Comparative Study of Liver Segmentation using U-Net and ResNet50

Medical tomography has been engaged to backing medical findings and medication. Radiography revels the decisive evidence to doctors in identifying dreadful tumors and potential ill-health findings, makes ease for surgeons to treat patients with optimal surgical methods. Particularly segmentation’s of tomography images to identify the area of interest such as tumor regions and organ boundaries etc. such that assuage complexity of understanding medical images. Automatic organ segmentation can be achieved with the help of Deep Learning algorithms this excites the attention of researchers to focus on these ace areas of research. This paper mainly focuses on two imperatives of DL architectures like U-Net and ResNet50 in the experiments so that efficacy of algorithms interim’s of comparative studies. Metrics used such as accuracy and Dice coefficients and the results were analyzed. Dataset which we acquired from Monai medical imaging, our motive is to introduce usage of Monai and PyTorch with the help of python programming language along with DL model for making segmentation of liver from computer assisted tomography scan. Results discloses that both ResNet50 and U-Net shows the tremendous work in purging false positive values. Moreover, accuracy metrics alone is not sufficient in making significant measure of efficacy about segmentation. The inferences are useful in calibrating surgical operations with respect to boundaries and critical regions which helps the doctors for making right choices to treat patients.