Automatic segmentation and classification of the liver tumor using deep learning algorithms

Abstract

Liver tumors are one of the life-threatening cancers with the fastest-growth rates worldwide. Early detection of tumors may therefore reduce morbidity and increase the survival rate. The development of automated techniques for the precise segmentation of hepatic tumors is essential for assisting doctors in tumor diagnosis and preoperative planning for surgical treatment of the liver which reduces the risk of surgical resection. The classification and segmentation of hepatic tumors in Computerized Tomography (CT) scan pose a great challenge due to noise, unclear boundaries, heterogeneity, and variability in tumor tissue appearance, shape, size, and location. In this study, we describe a novel method for automatic segmentation and classification of hepatic tumors in CT scan images using Deep Convolutional Neural Networks. For tumor segmentation, we created a modified Dense U-net model. The classification framework is based on a novel deep CNN with a pre-trained VGG-16 network to distinguish between normal and malignant liver tumors. The proposed system was evaluated based on MICCAI 2017 Liver Tumor Segmentation (LiTS) Challenge dataset and obtained the best result with a Dice Score of 95.40%, Jaccard Index of 92%, and accuracy of 92.60% for segmentation and the classification model has achieved an accuracy of 96%, Sensitivity of 95.80%, Specificity of 96.20% and Precision of 95.80%.