A Frame Work For Automatic Liver Image Segmentation Using CU-Net

Abstract

Medical image analytics became very important component in various medical practices. This process changed entire medical industry and facilitating researchers, doctors to understand more above human body for providing solutions to various unsolved problems. During the diagnosis of health issues doctors always preferring X-Rays, CT scan, MRI etc.. these images are widely helping doctors to know exact sate of the disease, tracking progress after recovery etc.  but in present scenario there is a high shortage of radiologists who can able to draw conclusions by manually analysing images on other hand certain kind of disease detection at early stage is not possible to detect through human eye so AI techniques  are very much required to automate image analysis process and to assist radiologist. Image segmentation is one of the crucial activity in image analysis early days various  machine learning techniques are used for image segmentationsuch as but clustering, simple thresholds, contour fitting ,SVM etc..   but these techniques are  failed when images having high noise. So to solve these issues in this paper we are proposing a deep learning model Called U-Net for liver image segmentation for separating cancer cells from the liver. To train the model we have taken CT scan images of 28 volumes and corresponding segment labels. Here the we have implemented CU-Net with five blocks encoding and 5 blocks Decoding  for image  reconstruction  by varying various hyper parameters such as Kernel regularization(L2 lamda),Batch normalization,  Drop out layers a . The model working very effectively in liver image segmentation.