Modified U-NET on CT images for automatic segmentation of liver and its tumor
R. V. Manjunath, Karibasappa Kwadiki
Abstract
Liver and tumor segmentation from abdominal CT images is a crucial phase in machine-assisted experimental involvements. FCNN has revealed extraordinary achievement in executing semantic segmentation by using convolutional layers for the whole planning and skip connections to combine diverse resolution features along with estimations in fruitful networks, such as UNet, ResUNet, etc. However here we recommend a system that select features automatically with convolutional layers, like modified Unet, and holds the spatial suggestion of each extracted feature. In the current study, an upgraded modified Unet procedure was applied on LITS 2D liver and tumor segmentation on CT images where it has achieved a dice similarity coefficient of 96.15% and 89.38% for liver and tumor, in addition, the algorithm was tested on a 3DIRCADb dataset where the system attains a DSC of 91.94% and 69.80% on CT images for segmentation of liver and its tumor effectively and the outcomes evidenced the efficiency of perfection.