Comparison of Data-Based Methods for Selecting Treatment Therapy in Hepatocellular Carcinoma

Yijie Zhao,Bin Zheng,Hao Ye,Jing Zhong,Haiwei Wang

DOI: https://doi.org/10.1109/safeprocess52771.2021.9693696

2021-01-01

Abstract:The initial treatment of patients with hepatocellular carcinoma (HCC) is highly correlated with their survival time. This study aimed to examine the pattern of clinical decisions of advanced physicians using data-based models to help inexperienced physicians select proper treatment. Data from 3348 patients with HCC collected from 16 top-level hospitals in China were used. Each patient report had 32 data entrances. Patients were initially treated surgery or transcatheter chemoembolization. These data were used to train seven different machine learning algorithms, including decision tree, random forest, linear SVM, Gaussian SVM, naive Bayes, k-nearest neighbors, and feedforward neural network. Outputs of these models in predicting the initial treatment on new patients were compared with data of accuracies, recalls, precisions, and receiver operating characteristic (ROC) curves between computer algorithms and human judgment. The accuracies of the above seven algorithms were 84.30%, 85.40%, 85.40%, 87.20%, 78.60%, 84.10%, and 87.40%, respectively. The area under the ROC curves of seven methods ranged between 0.88 and 0.94. In general, the Gaussian SVM and feedforward neural network showed the best performance. Therefore, big data and machine learning were useful in investigating physicians’ decision-making for treating patients with HCC.