Integrated prognostic framework for survival status, Cancer stages and time estimation in hepatocellular carcinoma using machine learning approaches

Abstract

Current medical limitations in assessing cancer survival status, stage, and duration necessitate advancements beyond physical indicators and conventional diagnostic techniques. Improving the quality of patient treatment is an ongoing objective of the medical community. Accurate prediction of survival status, survival duration, and stage in persons with early liver cancer is crucial for informed treatment and prognosis decisions. This study presents an innovative three-stage prognostic framework for liver cancer, addressing the shortcomings of existing diagnostic methods. The proposed model forecasts patient survival status (alive or deceased) and survival duration across seven intervals (0–6, 7–12, 13–24, 25–36, 37–48, 49–60 months, and beyond 5 years) and is assessed using metrics including Accuracy, Precision, Recall, AUC, and F1-score. The subsequent element assesses cancer stage classifications (I, UNK, II, IV, IIIA, IIIC, IIIB, IIINOS) employing similar metrics. Our experimental findings demonstrate enhanced prognostic accuracy via twelve popular based machine learning models, including Gradient Boosting, Random Forest, Support Vector Machine, Light Gradient Boosting Machine, Extreme Gradient Boosting, AdaBoost, Extra Trees, Logistic Regression, Linear Discriminant Analysis, Decision Tree, K-Nearest Neighbors, and Naive Bayes classifiers. Using a subset of SEER (Surveillance, Epidemiology, and End Results) data may produce diverse results; nonetheless, by univariate feature identification and correlation analysis, the models are constructed efficiently. Light Gradient Boosting Machine has exceptional efficacy in survival analysis, attaining 89.13% accuracy, 85.99% AUC, 88.93% recall, 86.99% precision, and 87.23% f1-score, surpassing other data-driven models. The Gradient Boosting model outperforms in cancer stage prediction, with 97.97% accuracy, 100% AUC, 97.97% recall, 98% precision, and 97.95% f1-score. For survival time prediction, the Logistic Regression model shows the highest performance with 97.07% accuracy, 99.43% AUC, 97.02% recall, 97.23% precision, 97.02% recall, and 95.77% f1-score. This framework incorporates interpretable model prediction analysis to identify significant variables influencing regression and classification outcomes inside of the three-stage liver cancer matrix. Consequently, our research presents a more transparent and dependable framework for machine learning–based cancer outcome prediction. To the best of our knowledge, this work proposes a unified three-stage approach that integrates liver cancer survival status, stage, and time classification and regression.