Visual Geometry Group based on U-Shaped Model for Liver/Liver Tumor Segmentation

Liver cancer is the primary reason of death around the globe. Manually detecting the infected tissues is a challenging and time-consuming task. The computerized methods help make accurate decisions and therapy processes. The segmentation accuracy might be increased to reduce the loss rate. Semantic segmentation performs a vital role in infected liver region segmentation. This article proposes a method that consists of two major steps; first, the local Laplacian filter is applied to improve the image quality. The second is the proposed semantic segmentation model in which features are extracted to the pre-trained VGG16 model and passed to the U-shaped network. This model consists of 51 layers: input, 23 convolutional, 4 max pooling, 4 transpose convolutional, 4 concatenated, 8 activation, and 7 batch-normalization. The proposed segmentation framework is trained on the selected hyperparameters that reduce the loss rate and increase the segmentation accuracy. The proposed approach more precisely segments the infected liver region. The proposed approach performance is accessed on two datasets such as 3DIRCADB and LiTS17. The proposed framework provides an average dice score of 0.98, which is far better compared to the existing works.