Performance analysis of liver tumor classification using machine learning algorithms

Liver is most important organ in human body. Mainly there are two types of liver cancers- “liver abscess (LA), and hepatocellular carcinoma (HCC)”. Computed tomography (CT) is used to identify these liver cancers. The shape of liver, tumor, and tumor location changes with patients, hence it is very difficult to classify liver tumors. Early detection of liver cancers such as LA and HCC are very essential to reduce the mortality rate. Medical image analysis techniques are used for identification of liver abnormalities. In this paper, different machine learning algorithms such as support vector machine (SVM), K-Nearest neighbour (KNN), decision tree (DT), ensemble, and naive Bayes (NB) are used to classify the tumor as LA, and HCC. The steps required for classification are “preprocessing, liver segmentation, feature extraction, and classification”. 68 CT images are collected from different hospitals in Tirupati to train the model, and the models are validated using accuracy, specificity, sensitivity, Matthew correlation coefficient (MCC), and F1-score etc. From the performance analysis of different classifiers, it is observations that accuracy of SVM classifier is improved by 10%, specificity is improved by 40%, sensitivity is improved by 28%, precision is improved by 66.67%, MCC is improved by 8%, F1-score is improved 4%, and kappa is improved by 20.14% compare to KNN, whereas error is reduced by 33%. SVM performance is also improved by 22.22%, 45.83%, 40%, 62%, 20%, and 70% with respect to accuracy, precision, specificity, MCC, F1-score, and kappa compared to DT classifier whereas 50% reduction in error. We can conclude that SVM classifier gives better performance compare to all other classifiers in the study.