems. In particular, fully convolutional neural networks (FCNs) have been shown to be suitable for performing image segmentation. Due to the large size of WSIs, most of the deep learning methods analyze the WSI by dividing it into small image patches and perform segmentation on each patch independently. Since, the neighboring patches share spatial correlation, the predicted segmentations might be inconsistent at the patch bordering regions. In this paper we propose an ensemble deep learning model for automating the segmentation of viable tumor region from WSIs of liver cancer which effectively addresses the above issues. The proposed approach helps in eliminating false positives and generating consistent segmentation. Post segmentation we propose a methodology for predicting the tumor burden by estimating the whole tumor region by constructing a convex hull around the viable tumor region.