23 Using a hybrid PSO-BP model to predict survival rate after partial hepatectomy

Abstract

Objectives Hepatocellular carcinoma (HCC) is the fifth most common cancer worldwide, and is the second most frequent cause of death from cancer. Partial hepatectomy is the standard curative treatment for HCC and has a 5-year survival rate of over 50%. However, given a high 5-year postoperative recurrence rate of approximately 40–70%, it is important to accurately predict survival rate after partial hepatectomy so that high-risk patients can be screened and decisions made on adjuvant therapy. Therefore, this study proposed a hybrid PSO-BP model for the prognosis of early HCC after partial hepatectomy. Methods Data on patients who were operated on between 2006 and 2015 were collected and prospectively studied. These data were randomly divided into a training set (75%) and a validation set (25%). Using the risk factors significantly related to survival in a multivariable Cox analysis model, the hybrid PSO-BP model which integrates BP’s non-linear capability and PSO’s global search ability was employed to predict survival rate after liver resection, the results were compared with the traditional Cox proportional hazards model, and their performances were evaluated using the area under curve (AUC) and CI (indicates confidence). Results Six factors including tumor size, tumor number, alpha-fetoprotein (AFP), microvascular invasion (MVI), tumor capsule and survival time were identified as significant risk factors based on linear regression analysis using the Cox model and were selected as input variables for the proposed hybrid PSO-BP model. Model evaluation results show that the proposed PSO-BP model demonstrates better performance with a larger AUC (0.887, p = 0.0102) and higher 95% CI: (0.771 to 0.845) than that of the traditional Cox proportional hazard model (AUC: 0.816; 95% CI: 0.667 to 0.794) during the training period. This finding was confirmed by the validation set. Acknowledgments This work was financially supported by Jiangxi Province Major Academic Disciplines and Technical Leaders Training Program (Grant No. 2014BCB22009) and the National Natural Science Foundation of China (Grant No. 81260231).