<p>The overgrowth of tumor cells in the liver results in Hepatocellular Carcinoma with unfamiliar symptoms in its earlier stages. Though the treatments facilitate transplantation, freezing, etc. the recognition of such tumors is to be made early using ultrasound images. This article proposes a Disparity Learning Network for Pixel Differentiation to recognize Hepatocellular Carcinoma from ultrasound image inputs. First, the conventional textural features are extracted from the input image from which the disparity for differential pixel distribution is estimated. This disparity is computed based on pixel absence in a well-distributed region and the corresponding error occurrence. In this case, the disparity network is constructed using one conditional and one training layer for disparity classification and new distribution training. In the classification process, the absence and high-density factors are differentiated for region-wise disparity estimation. Such estimation is used for training further classifications, for differentiating non-disparity regions. The maximum disparity pixel distributed region is recognized as the infected region from the given input.</p>