Deep learning based two-fold segmentation model for liver tumor detection

Abstract

Liver Tumour (LT) develops when healthy cells undergo abnormal DNA changes that cause them to grow and divide uncontrollably. In manual examination, evaluation might be changed by the unique perception of the observers, which depends on their expertise and subjectivity. Therefore, computer-aided intelligent tools are established to eliminate subjectivity and increase the performance. To overcome these challenges, a novel Two-fold Segmentation of Liver Tumour (TFSLT) model for accurately detecting the liver tumour using computed tomography (CT) images. Initially, the CT images are pre-processed using Normalized-Modified Anisotropic Diffusion Filtering (NMADF) Algorithm to reduce the noise artifacts. These pre-processed CT images are taken as input to the Canny Edge Detector (CED) for detecting the edges of the liver. Based on these edges, the first-fold segmentation process is performed using the Jaccard metric-based Watershed (JMWS) algorithm to accurately segment the liver region. Improved Deep Neural Network (IDNN) is utilized to classify the LT into normal, Hepatocellular carcinoma (HCC), Cholangio carcinoma (CC) and Metastatic tumour (MT). Modified Elephant Herd Optimization (MEHO) algorithm for the MEHO algorithm for selecting the features of the images. Finally, the Improved Expectation-Maximization (IEM) Algorithm as second-fold segmentation process to segment the different abnormal classes. The performance of the proposed TFSLT approach is assessed using the specific metrics like recall, precision, specificity, accuracy and F1 score. The experimental findings reveal that the proposed TFSLT approach achieves a better accuracy range of 99.57% for detecting LT in its early stages.