The need for CT scan analysis is growing for pre-diagnosis and therapy of abdominal organs. Automatic organ segmentation of abdominal CT scan can help radiologists analyze the scans faster and segment organ images with fewer errors. Automatic segmentation of organs is a challenging task due to the inter-patient variety of organ shapes and locations. This variety seems to be extremely large among patients with different body anatomy or unexpected disease. The organ detection and localization can be a vital step to help segmentation approaches in form of bounding boxes or location probability map. Most of the approaches tend to perform atlas registration to achieve a relatively appropriate location prior map while some methods perform regression forest algorithm to achieve a bounding box around each organ. In this thesis, we exploit DEEDS registration method and propose an adaptive 3D region growing with subject-specific conditions. The condition involves the intensity distribution of most probable voxels in the prior map along with location priors. We also incorporate the boundary of the target organ to restrict the region growing. In order to obtain strong edges and high contrast among organs, we propose effective contrast enhancement algorithm to facilitate more accurate segmentation. We compare our method with the method of hard thresholding on DEEDS prior map and also with the multi-atlas label fusion on the registered label with 13 organs. The registration process is a time-consuming preparation step making these methods inappropriate for medical tools. Regression forest approaches, perform organ localization in a reasonable time. However such methods are not robust enough to detect organs with large variations in shapes, locations, and appearances despite their use of hand-craft features. In this thesis, we also propose an efficient liver segmentation with our 3D to 2D fully convolutional network (3D-2D-FCN) with dice cost function. This network extracts the 3D surface feature of liver and is capable of localization and segmentation at the same time. By means of dice cost function, the network ignores a large number of background voxels, leading to balancing positive and negative voxels for better classification by network. Finally, the segmented mask is enhanced by means of the conditional random field on the organ’s border. Consequently, we segment a target liver in less than a minute and compare our method with a state-of-the-art approach. The experimental results show that our proposed method can outperform the existing state-of-the-art algorithm in time and accuracy. Keywords: Medical image processing, 3D localization and segmentation, Probabilistic atlas, medical registration algorithms, 3D fully convolutional deep network