ABSTRACT The liver is a very important and complex organ in our body. Efficient liver segmentation is a very important interesting and challenging task in the field of medical image processing. The abdomen consists of many organs among which the liver consists of very important parts, such as right lobule, left lobule and hepatic duct. In this research work, for computed tomography (CT) images we designed a feasible upstream approach to segment a liver region from abdomen using multi-level deep convolutional network and Fractal Residual Network (FRN). Our convolutional neural network is designed in such a way that it learns to allot different probabilities for each super pixel based on different regions of the liver by these indirectly different classes are created based on different intensities of super pixels. Further FRN is used to identify the tumor region and finally refinement of tumor segmentation is done with active contour model method. The proposed model (MDCN + FRN) is evaluated on CT images 125 patients from TCI dataset and achieved a dice similarity in an average of 0.89 in training and dice similarity in an average of 0.86 in testing. Compared to other segmentation methods, the experiment conducted shows better performance from proposed method.