Automatic Liver Damage Segmentation on 3D Images Using VGG16 and ResNet-50 Neural Networks

Abstract

Liver cancer is the second most dangerous cancer in the world. Medical experts still manually do most liver segmentations of Computed Tomography scans. Automatic segmentation of the liver and lesions is an important step towards computer-aided decision support systems. This thesis can be the basic foundation for developing related research at Can Tho University in the future. The topic has made positive scientific contributions to apply information technology into medicine. At the same time, it brings a big meaning to the society in general and the public health sector in particular. In this thesis, we have successfully deployed an algorithm for segmenting the liver and its lesions using two cascaded segmentation networks and one detector with VGG-16 and ResNet-50 Neural Networks. The liver images result illustrates the liver damage per each liver slice.