Developmental artificial neural network model to evaluate the preoperative safe limit of future liver remnant volume for HCC combined with clinically significant portal hypertension

Background & aims: Finding a way to comprehensively integrate the presence and grade of clinically significant portal hypertension, amount of preserved liver function and extent of hepatectomy into the guidelines for choosing appropriate candidates to hepatectomy remained challenging. This study sheds light on these issues to facilitate precise surgical decisions for clinicians. Methods: Independent risk factors associated with grade B/C post-hepatectomy liver failure were identified by stochastic forest algorithm and logistic regression in hepatitis B virus-related hepatocellular carcinoma patients. Results: The artificial neural network model was generated by integrating preoperative pre-ALB, prothrombin time, total bilirubin, AST, indocyanine green retention rate at 15 min, standard future liver remnant volume and clinically significant portal hypertension grade. In addition, stratification of patients into three risk groups emphasized significant distinctions in the risk of grade B/C post-hepatectomy liver failure. Conclusion: The authors' artificial neural network model could provide a reasonable therapeutic option for clinicians to select optimal candidates with clinically significant portal hypertension for hepatectomy and supplement the hepatocellular carcinoma surgical treatment algorithm. Plain language summary Hepatectomy involves removing the tumor from the liver and is considered the most effective treatment for hepatocellular carcinoma (HCC). Clinically significant portal hypertension is characterized by the presence of gastric and/or esophageal varices and a platelet count <100 x 10(9)/l with the presence of splenomegaly, which would aggravate the risk of post-hepatectomy liver failure, and is therefore regarded as a contraindication to hepatectomy. Over the past few decades, with improvement in surgical techniques and perioperative care, the morbidity of postoperative complications and mortality have decreased greatly. Current HCC guidelines recommend the expansion of hepatectomy to HCC patients with clinically significant portal hypertension. However, determining how to select optimal candidates for hepatectomy remains challenging. The authors' artificial neural network is a mathematical tool developed by simulating the properties of neurons with large-scale information distribution and parallel structure. Here the authors retrospectively enrolled 871 hepatitis B virus-related HCC patients and developed an artificial neural network model to predict the risk of post-hepatectomy liver failure, which could provide a reasonable therapeutic option and facilitate precise surgical decisions for clinicians.