Liver segmentation using 3D CT scans.
Anura Hiraman · ResearchSpace (University of KwaZulu-Natal) · 2018
Deep learning algorithms have shown ground breaking performance in the field of radiology.One of its primary uses is finding a region of interest and object or lesion detection which is a key part of diagnosis.Segmentation of organs or substructures allows for quantitative analysis of the organ or substructure.This will be used for segmentation of lesions which plays an important role in diagnosis and prognosis of liver diseases or abnormalities.The organ of interest in this research is the liver.Segmentation of the liver from CT scans plays an important role in the study of the liver functions and can assist in the diagnosis of liver diseases.Liver segmentation aims to accurately detect and delineate the liver, separating it from surrounding organs and isolating it for intricate analysis.A method for liver segmentation using 3D CT scans is proposed in this dissertation.The approach used for liver segmentation employs deep learning techniques.Convolutional neural networks (CNNs) are implemented to separate the liver from its background in the CT image.Firstly, a CNN is used to classify each slice of a 3D scan in order to remove slices that do not belong to the abdomen.This is done to obtain a region of interest that contains the liver that will be further processed during liver segmentation.Furthermore, the abdominal slices are processed by a CNN to segment the liver.The resulting segmented liver slices are then reassembled into a volume for post-processing which involves morphological operations.All 3D scan volumes used for experimental evaluation are taken from the Medical Image Computing and Computer Assisted Intervention (MICCAI) 2007 grand challenge datasets.The results obtained from evaluation indicate the effectiveness of the proposed liver segmentation method in accurately segmenting the liver from a 3D CT image.v