Predicting hepatocellular carcinoma survival with artificial intelligence

Despite the extensive research on hepatocellular carcinoma (HCC) exploring various treatment strategies, the survival outcomes have remained unsatisfactory. The aim of this research was to evaluate the ability of machine learning (ML) methods in predicting the survival probability of HCC patients. The study retrospectively analyzed cases of patients with stage 1-4 HCC. Demographic, clinical, pathological, and laboratory data served as input variables. The researchers employed various feature selection techniques to identify the key predictors of patient mortality. Additionally, the study utilized a range of machine learning methods to model patient survival rates. The study included 393 individuals with HCC. For early-stage patients (stages 1-2), the models reached recall values ​​of up to 91% for 6-month survival prediction. For advanced-stage patients (stage 4), the models achieved accuracy values ​​of up to 92% for 3-year overall survival prediction. To predict whether patients are ex or not, the accuracy was 87.5% when using all 28 features without feature selection with the best performance coming from the implementation of weighted KNN. Further improvements in accuracy, reaching 87.8%, were achieved by applying feature selection methods and using a medium Gaussian SVM. This study demonstrates that machine learning techniques can reliably predict survival probabilities for HCC patients across all disease stages. The research also shows that AI models can accurately identify a high proportion of surviving individuals when assessing various clinical and pathological factors.